Study design

The study had two phases: a screening phase and a clinical phase. In the screening phase⁹ a written symptom questionnaire was sent to a population based sample of children aged 7–12 years; 1633 (81%) questionnaires were returned. From these 1633 children, 229 with respiratory symptoms were enrolled in the Finnish part of the PEACE study⁹; this study was completed by 186 children. These 186 children and a further 128 randomly selected non-symptomatic controls were invited to participate in the clinical phase of the study; a total of 247 children (79% of those invited) agreed to take part (fig 1⇓).^10,11

• Download figure
• Open in new tab
• Download powerpoint

Figure 1

Flowchart of the study: sampling scheme and grouping of the children in the screening phase and after clinical assessment for asthma. *Number of participants from the 10% random sample. †In addition to subjects with incomplete data, six asthmatic children on inhaled steroids were excluded.

Approval for the screening phase of the study was given by the ethics committee of the National Public Health Institute, Helsinki, Finland, and for the clinical study by the ethics committee of the Kuopio University Hospital, Kuopio, Finland.

Grouping of children

Diagnostic tests

No short acting β₂ agonist, anticholinergic drugs, or cromones were administered for 12 hours before the tests, no long term β₂ agonist for 2 days, no theophylline or cough medication for 3 days, and most antihistamines were not given for 3 days before testing (except cetirizine and hydroxyzine (5 days) and astemizole (8 weeks)). Inhaled steroids were not withheld. Baseline FEV₁ had to be ≥70% predicted. In cases of acute respiratory infection the tests were postponed for 2 weeks.

Statistical analyses

The data were analysed using STATA release 6.0 (www.stata.com). One way ANOVA with Bonferroni multiple comparison test was used to compare spirometric baseline values between the groups. The Kruskal-Wallis test was used in the analyses of the free running test, methacholine inhalation challenge, and bronchodilatation test results due to the skewness of the data; the Wilcoxon rank sum test with Bonferroni correction was used in multiple comparisons. As a measure of association between pathological free running test and methacholine inhalation challenge results, the kappa statistic (κ) was calculated.

Multiple logistic regression models were used to evaluate the predictive ability of the symptoms and BHR test results in relation to clinical asthma.¹⁴ In building the model, we first included the symptoms suggestive of asthma (from the screening questionnaire). Secondly, dichotomised free running test and methacholine inhalation challenge results (pathological/normal; see above) were added to the model. The order of adding symptoms or BHR tests to the logistic model (see footnote to table 3⇓) was selected to represent a typical clinical decision making scheme—that is, the order was not based on statistical considerations. The area under the receiver operator characteristic (ROC) curve was used in the assessment of the predictive ability of each logistic model.

In validating the screening questionnaire, the sensitivity and specificity was calculated separately for each symptom for predicting BHR in the free running test and methacholine inhalation challenge and for predicting clinical asthma. In addition to questionnaire validation, the sensitivities and specificities of the free running test and the methacholine inhalation challenge, as well as of the combination of the symptoms and BHR tests, were calculated for predicting clinical asthma. The estimates of the sensitivity and specificity were adjusted for the sampling scheme to avoid verification bias.¹⁵ In addition, the Youden's index was calculated for each screening method (symptoms, free running test, methacholine inhalation challenge) or their combination in order to serve as an overall indicator of validity.^3,6

Demographic data and baseline spirometric values

The screening and diagnostic groups did not differ from each other by age, but there were more boys and more atopic individuals among the asthmatic children. In the baseline spirometric tests only MMEF in the asthmatic children was, on average, lower than in the other children (table 1⇓).

Free running test and bronchial lability index

The children who reported asthma and/or wheeze in the screening questionnaire had significantly larger ΔFEV₁(%) in the free running tests than the non-symptomatic children (table 1⇑). In fact, the children who did not report asthma or any respiratory symptoms in the screening questionnaire did not have marked (>15%) reductions in FEV₁. Interestingly, two children with cough as the only respiratory symptom had clear responses in the free running tests. Eleven (16.2%) children who reported asthma and/or wheeze, six (8.0%) with dry cough at night, and two (2.9%) of the children who did not report any respiratory symptoms in the screening questionnaire had pathological responses (≥10% fall) in the free running test (p=0.023).

Most children with clinical asthma had normal free running test results and only a few children had large responses. Twelve (35.3%) children with clinical asthma, three with possible asthma (9.7%), and four controls (2.7%) had a pathological response in the free running test (p<0.001). Using a 10% cut off point for the bronchial lability index (%), 12 children with clinical asthma (35.3%), three with possible asthma (9.7%), and seven controls (4.8%) had pathological responses (p<0.001).

