Tobacco Documents Online

, /l.prut.d fr.e tD. Pree.ediap.l c!. Nano.a Ac.aarr or aaeecu. VeL 43. N.. It, rp. 10T1-107/. D.e.eb.r, 1#37. SOME STATISTICAL OBSERVATIONS ON A CO-OPERATIVE STUDY OF HUMAN PULMONARY PATHOLOGY 1 BT EnwzN B. WneoN axn M..ttT H. Btrazz Ornca or NAVAI. Rsaa.aQ, SormK. Maautarvsmy, um ToaAcoo Ixavarar Rau.ecri Co+oarrss, Nsw Yosx R.od IrJon /d. losdewry, N..wRbsr 20, 1967 Because o(the reported persistent increase in tbe incidence of cancer of the lung. Dr. Stanley P. Reitnann, of the Iaakenau Hospital in PhUsdelphia, arraaged for the Scient.ific Advisory Board of the Tobacco Industry Rexarch Commitree a co- operative study by a number of patbologistein different oenters across tbe country to examine systematically the conditions at autopsy in the bronchial mucosae of run-of-the-mill (not specially selected) eaaea. He Laa given a general account of the design of the study and of its major findinga.'. It is the object of the pne-sent paper to give some of the etatistical det..ils, with comments upon them so far as the total picture can be put together at the time of writing. The conditions found on the pathologists' slidea.ve.re elasai5ed in ascending order of pathogenicity from normal to carcinoma in six ranks: 0, normal, 1, byperplasia; 2, squamous metaplasia; 3, atypical metaplasia; 4, earcinoma-in-aitu; 5, earcin- oma-ba.sed upon the worst condition found anywhere an the slide. It is obvious that in such a classification a considerable degree of judgment is involved, as the line betnw-n adjacent classes is not so sharply definable as to be interpreted in an identical manner by diEerent judges; indeed, it is not always possible for a single person -to repeat precisely a classification he has previously made.', ' Thus the pooling of data is almost certain to involve what statistieians know as "mixed classi5cation," but it is one way to get a sort of standard .citb which amalier sam- ple-s may be compared instead of comparing each and every asmple with every oth er. TABLE 1 Mazs PacWrt.. a Drtusarre. O.za C1ua s. A.a aaau 0 1 3 3 4 ~ Under 21 666 86 6 6.0 6.3 1.1 ... 45-44 1,240 70.2 11.9 11.5 6.7 0.7 4S-bi 3,982 70 0 12.4 14.0 2.3 0 1 1.3 6S and up 2,702 68.8 12.8 13.4 3.2 0.2 1.6 Total 5,690 70.9 12 0 17.6 3.0 0.1 1.2 25 and up 7,924 69.6 12.6 13.4 3.2 0.1 1.3 The pooled dats for maks by age are 6iven in Table 1. As Dr. Rennann pointed out' there is no substantW increase in the percentages in each class after age 25, except in class S(carcinoma), w+hert it would be expected. If it is desired to compare males and ferrlales, it Ki11 be necessary to restrict the coverage to tbose places which sent in data for both men and women, because it so happens that those pathologists reporting on only male autopsies tlad a slighil;• 101-3 11 , I CTRr1g10 C1 ' NN 00123* 413,12'

1074 PATt10LOCY: W11.SON AND BURTiE Paoc_ r.AS. different distribution of percentages by elaasi5cation than those reportillg both. It TABLE 2F Facass la4s.rras lh.rn.ene. O•rsa Cua.r Aa aK.ou 0 1 i 3 4 s Under 25 400 91.8 6.0 2.0 0.2 ... 25-44 803 88.2 6.2 6.2 0.4 . 45-64 1,157 81.8 8.3 7.4 1.4 0.2 0.9 6S and up 1,301 88.2 7.3 5.2 0.9 ... • 0.4 Total 3,661 85.9 7.0 6.9 0.9 0.1 0.4 25 aad up 3,261 85.1 7.1 6.3 1.0 0.1 0.5 TABLE 2\1 CoaaurolrntKO liaua tsaesWr.os Dvntsvno. O.ea CLs.u Asc auwn 0 1 3 i 4 t Under 25 61S $6.5 6.2 6.2 1.1 25--" 428 73.3 11.4 11.3 3.2 0.8 45-6t 2,871 74.1 10.1 13.1 1.6 0 1 1.0 65 and up 2,044 72.6 11.6 13.0 2.2 0.1 0.6 Total 6,458 74.7 10.4 12.1 2.0 0.04 0.8 25 and up b,843 73.4 10.9 12.8 2.1 0.1 0.8 is seen that at all ages women have higher percentages of normal slides and lower percentages in each other class except for an inaignificant excess in carcinoma-in-situ at ages 45--64 based on two cases. Further, it is seen that, for women as for