Indian J Med Res 122, August 2005, pp 153-164 Applicability of commonly used Caucasian prediction equations for spirometry interpretation in India A.N. Aggarwal, Dheeraj Gupta, Digamber Behera & S.K. Jindal Department of Pulmonary Medicine, Postgraduate Institute of Medical Education & Research Chandigarh, India Received December 30, 2003 Background & objectives: The applicability of Caucasian prediction equations in interpreting spirometry data in Indian patients has not been studied. The present study was undertaken to see if Caucasian and north Indian prediction equations can be used interchangeably while interpreting routine spirometric data. Methods: Forced vital capacity (FVC), forced expiratory volume in first second (FEV 1), and FEV1/FVC ratio were recorded from 14733 consecutive spirometry procedures in adults. Predicted values and lower limits of normality were calculated using regression equations previously derived at this centre, and four commonly used Caucasian equations described by Knudson, Crapo, European Community for Coal and Steel (ECCS) and the Third National Health and Nutrition Examination Survey (NHANES III). For men, 90 per cent of predicted values were also derived. Kappa estimates were used to study agreement, and Bland Altman analysis was performed to quantify differences, between interpretations from Indian and Caucasian equations. Receiver operating characteristic (ROC) curves were constructed to assess utility of using a fixed percentage of Caucasian predicted values in categorizing FVC or FEV1 as abnormal. Results: The use of Caucasian prediction equations (and 90% of predicted values in men) resulted in poor agreement with Indian equation in most height and age categories among both men and women. Bland Altman analysis revealed a large bias and wide confidence limits between Caucasian and Indian equations, indicating that the two cannot be used interchangeably. ROC analysis failed to yield good results with use of any single fixed percentage of Caucasian predicted value while categorizing FVC or FEV1. Interpretation & conclusion: Our results showed that the use of Caucasian prediction equations, or a fixed percentage of their predicted values, resulted in misinterpretation of spirometry data in a significant proportion of patients. There is a need to assess performance of more than one regression equation before choosing any single prediction equation. Key words Agreement - ethnic differences - prediction equation - respiratory function tests 153 154 INDIAN J MED RES, AUGUST 2005 The most important step in diagnosing abnormality of lung function in individuals is to define whether they are within or outside the healthy subjects range. For this purpose, the observed value of a patient is usually compared to a reference value obtained from prediction equations derived from studies on healthy people. It is often recommended that pulmonary function laboratories should employ prediction equations derived from subjects with a similar ethnic background as the patients1. Difficulties arise when either the patient population being investigated at a particular centre is markedly heterogeneous in terms of ethnic composition, or when no prediction equations are available for use in the patient population predominantly investigated at a centre. The problem gets compounded when computerized equipment used for pulmonary function testing provides a numeric output of results derived from a limited (and often only a single) set of prediction equations incorporated into the software. Often these equations are totally alien to the patient population being investigated. For example, the numeric output of results from spirometers built in the West, which are commonly used in India, is largely derived from Caucasian prediction equations. Though it is perhaps best to use different prediction equations for patients from different ethnic backgrounds, it is usually not possible to do so because of (i) non availability of spirometers with built-in Indian prediction equations, and/or modification of existing software, and (ii) extra time and manpower requirements for manual calculation of predicted values for each patient prior to actual interpretation of spirometry results. Clinicians in most instances, and even researchers, rely on the results obtained from Caucasian prediction equations incorporated into the software of spirometers, even though reference equations are available from several areas of India2,3. As an alternative, one can use a fixed percentage of the value derived from a Caucasian prediction equation when interpreting results for patients from a different ethnic background4. The percentages so recommended are largely arbitrary, and there has been no systemic analysis to show as to how well these