Incidental finding of pulmonary nodules and lung cancer screening studies

Only one study was undertaken in the UK (incidental findings on CT angiogram).3 Sixteen studies were from North America and Canada,4–19 13 from Europe20–32 and two from Korea and Japan.33 ,34 Nodule and lung cancer prevalence by country/continent in which the study was performed is shown in table 5.

Nodule prevalence in the largest screening study of 53 439 participants aged 55–74 years, with a history of at least 30 pack-years of smoking2 was 25.9% with a lung cancer prevalence of 1.1%. The prevalence of nodules and lung cancer in the screening and incidental finding studies are shown in table 6. Screening studies recruit asymptomatic people at higher risk of lung cancer, whereas studies of the incidental finding of pulmonary nodules include a mixture of younger patients at low risk of lung cancer (trauma studies) and people at higher risk who may have similar risk factors for lung cancer (cardiac CT).

Studies in patients with known cancer

A number of case series have examined the prevalence of malignant nodules in patients with known cancer. Interpretation of these studies is limited by their heterogeneous nature—in particular, reporting of the stage of the primary tumour, the time from diagnosis of the primary tumour in relation to the CT scan demonstrating nodules, the definition of nodule size and selection criteria for further investigation or follow-up.

Three studies grouped the primary sites according to predicted likelihood of metastasising to the lung based on the following groups:

1. Patients with squamous cell cancers of the head and neck.

2. Patients with lymphoma or leukaemia.

3. Patients with carcinomas of the urinary bladder, breast, uterine cervix, biliary tree, oesophagus, ovary, prostate or stomach.

4. Patients with carcinomas of the salivary glands, adrenal gland, colon, parotid gland, kidney, thyroid gland, thymus or uterus.

5. Patients with melanoma, sarcoma or testicular carcinoma.

Using multivariate analysis, one study36 found an association between the type of extrapulmonary cancer and the proportion of lung cancer/metastatic nodules. Groups 1–3 were more likely to have a lung primary and group 5 more likely to be metastatic, although numbers in each group were small. Another study37 did not find this association or an association with stage (III or IV) of the primary tumour. A third study38 found that group 4 patients were more likely to have a malignant nodule. A further study39 of patients with known extrapulmonary cancer referred for resection of a lung nodule found that 68% of resected nodules were malignant, of which 58% were non-small cell lung cancer (NSCLC); 113 nodules (10%) were metastases. Logistic regression analysis suggested that nodules were more likely to be NSCLC in patients aged >55 years, smokers and if the known cancer was breast or prostate.

Two case series reported subcentimetre lung nodules detected preoperatively on CT in patients who had undergone curative surgery for lung cancer in Asia. One study40 of 223 patients found that 26% of patients had nodules, of which 6% were malignant. Half of the malignant nodules were found in the primary tumour lobe. The second study41 reported nodule prevalence in 582 patients in the non-primary lobe which was not resected at the time of surgery. This study group included only patients with 24 months’ follow-up CT data (141 of 582 undergoing resection); 62 (44%) patients had a nodule and 3% were malignant. A study from the UK42 included 551 patients with lung cancer who had a staging CT scan and who were considered operable. Eighty-eight patients (16%) were found to have small non-calcified pulmonary nodules (size range 4–12 mm). Adequate follow-up (histological confirmation or CT follow-up for 24–48 months) was possible only in 25 patients who had a total of 36 nodules. Twenty-five nodules (70%) were subsequently confirmed to be benign, four (11%) were malignant and the nature of seven (19%) could not be determined.

Smyth et al43 reported histological findings from biopsy of suspicious lesions in 229 patients with previous malignant melanoma. They found that 88% of the biopsies were malignant; 69% were metastatic melanoma, 14% were new primary NSCLC and 5% were recurrent metastatic non-melanoma cancer. Multivariate analysis of predictive factors for melanoma metastases demonstrated ORs of 9.0 for stage III/IV disease, 3.44 for multiple nodules, 0.21 for smoking and 0.26 for previous non-melanoma cancer. Margolis et al44 retrospectively reviewed 116 patients with oesophageal cancer and found that 19% had solitary nodules and 3% had multiple nodules. Diagnosis was established in 19 of the 22 solitary pulmonary nodules (SPNs). None were metastases but 4 of 22 were lung cancer. All four cases of multiple nodules were classed as metastases without biopsy.

Summary

No studies were found that compared the features of pulmonary nodules according to the route of presentation. Thus the differences found will be influenced by selection bias and study protocol. The best evidence came from extrapolated evidence from CT screening trials where entry criteria were clearly defined. Thus the overall conclusion has to be that the route of presentation should not be an important factor in the management of pulmonary nodules.

Evidence statement

• The reported prevalence of non-calcified lung nodules is higher in screening studies than in studies reporting nodules as incidental findings on non-staging CT scans, but these differences are likely to reflect selection bias and protocol differences. Evidence level 3

• The reported prevalence of malignant nodules is similar in screening studies and in studies reporting nodules as incidental findings. Evidence level 3

• In screening studies, the prevalence of malignant nodules varies according to the screened population. Evidence level 2+

• The prevalence of malignant nodules may be higher in patients with extrapulmonary cancer, but studies are small and subject to selection bias. The relationship between the risk of nodule malignancy and the time from diagnosis of the primary tumour is not known owing to inconsistent reporting of this variable. Evidence level 3

• In patients with known extrapulmonary cancer there is conflicting evidence as to whether the primary site predicts whether the lung nodule is malignant or whether it is a metastasis or lung primary. Evidence level 3

• There is limited evidence relating to the aetiology of coexistent lung nodules in patients with known primary lung cancer. The reported prevalence of malignancy in sub-12 mm coexistent nodules in patients selected to undergo curative surgery is 3–11%. Evidence level 3

Recommendations

• Use the same diagnostic approach for nodules detected incidentally as those detected through screening. Grade D

• Consider using the presence of previous malignancy as a factor in the risk assessment for further investigation. Grade D

• Do not prioritise management of pulmonary nodules according to the route of presentation. Grade D

• Evaluate coexistent lung nodules detected in patients with known lung cancer otherwise suitable for radical treatment in their own right; these nodules should not be assumed to be malignant. Grade D

Thirty studies were identified that evaluated clinical and radiological characteristics of nodules in relation to probability of malignancy.4 ,36–39 ,43 ,45–68 Twenty-eight were retrospective case series, one was a screening study and one was based on a retrospective literature review. Eleven studies included patients with multiple pulmonary nodules.36–39 ,43 ,46 ,48 ,53 ,60 ,66 ,68 Only one study was conducted and validated in a UK population.68 Six studies included patients with known extrapulmonary cancer.36–39 43 68

The studies can be grouped into four categories:

1. Studies that evaluated clinical and radiological characteristics and/or described prediction models (n=18).

2. Studies that externally validated prediction models from category 1 (n=5).

3. Studies that compared prediction models with clinical judgement (n=2).

4. Studies that evaluated predictors of metastases versus primary lung cancer (n=5).

Studies that evaluated clinical and radiological characteristics and/or described prediction models

These studies had a wide range of inclusion criteria, differing demographic profiles, different criteria for labelling nodules as benign or malignant, and a wide variation in the prevalence of malignancy (1.8–75%).4 ,37 ,45–48 ,50 ,53 ,54 ,56–61 ,64–66 Overall these studies identified:

1. Eight clinical predictors of malignancy: age, current or ever smokers, time since quitting smoking, pack-years, family history of lung cancer, history of cancer >5 years before nodule detection, any history of previous cancer and haemoptysis.

2. Thirteen radiological predictors of malignancy: diameter, distance from pleura >10 mm, spiculation, ground-glass appearance, pleural indentation, vascular convergence, circumference diameter ratio, upper lobe location, volume, growth, air bronchogram, lymphadenopathy and cavity wall thickness.

3. Five radiological predictors for benign aetiology: calcification, smooth border, cavitation, satellite lesions and perifissural location.

4. Two biochemical predictors of malignancy (C reactive protein (CRP) and carcinoembryonic antigen (CEA))

Of these, nine predictors of malignancy (four clinical and five radiological) were identified consistently in two or more studies which reported multivariate analysis:

1. Age (OR=1.04–2.2 for every 10-year increment)

2. Current or former smoking status (OR=2.2–7.9)

3. Pack-years of smoking

4. Previous history of extrapulmonary cancer

5. Nodule diameter (OR approximately 1.1 for each 1 mm increment)

6. Spiculation (OR=2.1–5.7)

7. Upper lobe location

8. Pleural indentation

9. Volume doubling time <400 days.

Predictors of a benign aetiology included presence of a diffuse, central, laminated or popcorn pattern of calcification (OR=0.07–0.20) and perifissural location.

de Hoop et al45 specifically assessed perifissural nodules (PFNs) detected on CT screening in the NELSON study. These are homogeneous solid nodules, attached to a fissure with a lentiform or triangular shape and may be subpleural (figure 8). Seven hundred and ninety-four of the 4026 nodules (19.7%) detected at baseline screening were classified as PFNs, and were followed up according to the standard protocol. At first follow-up 66 PFNs (8.3% of all PFNs) grew with a volume doubling time (VDT) <400 days. One was resected and was proved to be a lymph node. None of the other PFNs turned out to be malignant after 5 years of follow-up. In a similar retrospective review, Ahn et al47 found 234 PFNs (28% of all non-calcified nodules) in 98 subjects participating in a CT screening study. None of the PFNs developed into cancer during the study 2-year follow-up period, or during 7½ years of follow-up thereafter. PFNs are thought to be intrapulmonary lymph nodes on the basis of their CT features and histological correlates. Four studies examined histologically confirmed intrapulmonary lymph nodes (n=38, 19, 18 and 11, respectively) and characterised their CT features.69–72 In all these studies and that of de Hoop, the nodules were relatively small (<10 mm). Caution may be required in larger PFNs (>10 mm) in the presence of known non-lung primary cancers as there is anecdotal evidence of malignancy in these nodules.

• Download figure
• Open in new tab
• Download powerpoint

Figure 8

Appearance of perifissural nodules (PFN) as defined in de Hoop et al. (Reproduced with permission.)

Gurney (Bayesian method)50 performed a retrospective literature review and applied the odds-likelihood ratio form of the Bayes theorem to calculate the probability of a nodule being benign or malignant. Only studies that included >100 patients were analysed but there was a wide variation in average nodule size and prevalence of malignancy, and studies were subject to methodological bias. A total of 15 malignant and 19 benign findings were identified for nodules. The most important predictors of malignancy were spiculation, diameter and cavity wall thickness, while predictors of a benign aetiology were VDT >465 days and calcification.

Five studies derived composite prediction models based on a combination of clinical and radiological factors using multivariate logistic regression analysis:

Swensen et al64 (Mayo Clinic model) evaluated, at a single centre in the USA, the probability of malignancy in 419 radiologically indeterminate SPNs that measured between 4 and 30 mm in diameter on CXR. Patients with a diagnosis of cancer within 5 years before the discovery of the nodule, and any history of lung cancer were excluded. Mean age of the patients was 62 years, 51% were male and 67% were current or past smokers. Sixty-five per cent of nodules were benign, 23% malignant and 12% were indeterminate. Three clinical characteristics (age, smoking status and history of cancer more than 5 years previously) and three radiological characteristics (diameter, spiculation and upper lobe location) were independent predictors of malignancy. The area under the receiver operating characteristic (ROC) curve (±SE) for the prediction model was 0.83 (±0.02). The model was validated on data from a separate group of 210 patients. The area under the ROC curve (±SE) for the validation set was 0.80 (±0.03). Calibration curves for the derivation and validation sets showed a good agreement between the predicted probability and the observed frequency of malignant SPNs.

Gould et al49 (Veterans Administration (VA) model) studied 375 patients enrolled from multiple centres in USA with SPNs measuring between 7 and 30 mm on CXR. Patients with a history of cancer, including lung cancer within 5 years were included but the authors were unable to identify patients who had a history of cancer more than 5 years before nodule detection. Mean (±SD) age of the patients was 65.9 (±10.7) years, 98% were male and 94% were current smokers or former smokers. Fifty-four per cent of SPNs were malignant and 46% benign. Independent predictors of malignant SPNs included a positive smoking history (OR=7.9; 95% CI 2.6 to 23.6), older age (OR=2.2 per 10-year increment; 95% CI 1.7 to 2.8), larger nodule diameter (OR=1.1 per 1 mm increment; 95% CI 1.1 to 1.2) and time since quitting smoking (OR=0.6 per 10-year increment; 95% CI 0.5 to 0.7). The area under the ROC curve (±SE) was 0.79 (±0.05) and the model was well calibrated.

Li et al59 studied 371 surgically resected SPNs ≤30 mm in diameter at a single centre in China. Median patient age was 57.1 years, 53% were male and 42% had a history of smoking. Patients with a diagnosis of cancer within 5 years before the discovery of the nodule were excluded. Fifty-three per cent of the nodules were malignant and 46% were benign. Independent predictors of malignancy included age (OR=1.07; 95% CI 1.05 to 1.09), diameter (OR=1.96; 95% CI 1.38 to 2.60), clear border (OR=0.25; 95% CI 0.13 to 0.45), calcification (OR=0.20; 95% CI 0.07 to 0.59), spiculation (OR=2.09; 95% CI 1.06 to 4.14) and family history of cancer (OR=3.55; 95% CI 1.26 to 9.97). The area under the ROC curve for the model (0.89; 95% CI, 0.78 to 0.99) was higher than those derived by Swensen et al and Gould et al. Although a history of smoking was a significant predictor of malignancy on univariate analysis, this was not significant on multivariate analysis. The authors hypothesised that this might have been owing to the high prevalence of adenocarcinoma in their population (67% of all malignant SPNs were adenocarcinomas). Data from an additional 62 patients were used to validate this model but the authors did not give any further details about ROC curves in the validation set. In addition, calibration curves were not reported for either the development or validation sets.

