The purpose of this report is to compare the results of

Comparison of Two Measures of Atherosclerosis in a Prospective Epidemiology Study Dwayne M. Reed, Jack P. Strong, Takuji Hayashi, W. P. ewman III, Richard E. Tracy, Miguel A. Guzman, and Grant. Stemmermann Two different methods of measuring atherosclerosis, the American Heart Association panel method and the International Atherosclerosis Project unaided visual estimation, were used to estimate the extent of atherosclerosis in the coronary arteries and aortas from men who participated in a prospective study of cardiovascular disease. The two methods produced quite different results of frequency distribution of levels of atherosclerosis. The correlations of the two measures of atherosclerosis were also relatively low for both the coronary arteries (0.63) and the aortas (0.7). The correlation of atherosclerosis with major risk factors was generally higher with the unaided visual estimation method than with the panel method, except for cigarette smoking. However, the associations of coronary atherosclerosis with autopsy-documented myocardial infarction and with clinical coronary disease were similar for the two measures. Thus, it is clear that the estimate of extent of atherosclerosis and its associations with major risk factors depends upon the method used to measure atherosclerosis. Epidemiologic studies that use only one method may miss some types of risk factor associations. (Arteriosclerosis 8:78-787, ovember/december 1988) The purpose of this report is to compare the results of using two different methods of measuring atherosclerosis in men who participated in a prospective study of cardiovascular disease. The methods were the American Heart Association (AHA) panel method 1 and the unaided visual estimation developed for the International Atherosclerosis Project (IAP). The comparisons included the frequency distributions of levels of atherosclerosis in coronary arteries and aortas, the associations of atherosclerosis with potential risk factors measured at entry into the study, and the associations of atherosclerosis with autopsy-verified myocardial infarction and clinical coronary heart disease (CHD). When this study was begun, there was little attempt to use standardized methods of measuring atherosclerosis in prospective epidemiologic investigations. Some studies used the two methods noted above, 3-6 while other studies used estimates of luminal narrowing 7 or various combinations of methods. 89 Thus, during the early stages of the i-hon-san project, a comparative study of cardiovascular disease among Japanese men in Japan, Hawaii, and California, pathologists from Hawaii and Japan worked together to evaluate four methods for estimation of the severity of atheroscle- Dwayne M. Reed is at the Honolulu Heart Program, 37 orth Kuakini Street, Honolulu, HI. Jack P. Strong, W.P. ewman III, Richard E. Tracy, and Miguel A. Guzman are at Louisiana State University Medical Center, ew Orleans, LA. Takuji Hayashi and Grant. Stemmermann are at the Pathology Department, Kuakini Medical Center, Honolulu, HI. This work was supported by HLBI Contract o. 01 HC 0901 and Grant o. HL 0879, and the Japan-U.S. Cooperative Agreement in the cardiovascular area. Address for correspondence: Dwayne M. Reed, M.D., Honolulu Heart Program, 37 orth Kuakini Street, Honolulu, HI 96817. Received ovember,1987; revision accepted May 19,1988. rosis. 10 Point counting, grid counting, the AHA panel method, and a modification of a gross visual method of Gore were compared. The gross visual method of Gore was found to have the greatest variation between observers, and the grid technique was quickly abandoned because it was too tedious and fatiguing. The pointcounting method had a high level of inter- and intraobserver reproducibility; however, the technique was judged unsatisfactory due to the difficulty of estimating the surface area involved when the lumen of the coronary vessels were stenotic or occluded. The panel method also had a high level of intra- and inter-observer reproducibility and was judged to be objective and easy to use. Thus, with the requirement that the method be reproducible and sensitive to differences in populations, the panel method was selected and has been the basis for geographic comparisons of atherosclerosis both in Japan and in Hawaii and for the study of antemortem characteristics associated with atherosclerosis. 611 Similarly, the unaided visual estimation method developed for the IAP has been shown to be a practical and reliable approach for population comparisons of atherosclerosis 1 and has been a valuable instrument in studies of atherosclerosis by investigators at the Louisiana State University Medical Center over the past years. 13 " 19 In 1986, over accumulated specimens of coronary arteries and aortas from the Honolulu Heart Program were sent to the Louisiana State University Medical Center Department of Pathology for estimation of the percent of intimal surface involved with atherosclerotic lesions by the IAP unaided visual method. It thus became possible to compare the results of using these two different methods in the same population. 78

