- Email: [email protected]
A Discriminant Analysis Applied to Methacholine Bronchoprovocation Testing Improves Classification of Patients as Normal, Asthma, or COPD
A Discriminant Analysis Applied to Methacholine Bronchoprovocation Testing Improves Classification of Patients as Normal, Asthma, or COPD*
ue W Greenspon, M.D., F.C.C.R;t and Edward Gracely, Ph.D.t Two discriminant functions, incorporating baseline measurements of pulmonary function and measures of airway responsiveness, were developed to improve patient classi6cation into groups of normal, asthma, or COPD. Accuracy of group classi&cation was compared between the usual laboratory method (single discriminating cut-ot1) and these new mathematically developed functions. Forty-Bve normal subjects, 27 asthmatic patients, and ten well-deBned COPD patients were entered into the analysis. Measurements of airway responsiveness were determined by measurement of both speci6c airway conductance (SGaw) and spirometry (FEV.) after sequential inhalation of methacholine. Beaulta: A single discriminant cut-ofT using measures of SGaw (PD35) or FEV. (PD20) does not sufficiently discrim-
Airway hyperresponsiveness is a distinguishing featore of asthma'> and the mechanisms involved in inducing airway hyperreactivity are active areas of research. Yet, laboratory bronchoprovocation tests using methacholine or histamine for diagnostic purposes in the evaluation of subclinical asthma have been questioned in their ability to adequately distinguish asthma from normal'-" and to predict which patient may significantly benefit from bronchodilator therapy 5 Histamine and methacholine bronchoprovocation tests that have been developed have shared a high sensitivity but a lower specificity in distinguishing subjects with asthma from normal subjects. The distribution of airway responsiveness in the normal population dictates that a simple cut-off point will include a significant number of false/positive test results," Bronchoprovocation agents like cold arr,7,8 exercise," distilled water," or hypertonic saline solution" have shown lower sensitivity but greater specificity than pharmacologic agents in distinguishing subjects with asthma from normal subjects. Although uniform guidelines have been recommended by the American Thoracic Society" and the society ofAllergy and Clinical Immunology, 13 various techniques are currently employed in clinical laboratories throughout the world. Since methodology is critical to the interpretation of test results, many laboratories have developed their own laboratory standards.
d
-From the Department of Pulmonary and Critical Care Medicine, the Medical College of Pennsylvania, Philadelphia. t Associate Professor of Medicine. *Assistant Professor of Community and Preventive Medicine. Manuscript received April 1, 1991; revision accepted March 13.
inate asthma from groups that contain normal and COPD subjects (67 to 71 percent predictive value). On the other
hand, our discriminant functions demonstrated improved patient classi6cation (positive predictive value, 88 to 89 percent). We conclude that bronchoprovocation tests used to evaluate the diagnosis of asthma should incorporate measures of baseline lung function into the analysis. This, we believe, is especially necessary when baseline lung function demonstrates minimal airOow obstruction and the possibility of other causes of airway disease exist. (Chat 1992; 102:1419-25)
I
CDU = cumulative dose units
I
The limitations of clinical bronchoprovocation testing may be due to several reasons: (1) there is a continuum of bronchial hyperresponsiveness in the normal population that overlaps with the asthmatic population such that a fully adequate discriminating cut-off is not perceptible. 3,5 (2) Airway responsiveness changes with age" and, perhaps more importantly, 15-18 with baseline airway caliber, which is not taken into account in the analysis. (3) Other diseases, such as COPD or bronchiolitis, which affect baseline airway caliber, will result in increased airway hyperresponsiveness I9 •20 even when patients may not respond clinically to bronchodilator therapy (4) Patients referred for testing with atypical chest complaints of cough, wheezing, or shortness of breath often have test results somewhere between the normal and asthmatic range (gray zone). These are difficult cases to classify and may represent mild asthma as well as early COPD, bronchiolitis, or within the expected range of normal. Over the past five years, we have collected data on airway responsiveness to methacholine in a population of normal men, as well as asthmatic and COPD patients. 2 1•22 As others have reported, we found a significant overlap between the normal and asthmatic population. COPD patients demonstrated a broad range of airway response from normal to asthmatic which is very dependent on baseline lung function. The purpose of this communication is to demonstrate that a mathematically developed discriminant function applied to tests of airway responsiveness that incorporate baseline lung function improves patient classification as normal, asthma, or COPD compared CHEST I 102 I 5 I NOVEMBER. 1992
1419
with a single cut-off level. Since one third of patients referred for bronchoprovocation testing to our laboratory have minimal airflow obstruction on baseline pulmonary function tests (yet FEV l greater than 70 percent), this type of analysis may help decrease the false/positive rate of bronchoprovocation testing, increasing the accuracy in recognizing asthma and a subset of patients that symptomatically improve with bronchodilator therap}'. METHODS
As part of other studies ongoing in our laboratory over the past five years, we have collected baseline pulmonary function data and airway responsiveness to methacholine data on 45 normal, 27 asthmatic, and 10 COPO subjects. Normal Subjects
Most normal subjects were recruited from an ongoing study evaluating age and airway responsiveness (pilot of NIA grant AG03934). Twenty-two subjects older than 60 years were physically active and screened to be free of medical illness. The remaining normal subjects were recruited from the hospital/medical school staff. Six normal subjects had smoked cigarettes but all had quit more than ten years prior to testing, and no subjects had smoked greater than ten pack-years. On questionnaire, no normal subjects had a history of allergic rhinitis, sinusitis, or asthma. Asthmatic Subjects
Allasthmatic subjects were recruited to be part ofother laboratory studies evaluating asthma. These patients were for the most part PRN users of inhaled bronchodilators. Ten subjects had been receiving no therapy for two weeks prior to testing. Nine subjects were regularly using inhaled sympathomimetics. Five patients were receiving a sustained-release theophylline preparation and three subjects had used steroids intennittently in the year prior to the study. All subjects gave a history typical of asthma and had histories of acute exacerbation that were relieved by bronchodilator therapy. COPD Subjects
Ten COPO subjects were recruited as part of a prior studya and selected to represent a wide range of objective functional impairment. AU subjects demonstrated airflow obstruction by pulmonary function tests. The diffusing capacity was reduced (53 ± 18 percent predicted), suggesting that our subjects probably had a signi6cant component of emphysema. In the COPO group, at the time of study, eight patients were receiving inhaled bronchodilators and five patients were receiving sustained theophylline. Three of ten subjects demonstrated a greater than 15 percent improvement in FEV. after inhaling isoproterenol 200 IJ.g, but this did not correlate with tests of airway responsiveness as previously reported." Since COPO subjects are not usually tested by bronchoprovocation testing, no other subjects were available for this analysis.
