The Accuracy of Patient History, Wheezing, and Laryngeal Measurements in Diagnosing Obstructive Airway Disease FREE

Despite the central importance of the initial clinical examination in the care of patients, its elements have rarely been subjected to rigorous evaluation. The evaluation of obstructive airway disease (OAD) is a typical example: a systematic review of the literature¹ identified 29 articles evaluating a total of 32 clinical signs for the detection of OAD (median of 1 sign, 2 clinicians, and 93 patients per study). However, only 1 of these studies² fulfilled standard criteria³ for classification as a methodologically rigorous study (an independent, blind comparison with a reference standard among an appropriate spectrum of consecutive patients). In that study, 2 physicians examined 164 consecutive patients in a preoperative evaluation clinic. They reported likelihood ratios (LRs) for several elements of the clinical examination, but none of the maneuvers was sufficiently sensitive to allow their absence to rule out OAD or sufficiently specific for their presence to rule in OAD.

The reported accuracies of commonly cited signs for OAD vary greatly between studies. For example, the detection of wheezing on auscultation has been evaluated in 7 studies: sensitivity ranged from 9% to 100%, specificity from 37% to 100%, and positive LRs varied from 0.9 to infinity.¹ Even the accuracy of the overall clinical impression (formed after obtaining complete patient history and conducting physical examination) for the detection of OAD is unclear, with sensitivity (50%-64%), specificity (64%-93%), and positive (1.4-7.3) and negative (0.4-0.8) LRs varying sharply between studies.¹ This situation led to calls in THE JOURNAL^{4 - 5} for larger, better studies of the clinical examination.

In an effort to obtain reliable information on the accuracy of the history and physical examination in diagnosing OAD, a multinational study involving investigators at various levels (primary, secondary, and tertiary care) was designed. In this study, the accuracy of several elements of the clinical examination in predicting OAD were investigated: patient self-reported history of chronic OAD, smoking history (yes/no, number of pack-years), wheezing on auscultation, laryngeal height (maximum and minimum), and laryngeal descent. A secondary objective was to assess whether it was possible to do large, fast, multicenter studies of the clinical examination using the Internet for clinician recruitment and data collection.

Investigators were recruited from various centers around the world via the Internet using the study group Web site (http://www.carestudy.com) and the evidence-based health care e-mail discussion group. All investigators joined the study in groups of 2 or more (at least 1 clinician and 1 spirometrist) and took responsibility for obtaining local ethics approval for the study. Investigator enrollment and data entry were done via a secure Internet-based data entry system, and data collation and analysis were done at the Centre for Evidence-Based Medicine at the University of Oxford in England.

Twenty investigator groups (46 investigators) enrolled consecutive patients (from November 3 to December 4, 1998) within 3 broad categories: patients who were known to have chronic OAD, patients who were suspected of having OAD, and patients who were neither known nor suspected of having OAD. Investigators were asked to enroll a minimum of 4 consecutive patients from each category. Known chronic OAD was defined as prior pulmonary function test results demonstrating forced expiratory volume in 1 second (FEV₁) less than the fifth percentile, FEV₁–forced vital capacity (FVC) ratio less than the fifth percentile, or FEV₁-FVC ratio less than 0.7; or patient self-report of a prior diagnosis of chronic OAD, emphysema, or chronic bronchitis; or patient taking inhaled bronchodilators and/or inhaled steroids for long periods. A case was defined as "suspected OAD" if the patient did not fulfill any criteria for known chronic OAD but was referred for suspected chronic OAD, or if the participating clinician thought that OAD was a diagnostic possibility before the structured examination. Patients with known or suspected OAD were eligible for enrollment during exacerbations of their disease if no bronchodilator treatment was given between the clinical examination and spirometry. Excluded were patients with purely reversible airway obstruction (ie, asthma); patients with a terminal illness whose goals of therapy were confined to comfort and dignity; patients younger than 18 years; patients with respiratory distress so severe that bronchodilators could not be withheld safely until after spirometry; patients who were medically unstable from other causes (eg, acute myocardial infarction, drug overdose); and patients who were unable to cooperate for the clinical examination or spirometry (eg, impaired cognition, level of consciousness, or language).

All patients underwent clinical examination and independent, blinded spirometry. The items chosen for this study were based on a review of the literature and consensus among the investigators. The items assessed included self-reported history of chronic OAD, smoking history, laryngeal height (the distance between the top of the thyroid cartilage and the suprasternal notch), laryngeal descent, and wheezing. Maximum laryngeal height was measured at the end of expiration, minimum laryngeal height at the end of inspiration.^{6 - 7} The difference between the maximum and minimum laryngeal heights is the laryngeal descent. A videotape of the laryngeal examination was provided on the study Web site for investigator training. Investigators listened for wheezes during respiration over 4 standardized areas (bilateral upper and lower back).⁸ Each patient also underwent spirometry within 30 minutes of the clinical examination (without intercurrent bronchodilator use) to assess FEV₁ and FVC values. A standard protocol for spirometry was used and the better result of 2 attempts was recorded. The spirometrists and clinicians were blind to the results of the others' investigations.

