- Email: [email protected]
Concepts of normality applied to the measurement of lung function
Concepts of Normality Applied to the Measurement of Lung Function
Lung function testing has become an integral part of the clinical assessment of pulmonary disease. However, increasing application of such testing in research into the origins of disease as well as in the assessment of health raises the issue of what is meant by normal respiratory function. In essence, this depends on the sources of variation in lung function measurements that are of interest (the signal) and those that are not (the noise). For instance, clinicians are primarily interested in variation due to disease, all other sources of variation being considered noise. Physiologists are concerned with sources of variation other than disease, and the interest of epidemiologists (and their definition of normality) varies according to the specific objectives of each particular study, report or program. Consideration of the sources of variation in lung function (within a subject, between subjects, and between populations) not only is useful in clarifying the concepts of normality for clinical application, physiologic studies, and epidemiologic purposes, but also is mandatory for an understanding of lung function in the transition between health and disease, a major thrust in all three areas of endeavour.
MARGARET R. BECKLAKE, M.D., F.R.C.P.(London) Montreal,
Quebec,
Canada
From the Departments of Epidemiology and Biostatistics, and of Medicine, McGill University, Montreal, Quebec, Canada. This work was supported by the Medical Research Council of Canada. Dr. Becklake is a Career Scientist of the Medical Research Council (Canada). Requests for reprints should be addressed to Dr. Margaret R. Becklake, Department of Epidemiology and Biostatistics, McGill University, 1110 Pine Avenue West, Montreal, Quebec, H3A lA3, Canada. Manuscript accepted March 7, 1985.
1158
June
1986
The American
Journal
Lung function testing, like chest radiography, has become an integral part of the clinical assessment of pulmonary disease [l]. In the clinical context, particularly when the disease process is advanced and when lung function information can be assessed in light of all other available information for an individual patient, interpretation of the results as normal or abnormal is usually not difficult. However, two developments have served to underline the need to clarify our concepts of normality. First is the increasing research effort into the origins of chronic nonmalignant respiratory disease, triggered by the evidence that the disease processes leading to disablement from chronic airflow limitation in middle-age start in early adulthood [2,3]. Second is the application of lung function testing in the assessment of health as opposed to disease, for instance in preemployment and annual physical examinations. The commercial availability of computerized apparatus for lung function testing has resulted in many such evaluations being carried out in workplaces and/or occupational groups, often by health personnel untrained in their use who rely on criteria and reference standards selected by the manufacturer and incorporated into the software of the apparatus as purchased. The purpose of this review is to clarify as far as possible the concepts of normality applied to the measurement of lung function; criteria of normality and the issue of what is “normal” including confidence limits have been addressed in several recent papers [4-61 and are not discussed here.
of Medicine
Volume
80
CONCEPTS
TABLE instrument Procedure
Observer Subject Interactions Other
I
Sources of Variation in Lung Function to Technical Factors
APPLIED
TO LUNG
FUNCTION
TESTS-BECKLAKE
with their search for (and research into) the factors responsible for the distribution of attributes (including abnormality and/or dysfunction and/or disease) within populations, with a view to developing control or preventive strategies. For instance, the American Thoracic Society proposed guidelines as to what constitutes an adverse respiratory health effect in response to envrionmental pollutants [9] to take into account the full range of potential response, including (1) increased pollutant burden without detectable health impact; (2) physiologic changes of uncertain significance; (3) pathophysiologic changes; (4) morbidity, and (5) mortality. Only the last three are considered adverse health effects.
Due
Within and between instruments [ 121 Number of trials; choice of results to be reported, e.g., best of three, average of best two of five, and so on [14,15] Administration of tests [ 161; evaluation of results [171 Comprehension; cooperation [18,19] Subject-observer; observer-instrument [ 161 Temperature [20,21]; altitude [12]
References indicate various components results and, hence, error.
OF NORMALITY
studies documenting the contribution of the of measurement to-variation in spirometric their potential as causes of measurement
VARIATION IN MEASUREMENT BASIC CONCEPTS The meaning of the word normal varies with the context in which it is used. In the popular (journalistic) use, it generally means ideal, conventional, or habitual. Statisticians use it in a numerical sense to describe distribution about a middling tendency. In the biologic context, it is concerned with the issue of human variation (which can, on occasions, be extreme) and the factors that account for it. Nevertheless, even in the various biomedical sciences, there is little unanimity about its meaning, perhaps because it involves defining health, itself an elusive concept. Thus the widely cited definition in the Constitution of the World Health Organization (1948) as “a state of complete physical, mental and social well being and not merely the absence of disease or infirmity” needs elaboration in specific contexts. To anatomists, normal variations in structure are those consistent with the maintenance of good health. To physiologists, normal variation of bodily functions are those that maintain the stability of the “internal milieu”: the focus is on the factors other than disease that cause these variations. Clinicians, on the other hand, concentrate on variations due to disease. They therefore wish to identify (so that they can disregard) variations that they loosely call physiologic. Interestingly enough, the word disease is not defined in most standard textbooks of medicine, although several authors [7,8] have proposed thoughtful if somewhat cumbersome definitions. Burrows [7], elaborating on earlier definitions, emphasized the distinction between diagnosis and disease, which he defined as “the sum of those abnormal phenomena displayed by a living organism that place the organism at a biological disadvantage and that are associated with one another on the basis of their frequent occurrence in other members of the species or on the basis of common causality.” By contrast, epidemiologists and those who practice public health have no general or universal definition for health or disease but define the terms precisely for each study, report, or program they undertake. This is in line
June
Information theory, which distinguishes signal (the information a test is designed to measure) from noise (measurement error), has been applied to other clinical diagnostic tests [IO]. In addition to measurement error, most if not all clinical measurements also comprise biologic variation (the basis of most biologic sciences) as well as variation due to disease if this is present (the basis of clinical medicine). To evaluate the contribution of the latter, clinicians must be aware of, and take into account, the former. Not only is there often little appreciation by clinicians of the range of both measurement error and biologic variation, but also there is often unjustified faith in the specificity of laboratory tests for detecting disease, particularly values measured on a continuous scale such as measurements of lung function. As a result, clinicians continue to strive for more precise definitions of health (or normality), and yet all the evidence suggests that the test ranges for disease, even frank disease, are hidden in the normal distribution curve-or, at best, deface the descending limb of the curve [ 1 I]. MEASUREMENT ERROR Measurement error is usually assessed as short-term repeatability of a test result within a time frame over which biologic variation is unlikely. Its magnitude varies according to the nature and complexity of the test [ 121. Table I lists the various technical components in the measurement of lung function; each has been shown to contribute to measurement error [ 13-221. For forced vital capacity, for instance, good laboratories maintain the variation between repeated measurements in normal subjects at approximately 3 percent [ 121, and for forced expiratory flow at 50 percent of forced vital capacity, at approximately 6 percent [ 121; for closing volume and for the slope of phase Ill in the single-breath nitrogen test, values closer to 20 percent and 50 percent, respectively, can be expected [ 121. When measurement error is random, it affects the internal validity of a study, i.e., diminishes the chances of detecting differences between study groups; when it is systematic, it affects the external
1986
The American
Journal
of Medicine
Volume
80
1159
CONCEPTS
TABLE
OF NORMALITY
II
APPLIED
TO LUNG
FUNCTION
TESTS-BECKLAKE
Sources of Biologic Variation in Measurement of Lung Function*
tween laboratories, it may also be responsible for failure to detect certain types of abnormality. For instance, the inability of a subject to produce a repeatable forced expiratory volume result in compliance with American Thoracic Society standards may, in itself, reflect abnormality and/or incipient disease. Thus, in a prospective study of Vermont granite workers, men who were unable to do so showed a greater loss of function over the next four years than men who were able to perform satisfactorily [28]. This point could have been missed had the study excluded subjects who were unable to perform a “satisfactory” test in accordance with American Thoracic Society instructions.