Testing Methods Static pulmonary function tests were performed on one of two spirometers (Gould 4000 rolling sealed spirometer or an Ohio 840 spirometer interfaced in line with an Apple 2E computer). Nonnal values for spirometry were from Crapo et alJ3 with a 10 percent correction for race applied to black individuals. The Deo predicted equation was based on data from Knudson et al. 1U Bronchoprovocation studies were performed according to a standard laboratory protocol. All subjects were free from upper respiratory tract infections for at least six weeks prior to the study Patients were instructed to discontinue oral theophylline therapy for 48 h and inhaled bronchodilator therapy for 12 h prior to 1420
methacholine testing. Only patients in stable conditions were entered into the study. Baseline measures of airway resistance and specific airway conductance were measured in a constant volume body plethysmograph (Warren E. Collins, Braintree, Mass). FEV. and FEF25-75 were recorded on a volume displacement spirometer. Subjects inhaled five breaths from FRC to TLC from a nebulizer (DeVilbiss 646, DeVilbiss Company, Somerset, Pa) powered by a dosimeter, (French-Rosenthal) of sequentially increasing concentrations of methacholine. Concentrations of 0.1, 0.5, 1.0, 5.0, 10.0, and 25.0 mg/ml were employed. Specific airway conductance (SGaw) and then FEV. were measured 3 min after each sequential dose until the FEV. fell by 20 percent. If there was no signi6cant fall in FEV. when the highest dose was inhaled, this concentration was administered a second time. The dose of methacholine that caused specific airway conductance to decrease by 35 percent (PD35) and the FEV. to fall by 20 percent (PD20) was recorded in each subject. We report for each test the log cumulative dose units of methacholine inhaled where one breath of 1 mg per milliliter methacholine = 1 dose unit. If after the highest dose of methacholine inhaled (333 CDU) there was no signi6cant response, a ceiling valve of >330 was used. In two subjects, a third inhalation of 25 mg/ml was delivered since they were perceived to be close to the physiologic end point. Single Discriminant Cut-off to Determine Ainooy &sponsiveness
A discriminant line best separating asthma from normal was generated by visual inspection. A separate cut-off was determined for values calculated using airway conductance (PD35) and for values utilizing spirometry (PD20). Patients with COPO were then evaluated as either matching the pulmonary and bronchial reactivity characteristics of the asthmatic" or "normal" groups by this scheme. U
Multiple Unear Dtscriminate Analysis
Linear discriminant analysis is a method for mathematically obtaining the "best" discrimination among groups of subjects using
one or a set of variables.· It is closely related to such parametric techniques as multiple regression and the multivariate analysis of variance. Like them, it assumes that the data are normally distributed and that relationships among variables are adequately captured by a linear (straight-line) function. For our analysis, we assumed an equal prevalence of each condition (asthma, COPD, or normal) occurring. We used raw numbers for all but the provocation tests for which logarithmically transformed values were used. Finally, we entered variables in a nonstepwise fashion. Six variables were entered into the analysis: log PD35, log PD20, FEV., FEV. predicted, and FEF percent predicted, SGaw. Since many laboratories throughout the country use only spirometry to measure airway responsiveness, a simplified discriminant function relying on only three variables was also analyzed (log P020, FEV. percent predicted, and FEV25-75 percent predicted). Statistics
Descriptive statistics of subject characteristics at baseline were expressed as mean± 1 SO. Groups were compared by analysis of variance (ANOVA) followed up with a version of the Tukey post hoc test for unequal sample sizes.· Differences were considered significant with p value <0.05. Correlations of airway responsiveness (P035 and P020) and other tests of airway function were analyzed for each group independently. Spearman's rank correlation was used for this analysis. Since many normal subjects exceeded the highest dose of methacholine inhaled in our laboratory, an upper limit of 330 units was recorded in these individuals. Unfortunately, this ceiling method was biased against our finding signmcant correlations in the normal population between lung function and airway responsiveness. To compare the results of the single discriminant cut-off method with the multiple discriminant functions, test sensitivity, specificity, Me1h8choIine 8ronctloprow)ca Testing (Greenspon, GraceIy)
Table I-Subject CharacteriBtica Normal
Asthma
COPD
45
27 29.7±8.1 3.46 ± 0.89 93±13* 80±5 85±16* 0.19±0.05 0.59±0.5O* 1.33 ± 0.63*
10 57.9±9.7*t 1.65±0.64*t 62±21*t 58±lO*t 27± 10*t 0.12±0.OO*t 1.38±0.75*t 1.43±0.83*
n Age, yr FEVI,L FEVI , %pred FEV/FVC FEF25-75, % pred
SGa\1v, em:"
S-I
Log PD3Ih CDU* Log PDIO , CDUI
49.2±18.4 3.54±0.89 105± 14 81±7 102±25 0.20 ± 0.04 2.00±47 2.45±0.18
*p
Subject characteristics by group classifications are listed in Table 1. Patients with COPD differed significantly on all variables from the normal group and by all variables except log PD20 from the asthma subjects. The asthma subjects differed from the normal group by all variables except FEV. (L), FEV.IFVC, and 3
2.5
2
-.... --- - -----
log P035
1
.5
0
3
2.5
--
2.25
..
1.5 ____B._________A 1.4
SGa~ As expected, the asthma group had the greatest degree of airway hyperresponsiveness (lowest PD35). The scatter plot in Figure 1 demonstrates the range of airway responsiveness by log PD35 and log PD20 for each group. The hatched line was visually selected as the best single discriminating point separating asthma from normal for each variable. In our laboratory, PD35s25 CDU (log PD35s1.4) or PD20
• •
2
•
• • •
• ••• •
log
PDm
- -
--- -
--~-------------------------
• • •
1.5
__________
• ••
...-
•
••
• •
•
•
1
•
.5
•
0
•
• -.5
-.5
• -1
-1
Normal
Asthma
COPO
Normal
Asthma
COPO
FIGURE 1. Airway response to methacholine by group. Results are presented as the dose of methacholine in cumulative dose units (CDU) necessary to cause a 35 percent reduction in SCaw (log PD35) or a 20 percent fall in FEV. (log PD20). CHEST I 102 I 5 I NOVEMBER, 1992
1421
Table 2-Correlation of A i n D G f l _ to Lung FUnction by Claa
0.13 (p=0.19) 0.38 (0.03) 0.76 «0.01)
Normal Asthma
COPD
FEV., % Pred
FEVlFVC
0.23 (0.06) 0.3 (0.01) 0.83 (0.<01)
0.22 (0.01) 0.33 (0.05) 0.47 «0.08)
FEF% Pred 0.23 (0.06) 0.49 (0.01) 0.7 (0.01)
SGaw
0.33 (0.01) 0.46 «0.01) 0.36 (0.15)
*Airway responsiveness is expressed as log PD35; similar results were found using log PD20.
group. Six variables were used to develop two canonical discriminant functions (see Table 3) which when plotted (Fig 3) best separate the three groups. An alternative method of classification is to apply a classification function (Fischer's linear discriminant function coefficient) which is provided in Appendix A. Using this discriminant analysis, 42/45 normal subjects were correctly classified, two were classified as "asthma;' and one was classified as "COPD." Of 27 asthmatics, 25 were correctly classified and two were rated as "normal," Most importantly, 100 percent of COPD subjects were correctly classified. More than 94 percent ofall subjects were appropriately classified. Since many laboratories do not make measures of SGa~ or calculate PD35, a simplified discriminant function using the three variables log PD20, FEV. percent predicted, and FEV25-75 percent predicted was developed (see Appendix B). By this method, 91 percent ofthe study population was correctly classified 3
• .•_.... ••
2.5
•• • • • •
2
•
•
15
• 5
(43145 normal, 23/27 asthma, 9/10 COPD). Table 4 summarizes the sensitivity, specificity, and predictive values for each method of analysis. DISCUSSION
By our laboratory method ofmethacholine bronchoprovocation testing, we have visually determined a single discriminant cut-off separating normal from asthma using both measures ofspirometry and specific airway conductance. Our PD20 cut-off for asthma of 180 CDU of methacholine is in keeping with prior reports in the literature. 28.29 Our PD35 cut-off for asthma of less than 25 CDU of methacholine is lower than reports in the literature. This is perhaps best explained by the fact that the normal population selected in this study was generally older than previous groups studied. Our results, however, are in keeping with other studies demonstrating greater overlap between normal and asthma with measurement of PD35 than with PD20. 30•31 Utilizing the PD20 measurement, we underclassified two asthmatic patients as normal, whereas the PD35 cut-off overclassified four normal patients as asthmatic. Hwe excluded the COPD group and only considered normal subjects with normal baseline pulmonary function tests, this analysis indicates that utilizing either of the single point cut-off values for PD20 or PD35 would be acceptable. The addition of a multiple discriminant analysis would add very little. However, in patients with COPD or those referred with chest complaints who have mildly abnormal baseline pulmonary function tests, a significant proportion will be misclassified into the asthma group by either single cut-offapproach. In our COPD group, 80 percent would be incorrectly classified as asthma using PD35 values; 60 percent would be miselassified by PD20 values. Table 3- CtJnonictJl DiacrimifItJfIt Function CoejJicienta
o
• •
-5
•• Normal
.·Asthma •• copo
-1
20
40
60
80
100
120
140
Baseline FEV, % predicted FIGURE 2. Linear regression comparing airway responsiveness to methacholine (log PD35) to one index of baselinie lung function (FEV. percent predicted) for each group. See Table 2 for statistics.