MEASURES OF ATHEROSCLEROSIS Reed et al. 783 Methods The Honolulu Heart Program is a prospective study of cardiovascular disease among 8006 men of Japanese ancestry, who were ages to 68 years at the time of the initial examination from 196 through 1968. Details of the examination method have been previously reported. 01 Since 196, rigorous efforts have been made to ascertain deaths. Daily telephone calls were made to island mortuaries, hospitals, and medical examiner's offices to discover impending autopsies among cohort men so that arrangements could be made to acquire the heart, aorta, and cerebral vessels. The incidence of coronary heart disease, stroke, and all causes of death was determined by a panel of study physicians on the basis of all available information obtained from hospital records, death certificates, autopsy reports, and personal physicians. The methods used for pathologic evaluation of the vessels and heart have been described earlier. ' 6 ' 1011 Following a standardized protocol, the coronary arteries and aorta were dissected free of fat and adventitial tissue, opened longitudinally, sewn to plastic sheets, and fixed in formalin. For estimating the degree of atherosclerosis by the AHA panel method, each vessel was compared to a panel of photographs showing increasing levels of atherosclerosis and was assigned a score from 1 (completely free of raised lesions) to 7 (severe atherosclerosis). When occlusive disease was severe enough to prevent longitudinal opening of the artery, the vessel was arbitrarily coded as 7. The readings were blind to any other autopsy or clinical findings. Reliability studies have shown a correlation of about 0.9 between duplicate annual readings. 6 Gross examinations of the hearts and microscopic examinations of seven sites of the heart were made. Scars or areas of acute necrosis of 0. cm or greater were classified as myocardial infarction lesions. Both old and new lesions were combined for analysis of association of coronary atherosclerosis. In 1986, all available coronary artery and aorta specimens, which had been stored in formalin, were shipped to the pathology laboratory in Louisiana for staining with Sudan IV and for visual grading of atherosclerosis by the method used for the IAP. The percentages of artery intimal surface area that were involved with fatty streaks, fibrous plaques, complicated lesions, and calcified lesions were estimated separately; however, for this comparison, fibrous plaques and complicated and calcified lesions were grouped together to form a category of percent of surface affected by raised lesions (% RL). For this comparison, the average scores for the three coronary arteries were used. Separate analyses of the individual coronary arteries resulted in patterns that were nearly identical to those presented for the mean scores. The thoracic and abdominal segments of the aorta were graded separately by the IAP visual examination method, and the scores were also averaged for comparison with the total aorta graded by the panel method. Statistical Analyses The statistical tests of association of risk factors with atherosclerosis scores were based on univariate and multivariate linear regression. effects were removed from the scores by adjustment for age at death and from the risk factors by adjustment for age at baseline examination when the risk factors were measured. Multiple regression coefficients relating scores to risk factors were standardized by multiplying the beta coefficients times the ratio of the standard deviation of the independent variable to the standard deviation of the dependent variable. The result was nearly identical to a partial correlation coefficient. To control for the potential effects of autopsy selection bias on the relationship of risk factors and atherosclerosis, we also performed analyses on the autopsied men who showed no evidence of cardiovascular disease or diabetes at the time of death. This is similar to the practice of using the "basal" group in the IAP Study. Results Among the specimens available for estimation of the extent of atherosclerosis, there were 09 sets of coronary arteries and 16 aortas from men who were free of any clinical evidence of coronary heart disease or stroke at the time of the baseline examination. The "basal" or non-cardiovascular disease (CVD) subgroup consisted of specimens of 116 coronary arteries and 78 aortas from men who had no clinical or autopsy evidence of any cardiovascular disease or diabetes. Table 1 provides details of the association of the two measures by showing the means and standard deviation of the % RL for the arteries classified by panel scores. For the coronary arteries, there was a general pattern of a doubling of % RL with each increase in panel scores from 1 to 3, followed by an approximate increase of 10% RL for each step of the rest of the scores. The increase of % RL with the panel scores for the aorta were more irregular. The correlation coefficient of 0.63 and the large standard deviations around the mean % RL for each coronary panel score indicate a relatively low level of correlation of the two methods. For the aorta, the correlation of 0.7 indicated a slightly stronger association of the two measures. Separate analyses of the frequency distribution of men by panel scores and % RL showed a skewed normal curve with a peak of for the panel scores, while the % RL had a flat distribution. An indication of their distributions can be seen in the number of columns of Table. Table shows mean panel scores and % RL by age at death and artery. For the coronary arteries, the general pattern of increasing levels of atherosclerosis at the younger ages, a middle plateau, and another increase at the older ages was seen with both measures. The correlation with age, however, was generally stronger for % RL than the panel score (0.7 vs. 0.11). The fact that some age groups had identical mean panel scores and strikingly different % RL (for example, age groups to 9 and 70 to 7) further indicates the lack of agreement between the two measures. Similar patterns were found for the aorta. In terms of severity, the age-specific % RL are similar to those reported for white men in ew Orleans. 16 Table 3 shows the standardized multiple regression coefficients (r) relating the two measures of atherosclerosis in the coronary arteries to selected risk factors mea-

78 ARTERIOSCLEROSIS VOL 8, o 6, OVEMBER/DECEMBER 1988 Table 1. umber of Men and Percent of Artery Surface Affected by Raised Lesions Coronary arteries (mean) Aorta Panel score 1 63 17±6 31 ±0 0 9±3 3 8 7±1 3±17 67±1 33 6±1 13 7±1 3 71 ±13 6 1 89±8 1 86±10 7 Total 09 93±3 1 ±1 16 98± 61±17 Correlation coefficient 0.63 0.7 =number of men, % RL=percent of surface affected by raised lesions. % RL values include±1 SD. Table. Panel Scores and Percent Raised Lesions at Death at death (yrs) 0 to to 9 60 to 6 6 to 69 70 to 7 7 to 8 37 37 8 38 1 Coronary arteries (mean) Panel score.7 3.1 3.1.9 3.1.0 33 7 9 9 7 1 9 6 10 =number of men, % RL=percent of surface affected by raised lesions. Aorta Panel score 3. 3.8.1.0.7. 9 61 71 8 Table 3. Standardized Multiple Regression Coefficients Relating Panel Scores and Percent Raised Lesions of Coronary Arteries to Selected Risk Factors Variable Panel score Total autopsy group (n=09) Body Mass Index on-cardiovascular disease group (n=116) Body Mass Index % RL=percent of surface affected by raised lesions. *p<0.0, tp<0.01, *p<0.001. 0.1* 0.13 0.3t 0.08 0.10 0.17 0. 0.6t -0.01 0.08 0.30$ 0.1* 0.33* 0.06 0.03 0.1 0.33$ 0.t 0.1 0.0 0.1 sured at baseline examination. For the total autopsy group shown at the top of Table 3, the % RL generally had higher levels of association than the panel scores, except for cigarette pack years and serum glucose levels. For the non-cvd group, the % RL had higher associations with age, serum cholesterol, and body mass index, while the panel scores had higher associations for blood pressure and cigarette pack years. In terms of making hypothesis testing judgments at a p value of 0.0 for this latter group, age and serum cholesterol would be judged "significant" using the % RL measure and "not significant" with the panel method, while cigarette smoking would be judged not significant by the % RL measure, and significant with the panel method. Similar comparisons are shown for the aorta in Table. For the total autopsy group, the r values for the panel