Source
Variation
Measurement error (includes all the technical factors listed in Table I) Diurnal (circadian) and seasonal effects
Intra-subject
[31-331 Endocrinologic effects [34,35] All the above Personal Size, sex, age [12,31]; Physical activity 1311, muscularity [ 181 Race and other genetic characteristics
Inter-subject
[31,36] Past and present health [ 1,371 Environmental Cigarette exposure, voluntary [31,38] and involuntary [39] Occupation [3 1,401 Residence, urban or rural [31] Urban air pollution [1,31], home pollution
BIOLOGIC VARIATION Biologic variation can, of course, be detected only if it exceeds variation due to measurement error. It can conveniently be thought of as intra-subject, inter-subject, and inter-population, depending on the focus of interest (see Table II). Besides measurement error, intra-subject variation may be due to factors such as posture [30], relationships to meals [31], circadian or diurnal variation [32], seasonal variation [33] possibly due to reactivity of airways to cold, and perhaps endocrinologic factors such as those that accompany the menstrual cycle [34,35]. All sources of intra-subject variation already mentioned may also cause inter-subject variation. Differences between subjects may also be due to differences in certain of their personal characteristics, These include age and sex, stature as reflected in height and weight [ 121, muscularity in so far as it affects respiratory muscle performance [ 121, level of physical activity and training [3 I], race and other genetic characteristics [ 121, and past and present health experience [ 1,371. Of the environmental factors, cigarette smoking is clearly the most important [ 1,31,38], including the effects of involuntary smoking [39]. Other environmental factors including occupation [1,40], exposure to urban air pollution [31] including home pollution [41], place of residence (urban or rural), and/or socioeconomic status [37,41,42], all of which may affect lung function. All the sources of intra- and inter-subject variation already mentioned may also cause inter-population variation insofar as they determine the inclusion of subjects in, or their exclusion from, one population or group compared with another. Even the frequently used grouping into smokers and nonsmokers may be bedeviled by this type of bias, since the decision to take up cigarette smoking appears to be related to lung function status, those who decide to do so having, on average, better function than those who do not [ 181. Likewise, the “healthy worker” effect describes the fact that employed persons often have a better health experience, including better lung function, than reference groups, again implying a selection process of the more fit into the work force [29]. Inter-
1411 Socioeconomic status [36,40,42] All the above Selection factors determining inclusion in, or exclusion from, the study populations 123,401
Inter-population
* Modified with permission from [29]; references indicate studies documenting the contribution of the various factors mentioned to biologic variation in lung function.
validity of a study, i.e., decreases the comparability of study results with those published elsewhere. Even in the case of forced vital capacity (perhaps the most widely used lung function test), all the components of measurement listed in Table I have been shown to contribute to measurement error. As a result, professional societies such as the American Thoracic Society and the Societas European Physiologica Respiratoire have developed guidelines [14,23] to minimize error in each component of measurement. These include: (1) criteria for instrument performance and advice on calibration procedures [ 14,231; (2) recommendations on procedure including number of trials to be carried out, criteria for technical rejection and selection of results to be reported, posture, use of nose-clips, and so on [23,24]; and (3) guidelines for technician training [25] and evaluation of results, although with the widespread introduction of microprocessors, this component of measurement error has diminished considerably [ 121. Lists of apparatus available [ 121 and reviews of their performance [26] are also published. Despite this, measurement error [13] and unexplained measurement drift over time [27] occur in even the most experienced laboratories and must be continually borne in mind in the interpretation of results. Although standardization of methodology has undoubtedly been remarkably successful in diminishing measurement error and increasing comparability of results be-
1160
June
1966
The American
Journal
of Medicine
Volume
60
CONCEPTS
OF NORMALITY
Proportion of Variation In Forced Vital Capacity Attributable Selected Published Studies of Adult White Men Reference fi”41 /:65;
Number of Subjects
Regression Age
Age Range
2751 156 517 870
25-74 20-64 15-79 20-84
-0.027 -0.032 -0.025 -0.014
[=I
2,150t 194
20-74 18-80
[481
125
15-91
-0.024 Nonlinear, weight -0.021
iii]
329t 86
25-70 17-35
[471
APPLlED
TO LUNG
to identified
FUNCTION
Personal
TESTS--BECKLAKE
Characteristics
In
Proportion of Variation” Explained Not Explained
Coefficients Height 0.051 0.046 0.048 0.058
0.67 0.41 0.33 0.42
0.33 0.59 0.67 0.58
0.062 includes and interactions 0.060
0.38 0.56
0.62 0.44
0.54
0.46
Nonlinear, -0.030 includes 0.084 weight, respiratory pressures, and smoking
0.71 0.29
0.71 0.29
* Calculated as R* and 1 - R*, respectively. + Includes smokers.
population variation due to health selection factors is also clearly relevant in the selection of reference lung function values, to be discussed later. PROPORTION ATTRIBUTABLE
OF INTER-SUBJECT VARIATION TO VARIOUS FACTORS
The proportion of inter-subject variation in one measure of lung function, the forced vital capacity, that can be explained on the basis of variation in personal characteristics such as age, height, and/or weight in selected studies [18,22,43-491 is shown in Table Ill. The table refers to data from studies on adult white men judged by more or less stringent criteria to be in good health; included are some of the more frequently used reference values applied in routine hospital laboratories [22,45-491. Table Ill also shows the proportion of variation in each study population that was explained by these personal characteristics expressed as the squared overall correlation coefficient (r*); this varied considerably between studies, from 0.29 to 0.7 1. The proportion not explained by these variables, i.e., 1 - r*, which varied from 0.29 to 0.71, must therefore be attributable to factors other than the personal characteristics contained in the regressions. Included would be those listed in Table II and presumably others as yet unidentified; indeed, much research into the origins of chronic airflow limitation is directed towards identifying and quantitating some of these unidentified sources of variation. Table IV shows estimates of the proportion of intersubject variation in forced vital capacity attributable to the sources already discussed; it is based on the information contained in Tables II and Ill and also on other published data [29]. Approximately 22 percent of overall variation can reasonably be attributed to variation in body characteristics reflected in height and weight, approximately 8 percent can be attributed to age (if the age span covers childhood to old age), and approximately IO percent can
June
be attributed to race [31,36]. Within the residual (i.e., unexplained) variation, estimated at approximately 27 percent, must lie all the other factors, personal and environmental, known to contribute to inter-subject variation. Smoking is probably the most important single factor, accounting for perhaps a third to a half of the residual variation [22,46,47]; a further 5 percent may perhaps be due to childhood illnesses (before the age of 16 years) [37], approximately 5 percent may be due to variation in respiratory pressures [ 181, and the remainder is due to other factors known and unknown. SIGNAL
VERSUS
NOISE
Do the sources of variation listed in Table IV represent variation in “normal” function or do they represent “abnormal” variation? This depends on the sources of variation that are of interest (the signal) and the sources that are not (the noise), and this in turn depends on the context in which the question is asked. For instance, clinicians interested in the diagnosis of disease and its precursors
TABLE IV
Estimates of Proportion of inter-Subject Variation in Forced Vital Capacity Attributed to Identified Factors* Factor
Sex Age Height Weight Race Technical (measurement Unexplained+ Total
Proportion of Attributable
Variation
up to 0.30 0.08 0.20 0.02 1
up to 0.30 0.10 0.03 0.27 1.00
error)
* Based on published data in Tables II and Ill and [29]. t Residual variation, includes ail the factors listed in Table II such as smoking (active and passive), childhood illnesses, respiratory pressures, and so on.
1986
The American
Journal
of Medicine
Volume
80
1161
CONCEPTS
OF NORMALITY
APPLIED
TO LUNG
FUNCTION
TESTS-SECKLAKE
vfoulci consider all of the factors listed on Table II as “normal” variation (or noise), with the possible exception of past and present health. In practice, they therefore seek a convenient way to take them all into account (usually through prediction formulas) so that variation due to disease (the signal) can be identified. To reinforce antismoking advice, they may also wish to include the effects of smoking as part of the signal and will therefore seek reference values (and prediction formulas) based on nonsmoking subjects. On the other hand, if the objective is to identify the influence of environmental agents over and above that of smoking, then the effects of smoking become part of the noise, and it may be useful to use prediction formulas that take it into account [50,51]. Similar issues arise in the context of physiologic and epidemiologic studies. For instance, in a physiologic study to examine the influence of posture on lung function (the signal), the effects of postprandial changes in lung volume would be considered noise. By contrast, in another study, the effects of ingesting a meal might be the signal, and the effects of posture would be noise. Similarly, in an epidemiologic survey of a work force, the effects of an occupational exposure might be the signal in one study, with smoking included in the noise; in another study, their roles might be reversed.