approximate the predicted lung function in the patient population being studied. The present study was designed to assess, both qualitatively and quantitatively, the impact of applying commonly used Caucasian prediction equations for the interpretation of results of routine spirometry performed in patients in India. Material & Methods Study population: The Pulmonary Function Laboratory at the Postgraduate Institute of Medical Education & Research, Chandigarh, north India, offers spirometry as a routine service to both outpatients and inpatients, and approximately 3500 procedures are performed annually. Records of all consecutive patients referred for spirometry over a four year period (1999-2002) were retrieved. Reasons for performing spirometry as well as other clinical details were not analyzed further. Age, gender, height and spirometry data were recorded for all patients using a computer software previously developed by us5. A total of 15106 patients underwent spirometry during this period. After exclusion of 22 incomplete records and 351 patients aged less than 15 yr, 14733 records were available for further analysis. There were 8155 men (age 15-97 yr, height 123-215 cm) and 6578 women (age 15-90 yr, height 121-181 cm) (Table I). Spirometry and its interpretation: All subjects performed spirometry on a dry rolling seal spirometer (Spiroflow; P.K. Morgan Ltd.; Kent, UK). Spirometric indices such as forced vital capacity (FVC), forced expiratory flow in first second (FEV 1) and peak expiratory volume were measured using American Thoracic Society guidelines, and the highest measurements from among three technically acceptable and reproducible maneuvers were expressed at body temperature and pressure saturated with water vapor 6. Height was measured using a stadiometer and expressed to the nearest centimetre. The spirometer calibration was frequently checked to ensure performance. Predicted values for FVC, FEV 1 and FEV1/FVC ratio were generated using previously defined prediction equations, as detailed below. Lower limits of normal (LLN) for FEV1, FVC and FEV1/ FVC were calculated using lower 95 per cent confidence limits derived from the regression equation being used, and were computed as the difference between the predicted value and 1.645 times the standard error of estimate (SEE) or residual standard deviation (RSD) of the regression equation1. Any observed value lower than its corresponding LLN was considered abnormal. AGGARWAL et al: CAUCASIAN EQUATIONS FOR SPIROMETRY IN INDIA 155 Table I. Height and age distribution in the study population Men Women Total 116 (1.4) 1453 (17.8) 4420 (54.2) 2013 (24.7) 153 (1.9) 1981 (30.1) 3754 (57.1) 814 (12.4) 28 (0.4) 1 (0.0) 2097 (14.2) 5207 (35.3) 5234 (35.5) 2041 (13.9) 154 (1.0) 15-24 25-34 35-44 45-54 55-64 65-74 >75 817 (10.0) 1010 (12.4) 1357 (16.6) 1771 (21.7) 1629 (20.0) 1185 (14.5) 386 (4.7) 647 (9.8) 1103 (16.8) 1472 (22.4) 1526 (23.2) 1074 (16.3) 611 (9.3) 145 (2.2) 1464 (9.9) 2113 (14.3) 2829 (19.2) 3297 (22.4) 2703 (18.3) 1796 (12.2) 531 (3.6) Total 8155 6578 14733 Height (cm): <150 151-160 161-170 171-180 > 180 Age (yr): Values in parentheses are percentages FEV 1, FVC and FEV 1/FVC ratio were used as basic parameters to interpret spirometry data 1,5. A spirometry record with FEV 1/FVC ratio less than the LLN for that subject was categorized as having an obstructive pattern. Although a true restrictive defect can be diagnosed only on demonstration of a reduced total lung capacity, a restrictive pattern was inferred from spirometry results for categorization comparison purposes only. A record with FVC less than the LLN, associated with a normal FEV 1 /FVC ratio was categorized as having a restrictive defect. Static lung volumes and total lung capacity were not routinely estimated in subjects having restrictive defects on spirometry. Prediction equations: Four sets of prediction equations were used in this study. Regression equations previously derived at this centre, and routinely used at the Pulmonary Function Laboratory, were considered as the gold standard7. These equations were generated from spirometry studies performed on 962 healthy non smoking north Indian adults, aged 15-74 yr, using a water seal spirometer. Predicted values were also derived from four other previously described regression equations commonly used in Caucasian subjects. The Crapo equations were derived from 251 non smoking American subjects, aged 15-91 yr and residing in Utah 1400 m above sea level, using a water seal spirometer8. The Knudson equations were obtained from 746 American white non smoking subjects, aged 8-90 yr and residing in Arizona, using a pneumotachygraph device 9 . The European Community for Coal and Steel (ECCS) equations are summary equations derived for Caucasian subjects