MEASURES OF ATHEROSCLEROSIS Reed et al. 78 Table. Standardized Multiple Regression Coefficients Relating Panel Scores and Percent Raised Lesions of Aortas to Selected Risk Factors Variable Panel score Total autopsy (n=19) 0.t 0.0t 0.t 0.f 0.08 on-cardiovascular disease group (n=78) 0.1 0.3* 0.19 0.3* 0.07 % RL=percent of surface affected by raised lesions. *p<0.0, tp<0.01, *p<0.001. Thoracic aorta 0.3* 0.6* 0.3f 0.33* 0.1* 0.6* 0.31 f 0.13 0.9f 0.1 Abdominal aorta 0.t 0.16* 0.7* 0.7* 0.13 0.1 0.1 0.19 0.3* Whole aorta 0.7* 0.11 0.f 0.30* 0.1 0.0* 0.3* 0.16 0.311 0.11 Table. -adjusted Percentages of Men with CHD and Pathology-determined Myocardial Infarction Mean coronary panel score % with CHD 1 63 3 8 1 19 13 30 en- Total 09 1 T value 6.0* Mean % RL Oto 1 1 to 9 30 to 33 to 9 3 11 60 to 7 3 7 to 89 3 >90 0 Total 09 1 T value 6.0* % RL=percent of surface affected by raised lesions. CHD=coronary heart disease. *p<0.001 from regression models with age. % with myocardial infarction 6 3 3 0 69 3 6.* 7 11 7 6 3.3* score were similar to those for the average % RL for the two parts of the aorta, except for serum glucose. For the non-cvd group, the values were similar except for age. Table shows the age-adjusted percentages of autopsied men with clinical evidence of CHD and with autopsyverified myocardial infarction by levels of atherosclerosis. For both measures, there was a linear increase in clinical CHD and myocardial infarction with increasing levels of atherosclerosis, and the levels of statistical association were similar. Discussion Ideally, epidemiologic studies of atherosclerosis in different populations would use a standardized measure of atherosclerosis that is accurate, reliable, and practical to use. Both of the measures examined here have been found to fare quite well from this perspective, 6101-19 but it is clear from the present study that these measures are different. The correlation coefficient of 0.63 between the panel scores and % RL for coronary arteries was relatively lower than the correlation of 0.89 reported

786 ARTERIOSCLEROSIS VOL 8, o 6, OVEMBER/DECEMBER 1988 earlier between panel scores and the point-counting estimate. 10 The large standard deviation of over 0% RL for many of the panel scores further indicates a level of difference between the two measures. We know of at least two possible explanations for this difference. The first relates to the possible uneven intervals between the ranks of the panel scores. Although the photographs of vessels used for the panel method were selected for increasing severity by a gross visual method, there was no claim that the intervals between ranks were equal, 1 and one report documented that they were not equal. 10 Our findings also indicate a lack of equal intervals; for example, there was a difference of 6% RL between the coronary panel scores of and 3, compared to a difference of % RL between panel scores of 6 and 7 (Table 1). The other discrepancy is due to the fact that any vessel with a severe occlusion was arbitrarily coded as Grade 7 on the panel scores but not given any special treatment with the visual estimation. When there are differences between two measures, it seems natural to try to determine which one is "better" in terms of an accurate indicator of the degree of atherosclerosis. Unfortunately, this is not possible without an independent "gold standard" for reference. Comparisons of the associations of the measures of coronary atherosclerosis with autopsy-determined myocardial infarction offers a practical indication of the accuracy of the different measures. As noted in Table, there were no meaningful differences in the patterns of progressive increases in the proportion of men with myocardial infarction for increases in either the panel scores and % RL, and the levels of statistical associations were quite similar. Thus, from this perspective neither method is better. The associations of the measures of atherosclerosis with antecedent risk factors in multiple regression models offer another opportunity for comparison; however, it should be emphasized that strength of association indicates which relationships are more nearly linear, but not necessarily which measure is better. For the total autopsy group, the mean % RL in the coronary arteries generally had higher correlations with risk factors than did the panel scores. Among the non-cvd group, the associations were higher with % RL for age and serum cholesterol, and lower for cigarette smoking. These differences could easily be due to the differences in the methods noted above, and our general inference is that if a risk factor is associated with either of the measures, it is worth further consideration. Thus, blood pressure, serum cholesterol, and cigarette smoking appeared to be independent antecedent predictors for atherosclerosis in both the coronary arteries and the aortas. In univariate, but not multivariate, analyses, serum glucose and body mass index were also significant. In conclusion, it is clear from these analyses that estimates of the extent of atherosclerosis and the degree of association of atherosclerosis with risk factors depend upon the method used for estimation of atherosclerosis. Thus, epidemiologic studies that use only a single estimate may miss some types of risk factor association. Several of the current studies seemed to have anticipated this finding, since they include a combination of the