Article Text

PDF

Prognostic significance of hypoxia-inducible factor-1α, TWIST1 and Snail expression in resectable non-small cell lung cancer
  1. J-J Hung1,3,4,6,
  2. M-H Yang1,7,8,
  3. H-S Hsu2,6,8,
  4. W-H Hsu3,6,
  5. J-S Liu4,
  6. K-J Wu1,5,8
  1. 1
    Institute of Clinical Medicine, National Yang-Ming University, Taipei, Taiwan
  2. 2
    Institute of Emergency and Critical Care Medicine, National Yang-Ming University, Taipei, Taiwan
  3. 3
    School of Medicine, National Yang-Ming University, Taipei, Taiwan
  4. 4
    Department of Surgery, Cathay General Hospital and School of Medicine, Fu Jen Catholic University, Taipei, Taiwan
  5. 5
    Institute of Biochemistry and Molecular Biology, National Yang-Ming University, Taipei, Taiwan
  6. 6
    Division of Thoracic Surgery, Department of Surgery, Taipei Veterans General Hospital, Taipei, Taiwan
  7. 7
    Division of Hematology-Oncology, Department of Medicine, Taipei Veterans General Hospital, Taipei, Taiwan
  8. 8
    Genomic Research Center, Taipei Veterans General Hospital, Taipei, Taiwan
  1. Correspondence to Dr K-J Wu, Institute of Biochemistry and Molecular Biology, National Yang-Ming University, No. 155, Section 2, Li-Nong Street, Peitou, Taipei 112, Taiwan; kjwu2{at}ym.edu.tw

Abstract

Background: Metastasis is the most common cause of disease failure and mortality for non-small cell lung cancer (NSCLC) after surgical resection. Snail and TWIST1 are epithelial-mesenchymal transition (EMT) regulators which induce metastasis. Intratumoral hypoxia followed by stabilisation of hypoxia-inducible factor 1α (HIF-1α) promotes metastasis through regulation of certain EMT regulators. The aim of this study was to evaluate the prognostic value of HIF-1α, TWIST1 and Snail expression in patients with resectable NSCLC.

Methods: A retrospective analysis of 87 patients with resectable NSCLC from Taipei Veterans General Hospital between 2003 and 2004 was performed using immunohistochemistry to analyse HIF-1α, TWIST1 and Snail expression. The association between HIF-1α, TWIST1 and Snail expression and patients’ overall and recurrence-free survivals was investigated.

Results: Overexpression of HIF-1α, TWIST1 or Snail was shown in 32.2%, 36.8% and 55.2% of primary tumours, respectively. Overexpression of HIF-1α, TWIST1 or Snail in primary NSCLCs was associated with a shorter overall survival (p = 0.005, p = 0.026, p = 0.009, respectively), and overexpression of HIF-1α was associated with a shorter recurrence-free survival (p = 0.016). We categorised the patients into four groups according to the positivity of HIF-1α/TWIST1/Snail to investigate the accumulated effects of these markers on survival. Co-expression of more than two markers was an independent prognostic indicator for both recurrence-free survival and overall survival (p = 0.004 and p<0.001, respectively, by multivariate Cox proportional hazards model).

Conclusions: Co-expression of more than two markers from HIF-1α, TWIST1 and Snail is a significant prognostic predictor in patients with NSCLC.

Statistics from Altmetric.com

Non-small cell lung cancer (NSCLC) is the leading cause of cancer death worldwide. Surgical resection is the treatment of choice for early-stage NSCLC.1 Tumour recurrence and metastasis are the most common events encountered after resection that lead to mortality.2 3 4 Chemotherapy and radiotherapy are common treatment modalities applied to patients with recurrent lung cancer.4 5 However, the combination modality did not significantly improve patients’ survival. Many molecular markers have been shown in the literature to predict prognosis and survival of patients with NSCLC.6 7 8 A constellation of 3–5 markers or >20 markers in NSCLC have been reported by different groups with little overlap.9 10 11 12 13 Since tumour metastasis is the main obstacle for long-term survival after surgical resection, identification of molecular markers related to metastasis may better reflect and predict the prognosis and survival in patients with NSCLC.