Yonemori et al65 studied 452 surgically resected SPNs ≤30 mm in diameter at a single centre in Japan. Mean patient age was 62 years, 55% were male and 49% had a history of current or past smoking. Any SPN diagnosed as metastatic extrapulmonary cancer was excluded. Patients with a history of cancer more than 5 years previously were included, but it was unclear if those with cancer within 5 years of nodule detection were also included. Seventy-five per cent of the nodules were malignant and 25% were benign. Independent predictors of malignancy identified were level of serum CRP, level of CEA, presence of spiculation, the CT bronchus sign (where a bronchus is seen to enter the nodule) and the absence of calcification; ORs and 95% CIs were not reported. The area under the ROC curve for the prediction model was 0.96, and 0.94 if biochemical variables (CRP and CEA) were not included. The model was validated on data from a separate group of 148 patients. The area under the ROC curve for the validation set was 0.84. CIs for the development and validation sets were not reported.

McWilliams et al46 (Brock University model) analysed data from two cohorts of participants undergoing low-dose CT screening. The development dataset included participants in the Pan-Canadian Early Detection of Lung Cancer Study (PanCan) while the validation dataset included participants involved in chemoprevention trials at the British Columbia Cancer Agency (BCCA). All participants were current or former smokers between 50 and 75 years of age without a history of lung cancer. The final outcomes of all nodules of any size that were detected on baseline low-dose CT scans were tracked. Parsimonious and fuller multivariate logistic regression models were prepared to estimate the probability of lung cancer. In the PanCan dataset, 1871 people had 7008 nodules, of which 102 were malignant, and in the BCCA dataset, 1090 people had 5021 nodules, of which 42 were malignant. Among those with nodules, the rates of cancer in the two datasets were 5.5% and 3.7%, respectively. Predictors of cancer in the model included older age, female sex, family history of lung cancer, emphysema, larger nodule size, location of the nodule in the upper lobe, PSN type, lower nodule count and spiculation. The final parsimonious and full models demonstrated areas under the ROC curve of more than 0.91 to 0.98 with good calibration, even for nodules that were ≤10 mm.

Herder et al55 performed an external validation of the Mayo clinic model and quantified the potential added value of fluorodeoxyglucose-positron emission tomography (FDG-PET) scanning in a population of patients with radiologically indeterminate pulmonary nodules. They demonstrated improved accuracy of the Mayo model by addition of a factor relating to a four-point intensity scale of FDG avidity. This model is described in greater detail in the section ‘FDG PET-CT and clinical risk prediction models’.

Studies that externally validated prediction models

Four studies externally validated the Mayo model, two validated the Bayesian method,50 two validated the VA model, and one study validated the Brock and Herder models. Three studies included patients referred for PET scan, and one included patients who had surgically resected nodules. The prevalence of malignancy varied from 40.6% to 73%. The area under the ROC curve (AUC) results for the Mayo model were 0.79–0.90, for the VA model 0.73 to 0.74 and for the Bayesian method 0.80 (one study did not quote the AUC for the Bayesian method).

Dewan et al52 compared the accuracy of predicting the probability of cancer in 52 patients with SPNs using Bayesian analysis and PET. Three patients with extrapulmonary malignancy were included. PET, as a stand-alone test, was better at classifying nodules as malignant or benign than either Bayesian analysis alone or Bayesian analysis plus PET scan.

Herder et al55 validated the Mayo model in conjunction with PET scanning in 106 patients referred for PET evaluation of an indeterminate lung nodule. Patients with prior malignancy within the past 5 years were excluded. The addition of PET scan findings (classified using a four-point scale) increased the AUC by 13% from 0.79 to 0.92.

Schultz et al62 compared the Mayo and VA models in 151 patients undergoing PET evaluation of lung nodules (maximum one on CXR and six on CT). The area under the ROC curve for the Mayo Clinic model (0.80; 95% CI 0.72 to 0.88) was higher than that of the VA model (0.73; 95% CI 0.64 to 0.82), but this difference was not statistically significant. Calibration curves showed that the Mayo model underestimated, while the VA model overestimated, the probability of malignancy.

Isbell et al67 evaluated the Mayo and Bayesian models in patients with pulmonary nodules referred for surgical resection. Area under the ROC curve was 0.78 (95% CI 0.70 to 0.85) for the Mayo model, and 0.80 (95% CI 0.73 to 0.87) for the Bayesian model. The Mayo model was well calibrated for the two highest quintiles of probabilities but underestimated the probability of malignancy for the lower quintiles. The Bayesian model underestimated probability for the lower quintiles and overestimated for the higher quintiles.

Al-Ameri et al68 compared the performance of four prediction models (Mayo, VA, Brock University and the model described by Herder et al) in a cohort of 244 patients with pulmonary nodules detected in routine clinical practice in a UK teaching hospital. Of the three CT based scores, the Mayo and Brock models performed similarly, and both were significantly more accurate than the VA model. The AUCs were 0.89 (95% CI 0.85 to 0.94) for the Mayo model, 0.90 (95% CI 0.86 to 0.95) for the Brock model, and 0.74 (95% CI 0.67 to 0.80) for the VA model. In patients undergoing FDG PET-CT, the Herder model had significantly higher accuracy than the other three models (AUC 0.92; 95% CI 0.87 to 0.97). When analysis was extended to include patients outside the original described inclusion criteria for each model, the accuracy remained high, especially for the Herder score (AUC 0.92). For subcentimetre nodules, AUC values for Mayo and Brock models were 0.79 (95% CI 0.63 to 0.95) and 0.85 (95% CI 0.77 to 0.94), respectively.

Studies that compared prediction models with clinical judgement

Swensen et al63 compared four physicians’ clinical judgements with the Mayo model in 100 patients with indeterminate pulmonary nodules. Although ROC analysis showed no statistically significant difference between the two, calibration curves revealed that physicians overestimated the probability of malignancy in patients with a low risk of malignant disease. Gurney compared the accuracy of four expert radiologists using clinical judgement with two other radiologists using Bayesian analysis in 66 patients with pulmonary nodules.51 The latter performed significantly better than the former (p<0.05) and misclassified fewer malignant nodules as benign.

Studies that specifically evaluated predictors of metastases versus primary lung cancer

A number of case series have examined the prevalence of malignant nodules in patients with known extrapulmonary cancer—these are described in the section on route of detection of pulmonary nodules.36–44 Because of their heterogeneous nature, these studies provide conflicting evidence as to whether the primary site predicts whether the lung nodule is malignant or whether it is a metastasis or lung primary.

Patients with multiple pulmonary nodules

The Brock model is the only multivariate model that included an analysis of multiple pulmonary nodules.46 In this model the presence of multiple nodules had a small negative effect on the likelihood of malignancy in any one nodule. The remaining studies were small case series48 ,53 ,60 ,66 which did not report multivariable analysis, and were based on specific patient populations which can broadly be divided into three groups:

1. Immunosuppressed patients including those with AIDS and post-transplant settings.

2. Nodules in patients with suspected or proven pulmonary infections (eg, TB, histoplasmosis and other fungal diseases).

3. Nodules in the setting of known diffuse parenchymal disease.

In the NELSON trial the nodule management algorithm was determined according to the largest nodule when more than one nodule was present. This is the best evidence for the effectiveness of this approach.29

Limitations and choice of predictive models

The accuracy and clinical utility of predictive models depend on the case mix of the population in which it was derived and the prevalence of malignancy in that population. The applicability of the predictors identified will depend on the methods used to identify the events (ie, nodules) and the method of evaluation (essentially CT or CXR). The clinical characteristics and results of the studies that developed predictive models are summarised in table 7.

The Mayo model was developed in a cohort of patients with lung nodules who were originally managed in the 1980s at a single tertiary care centre in the USA.64 The investigators excluded patients with a history of lung cancer or a history of extrathoracic cancer within 5 years, and 12% of the patients did not have a final diagnosis. The VA model did include patients with a history of lung cancer or a history of extrathoracic cancer within 5 years but evaluated a relatively smaller number of clinical predictors.49 The population in the VA model comprised mainly older male smokers, and nodule size range was 7–30 mm, hence the accuracy of this model is unknown in nodules that are smaller than 7 mm in diameter.

Herder et al55 validated the Mayo model in 106 patients referred for PET evaluation of an indeterminate lung nodule. Patients with prior malignancy within the past 5 years were excluded. The addition of PET scans findings (classified using a four-point scale) increased the AUC by 13% from 0.79 to 0.92. In a non-screening population, this score demonstrates the highest accuracy.

The Brock model has the highest AUC but was based on a screening population.46 All participants were current or former smokers, hence smoking status was not included as a variable in the final predictive models. In addition, this model did not incorporate PET scan findings as an additional predictive variable. The prevalence of malignancy (5.5% in the PanCan cohort, and 3.7% in the BCCA cohort) was significantly lower than that reported by Swensen (23%), Gould (54%) and Herder (57%).

Figure 9 shows how the various models compare for a 70-year-old man with a spiculated upper lobe nodule according to nodule diameter. The models perform very differently across the whole range of diameters. The Brock model shows a much lower probability of malignancy for smaller nodules and is the only model with a large number of smaller nodules in the derivation population. Despite being developed in an exclusively smoking or ex-smoking population, the likelihood of malignancy using the Brock tool is consistently below the likelihood of the Mayo tool even when the latter was calculated for a non-smoking patient.

• Download figure
• Open in new tab
• Download powerpoint

Figure 9

Predicted probability of malignancy according to nodule size in a 70-year-old man (spiculate nodule in upper lobe). VA, Veterans Administration.

The validation study by Al-Ameri et al68 is the only study to validate the Brock and Herder models, and the only analysis of the performance of any models in a UK population. In patients undergoing FDG PET-CT, the Herder model was clearly the most accurate in predicting malignant risk, even when used in a cohort not restricted by the inclusion criteria of the model (ie, including patients with a previous history of lung cancer or an extra-thoracic cancer within the past 5 years). For smaller subcentimetre nodules, the highest accuracy was seen for the Brock score.

Patients with smaller pulmonary nodules

A consistent finding from the studies considered in the section ‘Risk prediction models’ is the strong effect of size on predicting malignancy in a nodule. However, as discussed above, the prediction tools vary considerably in their estimates of malignant risk for very small nodules. Thus a non-spiculated 4 mm upper lobe nodule in a 70-year-old smoking woman has a malignant probability of 0.3% according to the Brock model, 11.9% according to the Mayo model and 39.1% according to the VA model.

The largest body of data relating to small nodules comes from the CT screening studies. Both NLST and NELSON studies have published rates of malignancy by nodule size alone in screened populations.

In a report of the initial CT findings (prevalence screen) from 26 309 patients randomised to the CT screening arm of NLST, 3668 patients were found to have a nodule of between 4 and 6 mm in diameter, of which only 18 were subsequently confirmed as lung cancer (positive predictive value (PPV) 0.5%; 95% CI 0.3% to 0.7%).19 In a subsequent report, Aberle et al73 described the results of the two incidence screenings in NLST. Of 24 715 patients undergoing low-dose CT at the first incidence screening round, 3822 were found to have a nodule of 4–6 mm in diameter, of which 12 were subsequently confirmed as lung cancer (PPV 0.3%; 95% CI 0.2% to 0.5%). Of 24 102 patients undergoing low-dose CT at the second incidence screen, 2023 were found to have a nodule of 4–6 mm in diameter, of which 15 were subsequently confirmed as lung cancer (PPV 0.7%; 95% CI 0.4% to 1.1%).

Horeweg et al74 reported results from 7155 Dutch participants in the NELSON study who underwent CT in the first or second rounds. The risk of developing lung cancer over a 2-year period was quantified for ranges of nodule size (volume/diameter) and VDT. Lung cancer probabilities were calculated using both screen-detected lung cancers, and interval cancers identified through linkages to the Dutch Cancer Registry. Over two rounds of screening, 6394 nodules were detected on 14 024 scans. The 2-year lung cancer probability was 1.3% for all participants (95% CI 1.2% to 1.5%). Participants without any pulmonary nodule (54.4%) had a lung cancer probability of 0.4% (95% CI 0.3% to 0.6%).

Patients with a nodule with a volume of ≥100 mm³ had a significantly higher chance of being diagnosed with lung cancer than those patients without nodules. However, there was no difference in risk of lung cancer between patients with smaller nodules (<100 mm³) and patients with no nodules. Similar findings were shown by nodule diameter, with the smallest nodule size associated with a significantly increased risk compared with patients with no nodules being 5–6 mm (PPV 0.9%; 95% CI 0.5% to 1.6%, p=0.03). Patients with smaller nodules (<5 mm diameter) had no increased risk compared with patients with no nodules. By subdividing the population by nodule volume and diameter, and comparing risk with that in patients without nodules, Horeweg et al74 were able to define an appropriate size cut-off point for discharging small nodules without any follow-up. They concluded that nodules <5 mm in diameter or <100 mm³ volume do not require any CT surveillance, as they are not associated with a significantly increased risk of lung cancer. However, two other studies have reported variation in absolute volume measurement between volumetric software packages.75 ,76 Therefore, until there is better agreement confirmed between packages it might be safer to reduce the threshold to 80 mm³. Subjects entered into screening trials have a greater baseline risk of malignancy than the general population but these findings probably apply to lower risk populations as well since the nodules below the stated size and volume cut-off point conferred no extra risk of malignancy and may therefore confer no extra risk irrespective of baseline risk. Horeweg et al also found that the chance of developing lung cancer after two screening rounds was 2.4% for nodules between 100 and <300 mm³ in volume and 16.9% for nodules ≥300 mm³. The corresponding chance of lung cancer for a diameter of 5 to <8 mm was 1.0% and for ≥8 mm, 9.7%. Thus it might be argued that at least for nodules <300 mm³ or <8 mm diameter, where PET-CT is less valuable (see section ‘Further imaging in management of pulmonary nodules’), CT follow-up is indicated without further risk assessment.