1422
FEV. FEV., %pred FEF, %pred SGavv, cm- I s-·
LogPD35 LogPD20
Function 1*
Function 2*
0.52 0.20 0.10 0.46 -0.97 -0.25
0.02 0.31 0.60 0.02 -0.02 0.50
*Variables are multiplied by the coefficients in the table after first being standardized to a mean of 0 and a standard deviation of 1 (converted to Z scores). The products are summed to find the value plotted in Figure 3. Function 1 represents the x-axis value and function 2 represents the y-axisvalue in Figure 3. Methacholine 8ronchoprow)cat Testing(Greenspon, GraceIy)
Results from the VA nonnative aging study33 suggest that there is a synergistic effect of smoking and atopy on nonspecific airway responsiveness. Yet, Burney et • 34 al have shown that the association of skin test • • 2.0 sensitivity and airway hyperresponsiveness diminishes • •• • •• with age. Since our normal group was currently ••• •• •• • • nonsmoking and somewhat older, the presence of • ••• • atopy may have had little overall effect. However, it is o •• •• •• •• •• possible that a discriminant function separating asthma • •• • • • Mil• from normal could be improved by considering other ••• • nonphysiologic parameters like age, smoking history, -2.0 ••• • IgE level, eosinophil count, and skin test reactivity. The use of a linear discriminant function with the • • • present data should be regarded as illustrative rather than definitive. This technique depends rather heavily -4.0 • •• •• Normal on the existence of a normal distribution of data within •• Asthma each group. It is not robust because individual subjects, • rather than means, are being compared and classified. -6.0 The logarithmic transformation, which often normalizes the numbers, does not do so herein because -6.0 -4.0 -2.0 o 2.0 4.0 6.0 of an artificial concentration ofP035 and P020 values FIGURE 3. Plot of the multiple discriminant functions. Normal, at 330, the upper limit in our laboratory. There is no asthma, and COPD groups can be visually separated. The x-axis technique designed to handle artificial distributions value is calculated from the sum of function 1 in Table 3. The y-axis value represents the sum of function 2, Table 3. such as this one. Nevertheless, the obtained results provide surprisingly good classifications. Of interest is the finding of positive methacholine The existence of relatively small sample sizes makes challenge test results in 65 percent of patients with both the discriminant results and the eyeball results mild chronic obstructive lung disease who have been look better than they are. The eyeball approach, in entered into the Lung Health Stud~32 The basis for a particular, capitalizes on minor fluctuations in the data positive test in this population is probably related to that will often not be present in a second sample. The the lower starting airway caliber at baseline. 16 •19 ,m discriminant analysis is slightly less subject to this Moreno et al'? have used a modeling analysis of in problem because it imposes a distribution on the data, vitro and in vivo airway responsiveness and hypothetthus limiting its ability to utilize random differences. ically demonstrate that a change in airway caliber can Nevertheless, since the imposed distribution is itself tailored to the given data, the discriminant results will totally explain a shift of the dose response curve to the left. Although direct measures of airway caliber also not generalize perfectly to another sample. Our six-variable discriminant function applied to were not made in this study; the strong linear correlation between baseline lung function and airway CO PO subjects classified 100 percent of the cases correctly into a separate group. This kind of analysis responsiveness in the CO PO group as noted in this and other studies confirms this important mechanism. can be added to the several other tests that reportedly One limitation of this study is that atopic status, will help discriminate asthma from COPD. For exwhich may contribute to airway hyperresponsiveness ample, asthmatics respond to histamine and methain the normal population, was not carefully evaluated. choline in a one-to-one relationship, but COPD paEvaluation of atopic status includes prick test skin tients show a much greater response to histamine than sensitivity, total IgE level, and eosinophil count. to methacholine. COPD subjects also fail to respond Table 4-MetluJcholine Bronchoprooocation Teating: Seruitioity, Specificity, Grad Predictive Valaa ofMethod. to 4.0
_..
._...
..copo
Identify An,",",·
Sensitivity Specificity (+ ) Predictive value ( - ) Predictive value
LogPD20
LogPD35
93 80 67
96 82
96
71 98
Discriminant Function
SimpliBedt Discriminant Function
93 95 89 97
85 95 88 94
• All values are percentages. tSpirometry values only. CHEST I 102 I 5 I NOVEMBER, 1992
1423
to inhalation of the a-agonist methoxamine." Whereas 80 percent of asthma subjects have bronchoconstriction in response to breathing cold dry air during exercise, less than 10 percent of COPD subjects will respond even when their methacholine responsiveness is in the asthma range.i" The advantage of our discriminant function is that no other test need be given beyond measurements of baseline lung function. We routinely measure both PD35 and PD20. However, failure to measure specific airway conductance would only decrease the sensitivity for recognizing COPD to 90 percent (9/10). The positive predictive value for identifying asthma remains very good (88 percent). There are several advantages of utilizing a multiple discriminant function analysis. Since a significant proportion of patients referred to our laboratory for evaluation of atypical chest symptoms have mild abnormal pulmonary function test results at baseline, this type of analysis may classify these patients as COPD rather than asthma. The COPD classification would indicate an appropriate level of airway responsiveness for the degree of airflow obstruction. It would also predict that these patients would be less likely to respond to standard bronchodilator therapy This hypothesis, however, needs further testing. Secondly, even in patients with COPD, airway responsiveness may be out of proportion to the degree of alteration in lung function. These patients may need more aggressive management to control airway inflammation since this is our current understanding for the development of airway hyperresponsiveness in asthma. Future studies evaluating the role of airway responsiveness in predicting the decline in lung function in COPD will hopefully clarify the importance of airway inflammation and hyperresponsiveness in this group of patients. We are not suggesting that this equation be adopted by all laboratories to improve patient classification. Rather, we alert clinicians to the hazards of test interpretation without relating to the baseline lung function. Each laboratory performing these tests ApPENDIX A*
Normal (1) FEV. FEV., % pred FEF25-75, % pred SGaw LogPD35 LogPD20 Constant
1.225 0.427 0.017 76.77 -3.780 5.512 -37.25
Asthma (2)
3.086 0.433 -0.014 104.2 -8.5 2.468
-34.86
COPD (3) 0.194 0.323 -0.102 61.699 -0.096 2.01 -15.01
*To classify subjects into one of three groups (normal, asthma, COPD), measures of baseline lung function and test results of airway responsiveness are entered into the following three equations. Variables are multiplied by the coefficients in the table. The products are summed and the constant (being a negative number) is subtracted. For all variables, raw whole numbers are used or log values. The equation that generates the largest summed result designates the patient's classification.