Epithelial-mesenchymal transition (EMT) is considered to be one of the major molecular mechanisms inducing tumour invasion and metastasis.14 15 Repression of E-cadherin is a hallmark of the EMT process.15 Many EMT regulators including Snail, TWIST1, Slug, Zeb1, SIP1 and E47 have been shown to induce EMT through the repression of E-cadherin expression.16 17 18 19 Increased expression of Snail or TWIST1 was associated with tumour recurrence, metastasis and poor prognosis in different types of human cancers.19 20 21 22 23 24 However, the roles of these two markers in NSCLC remain unknown. Intratumoral hypoxia, followed by activation of hypoxia-inducible factor 1 (HIF-1), is one of the most important mechanisms promoting tumour aggressiveness, metastasis and poor prognosis.14 25 The hypoxic response is mainly mediated by a heterodimer complex (HIF-1) consisting of two basic helix-loop-helix (bHLH) transcription factors (HIF-1α and HIF-1β). HIF-1α is a cytoplasmic protein regulated by oxygen levels, whereas HIF-1β (also known as ARNT) is a constitutively expressed nuclear protein.19 26 Increased HIF-1α expression correlates with metastasis, poor prognosis and resistance to treatment in a variety of tumours including NSCLC.9 27 28 29 HIF-1α stabilisation was shown to induce the expression of certain EMT regulators.19 However, the combined use of HIF-1α, TWIST1 and Snail as prognostic markers in NSCLC has not been investigated.

We have previously demonstrated that HIF-1α regulates the expression of TWIST1 by binding directly to the hypoxia response element in the TWIST1 proximal promoter.30 Knockdown of TWIST1 or HIF-1α by short-interference RNA reverts EMT and metastatic phenotypes in lung cancer H1299 cells.30 Co-expression of HIF-1α, TWIST1 and Snail in primary tumours of patients with head and neck squamous cell carcinoma correlates with the highest percentage of metastasis and the worst prognosis.30 In this report we show that increased HIF-1α, TWIST1 or Snail expression is observed in a significant percentage of patients with NSCLC using immunohistochemistry. Overexpression of HIF-1α, TWIST1 or Snail correlated with poor overall survival in patients with NSCLC. Co-expression of any two or all markers from HIF-1α, TWIST1 and Snail in primary tumours of patients with NSCLC correlated with a significantly worse prognosis. These results demonstrated the prognostic value of the three markers to predict the overall and recurrence-free survivals in patients with resectable NSCLC.

Methods

Patients and treatment

From January 2003 to December 2004, 87 patients undergoing surgical resection for NSCLC at Taipei Veterans General Hospital were enrolled in this study. The preoperative staging investigation was performed as previously described.31 Chest and upper abdomen CT scans and bronchoscopy were routinely performed before operation. A whole-body bone scan and CT scan of brain were used to exclude possible metastasis. Mediastinoscopy was performed only when enlarged mediastinal lymph nodes (diameter over 1.0 cm) were shown by CT scanning. Patients with suspected distant metastasis were excluded from surgical procedures. All patients underwent complete resection of lung cancer with mediastinal lymph node dissection. The resected specimens and all dissected mediastinal lymph nodes were sent to the pathologists for pathological staging. TNM classification of the International Union Against Cancer was used for determination of disease stages.32 The criteria of adjuvant therapy included advanced stage (chemotherapy) and better local disease control (radiotherapy). All patients were followed up at the outpatient department quarterly in the first 2 years and semi-annually thereafter.

Immunohistochemistry

The specimen processing and immunohistochemistry procedures were performed as previously described.22 30 33 Tumour and neighbouring normal tissue were cut into 6 μm sections from the NSCLC specimens for immunohistochemistry analysis. The samples were fixed in acetone, air-dried and followed by bath in TBS solution (pH 7.6). The endogenous peroxidase activity was blocked with 3% hydrogen peroxide for 5 min. For HIF-1α immunohistochemistry staining, a mouse monoclonal anti-HIF-1α antibody (catalogue number ab8366, Abcam Ltd) was used at the dilution of 1:50 and incubated at 4°C overnight. Tissue sections were also stained with a rabbit polyclonal antibody against TWIST1 (catalogue number ab50581, Abcam Ltd) or Snail (catalogue number ab17732, Abcam Ltd) at the dilution of 1:100 and incubated for 1 h, respectively. Sections were again incubated with a biotinylated secondary antibody for 10 min. The sections were then visualised using streptavidin-horseradish peroxidase conjugate (DAKO LSAB kit; DAKO, Los Angles, California, USA) with 3-amino-9-ethylcarbazole as the chromogen. Finally, all slides were counterstained with haematoxylin.