aged 5-70 yr from previously published reference equations 10. The NHANES III (the third National Health and Nutrition Examination Survey) equations are the most recently described of the four Caucasian equations evaluated, and were derived from 2281 Caucasian subjects randomly selected from 81 American counties, using a dry rolling-seal spirometer11. All five equations predict FEV1, FVC and FEV1/FVC based on gender, age and height of a subject as primary variables. All equations, except the Crapo and ECCS equations, are nonlinear with respect to age. In addition, it has been suggested that reducing corresponding Caucasian values by 10 per cent might approximate predicted values for FEV1 and FVC in north Indian men4; no similar guidelines exist for women. Hence, LLN for each parameter was also calculated after reducing predicted values obtained from Crapo, Knudson, ECCS and NHANES III equations by 10 per cent each. Thus nine expressions each of LLN for FEV1, FVC and FEV1/FVC were calculated for each subject (LLN from north Indian equation, and actual and 90% of LLN each from Crapo, Knudson, ECCS and NHANES III equations). 156 INDIAN J MED RES, AUGUST 2005 Statistical analysis: Spirometry was interpreted for each patient using each of the nine schemes for defining LLN. Spirometry results (normal, obstructive or restrictive patterns) obtained using north Indian and various Caucasian equations were compared after construction of contingency tables. Agreement between classification of spirometry results using north Indian and other equations were calculated using the Kappa estimate12. Bootstrapping was employed to check the robustness of Kappa estimate. For this purpose, study population was resampled with replacement 1000 times and Kappa estimates recalculated in each instance; 95 per cent confidence limits for Kappa estimate were then computed as the 2.5th and 97.5th percentile of the 1000 Kappa estimates obtained for each set of north Indian and Caucasian equations 13. The proportion of subjects having different interpretations using north Indian and another Caucasian equation was calculated for all combinations of prediction strategies as a measure of discordance. This group represented individuals whose interpretation changed if another scheme for calculating LLN was used in place of the standard north Indian equation. To assess how closely the LLNs obtained by Caucasian equations match those obtained by the north Indian equation, we calculated the difference (bias) between values estimated by the north Indian and each Caucasian equation. In order to assess if the values determined by the different prediction equations can be used interchangeably across different age and height groups, we calculated the limits of agreement for each set of north Indian and Caucasian equation 14. In addition, we also evaluated if using a fixed percentage of the Caucasian LLNs instead of north Indian LLNs can result in useful approximations. For this purpose we expressed the LLNs obtained by north Indian equations as a percentage of LLNs obtained by different Caucasian equations for both FVC and FEV1. We then calculated the mean of this value for both men and women in different age and height groups to search for any consistent trends. Receiver-operating characteristic (ROC) curves were also constructed to assess the utility of using such a fixed percentage as a discriminator in categorizing FVC or FEV 1 as abnormal. For this purpose sensitivity and specificity of each possible cut off percentage available from the data was calculated in classifying observed FVC or FEV 1 values as abnormal. The distribution of probability for false positives (1 - specificity) versus true positives (sensitivity) was plotted for each of the four sets of comparison data to obtain ROC curves; area under each curve was calculated using the trapezoidal method15. All mathematical computations and statistical procedures were performed using the software SPSS 10.0 (SPSS Inc., Chicago, IL). Results Using north Indian prediction equations, 4018 (27.3%) and 3901 (26.5%) spirometry records were interpreted as having obstructive and restrictive defects respectively. Among the 6814 normal records, 2728 (40.0%), 961 (14.1%), 699 (10.3%) and 3584 (52.6%) were classified as being abnormal using the Crapo, Knudson, ECCS and NHANES III equations respectively (Table II). Misclassification was worst for obstructive defects using the ECCS equation (31.9%) and worst for restrictive defects using the Knudson equation (21.0%) (Table II). Overall, the Crapo, Knudson, ECCS and NHANES III equations misclassified results in 23.1, 16.7, 14.7 and 29.1 per cent subjects respectively. For men, corresponding Kappa estimates of agreement were best for ECCS equation and worst for NHANES III