1424
,ApPENDIX
Normal (1) FEV1 , 'II pred FEF25-75, % pred LogPD20 Constant
B* Asthma (2)
0.389 0.085 5.227 -32.38
0.397 0.073 0.359
-22.86
COPD(3) 0.288 -0.041 3.385 -12.07
*This simplified discriminant function just incorporates measures of spirometry with one measure of airway responsiveness (PD20). The equation that generates the largest summed result designates the patients classmcation. For all percent, use whole numbers. The classification functions listed above are available on floppy disk (IBM compatible) upon request to the author. Example 1: A patient is studied with FEV1 =2.9 FEV1 'II pred = 93 FEF25-75 % pred=65 SGaw=0.19 PD35 (CDU) = 10 PD20 (CDU) = 75 Equation (1) Normal = 28.27 Equation (2) Asthma = 29.4 Equation (3) COPD=24.35 Therefore the patient is classified asthma. Example 2: A patient is studied with FEV. =2.5 FEV. % pred=74 FEF 'II pred = 50 SGaw=0.19 PD35 (CDU) = 20 PD20 (CDU) =75 Equation (1) Normal = 18.7 Equation (2) Asthma = 18.01 Equation (3) COPD = 19.96 Therefore the patient is classified COPD.
should attempt to study laboratory controls, including asthma and normal subjects. Laboratory measures and techniques differ with each institution such that each laboratory may need to develop its own reference standards and discriminant function. The utility of this method in evaluating clinical cases with airway responsiveness results that fall in the gray zone is under prospective evaluation. ACKNOWLEDGMENTS: The writers wish to thank Barbara Sewell for her excellent secretarial support on this manuscript and Adele Kaplan, Ph. D., for her statistical advice. REFERENCES
1 Boushey HA, Holtzman MJ, Sheller JR, Nadel JA. Bronchial hyperreactfvity Am Rev Respir Dis 1980; 121:389-413 2 Hargraeve FE, Ryan G, Thompson NC, O'Byrne PM, Latimore R, Juniper EF, et al, Bronchial responsiveness to histamine or methacholine in asthma: measurement and clinical significance. J Allergy Clio Immunoll981; 68:342-55, 347-55 3 Popa ~ Singleton J. Provocation dose and discriminant analysis in histamine bronchoprovocation. Chest 1988; 94:466-75 4 Filuk RB, Serretta L, Anthonisen NR. Comparison of response to methacholine and cold air in patients suspected of having asthma. Chest 1989; 95:948-52 5 Chbebra SK, Gaur SN, Khanna AK. Clinical significance of nonspecific bronchial hyperresponsiveness in asthma. Chest 1989; 96:596-600 6 Cockroft D~ Bercheid BA, Murdock KY. Unimodol distribution ofbroncbial responsiveness to inhaled histamine in a random Methacholine 8ronchoprtM)cat Testing (GteenspotJ, GraceIy)
human population. Chest 1983; 83:751-54 7 O'Byrne PM, Ryan G, Morris M, McCormack D, Jones NL, Morse JLC, et ale Asthma induced by cold air and its relation to non specific bronchial responsiveness to methacholine. Am Rev Respir Dis 1982; 125:281-85 8 Weiss J~ Rossing TH, McFadden BR, Ingram RH. Relationship between bronchial responsiveness to hyperventilation with cold and methacholine in asthma. J Allergy Clin Immunol 1983; 72:140-44 9 Chatham M, Bleecker ER, Smith PL, Rosenthal RR, Mason ~ Normal PS. A comparison of histamine, methacholine and exercise airway reactivity in normal and asthmatic subjects. Am Rev Respir Dis 1982; 126:235-40 10 Bascom R, Bleecker ER. Bronchooonstriction induced by distilled water. Am Rev Respir Dis 1986; 134:248-53 11 Araki H, Sly PD. Inhalation of hypertonic saline as a bronchial challenge in children with mild asthma and normal children. J Allergy Clin Immunoll989; 84:99-107 12 ATS News Spring, 1980; 11-19 13 Chai H, Farr RS, Froehlich LA, Mathison DA, McLean JA, Rosenthal RR, et ale Standardization of bronchial inhalation challenge procedures. J Allergy Clin Immunoll975; 56:323-27 14 Hopp RJ, Bentra A, Nair NM, Townley RG. The effect of age on methacholine response. J Allergy Clin Immunoll985; 76:60913 15 Sparrow D, O'Conner G, Colton T, Barry CL, Weiss ST. The relationship of non-specific bronchial responsiveness to the occurrence of respiratory symptoms and decreased levels of pulmonary function. Am Rev Respir Dis 1987; 135:1255-60 16 Rijcken B, Schoten J~ Weiss ST, Speizer PE, Vanderlende R. The relationship between airway responsiveness to histamine and pulmonary function in a random population sample. Am Rev Respir Dis 1988; 132:826-32 17 Moreno RH, HoggJC, Pare PD. Mechanisms of airway narrowing. Am Rev Respir Dis 1986; 133:1171-80 18 James AL, Pare PO, Hogg JL. The mechanisms of aiJway narrowing in asthma. Am Rev Respir Dis 1989; 139:242-46 19 Ramsdell ~ Nachtwey FJ, Moser KM. Bronchial hyperreactivity in chronic obstructive bronchitis. Am Rev Respir Dis 1982; 126:829-32 20 Yan K, Solome CM, Woolcock AJ. Prevalence and nature of bronchiole hyperresponsiveness in subjects with chronic 0bstructive pulmonary disease. Am Rev Respir Dis 1985; 132:259
22
23
24 25 26 27 28
29
30 31
32 33
34
35
36
inhibition of mechacholine induced bronchooonstriction. Am Rev Respir Dis 1986; 133:735-39 Greenspon L~ Parrish B. Inhibition of mechacholine induced bronchoconstriction in patients with chronic obstructive pulmonary disease. Am Rev Respir Dis 1988; 137:281-85 Crapo RO, Morris AH, Gardner RM. Reference spirometric values using feelings and equipment that meet ATS recommendation. Am Rev Respir Dis 1981; 123:659-64 Knudson RJ, Kaltenborn WT, Knudson DE, Burrows B. The single breath carbon monoxide diffusing capacity. Am Rev Respir Dis 1987;135:805-11 Sokal RR, RohlfEJ. Biometry. New York: WH Freeman and Co, 1969:488-90 Jaccard J, Becker MA, Wood G. Pairwise multiple comparison procedures: a review Psychol BuD 1984; 96:589-96 Pratter MR, Hingston DM, Irwin RS. Diagnosis of bronchial asthma by clinical evaluation. Chest 1983; 84:42-7 Hopp JR, Bewtra AR, Nair NM, Townley RG. Specificity and sensitivity of methacholine inhalation challenge in normal and asthma children. J Allergy Clio Immunoll984; 74:143-48 Townley R, Bewtra A, Nair N, Bradley F, Walt G. Methacholine inhalation challenge studies. J Allergy Clin Immunol 1979; 64:569-74 Fish JE, Kelly JF. Measurement of responsiveness in bronchoprovocation testing. J Allergy Clio Immunoll979; 64:592-96 Cockroft: D~ Berscbeid BA. Measurement of responsiveness to inhaled histamine: comparison of FEV I and SGavv. Ann Allergy 1983; 51:374-77 Anthonisen NR. Lung health study. Am Rev Respir Dis 1989; 140:871-72 O'Conner GT, Sparrow D, Segal MR, Weiss ST. Smoking, atopy and methacholine airway responsiveness among middle aged and elderly men. Am Rev Respir Dis 1989; 140:1520-26 Burney PG, Britton JR, Chinn S, Tahers6eld AE, Papacosta AO, Kelson MC, et aI. Epidemiology of bronchial reactivity in an adult population: results from a community study. Thorax 1987; 42:38-44 Toit JI, Woollock AJ, Salome CM, Sundrum R, Blanc JL. Characteristics of bronchial hyperresponsiveness in smokers with chronic airftow limitation. Am Rev Respir Dis 1986; 134:498-501 Amup ME, Mendulla LA, Anthonisen NR. Effects of cold air hyperpnea in patients with chronic obstructive lung disease. Am Rev Respir Dis 1983; 128:236-39
21 Greenspon L~ Morrissey WL. Factors that contribute to
CHEST I 102 I 5 I NOVEMBER. 1992
1425
ue W Greenspon, M.D., F.C.C.R;t and Edward Gracely, Ph.D.t Two discriminant functions, incorporating baseline measurements of pulmonary function and measures of airway responsiveness, were developed to improve patient classi6cation into groups of normal, asthma, or COPD. Accuracy of group classi&cation was compared between the usual laboratory method (single discriminating cut-ot1) and these new mathematically developed functions. Forty-Bve normal subjects, 27 asthmatic patients, and ten well-deBned COPD patients were entered into the analysis. Measurements of airway responsiveness were determined by measurement of both speci6c airway conductance (SGaw) and spirometry (FEV.) after sequential inhalation of methacholine. Beaulta: A single discriminant cut-ofT using measures of SGaw (PD35) or FEV. (PD20) does not sufficiently discrim-
Airway hyperresponsiveness is a distinguishing featore of asthma'> and the mechanisms involved in inducing airway hyperreactivity are active areas of research. Yet, laboratory bronchoprovocation tests using methacholine or histamine for diagnostic purposes in the evaluation of subclinical asthma have been questioned in their ability to adequately distinguish asthma from normal'-" and to predict which patient may significantly benefit from bronchodilator therapy 5 Histamine and methacholine bronchoprovocation tests that have been developed have shared a high sensitivity but a lower specificity in distinguishing subjects with asthma from normal subjects. The distribution of airway responsiveness in the normal population dictates that a simple cut-off point will include a significant number of false/positive test results," Bronchoprovocation agents like cold arr,7,8 exercise," distilled water," or hypertonic saline solution" have shown lower sensitivity but greater specificity than pharmacologic agents in distinguishing subjects with asthma from normal subjects. Although uniform guidelines have been recommended by the American Thoracic Society" and the society ofAllergy and Clinical Immunology, 13 various techniques are currently employed in clinical laboratories throughout the world. Since methodology is critical to the interpretation of test results, many laboratories have developed their own laboratory standards.
