Tobacco Documents Online

SOtE STATISTICAL OBSERVATIONS ON A C00FERATIVE SNDY OF EU14AN PUIMNARY PATHOLOGY, II By Edwin B. Wilson and Mary E. Burke In our first paper .(1) we gave some genera.l average data for the readings of eight pathologists in eight different cities on slides ma.de from sections taken irp standard positions in run-of-the-mill lungs at autopsy using the classifications: normal, typerplaaia, metaplasia, atypical metaplasia, cercinoma-in-situ, and carcinoma. As carcinoem-in-situ vas found so rarely by any of the pathologists, that classification vill be combined vith atypical metaplasia in this continuation of the study; there vill be only 5 groups and their rank indices (2) vill be 0, 1, 2, 3, 4. When we became convinced that the'classification vas being made on different bases by the different pathologists,'ve asked 0.11 twelve to read a selected sample of 40 slides. This they kindly did, and ve reported on the considerable statistical differences in the readings. As the main object in all the work bad been to.obtain comparable data in the twelve cities for the degree of pathology in the lungs examined, ve stated that it vould be vell to have a conaiderable sample of the slides from all cities read by several pathologists. The need for this is clear from the differences shown in Thble 1 for the percentages of their slides placed in the 5 groups by the pathologists in eight of the'twelve cities (3). Table 1 Percenta€e Distributions for Ma.les Age 25 and Up P,ea?er Slides 0 1 2 k Index J: 909 28.8 53.6 11.7 4.2 1.8 .97 D: 91+1 57.1 21.1 7.7 11.1 3.0 .82 A: 408 38.7 46.1 15.0 0.0 0.2 .77 E: 630 66.7 9.7 18.7 3.6 1.3 .63 E: 223 65.9 9.4 21.1 2.6 0.9 .63 L: 2495 76.4 6.9 11.9 3.3 1.5 .47 1: 669 74.4 8.4 16.3 0.9 0.0 L4 H: 1418 81.8 9.7 8.0 0.4 0.1 .27 Mean 61.2 20.6 13.8 3•3 1.1 .62 We were fortunate enough to find t2iree of the pathologists v-bo were willing to read a sample of 609 slides drawn from the different cities by randoT processes (k). We included the 40 slides read by all. The present paper is to report on the results of this rereading, and the two sets of slides will be treated separately. The gross results a.re in Tables 2 and 3. / yT~ 1416 . LUt:i iJ~a::1f,L: MlNNES0TA TOBACCO LITIGATION t C T RZ H H D 0 ~~~~~

- 2 - Table 2 Distribution of Total of 609 Slides on Rereading Reader Slides 0 1 2 L k Index A: 609 359 93 127 14 16 .744 E: 609 348 25 212 6 18 .885 L: 6o9 359 88 133 7 22 .760 Total 1827 1066 206 472 27 56 .796 Reader A is high in atypicals (3) and Reader E is low in hyperplasia (1) and high in metaplasia (2) compared with the other two. Tab1e 3 Distribution of 40 Slides on Rereading Reader Slides 0 1 2 2 4 Index A: 40 4 4 27 3 2 1.875 E: 40 5 2 28 2 3 1.900 L: 4o 5 6 25 1 3 1.775 Total 120 lI+ 12 80 6 8 1.850 In this small aasrple, distributed very differently froc the large one, the differences noticeable in the latter are not in evidence; but the distribution is significantly di2Yexent h-om that previously Sou.^d by all twelve patholobists, viz., 48, 120, 223, 57, 32; it is not significantly different from what was fo•.:...^d by these three pathologists, viz., 16, 20, 64, 10, 10. The rereadings of the 40 alides by the three readers and their original readings have these properties: Reader Mean II Mean I Mean II - Mean I Correlation A: 1.875 1.800 - .075 .1.14 0.65 E: 1.900 2.125 - .225 .103 0.76 L: 1.775 1.525 -..250 .091 0.86 The first reader has not changed his mean significantly, the second has decreased his and the third increaaed his, each significantly. The means have thus come closer together. The aelf -correlation coefficients ves-y iro= o.65 to 0.86. , I CTR' 1417 CONFIDENTIAL: MINNESOTA - TOBACCO LITIGATION , ".,, T,,,s ,, I, , 0~.~ 3~, ~-319

-3- In the random sample, the numbers of slides be7~ing to A, E, L respectively were 73, 72 and 60. The comparison of the ~er~tabings by each of his own slides is as follows: A: 0.548 0.644 - .096 E: 0.792 0.76k -~ .028 L: 0.950 1.333 - .383 It is seen that the three pathologists are readitig their t as they did before and that the selS-correlstion coeSficiea -out the same magnitude as for the 40 slides. With this background ve may turn to the s -the percentages over classes which results from using the three pathologists as a basis. The method is similar to th" death rates for age and sex against the age and sex distri population. In Table 1, J put 28.8p of his slides in the t~o~s drawn for J from his 909 slides and presented to the three other slides contained 32 normA1s, 59 hyperplasias, 17 me and 3 carcinomas. These were distributed by the three pa as follows: •Rank Number 0 1 2. . 0 32 31 2/3 1/3 0 0 1 59 38-1/3 13 7-1/3 1/3 '0 2 17 3 2/3 10-2/3 2-2/3 0 3 5 2 1/3 1-2/3 0 1 3 0 0 2/3 1/3 2 We have to e.ss:=e that all J's slides of each class would have been distri- buted in these same proportions had they all been reread. 2hu6 his 28.8N of no:--v-ls in Table 1 would have been as 31/32 of 28&~ ^o:ml, 1/48 of 28.8:i~ hyperplasia and 1/9%' metapiasia. In this vay one calculates the table: 0 1 2 3 4 26.8 53.6 11.7 4.2 1.8 0 27 . 0. 0. 3 0.0 0. C 1 34.8 11.8 6.7 0.3 0.0 2 2.1 0.4 7.3 1.8 0.0 3 1.7 0.3 . 1.4 0.0 0.8 4 0.0 0.0 0.4 0.2 1.2 Total 66. 13.1 16.1 2.3 2.0 The comparison of J's original percentages at: be.tqp Jof this table with the ad3ustment by the averaged readings of.~tie ',patholo- gists reveals the fact that they would have read his slides ~EiTferently, and would indeed have given them a percentage distribution tf~ ~` +lar 2ror• the mean. This does not mean that J was wrong and they are ri.g'ht; it only means that there is a difference. Treating all eight in theAame vay, Tab)~ 1 as adjusted becomes Table 4. I rrR: 144 8 COk?IDGNTIAL: N,!(VNESOTA, TOBACCO LITfGATfOPJ -- ~ &MLW no~ CT P, N N ~'~0 4 r4 ~'.~