The sample size contributing to these estimates is considerably larger than those used to produce the risk prediction tools described above. The higher risks assigned to these small nodules from the Mayo and VA models are likely to be erroneous in this context, and so the Brock model is preferred.

Summary

There have been several validated risk prediction models developed to assist in the management of pulmonary nodules. Earlier risk models have been improved considerably by the addition of PET findings while new models based on larger datasets and using more modern imaging have generated more reliable data to inform the recommendations on risk prediction and subsequent management (see also algorithm 1, initial assessment). The best evidence to guide recommendations comes from CT screening trials that selected subjects at relatively high risk of lung cancer.

Evidence statement

• Clinical predictors of lung cancer in patients presenting with pulmonary nodules include:

  • increasing age

  • history of smoking

  • number of pack years smoked. Evidence level 2+

• Radiological (CT) predictors of lung cancer in patients presenting with pulmonary nodules include:

  • increasing nodule diameter

  • spiculation

  • pleural indentation

  • upper lobe location. Evidence level 2+

• Nodules with diffuse, central, laminated or popcorn pattern of calcification or macroscopic fat can be considered benign. Evidence level 2+

• A homogeneous, smooth, solid nodule with a lentiform or triangular shape either within 1 cm of a fissure (perifissural) or the pleural surface (subpleural) can be considered benign. Evidence level 2+

• In the NLST and NELSON, the prevalence of lung cancer among patients with 4–6 mm nodules was 0.5% and in NELSON, malignancy risk was no different from the subjects without nodules where nodules measured <5 mm or <100 mm³, with better accuracy for volume measurements. Evidence level 2+

• There is variation between different volumetry software packages such that the threshold of 100 mm³ found in NELSON could be as low as 80 mm³ depending on the software. Evidence level 3

• In NELSON, the risk of lung cancer among nodules of 100 mm³ to <300 mm³ and 5 to <8 mm diameter was found to be 2.4% and 1.0%, respectively. Evidence level 2+

• Prediction models for pulmonary nodules based on clinical and radiological parameters have been externally validated. In the only validation study performed in a UK population, the Herder model (incorporating nodule FDG avidity) performed significantly better than other models (Mayo, Brock, Veterans Administration). In subcentimetre nodules, the Brock score had the highest accuracy (AUC value). Evidence level 2+

• The use of clinical prediction models is more accurate than clinicians’ individual clinical judgement in estimating the probability of malignancy in patients with pulmonary nodules. Evidence level 3

• In patients with known extrapulmonary cancer who have pulmonary nodules at presentation, there is limited evidence for the role of clinical and radiological factors in differentiating nodules that are primary lung cancer or metastases. Evidence level 3

• There is limited evidence outside the screening population for determining aetiology and management in patients with multiple pulmonary nodules. Evidence level 3 supported by 2+

• In a screening population the presence of multiple pulmonary nodules was found to indicate a lower risk of malignancy. Evidence level 2+

• In the NELSON screening trial, effective management of subjects with multiple nodules was achieved as determined by the management of the largest nodule. Evidence level 2+

Recommendations

• Do not offer follow-up or further investigation for people with nodules with diffuse, central, laminated or popcorn pattern of calcification or macroscopic fat. Grade C

• Do not offer nodule follow-up or further investigation for people with perifissural or subpleural nodules (homogeneous, smooth, solid nodules with a lentiform or triangular shape either within 1 cm of a fissure or the pleural surface and <10 mm). Grade C

• Consider follow-up of larger intrapulmonary lymph nodes, especially in the presence of a known extrapulmonary primary cancer. Grade D

• Do not offer nodule follow-up for people with nodules <5 mm in maximum diameter or <80 mm³ volume. Grade C

• Offer CT surveillance to people with nodules ≥5 mm to <8 mm maximum diameter or ≥80 mm³ to <300 mm³ volume. Grade C

• Use composite prediction models based on clinical and radiological factors to estimate the probability that a pulmonary nodule (≥8 mm or ≥300 mm³) is malignant. Grade C

• Use the Brock model (full, with spiculation) for initial risk assessment of pulmonary nodules (≥8 mm or ≥300 mm³) at presentation in people aged ≥50 or who are smokers or former smokers. Grade C

• Consider the Brock model (full, with spiculation) for initial risk assessment of pulmonary nodules (≥8 mm or ≥300 mm³) in all patients at presentation. Grade D

• Base the risk assessment of people with multiple pulmonary nodules on that of the largest nodule. Grade C

• Nodule malignancy risk prediction models should be validated in patients with known extrapulmonary cancer. RR

• Further analysis of variation in volumetry measurements by different software packages should be undertaken and methods developed for standardisation. RR

How should nodule change be assessed?

Pulmonary nodule size has traditionally been assessed by measuring the largest transverse cross-sectional diameter. The VDT of a nodule can then be estimated from the difference in nodule diameter between baseline and follow-up CT and the time interval between these two scans, using a simple exponential growth model that assumes uniform three-dimensional (3D) tumour growth. Within the past 15 years, volumetric analysis (calculated either manually or by a semiautomated/automated method) has been increasingly reported as an alternative tool with which to assess nodule growth.

In a retrospective case series, Revel et al77 assessed variability in 2D CT measurements of 54 pulmonary nodules in 24 patients both between readers and in the same reader's measurements at different times. Both intra- and inter-reader agreement for 2D measurements were found to be poor, with a change in size of <1.7 mm only having a 5% chance of corresponding to an actual change in nodule size. Korst et al78 compared automated 3D volumetric estimates of pulmonary nodule growth rate with those derived from 2D measurement of nodule diameter in a retrospective case series of 87 nodules in 69 patients seen in a routine review clinic. Although correlation overall between these measurements was good, greater divergence was seen between these two methods for irregular nodules, or where the time interval between scans was shorter (<100 days). Of the cases where volumetric analysis would have changed management and prompted a biopsy (6.2% of all cases), 43% of nodules had an eventual malignant diagnosis.

Ko et al79 compared semiautomated 3D volumetric analysis against standard calliper cross-sectional diameter measurement of 123 lung nodules in a retrospective analysis of 59 patients recruited through a CT lung cancer screening programme. Abnormal growth was detected in nodules subsequently proved to be malignant at a much shorter time interval (183±158 days) by 3D volumetry than by standard radiological diagnosis (344±284 days), suggesting greater sensitivity of the volumetric technique. Similarly, Revel et al80 compared automated 3D volumetric analysis against 2D calliper measurement of 63 solid lung nodules in a retrospective case series. The sensitivity for volumetric-calculated doubling time for malignancy (with a 2-month median interval for rescan) was 91% (95% CI 0.59 to 1.00) compared with a sensitivity of manual diameter-change measurement of 54% (95% CI 0.23 to 0.83).

In addition to growth in the size of a nodule, changes in other parameters have been evaluated. de Hoop et al81 retrospectively compared diameter, volume and mass measurements of 52 GGNs detected in a lung cancer screening trial. Of the three parameters, mass measurements showed the least intra- and interobserver variation. Furthermore, in a subgroup of 13 malignant GGNs subsequently resected, changes in mass were seen significantly earlier than changes in volume or diameter (mean time 425, 673 and 713 days, respectively, p=0.02).

Xu et al82 retrospectively analysed 372 indeterminate solid intraparenchymal nodules in 312 patients recruited to the NELSON lung cancer screening trial. Although baseline density did not differ between nodules with an eventual benign or malignant diagnosis, malignant nodules showed a statistically significant increase in density during CT follow-up compared with benign nodules (median change 12.8 Hounsfield units (HU) vs −0.1 HU, respectively, p<0.05). However, there was significant overlap in density changes between benign and malignant nodules, indicating that density change alone is unlikely to be sufficiently specific or sensitive to accurately identify malignant nodules.

What is the appropriate time interval between surveillance scans?

In seeking to discriminate between benign and malignant nodule growth patterns, CT surveillance aims to have a high sensitivity for detecting nodule growth consistent with malignancy at the earliest opportunity, while maintaining high specificity and minimising false-positive referrals (nodules deemed to have grown but which have a subsequent benign diagnosis). The optimal interval between surveillance scans will relate to the reliability of detecting percentage volume change taking into account artefact, and the doubling time threshold between growing and stable nodules.

In a retrospective analysis of patients recruited to a lung cancer screening programme, Kostis et al83 reviewed 115 pulmonary nodules deemed stable over 2 years’ observation. They assessed error in 3D volumetric assessment of nodule size due to artefact and other factors, to determine whether apparent growth might simply reflect measurement errors in stable nodules. They then derived the critical time to follow-up CT—that is, the earliest point at which growth in a nodule of a given size can be reliably identified with repeat CT. As expected, the percentage SD of nodule size estimate increased with decreasing nodule size. The critical time to follow-up CT was calculated as 12 months for nodules with initial diameter 2–5 mm, 5 months for nodules 5–8 mm and 3 months for nodules 8–10 mm.

Although detecting nodule growth at the earliest opportunity is preferable enabling prompt treatment to be offered, there is evidence that the accuracy of growth rate measurement and assessment of malignant risk improves with a greater time interval between surveillance scans. Thus Ko et al79 demonstrated a reduction in SD of growth rate estimate with increasing time between scans (SD of 47% at 6 months, 30% at 1 year and 20% at 2 years).

Xu et al84 retrospectively evaluated 891 indeterminate nodules detected during the NELSON CT screening trial. VDT was assessed at 3-month and 12-month interval scans, and nodules with a VDT of <400 days at either time point were referred for further investigation. Overall, 78 nodules were referred owing to a VDT of <400 days—68 nodules at 3 months and 10 at 12 months. The proportions of nodules with an eventual malignant diagnosis referred at 3 months and 12 months were 15% and 50%, respectively, indicating greater specificity for assessment of malignant risk at the later time point.

Zhao et al85 retrospectively reviewed characteristics of resolving pulmonary nodules detected at initial scan in a CT screening programme. Of a total of 964 indeterminate nodules initially detected, 10.1% (97) disappeared at subsequent screening. The majority of resolving nodules (75/97 (77%)) had disappeared by 3 months. Features predicting resolution were non-peripheral location, spiculation and larger nodule size (≥8 mm vs <8 mm). The last two factors are also predictors of malignancy, therefore limiting the extent to which baseline features can be used to predict which nodules will subsequently resolve.

What growth rates are reported for malignant pulmonary nodules?

A number of studies describing VDTs for malignant pulmonary nodules were reviewed. Considering only studies with 50 or more cases, five reports of doubling times for nodules subsequently confirmed as lung cancer were identified and details of their findings are shown in table 8. Two studies retrospectively reviewed growth rates of lung cancers detected in routine clinical practice86 ,87 and three studies reviewed cancers detected by CT screening.88–90 Of the studies reviewed, two used 2D diameter assessment of size only, two used manual volumetric analysis and one used automated/semiautomated volumetric analysis.

All studies showed a wide range of growth rates for lung cancers. VDTs were reported by histological subtype of tumour and the mean/median values are shown in table 9. Direct comparison of these studies is limited by differences in the methods of volume estimation (2D diameter measurement vs manual or automated 3D volumetry), histological definitions (two studies grouped adenocarcinomas and bronchoalveolar cell carcinoma (BAC)/AIS together) and presentation of data (parametric or non-parametric variables). Despite these limitations, consistent patterns were seen between histological subtypes with progressively longer VDTs quoted for small cell carcinoma, squamous cell carcinoma, adenocarcinoma, BAC/AIS, respectively.

Two studies88 ,89 compared VDT by radiological appearance of the nodule, and showed shorter VDTs for solid versus SSNs or pGGNs. In one of the largest series of malignant nodules, Henschke et al reported that all lung cancers detected as solid nodules in the International Early Lung Cancer Action Program (I-ELCAP) screening programme had VDTs of <400 days (n=99), whereas 3 of the 12 SSNs detected had VDTs of >400 days (413, 531, 884 days).89

The upper limit of VDT for malignant nodules was high in some series owing to the presence of very slow growing nodules that turned out to be lung cancers: 884, 1435 and 1733 days.88–90 Cancers with long VDTs tended to present as SSNs and were associated with BAC/AIS on eventual histology. In two studies, regression on nodules subsequently diagnosed as cancer was described (leading to negative VDTs).

What is an appropriate cut-off point for nodule growth rate to allow discrimination of benign and malignant nodules?