d
-From the Department of Pulmonary and Critical Care Medicine, the Medical College of Pennsylvania, Philadelphia. t Associate Professor of Medicine. *Assistant Professor of Community and Preventive Medicine. Manuscript received April 1, 1991; revision accepted March 13.
inate asthma from groups that contain normal and COPD subjects (67 to 71 percent predictive value). On the other
hand, our discriminant functions demonstrated improved patient classi6cation (positive predictive value, 88 to 89 percent). We conclude that bronchoprovocation tests used to evaluate the diagnosis of asthma should incorporate measures of baseline lung function into the analysis. This, we believe, is especially necessary when baseline lung function demonstrates minimal airOow obstruction and the possibility of other causes of airway disease exist. (Chat 1992; 102:1419-25)
I
CDU = cumulative dose units
I
The limitations of clinical bronchoprovocation testing may be due to several reasons: (1) there is a continuum of bronchial hyperresponsiveness in the normal population that overlaps with the asthmatic population such that a fully adequate discriminating cut-off is not perceptible. 3,5 (2) Airway responsiveness changes with age" and, perhaps more importantly, 15-18 with baseline airway caliber, which is not taken into account in the analysis. (3) Other diseases, such as COPD or bronchiolitis, which affect baseline airway caliber, will result in increased airway hyperresponsiveness I9 •20 even when patients may not respond clinically to bronchodilator therapy (4) Patients referred for testing with atypical chest complaints of cough, wheezing, or shortness of breath often have test results somewhere between the normal and asthmatic range (gray zone). These are difficult cases to classify and may represent mild asthma as well as early COPD, bronchiolitis, or within the expected range of normal. Over the past five years, we have collected data on airway responsiveness to methacholine in a population of normal men, as well as asthmatic and COPD patients. 2 1•22 As others have reported, we found a significant overlap between the normal and asthmatic population. COPD patients demonstrated a broad range of airway response from normal to asthmatic which is very dependent on baseline lung function. The purpose of this communication is to demonstrate that a mathematically developed discriminant function applied to tests of airway responsiveness that incorporate baseline lung function improves patient classification as normal, asthma, or COPD compared CHEST I 102 I 5 I NOVEMBER. 1992
1419
with a single cut-off level. Since one third of patients referred for bronchoprovocation testing to our laboratory have minimal airflow obstruction on baseline pulmonary function tests (yet FEV l greater than 70 percent), this type of analysis may help decrease the false/positive rate of bronchoprovocation testing, increasing the accuracy in recognizing asthma and a subset of patients that symptomatically improve with bronchodilator therap}'. METHODS
As part of other studies ongoing in our laboratory over the past five years, we have collected baseline pulmonary function data and airway responsiveness to methacholine data on 45 normal, 27 asthmatic, and 10 COPO subjects. Normal Subjects
Most normal subjects were recruited from an ongoing study evaluating age and airway responsiveness (pilot of NIA grant AG03934). Twenty-two subjects older than 60 years were physically active and screened to be free of medical illness. The remaining normal subjects were recruited from the hospital/medical school staff. Six normal subjects had smoked cigarettes but all had quit more than ten years prior to testing, and no subjects had smoked greater than ten pack-years. On questionnaire, no normal subjects had a history of allergic rhinitis, sinusitis, or asthma. Asthmatic Subjects
Allasthmatic subjects were recruited to be part ofother laboratory studies evaluating asthma. These patients were for the most part PRN users of inhaled bronchodilators. Ten subjects had been receiving no therapy for two weeks prior to testing. Nine subjects were regularly using inhaled sympathomimetics. Five patients were receiving a sustained-release theophylline preparation and three subjects had used steroids intennittently in the year prior to the study. All subjects gave a history typical of asthma and had histories of acute exacerbation that were relieved by bronchodilator therapy. COPD Subjects
Ten COPO subjects were recruited as part of a prior studya and selected to represent a wide range of objective functional impairment. AU subjects demonstrated airflow obstruction by pulmonary function tests. The diffusing capacity was reduced (53 ± 18 percent predicted), suggesting that our subjects probably had a signi6cant component of emphysema. In the COPO group, at the time of study, eight patients were receiving inhaled bronchodilators and five patients were receiving sustained theophylline. Three of ten subjects demonstrated a greater than 15 percent improvement in FEV. after inhaling isoproterenol 200 IJ.g, but this did not correlate with tests of airway responsiveness as previously reported." Since COPO subjects are not usually tested by bronchoprovocation testing, no other subjects were available for this analysis.