WHY DO REFERENCE VALUES VARY? Reference values are usually based on the analysis of pertinent personal and environmental factors in populations judged to be healthy according to more or less wellspecified criteria, which may vary from study to study [23]. The analysis is also usually, as Rurrows et al [54] have pointed out, carried out for descriptive rather than predictive purposes. Of the sources of inter-population differences likely to affect derived prediction formulas, some are obvious-inter-population differences in history of previous respiratory illness, as well as differences in place of residence, socioeconomic status, and environmental pollution levels. Less obvious are the frequency distribution within the population of the factors analyzed and the method of analysis [55]. $or instance, for the studies listed in Table Ill, the age coefficient was greatest (and, by implication, the rates of aging fastest) in Cotes et al’s [44] subjects and least in Cherniak and Raber’s [46] subjects (implying the slowest rates of aging). These differences can, however, be reasonably attributed to differences in the age distribution of the two study populations; in the latter study 1461, the population was weighted towards the younger ages, with more than half of the subjects 35 years or less, and this weighting is obviously reflected in the regression coefficient. In the former study [44], the opposite pertained, with more than half of the subjects between 55 and 65 years of age. Likewise, the method of analysis used also influences the findings. For instance, if linear analyses are used to describe the agerelated changes in studies confined to adults, over-prediction of values in young persons is likely [22]; nonlinear analyses more correctly model the age-related changes that occur in early adulthood [6,18,22,54-561 and should therefore be used in dealing with persons under the age of 35 years. CHOICE OF REFERENCE VALUES
DEALING Wll-bj NOISE The general approach to dealing with noise will depend on the context, in particular whether the issue is clinical or epidemiologic. In either event, it is important to identify all relevant potential sources of noise to be taken into account. As already indicated, measurement error will always be noise and should be minimized by the techniques referred to earlier. In population (epidemiologic) studies, comparison between population groups is usually required to test the hypothesis‘under study. For instance, to assess the effects of an occupational exposure, exposed and non-exposed groups may be compared with all other sources of inter-population differences taken into account. This may be done (1) by matching, for example, by smoking habits and age, although this may not be possible when groups are small; (2) by statistical analysis, for instance, using the various forms of covariance and regression analysis; or (3) by the use of outside reference values that take into account some of the sources of variation usually considered noise, for example, age and stature [52]. The first two methods are commonly used in epidemiologic studies, and the third is used in clinical laboratories. However, because of the considerable variation between published reference values or prediction formulas, graphically described by Glindmeyer [53] as “predictable confusion” a clinical laboratory director or epidemiologist in charge of a study faces a major dilemma in ‘making the most appropriate choice.
1162
June
1966
The
American
Journal
of Medicine
Volume
Many would agree with the conclusion reached by Clausen [ 121 in a recently published and valuable text on pulmonary function testing that “because of inexplicable differences (in published normal values), there is no recommended set of prediction equations applicable to all laboratories and all patient populations.” Nevertheless, a reasoned choice made by a responsible laboratory director (or epidemiologist) is preferable to leaving the choice to the whim and/or judgment of the manufacturers of automated equipment. Certain guidelines can be offered based on methodologic, epidemiologic, and conceptual considerations. The first is obvious: the methods used (apparatus, procedure, and analysis) to generate the reference values should be comparable-and, preferably, identical-to those to be used in the laboratory seeking to apply the reference values. Second, the population (or sample) that generated
60
CONCEPTS
the reference values should be as closely matched as possible to the population from which the patients or subjects to be assessed are drawn. For instance, for the residents of Alice Springs, Queensland, Australia, their lung function may be more appropriately assessed in relation to that of the residents of Busselton, West Australia [57] who do not smoke by inclination, than in relation to that of either the residents of Tucson, Arizona, who also do not smoke by inclination [49], or the residents of Salt Lake City, Utah, who do not smoke by religious conviction [48]. Third, and most important, although usually least clearly articulated, is the conceptual issue of the purpose underlying the comparison to be made. If it is for clinical (diagnostic) purposes, the choice of reference values, paradoxically, is probably relatively unimportant, because the tests of function are simply one item of information in an array of other clinical information about an individual subject upon which an assessment will be based. A good clinician will assign the correct weighting to a borderline lung function result in reaching a final opinion. By contrast, borderline values may be much harder to interpret in the context of a health maintenance (or periodic) examination. Such an examination has the characteristics of a screening examination, including ethical as well as scientific considerations [58]. In contrast to the clinical context in which the subject seeks out the physician who is not therefore, to use the graphic phrase of Cochrane and Holland [58], “responsible for defects in medical knowledge,” the opposite applies in the context of a screening examination. Here, the criteria for subject (case) identification have nothing to do with “normal” variation but depend on whether a given test value identi-
OF NORMALITY
APPLIED
TO LUNG
FUNCTION
TESTS-BECKLAKE
fies a condition at a stage at which intervention (for example, treatment or removal from exposure) would improve the subject’s future health prospects. So far, there is insufficient information on which to base the use of lung function tests as a screening tool, although this has been identified as a priority area for study [59]. Meanwhile, lung function tests continue to be used and interpreted daily in clinical laboratories. Action based on these interpretations is likely to be reasonable, given that clinicians integrate these results into an overall medical judgment based on all available information about the subject. Indeed, it cannot be too strongly emphasized that these judgments should be based more on clinical experience than on strict adherence to defined (usually arbitrary) limits of “normality.” Numerical (statistical) criteria for such limits are also hampered by the fact that the greater the number of laboratory tests of lung function or any other bodily function [60] performed in healthy subjects, the greater the chance of abnormal results. Nor is there any evidence that further improvement in the precision of prediction formulas (that is, reduction in the proportion of unexplained variation) will occur until other sources of inter-subject variation are identified, quantified, and incorporated into the formulas. At present, therefore, interpretation of lung function test results in the clinical context should remain a matter of clinical judgment. ACKNOWLEDGMENT I thank the following colleagues for a critical review of the text: Dr. A. S. Buist (Portland, Oregon), Professor S. Benatar (Cape Town), and Drs. P. Ernst and R. Dales (Montreal).
REFERENCES
2. 3. 4.
5.
6.
7. a. 9.
Bates DV, Macklem PT, Christie RV: Respiratory function in health and disease, 2nd ed. Philadelphia: WB Saunders, 1977; 37. Fletcher C, Peto R: The natural history of chronic airflow obstruction. Br Med J 1977; 1: 1645-1648. Burrows B: An overview of obstructive lung disease. Med Clin North Am 1981; 65: 455-471. Miller A, Thornton JC: The interpretation of spirometric measurements in epidemiologic surveys. Environ Res 1980; 23: 444-468. Pennock BE, Cottrell JJ, Rogers RM: Pulmonary function testing. What is normal? Arch Intern Med 1983; 143: 2123-2127. Buist AS: Evaluation of lung function: concepts of normality. In: Simmons DH, ed. Pulmonology, vol 4. New York: John Wiley, 1982; 141-165. Burrows B: Standard pulmonary nomenclature and symbols. Ann Intern Med 1975; 83: 419-420. Campbell EJM, Scadding JG, Roberts RS: The concept of disease. Br Med J 1979; 2: 757-762. American Thoracic Society: Guidelines as to what constitutes an adverse respiratory health effect, with special reference to air pollution. Am Thorac Sot News 1984; 10: 6-8.
10.
11. 12. 13.
14. 15.
16.
17.
18.