Sensitivity, specificity, and LRs for each element of the clinical examination were calculated using spirometry as the gold standard (OAD was defined as an FEV₁ and FEV₁-FVC ratio less than the fifth percentile).⁹ Percentile flow rates, adjusted for age, sex, and height, were calculated using the regression equation of Crapo et al.¹⁰ Continuous measurements (age, pack-years of smoking, and measurements of laryngeal position and descent) were categorized, either according to cut points previously published or values derived from noting obvious inflection points on receiver operating characteristic (ROC) curves. Cut points were chosen such that the slopes of the ROC curve based on the selected cut points mirrored those in the full ROC curve. The relationships between each diagnostic element and OAD were tested using χ² tests and the Fisher exact test for dichotomous features, the χ² test for trend for categorical variables, and the t test for continuous variables.

Multivariate analyses were carried out using the method of Spiegelhalter and Knill-Jones (which adjusts for confounding from related diagnostic elements),^{11 - 12} and a reduced multivariate model was produced by grouping categories with similar LRs within each element and only selecting diagnostic elements with adjusted LRs greater than 2 or less than 0.5. All analyses were done using statistical software.¹³

A total of 332 patients were recruited by 25 investigator groups from 14 countries. Twenty-three patients were excluded from further analysis because they had a primary diagnosis of asthma; no other protocol violations were identified. Thus, the final sample size was 309. After the closing of the study, a subset of investigators (chosen because of outlying results) were asked to submit their original data collection sheets to be checked against the database—11 of 680 data points were incorrectly entered (error rate, 1.6%).

Patient demographics are outlined in Table 1 and the distribution of FEV₁ values are illustrated in Figure 1. On objective testing, more than half (162 [52%]) of the patients had an FEV₁ and FEV₁-FVC ratio less than the fifth percentile. The accuracy of the various elements of the clinical examination assessed is outlined in Table 2. Whereas we used FEV₁ and FEV₁-FVC ratio less than the fifth percentile as the reference standard, given the controversy regarding the spirometric definitions of OAD,⁵ the LRs for each element of the clinical examination also were calculated using other reference standards (FEV₁-FVC ratio <0.7, FEV₁-FVC ratio <0.8, FEV₁ value < the fifth percentile, or FEV₁-FVC ratio <the fifth percentile alone). The accuracies reported in Table 2 did not change appreciably: for example, the positive LRs for wheezing were 1.9, 3.1, 2.6, and 2.1, respectively, using the alternate reference standards. The mean minimum and maximum laryngeal heights were significantly smaller in patients with FEV₁ values and FEV₁-FVC ratio less than the fifth percentile than those with higher FEV₁ values (3.8 vs 4.6 cm, P<.001, and 5.4 vs 6.3 cm, P<.001). Laryngeal descent was not significantly associated with OAD diagnosis, even when the analysis was restricted to subgroups of patients with more severe obstruction (data not shown). There was no heterogeneity in accuracy across countries, investigator groups, examiner experience), or point of contact (primary or secondary/tertiary care). Moreover, the accuracy of the tested elements was similar even after exclusion from analysis of the 76 patients with known chronic OAD (Table 2).

The ROC curve for smoking demonstrated that the most appropriate cut point was at 40 pack-years (data available on request).

The reduced multivariate model included 4 items (Table 3). In patients with all 4 items (self-reported history of chronic OAD, smoked >40 pack-years, older than 45 years, and maximum laryngeal height ≤4 cm), the LR for the diagnosis of OAD is 220 (essentially ruling in the diagnosis). In patients without any of these 4 characteristics, the LR is 0.13 (essentially ruling out the diagnosis). A multivariate model derived from the 233 patients without known chronic OAD included the same 3 items (Table 3). In particular, wheezing and laryngeal descent did not enter the model even after exclusion of known chronic OAD patients.

We evaluated the accuracy of several elements of the clinical examination in diagnosing OAD. In terms of history, the most useful points to rule in a diagnosis of OAD are self-reported history of chronic OAD and smoking in excess of 40 pack-years. Age younger than 45 years virtually ruled out the diagnosis of OAD (given that patients with asthma were excluded). On physical examination, auscultated wheezing and maximal laryngeal height of 4 cm or less increased the likelihood that OAD was present but did not do so sufficiently to resolve the diagnostic process. For example, in a patient with a prior likelihood of 10%—the prevalence of chronic OAD among smokers¹⁴ —the presence of wheezing raises the probability of OAD to only 23% (similarly, a maximum laryngeal height of ≤4 cm only increases the probability to 28%). Laryngeal descent was not helpful in either ruling in or ruling out the diagnosis of OAD. Although it may seem tautologous to include "history of chronic OAD" in a prediction rule for OAD, it must be acknowledged that clinicians usually collect history prior to physical examination or further diagnostic testing, and thus it is important to evaluate the accuracy of this element of the clinical assessment. Furthermore, in testing the accuracy of a symptom or sign, individuals representing a full spectrum of disease should be included. Finally, the accuracy of the tested elements did not change even after exclusion of patients with known chronic OAD, and inclusion of this factor more closely reflects actual practice.