Methacholine inhalation challenge

The methacholine inhalation challenge was pathological in 16 children who reported asthma and/or wheeze (23.5%), in five with dry cough at night (6.7%), and in two who did not report any respiratory symptoms (2.9%) in the screening questionnaire (p<0.001). Similar results were obtained when the methacholine challenge results were compared between the diagnostic groups; a pathological result was found in 16 children with clinical asthma (47.1%), in five with possible asthma (16.1%), and in two controls (1.4%) (p<0.001). In addition, mild to moderate reactions (PD₂₀FEV₁ 400–4900 μg) were common in children with clinical asthma, so that only five of these 34 children (14.7%) did not reach the 20% fall in FEV₁ during the methacholine challenge—that is, their PD₂₀FEV₁ was >4900 μg.

Association between free running test and methacholine challenge results

The results of the methacholine challenge (PD₂₀FEV₁) and of the free running test (ΔFEV₁(%)) correlated poorly although statistically significantly with each other (Spearman's r=0.31, p<0.0001). The correlation tended to be higher in the children with clinical asthma (r=0.37), intermediate in children with possible asthma (r=0.25), and lowest in control children (r=0.12). There was moderate agreement between the pathological results of the free running test and the methacholine challenge (κ=0.42, p<0.0001; table 2⇓).

Added value of BHR tests in predicting clinical asthma

We were able to predict asthma almost as accurately by the symptom history alone—that is, the symptoms reported in the screening questionnaire—as if we supplemented the decision by the results of the BHR tests (fig 2⇓). However, all of the selected variables significantly contributed to the model as assessed by the likelihood ratio test (data not shown).

• Download figure
• Open in new tab
• Download powerpoint

Figure 2

Receiver operating characteristic (ROC) curves for the multivariate models relating clinical asthma to the questionnaire data (Q) or to both the questionnaire and BHR data (Q & C). The areas under the ROC curves were 0.895 (Q) and 0.936 (Q & C), respectively. Variables included in the models are given as footnotes in table 3⇓; model Q, symptoms 1–4; model Q & C, symptoms 1–4 and BHR tests 5–6.

Sensitivity and specificity analysis

Individual questions on shortness of breath, wheezy chest, and dry cough at night were highly specific but not very sensitive for the pathological free running test and methacholine challenge results, so the corresponding Youden's indices ranged between 0.16 and 0.46. In contrast, these questions had a relatively high sensitivity and specificity for clinical asthma with a Youden's index of 0.63–0.75 (table 3⇓).

Both the free running test and the methacholine inhalation challenge were equally specific for clinical asthma, but the methacholine challenge had higher sensitivity. The sampling adjusted positive and negative predictive values of the free running test for clinical asthma were 0.28 and 0.98, respectively; the corresponding values for the methacholine challenge were 0.35 and 0.98. The combination of the free running test or methacholine challenge with the symptom of “wheezy chest apart from colds in the past 12 months” gave almost complete specificity for clinical asthma, but the sensitivity was low. The Youden's index was therefore only 0.24–0.37 (table 3⇑).

Definition of clinical asthma

There is no “gold standard” for the diagnosis of asthma in children. In our study subjects had to have both subjective symptoms and objective evidence of variable airway obstruction to be defined as having asthma.¹⁰ We therefore believe that our clinical definition was highly specific for asthma, as suggested also by the ROC analysis. To see whether this rather strict definition led to underdiagnosis we followed the children with “possible asthma” for 2 years.¹⁹ About one third of these children continued to have symptoms and showed objective evidence of asthma during the follow up, one third continued to have symptoms with no objective evidence of asthma, and one third grew out of their symptoms.¹⁹

The first third probably represented children who actually had mild asthma in the baseline study but in the cross sectional setting we were unable to demonstrate BHR because of the fluctuating course of the disease. When these nine children were included in the clinical asthma group in the analyses the results remained very similar. The areas under the ROC curves were only slightly though constantly lower than in the original models (see fig 2⇑); 0.86 for the model with symptoms alone and 0.91 for the model with both symptoms and BHR test results.