- k - Table 4 Adjusted Percentage Distributions for 2'.ales Age 25 and Up Reader Slides 0 _ 1 2 3 4 Index J: 116 66.6 13.1 16.1 2.3 2.0 .60 D: 74 62.7 13.3 18.0 0.8 5.1 .72 A: 73 63.9 15.3 19.9 0.0 0.9 -.59 E: 72 68.8 6.5 22.5 1.8 0.8 .58 B: 75 74.3 7.6 14.o 1.0 3.0 .51 L: 60 60.9 12.2 21.0 0.9 5.0 .78 I: 70 71.1 5.2 20.3 0.5 3.0 .58 x: 69 68.8 13.3 13.6 0.6 3.7 .57 Mean 609 67.1 10.8 18.2 1.0 2.9 .62 When one compares Tables 1 and 4, bearing in mind that had an,y three other rathologtets reread the slides, the adjustments would have been different (6), and further bearing i.n mind that the adjustments have been made by scaling up sAmples in the different cities of from 60 to 75 with one exceptionally large one of 116, it is obvious that most of the differences between the eight cities have disappeared and that it would be very difficult to separate out from the adjusted percentages items which proved that the pathological conditions of the lungs in the different cities were in fact different (7). Even when comparisons of general morbidity or mortality conditions in different places are dubious because of differences in reporting, the analysis of local reports by thoeefaniliar with local conditions has value. We hope that the individual pathologists who have been good enough to engage in this survey vill work up their data in any way they please. We shall be glad if our study fu.-nishes them something of value for theirs. Footnotes (1) These Proceedings, 43, 1073-78, 1957. (This reference is to the ProceedSngs of the National Academy of Sciences in which the previous note was published.) It is expected that the present one will be printed in the February issue of that journal, 1959. (2) This vill =ke the mean indYces, standard deviations and corre- lation coefficients of the previoua paper not strictly comparable with those here, but the comparison will not have to be made. (3) It has been necessary to omit four of the twelve cities. One of the cooperating pathologists failed to send in the data from his city; one had so few cases that it seemed better not to include his city in the rereading; one sent in no slides to be reread;-one had used the Swiss roll Snstead of the standard sections and we feared this might introduce non- comparability. , I CTR•.1419 CONFIDENTlAL: MINNESOTA TOBACCO LITIGATION ~ , T R 0 0 3 41_ sII

-5- (L) The 609 slides are not strictly random because a few more had been drawn randomly of which some had to be discarded because at least two of the three rereaders felt that they were not good enoug.h to be read at all. It is, however, our belief that this loss did not seriously disturb the randomness. (5) Tbe self-correlation coefficients have long been used by psycho- metrists, educational testers, and others to give one estimate of the reproducibility of the data. See, for example, C. Spearman, "Me Abilities of Man, Tbeir Ziature and Heasurement"; T. L. Kelley, "Crossroad.s in the Mind of Man;-A Study of Differentiable Mental Abilities"; J. P. Guilford, "Psychometric Methods." On pages 411 et se . of the last is given a brief general discuasion of various concepts re~ted to reliability. Our index is a rank index, an index of ordinal position. So are mazy, if not moat, of the grades or marks which teachers use. It mny be questionable whether one should treat ranks as cardinal nunbers, but that is widely done as we are doing it. (6) The two cities, J and H, top and bottom of Zhble 1, which showed the highest and the lowest values of the index, and the lowest and highest percentages of normals, were each first adjusted by using the rereadings of each of the three pathologists, and the results were in fact different, as asist be expected; but a study of their similarities indicated that an averaging of the findings of the three pathologists should give not only a stabler result but one which would give a standardization worth having. (7) Consider, for exaaple, what rereading by A, E, L has done to their ovn previous distributions: A old A new E old E new L old L nev Index 0.77 0.59 0.63 0.58 0.47 0.78 % Normal 38.7 63.9 66.7 68.5 76.4 60.9 For these three, the aees index vas 0.62 and has become 0.65 - arn insigni- ficaat char.ge. The differences from the old mean index were 1.15, 4.01, -.15; fro= the new, -.06, - .07, 4 .13. Descriptively the correlatio.^n is negative, though not significant. The differences from the respective meams of percent r.orr.al were - 21. 9, 46.1, 415.8, and become - 0.5, ~ L.1, - 3.5 a_-id again the correlation is negative. This is sitap]~y an indication of the differences inherent in passing judgments on'the slidea. If we correlated the two sets of percent norr&l in Tables 1 and 4, we would find ra 0.24; and if we correlated the two sets of indices, r 0.05. The striking phenomenon is to notice hov much the standardization has reduced the scatter. i s * / CT R; 14 2 0 CONFIDEN7]AL: MINNESO'FK TOBACCO LITIGATION '` 0 0 -*.3' 4 4 2_