Immunohistochemical scoring

The interpretation of immunohistochemistry results for HIF-1α, TWIST1 and Snail was performed independently by two pathologists, according to the criteria described previously.22 30 33 The pathologists scoring the immunohistochemistry were blinded to the patients’ outcome. The immunoreactivity of HIF-1α, TWIST1 and Snail was graded from 0 to 3+ (0, no staining; 1+, 1–25%; 2+, 26–50%; 3+, >50% nuclear staining) according to nuclear expression and only 3+ (>50% nuclear staining) was considered as a positive immunohistochemistry result.22 30 33

Statistical analysis

The relationship between HIF-1α, TWIST1 and Snail expression and clinical-pathological characteristics was analysed with a χ2 test. The overall survival and disease-free survival were calculated by the Kaplan-Meier method. Univariate and multivariate analyses were performed by means of the Cox proportional hazards model using SPSS software Version 12.0 (SPSS, Chicago, Illinois, USA). Variables with p value <0.1 after the univariate analysis were entered into multivariate analysis. The log-rank test and Cox proportional hazards model were used to make group comparisons for applying the HIF-1α/TWIST1/Snail prognostic model to predict the prognosis of early-stage NSCLC and outcome of patients receiving adjuvant chemotherapy. Statistical analysis was considered to be significant when the probability value was <0.05.

Results

Overexpression of HIF-1α, TWIST1 and Snail and their correlation with clinicopathological factors in resectable NSCLC

The characteristics of the 87 patients with NSCLC are listed in table 1. Adjuvant therapies included chemotherapy alone for 22 patients, radiation alone for 5 patients, and a combination of chemotherapy and radiotherapy for 4 patients. With a median follow-up time of 43.2 months (37.1 ± 15.1), the 4-year overall survival rate was 74.0%. Tumour recurrence developed in 34 (39.1%) patients during follow-up. To determine the expression of HIF-1α, TWIST1 and Snail in NSCLC samples, immunohistochemical analysis of HIF-1α, TWIST1 and Snail expression was performed in 87 sets of NSCLC samples. A representative case of immunohistochemical staining of all three markers is shown in fig 1. Overexpression of HIF-1α, TWSIT1 and Snail (⩾50% nuclear expression in tumour cells) was shown in 32.2%, 36.8% and 55.2% of lung tumour samples, respectively (table 1).

Figure 1

Immunohistochemical staining of co-expression of hypoxia-inducible factor 1α (HIF-1α), TWIST1 and Snail in corresponding normal tissue (N) and primary tumour (T) of a representative patient with non-small cell lung cancer. The samples prepared for co-expression analysis were cut and examined at the same region. Black arrows indicate the nuclear expression of HIF-1α, TWIST1 and Snail. Photographs were taken at magnifications of 400×. Scale bars represent 100 μm.

Table 1

Characteristics and univariate analyses of 87 patients with lung cancer

Correlation between clinicopathological variables (age, gender, TNM stage, histological type and extent of pulmonary resection) and HIF-1α, TWIST1 and Snail expression are shown in table 2. HIF-1α overexpression was marginally associated with TNM stage (stage II–IV) (p = 0.05). There was no correlation between TNM stage and TWIST1 or Snail overexpression (p = 0.364 and p = 0.377, respectively). TWIST1 overexpression was associated with the histological type of non-adenocarcinoma (p = 0.026). Snail overexpression was marginally associated with extent of pulmonary resection (p = 0.054). The association between HIF-1α, TWIST1 and Snail expression was demonstrated by the Pearson χ2 test. Although overexpression of HIF-1α was not significantly associated with TWIST1 overexpression, it tended to correlate better with TWIST1 overexpression (p = 0.078) than with Snail overexpression (p = 0.474, table 3). Furthermore, there was no correlation between TWIST1 overexpression and Snail overexpression (p = 0.135, table 3).