equation (Table III). Among women, agreement was best with Knudson and worst with NHANES III equation. The Crapo and Knudson equation showed poorest agreement and worst misclassification in tall men and men at extremes of age; use of 90 per cent of predicted LLN improved agreement for the Crapo, but not for the Knudson, equations. The ECCS equation showed fairly good agreement, and the NHANES III equation showed poor agreement, across the entire range of age and height for men. All four Caucasian equations demonstrated poor agreement in women of all age and height categories, with high misclassification rates (Table III). Use of 90 per cent of predicted values in men improved overall agreement when using Crapo or NHANES III equation, but worsened it when using ECCS or Knudson equation (Table III). The Crapo and NHANES III equations consistently overpredicted LLN for both FVC and FEV1 in men and women in almost all age and height categories, AGGARWAL et al: CAUCASIAN EQUATIONS FOR SPIROMETRY IN INDIA 157 Table II. Comparison of spirometry interpretation by Caucasian and north Indian prediction equations North Indian 7 Normal Obstruction Restriction Normal Obstruction Restriction 4086 (60.0) 622 (9.1) 2106 (30.9) 132 (3.3) 3761 (93.6) 125 (3.1) 0 412 (10.6) 3489 (89.4) Knudson 9 Normal Obstruction Restriction 5853 (85.9) 278 (4.1) 683 (10.0) 448 (11.1) 3332 (82.9) 238 (5.9) 588 (15.1) 231 (5.9) 3082 (79.0) ECCS 10 Normal Obstruction Restriction 6115 (89.7) 0 699 (10.3) 814 (20.3) 2735 (68.1) 469 (11.7) 185 (4.7) 0 3716 (95.3) NHANES III11 Normal Obstruction Restriction Total 3230 (47.4) 25 (0.4) 3559 (52.2) 6814 274 (6.8) 3140 (78.1) 604 (15.0) 4018 7 (0.2) 20 (0.5) 3874 (99.3) 3901 Crapo 8 Data presented as number of patients and respective percentages (in parenthesis) ECCS, European Community for Coal and Steel; NHANES III, Third National Health and Nutrition Examination Survey with the latter giving slightly better results in the youngest subjects (Table IV). The differences between predicted LLNs for FVC and FEV 1 increased with height and decreased with age in both men and women. Although use of 90 per cent of predicted LLN lowered these differences in men, the limits of agreement still remained wide, and were more than 150 ml in most instances (Table IV). The Knudson equation consistently underpredicted LLN for FVC and FEV1 in short and old men, and overpredicted them in women of all age and height categories (Tables IV). The limits of agreement were wide in all instances and were only marginally improved by the use of 90 per cent of predicted LLN in men. The ECCS equation consistently overpredicted LLN for FVC and FEV 1 for short men and most women (Tables IV). The differences as well as limits of agreement were worsened by use of 90 per cent of predicted LLN in men. A few important observations emerged on expressing the LLN derived from north Indian equations as a percentage of the corresponding LLN derived from a Caucasian equation (Table V). Firstly, the resulting percentages were much lower for women as compared to men for all Caucasian equations in most instances. Secondly, there was considerable difference in the values obtained across various categories of age and height. Thirdly, the percentages were lower for FEV1 as compared to FVC. While the LLN for FVC obtained by north Indian equation was close to 90 per cent of LLN obtained by Caucasian equations in men of several age and height groups, the same was not true for FEV1. Fourthly, there was also considerable difference between results obtained from different equations. For example, the values obtained by Knudson and ECCS equations were higher than those obtained by north Indian equations in older and shorter men; this trend was not observed with the Crapo or the NHANES III equations (Table V). Overall, no consistent percentages were noted for any of the Caucasian equation used. Construction of ROC curves also demonstrated that the use of a fixed percentage of LLN from a Caucasian equation was a poor discriminator in identifying abnormal values of FVC or FEV1, as areas under the curves in nearly all age and height groups were close to 0.5. Areas under the ROC curve for the entire population for Crapo, Knudson, ECCS and NHANES III equations were 0.583, 0.589, 0.553 and 0.598 respectively for FVC, and 0.589, 0.593, 0.614 and 0.593 respectively for FEV1 (Fig.). 