The presence of extremely long VDTs for some lung cancers, and the observation that a proportion of malignant nodules reduce in size on interval screening, indicates that there is no upper limit of VDT above which nodules can be guaranteed to be benign. Similarly, the observation that some malignant nodules show a long period of radiological stability before growing means that it is not possible to define a period of surveillance during which stability will completely exclude the possibility of malignancy. Current practice is guided by recommendations from the Fleishner Society which recommend follow-up for either 12 or 24 months depending on initial nodule size and patient risk.91 Stability over 2 years of follow-up has traditionally been regarded as indicative of benign disease, having first been proposed on the basis of CXR follow-up of nodules in the 1950s,92 although the evidence underlying this assumption has been questioned.93

Although some cancers grow very slowly, or grow after a prolonged period of stability, it is not practical to follow up every nodule indefinitely for fear of missing an occasional cancer. Studies reporting growth rate of lung cancers will obviously consider only nodules with an eventual malignant diagnosis and are therefore all retrospective in nature. These data do not facilitate a prospective assessment of risk in any given pulmonary nodule with a known growth rate. Instead, studies of populations of all nodules (both benign and malignant) by growth rate are of more use in providing clinicians and patients with accurate information on malignant risk on which to base decisions about management and follow-up.

Ashraf et al94 reported a series of 54 indeterminate pulmonary nodules identified through the Danish Lung Cancer Screening Trial. They classified nodules according to VDT derived by automated 3D volumetric analysis on repeat CT scanning at 3 months, and FDG avidity of nodules at PET-CT. Nodules were grouped into those with a VDT <1 year, and those with a VDT >1 year or regressing. Seventeen nodules had a VDT <1 year (31% of total group), of which 14 were malignant (82%). Six of the remaining 37 nodules which either regressed or had a VDT >1 year were malignant (16%). The VDTs of these slow-growing malignant nodules were not reported. The VDT estimates were made on the basis of a 3-month interval scan. As discussed above, there appears to be greater error in VDT estimates made at 3 months compared with 12 months, and it is unclear to what extent this affected the sensitivity of 3-month VDT assessment for detecting malignant nodules.79 ,84 ,95

By far the largest series of pulmonary nodules followed up by volumetric analysis comes from the NELSON study. The trial used VDT (calculated by automated volumetric analysis after a 3-month or 12-month interval) to guide management of indeterminate pulmonary nodules (50–500 mm³) so that patients with nodules with a VDT of <400 days were referred to a chest physician for investigation and diagnosis, whereas those with nodules with a VDT >400 days were considered benign and re-entered the screening programme.96 At least a 25% change in volume was required to indicate a significant change.97 It should be noted that where the automated software was unable to calculate volume, VDT was measured by manually measuring maximum diameter in three perpendicular planes. Thus any conclusions about follow-up periods relying on diameter measurements can only apply when VDT is calculated using this method.

Horeweg et al74 reported the follow-up of 2500 nodules where VDT was calculated. As discussed in the section ‘Patients with smaller pulmonary nodules’, the 2-year lung cancer probability was 1.3% for all participants (95% CI 1.2% to 1.5%) in the screening programme, whereas participants without any pulmonary nodules (54.4%) had a 2-year lung cancer probability of 0.4% (95% CI 0.3% to 0.6%). Participants with slowly growing nodules (VDT >600 days), stable, shrunken or resolved nodules had a low probability of lung cancer (0.0–1.0%). Lung cancer probability was not significantly increased for participants with nodule VDTs of ≥600 days (0.8%; 95% CI 0.4% to 1.7%) compared with participants without nodules (p=0.06). Lung cancer probability was significantly increased for participants with nodule VDTs of 400–600 days (4.0%; 95% CI 1.8% to 8.3%, p<0.0001) and was even higher in participants with nodule VDTs of ≤400 days (9.9%; 95% CI 6.9% to 14.1%, p<0.0001). Analysis was not presented by nodule morphology (solid vs sub-solid, etc).

Although there is a trend to increased cancer risk in patients with slowly growing nodules (with VDT >600 days), the risk of malignancy in this situation is very small (0.8%). In considering more aggressive management in this situation, this risk must be considered alongside the operative mortality of thoracoscopic wedge resection (0.4% inpatient mortality according to the Society for Cardiothoracic Surgery in Great Britain and Ireland, with possible higher 90-day mortality) (www.bluebook.scts.org).

The NELSON trial is the only screening study to date to have prospectively assessed growth using automated volumetry in a defined protocol. Thus despite this being a single publication, it is unlikely that any future studies will be able to provide this level of prospective data for such a large number of patients with pulmonary nodules.

There is little evidence for the management of new nodules that appear in follow-up CTs. Here, the risk of malignancy will depend on the growth rate and it should be noted that rapid growth may imply an inflammatory process rather than malignancy.

Summary

A repeat CT at 3 months will reliably detect growth in larger nodules, and will also demonstrate resolution in the majority of resolving nodules. Automated or semiautomated volumetry is more accurate than diameter measurements and accuracy of VDT assessment is better after 1 year than 3 months, especially for small nodules (<6 mm). Some lung cancers have very long VDTs, show prolonged periods of stability or even reduce in size on interval screening so that there is no upper limit of VDT above which nodules can be guaranteed to be benign. The approach to this problem, as recently suggested in a NELSON publication, may be to compare the risk of malignancy with that of the baseline risk of malignancy to define a point where follow-up is no longer indicated, given the absence of national screening programmes. A consistent finding in the studies quoted above is the slow rate of growth for SSNs, and therefore recommendations for duration of follow-up are distinct for this subgroup and are considered in the next section of the guideline. The evidence quoted for assessment of VDT and duration of follow-up relates to assessment of the risk of lung cancer. There is no published evidence informing assessment of the likelihood of lung metastasis from extrapulmonary malignancy according to VDT, and no evidence to guide appropriate duration of surveillance follow-up to exclude malignancy in this setting.

Evidence statement

• Repeat CT scans to assess interval growth have greater sensitivity and specificity for detecting malignancy at 1 year than scans at earlier time points. Evidence level 2+

• The majority of pulmonary nodules that eventually resolve have done so after a 3-month interval. Evidence level 3

• Accuracy of growth detection at 3 months reduces with smaller nodule size. Evidence level 3

• The growth rate of malignant nodules differs by histological subtypes and CT morphology. Small cell and squamous cell carcinomas tend to have shorter VDTs than adenocarcinoma. Evidence level 3

• Malignant nodules show wide ranges of growth rates, with some demonstrating regression at times. There is therefore no growth rate threshold beneath which, nor duration of radiological stability beyond which, malignancy is definitely excluded. Evidence level 3

• In the NELSON screening study, patients with nodules with a VDT <400 days and 400–600 days measured after a 3- or 12-month interval, had 2-year cancer probabilities of 9.7% and 4.1%, respectively, significantly greater than the cancer risk of subjects without nodules (0.4%) and the screened population as a whole (1.3%). Evidence level 2+

• The same study showed that the 2-year risk of lung cancer was 0.8% when the VDT was >600 days, not significantly higher than for subjects without nodules. Evidence level 2+

• In NELSON, where diameter measurements were used to calculate VDT, maximum diameter was measured in three planes. Evidence level 2+

• At least a 25% change in volume is required before the change can be regarded as significant. Evidence level 2+

• Duration of follow-up to ensure stability of nodules is not known for 2D diameter measurements. Evidence level 3

Recommendations

• Where initial risk stratification assigns a nodule a chance of malignancy of <10%, assess growth rate using interval CT with capability for automated volumetric analysis. Grade C

• Assess growth for nodules ≥80 mm³ or ≥6 mm maximum diameter by calculating VDT by repeat CT at 3 months and 1 year. Grade C

• Use a ≥25% volume change to define significant growth. Grade C

• Assess growth for nodules of ≥5 mm to <6 mm maximum diameter by calculating VDT by repeat CT at 1 year. Grade C

• Offer further diagnostic investigation (biopsy, imaging or resection) for patients with nodules showing clear growth or a VDT of <400 days (assessed after 3 months, and 1 year). Grade C

• Discharge patients with solid nodules that show stability (<25% change in volume) on CT after 1 year. Grade C

• If 2D diameter measurements are used to assess growth, follow-up with CT for a total of 2 years. Grade D

• Consider ongoing yearly surveillance or biopsy for people with nodules that have a VDT of 400–600 days, according to patient preference. Grade C

• Consider discharge or ongoing CT surveillance for people who have nodules with a VDTs of >600 days, taking into account patient preference and clinical factors such as fitness and age. Grade C

• Where nodules are detected in the context of an extrapulmonary primary cancer, consider the growth rate in the context of the primary and any treatment thereof. Grade D

Prevalence of SSNs

The prevalence of SSNs is difficult to extract from most studies as it is not directly reported. In the PanCan dataset 1871 of 2537 subjects had nodules detected over the screening rounds, and 15.9% of nodules were pGGNs and 4.3% PSNs. In the BCCA dataset the proportions were 9.3% and 0.9%, respectively.46 If these proportions are applied to the prevalence of nodules in the much larger NLST2 (over 27 000 subjects), where the average proportion of subjects with nodules ≥4 mm was 24%, the prevalence of pGGNs detected will lie in the range 2.2–3.8% and PSNs would be found in 0.2–1% of CTs. This broadly agrees with the original report from ELCAP102 (1000 subjects), where it was found that 2.8% of baseline CTs detected pGGNs and 1.6% detected PSNs. In a review of 60 000 CTs Matsuguma et al found only 98 pGGNs (0.16%) and 76 (0.13%) PSNs103; this may reflect the different population including a higher proportion of non-smokers.

Histopathological correlates of SSNs

Only six studies were identified where consecutive cases were resected and most of these reported histology according to the previous classification of adenocarcinoma. Two more recent studies that reported on SSNs ≤20 mm diameter are shown in table 10.103 ,104 One study looked at a CT-detected series and the other resected all lesions. Amongst the resected lesions the spectrum of pathology was similar, although the overall rates of invasive carcinoma were much lower in the CT-detected series, probably reflecting selection bias in the resected series.

Proportion malignant

The best evidence for the proportion of SSNs malignant, when detected by CT, comes from the screening trials, although again most report the previous classification of adenocarcinoma. In the PanCan dataset,46 1.9% (21/1105) of pGGNs and 6.6% (20/303) of PSNs were malignant and in the BCCA the numbers were lower but the rates were 1.3% (6/467) and 22.2% (10/45), respectively.

Predictors of malignancy and growth pattern

Twenty-four studies with 50 or more cases looked at predictors of malignancy.46 ,81 ,103–124 Initial size of lesion and growth are predictive of malignancy. A previous history of lung cancer was also found to be an independent predictor in three studies.103 ,106 ,116 McWilliams et al46 were able to show that pGGNs, although more often malignant than solid nodules, actually conferred a lower chance of being malignant when adjusted for other factors in the risk prediction model. This was not the case for PSNs, where the part solid nature was an independent predictor of malignancy. Despite this, even PSNs ≤5 mm in maximum diameter, with other adverse factors added, had no more than a 1% risk of being malignant over 2–4 years for people aged <78. Two further studies noted that older age was associated with an increased risk of malignancy.113 ,114 Three studies showed that a peripheral eosinophilia was predictive of benignity.112 ,114 ,125 Morphological features predictive of malignancy, other than initial size, were pleural retraction or indentation and a bubble-like appearance in a pGGN.

Some studies evaluated the outcome of nodules that were persistent. Lee et al analysed the long-term progression of 175 SSNs that persisted for more than 2 years in 114 patients.113 The mean initial size was 7.8±4.4 mm and median follow-up duration was 45 months. Forty-six (26.3%) SSNs showed significant size increases (≥2 mm in the longest diameter) with a mean VDT of 1041 days. In a multivariate analysis, large size (≥10 mm), PSNs and old age (≥65 years) were risk factors for significant size increase, with ORs (95% CI) of 6.46 (2.69 to 15.6), 2.69 (1.11 to 6.95) and 2.55 (1.13 to 5.77), respectively. SSNs with character changes from pure to mixed or mixed to solid showed more rapid volume expansion. The authors concluded that SSNs which persisted for several years showed an indolent course but noted that larger lesions with a solid portion in male or elderly individuals may be cause for more concern. In a retrospective study of 93 individuals with 126 PSNs identified from 16 777 individuals who underwent chest CT, 69.8% of PSNs were transient. Multivariate analysis showed that young patient age, detection of the lesion at follow-up, blood eosinophilia, lesion multiplicity, ill-defined border and large solid portion (the latter OR was only 1.05) were significant independent predictors of transient PSNs.112 A further study followed up 120 SSNs (91 pGGNs) for a median of 4.2 years that had been observed without treatment for a minimum of 6 months.111 Of these SSNs 28% grew by ≥2 mm and all of these had grown by 3 years. Independent predictors of growth were smoking history (OR=6.51) and initial size >10 mm (OR=4.06).

Two studies have developed logistic regression models specifically for PSNs. One found an ROC of 0.93 for distinguishing transient from persistent SSNs; however, the model was strongly influenced by eosinophilia and lesion multiplicity.125 The other developed a model to predict invasive versus non-invasive adenocarcinoma confirmed by resection. This showed an ROC of 0.9 but was strongly influenced by whether the lesion had a spiculated margin or not (OR=26.8) or a lobulated border (OR=2.9).122

In the review by Matsuguma et al103 of more than 60 000 CTs it is stated that the usual policy of the institution was to offer patients resection for all sub-solid lesions >15 mm diameter and lesions showing >2 mm growth or development of a ≥2 mm solid area in a pGGN. Some patients did not have resection in view of comorbidities. The authors found that 18 (10%) of 174 nodules reduced in size and 41 (24%) showed growth. All except one SSN that showed growth was malignant. Estimates were made of cumulative percentages of growing nodules at 2 and 5 years (13% and 23% in pGGNs and 38% and 55% in PSNs). The multivariate analysis showed that size >10 mm and history of lung cancer were independent predictors of malignancy in SSNs. The authors reported that none of the pGGNs were invasive adenocarcinoma, with only 4% MIA.