Table 2

Association between patterns of hypoxia-inducible factor 1α (HIF-1α), TWIST1 and Snail expression and clinicopathological variables

Table 3

Association between hypoxia-inducible factor 1α (HIF-1α), TWIST1 and Snail expression in patients with lung cancer

Overexpression of HIF-1α, TWIST1 and Snail as prognostic factors in patients with NSCLC

To investigate the prognostic impact of HIF-1α, TWIST1 and Snail overexpression in NSCLC, Kaplan-Meier survival analyses were carried out and the differences in survival between groups were examined. The results showed that overexpression of HIF-1α, TWIST1 or Snail alone in primary NSCLCs was associated with a shorter overall survival (p = 0.005, p = 0.026, p = 0.009, respectively; fig 2A–C). Considering recurrence-free survival, overexpression of HIF-1α was associated with a shorter recurrence-free period (p = 0.016) whereas overexpression of TWIST1 or Snail did not influence recurrence-free survival (p = 0.479 and p = 0.336, respectively).

Figure 2

Kaplan-Meier survival analysis in patients with non-small cell lung cancer (NSCLC) with (A) hypoxia-inducible factor 1α (HIF-1α) (−) vs HIF-1α (+), (B) TWIST1 (−) vs TWIST1 (+), and (C) Snail (−) vs Snail (+) in primary tumours. Overexpression of HIF-1α, TWIST1 or Snail in primary NSCLCs was associated with a shorter overall survival in the respective groups.

Generation of a prognostic prediction model for patients with NSCLC using a combination of HIF-1α/TWIST1/Snail staining results

To investigate the accumulative effects of HIF-1α, TWIST1 and Snail expression on the prognosis of NSCLC, we divided these 87 patients into four groups according to the number of positive markers from HIF-1α, TWIST1 and Snail overexpression. The patients were scored according to the number of positive markers: 0 (none positive), 1 (one positive), 2 (two positive) and 3 (co-expression of all three markers). Kaplan-Meier overall survival curves were generated and differences between the four groups were examined. The results showed that patients with overexpression of any two of HIF-1α, TWIST1 and Snail (score 2) or all of the three markers (score 3) had a worse overall survival (fig 3A). We therefore divided the patients into the following two groups: score 0–1 vs score 2–3. The result showed that patients with score 2–3 had a significantly shorter overall survival (fig 3B). A similar result was shown in recurrence-free survival. Patients who scored 2–3 were correlated with a shorter recurrence-free survival compared with those who scored 0–1 (fig 3C).

Figure 3

Kaplan-Meier survival analysis in patients with non-small cell lung cancer (NSCLC) according to the number of markers including hypoxia-inducible factor 1α (HIF-1α), TWIST1 and Snail which had increased expression. (A) The patients were divided into four groups: HIF-1α(−)/TWIST1(−)/Snail(−) (group1), any one of HIF-1α, TWIST1 or Snail overexpression (group 2), any two of HIF-1α, TWIST1 or Snail overexpression (group 3) and HIF-1α(+)/TWIST1(+)/Snail(+) (group 4). Any two of HIF-1α, TWIST1 or Snail overexpression (group 3) and HIF-1α(+)/TWIST1(+)/Snail(+) (group 4) had a shorter overall survival compared with the other groups. (B) The patients were re-divided into two groups: none or one of HIF-1α, TWIST1 or Snail overexpression (group 1) and any two or all of HIF-1α, TWIST1 or Snail overexpression (group 2). Co-expression of any two or all of HIF-1α, TWIST1 and Snail markers (group 2) had a significantly worse overall survival. (C) The grouping method used in (B) was applied for recurrence-free survival analysis. Co-expression of any two or all of HIF-1α, TWIST1 and Snail markers (group 2) had a significantly worse recurrence-free survival.

Univariate analyses indicated that TNM stage had a significant impact on overall survival (p = 0.030) and recurrence-free survival (p = 0.001, table 1). TNM stage and HIF-1α/TWIST1/Snail co-expression pattern were entered into multivariate analyses. To control for potential confounders, age, gender, histological type and extent of pulmonary resection were also entered into multivariate analyses. Multivariate analyses showed that TNM stage (p = 0.008) and HIF-1α/TWIST1/Snail co-expression pattern (p<0.001) were independent prognostic markers for overall survival (table 4). TNM stage (p<0.001) and HIF-1α/TWIST1/Snail co-expression pattern (p = 0.004) were also significant independent predictors for recurrence-free survival (table 4; see table 1 in online supplement for complete results of multivariate analyses). The prognostic effect of co-expression of more than two markers was confirmed by the Cox proportional hazard model. It was an independent prognostic factor for overall survival as well as recurrence-free survival.