Testing Methods Static pulmonary function tests were performed on one of two spirometers (Gould 4000 rolling sealed spirometer or an Ohio 840 spirometer interfaced in line with an Apple 2E computer). Nonnal values for spirometry were from Crapo et alJ3 with a 10 percent correction for race applied to black individuals. The Deo predicted equation was based on data from Knudson et al. 1U Bronchoprovocation studies were performed according to a standard laboratory protocol. All subjects were free from upper respiratory tract infections for at least six weeks prior to the study Patients were instructed to discontinue oral theophylline therapy for 48 h and inhaled bronchodilator therapy for 12 h prior to 1420
methacholine testing. Only patients in stable conditions were entered into the study. Baseline measures of airway resistance and specific airway conductance were measured in a constant volume body plethysmograph (Warren E. Collins, Braintree, Mass). FEV. and FEF25-75 were recorded on a volume displacement spirometer. Subjects inhaled five breaths from FRC to TLC from a nebulizer (DeVilbiss 646, DeVilbiss Company, Somerset, Pa) powered by a dosimeter, (French-Rosenthal) of sequentially increasing concentrations of methacholine. Concentrations of 0.1, 0.5, 1.0, 5.0, 10.0, and 25.0 mg/ml were employed. Specific airway conductance (SGaw) and then FEV. were measured 3 min after each sequential dose until the FEV. fell by 20 percent. If there was no signi6cant fall in FEV. when the highest dose was inhaled, this concentration was administered a second time. The dose of methacholine that caused specific airway conductance to decrease by 35 percent (PD35) and the FEV. to fall by 20 percent (PD20) was recorded in each subject. We report for each test the log cumulative dose units of methacholine inhaled where one breath of 1 mg per milliliter methacholine = 1 dose unit. If after the highest dose of methacholine inhaled (333 CDU) there was no signi6cant response, a ceiling valve of >330 was used. In two subjects, a third inhalation of 25 mg/ml was delivered since they were perceived to be close to the physiologic end point. Single Discriminant Cut-off to Determine Ainooy &sponsiveness
A discriminant line best separating asthma from normal was generated by visual inspection. A separate cut-off was determined for values calculated using airway conductance (PD35) and for values utilizing spirometry (PD20). Patients with COPO were then evaluated as either matching the pulmonary and bronchial reactivity characteristics of the asthmatic" or "normal" groups by this scheme. U
Multiple Unear Dtscriminate Analysis
Linear discriminant analysis is a method for mathematically obtaining the "best" discrimination among groups of subjects using
one or a set of variables.· It is closely related to such parametric techniques as multiple regression and the multivariate analysis of variance. Like them, it assumes that the data are normally distributed and that relationships among variables are adequately captured by a linear (straight-line) function. For our analysis, we assumed an equal prevalence of each condition (asthma, COPD, or normal) occurring. We used raw numbers for all but the provocation tests for which logarithmically transformed values were used. Finally, we entered variables in a nonstepwise fashion. Six variables were entered into the analysis: log PD35, log PD20, FEV., FEV. predicted, and FEF percent predicted, SGaw. Since many laboratories throughout the country use only spirometry to measure airway responsiveness, a simplified discriminant function relying on only three variables was also analyzed (log P020, FEV. percent predicted, and FEV25-75 percent predicted). Statistics
Descriptive statistics of subject characteristics at baseline were expressed as mean± 1 SO. Groups were compared by analysis of variance (ANOVA) followed up with a version of the Tukey post hoc test for unequal sample sizes.· Differences were considered significant with p value <0.05. Correlations of airway responsiveness (P035 and P020) and other tests of airway function were analyzed for each group independently. Spearman's rank correlation was used for this analysis. Since many normal subjects exceeded the highest dose of methacholine inhaled in our laboratory, an upper limit of 330 units was recorded in these individuals. Unfortunately, this ceiling method was biased against our finding signmcant correlations in the normal population between lung function and airway responsiveness. To compare the results of the single discriminant cut-off method with the multiple discriminant functions, test sensitivity, specificity, Me1h8choIine 8ronctloprow)ca Testing (Greenspon, GraceIy)
Table I-Subject CharacteriBtica Normal
Asthma
COPD
45
27 29.7±8.1 3.46 ± 0.89 93±13* 80±5 85±16* 0.19±0.05 0.59±0.5O* 1.33 ± 0.63*
10 57.9±9.7*t 1.65±0.64*t 62±21*t 58±lO*t 27± 10*t 0.12±0.OO*t 1.38±0.75*t 1.43±0.83*
n Age, yr FEVI,L FEVI , %pred FEV/FVC FEF25-75, % pred
SGa\1v, em:"
S-I
Log PD3Ih CDU* Log PDIO , CDUI
49.2±18.4 3.54±0.89 105± 14 81±7 102±25 0.20 ± 0.04 2.00±47 2.45±0.18
*p
Subject characteristics by group classifications are listed in Table 1. Patients with COPD differed significantly on all variables from the normal group and by all variables except log PD20 from the asthma subjects. The asthma subjects differed from the normal group by all variables except FEV. (L), FEV.IFVC, and 3
2.5
2
-.... --- - -----
log P035
1
.5
0
3
2.5
--
2.25
..
1.5 ____B._________A 1.4
SGa~ As expected, the asthma group had the greatest degree of airway hyperresponsiveness (lowest PD35). The scatter plot in Figure 1 demonstrates the range of airway responsiveness by log PD35 and log PD20 for each group. The hatched line was visually selected as the best single discriminating point separating asthma from normal for each variable. In our laboratory, PD35s25 CDU (log PD35s1.4) or PD20
• •
2
•
• • •
• ••• •
log
PDm
- -
--- -
--~-------------------------
• • •
1.5
__________
• ••
...-
•
••
• •
•
•
1
•
.5
•
0
•
• -.5
-.5
• -1
-1
Normal
Asthma
COPO
Normal
Asthma
COPO
FIGURE 1. Airway response to methacholine by group. Results are presented as the dose of methacholine in cumulative dose units (CDU) necessary to cause a 35 percent reduction in SCaw (log PD35) or a 20 percent fall in FEV. (log PD20). CHEST I 102 I 5 I NOVEMBER, 1992
1421
Table 2-Correlation of A i n D G f l _ to Lung FUnction by Claa
0.13 (p=0.19) 0.38 (0.03) 0.76 «0.01)
Normal Asthma
COPD
FEV., % Pred
FEVlFVC
0.23 (0.06) 0.3 (0.01) 0.83 (0.<01)
0.22 (0.01) 0.33 (0.05) 0.47 «0.08)
FEF% Pred 0.23 (0.06) 0.49 (0.01) 0.7 (0.01)
SGaw
0.33 (0.01) 0.46 «0.01) 0.36 (0.15)
*Airway responsiveness is expressed as log PD35; similar results were found using log PD20.
group. Six variables were used to develop two canonical discriminant functions (see Table 3) which when plotted (Fig 3) best separate the three groups. An alternative method of classification is to apply a classification function (Fischer's linear discriminant function coefficient) which is provided in Appendix A. Using this discriminant analysis, 42/45 normal subjects were correctly classified, two were classified as "asthma;' and one was classified as "COPD." Of 27 asthmatics, 25 were correctly classified and two were rated as "normal," Most importantly, 100 percent of COPD subjects were correctly classified. More than 94 percent ofall subjects were appropriately classified. Since many laboratories do not make measures of SGa~ or calculate PD35, a simplified discriminant function using the three variables log PD20, FEV. percent predicted, and FEV25-75 percent predicted was developed (see Appendix B). By this method, 91 percent ofthe study population was correctly classified 3
• .•_.... ••
2.5
•• • • • •
2
•
•
15
• 5
(43145 normal, 23/27 asthma, 9/10 COPD). Table 4 summarizes the sensitivity, specificity, and predictive values for each method of analysis. DISCUSSION
By our laboratory method ofmethacholine bronchoprovocation testing, we have visually determined a single discriminant cut-off separating normal from asthma using both measures ofspirometry and specific airway conductance. Our PD20 cut-off for asthma of 180 CDU of methacholine is in keeping with prior reports in the literature. 28.29 Our PD35 cut-off for asthma of less than 25 CDU of methacholine is lower than reports in the literature. This is perhaps best explained by the fact that the normal population selected in this study was generally older than previous groups studied. Our results, however, are in keeping with other studies demonstrating greater overlap between normal and asthma with measurement of PD35 than with PD20. 30•31 Utilizing the PD20 measurement, we underclassified two asthmatic patients as normal, whereas the PD35 cut-off overclassified four normal patients as asthmatic. Hwe excluded the COPD group and only considered normal subjects with normal baseline pulmonary function tests, this analysis indicates that utilizing either of the single point cut-off values for PD20 or PD35 would be acceptable. The addition of a multiple discriminant analysis would add very little. However, in patients with COPD or those referred with chest complaints who have mildly abnormal baseline pulmonary function tests, a significant proportion will be misclassified into the asthma group by either single cut-offapproach. In our COPD group, 80 percent would be incorrectly classified as asthma using PD35 values; 60 percent would be miselassified by PD20 values. Table 3- CtJnonictJl DiacrimifItJfIt Function CoejJicienta
o
• •
-5
•• Normal
.·Asthma •• copo
-1
20
40
60
80
100
120
140
Baseline FEV, % predicted FIGURE 2. Linear regression comparing airway responsiveness to methacholine (log PD35) to one index of baselinie lung function (FEV. percent predicted) for each group. See Table 2 for statistics.