158 INDIAN J MED RES, AUGUST 2005 Table III. Kappa estimate of agreement, and percentage of misclassified results, for spirometry interpretation between north Indian and Caucasian prediction equations. Regression equation Crapo 8 Age (yr) Actual <20 >20 90%* <20 >20 Knudson 9 Actual <20 >20 90%* <20 >20 ECCS 10 Actual <20 >20 90%* <20 >20 NHANES III11 Actual <20 >20 90%* <20 >20 Kappa estimate of agreement (95% confidence limits) Percentage of misclassified results Men Women Men Women 0.623 (0.617-0.629) 0.698 (0.684-0.712) 0.850 (0.805-0.892) 0.793 (0.781-0.805) 0.565 (0.501- 0.629) 0.620 (0.604-0.636) 23.3 29.8 20.4 25.9 0.857 (0.813-0.901) 0.756 (0.744-0.768) 0.752 (0.698-0.806) 0.706 (0.692-0.720) 0.866 (0.820-0.912) 0.702 (0.686-0.718) 0.728 (0.670-0.786) 0.875 (0.865-0.885) 0.897 (0.859-0.935) 0.788 (0.776-0.800) 0.556 (0.488-0.624) 0.648 (0.632-0.664) 0.744 (0.688-0.800) 0.674 (0.660-0.688) 0.852 (0.808-0.894) 0.870 (0.860-0.880) 0.675 (0.611-0.739) 0.402 (0.384-0.418) 9.5 13.7 8.9 8.6 15.8 19.0 15.7 18.7 16.0 29.2 8.1 22.0 6.2 13.3 15.3 21.4 22.1 42.5 9.1 8.5 ECCS, European Community for Coal and Steel; NHANES III, Third National Health and Nutrition Examination Survey *Results from using 90 per cent of predicted values of forced vital capacity (FVC) and forced expiratory volume in first second (FEV 1) of respective Caucasian prediction equations Discussion Differences in pulmonary function among various racial and ethnic groups are well known, and differences in body proportions, chest wall anatomy, mechanical properties of the thorax, and parenchymal lung development have been postulated as contributory factors 1,11,16-26 . Most reference equations used to predict normal lung function have been derived in American or European subjects, and may not be suitable for use in other populations. It has been generally suggested that lung volumes in Indian subjects are approximately 10 per cent smaller than corresponding values in Caucasians4. The origin of this figure is not clear, but probably stems from a single small study on Pakistani workers resident in United Kingdom27. Though such ‘ethnic discounting’ has been widely recommended, there is no firm evidence that it provides accurate estimates28. This single figure also does not account for the heterogeneity among Indian population. It is well known that north Indians have higher lung volumes than south Indians 7. This study was conducted to evaluate how well commonly used Caucasian reference Regression equation Crapo 8 FVC (l) <20 0.40 (0.18 to 0.61) 0.48 (0.38 to 0.58) 0.60 (0.42 to 0.78) 0.64 (0.54 to 0.73) >20 0.26 (0.06 to 0.47) 0.36 (0.18 to 0.55) 0.46 (0.25 to 0.67) 0.50 (0.25 to 0.75) <20 0.01 (-0.09 to 0.12) 0.14 (0.11 to 0.16) >20 -0.06 (-0.15 to 0.04) 0.09 (0.00 to 0.19) <20 -0.03 (-0.43 to 0.37) -0.08 (-0.54 to 0.38) 0.03 (-0.27 to 0.34) 0.10(-0.22 to 0.42) >20 -0.20 (-0.86 to 0.47) -0.08 (-0.63 to 0.48) 0.20 (0.09 to 0.31) 0.30 (0.19 to 0.42) <20 -0.37 (-0.68 to -0.05) -0.37(-0.74 to 0.00) >20 -0.49(-0.99 to 0.03) -0.30 (-0.73 to 0.12) <20 0.28 (0.11 to 0.45) 0.32 (0.20 to 0.45) 0.42 (0.30 to 0.54) 0.35 (0.19 to 0.51) >20 0.00 (-0.29 to 0.29) 0.06 (-0.26 to 0.38) 0.16 (-0.11 to 0.42) 0.23 (-0.05 to 0.51) <20 -0.09 (-0.16 to -0.02) 0.00 (-0.05 to 0.04) >20 -0.30(-0.47 to -0.12) -0.18 (-0.40 to 0.04) Actual <20 0.22 (-0.36 to 0.79) 0.20 (-0.28 to 0.68) 0.36 (0.17 to 0.56) 0.38 (0.16 to 0.61) >20 <20 0.52 (0.23 to 0.80) -0.15 (-0.65 to 0.36) 0.34 (-0.02 to 0.69) -0.12 (-0.54 to 0.30) 0.65 (0.50 to 0.80) 0.55 (0.36 to 0.74) 90%* >20 0.17 (-0.02 to 0.36) 0.07 (-0.20 to 0.33) Actual 90%* ECCS 10 Actual 90%* NHANES III 11 Women FEV1 (l) 90%* Knudson Men FVC (l) Actual 9 Age (yr) ECCS European Community for Coal and Steel; NHANES III, Third National Health and Nutrition Examination Survey FVC and FEV1 values expressed in litres *Results from using 90 per cent of predicted values of FVC and FEV1 of respective Caucasian prediction equations FEV 1 (l) AGGARWAL et al: CAUCASIAN EQUATIONS FOR SPIROMETRY IN INDIA Table IV. Mean bias and limits of agreement for lower limit of normal for forced vital capacity (FVC) and forced expiratory vloume in first second (FEV 1) on comparison of Caucasian with north Indian prediction equations 159 160 INDIAN J MED RES, AUGUST 2005 Table V. Mean predicted values of forced vital capacity (FVC) and forced expiratory volume in first second (FEV1) obtained by north Indian equations, expressed as a percentage of those obtained by Caucasian equations Crapo 8 FVC FEV 1 Knudson 9 ECCS 10 NHANES III9 FVC FEV 1 FVC FEV 1 FVC FEV 1 76.9±8.0 Age (yr): 15-24 Men 89.8±1.6 76.5±3.0 96.3±9.2 82.1±9.9 92.8±1.5 79.0±2.8 90.3±7.8 Women 79.9±0.9 69.6±1.9 96.0±6.3 83.8±7.1 85.4±0.9 74.4±1.6 84.6±3.5 73.7±4.2 Men 90.5±1.3 74.5±1.8 98.9±9.1 81.5±8.3 94.9±1.6 78.1±1.9 83.8±0.9 69.0±0.8 Women 80.2±1.1 66.4±1.6 91.7±1.6 76.0±2.2 87.6±1.2 72.6±1.4 78.4±1.8 65.0±2.1 Men 91.5±1.4 73.3±1.7 103.9±10. 83.3±8.9 97.5±1.9 78.1±2.0 83.4±0.9 66.8±0.6 Women 80.5±1.2 63.9±1.3 91.3±1.7 72.5±1.8 90.4±1.7 71.7±1.4 74.4±1.9 59.1±1.8 Men 92.3±1.7 72.2±1.7 110.5±14. 