June
1966
Diamond GA, Hirsch M, Forrester JS, et al: Application of information theory to clinical diagnostic testing. Circulation 1981; 63: 915-921. Cochrane AL: The history of the measurement of ill-health. Int J Epidemiol 1972; 1: 89-92. Clausen JL, ed: Pulmonary function testing: guidelines and controversies. New York: Academic Press, 1982; l-367. Graham WGB, O’Grady RV, Dubuc 8: Pulmonary function loss in Vermont granite workers: a long term follow-up and critical appraisal. Am Rev Respir Dis 1981; 123: 25-28. Ferris BG: Epidemiology standardization project. Am Rev Respir Dis 1978; G(part 2):1-20. Becklake MR, Leclerc M, Strobach H, Swift J: The N2 closing volume test in population studies: sources of variation and reproducibility. Am Rev Respir Dis 1975; 111: 141-147. Ducic S, Swift J, Martin R, Macklem PT: Appraisal of a new test: between-technician variation In the measurement of closing volume. Am Rev Respir Dis 1975; 112: 621-627. Snider GL, Rieger RA, Desmas T, et al: Variation in the measurement of spirograms. Am J Med Sci 1967; 254: 679-684. Leech JA, Ghezzo H, Stevens DA, Becklake MR: Respiratory pressures and function in young adults. Am Rev Respir
The American
Journal
of Medicine
Volume
80
1163
CONCEPTS
19.
20.
21. 22.
23. 24.
25.
26.
27.
28.
29.
30.
31.
32.
33.
34.
35. 36. 37.
38.
39.
40.
1164
OF NORMALITY
APPLIED
TO LUNG
FUNCTION
TESTS-BECKLAKE
Dis 1983; 128: 17-23. Nickerson BG, Lemen RJ, Gerdes CB, Wegmann MJ, Robertson G: Within-subject variability and percent change for significance of spirometry in normal subjects and patients with cystic fibrosis. Am Rev Respir Dis 1980; 122: 859-866. Perks WH, Sopwith T, Brown D, Jones CH, Green M: Effects of temperature on Vitalograph spirometer readings. Thorax 1983; 38: 592-594. Pincock AC, Miller MR: The effect of temperature on recording spirograms. Am Rev Respir Dis 1983; 128: 894-898. Schoenberg JB, Beck GL, Bouhuys A: Growth and decay of pulmonary function in healthy blacks and whites. Respir Physiol 1978; 33: 367-393. Quanjer P, ed: Standardized lung function testing. Bull Eur Physiopathol Respir 1983; 19(suppl 5): 7-92. Gardner RM, Chairman: ATS Statement. Snowbird workshop on standardization of spirometry. Am Rev Fiespir Dis 1979; 119: 831-838. ATS Committee on proficiency standards for clinical pulmonary laboratories: Pulmonary function laboratory personnel standards. Am Thorac Sot News 1982; 8: 8-9. Gardner RM, Hankinson JL, West BJ: Evaluating commercially available spirometers. Am Rev Respir Dis 1980; 121: 73-82. Diem JE, Jones RN, Hendrick DJ, et al: Five year longitudinal study of workers employed in a new toluene diisocyanate manufacturing plant. Am Rev Respir Dis 1982; 128: 420-428. Eisen E, Wegman DH, Louis TA: Effect of selection in a prospective study of forced expiratory volume in Vermont granite workers. Am Rev Respir Dis 1983; 128: 587-591. Becklake MR: Epidemiologic studies in human populations. In: Witschi HP, Brain JD, eds. Handbook of experimental pharmacology, vol 75. The toxicology of inhaled materials. Berlin: Springer-Verlag, 1985; 115-147. Pierson DJ, Dick NP, Petty TL: A comparison of spirometric values with subjects in standing and sitting positions. Chest 1976; 70: 17-20. Cotes JE: Lung function: assessment and application in medicine, 4th ed. Oxford: Blackwell Scientific Publications, 1978; I-605. Hetzel MR, Clark TJH: Comparison of normal and asthmatic circadian rhythms in peak expiratory flow rate. Thorax 1980; 35: 732-738. McKerrow CB, Rossiter CE: An annual cycle in the ventilatory capacity of men with pneumoconiosis and of normal subjects. Thorax 1968; 23: 340-349. Leidinger MJ, Cota KA, Taussig LM: Lung function throughout the menstrual cycle in normal and asthmatic women (abstr). Chest 1979; 76: 372. Gibbs CJ, Coutts H, White RJ: Bronchial reactivity during the menstrual cycle (abstr). Thorax 1984; 39: 232. Myers JE: Differential ethnic standards for lung functions, or one standard for all? S Afr Med J 1984; 65: 768-772. Burrows 8, Knudson RJ, Lebowitz MD: The relationship of childhood respiratory illness to adult obstructive airway disease. Am Rev Respir Dis 1977; 115: 751-760. A report of the Surgeon General: The health consequences of smoking. Chronic obstructive lung disease. U.S. Department of Health and Human Services, Public Health Service, Rockville, Maryland, 1984; 17-180. Weiss ST, Tager IB, Schenker M, Speizer FE: The health effects of involuntary smoking. Am Rev Respir Dis 1983; 128: 933-942. Kauffmann F, Drouet D, Lellouch J, Brille D: Occupational
June
1966
The American
Journal
of Medicine
Volume
41.
42.
43.
44.
45.
46.
47. 48.
49.
50.
51.
52.
53. 54.
55.
56.
57.
58. 59.
60.
60
exposure and 12 year spirometric changes among Paris area workers. Br J Ind Med 1982; 39: 221-232. Ware JH, Dockery DW, Spiro A, Speizer FE, Ferris BG: Passive smoking, gas cooking and respiratory health of children living in six cities. Am Rev Respir Dis 1984; 129: 366-374. Speizer FE, Tager IB: Epidemiology of chronic mucus hypersecretion and obstructive airways disease. Epidemiol Rev 1979; 1: 124-142. Ferris BG, Anderson DO, Zinkmantel R: Prediction values for screening tests of respiratory function. Am Rev Respir Dis 1965; 91: 252-261. Cotes JE, Rossiter CE, Higgins ITT, Gilson JG: Average normal values for the forced expiratory volume in white Caucasian males. Br Med J 1966; 1: 1016-1019. Morris JF, Koski A, Johnson LC: Spirometric standards for healthy nonsmoking adults. Am Rev Respir Dis 1971; 103: 57-67. Cherniak RM, Raber MB: Normal standards for ventilatory function using an automated wedge spirometer. Am Rev Respir Dis 1972; 106: 38-46. Higgins MW, Keller JB: Seven measures of ventilatory lung function. Am Rev Respir Dis 1973; 108: 258-272. Crapo RO, Morris AH, Gardner RM: Reference spirometric values using techniques and equipment that meets ATS recommendations. Am Rev Respir Dis 1981; 123: 659-664. Knudson RJ, Lebowitz MD, Holberg CV, Burrows B. Changes in the normal maximal expiratory flow-volume curve with growth and aging. Am Rev Respir Dis 1983; 127: 725-734. Grimes CA, Hanes B: Influence of cigarette smoke on the spirometer evaluation of employees of a large insurance company. Am Rev Respir Dis 1973; 108: 273-282. Burrows B, Knudson RJ, Cline MG, Lebowitz MD: Quantitative relationships between cigarette smoking and ventilatory function. Am Rev Respir Dis 1977; 115: 195-205. Berry G: Statistical analysis. In: Weill H, Turner-Warwick M, eds. Occupational lung diseases: research approaches and methods. New York: Marcel Dekker, 1981; 427-464. Glindmeyer HW: Predictable confusion. J Occup Med 1981; 23: 845-849. Burrows B, Cline MG, Knudson RJ, Taussig LM, Lebowitz MD: A descriptive analysis of the growth and decline of the FVC and FEV,. Chest 1983; 83: 717-724. Becklake MR, Permutt S: Evaluation of tests of lung function for screening for early detection of chronic obstructive lung disease. In: Macklem PT, Permutt S, eds. The lung in the transition between health and disease. New York: Marcel Dekker, 1979; 345-388. Hurwitz S, Allen J, Liben A: Lung function in young adults: evidence for differences in the chronological age at which the various functions start to decline. Thorax 1980; 35: 615-619. Woolcock AJ, Leeder SR, Armstrong JG, Peat JK, Colman M, Cullen KJ: The singlebreath nitrogen test in rural and urban smokers and nonsmokers. Bull Eur Physiopathol Respir 1978; 14: 127-135. Cochrane AL, Holland WW: Validation of screening procedures. Br Med Bull 1971; 27: 3-8. National Heart, Lung and Blood Institute (RFP-NHLBI-84-11): Early intervention for chronic obstructive pulmonary disease. Washington: U.S. Department of Health and Human Services, 1984. Sackett DL: Clinical diagnosis and the clinical laboratory. Clin Invest Med 1978; 1: 37-43.