Hiramatsu et al, again in the setting of a large Japanese thoracic surgical centre, studied 184 patients referred with SSNs.106 Of these, 17 underwent immediate investigation and 10 were lost to follow-up. Of the remaining 157, at the initial 3-month follow-up CT, six nodules had resolved but six had already shown obvious growth, while another three showed metastases and four progression of another malignancy. Fifty SSNs that were ≤10 mm in patients with no history of lung cancer did not grow at 3.5 years.

Ichinose et al,104 again from Japan, looked at 191 resected lesions in 160 patients. They found that the proportion of lesions that were malignant was higher, probably owing to the different selection criteria. They noted that eight of 14 pGGNs with an standardised uptake value (SUV) max of 0.8 were malignant compared with four of 52 that were below this threshold.

In a third study from Japan, Takahashi et al115 followed up 150 pGGNs in 111 patients. Patients had high-resolution CT (HRCT) at various intervals but scans were evaluated at first detection, 2 years later and at final follow-up. After a mean of 66 months 19 (12.7%) nodules increased in size by ≥2 mm. Six of the 19 nodules that increased were stable at 2 years. The time to growth was 39.9 months in these nodules and 16.9 months in those 13 nodules deemed to have grown at 2 years. The authors concluded that more than 2 years’ follow-up was necessary to detect growth. The final diagnosis of the growing nodules was determined in seven of 19 (and one further nodule that was stable at 2 years and decreased at follow-up). These were BAC in four and mixed subtype adenocarcinoma in three. All resected patients were alive after a mean of 32 months.

Lee et al113 followed up 175 SSNs that had been stable on HRCT for 2 years and found that after a median follow-up period of 45 months, 26% had enlarged by ≥2 mm. Large size, part-solid type and increasing age were predictors of growth. Kobayashi et al111 followed up 120 SSNs and found that after a median of 4.2 years, 28% had enlarged by ≥2 mm. Smoking and large size were the most important predictors of growth. Chang et al118 found that 12 (9.8%) of 122 pGGN lesions that were followed up for more than 2 years grew and that the 11 that were surgically biopsied were lung cancer. Larger size and the development of a solid component were predictors of growth. The mean VDT of growing pGGNs was 769 days. Most studies used linear measurements to assess growth, but another- study showed that change in mass was a more reliable measure.81

Nakao et al126 performed a prospective trial of limited resection of 50 SSNs ≤2 cm with no pleural indentation or vascular convergence; 40 were adenocarcinoma. There were no recurrences before 5 years but four after 10 years. The authors concluded that long-term follow-up was indicated after limited resection and that the latter should only be done in a trial setting. Silva et al124 reported the experience from the Italian Multicentric Italian Lung Detection (MILD) trial—lung cancer randomised controlled CT screening trial. Seventy-six SSNs were found in 56 participants. Of 48 pGGNs, 31% resolved, 8.3% reduced in size, 43.8% remained stable and 16.7% progressed over an average follow-up period of 50 months. Of these pGGNs, 81% were <10 mm. For PSNs, with a solid component <5 mm, 11.5% resolved, 42.3% remained stable and 46.2% progressed. Lung cancer was found in one pGGN and three PSNs.

Prognosis

The reported prognosis of SSNs is very good, irrespective of the selection criteria. Case series of more than 100 SSNs, whether all resected or managed by follow-up and resection of selected nodules, report few deaths due to cancer109 ,110 ,115 One study found that even when SSNs with suspicious cytology are observed, the outcome is still good.127 These observations have led some authors to suggest that less aggressive treatment is more appropriate. Patz et al128 reported overdiagnosis in the NLST to be a maximum of 18.5% overall, but found this rose to 79% (62%–94%) when BAC was detected by CT screening. Many of these tumours, in the new international classification, would have been designated MIA. The CT correlate would be mostly SSNs. Careful evaluation to identify more indolent tumours has been advocated as a way to reduce overdiagnosis and the associated potential harms by the use of imaging follow-up and minimally invasive surgery.128

Lymph node metastases

Maeyashiki et al129 looked at 398 consecutive clinical stage 1A lung cancers undergoing resection with 263 SSNs. They found that the size of the consolidation (solid component) and the presence of an air bronchogram were independent predictors of lymph node metastases and that 16% of PSNs had nodal metastases. Node metastases occurred in 9.8% of lesions <20 mm diameter and in 22.1% of lesions ≥20 mm (including solid lesions). None of the pGGNs (n=30) or PSNs (number not given) with solid component ≤10 mm had nodal metastases. Ichinose et al104 reported that one of 114 pGGNs showed lymph node involvement. A further study of 57 SSNs showed that the proportion of the solid component was predictive of nodal metastases and that there were no metastases in the 15 SSNs where the solid component was ≤25% of the total diameter.98

Multiple SSNs

SSNs are frequently multiple. One study of 193 SSNs compared the features of single versus multiple nodules.109 Multiple SSNs were more frequently AAH or BAC, and occurred more often in women and non-smokers. However, the authors did not think the differences were enough to recommend a different approach to management. Another study looked at multiple pGGNs in 73 patients undergoing resection for BAC and found that all but one remained stable over a 40-month median follow-up.108 In a study of PSNs, nodules were more likely to be benign if multiple.112

SSNs and staging

Two recent studies have shown that measurement of the solid component of malignant PSNs is a better predictor of prognosis than total diameter,123 ,129 one study suggesting a change to the T descriptor of the Tumour Node Metastasis (TNM)—staging system for lung cancer (table 11).123

Summary

SSNs merit separate consideration from solid nodules because they represent lesions that confer a better prognosis but paradoxically are more likely to be malignant than their pure solid counterparts, with part-solid nature being an independent predictor of malignancy. There are now well-established baseline predictors of malignancy, and growth or the development of a solid component in pGGNs are strong predictors. Many SSNs have slow growth rates and may remain stable for years. The management of these nodules is therefore uncertain as the prognosis may be good even when confirmed adenocarcinomas are seen. This may suggest that a less aggressive approach is indicated for these lesions.

Evidence statement

• The prevalence of pGGNs and PSNs in high-risk screening cohorts is 2–4% and 0.2–1%, respectively. This may be less in some populations that have more non-smokers. Evidence level 2++

• The pathological correlates of SSNs are AAH, AIS, MIA and invasive adenocarcinoma. pGGNs are more often AAH, AIS and MIA and PSNs more often invasive adenocarcinoma. Evidence level 3

• The majority of studies have assessed SSNs by high-resolution (thin-section) CT. Evidence level 3

• Baseline factors consistently associated with malignancy in SSNs are older age, previous history of lung cancer, size of nodule and part-solid nature. Evidence level 2++

• Other baseline factors that may be predictive of malignancy are size of the solid component in PSNs, pleural indentation and bubble-like appearance. Evidence level 3

• SSNs are more likely to be malignant than solid nodules; however, only PSNs are independent predictors of malignancy. Evidence level 2++

• SSNs may resolve after initial follow-up at 3 months. Factors predictive of resolution of PSNs are younger age, peripheral eosinophilia, lesion multiplicity and an ill-defined border. Evidence level 3

• About a quarter of SSNs will show growth; PSNs grow more often than pGGNs. Around a quarter of SSNs may grow after being stable for ≥2 years. Evidence level 3

• Growth of SSNs is strongly predictive of malignancy; defined as ≥2 mm in maximum diameter. Larger size, current smoking and part solid nature are predictors of growth. Change in mass of SSNs, measured on CT, may be an early indicator of growth. Evidence level 3

• The appearance of a new solid component in a pGGN or enlargement of a solid component (≥2 mm in maximum diameter) is predictive of malignancy. Evidence level 3

• The prognosis of resected SSNs is excellent (95–100% 5-year survival) and may remain good even when resection is delayed following imaging follow-up. Evidence level 3

• PET-CT may have a role in the management of SSNs using lower SUV thresholds. Evidence level 3

• The rate of lymph node metastases in SSNs is related to the size of the solid component; the rate is <1% for pGGNs and where the solid component is <10 mm. Evidence level 3

Recommendations

• Do not follow-up SSNs that are <5 mm in maximum dimeter at baseline. Grade C

• Reassess all SSNs with a repeat thin-section CT at 3 months. Grade D

• Use the Brock risk prediction tool to calculate risk of malignancy in SSNs ≥5 mm that are unchanged at 3 months. Grade C

• Consider using other factors to further refine the estimate of risk of malignancy, including smoking status, peripheral eosinophilia, history of lung cancer, size of solid component, bubble-like appearance and pleural indentation. Grade D

• Offer repeat low-dose, thin-section CT at 1, 2 and 4 years from baseline where the risk of malignancy is approximately <10%. Grade D

• Discuss the options of observation with repeat CT, CT-guided biopsy, or resection/non-surgical treatment with the patient where the risk of malignancy is approximately >10%; consider factors such as age, comorbidities and risk of surgery. Grade D

• Consider using changes in mass of SSNs to accurately assess growth. Grade D

• Consider resection/non-surgical treatment or observation for pGGNs that enlarge ≥2 mm in maximum diameter; if observed, repeat CT after a maximum of 6 months. Take into account patient choice, age, comorbidities and risk of surgery. Grade D

• Favour resection/non-surgical treatment over observation for PSNs that show enlargement of the solid component, or for pGGNs that develop a solid component. Take into account patient choice, age, comorbidities and risk of surgery. Grade D

• Favour resection/non-surgical treatment over observation where malignancy is pathologically proven. Take into account patient choice, age, comorbidities and risk of surgery. Grade D

PET and PET-CT

PET-CT is a cross-sectional imaging technique that provides both anatomical and functional information. It has become firmly established in the management pathways of several malignancies, including lung cancer.130 FDG is the preferred radiopharmaceutical agent for oncological PET-CT. It is a glucose analogue that is injected and taken up and trapped within metabolically active cells; tumour cells have differentially increased glucose use and display increased tracer uptake. However, false-positive uptake is seen in both infective and inflammatory conditions, such as TB and sarcoidosis, whereas false-negative observations are associated with certain types of malignancy, including adenocarcinomas with a significant bronchoalveolar or mucinous component and well-differentiated carcinoid tumours.

A large proportion of the literature focuses on FDG PET alone before the introduction of integrated PET-CT scanners, which are now widely available throughout the UK. The CT component of the examination improves anatomical localisation and can provide additional growth/morphological information that may strengthen a diagnosis of lung malignancy or raise the possibility of alternative benign diagnoses. Nevertheless, the GDG considered the FDG PET only literature still relevant to current practice with FDG PET-CT.

Single-photon emission CT

SPECT imaging using ^99mTc-labelled radiopharmaceutical agents is similar in principle to PET imaging, where the distribution of injected radiopharmaceutical agent and emission of gamma photons is used to create representative cross-sectional images. This permits more accurate localisation of tracer uptake in comparison with planar imaging, but the spatial resolution of SPECT remains lower than PET. Studies have assessed the utility of ^99mTc-depreotide SPECT, a somatostatin analogue, as an alternative to FDG PET for the evaluation of SPNs given that malignant nodules have a greater expression of somatostatin receptors than benign nodules. Cronin et al,132 in a pooled analysis of seven studies (439 nodules), reported a sensitivity, specificity and area under the ROC curve of 95%, 82% and 0.94, respectively, which was not significantly different from the results obtained with FDG PET. Naalsund and Maublant150 in a multicentre prospective study analysing 118 nodules, reported slightly inferior results with a sensitivity, specificity and diagnostic accuracy of 89%, 67% and 81%, respectively. However, since October 2010, ^99mTc-depreotide is no longer commercially available in Europe.

MRI

MRI is a non-ionising cross-sectional imaging technique which is reliant upon the variable excitation and relaxation of hydrogen atoms— that is, protons, in response to a radiofrequency pulse, while in a static magnetic field. Imaging the lungs is problematic owing to inherent low proton density, resulting in poor image contrast, numerous air–soft tissue interfaces which result in signal loss and distortion and cardiac and respiratory motion, causing image blur. However, technological advances incorporating faster image sequences and functional imaging sequences including DW-MRI and DCE-MRI have changed this.

Studies assessing the accuracy of ultrafast MRI techniques using a HASTE sequence for the detection of pulmonary nodules, in comparison with the ‘gold standard’ of CT, have reported reliable detection of pulmonary nodules >5 mm.151 ,152 However, no studies have assessed the ability of HASTE MRI to differentiate between benign and malignant SPNs.

Dynamic contrast-enhanced CT

DCE-CT provides similar information to that obtained with DCE-MRI using iodinated contrast material, with the overall contrast enhancement of malignant nodules usually higher than that of benign nodules. Cronin et al132 in a pooled analysis of 10 studies (1167 nodules) reported a sensitivity, specificity and area under the ROC curve of 93%, 76% and 0.93, respectively, which was comparable with the diagnostic performance of FDG PET. Smaller prospective studies have reported similar results but with lower specificities and differing cut-off values, probably attributable to varying image acquisition parameters.159 ,160 Single-centre prospective studies have also looked at the first-pass perfusion DCE-CT assessing tissue haemodynamics based on perfusion parameters to differentiate between benign and malignant nodules. Sitartchouk et al161 in a single-centre study of 57 nodules, first reported the utility of perfusion parameters in differentiating between benign and malignant SPNs. Li et al162 in a similar sized study of 77 nodules confirmed this potential with sensitivities, specificities and accuracies to diagnose malignancy of 91.3–93.5%, 81.8–90.9% and 88.2–92.6%, respectively. More recently, Ohno et al in two prospective comparative studies between perfusion DCE-CT and FDG PET-CT have suggested that perfusion CT is more specific and accurate than FDG PET-CT.163 ,164 However, in both studies an SUV_max cut-off point was used to determine malignancy on FDG-PET-CT with nodule characteristics on the CT component of the examination not obviously used to make this decision. Further comparative studies with FDG PET-CT using optimal image interpretation are required before considering DCE-CT as a viable alternative for diagnosing malignancy in SPNs.