IAP unaided visual estimation with other more sensitive measures. A current cooperative effort of 1 laboratories in the United States for the investigation of the "Pathobiological Determinants of Atherosclerosis in Youth," for example, utilizes electronic image analysis on photographs and x-rays of arterial specimens to complement unaided visual estimates of the extent of lesions. The AHA panel method has not become popular, perhaps because it does not provide separate estimates of the severity of different types of lesions. Because of this lack of use in other studies, and because of the generally lower levels of linear association with risk factors, there is little reason to recommend the use of the AHA panel method for future epidemiologic studies. Finally, we would like to note the value of careful collection, preparation, storage, and maintenance of specimens, which is demonstrated by the opportunities provided to re-evaluate specimens in this study by an additional systematic method. References 1. McGill HC Jr, Brown BW, Gore I, et al. Grading human atherosclerotic lesions using a panel of photographs. Circulation 1968;37:-9. Guzman MA, McMahan CA, McGIII HC Jr, et al. Selected methodologic aspects of the International Atherosclerosis Project. Lab Invest 1968::79-97 3. Holme I, Enger SC, Helgeland A, et al. Risk factors and raised atherosclerotic lesions in coronary and cerebral arteries. Statistical analysis from the Oslo Study. Arteriosclerosis 1981:1:0-6. Sorlie PD, Garcia-Palmleri MR, Castlllo-Staab Ml, et al. The relation of antemortem factors to atherosclerosis in autopsy. The Puerto Rico Heart Health Program. Am J Pathol 1981:103:-. Rhoads GG, Blackwelder WC, Stemmermann G, et al. Coronary risk factors and autopsy findings in Japanese- American men. Lab Invest 1978:38:30-311 6. Stemmermann G, Steer A, Rhoads GG, et al. A comparative pathology study of myocardial lesions and atherosclerosis in Japanese men living in Hiroshima, Japan and Honolulu, Hawaii. Lab Invest 1976:3:9-600 7. Okumlya, Tanaka K, Ueda K, et al. Coronary atherosclerosis and antecedent risk factors: pathologic and epidemiology study in Hisayama, Japan. Am J Cardiol 198:6:6-66 8. Feinleib M, Kannel WB, Tedeschl CG, et al. The relation of antemortem characteristics to cardiovascular findings at necropsy. The Framingham Study. Atherosclerosis 1979; 3:-7 9. Rickert RR, Johnson KG, Kato H, Yamamoto T, Yano K. Atherosclerosis in a defined Japanese population: a clinicopathologic appraisal. Am J Clin Pathol 1968;9:371 10. Gould SE, Hayashi T, Tashiro T, Tanlmura H, akashlma T, Shohojl T. Coronary heart disease and stroke: atherosclerosis in Japanese men living in Hiroshima, Japan and Honolulu, Hawaii. Arch Pathol 197;93:98-10 11. Reed DM, MacLean CJ, Hayashi T. Predictors of atherosclerosis in the Honolulu Heart Program. 1. Biologic, dietary and lifestyle characteristics. Am J Epidemiol 1987; 16:1-1. Guzman MA, McMahan CA, Strong JP. Unaided visual estimation of atherosclerotic lesions. Biological variability compared with grading variability. Lab Invest 197; 31:398-0 13. Strong JP, McGill HC Jr. The natural history of coronary atherosclerosis. Am J Pathol 196;0:37-9 1. Strong JP, Richards ML. Cigarette smoking and atherosclerosis in autopsied men. Atherosclerosis 1976;3:1-76

MEASURES OF ATHEROSCLEROSIS Reed et al. 787 1. Moore MC, Guzman MA, Schilling PE, Strong JP. Dietary atherosclerosis study on deceased persons: relation of selected dietary components to raised coronary lesions. J Am Diet Assn 1976;68:16-3 16. Strong JP, Guzman MA. Decrease in coronary atherosclerosis in ew Orleans. Lab Invest 1980;3:97-301 17. Strong JP, Oalmann MC, ewman WP III, et al. Coronary heart disease in young black and white males in ew Orleans: community pathology study. Am Heart J 198; 108:77-79. Ishii T, ewman WP III, Guzman MA, Hosoda Y, Strong JP. Coronary and aortic atherosclerosis in young men from Tokyo and ew Orleans. Lab Invest 1986;:61-6 19. Solberg LA, Ishii T, Strong JP, et al. Comparison of atherosclerosis in middle-aged orwegian and Japanese men. Lab Invest 1987;6:1-6 0. Kagan A, Harris BR, Winkelstein W Jr, et at. Epidemiologic studies of coronary heart disease and stroke in Japanese men living in Japan, Hawaii and California: demographic, physical, dietary and biochemical characteristics. J Chronic Dis 197;7:-36 1. Yano K, Reed DM, McGee D. Ten-year incidence of coronary heart disease in the Honolulu Heart Program: relationship to biologic and lifestyle characteristics. Am J Epidemiol 198;119:63-666. Solberg LA, Strong JP. Risk factors and atherosclerotic lesions: a review of autopsy studies. Atherosclerosis 1983; 3:7-198 Index Terms: autopsied cases coronary atherosclerosis blood pressure serum cholesterol cohort