1422
FEV. FEV., %pred FEF, %pred SGavv, cm- I s-·
LogPD35 LogPD20
Function 1*
Function 2*
0.52 0.20 0.10 0.46 -0.97 -0.25
0.02 0.31 0.60 0.02 -0.02 0.50
*Variables are multiplied by the coefficients in the table after first being standardized to a mean of 0 and a standard deviation of 1 (converted to Z scores). The products are summed to find the value plotted in Figure 3. Function 1 represents the x-axis value and function 2 represents the y-axisvalue in Figure 3. Methacholine 8ronchoprow)cat Testing(Greenspon, GraceIy)
Results from the VA nonnative aging study33 suggest that there is a synergistic effect of smoking and atopy on nonspecific airway responsiveness. Yet, Burney et • 34 al have shown that the association of skin test • • 2.0 sensitivity and airway hyperresponsiveness diminishes • •• • •• with age. Since our normal group was currently ••• •• •• • • nonsmoking and somewhat older, the presence of • ••• • atopy may have had little overall effect. However, it is o •• •• •• •• •• possible that a discriminant function separating asthma • •• • • • Mil• from normal could be improved by considering other ••• • nonphysiologic parameters like age, smoking history, -2.0 ••• • IgE level, eosinophil count, and skin test reactivity. The use of a linear discriminant function with the • • • present data should be regarded as illustrative rather than definitive. This technique depends rather heavily -4.0 • •• •• Normal on the existence of a normal distribution of data within •• Asthma each group. It is not robust because individual subjects, • rather than means, are being compared and classified. -6.0 The logarithmic transformation, which often normalizes the numbers, does not do so herein because -6.0 -4.0 -2.0 o 2.0 4.0 6.0 of an artificial concentration ofP035 and P020 values FIGURE 3. Plot of the multiple discriminant functions. Normal, at 330, the upper limit in our laboratory. There is no asthma, and COPD groups can be visually separated. The x-axis technique designed to handle artificial distributions value is calculated from the sum of function 1 in Table 3. The y-axis value represents the sum of function 2, Table 3. such as this one. Nevertheless, the obtained results provide surprisingly good classifications. Of interest is the finding of positive methacholine The existence of relatively small sample sizes makes challenge test results in 65 percent of patients with both the discriminant results and the eyeball results mild chronic obstructive lung disease who have been look better than they are. The eyeball approach, in entered into the Lung Health Stud~32 The basis for a particular, capitalizes on minor fluctuations in the data positive test in this population is probably related to that will often not be present in a second sample. The the lower starting airway caliber at baseline. 16 •19 ,m discriminant analysis is slightly less subject to this Moreno et al'? have used a modeling analysis of in problem because it imposes a distribution on the data, vitro and in vivo airway responsiveness and hypothetthus limiting its ability to utilize random differences. ically demonstrate that a change in airway caliber can Nevertheless, since the imposed distribution is itself tailored to the given data, the discriminant results will totally explain a shift of the dose response curve to the left. Although direct measures of airway caliber also not generalize perfectly to another sample. Our six-variable discriminant function applied to were not made in this study; the strong linear correlation between baseline lung function and airway CO PO subjects classified 100 percent of the cases correctly into a separate group. This kind of analysis responsiveness in the CO PO group as noted in this and other studies confirms this important mechanism. can be added to the several other tests that reportedly One limitation of this study is that atopic status, will help discriminate asthma from COPD. For exwhich may contribute to airway hyperresponsiveness ample, asthmatics respond to histamine and methain the normal population, was not carefully evaluated. choline in a one-to-one relationship, but COPD paEvaluation of atopic status includes prick test skin tients show a much greater response to histamine than sensitivity, total IgE level, and eosinophil count. to methacholine. COPD subjects also fail to respond Table 4-MetluJcholine Bronchoprooocation Teating: Seruitioity, Specificity, Grad Predictive Valaa ofMethod. to 4.0
_..
._...
..copo
Identify An,",",·
Sensitivity Specificity (+ ) Predictive value ( - ) Predictive value
LogPD20
LogPD35
93 80 67
96 82
96
71 98
Discriminant Function
SimpliBedt Discriminant Function
93 95 89 97
85 95 88 94
• All values are percentages. tSpirometry values only. CHEST I 102 I 5 I NOVEMBER, 1992
1423
to inhalation of the a-agonist methoxamine." Whereas 80 percent of asthma subjects have bronchoconstriction in response to breathing cold dry air during exercise, less than 10 percent of COPD subjects will respond even when their methacholine responsiveness is in the asthma range.i" The advantage of our discriminant function is that no other test need be given beyond measurements of baseline lung function. We routinely measure both PD35 and PD20. However, failure to measure specific airway conductance would only decrease the sensitivity for recognizing COPD to 90 percent (9/10). The positive predictive value for identifying asthma remains very good (88 percent). There are several advantages of utilizing a multiple discriminant function analysis. Since a significant proportion of patients referred to our laboratory for evaluation of atypical chest symptoms have mild abnormal pulmonary function test results at baseline, this type of analysis may classify these patients as COPD rather than asthma. The COPD classification would indicate an appropriate level of airway responsiveness for the degree of airflow obstruction. It would also predict that these patients would be less likely to respond to standard bronchodilator therapy This hypothesis, however, needs further testing. Secondly, even in patients with COPD, airway responsiveness may be out of proportion to the degree of alteration in lung function. These patients may need more aggressive management to control airway inflammation since this is our current understanding for the development of airway hyperresponsiveness in asthma. Future studies evaluating the role of airway responsiveness in predicting the decline in lung function in COPD will hopefully clarify the importance of airway inflammation and hyperresponsiveness in this group of patients. We are not suggesting that this equation be adopted by all laboratories to improve patient classification. Rather, we alert clinicians to the hazards of test interpretation without relating to the baseline lung function. Each laboratory performing these tests ApPENDIX A*
Normal (1) FEV. FEV., % pred FEF25-75, % pred SGaw LogPD35 LogPD20 Constant
1.225 0.427 0.017 76.77 -3.780 5.512 -37.25
Asthma (2)
3.086 0.433 -0.014 104.2 -8.5 2.468
-34.86
COPD (3) 0.194 0.323 -0.102 61.699 -0.096 2.01 -15.01
*To classify subjects into one of three groups (normal, asthma, COPD), measures of baseline lung function and test results of airway responsiveness are entered into the following three equations. Variables are multiplied by the coefficients in the table. The products are summed and the constant (being a negative number) is subtracted. For all variables, raw whole numbers are used or log values. The equation that generates the largest summed result designates the patient's classification.