Risk thresholds for further investigation of risk of malignancy

Louie et al165 developed a Markov model to determine appropriate thresholds for deciding between management strategies in patients unfit for surgical resection. They modeled the pre-test probability of malignancy below which CT surveillance was appropriate and above which PET-CT should be performed, and proposed 17% as an appropriate cut-off between these two strategies. The GDG assessed the effect of increasing the threshold for PET-CT from 10% (as proposed previously) to 17% in a cohort of British patients with incidentally detected pulmonary nodules (unpublished data) and found that a malignant diagnosis would have been delayed for 23% of patients affected by this change. Thus a 10% threshold for proceeding to PET-CT was preferred. Louie et al also modeled the threshold above which it was appropriate to proceed to treatment without biopsy confirmation of malignancy, suggesting 85% probability of malignancy as an appropriate cut-off. Again, when these thresholds were used in a UK population, use of a 70% cut-off for treatment resulted in only a small increase in treatment of benign disease and reduced the chance of treatment delay, so the GDG considered that a lower threshold was appropriate (see further discussion in ‘Non-surgical treatment without pathological confirmation section’.) Thus figure 1 specifies a range of 10–70% where biopsy is preferred.

Summary

PET-CT remains the preferred investigation in the further evaluation of pulmonary nodules, partly because it is widely available and no alternative investigation shows superiority. Further research is required to evaluate alternative techniques, such as DCE-CT, which may be more cost-effective. PET-CT is less useful in smaller nodules but these have a lower risk of malignancy and can be managed by further imaging follow-up. In the assessment of risk, the Mayo model is substantially improved by the addition of PET-CT as described by Herder et al and when used with an ordinal scale to categorise FDG uptake.

Evidence statement

• Pre-test probability of malignancy influences interpretation of PET-CT, with high-risk individuals at risk of false-negative results, and low-risk individuals at risk of false-positive results. Evidence level 3

• In a meta-analysis PET has shown a sensitivity of 93.9% and specificity of 88.5% for determining malignancy from a pooled cohort of studies including patients with low to high risk. Evidence level 2++

• PET has a good sensitivity and moderate specificity for determining a malignant nodule in patients with a high risk of malignancy with a pulmonary nodule of uncertain aetiology of ≥10 mm, with more limited evidence for nodules <10 mm. Further imaging to assess growth increases the sensitivity of determining malignancy. Evidence level 1− supported by 2++

• PET has a lower sensitivity and higher false-negative rate in SSNs. Evidence level 2++ and 3

• Methods of assessing FDG uptake include qualitative visualisation, semiquantitative analysis and the measurement of SUVs; all have similar accuracy. The Herder model employed an ordinal scale. Evidence level 3

• Risk prediction models are improved by the addition of information from PET-CT; the Herder model AUC improved from 0.79 to 0.92. Evidence level 3

• MRI does not have a routine place in assessing pulmonary nodules outside of research studies. Evidence level 2++ and 3

• SPECT does not show any advantage over PET-CT in the assessment of pulmonary nodules. Evidence level 2++ and 3

• DCE-CT has a high sensitivity but low specificity for determining malignancy. Evidence level 2++ and 3

Recommendations

• Offer a PET-CT to patients with a pulmonary nodule with an initial risk of malignancy of >10% (Brock model) where the nodule size is greater than the local PET-CT detection threshold. Grade B

• Ensure that PET-CT reports include the method of analysis. Grade D

• Use a qualitative assessment with an ordinal scale to define FDG uptake as absent, faint, moderate or high using the following guide:

  • Absent—Uptake indiscernible from background lung tissue

  • Faint—Uptake less than or equal to mediastinal blood pool

  • Moderate—Uptake greater than mediastinal blood pool

  • Intense—Uptake markedly greater than mediastinal blood pool. Grade D

• Reassess risk after PET-CT using the Herder prediction tool. Grade B

• After reassessment of risk:

  • Consider CT surveillance for people who have nodules with a chance of malignancy <10%

  • Consider image-guided biopsy where the risk of malignancy is assessed to be between 10 and 70%; other options are excision biopsy or CT surveillance guided by individual risk and patient preference

  • Offer people surgical resection as the favoured option where the risk that the nodule is malignant is >70%; consider non-surgical treatment for people who are not fit for surgery. Grade C

• Do not use MRI, SPECT or DCE-CT to determine whether a nodule is malignant where PET-CT is an available alternative. Grade D

• Further research is needed into the most effective follow-up pathway in low-to-medium risk patients and for those with pGGNs. RR

• Further research should be undertaken into the use of PET-CT in the evaluation of pGGNs using lower SUV cut-off values. RR

Biomarkers

Five studies were identified assessing the role of biomarkers in determining the likelihood of a pulmonary nodule being malignant. Four of these studies measured circulating markers, and one involved analysis of bronchoalveolar lavage fluid.

Chu et al166 described a cross-sectional study that looked at combining tumour markers (squamous cell carcinoma antigen, CEA, cytokeratin 19 fragment antigen and neuron-specific enolase) in 805 patients with suspicious pulmonary masses, of whom 444 were found to have stage I lung cancer. The test performed poorly, with a demonstrated sensitivity of only 23.2%. Shen et al167 reported a cross-sectional study looking at the role of plasma microRNAs in 156 patients with SPNs found that plasma microRNA could distinguish between malignant and benign nodules with 75% sensitivity and 85% specificity (ROC AUC 0.86). Daly et al168 looked at a panel of seven biomarkers (interleukin (IL) 6, IL-10, IL-1ra, sIL-2Rα, stromal cell-derived factor-1 α+β, tumour necrosis factor α and macrophage inflammatory protein 1α) in a validation cohort of 81 patients (61 with benign nodules, 20 with malignant nodules). They found that the sensitivity and specificity for diagnosing malignancy were 95% and 23.3%, respectively, with a negative predictive value of 93.8%.

Higgins et al assessed the performance of a variant form of Ciz1 (a nuclear matrix protein) as a circulating biomarker for stage 1 lung cancer. In a validation cohort of 160 individuals comprising patients with stage I lung cancer, benign pulmonary nodules, inflammatory lung disease and age-matched smoking controls, the test performed well with 95% sensitivity and 74% specificity. Currently, however, measurement requires Western blot analysis which is not easily applied outside a research laboratory, and the authors acknowledge that the assay would need to switch platforms and be validated further before widespread use.169

The value of lactate dehydrogenase (LDH) in bronchoalveolar lavage fluid for discriminating between benign and malignant pulmonary nodules was assessed in a prospective case–control trial with 21 controls, 17 patients with benign SPNs and 42 with malignant lesions.170 LDH levels were significantly higher in those with malignant SPNs than in those with benign lesions or controls, although this study excluded those with a smoking history.

Three studies evaluated the performance of a panel of circulating autoantibodies in predicting patients with lung cancer. This test is commercially available (EarlyCDT-Lung) and is marketed as a tool to risk stratify patients with pulmonary nodules. Boyle et al171 described use of a panel of six circulating autoantibodies in three cohorts of patients with newly diagnosed lung cancer (n=145, 241, 269), each matched to control individuals (although little clinical information was supplied about the control groups). Sensitivities of 36–39% and specificities of 89–91% were reported for the three cohorts. Lam et al reported a similar study with the same panel of autoantibodies measured in four cohorts of patients with newly diagnosed lung cancer from Europe and North America.172 Results were compared with control populations for three of the four cohorts, with control populations matched by age, sex and (in two cohorts) smoking history. Overall, results were similar to those of the study by Boyle et al,171 with sensitivities of 34–43% and specificities of 87–89% for lung cancer.

A subsequent report described the addition of a seventh autoantibody to the panel and assessed performance in routine clinical practice in 1613 patients for whom US physicians ordered the EarlyCDT-Lung test.173 Sensitivity and specificity for lung cancer were 37% and 91%, respectively, with a positive test increasing the chance of lung cancer diagnosis by a factor of 5.4. No data are presented for baseline clinical or demographic data for the whole population. Of all patients tested (irrespective of test results), 3.8% were diagnosed with lung cancer within 6 months, which is considerably in excess of the rates of lung cancer seen in CT screening studies. From the information provided in the study, it is impossible to comment on the reasons for such a high rate of lung cancer diagnosis. To date, there are no published studies reporting the performance of the Early CDT-Lung test in prospectively recruited populations at high risk of developing lung cancer according to predefined criteria. Trials are ongoing, including a study in the UK using the test as a pre-CT screening tool. Of relevance to this guideline, there are no reports evaluating the performance of this test in a cohort of patients with pulmonary nodules, and thus its efficacy in discriminating malignant from benign nodules is unknown.

Flexible bronchoscopy

van't Westeinde et al174 evaluated the use of conventional bronchoscopy in the investigation of 308 consecutive patients with a positive CT screen enrolled in the NELSON trial. There were 318 suspicious lesions, with mean diameter of 14.6 mm (only 2.8% were >3 cm). The sensitivity of bronchoscopy was only 13.5% (95% CI 9.0% to 19.6%) with a negative predictive value of 47.6%. The authors do not recommend routine use of bronchoscopy for positive test results in a screening programme.

Image-guided biopsy

Interpretation of CT-guided biopsy results

Whilst there is clear heterogeneity in the studies considered together in table 13, the pooled data gives an overall assessment of the performance of CT-guided biopsy in reported clinical practice. Overall sensitivity is 90.7% (95% CI 88.8% to 92.4%), specificity 93.9% (95% CI 91.1% to 96.0%), PPV 97.4% (95% CI 96.2% to 98.3%) and NPV 79.9% (95% CI 76.0% to 83.4%). The negative predictive value is of particular importance, as clinicians often have to make a decision about management of a non-malignant biopsy result, mindful of the possibility of a false-negative result.

View this table:

• View inline
• View popup

Table 13

Case series of CT-guided biopsy of pulmonary nodules

As with any diagnostic test, the post-test probability of malignancy (after a non-malignant CT biopsy) will depend on the pre-test probability and the negative likelihood ratio (calculated here as 0.10, 95% CI 0.08 to 0.12). The effect of a negative (ie, non-malignant) biopsy on the post-test probability of cancer is shown in figure 10. Where the pre-test probability is high (eg, 90%) there is still approximately a 50% chance of malignancy even after a non-malignant biopsy (exact value 47.0%, 95% CI 41.9% to 51.9%). This has recently been confirmed in the largest retrospective series, where there was a 90% prevalence of malignancy. However, in this series half of the lesions were outside the definition of pulmonary nodules as they were greater than 30 mm. The authors emphasised the importance of considering repeat biopsies as they showed that repeat biopsies usually confirm the diagnosis of malignancy.199 However, if the pre-test probability of cancer is only 50%, then the chance of malignancy drops to about 10% after a non-malignant biopsy (exact value 9.0%, 95% CI 7.4% to 10.7%).

• Download figure
• Open in new tab
• Download powerpoint

Figure 10

The effect of a negative CT-guided percutaneous biopsy (CTgBx) on the probability of a pulmonary nodule being malignant.

The impact of CT-guided biopsy findings on clinical decision-making was investigated by Baldwin et al.189 Clinicians were presented with 114 patient scenarios with and without the results of CT-guided biopsy of pulmonary nodules, and asked to specify management. The proportion of successful decisions (against known outcomes) was assessed. Agreement between clinicians on the need for surgery increased with biopsy result information compared with CT findings alone (κ value 0.57 vs 0.44, respectively). The major benefit of knowing the CT-guided biopsy result was a reduction in unnecessary surgery, especially when the clinical perception of pre-test probability of malignancy was intermediate (31–70%).

Summary

Although some biomarkers show interesting early results, further studies are required to validate their performance prospectively in clearly defined patient populations before they can be recommended for clinical use. Standard bronchoscopy has a very low yield but this can be increased with the image-guidance techniques described (fluoroscopy, rEBUS and ENB). However, no studies compared performance between these various techniques, and direct comparison between yields described in these series is limited by significant heterogeneity in inclusion criteria, and multiple confounding factors. The reported yields (65–84% for ENB and 46–77% for rEBUS) were less than those for CT-guided percutaneous transthoracic biopsy (pooled 91%), although the latter has a much higher pneumothorax rate (6.6% requiring chest drain in the largest series). The latter may be important for some patients although ENB and to a lesser extent rEBUS may be very time-consuming and are not as widely available as percutaneous biopsy. Figure 10 shows how in a percutaneous biopsy, the pre-test probability of malignancy is altered by a negative biopsy. This may be important when explaining the possible implications to patients.