Using multivariate analysis, we developed a 4-variable model for diagnosing OAD. The LRs for each of these variables can be multiplied (as they are adjusted to account for their nonindependence) to generate an LR for an individual patient.^{11 - 12} For example, in a 65-year-old patient with self-reported chronic OAD, a 45-pack-year smoking history, and a maximum laryngeal height of 3 cm, the LR is 220. Thus, even if the pretest probability was only 10%, the constellation of symptoms and signs increases his/her posttest probability to 96%. Although this may obviate the need for spirometry for diagnostic purposes, it does play a useful role in identifying the severity of disease and the effects of therapy.

Our study adds substantially to the literature on the rigorous evaluation of the clinical examination for OAD (it triples the numbers of patients in such studies and increases the numbers of clinicians 10-fold). Furthermore, our findings are generally consistent with the literature. For example, Badgett and colleagues¹⁴ found that the only useful items on history were self-reported history of chronic OAD (positive LR, 3.1) and smoking more than 70 pack-years (positive LR, 8.0). Others⁵ have reported, as we did, that a history of never smoking significantly decreases the likelihood of OAD, but the negative LR is insufficient to allow the diagnosis to be definitively ruled out. Although our study contradicts previous studies that suggested increased tracheal descent was a useful sign in identifying OAD,^{7 ,15} these were unblinded studies of nonconsecutive patients and thus subject to potential selection and measurement bias. In our study, the presence of wheezing was not as useful in diagnosing OAD as other investigators have reported it to be. In their study of 164 patients, Holleman and colleagues² reported a positive LR of 12 for wheezing; however, the 95% confidence interval ranged from 1.7 to 98. Our results are consistent with this estimate and, given our larger sample size, serve to refine the previously published estimates. The ability of our model to predict OAD is similar to previously published models (one² incorporated years of smoking exposure, patient-reported wheezing, and auscultated wheezing, and the other¹⁴ incorporated pack-year smoking history, self-reported history of chronic OAD, and decreased breath sounds); however, our model is somewhat better at ruling out OAD than were those models. For example, in a 40-year-old nonsmoking patient without a prior diagnosis of chronic OAD and with a maximum laryngeal height of 7 cm, the LR of 0.13 virtually rules out the diagnosis of OAD.

This study established the feasibility of using the Internet to recruit investigators and conduct studies of the clinical examination. This approach allowed the rapid accrual of patients to this study (at a rate more than 20 times faster than that of the only other methodologically rigorous study² of the clinical examination in OAD) and has resulted in the development of a practice-based research network^{16 - 17} of clinicians interested in performing other studies of the clinical examination.

However, there were some limitations to our study. First, we did not assess interrater reliability. This was a deliberate exclusion, as the primary focus of this initial study was to evaluate the accuracy of elements of the clinical examination and prove the feasibility of the study design. To achieve these objectives, the study was designed such that data collection would be brief. We decided a priori to defer assessment of interobserver variation for a future study in which we will only evaluate those signs that have been shown to be accurate. Second, to participate, investigators had to have access to the Internet. Critics might be concerned that this may affect the applicability of our results to patients or clinicians in other settings. However, our patients and results are similar to those in other studies,^{5 ,14 - 15} suggesting that our findings are generalizable. Third, the applicability of our model for clinical practice has yet to be determined, although the preliminary observations from our data set suggest it holds significant promise. If it had been used to assess the patients in our study, a diagnosis (ie, probability of OAD >90% or <10%) could have been made and spirometry for diagnosis avoided in 48% of them. Furthermore, we derived LRs for the tested elements of the clinical examination (rather than sensitivity or specificity) to permit the ready extrapolation of our results to other settings with different prevalence of disease. Finally, our model must be validated in an independent sample of patients, and such a study is being planned.¹⁸

In summary, our results suggest that less emphasis should be placed on the presence of wheezing or exaggerated laryngeal descent in making a diagnosis of OAD. We found that a combination of 4 symptoms/signs (self-reported history of chronic OAD, pack-year smoking history, age, and maximum laryngeal height) can be used to predict airway obstruction. In those settings in which spirometry is readily available, it should be used because it takes only slightly longer to do than the clinical examination, definitively establishes the diagnosis of airway obstruction, and provides prognostic information. However, in those settings in which spirometry is unavailable, our model provides useful diagnostic support for the clinician. Future studies are under way to evaluate other signs and symptoms that have been described for the diagnosis of OAD and to test our model in an independent sample of patients.