1424
,ApPENDIX
Normal (1) FEV1 , 'II pred FEF25-75, % pred LogPD20 Constant
B* Asthma (2)
0.389 0.085 5.227 -32.38
0.397 0.073 0.359
-22.86
COPD(3) 0.288 -0.041 3.385 -12.07
*This simplified discriminant function just incorporates measures of spirometry with one measure of airway responsiveness (PD20). The equation that generates the largest summed result designates the patients classmcation. For all percent, use whole numbers. The classification functions listed above are available on floppy disk (IBM compatible) upon request to the author. Example 1: A patient is studied with FEV1 =2.9 FEV1 'II pred = 93 FEF25-75 % pred=65 SGaw=0.19 PD35 (CDU) = 10 PD20 (CDU) = 75 Equation (1) Normal = 28.27 Equation (2) Asthma = 29.4 Equation (3) COPD=24.35 Therefore the patient is classified asthma. Example 2: A patient is studied with FEV. =2.5 FEV. % pred=74 FEF 'II pred = 50 SGaw=0.19 PD35 (CDU) = 20 PD20 (CDU) =75 Equation (1) Normal = 18.7 Equation (2) Asthma = 18.01 Equation (3) COPD = 19.96 Therefore the patient is classified COPD.
should attempt to study laboratory controls, including asthma and normal subjects. Laboratory measures and techniques differ with each institution such that each laboratory may need to develop its own reference standards and discriminant function. The utility of this method in evaluating clinical cases with airway responsiveness results that fall in the gray zone is under prospective evaluation. ACKNOWLEDGMENTS: The writers wish to thank Barbara Sewell for her excellent secretarial support on this manuscript and Adele Kaplan, Ph. D., for her statistical advice. REFERENCES
1 Boushey HA, Holtzman MJ, Sheller JR, Nadel JA. Bronchial hyperreactfvity Am Rev Respir Dis 1980; 121:389-413 2 Hargraeve FE, Ryan G, Thompson NC, O'Byrne PM, Latimore R, Juniper EF, et al, Bronchial responsiveness to histamine or methacholine in asthma: measurement and clinical significance. J Allergy Clio Immunoll981; 68:342-55, 347-55 3 Popa ~ Singleton J. Provocation dose and discriminant analysis in histamine bronchoprovocation. Chest 1988; 94:466-75 4 Filuk RB, Serretta L, Anthonisen NR. Comparison of response to methacholine and cold air in patients suspected of having asthma. Chest 1989; 95:948-52 5 Chbebra SK, Gaur SN, Khanna AK. Clinical significance of nonspecific bronchial hyperresponsiveness in asthma. Chest 1989; 96:596-600 6 Cockroft D~ Bercheid BA, Murdock KY. Unimodol distribution ofbroncbial responsiveness to inhaled histamine in a random Methacholine 8ronchoprtM)cat Testing (GteenspotJ, GraceIy)
human population. Chest 1983; 83:751-54 7 O'Byrne PM, Ryan G, Morris M, McCormack D, Jones NL, Morse JLC, et ale Asthma induced by cold air and its relation to non specific bronchial responsiveness to methacholine. Am Rev Respir Dis 1982; 125:281-85 8 Weiss J~ Rossing TH, McFadden BR, Ingram RH. Relationship between bronchial responsiveness to hyperventilation with cold and methacholine in asthma. J Allergy Clin Immunol 1983; 72:140-44 9 Chatham M, Bleecker ER, Smith PL, Rosenthal RR, Mason ~ Normal PS. A comparison of histamine, methacholine and exercise airway reactivity in normal and asthmatic subjects. Am Rev Respir Dis 1982; 126:235-40 10 Bascom R, Bleecker ER. Bronchooonstriction induced by distilled water. Am Rev Respir Dis 1986; 134:248-53 11 Araki H, Sly PD. Inhalation of hypertonic saline as a bronchial challenge in children with mild asthma and normal children. J Allergy Clin Immunoll989; 84:99-107 12 ATS News Spring, 1980; 11-19 13 Chai H, Farr RS, Froehlich LA, Mathison DA, McLean JA, Rosenthal RR, et ale Standardization of bronchial inhalation challenge procedures. J Allergy Clin Immunoll975; 56:323-27 14 Hopp RJ, Bentra A, Nair NM, Townley RG. The effect of age on methacholine response. J Allergy Clin Immunoll985; 76:60913 15 Sparrow D, O'Conner G, Colton T, Barry CL, Weiss ST. The relationship of non-specific bronchial responsiveness to the occurrence of respiratory symptoms and decreased levels of pulmonary function. Am Rev Respir Dis 1987; 135:1255-60 16 Rijcken B, Schoten J~ Weiss ST, Speizer PE, Vanderlende R. The relationship between airway responsiveness to histamine and pulmonary function in a random population sample. Am Rev Respir Dis 1988; 132:826-32 17 Moreno RH, HoggJC, Pare PD. Mechanisms of airway narrowing. Am Rev Respir Dis 1986; 133:1171-80 18 James AL, Pare PO, Hogg JL. The mechanisms of aiJway narrowing in asthma. Am Rev Respir Dis 1989; 139:242-46 19 Ramsdell ~ Nachtwey FJ, Moser KM. Bronchial hyperreactivity in chronic obstructive bronchitis. Am Rev Respir Dis 1982; 126:829-32 20 Yan K, Solome CM, Woolcock AJ. Prevalence and nature of bronchiole hyperresponsiveness in subjects with chronic 0bstructive pulmonary disease. Am Rev Respir Dis 1985; 132:259
22
23
24 25 26 27 28
29
30 31
32 33
34
35
36
inhibition of mechacholine induced bronchooonstriction. Am Rev Respir Dis 1986; 133:735-39 Greenspon L~ Parrish B. Inhibition of mechacholine induced bronchoconstriction in patients with chronic obstructive pulmonary disease. Am Rev Respir Dis 1988; 137:281-85 Crapo RO, Morris AH, Gardner RM. Reference spirometric values using feelings and equipment that meet ATS recommendation. Am Rev Respir Dis 1981; 123:659-64 Knudson RJ, Kaltenborn WT, Knudson DE, Burrows B. The single breath carbon monoxide diffusing capacity. Am Rev Respir Dis 1987;135:805-11 Sokal RR, RohlfEJ. Biometry. New York: WH Freeman and Co, 1969:488-90 Jaccard J, Becker MA, Wood G. Pairwise multiple comparison procedures: a review Psychol BuD 1984; 96:589-96 Pratter MR, Hingston DM, Irwin RS. Diagnosis of bronchial asthma by clinical evaluation. Chest 1983; 84:42-7 Hopp JR, Bewtra AR, Nair NM, Townley RG. Specificity and sensitivity of methacholine inhalation challenge in normal and asthma children. J Allergy Clio Immunoll984; 74:143-48 Townley R, Bewtra A, Nair N, Bradley F, Walt G. Methacholine inhalation challenge studies. J Allergy Clin Immunol 1979; 64:569-74 Fish JE, Kelly JF. Measurement of responsiveness in bronchoprovocation testing. J Allergy Clio Immunoll979; 64:592-96 Cockroft: D~ Berscbeid BA. Measurement of responsiveness to inhaled histamine: comparison of FEV I and SGavv. Ann Allergy 1983; 51:374-77 Anthonisen NR. Lung health study. Am Rev Respir Dis 1989; 140:871-72 O'Conner GT, Sparrow D, Segal MR, Weiss ST. Smoking, atopy and methacholine airway responsiveness among middle aged and elderly men. Am Rev Respir Dis 1989; 140:1520-26 Burney PG, Britton JR, Chinn S, Tahers6eld AE, Papacosta AO, Kelson MC, et aI. Epidemiology of bronchial reactivity in an adult population: results from a community study. Thorax 1987; 42:38-44 Toit JI, Woollock AJ, Salome CM, Sundrum R, Blanc JL. Characteristics of bronchial hyperresponsiveness in smokers with chronic airftow limitation. Am Rev Respir Dis 1986; 134:498-501 Amup ME, Mendulla LA, Anthonisen NR. Effects of cold air hyperpnea in patients with chronic obstructive lung disease. Am Rev Respir Dis 1983; 128:236-39
21 Greenspon L~ Morrissey WL. Factors that contribute to
CHEST I 102 I 5 I NOVEMBER. 1992
1425