Lung function testing has become an integral part of the clinical assessment of pulmonary disease. However, increasing application of such testing in research into the origins of disease as well as in the assessment of health raises the issue of what is meant by normal respiratory function. In essence, this depends on the sources of variation in lung function measurements that are of interest (the signal) and those that are not (the noise). For instance, clinicians are primarily interested in variation due to disease, all other sources of variation being considered noise. Physiologists are concerned with sources of variation other than disease, and the interest of epidemiologists (and their definition of normality) varies according to the specific objectives of each particular study, report or program. Consideration of the sources of variation in lung function (within a subject, between subjects, and between populations) not only is useful in clarifying the concepts of normality for clinical application, physiologic studies, and epidemiologic purposes, but also is mandatory for an understanding of lung function in the transition between health and disease, a major thrust in all three areas of endeavour.
MARGARET R. BECKLAKE, M.D., F.R.C.P.(London) Montreal,
Quebec,
Canada
From the Departments of Epidemiology and Biostatistics, and of Medicine, McGill University, Montreal, Quebec, Canada. This work was supported by the Medical Research Council of Canada. Dr. Becklake is a Career Scientist of the Medical Research Council (Canada). Requests for reprints should be addressed to Dr. Margaret R. Becklake, Department of Epidemiology and Biostatistics, McGill University, 1110 Pine Avenue West, Montreal, Quebec, H3A lA3, Canada. Manuscript accepted March 7, 1985.
1158
June
1986
The American
Journal
Lung function testing, like chest radiography, has become an integral part of the clinical assessment of pulmonary disease [l]. In the clinical context, particularly when the disease process is advanced and when lung function information can be assessed in light of all other available information for an individual patient, interpretation of the results as normal or abnormal is usually not difficult. However, two developments have served to underline the need to clarify our concepts of normality. First is the increasing research effort into the origins of chronic nonmalignant respiratory disease, triggered by the evidence that the disease processes leading to disablement from chronic airflow limitation in middle-age start in early adulthood [2,3]. Second is the application of lung function testing in the assessment of health as opposed to disease, for instance in preemployment and annual physical examinations. The commercial availability of computerized apparatus for lung function testing has resulted in many such evaluations being carried out in workplaces and/or occupational groups, often by health personnel untrained in their use who rely on criteria and reference standards selected by the manufacturer and incorporated into the software of the apparatus as purchased. The purpose of this review is to clarify as far as possible the concepts of normality applied to the measurement of lung function; criteria of normality and the issue of what is “normal” including confidence limits have been addressed in several recent papers [4-61 and are not discussed here.
of Medicine
Volume
80
CONCEPTS
TABLE instrument Procedure
Observer Subject Interactions Other
I
Sources of Variation in Lung Function to Technical Factors
APPLIED
TO LUNG
FUNCTION
TESTS-BECKLAKE
with their search for (and research into) the factors responsible for the distribution of attributes (including abnormality and/or dysfunction and/or disease) within populations, with a view to developing control or preventive strategies. For instance, the American Thoracic Society proposed guidelines as to what constitutes an adverse respiratory health effect in response to envrionmental pollutants [9] to take into account the full range of potential response, including (1) increased pollutant burden without detectable health impact; (2) physiologic changes of uncertain significance; (3) pathophysiologic changes; (4) morbidity, and (5) mortality. Only the last three are considered adverse health effects.
Due
Within and between instruments [ 121 Number of trials; choice of results to be reported, e.g., best of three, average of best two of five, and so on [14,15] Administration of tests [ 161; evaluation of results [171 Comprehension; cooperation [18,19] Subject-observer; observer-instrument [ 161 Temperature [20,21]; altitude [12]
References indicate various components results and, hence, error.
OF NORMALITY
studies documenting the contribution of the of measurement to-variation in spirometric their potential as causes of measurement
VARIATION IN MEASUREMENT BASIC CONCEPTS The meaning of the word normal varies with the context in which it is used. In the popular (journalistic) use, it generally means ideal, conventional, or habitual. Statisticians use it in a numerical sense to describe distribution about a middling tendency. In the biologic context, it is concerned with the issue of human variation (which can, on occasions, be extreme) and the factors that account for it. Nevertheless, even in the various biomedical sciences, there is little unanimity about its meaning, perhaps because it involves defining health, itself an elusive concept. Thus the widely cited definition in the Constitution of the World Health Organization (1948) as “a state of complete physical, mental and social well being and not merely the absence of disease or infirmity” needs elaboration in specific contexts. To anatomists, normal variations in structure are those consistent with the maintenance of good health. To physiologists, normal variation of bodily functions are those that maintain the stability of the “internal milieu”: the focus is on the factors other than disease that cause these variations. Clinicians, on the other hand, concentrate on variations due to disease. They therefore wish to identify (so that they can disregard) variations that they loosely call physiologic. Interestingly enough, the word disease is not defined in most standard textbooks of medicine, although several authors [7,8] have proposed thoughtful if somewhat cumbersome definitions. Burrows [7], elaborating on earlier definitions, emphasized the distinction between diagnosis and disease, which he defined as “the sum of those abnormal phenomena displayed by a living organism that place the organism at a biological disadvantage and that are associated with one another on the basis of their frequent occurrence in other members of the species or on the basis of common causality.” By contrast, epidemiologists and those who practice public health have no general or universal definition for health or disease but define the terms precisely for each study, report, or program they undertake. This is in line
June
Information theory, which distinguishes signal (the information a test is designed to measure) from noise (measurement error), has been applied to other clinical diagnostic tests [IO]. In addition to measurement error, most if not all clinical measurements also comprise biologic variation (the basis of most biologic sciences) as well as variation due to disease if this is present (the basis of clinical medicine). To evaluate the contribution of the latter, clinicians must be aware of, and take into account, the former. Not only is there often little appreciation by clinicians of the range of both measurement error and biologic variation, but also there is often unjustified faith in the specificity of laboratory tests for detecting disease, particularly values measured on a continuous scale such as measurements of lung function. As a result, clinicians continue to strive for more precise definitions of health (or normality), and yet all the evidence suggests that the test ranges for disease, even frank disease, are hidden in the normal distribution curve-or, at best, deface the descending limb of the curve [ 1 I]. MEASUREMENT ERROR Measurement error is usually assessed as short-term repeatability of a test result within a time frame over which biologic variation is unlikely. Its magnitude varies according to the nature and complexity of the test [ 121. Table I lists the various technical components in the measurement of lung function; each has been shown to contribute to measurement error [ 13-221. For forced vital capacity, for instance, good laboratories maintain the variation between repeated measurements in normal subjects at approximately 3 percent [ 121, and for forced expiratory flow at 50 percent of forced vital capacity, at approximately 6 percent [ 121; for closing volume and for the slope of phase Ill in the single-breath nitrogen test, values closer to 20 percent and 50 percent, respectively, can be expected [ 121. When measurement error is random, it affects the internal validity of a study, i.e., diminishes the chances of detecting differences between study groups; when it is systematic, it affects the external
1986
The American
Journal
of Medicine
Volume
80
1159
CONCEPTS
TABLE
OF NORMALITY
II
APPLIED
TO LUNG
FUNCTION
TESTS-BECKLAKE
Sources of Biologic Variation in Measurement of Lung Function*
tween laboratories, it may also be responsible for failure to detect certain types of abnormality. For instance, the inability of a subject to produce a repeatable forced expiratory volume result in compliance with American Thoracic Society standards may, in itself, reflect abnormality and/or incipient disease. Thus, in a prospective study of Vermont granite workers, men who were unable to do so showed a greater loss of function over the next four years than men who were able to perform satisfactorily [28]. This point could have been missed had the study excluded subjects who were unable to perform a “satisfactory” test in accordance with American Thoracic Society instructions.