men, there are no really substantial changes after age 25. As to the slides for cases.cith primary lung cancer in males, the numbers and per- centages are given in Table 3. Comparison with the last line of Table I shows t.tiat TABLE 3 Au anu 0 r'°K"i'as D'.miaro.' a. Ci...u s 25 and up 596 50.3 8.4 18.6 6.5 1.2 16 0 the percentages of normal and hsperplasia are down, n•hereas the percentages of metaplasia, squamous, and atypical, and of earcinoma-in-situ and carcinoma are up, ae taight be especte•d_ Even if adjustment be made for the 16 per cent of car- einoma by dividing by 0.84, the statement rerrlairts true that normal and hyper- plasia are dowv and metaplasia, squamous, and atypical, and cnrcinoma-ili-situ are up; for the percentages become 59.9, 10.0, 22.1, 6.5, and 1.5 compared %%ith 69.6, 12.5, 13.4, 3.2, 0.1. Unfortunately, these results cannot be compared with Auerbach's' because his classification and his method of collecting data are both dif- ferent from those employed in the study. Corresponding to tbe l.ast lines of Table 1(all males 25 and up) and Table 2F (all females 25 and up) the different catiea for which we have records of individual percentages give the results in Table 4 for the means and for the standard do- TABLE 4 0 t rsac.n.os D?rru.rno. Or ~ Cusa. , Atalr. 65 2 f 13 S 15 6 t 13.1 15 0 * 5.3 2 0 * 2 0 .15 f.10 1.1 f 1 0 Femake 81.2 3 0.2 12.0 f 10.4 S 8 * 2.0 0.7 * 1.0 OJ x.02 0 2 f 0.4 I CrR,1411 0 0 03 43 :J"

VOL. 4.3, 1957 PATXOLOCY: WILSON AND BURKE 1076 riationa of the distributions about them. Although the me1r of the percentages do not differ materia.lly from the mean percentages, the relativdy luge estimated standard errors of the distributions about them are noteworthy and ssgnify either that the patbological conditions in the different places u+e difterti.nt or that the dif- fenent pathologists were classifying differently or both, and tbere is no way to dis- tribute the variation over these two sources. An experiment was made by asking each of the twelve participating pat.hologists to send in a few alides, some typical and others problematical. From tbete, Dr. Reilnann selected 40 which were elaasi5ed independently by all tweh•e. There were therefore 480 items spread over the 6 cla.sxa 0-5, respectively, as 48, 120, 223, 50, 7, 32. Division by 12 gives the average values 4, 10, 18.6, 4.2, 0.6, and 2.7. For simplicity, one may combine 0 and 1(normal and byperplasia) to get 14, keep 2 (squamotu metaplasia) at 18.6, and combine the three moet serious conditions 3, 4, and 5 at 7.4. A chi-aquare on the table of 3 X 12 observed values gives about 75 for 22 degrees of freedom, wbich is very significant of diffesences in the classifica- tion.' There is no way to estimate what would have been the diHerences or their signifi- cance, had the tviceh•e participating pathologists each read the 12,000 or so slides in the main study or any random sample of them aufficiently large to establis)z a reasonably stable set of data for the purpose of comparison between them. If one were satisfied to make three groups, as above, namely, clisses 0 and 1, class 2, a.n.d classes 3, 4, and 5, it would be well to have 400 or 500 randomly selected slide: read by several pathologists. Another way to deal with the rnatter is to make a correlation table of the pairs of clissi5cations by the 66 pairs of classifiers. The table may be corstr,rcted i: a symmetrical way, resulting in 132 X 40 - 5,280 entries, eicb reader appearing as Reader I and as Reader II. TABLE S lrw.ceruc Couu.anov TsLs roa Rasca. 1 1 7 4 < TeW 0 190 189 115 34 d?8 1 189 6fr1 {09 129 1 8 1, 320 2 115 409 1,695 206 19 6 2,453 3 34 129 908 138 13 28 bS0 4 ... 1 19 13 12 82 7_1 6 ... 