Table 4

Multivariate analyses for recurrence-free and overall survivals of 87 patients with lung cancer

Application of the HIF-1α/TWIST1/Snail prognostic model in predicting the prognosis of early-stage NSCLC and outcome of patients receiving postoperative adjuvant chemotherapy

The predictive value and application of the generated three-marker model in early and advanced NSCLC were examined. For early-stage NSCLC (stages I and II, n = 51) in our study, patients with score 2–3 (n = 20) had significantly worse overall survival than those with score 0–1 (n = 31) (hazard ratio (HR) 5.36, 95% CI 1.08 to 26.57; p = 0.040). However, no difference was observed in recurrence-free survival (HR 2.07; 95% CI 0.78 to 5.54; p = 0.147). Among the 51 patients with early-stage NSCLC, 32 (62.7%) had adenocarcinoma. For early-stage (stages I and II) adenocarcinoma, patients with score 2–3 (n = 10) had significantly worse overall survival than those with score 0–1 (n = 22; p = 0.045, log-rank test). The difference was only marginally significant by the Cox proportional hazards model (HR 7.23; 95% CI 0.75 to 69.59; p = 0.087). No difference was observed in recurrence-free survival (HR 1.58; 95% CI 0.39 to 6.34; p = 0.519). For stage I NSCLC (n = 38), there is a trend toward worse overall survival in patients with score 2–3 (n = 15) than those with score 0–1 (n = 23; p = 0.064, log-rank test). However, there was no significant difference by the Cox proportional hazards model (HR 6.12; 95% CI 0.68 to 54.78; p = 0.105). There was no difference in recurrence-free survival between the two groups (HR 1.99; 95% CI 0.53 to 7.42; p = 0.305). Of the 38 patients with stage I NSCLC, 23 (60.5%) had adenocarcinoma. At the last follow-up session only one of the patients with stage I adenocarcinoma was dead. Three patients had tumour recurrence, including the patient who died. Since only one patient was dead, the difference in overall survival between patients with score 2–3 and those with score 0–1 was not calculated. There was no significant difference in recurrence-free survival (HR 1.31; 95% CI 0.12 to 14.46; p = 0.827) between patients with stage I adenocarcinoma with score 2–3 (n = 6) and those with score 0–1 (n = 17).

In this cohort, 26 patients with advanced NSCLC (ie, stage III–IV) received adjuvant therapy after surgery. Compared with the cases without adjuvant therapy, patients undergoing adjuvant chemotherapy after surgical resection had a favourable overall survival compared with those without treatment (HR 0.29; 95% CI 0.09 to 0.92; p = 0.035). To test whether the generated three-marker model can predict the outcome of patients receiving adjuvant therapy, we examined the effect of HIF-1α/TWIST1/Snail co-expression pattern on overall and recurrence-free survivals in patients receiving adjuvant therapy. Patients with score 2–3 (n = 8) survived for a shorter time than those with score 0–1 (n = 18; HR 7.89; 95% CI 1.52 to 40.96; p = 0.014). However, recurrence-free survival between the two groups was similar (HR 1.94; 95% CI 0.72 to 5.27; p = 0.193). These results suggest that co-expression of more than two markers could be used as a predictor of poor prognosis in early-stage patients and poor outcome in patients receiving adjuvant therapy.

Discussion

This report investigated the prognostic role of HIF-1α, TWIST1 and Snail expression in patients with resectable NSCLC. Our results showed that overexpression of HIF-1α, TWIST1 or Snail in primary NSCLC was associated with a shorter overall survival. HIF-1α overexpression was associated with a shorter recurrence-free survival. TNM stage and HIF-1α/TWIST1/Snail co-expression pattern were significant independent prognostic indicators for both overall and recurrence-free survivals in multivariate analyses. Co-expression of any two or all of HIF-1α, TWIST1 and Snail correlated with significantly worse overall and recurrence-free survival in our study.