Source
Variation
Measurement error (includes all the technical factors listed in Table I) Diurnal (circadian) and seasonal effects
Intra-subject
[31-331 Endocrinologic effects [34,35] All the above Personal Size, sex, age [12,31]; Physical activity 1311, muscularity [ 181 Race and other genetic characteristics
Inter-subject
[31,36] Past and present health [ 1,371 Environmental Cigarette exposure, voluntary [31,38] and involuntary [39] Occupation [3 1,401 Residence, urban or rural [31] Urban air pollution [1,31], home pollution
BIOLOGIC VARIATION Biologic variation can, of course, be detected only if it exceeds variation due to measurement error. It can conveniently be thought of as intra-subject, inter-subject, and inter-population, depending on the focus of interest (see Table II). Besides measurement error, intra-subject variation may be due to factors such as posture [30], relationships to meals [31], circadian or diurnal variation [32], seasonal variation [33] possibly due to reactivity of airways to cold, and perhaps endocrinologic factors such as those that accompany the menstrual cycle [34,35]. All sources of intra-subject variation already mentioned may also cause inter-subject variation. Differences between subjects may also be due to differences in certain of their personal characteristics, These include age and sex, stature as reflected in height and weight [ 121, muscularity in so far as it affects respiratory muscle performance [ 121, level of physical activity and training [3 I], race and other genetic characteristics [ 121, and past and present health experience [ 1,371. Of the environmental factors, cigarette smoking is clearly the most important [ 1,31,38], including the effects of involuntary smoking [39]. Other environmental factors including occupation [1,40], exposure to urban air pollution [31] including home pollution [41], place of residence (urban or rural), and/or socioeconomic status [37,41,42], all of which may affect lung function. All the sources of intra- and inter-subject variation already mentioned may also cause inter-population variation insofar as they determine the inclusion of subjects in, or their exclusion from, one population or group compared with another. Even the frequently used grouping into smokers and nonsmokers may be bedeviled by this type of bias, since the decision to take up cigarette smoking appears to be related to lung function status, those who decide to do so having, on average, better function than those who do not [ 181. Likewise, the “healthy worker” effect describes the fact that employed persons often have a better health experience, including better lung function, than reference groups, again implying a selection process of the more fit into the work force [29]. Inter-
1411 Socioeconomic status [36,40,42] All the above Selection factors determining inclusion in, or exclusion from, the study populations 123,401
Inter-population
* Modified with permission from [29]; references indicate studies documenting the contribution of the various factors mentioned to biologic variation in lung function.
validity of a study, i.e., decreases the comparability of study results with those published elsewhere. Even in the case of forced vital capacity (perhaps the most widely used lung function test), all the components of measurement listed in Table I have been shown to contribute to measurement error. As a result, professional societies such as the American Thoracic Society and the Societas European Physiologica Respiratoire have developed guidelines [14,23] to minimize error in each component of measurement. These include: (1) criteria for instrument performance and advice on calibration procedures [ 14,231; (2) recommendations on procedure including number of trials to be carried out, criteria for technical rejection and selection of results to be reported, posture, use of nose-clips, and so on [23,24]; and (3) guidelines for technician training [25] and evaluation of results, although with the widespread introduction of microprocessors, this component of measurement error has diminished considerably [ 121. Lists of apparatus available [ 121 and reviews of their performance [26] are also published. Despite this, measurement error [13] and unexplained measurement drift over time [27] occur in even the most experienced laboratories and must be continually borne in mind in the interpretation of results. Although standardization of methodology has undoubtedly been remarkably successful in diminishing measurement error and increasing comparability of results be-
1160
June
1966
The American
Journal
of Medicine
Volume
60
CONCEPTS
OF NORMALITY
Proportion of Variation In Forced Vital Capacity Attributable Selected Published Studies of Adult White Men Reference fi”41 /:65;
Number of Subjects
Regression Age
Age Range
2751 156 517 870
25-74 20-64 15-79 20-84
-0.027 -0.032 -0.025 -0.014
[=I
2,150t 194
20-74 18-80
[481
125
15-91
-0.024 Nonlinear, weight -0.021
iii]
329t 86
25-70 17-35
[471
APPLlED
TO LUNG
to identified
FUNCTION
Personal
TESTS--BECKLAKE
Characteristics
In
Proportion of Variation” Explained Not Explained
Coefficients Height 0.051 0.046 0.048 0.058
0.67 0.41 0.33 0.42
0.33 0.59 0.67 0.58
0.062 includes and interactions 0.060
0.38 0.56
0.62 0.44
0.54
0.46
Nonlinear, -0.030 includes 0.084 weight, respiratory pressures, and smoking
0.71 0.29
0.71 0.29
* Calculated as R* and 1 - R*, respectively. + Includes smokers.
population variation due to health selection factors is also clearly relevant in the selection of reference lung function values, to be discussed later. PROPORTION ATTRIBUTABLE
OF INTER-SUBJECT VARIATION TO VARIOUS FACTORS
The proportion of inter-subject variation in one measure of lung function, the forced vital capacity, that can be explained on the basis of variation in personal characteristics such as age, height, and/or weight in selected studies [18,22,43-491 is shown in Table Ill. The table refers to data from studies on adult white men judged by more or less stringent criteria to be in good health; included are some of the more frequently used reference values applied in routine hospital laboratories [22,45-491. Table Ill also shows the proportion of variation in each study population that was explained by these personal characteristics expressed as the squared overall correlation coefficient (r*); this varied considerably between studies, from 0.29 to 0.7 1. The proportion not explained by these variables, i.e., 1 - r*, which varied from 0.29 to 0.71, must therefore be attributable to factors other than the personal characteristics contained in the regressions. Included would be those listed in Table II and presumably others as yet unidentified; indeed, much research into the origins of chronic airflow limitation is directed towards identifying and quantitating some of these unidentified sources of variation. Table IV shows estimates of the proportion of intersubject variation in forced vital capacity attributable to the sources already discussed; it is based on the information contained in Tables II and Ill and also on other published data [29]. Approximately 22 percent of overall variation can reasonably be attributed to variation in body characteristics reflected in height and weight, approximately 8 percent can be attributed to age (if the age span covers childhood to old age), and approximately IO percent can
June
be attributed to race [31,36]. Within the residual (i.e., unexplained) variation, estimated at approximately 27 percent, must lie all the other factors, personal and environmental, known to contribute to inter-subject variation. Smoking is probably the most important single factor, accounting for perhaps a third to a half of the residual variation [22,46,47]; a further 5 percent may perhaps be due to childhood illnesses (before the age of 16 years) [37], approximately 5 percent may be due to variation in respiratory pressures [ 181, and the remainder is due to other factors known and unknown. SIGNAL
VERSUS
NOISE
Do the sources of variation listed in Table IV represent variation in “normal” function or do they represent “abnormal” variation? This depends on the sources of variation that are of interest (the signal) and the sources that are not (the noise), and this in turn depends on the context in which the question is asked. For instance, clinicians interested in the diagnosis of disease and its precursors
TABLE IV
Estimates of Proportion of inter-Subject Variation in Forced Vital Capacity Attributed to Identified Factors* Factor
Sex Age Height Weight Race Technical (measurement Unexplained+ Total
Proportion of Attributable
Variation
up to 0.30 0.08 0.20 0.02 1
up to 0.30 0.10 0.03 0.27 1.00
error)
* Based on published data in Tables II and Ill and [29]. t Residual variation, includes ail the factors listed in Table II such as smoking (active and passive), childhood illnesses, respiratory pressures, and so on.
1986
The American
Journal
of Medicine
Volume
80
1161
CONCEPTS
OF NORMALITY
APPLIED
TO LUNG
FUNCTION
TESTS-SECKLAKE
vfoulci consider all of the factors listed on Table II as “normal” variation (or noise), with the possible exception of past and present health. In practice, they therefore seek a convenient way to take them all into account (usually through prediction formulas) so that variation due to disease (the signal) can be identified. To reinforce antismoking advice, they may also wish to include the effects of smoking as part of the signal and will therefore seek reference values (and prediction formulas) based on nonsmoking subjects. On the other hand, if the objective is to identify the influence of environmental agents over and above that of smoking, then the effects of smoking become part of the noise, and it may be useful to use prediction formulas that take it into account [50,51]. Similar issues arise in the context of physiologic and epidemiologic studies. For instance, in a physiologic study to examine the influence of posture on lung function (the signal), the effects of postprandial changes in lung volume would be considered noise. By contrast, in another study, the effects of ingesting a meal might be the signal, and the effects of posture would be noise. Similarly, in an epidemiologic survey of a work force, the effects of an occupational exposure might be the signal in one study, with smoking included in the noise; in another study, their roles might be reversed.