86.5±11.3 100.5±2.6 78.6±2.3 83.8±1.0 65.6±0.5 Women 80.6±1.2 61.9±1.0 90.5±1.6 69.5±1.4 93.7±2.3 72.0±1.5 71.7±2.0 55.1±1.7 Men 92.9±1.8 71.3±1.6 118.9±50.9 91.3±39.7 103.5±3.2 79.5±2.6 85.4±1.1 65.6±0.5 Women 80.5±1.5 60.8±1.0 89.2±1.7 98.1±3.9 74.1±2.7 70.1±2.5 53.0±2.1 Men 93.3±2.2 70.8±1.8 127.7±26.4 96.9±20.2 106.9±4.6 81.1±3.6 88.3±1.5 67.0±0.9 Women 79.8±1.6 60.8±1.1 87.4±2.4 66.6±1.8 103.3±6.0 78.7±4.7 70.5±2.9 53.7±2.4 Men 93.4±2.6 70.6±2.0 152.1±102 115.0±78. 111.8±7.2 84.5±5.6 95.0±4.6 71.8±3.7 Women 78.0±1.8 62.4±2.3 90.8±3.9 72.7±4.7 111.5±12.3 89.5±12.8 75.0±4.6 60.1±5.5 Men 98.0±4.3 80.9±5.2 206.8±249 166.4±190 110.1±15. 90.4±8.3 87.7±11.1 72.9±13.4 Women 81.6±1.3 64.5±3.4 92.1±4.1 72.9±6.7 97.0±8.8 76.5±5.9 71.7±5.4 56.9±7.4 Men 94.5±1.7 74.7±2.2 134.7±27. 106.0±18. 105.0±6.7 82.9±2.7 85.3±5.2 67.5±5.3 Women 79.9±0.7 63.3±2.9 90.6±2.8 71.9±5.2 92.0±4.8 72.8±2.4 75.1±4.3 59.7±6.2 Men 92.0±1.2 72.5±2.0 110.5±11.5 86.9±7.0 100.6±4.8 79.2±1.6 86.0±3.9 67.8±3.9 Women 78.7±0.6 62.7±2.6 89.4±2.6 71.2±4.6 89.1±3.3 70.8±1.4 77.4±3.7 61.8±5.5 Men 90.2±0.9 71.2±1.8 97.8±6.0 77.1±3.4 97.2 ±3.6 76.7±0.9 86.0±3.2 67.9±3.4 Women 77.6±0.6 61.7±3.0 88.8±4.1 70.8±6.2 87.8±3.1 69.7±1.3 78.9±4.4 62.9±6.4 > 180 Men 88.8±0.7 70.0±1.7 90.5±3.8 71.3±2.6 95.1±3.1 74.9±0.7 85.9±2.9 67.7±2.9 Total Men 92.0±2.1 72.6±2.6 112.7±37. 88.7±28.0 100.6±5.9 79.3±3.2 85.9±4.2 67.8±4.4 Women 80.3±1.3 63.6±3.1 90.9±3.3 72.1±5.6 93.1±6.7 73.6±4.2 74.4±5.0 59.1±6.7 25-34 35-44 45-54 55-64 65-74 >75 67.4±1.2 Height (cm): <150 151-160 161-170 171-180 ECCS- European Community for Coal and Steel; NHANES III, Third National Health and Nutrition Examination Survey AGGARWAL et al: CAUCASIAN EQUATIONS FOR SPIROMETRY IN INDIA 161 (A) (B) Fig. Receiver operating characteristic curves to assess if use of a fixed percentage of Caucasian predicted value (References 8-11) can be used to correctly categorize forced vital capacity ( A) or forced expiratory volume in first second (B) as abnormal when compared to north Indian equations (Reference 7). All plots have area under curve close to 0.5, indicating that this cannot be achieved at any specified percentage. ECCS European Community for Coal and Steel; NHANES III, Third National Health and Nutrition Examination Survey. 162 INDIAN J MED RES, AUGUST 2005 equations perform, in comparison to already available north Indian prediction equations, in interpretation of routine spirometry data at our institute. Few investigators have tried to establish the utility of Caucasian prediction equations in other population subsets. Almost all of these studies have suffered from common methodological problems that preclude the extrapolation of their results into routine clinical practice. Most studies correlated values obtained from ‘external’ prediction equations with values obtained on a small number of native healthy subjects18,22,23,25. It is well known that there can be clinically important disagreement in results despite statistically good correlation14. More importantly, these results have not been verified on patients undergoing spirometry. In other instances, predicted values from two equations have been compared for a specified age and/or height in men and women 7,21, or mean values have been compared18,22. Such results hold true only for a small proportion of the large spectrum of subjects studied. Most studies also do not take into account the complex interplay between different variables in predicting lung volumes. analysis was that predicted north Indian FEV1 values were much lower than Caucasian values, in comparison to corresponding FVC values. Based solely on this observation, we would hesitate to recommend a uniform reduction in predicted values for both these lung volumes for approximation from Caucasian prediction equations. Even otherwise, there is wide variability in individual differences in both FVC and FEV 1 across different age and height categories, and again, a single fixed percentage value is unlikely to be useful for the entire spectrum of patients. In fact, the use of a fixed reduction factor of any magnitude will be a poor discriminator in labeling either FVC or FEV1 as abnormal, since areas under corresponding ROC curves are quite close to the 0.5 value obtained purely by chance. Data on the utility of Caucasian equations in interpreting spirometry results in patients of other populations are sparse. In a small study on 442 Spanish patients at a New Mexico hospital, local Hispanic prediction equations were compared to three Caucasian equations in categorizing FVC and FEV1 as normal or abnormal29. In this report, 2-10 per cent results were misclassified, depending on the reference equation used. Subjects with discordant classifications tended to be at extremes of age and height distributions. On average, the differences between the predicted values from the three sets of external equations and local Hispanic