Clinical implications

When our results are considered from the primary care perspective, it seems that in most cases the diagnosis of asthma could be made with reasonable accuracy by taking a careful symptom history. A diagnosis of asthma made by trained primary care physicians can considerably shorten the delay in starting anti-inflammatory therapy.⁵ Although our results stress the role of symptoms in the diagnosis of asthma, sensitive and specific tests would obviously be beneficial to reach an objective diagnosis in clinical practice. Such tests would be especially useful in the evaluation of children with atypical symptoms or in the determination of asthma severity. The objective evidence of asthma would also be helpful in motivating the patient to undergo long term treatment, or it may be required for receiving social security and other benefits.

Our results show that in most cases the current BHR tests cannot provide such evidence. In order to avoid delays in the diagnosis of asthma, children with an obvious symptom history but no objective evidence of BHR should therefore be carefully followed and anti-inflammatory medication should be started when necessary.⁵ Although repeated BHR tests may help in reaching the diagnosis,¹⁹ the development of more sensitive and specific diagnostic tests²⁰ such as sputum biomarkers or measurement of NO in exhaled air may assist in identifying asthma cases at an early phase of the disease.

Epidemiological implications

Our results on the validation of the written symptom questionnaire confirm and expand previous findings.^3,6,8 BHR alone, or in combination with the symptom history, was found to be a very insensitive measure of asthma. Similar findings have been reported by Jenkins et al⁷ who compared an interview based physician's diagnosis of asthma with a symptom questionnaire and the results of a hypertonic saline challenge test and found that the symptoms alone had the best agreement with the physician's diagnosis. BHR alone, or in combination with symptoms, had a high specificity but low sensitivity, whereas the questionnaire had both a reasonably high sensitivity and specificity.⁷ On the basis of our present results and the results of Jenkins et al⁷ we disagree with the suggestions that symptom history should be combined with BHR when defining asthma in epidemiological studies.^2,21 Such suggestions have also been criticised in the study of Demissie et al.¹⁸ They used the Bayesian approach for sensitivity and specificity analysis and found that the exercise challenge test was clearly inferior to the symptom questionnaire in identifying asthmatic children in a community setting.

In addition to our study, only a few other studies have validated their symptom questionnaires by BHR tests or clinical assessment³ including two that have validated the currently used questionnaires in the ISAAC study.^6,8 Both of these studies used only one type of BHR test on a rather small random population sample resulting in a small number of children with asthma. Our study had a relatively large source population from which we sampled symptomatic and non-symptomatic children into the clinical study. This kind of sampling selects more asthmatic children into the study, but it still retains the benefits of a population based sample.³ On the other hand, it is necessary to adjust the results by the sampling scheme to avoid verification bias.¹⁵ Another strength of our study was that we performed two different BHR tests and a careful clinical evaluation to differentiate “true” clinical asthma cases from the population of symptomatic and asymptomatic children. Taking the validation results together, our study gives further support to the evidence that use of a symptom questionnaire is a highly valid method for studying the prevalence of asthma in epidemiological studies.^3,7,17,18

Free running test and methacholine inhalation challenge

In our study the free running test was specific for asthma and profound responses in terms of ΔFEV₁(%) were found only among the children with clinical asthma. The few studies reporting false positive reactions in free running tests have used PEF measurements in the response follow up instead of FEV₁.^22,23 On the other hand, the sensitivity of the free running test was low. Previous studies have suggested that exercise challenge is useful in the characterisation of asthmatic children and in the differential diagnosis of asthma from other lung diseases.^24–,²⁶ Some authors have even suggested the free running test as a screening tool for undiagnosed asthma,²⁷ but its usefulness is certainly limited by its minor additional value after symptoms are known and the low sensitivity, as shown in our study and in some earlier studies.^28,29

The methacholine inhalation challenge seemed to be almost as specific for asthma as the free running test but it was more sensitive, which is in accordance with previous observations.^3,4 Similarly, our findings support the previously reported complexity in the association between the responses of asthmatic children to physical and pharmacological stimuli.³⁰ Pathological responses to both exercise and methacholine were rare, indicating that these tests reflect different pathophysiological mechanisms in the asthmatic airways. The high specificity of the methacholine inhalation challenge in our study was probably due to the rather strict criterion for a positive test result (PD₂₀FEV₁ ≤400 μg).

In summary, our study showed that a careful symptom history still forms the most important basis for defining asthma in both clinical and epidemiological settings. Tests of BHR give only marginal “added value” to the diagnostic accuracy based on symptom history. We therefore emphasise that the diagnosis of childhood asthma should not be overlooked in cases with an obvious symptom history but no objective evidence of BHR. We conclude that a written symptom questionnaire is a valid method for studying the epidemiology of asthma, and BHR should not be required for defining asthma in epidemiological studies.