WHY DO REFERENCE VALUES VARY? Reference values are usually based on the analysis of pertinent personal and environmental factors in populations judged to be healthy according to more or less wellspecified criteria, which may vary from study to study [23]. The analysis is also usually, as Rurrows et al [54] have pointed out, carried out for descriptive rather than predictive purposes. Of the sources of inter-population differences likely to affect derived prediction formulas, some are obvious-inter-population differences in history of previous respiratory illness, as well as differences in place of residence, socioeconomic status, and environmental pollution levels. Less obvious are the frequency distribution within the population of the factors analyzed and the method of analysis [55]. $or instance, for the studies listed in Table Ill, the age coefficient was greatest (and, by implication, the rates of aging fastest) in Cotes et al’s [44] subjects and least in Cherniak and Raber’s [46] subjects (implying the slowest rates of aging). These differences can, however, be reasonably attributed to differences in the age distribution of the two study populations; in the latter study 1461, the population was weighted towards the younger ages, with more than half of the subjects 35 years or less, and this weighting is obviously reflected in the regression coefficient. In the former study [44], the opposite pertained, with more than half of the subjects between 55 and 65 years of age. Likewise, the method of analysis used also influences the findings. For instance, if linear analyses are used to describe the agerelated changes in studies confined to adults, over-prediction of values in young persons is likely [22]; nonlinear analyses more correctly model the age-related changes that occur in early adulthood [6,18,22,54-561 and should therefore be used in dealing with persons under the age of 35 years. CHOICE OF REFERENCE VALUES
DEALING Wll-bj NOISE The general approach to dealing with noise will depend on the context, in particular whether the issue is clinical or epidemiologic. In either event, it is important to identify all relevant potential sources of noise to be taken into account. As already indicated, measurement error will always be noise and should be minimized by the techniques referred to earlier. In population (epidemiologic) studies, comparison between population groups is usually required to test the hypothesis‘under study. For instance, to assess the effects of an occupational exposure, exposed and non-exposed groups may be compared with all other sources of inter-population differences taken into account. This may be done (1) by matching, for example, by smoking habits and age, although this may not be possible when groups are small; (2) by statistical analysis, for instance, using the various forms of covariance and regression analysis; or (3) by the use of outside reference values that take into account some of the sources of variation usually considered noise, for example, age and stature [52]. The first two methods are commonly used in epidemiologic studies, and the third is used in clinical laboratories. However, because of the considerable variation between published reference values or prediction formulas, graphically described by Glindmeyer [53] as “predictable confusion” a clinical laboratory director or epidemiologist in charge of a study faces a major dilemma in ‘making the most appropriate choice.
1162
June
1966
The
American
Journal
of Medicine
Volume
Many would agree with the conclusion reached by Clausen [ 121 in a recently published and valuable text on pulmonary function testing that “because of inexplicable differences (in published normal values), there is no recommended set of prediction equations applicable to all laboratories and all patient populations.” Nevertheless, a reasoned choice made by a responsible laboratory director (or epidemiologist) is preferable to leaving the choice to the whim and/or judgment of the manufacturers of automated equipment. Certain guidelines can be offered based on methodologic, epidemiologic, and conceptual considerations. The first is obvious: the methods used (apparatus, procedure, and analysis) to generate the reference values should be comparable-and, preferably, identical-to those to be used in the laboratory seeking to apply the reference values. Second, the population (or sample) that generated
60
CONCEPTS
the reference values should be as closely matched as possible to the population from which the patients or subjects to be assessed are drawn. For instance, for the residents of Alice Springs, Queensland, Australia, their lung function may be more appropriately assessed in relation to that of the residents of Busselton, West Australia [57] who do not smoke by inclination, than in relation to that of either the residents of Tucson, Arizona, who also do not smoke by inclination [49], or the residents of Salt Lake City, Utah, who do not smoke by religious conviction [48]. Third, and most important, although usually least clearly articulated, is the conceptual issue of the purpose underlying the comparison to be made. If it is for clinical (diagnostic) purposes, the choice of reference values, paradoxically, is probably relatively unimportant, because the tests of function are simply one item of information in an array of other clinical information about an individual subject upon which an assessment will be based. A good clinician will assign the correct weighting to a borderline lung function result in reaching a final opinion. By contrast, borderline values may be much harder to interpret in the context of a health maintenance (or periodic) examination. Such an examination has the characteristics of a screening examination, including ethical as well as scientific considerations [58]. In contrast to the clinical context in which the subject seeks out the physician who is not therefore, to use the graphic phrase of Cochrane and Holland [58], “responsible for defects in medical knowledge,” the opposite applies in the context of a screening examination. Here, the criteria for subject (case) identification have nothing to do with “normal” variation but depend on whether a given test value identi-
OF NORMALITY
APPLIED
TO LUNG
FUNCTION
TESTS-BECKLAKE
fies a condition at a stage at which intervention (for example, treatment or removal from exposure) would improve the subject’s future health prospects. So far, there is insufficient information on which to base the use of lung function tests as a screening tool, although this has been identified as a priority area for study [59]. Meanwhile, lung function tests continue to be used and interpreted daily in clinical laboratories. Action based on these interpretations is likely to be reasonable, given that clinicians integrate these results into an overall medical judgment based on all available information about the subject. Indeed, it cannot be too strongly emphasized that these judgments should be based more on clinical experience than on strict adherence to defined (usually arbitrary) limits of “normality.” Numerical (statistical) criteria for such limits are also hampered by the fact that the greater the number of laboratory tests of lung function or any other bodily function [60] performed in healthy subjects, the greater the chance of abnormal results. Nor is there any evidence that further improvement in the precision of prediction formulas (that is, reduction in the proportion of unexplained variation) will occur until other sources of inter-subject variation are identified, quantified, and incorporated into the formulas. At present, therefore, interpretation of lung function test results in the clinical context should remain a matter of clinical judgment. ACKNOWLEDGMENT I thank the following colleagues for a critical review of the text: Dr. A. S. Buist (Portland, Oregon), Professor S. Benatar (Cape Town), and Drs. P. Ernst and R. Dales (Montreal).
REFERENCES
2. 3. 4.
5.
6.
7. a. 9.
Bates DV, Macklem PT, Christie RV: Respiratory function in health and disease, 2nd ed. Philadelphia: WB Saunders, 1977; 37. Fletcher C, Peto R: The natural history of chronic airflow obstruction. Br Med J 1977; 1: 1645-1648. Burrows B: An overview of obstructive lung disease. Med Clin North Am 1981; 65: 455-471. Miller A, Thornton JC: The interpretation of spirometric measurements in epidemiologic surveys. Environ Res 1980; 23: 444-468. Pennock BE, Cottrell JJ, Rogers RM: Pulmonary function testing. What is normal? Arch Intern Med 1983; 143: 2123-2127. Buist AS: Evaluation of lung function: concepts of normality. In: Simmons DH, ed. Pulmonology, vol 4. New York: John Wiley, 1982; 141-165. Burrows B: Standard pulmonary nomenclature and symbols. Ann Intern Med 1975; 83: 419-420. Campbell EJM, Scadding JG, Roberts RS: The concept of disease. Br Med J 1979; 2: 757-762. American Thoracic Society: Guidelines as to what constitutes an adverse respiratory health effect, with special reference to air pollution. Am Thorac Sot News 1984; 10: 6-8.
10.
11. 12. 13.
14. 15.
16.
17.
18.
June
1966
Diamond GA, Hirsch M, Forrester JS, et al: Application of information theory to clinical diagnostic testing. Circulation 1981; 63: 915-921. Cochrane AL: The history of the measurement of ill-health. Int J Epidemiol 1972; 1: 89-92. Clausen JL, ed: Pulmonary function testing: guidelines and controversies. New York: Academic Press, 1982; l-367. Graham WGB, O’Grady RV, Dubuc 8: Pulmonary function loss in Vermont granite workers: a long term follow-up and critical appraisal. Am Rev Respir Dis 1981; 123: 25-28. Ferris BG: Epidemiology standardization project. Am Rev Respir Dis 1978; G(part 2):1-20. Becklake MR, Leclerc M, Strobach H, Swift J: The N2 closing volume test in population studies: sources of variation and reproducibility. Am Rev Respir Dis 1975; 111: 141-147. Ducic S, Swift J, Martin R, Macklem PT: Appraisal of a new test: between-technician variation In the measurement of closing volume. Am Rev Respir Dis 1975; 112: 621-627. Snider GL, Rieger RA, Desmas T, et al: Variation in the measurement of spirograms. Am J Med Sci 1967; 254: 679-684. Leech JA, Ghezzo H, Stevens DA, Becklake MR: Respiratory pressures and function in young adults. Am Rev Respir
The American
Journal
of Medicine
Volume
80
1163
CONCEPTS
19.