Evidence statements

• Biomarkers do not offer sufficient accuracy to differentiate malignant from benign nodules. Evidence level 3

• The diagnostic yield of bronchoscopy in the investigation of pulmonary nodules is low. Evidence level 3

• Diagnostic yield of bronchoscopy may be increased by the use of fluoroscopy, electromagnetic navigation and radial endobronchial ultrasound and in the presence of a CT bronchus sign. However, yield remains relatively low for lesions <2 cm in the peripheral third of the lung. Evidence level 3

• The diagnostic yield from CT-guided biopsy of pulmonary nodules decreases with decreasing size of the lesion. Evidence level 2+

• Techniques such as multiplanar reconstructed images and C-arm cone-beam CT may increase the yield. Evidence level 3

• Pneumothorax is the most common complication of CT-guided biopsies; by far the largest study showed an incidence of 15%, with 6.6% of patients requiring an intercostal drain insertion. Consistent factors that increase the risk are lower FEV₁ and presence of emphysema along the needle tract. Evidence level 3

• The post-test probability of malignancy after a negative lung biopsy depends on the pre-test probability. Evidence level 3

• Repeating biopsies in patients with nodules with a high probability of malignancy showed a high confirmation rate of malignancy. Evidence level 3

Recommendations

• Do not use biomarkers in the assessment of pulmonary nodules. Grade D

• Consider bronchoscopy in the evaluation of pulmonary nodules with bronchus sign present on CT. Grade D

• Consider augmenting the yield from bronchoscopy using either radial endobronchial ultrasound, fluoroscopy or electromagnetic navigation. Grade D

• Offer percutaneous lung biopsy in cases where the result will alter the management plan. Grade C

• Consider the use of other imaging techniques such as C-arm cone beam CT and MPR to improve diagnostic accuracy. Grade D

• Consider the risk of pneumothorax when deciding on a transthoracic needle biopsy. Grade C

• Interpret negative lung biopsies in the context of the pre-test probability of malignancy. Grade D

• Consider repeating percutaneous lung biopsies where the probability of malignancy is high. Grade D

• Research should be undertaken into the application of new and existing biomarkers in the evaluation of pulmonary nodules. RR

Optimal surgical management for nodules confirmed to represent lung cancer

Localisation techniques for pulmonary nodules

If limited resection is planned, nodules that are either of small size, located deep to the visceral pleura, or of ground-glass morphology may be difficult to locate at thoracoscopic surgery. A number of techniques have been developed to facilitate localisation of these nodules. Some techniques involve preoperative marking of nodules and include CT-guided hookwire/needle/microcoil insertion, Lipiodol injection (lipid-soluble contrast medium with subsequent intraoperative fluoroscopy), methylene blue injection (to identify the overlying visceral pleura to guide resection) or radio-tracer injection (using ^99mTc macro-aggregated albumin with subsequent use of intraoperative gamma probe). An alternative approach has been to use intraoperative ultrasonography to identify the nodule in a collapsed lung during single lung ventilation.

Considering only reports with 50 or more patients, seven case series were identified for CT-guided wire localisation or equivalent,240–246 three for Lipiodol marking,247–249 one for methylene blue injection,250 two for radio-tracer injection251 ,252 and one for transthoracic ultrasonography.253 One small randomised trial comparing hookwire and radiotracer was identified.254 A summary of these reports is shown in table 15.

The inclusion criteria whereby localisation was deemed necessary varied between case studies. Some studies stipulated a maximum size of nodule (usually 10 mm but 25 mm in one study), or distance from visceral pleura (range 5–15 mm). Some studies included SSNs, whereas others left requirement for localisation to the discretion of the surgeon. The outcome measures varied also, with some reporting successful localisation, whereas others required thoracoscopic resection for success. Success rates according to these disparate criteria range from 84% to 100%. Accepting the limitations of comparison between these series, no one technique appeared more efficacious than any other.

Thirteen of the 14 studies of preoperative localisation reported complications. The remaining study did not discuss complications at all.242 Pneumothorax was reported in all 13 series, with rates of 4–49.1%, although most of these were asymptomatic and did not require treatment. Five studies quoted the rate of pneumothorax requiring chest drain (1.2–6%), and pulmonary haemorrhage (7–29.8%). Pain was reported in two series (7, 11%) and dislodgement of wire/coil in four series (1.8–7.5%). One patient (0.6%) undergoing Lipiodol injection developed a haemopneumothorax requiring immediate operation.247 The one report of intraoperative ultrasound localisation described no complications.253 Ultrasound successfully localised 94% of nodules, but this was more difficult when the surrounding lung was emphysematous.

Gonfiotti et al254 reported a small randomised trial of hookwire versus radiotracer localisation for resection of nodules <2 cm in diameter (n=25 in each arm). The hookwire technique successfully located 84% of nodules compared with 96% with radiolabelling. Twenty-four per cent of patients in the hookwire group developed a pneumothorax compared with 4% in the radiolabelling group (none needed insertion of chest drain). No specific details were given of the randomisation process. No significant differences were reported between the two groups, reflecting the small sample size.

Surgical management of SSNs

Six case series (all from Japan) were identified specifically reporting outcomes for patients with SSNs undergoing surgical resection.256–261 The studies differed in their inclusion criteria, with some reporting outcomes for a combined sub-solid cohort (pGGNs and PSNs), other reports subdividing these two populations or reporting outcomes according to the ratio between consolidation and solid tumour, and other studies considering only small nodules (<15 or <20 mm). The surgical management differed also, with some case series reporting outcomes for lobectomy versus sublobar resection, whereas others compared segmentectomy and wedge resection.

The consistent finding between all case series was excellent long-term prognosis from sublobar resection with low rates of local recurrence. Thus Tsutani et al258 reported 3-year overall survival of 98.7% and 98.2% for ground-glass opacity dominant tumours (>50% ground-glass component) undergoing wedge resection (n=93) or segmentectomy (n=56), respectively. Three-year recurrence-free survival was 98.1% and 96.7%, respectively. Iwata et al259 described a case series of patients undergoing segmentectomy for NSCLC, which included 38 patients with ground-glass opacity dominant tumours (>50% ground-glass component), none of whom died in the follow-up period (mean follow-up 34.6 months). Yano et al261 described a case series of 810 patients with stage IA lung cancer with a consolidation/tumour ratio <0.25, reporting 5-year overall survival of 96.7% and disease-free survival of 96.5%.

Summary

Lobectomy appears to be associated with improved outcomes compared with sublobar resection in one RCT and three retrospective cohort studies of early-stage lung cancer, with one cohort study, part of a screening programme, showing equivalence. Anatomical segmentectomy was found to be oncologically equivalent to lobectomy in one retrospective cohort study and two retrospective propensity matched analyses. In a meta-analysis of observational studies, segmentectomy had worse survival for stage I and stage IA tumours, but equivalent survival for tumours ≤2 cm in diameter. One cohort study and two subgroup analyses have suggested worse outcome for wedge resection than for segmentectomy. Further prospective, randomised evidence would clarify the relative oncological performance of anatomical segmentectomy and lobectomy.

Localisation techniques seem to be a necessary aid for resection of smaller nodules and no one technique was identified as better than the others.

There is very limited evidence to suggest nodal dissection may not be necessary for nodules <10 mm, or for PSNs with a solid component <10 mm.

The evidence comparing lobar and sublobar resection in SSNs was limited to case series, but the excellent survival and low rates of recurrence from sublobar resections in these series suggest that there may be little to be gained by extending to a lobectomy. Unfortunately, there was inconsistency in the inclusion criteria reported relating to the cut-off point or inclusion of PSNs (eg, >50% ground-glass component vs consolidation/tumour ratio <0.25). Therefore the recommendation for sublobar resection can only be confidently made for pGGNs. In the absence of specific evidence for PSNs with consistent definitions, these should probably be surgically managed in the same way as solid nodules.

From the limited evidence available, the rate of lymph node metastases is low and related to size of the solid component in PSNs. The rate in pGGNs is negligible (see section Management of SSNs, sub-section lymph node metastases).

Evidence statements

• VATS wedge resection with intraoperative frozen section has a high diagnostic sensitivity and specificity and generally low complication/mortality rates. Evidence level 3

• Case series dealing with the problem of whether to proceed to surgical resection without preoperative biopsy are limited by confounding factors. Evidence level 3

• Benign resection rates vary considerably between published case series. Evidence level 3

• Excision biopsy of benign lesions has been shown to lead to change in treatment. Evidence level 3

• Lobectomy for early-stage lung cancer was associated with reduced locoregional recurrence and probable improved survival compared with combined  results for wedge resection and segmentectomy, in a randomised controlled trial. Patients were not diagnosed or staged contemporarily. Evidence level 2+

• Anatomical segmentectomy is associated with reduced locoregional recurrence and possibly improved survival compared with non-anatomical wedge resection for early-stage lung cancer. Evidence level 2+

• There is emerging evidence to demonstrate oncological equivalence of segmentectomy and lobectomy for tumours ≤2 cm in diameter. Evidence level 2+

• There is no evidence to suggest superiority of any particular localisation technique for impalpable nodules, and no consistent criteria for when these should be used. Complications rates in some case series are high, although mostly relate to asymptomatic pneumothorax or pulmonary haemorrhage not requiring specific treatment. Evidence level 3

• Despite heterogeneity in inclusion criteria and details of surgical management, case series of SSNs undergoing predominantly sublobar resection report very good long-term prognosis. Evidence level 3

Recommendations

• Surgical resection of pulmonary nodules should preferentially be by VATS rather than by an open approach. Grade C

• Offer lobectomy (to patients fit enough to undergo the procedure) as definitive management of a pulmonary nodule confirmed as lung cancer preoperatively or after wedge resection and intraoperative frozen section analysis during the same anaesthetic procedure. Grade C

• Consider anatomical segmentectomy where preservation of functioning lung tissue may reduce the operative risk and improve physiological outcome. Grade D

• Consider a diagnostic anatomical segmentectomy for nodules <2 cm in diameter without nodal disease when there has been no pathological confirmation and frozen section is not possible. Grade D

• Use localisation techniques, depending on local availability and expertise, to facilitate limited resection of pulmonary nodules. Grade D

• Consider sublobar resection for pGGNs deemed to require surgical resection owing to the excellent long-term prognosis and low risk of local relapse. Grade D

• Prospective trials should compare complications and oncological outcomes from lobectomy versus anatomic segmentectomy in appropriately selected patients. RR

Outcome of patients treated without pathological confirmation

Four retrospective cohort studies specifically compared outcomes in patients with clinically diagnosed lung cancer (CDLC) versus patients with pathologically proven NSCLC.262–265 Summary details of the studies are shown in table 16. In all four studies, a clinical diagnosis of lung cancer was made on the basis of clinical characteristics, CT findings (including progressive enlargement) and FDG avidity on PET-CT scan, and three studies explicitly recorded the decision being made by multidisciplinary team consensus.262–264

Takeda et al262 reported similar 3-year local control, progression-free survival, cause-specific survival and overall survival rates between CDLC and pathologically proven NSCLC, suggesting that few benign lesions were likely to have been included in the CDLC group. Patients with CDLC did not undergo histological confirmation because of negative biopsy results, increased risk of biopsy or patient choice. No quantitative model was used to define the CDLC group, limiting direct comparison with other studies. Verstegen et al263 reported a retrospective analysis from a prospectively collected institutional database for patients undergoing stereotactic ablative body radiotherapy (SABR) for proven or suspected stage I NSCLC. Outcomes (shown in table 16) were similar between the two groups. No quantitative prediction model was prospectively used to define the CDLC population, but retrospective use of prediction models from Swensen et al64 and Herder et al55 indicated a mean probability of malignancy of 92.5% (95% CI 91.8% to 93.3%) in the CDLC group and 94.8% (95% CI 94.2% to 95.4%) in the NSCLC group. Potential confounders included lower FEV₁ and a higher proportion of T1 tumours in the CDLC group (although subgroup analysis was performed by T stage showing no differences in outcome). Additionally, the high proportion of patients with other previous malignancy (34%) raised the possibility that some presumed primary lung cancers were instead metastatic recurrence. A third retrospective cohort by Haidar et al264 reported outcomes for 55 patients undergoing SABR for early-stage lung cancer. The groups were well matched according to the limited clinical information supplied, and over a mean follow-up of 24 months, local control, actuarial 1- and 2-year survival and toxicities did not differ between the two groups. Finally, a retrospective cohort study by Stephans et al265 was identified which compared outcomes after two different SABR protocols for patients with stage I NSCLC. As a secondary analysis, outcomes were compared between clinically and pathologically diagnosed lung cancers, with no significant difference shown in overall survival (p=0.37). Patient characteristics for the NSCLC and CDLC groups were not reported, thereby limiting the ability to identify and assess possible confounding variables.

Of the four studies considered, three were explicit about potential confounding variables (and in Verstegen et al, attempted to minimise one confounding factor). All four studies were consistent in reporting similar outcomes between pathologically confirmed and CDLC treated with SABR, thereby tending to support the accuracy of clinically diagnosed cases when made by a multidisciplinary assessment of clinical and radiological criteria.

A recent study developed a decision tree and Markov model comparing the relative merits of surveillance, a PET-CT scan directed SABR strategy without histological confirmation and a PET-CT scan–biopsy–SABR strategy. The authors concluded that when there are concerns about biopsy-related morbidity, a PET scan–SABR policy is warranted when the pre-test probability of malignancy in pulmonary nodules exceeds 85%.165 However, the estimated complication rate might have been below that expected in people with comorbidities sufficient to make percutaneous biopsy a concern.