equation were small, but the range of differences was wide. We also attempted to quantify agreement between predicted values of different spirometric parameters using north Indian and different Caucasian equations. This approach has definite advantages over merely reporting correlation or regression coefficients, as it provides a numerical estimate of how similar are values obtained from two distributions, and whether results from the two approaches can be used interchangeably14. We observed wide variations in the mean bias for all spirometric parameters studied. We also calculated the limits of agreement, which represent a numerical expression of range in which 95 per cent of the bias values are likely to be situated. We documented wide limits of agreement in most age and height categories, and more so in women. The range was more than 150 ml of FVC and FEV1 in several categories of age and height, and thus clearly beyond the normal variability accepted in the measurement of these lung volumes. These findings suggested a rather poor agreement between Caucasian and north Indian predicted values. Our results highlighted some important facts about use of Caucasian prediction equations in other populations. It is quite evident that these equations led to misinterpretation of spirometry data in a significant proportion of patients and this might result in inappropriate diagnosis and/or management. The use of 90 per cent of predicted Caucasian values for FVC and FEV1 in men also did not totally ameliorate the problem, and actually worsened it in some instances. An important aspect brought out by our The main strength of our study is the large number of spirometry records included for analysis that ensured adequate representation from all categories of age and height in both men and women allowing confident identification of even small differences without any bias. In fact, our estimates appear much more precise and associated with much smaller error distributions than previous studies23,29. The patients studied herein represent a population that undergoes routine pulmonary function testing at our centre. The AGGARWAL et al: CAUCASIAN EQUATIONS FOR SPIROMETRY IN INDIA relatively higher proportion of normal spirometric records in this study is probably a reflection of the fact that a large number of patients from other departments are routinely investigated at our laboratory, and spirometry is a screening procedure for several of them (for example, as part of routine preoperative workup). Though the results from this study can only be strictly applied to a similar population tested with similar instruments and procedures, the conclusions can perhaps be generalized to other situations as well with minor differences. Our analysis also provides a scheme that could be used to ascertain how well (or poorly) existing regression equations predict lung function in a different set of patient population. The absence of concurrent controls may be regarded by same as a limitation of our study. We preferred to use regression equations previously desired by us from asymptomatic non smokers selected from the general population, rather than asymptomatic patients. Moreover, it is not a standard procedure to derive regression equations for population use from patients attending hospitals. It is also well known that different Caucasian prediction equations yield different results and classify subjects differently when applied to a study population 30-32. The fact reinforces the utility of a strategy of assessing performance of more than one regression equation in one’s own clinical practice before choosing any single prediction equation. References 1. 2. 3. (No authors listed). Lung function testing: selection of reference values and interpretative strategies. American Thoracic Society. Am Rev Respir Dis 1991; 144 : 1202-18. Boro AK, Sharma SK, Pande JN. Comparison of single breath and steady state methods for the measurement of pulmonary diffusing capacity for carbon monoxide in normal subjects, patients with bronchial asthma and chronic obstructive airway disease. Indian J Chest Dis Allied Sci 1992; 34 : 1-5. Chowgule RV, Shetye VM, Parmar JR, Bhosale AM, Khandagale MR, Phalnitkar SV, et al. Prevalence of respiratory symptoms, bronbchial hyperreactivity, and asthma in a mega city. Results of the European Community respiratory health survey in Mumbai (Bombay). Am J Respir Crit Care Med 1998; 158 : 547-54. 163 4. Lung function throughout life: determinants and reference values. In: Cotes JE, editor. Lung function. Assessment and application in medicine. 5th ed. Oxford: Blackwell Scientific Publications; 1993 p. 445-513. 5. Aggarwal AN, Gupta D, Jindal SK. Development of a simple computer program for spirometry interpretation. J Assoc Physicians India 2002; 50 : 567-70. 