20.
21. 22.
23. 24.
25.
26.
27.
28.
29.
30.
31.
32.
33.
34.
35. 36. 37.
38.
39.
40.
1164
OF NORMALITY
APPLIED
TO LUNG
FUNCTION
TESTS-BECKLAKE
Dis 1983; 128: 17-23. Nickerson BG, Lemen RJ, Gerdes CB, Wegmann MJ, Robertson G: Within-subject variability and percent change for significance of spirometry in normal subjects and patients with cystic fibrosis. Am Rev Respir Dis 1980; 122: 859-866. Perks WH, Sopwith T, Brown D, Jones CH, Green M: Effects of temperature on Vitalograph spirometer readings. Thorax 1983; 38: 592-594. Pincock AC, Miller MR: The effect of temperature on recording spirograms. Am Rev Respir Dis 1983; 128: 894-898. Schoenberg JB, Beck GL, Bouhuys A: Growth and decay of pulmonary function in healthy blacks and whites. Respir Physiol 1978; 33: 367-393. Quanjer P, ed: Standardized lung function testing. Bull Eur Physiopathol Respir 1983; 19(suppl 5): 7-92. Gardner RM, Chairman: ATS Statement. Snowbird workshop on standardization of spirometry. Am Rev Fiespir Dis 1979; 119: 831-838. ATS Committee on proficiency standards for clinical pulmonary laboratories: Pulmonary function laboratory personnel standards. Am Thorac Sot News 1982; 8: 8-9. Gardner RM, Hankinson JL, West BJ: Evaluating commercially available spirometers. Am Rev Respir Dis 1980; 121: 73-82. Diem JE, Jones RN, Hendrick DJ, et al: Five year longitudinal study of workers employed in a new toluene diisocyanate manufacturing plant. Am Rev Respir Dis 1982; 128: 420-428. Eisen E, Wegman DH, Louis TA: Effect of selection in a prospective study of forced expiratory volume in Vermont granite workers. Am Rev Respir Dis 1983; 128: 587-591. Becklake MR: Epidemiologic studies in human populations. In: Witschi HP, Brain JD, eds. Handbook of experimental pharmacology, vol 75. The toxicology of inhaled materials. Berlin: Springer-Verlag, 1985; 115-147. Pierson DJ, Dick NP, Petty TL: A comparison of spirometric values with subjects in standing and sitting positions. Chest 1976; 70: 17-20. Cotes JE: Lung function: assessment and application in medicine, 4th ed. Oxford: Blackwell Scientific Publications, 1978; I-605. Hetzel MR, Clark TJH: Comparison of normal and asthmatic circadian rhythms in peak expiratory flow rate. Thorax 1980; 35: 732-738. McKerrow CB, Rossiter CE: An annual cycle in the ventilatory capacity of men with pneumoconiosis and of normal subjects. Thorax 1968; 23: 340-349. Leidinger MJ, Cota KA, Taussig LM: Lung function throughout the menstrual cycle in normal and asthmatic women (abstr). Chest 1979; 76: 372. Gibbs CJ, Coutts H, White RJ: Bronchial reactivity during the menstrual cycle (abstr). Thorax 1984; 39: 232. Myers JE: Differential ethnic standards for lung functions, or one standard for all? S Afr Med J 1984; 65: 768-772. Burrows 8, Knudson RJ, Lebowitz MD: The relationship of childhood respiratory illness to adult obstructive airway disease. Am Rev Respir Dis 1977; 115: 751-760. A report of the Surgeon General: The health consequences of smoking. Chronic obstructive lung disease. U.S. Department of Health and Human Services, Public Health Service, Rockville, Maryland, 1984; 17-180. Weiss ST, Tager IB, Schenker M, Speizer FE: The health effects of involuntary smoking. Am Rev Respir Dis 1983; 128: 933-942. Kauffmann F, Drouet D, Lellouch J, Brille D: Occupational
June
1966
The American
Journal
of Medicine
Volume
41.
42.
43.
44.
45.
46.
47. 48.
49.
50.
51.
52.
53. 54.
55.
56.
57.
58. 59.
60.
60
exposure and 12 year spirometric changes among Paris area workers. Br J Ind Med 1982; 39: 221-232. Ware JH, Dockery DW, Spiro A, Speizer FE, Ferris BG: Passive smoking, gas cooking and respiratory health of children living in six cities. Am Rev Respir Dis 1984; 129: 366-374. Speizer FE, Tager IB: Epidemiology of chronic mucus hypersecretion and obstructive airways disease. Epidemiol Rev 1979; 1: 124-142. Ferris BG, Anderson DO, Zinkmantel R: Prediction values for screening tests of respiratory function. Am Rev Respir Dis 1965; 91: 252-261. Cotes JE, Rossiter CE, Higgins ITT, Gilson JG: Average normal values for the forced expiratory volume in white Caucasian males. Br Med J 1966; 1: 1016-1019. Morris JF, Koski A, Johnson LC: Spirometric standards for healthy nonsmoking adults. Am Rev Respir Dis 1971; 103: 57-67. Cherniak RM, Raber MB: Normal standards for ventilatory function using an automated wedge spirometer. Am Rev Respir Dis 1972; 106: 38-46. Higgins MW, Keller JB: Seven measures of ventilatory lung function. Am Rev Respir Dis 1973; 108: 258-272. Crapo RO, Morris AH, Gardner RM: Reference spirometric values using techniques and equipment that meets ATS recommendations. Am Rev Respir Dis 1981; 123: 659-664. Knudson RJ, Lebowitz MD, Holberg CV, Burrows B. Changes in the normal maximal expiratory flow-volume curve with growth and aging. Am Rev Respir Dis 1983; 127: 725-734. Grimes CA, Hanes B: Influence of cigarette smoke on the spirometer evaluation of employees of a large insurance company. Am Rev Respir Dis 1973; 108: 273-282. Burrows B, Knudson RJ, Cline MG, Lebowitz MD: Quantitative relationships between cigarette smoking and ventilatory function. Am Rev Respir Dis 1977; 115: 195-205. Berry G: Statistical analysis. In: Weill H, Turner-Warwick M, eds. Occupational lung diseases: research approaches and methods. New York: Marcel Dekker, 1981; 427-464. Glindmeyer HW: Predictable confusion. J Occup Med 1981; 23: 845-849. Burrows B, Cline MG, Knudson RJ, Taussig LM, Lebowitz MD: A descriptive analysis of the growth and decline of the FVC and FEV,. Chest 1983; 83: 717-724. Becklake MR, Permutt S: Evaluation of tests of lung function for screening for early detection of chronic obstructive lung disease. In: Macklem PT, Permutt S, eds. The lung in the transition between health and disease. New York: Marcel Dekker, 1979; 345-388. Hurwitz S, Allen J, Liben A: Lung function in young adults: evidence for differences in the chronological age at which the various functions start to decline. Thorax 1980; 35: 615-619. Woolcock AJ, Leeder SR, Armstrong JG, Peat JK, Colman M, Cullen KJ: The singlebreath nitrogen test in rural and urban smokers and nonsmokers. Bull Eur Physiopathol Respir 1978; 14: 127-135. Cochrane AL, Holland WW: Validation of screening procedures. Br Med Bull 1971; 27: 3-8. National Heart, Lung and Blood Institute (RFP-NHLBI-84-11): Early intervention for chronic obstructive pulmonary disease. Washington: U.S. Department of Health and Human Services, 1984. Sackett DL: Clinical diagnosis and the clinical laboratory. Clin Invest Med 1978; 1: 37-43.