Many groups have used different sets of markers to predict the prognosis and survival of patients with NSCLC with some success.6 7 8 9 10 11 12 13 However, there is little overlap between the markers presented by different groups. Our study is the first to predict overall and recurrence-free survival in NSCLC by using a combination of metastasis-related markers. Our results showed that the expression of HIF-1α tended to correlate with TWIST1 expression but not with Snail expression. This observation was consistent with our previous results demonstrating the direct regulation of TWIST1 expression by HIF-1α.30 In our previous study we have also shown that knockdown of TWIST1 or HIF-1α by short-interference RNA reverts EMT and metastatic phenotypes in lung cancer H1299 cells.30 In the case of head and neck cancer, co-expression of HIF-1α, TWIST1 and Snail correlates with metastasis and the worst outcome.30 In this report, overexpression of HIF-1α was associated with worse overall and recurrence-free survival, whereas overexpression of TWIST1 or Snail was only associated with a worse overall survival in NSCLC. By combining HIF-1α, TWIST1 and Snail, we found that co-expression of any two or all of the three markers predicted worse overall and recurrence-free survival in patients with NSCLC. It is possible that different types of cancers use different signalling pathways to reach transformation and mediate metastasis. Our results provided the scenario that staining with two different markers will be suitable to reach prognostic significance in NSCLC.

There is increasing evidence to support the role of postoperative adjuvant chemotherapy in locally advanced NSCLC. However, the effect of adjuvant chemotherapy in early-stage NSCLC remains to be determined.34 35 The lung cancer community is trying to identify poor prognostic factors in stage I NSCLC for adjuvant therapy. In our cohort we analysed the predictive ability of the three-marker model in patients with early-stage NSCLC as well as in those with advanced NSCLC receiving adjuvant chemotherapy. For early-stage NSCLC or adenocarcinoma (stage I and II), the three-marker model demonstrated a difference in overall survival. For stage I NSCLC there is a trend toward worse overall survival in patients with score 2–3. Because only one of the patients with stage I adenocarcinoma had died at the last follow-up session, the difference in overall survival between patients with score 2–3 and those with score 0–1 was not calculated. The difference in recurrence-free survival was not statistically significant in stage I adenocarcinoma in our study (score 2–3 vs score 0–1). However, the number of patients in this group is relatively small. Prospective multi-institutional studies with long-term follow-up are required to further validate our prognostic model in predicting the prognosis of patients with stage I adenocarcinoma. For patients with adenocarcinoma our prognostic model could only be used to predict survival of patients with stage I+II but not those with stage I disease. A summary of published studies is shown in table 5. However, our model has the advantage of using only three markers compared with other studies predicting the prognosis of patients with adenocarcinoma with numerous markers (10 and 50 genes, table 5).11 13 Our study also showed that co-expression of more than two markers was a predictor for poor outcome after adjuvant therapy in advanced NSCLC. These results suggest that the three-marker model may be able to identify the poor prognostic cases in early-stage NSCLC as well as those with advanced disease receiving adjuvant therapy. Adjuvant therapy may be considered in patients with early-stage NSCLC with co-expression of more than two markers after surgical resection. More intensive treatment may also be indicated in patients with advanced disease with co-expression of more than two markers.

Table 5

Comparison of published reports with the current study on predictors of survival in early-stage non-small cell lung cancer (NSCLC)

In conclusion, our results show that co-expression of any two or all of HIF-1α, TWIST1 and Snail is a significant prognostic marker to predict overall and recurrence-free survival in patients with resectable NSCLC. It is also a maker independent of TNM stage. The information generated will be valuable for the diagnosis, prognosis and management of patients with NSCLC.

Acknowledgments

We are grateful to Drs Jin-Fan Chen and Chih-Hao Sun of the Department of Pathology and Laboratory Medicine, Taipei Veterans General Hospital for providing expert opinions on pathological reading. We also thank Ms I-Chun Hung and Mr Yuan-Ren Jeng for their technical assistance.

REFERENCES

View Abstract

Footnotes

  • Funding This work was supported in part by the National Science Council (NSC97-2320-B-010-029, NSC97-2311-B-010-007) (K-JW), NRPGM (DOH97-TD-G-111-038, DOH98-TD-G-111-027) (K-JW), Taipei Veterans General Hospital (97-C1-032, 97-ER2-008) (M-HY), Cathay General Hospital (CGH-MR-9711) (J-JH), a grant from the Ministry of Education, Aim for the Top University Plan (98-AC-T-509) (K-JW), National Health Research Institutes (NHRI-EX98-9611BI) (K-JW).

  • Competing interests None.

  • Provenance and peer review Not commissioned; externally peer reviewed.

  • Ethics approval Informed consent was obtained in writing before patient enrolment. This study has been approved by the Institutional Review Board of Taipei Veterans General Hospital.

  • M-HY and H-SH contributed equally to this article.

  • ▸ Complete results of multivariate analyses are published online only at http://thorax.bmj.com/content/vol64/issue12

Request permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.