6. (No authors listed). Standardization of spirometry, 1994 update. American Thoracic Society. Am J Respir Crit Care Med 1995; 152 : 1107-36. 7. Jindal SK, Wahi PL. Pulmonary function laboratory in the tropics: needs, problems and solutions. In: Sharma OP, editor. Lung disease in the tropics. New York: Marcel Dekker; 1991 p. 523-42. 8. Crapo RO, Morris AH, Gardner RM. Reference spirometric values using techniques and equipment that meet ATS recommendations. Am Rev Respir Dis 1981; 123 : 659-64. 9. Knudson RJ, Lebowitz MD, Holberg CJ, Burrows B. Changes in the normal maximal expiratory flow-volume curve with growth and aging. Am Rev Respir Dis 1983; 127 : 725-34. 10. European Community for Coal and Steel. Standardization of lung function tests. Bull Eur Physiopathol Respir 1983; 19 (Suppl): 1-93. 11. Hankinson JL, Odencrantz JR, Fedan KB. Spirometric reference values from a sample of the general U.S. population. Am J Respir Crit Care Med 1999; 159 : 179-87. 12. Fleiss JL. The measurement and control of misclassification error. In: Fleiss JL, editor. Statistical methods for rates and proportions. 2nd ed. New York: John Wiley; 1973 p. 140-54; 1981 p. 213. 13. Efron B. Better bootstrap confidence intervals. J Am Stat Assoc 1987; 82 : 171-85. 14. Bland JM, Altman DG. Statistical methods for assessing agreement between two methods of clinical measurement. Lancet 1986; 1 : 307-10. 15. Hanley JA, McNeil BJ. The meaning and use of the area under a receiver operating characteristic (ROC) curve. Radiology 1982; 143 : 29-36. 16. Oscherwitz M, Edlavitch SA, Baker TR, Jarboe T. Differences in pulmonary functions in various racial groups. Am J Epidemiol 1972; 96 : 319-27. 17. Schoenberg JB, Beck GJ, Bouhuys A. Growth and decay of pulmonary function in healthy blacks and whites. Respir Physiol 1978; 33 : 367-93. 164 INDIAN J MED RES, AUGUST 2005 18. B i b i H , G o l d s m i t h J R , Va r d i H . R a c i a l o r e t h n i c variation in spirometric lung function norms. Recommendations based on study of Ethiopian Jews. Chest 1988; 93 : 1026-30. 25. Milivojevic-Poleksic L, Wells AU, Moody A, Fergusson W, Tukuitonga C, Kolbe J. Spirometric lung volumes in the adult Pacific Islander population: comparison with predicted values in a European population. Respirology 2001; 6 : 247-53. 19. Schwartz JD, Katz SA, Fegley RW, Tockman MS. Sex and race differences in the development of lung function. Am Rev Respir Dis 1988; 138 : 1415-21. 26. Donnelly PM, Yang TS, Peat JK, Woolcock AJ. What factors explain racial differences in lung volumes? Eur Respir J 1991; 4 : 829-38. 20. Mathur N, Rastogi SK, Gupta BN, Husain T. A global comparison of predicting equations on spirometry in the male population. Int J Epidemiol 1990; 19 : 331-8. 27. Malik MA, Moss E, Lee WR. Prediction values for the ventilatory capacity in male West Pakistani workers in the United Kingdom. Thorax 1972; 27 : 611-9. 21. Yang TS, Peat J, Keena V, Donnelly P, Unger W, Woolcock A. A review of the racial differences in the lung function of normal Caucasian, Chinese and Indian subjects. Eur Respir J 1991; 4 : 872-80. 28. White NW. ‘Ethnic discounting’ and spirometry. Respir Med 1995; 89 : 312-3. 22. Crapo RO, Jensen RL, Oyunchimeg M, Tsh T, DuWayne Schmidt C. Differences in spirometry reference values: a statistical comparison of a Mongolian and a Caucasian study. Eur Respir J 1999; 13 : 606-9. 23. Ip MS, Karlberg EM, Karlberg JP, Luk KD, Leong JC. Lung function reference values in Chinese children and adolescents in Hong Kong. I. Spirometric values and comparison with other populations. Am J Respir Crit Care Med 2000; 162 : 424-9. 24. Korotzer B, Ong S, Hansen JE. Ethnic differences in pulmonary function in healthy nonsmoking AsianAmericans and European-Americans. Am J Respir Crit Care Med 2000; 161 : 1101-8. 29. Shaffer BA, Samet JM, Coultas DB, Stidley CA. Prediction of lung function in Hispanics using local ethnic-specific and external non-ethnic-specific prediction equations. Am Rev Respir Dis 1993; 147 : 1349-53. 30. Harber P, Schnur R, Emery J, Brooks S, Ploy-Song-Sang Y. Statistical “biases” in respiratory disability determinations. Am Rev Respir Dis 1983; 128 : 413-8. 31. Lebowitz MD, Holberg CJ. Comparisons of spirometric reference values and the proportion of abnormal subjects among male smokers and those symptomatic in a community population. Am Rev Respir Dis 1990; 141 : 1491-6. 32. Quadrelli S, Roncoroni A, Montiel G. Assessment of respiratory function: influence of spirometry reference values and normality criteria selection. Respir Med 1999; 93 : 523-35. Reprint requests: Dr S.K. Jindal, Professor & Head, Department of Pulmonary Medicine Postgraduate Institute of Medical Education & Research, Chandigarh 160012, India e-mail: skjindal@indiachest.org