Treatment modalities

Publications relating to non-surgical treatment modalities for pulmonary nodules largely comprised case series and poor-quality retrospective cohort studies for patient populations with presumed or pathologically proved malignancy. The majority of studies considered RFA (n=25 studies)266^–290 and SABR (n=14)262 ,265 ,291–302. Other publications reported outcomes from conventional radiotherapy (n=3),281 ,297 ,303 percutaneous cryotherapy (n=1),276 microwave ablation (n=2)304 ,305 inhaled corticosteroids (n=2)306 ,307 and antibiotics (n=1).308

Comparison of outcomes between the case series and cohort studies reviewed was severely limited by a number of problems. First, the reviewed studies considered heterogeneous populations, with some reporting outcomes from early-stage lung cancer only, the majority of reports considering a mixed population of lung cancer and metastases from other solid tumours, and two series considering metastases alone. Second, for case series of pulmonary metastases, there was significant variability in the tissue types considered and the number of metastatic lesions treated. Third, there was variability between studies in the proportion of patients with pathologically proven malignancy (lung cancer or other metastatic disease) and those where malignancy was presumed on the basis of clinical and radiological criteria. Fourth, where patients were treated for presumed malignancy without pathological confirmation, the criteria on which these presumed diagnoses were made were often not explicitly defined, and in the reports where they were defined often varied between cases series. Fifth, patients in some case series received systemic treatment together with local ablative treatment, thus confounding comparison of overall survival. Sixth, some series reported repeated treatments with ablative therapy after disease progression. Finally, the length of follow-up and the outcome parameters reported (which included overall survival, progression-free survival and disease-specific survival) varied between studies. Overall median survival was the most frequently reported parameter. For patients treated for presumed or proven lung cancer, overall median survival after RFA varied between 21 and 44 months, and after stereotactic radiotherapy varied between 24 and 54 months. However for the reasons described above direct comparison between these quoted figures is not appropriate.

While most studies reported outcomes from just one treatment modality, one multicentre and four single-centre retrospective cohort studies compared different treatment modalities for patients with pulmonary nodules presumed or pathologically proved to be early-stage NSCLC.

Verstegen et al309 performed a retrospective cohort analysis of patients treated for stage I–II NSCLC treated with VATS in six hospitals or SABR at a central hospital. Sixty-four cases of each (from 86 VATS and 527 SABR patients) were selected for analysis by investigators who were blinded to the outcome using a propensity score-matched analysis to reduce bias and confounding by matching on multiple variables. This excluded patients with severe COPD (GOLD score 4), previous or synchronous lung malignancy. Outcomes were analysed on an intention-to-treat basis with cases with an eventual benign diagnosis after VATS included. There was no difference in 3-year overall survival and freedom from progression rate. Local/regional control appeared to be better in the SABR group (p=0.037) and treatment-related toxicity appeared to be less in the SABR group. Shorter median follow-up in the VATS arm (16 vs 32 months) and operator experience with the VATS technique are both possible confounding factors.

Hsie et al281 retrospectively compared outcomes in 96 patients with pathologically confirmed stage I NSCLC not suitable for standard surgical resection (lobectomy/pneumonectomy) and treated with either limited surgical resection, RFA or conventional radical radiotherapy. Patients were assigned to treatment groups by clinician preference and the cohorts were not well matched. Significant confounding factors were worse performance status, lower FEV₁ and greater use of long-term oxygen treatment in the radiotherapy group. Three-year survival was 63% for limited resection and 55% for radiotherapy (no quoted figure for RFA owing to small patient numbers), leading the authors to conclude that survival is reasonable for patients not suitable for standard surgical resection.

Crabtree et al294 retrospectively compared outcomes from 538 patients with stage I NSCLC treated with surgery or SABR in a single-centre study. Treatments were assigned by clinical preference and major confounders were differences between the cohorts in age, comorbidity, pulmonary function tests and the proportion of patients with pathological confirmation of malignancy (100% vs 80% for surgery vs SABR). Three-year overall survival was 68% and 32% for patients receiving surgery and SABR, respectively. Propensity analysis matching 57 high-risk surgical patients with 57 patients undergoing SABR showed no significant difference in disease-free survival (77% vs 86%) or overall survival (54% vs 38%) at 3 years.

Widder et al297 reported survival and quality of life data for two cohorts of patients treated with radiotherapy for inoperable stage I lung cancer. Twenty-seven patients treated with 3D conformal radiotherapy (CR) between 1994 and 1996 were compared with 202 patients treated with SABR between 2006 and 2009. Confounding factors included a lack of PET-CT imaging for the CR radiotherapy group, differences in rates of pathological confirmation (74% for the CR group vs 29% for the SABR group), and differences in performance status and age. Two-year overall survival was significantly better in the SABR than CR group (72% vs 48%, HR=2.6, 95% CI 1.5 to 4.8, p<0.01).

McGarry et al303 retrospectively compared outcomes for 128 patients with stage I/IIa NSCLC treated with surgery, radiotherapy (curative or palliative) and observation only, reporting median survival times of 46.2 months, 19.9 months and 14.2 months, respectively. The study demonstrates poor outcome from observation only for lung cancer, but the substantial confounding factors prevent meaningful comparison among groups. These confounders were not explicitly described in the report and no attempt was made to correct for them.

All five cohort studies are subject to selection bias and confounders. However, Verstegen et al made substantial efforts to try to compensate for these factors. Propensity score matching was used to reduce bias by matching patients according to numerous baseline variables, and analysis was performed on an intention-to-treat basis of clinical diagnosis, irrespective of the final histological result. The four single-centre cohort studies281 ,294 ,297 ,303 have significant selection bias and major confounding factors, which preclude direct comparison between outcomes in the groups studied.

Two RCTs assessed the effect of inhaled corticosteroids on nodule size in patients with persistent indeterminate pulmonary nodules. Veronesi et al306 randomised 202 patients to inhaled budesonide 800 µg twice a day or placebo for 12 months and showed no effect on pre-existing nodule size or the development of new nodules. van der Berg et al307 randomised patients with evidence of bronchial squamous metaplasia/dysplasia and either >20 pack-year history of smoking or previous history of lung or head and neck cancer to inhaled fluticasone. Again no effect was seen on either previously detected nodules or the development of new nodules.

Khokhar et al308 retrospectively reviewed patients with pulmonary nodules to see whether antibiotic prescription was associated with a change in CT appearance of nodules on a follow-up scan. No significant difference was seen in nodule appearance between 34 patients who received antibiotics and 109 patients who did not. The authors concluded that their data did not support routine use of antibiotics in patients found to have pulmonary nodules on CT scan. Significant selection bias and confounding variables were present.

Harms of treatments

The potential harms of treatments for presumed or proven malignant nodules have been reported in a number of case series. There was wide variability between studies in the frequency of reported complications, which related in part to the different criteria used to define/grade these complications. For example, some case series reported any haemoptysis following RFA treatment, whereas others reported only significant or major bleeding without specifically defining the relevant criteria. Similarly, pneumothoraces were classed as minor, major or sometimes only reported if intercostal drain insertion was required.

The frequency of complications in case series of patients treated with RFA266 ,267 ,271–273 ,275–278 ,280 ,281 ,283–287 ,289 ,290 ,304 ,305 was as follows: pneumothorax was the most commonly reported complication with rates varying from 9% to 54% in 19 case series. Other reported complications after RFA were bleeding (0.7–26%), pleural effusion (1.8–19%), pneumonia (1.8–12%), pleuritis (0.6–4.3%), lung abscess (0.3–3.1%), haemothorax (3.0%), severe pain (2%), bronchopleural fistula (1.5–1.8%), acute respiratory distress syndrome (1.5%) and pericardial tamponade (0.9%). Procedure-related mortality varied from 0% to 0.9% in seven case series, although one series reported a 30-day procedure-related mortality of 2.6%.289

Currently, in the UK, lung SABR is used only for peripheral lesions, and treatment is generally very well tolerated provided that organ at risk tolerances are adhered to. The main acute toxicities are fatigue, chest pain, skin erythema and cough, but these side effects are almost always mild (<grade 3) and self-limiting. Severe radiation pneumonitis —that is, grade 3 (requiring oxygen, severe symptoms±limiting self-care), is uncommon (range 1–2.8%) and grade 2 (symptomatic requiring medical intervention±limiting activities of daily living) or less ranges from 1% to 11%.310–312 The incidence of radiation pneumonitis does not appear to be higher in patients with poor pulmonary function.313–316 There is no strong evidence for absolute dose constraints, though in one large institutional series the risk of pneumonitis was higher in tumours with a large radiotherapy volume (internal target volume >145 ml) and when the contralateral lung receives a mean dose of >3.6 Gy.317 Guckenberger et al318 observed a dose relationship between ipsilateral lung dose and the development of radiation pneumonitis. Patients developing pneumonitis had an ipsilateral mean lung dose of 12.5±4.3 Gy compared with a mean dose of 9.9±5.8 Gy in unaffected patients. Ideally, the mean lung doses should be low in SABR and these figures are only a guide as they are based on relatively small numbers of patients and events. To minimise the risk of pneumonitis the UK SABR consortium has produced strict planning guidelines, which include limits for lung doses. Rib fracture and chest wall pain are the main late side effects with varying incidence depending on the dose fractionation scheme used. In one large single institutional series of >500 patients using a risk adaptive dose schedule with reduced doses for lesions close to or involving the chest wall, severe (grade 3 or higher) chest wall toxicity was rare ≤ 2% and grade 2 or less toxicity was <10%.291

Currently, SABR is not routinely used for lesions close to central mediastinal structures. This practice is based on a phase II study by Timmerman et al which showed that for central lesions treated with SABR the rates of severe toxicity (grade 3 or higher) were 46% at 2 years compared with 17% for peripheral lesions. Toxicities included decline in pulmonary function tests, pneumonias, pleural effusions, apnoea, skin reaction and treatment-related deaths.295 A more recent systematic review of SABR for central lesions showed lower rates of toxicity319 and a European Organisation for Research and Treatment of Cancer (EORTC) study started in November 2014 (trial reference EORTC -222113-08113-ROG-LCG) to evaluate SABR for central lesions.

Summary

Four retrospective case series provide evidence that for SABR, outcomes are similar for people with nodules that are not pathologically confirmed as malignant and for those with prior confirmation. Most evidence found was for SABR and RFA, although the variability in case definition, pathological confirmation, proportion of primary and secondary cancer, selection criteria and concomitant treatment made comparison inappropriate. One study used propensity score matching in a comparison of SABR and VATS for stage I–II lung cancer and showed a similar 3-year outcome. Non-surgical treatments show marked variation in the frequency of harms, something that is likely to be strongly influenced by case selection and the technique employed.

Evidence statements

• Where biopsy of a pulmonary nodule is either non-diagnostic or not possible, patients diagnosed with clinical lung cancer by a multidisciplinary assessment on the basis of clinical and radiological criteria appear to have similar outcomes to patients with pathological confirmation of malignancy following SABR. Evidence level 2+

• Inhaled corticosteroids have no effect on pre-existing nodule size or on the development of new pulmonary nodules on CT scan. Evidence level 1+

• There is no evidence to support routine use of antibiotics in the management of indeterminate pulmonary nodules. Evidence level 3

• Treatment of central lesions with SABR has been associated with higher rates of toxicity than with SABR for peripheral lesions. This may be alleviated by using risk-adapted dose schedules and is the subject of ongoing clinical trials. Evidence level 3

• In the treatment of pulmonary nodules, proved or presumed to be malignant, SABR and RFA have low rates of acute mortality. Reported rates of morbidity are highly variable. Evidence level 3

Recommendations

• For people who are unfit for surgery who have pulmonary nodule(s) with a high probability of malignancy, where biopsy is non-diagnostic or not possible, consider treatment with SABR or RFA, if technically suitable. Grade C

• For people who are unfit for surgery who have pulmonary nodule(s) with a high probability of malignancy, where biopsy is non-diagnostic or not possible, consider treatment with conventional radical radiotherapy if not suitable for SABR or RFA. Grade D

• Do not use inhaled corticosteroids in the management of indeterminate pulmonary nodules. Grade B

• Do not use antibiotics in the management of indeterminate pulmonary nodules. Grade D

• Consider prospective randomised trials of local treatments for pathologically proved or clinically diagnosed early-stage lung cancer and pulmonary oligometastases. RR

• For prospective randomised trials of interventions for pathologically proved or clinically diagnosed early-stage lung cancer include assessment of harms. RR

Advances in imaging technology

Imaging technology is improving rapidly and evidence reviewed for the purposes of guideline development often reports on technology that has effectively become out of date, although may still be in use clinically. Thus recommendations have to be interpreted in the light of current technology.

The most significant technological changes over the lifetime of this guideline may be:

1. The introduction of CT scanners using iterative reconstruction that will substantially reduce the effective radiation dose. Possible consequences are a lowering of the threshold for performing CT scans, thus increasing the number of incidentally detected nodules, and a more permissive approach to follow-up examinations.

2. The increased use of perfusion CT to assess nodule vascularity and perfusion. This may have implications for prognostication, and differentiation between benign and malignant nodules

3. Improved image data reconstruction, such as nodule surface textural analysis. This may aid differentiation of benign from malignant nodules.

4. Changes in PET-CT scanner construction and image processing. This is likely to produce substantial improvements in the accuracy of characterisation of pulmonary nodules. Appendix 2 shows specific likely future developments in PET-CT.

5. The method of reporting positivity in PET-CT may be optimised; methods may include modification of absolute SUV cut-off points according to nodule type. There may be different values for solid and SSNs and nodule size with lower levels for nodules with greater ground-glass components and for smaller nodules.

Method of detection

Extensive publications on the detection of pulmonary nodules using CXR and CT scan are available. These include new techniques to improve detection by CXR, such as subtraction methods and computer-aided detection (CAD), as well as improved detection by CT in comparison with CXR. The latter includes CAD in CT and reconstruction algorithms such as MPR, maximum intensity projection (MIP), and volume rendering (VR). Aside from the reconstruction algorithms none of these techniques are in use in routine clinical practice and remain areas for research. These guidelines focus on nodule characterisation once detected but it is known that nodules are better detected and characterised if a CT scan maximum section thickness of 1.25 mm323–325 is used and if they are reported using software reconstruction algorithms including MPR, or MIP or VR review.326–337

Factors influencing the accuracy of measurements