8 6 28 32 278 3S2 Total b28 1,320 2,453 650 77 352 6,280 In this table 54.9 per cent of the entries are in the main diagonal, showing that degree and only that degree of agreement between the pairs of pathologists. If the two adjacent diagonals be included, the result is 87.1 per cent of the entries, 9ndi- cating that 12.9 per cent of the entries were for cases in which two pathologists dif- fered by two or more classes in their estimates of the most .erioue condition on a slide.' One cannot convert the correlation table into a atandard correldtion coef- ficient without implying that the six classificatiorts 0-5 are proper numerical mea.- ures of the relative pathogenicities.' But if one uses this assumption, rl.ll - 0.G92 with S per cent and 95 per cent points at r- 0.576 and r - 0.787 as figured by Fisher's r-tronsformation. And on this index the mean is 1.883 (*0.190 ~ =L 1.183,---.r3-9). I crR. 1412 0 C `Q~ ~'IPI ~'~03~'~ ~41

1076 PATXOIQGY: WILSON .4ND BUR1;E Paoc. \.AS. If one is willing to rank the di8'ereat readers by using this numerical measure of relative patbogenicity aa an index, one finds for the twelve the following: 1.676 f 0.196 1.600 * 0.211 ' 1.650 f 0.201 . 1.615 * 0.201 1.700 * 0.173 1.Ei6 t 0.169 1.900 * 0.179 1.925 * 0.1117 1.950 * 0.179 2.125 * 0.161 2.225 * 0.198 l.400 CL 0.167 It was interesting to us to note that the order of these means for the different pathologists correlated well with the order of leniency appearing in their reading of all their own slides. This stiggests that the distributions of their findi.ats in their own hospitals are due in part to their habits of interpreting their slides and that interinstitution comparisons of their findings should not be made without some a!- lowances for these differences, Le., for what astronomers are wont to call the r`personal equation" of the observer. Classifying an individual alide is a "clinical" problem, as is any exercise of judg- ment about any individual cax,wbether pathologic, meteorologic, social, economic, medical, or psychological. Statistics deals with a population, and there may be such wide scatter about mearu or trends that the statistical findings may be of little use to the cli.nician. Moreover, no statistical study can include all the %-ariables which may be acting. An important function of the clinician is to be alert to the special importance for the particular case of some variable which is not in the statisticians' firqdi.ngs for populations but may be controlling in the case before him: and, moreover, there are aspects of any clinical problem which may be sensed b* % - the keen and experienced clinician but are not yet generally incorporntea into standard statistical methods.' Even hnd it.been practicable, the co-operating pathologists should not h~ve been forced to agreement with respect to standards of classification for this atud}• at the danger of modifying their indMdual standards; for they were all experienced scientists, primarily responsible for long sesies of patbological deterrnin;.tiorts in their respective hospit.aL and ehould, therefore, be concerned n•ith the continuity of stsndards in those series. Science is built upon the deductions one can draa- f rom data freely assembled by different scientists with proper allowance for the inevitable differences between scient.ista. One can come to forced agreements, gi.-ing an arti- ficial sense of security. The co-operati.ng pathologists, after getting well into the study, were unanimously of the opinion that through it they n•ere lear.ting atuch that a•as new to them about pulroona.ry pathology and profiting greatly from their mutual contacts relative to it. The original hope of Dr. Reims.nrt's plan was to correlate the pathological findings with such environmental variables as amount and type of smoking, rural or urban residence, Eeograpbic location, occupation, etc. We hope that developments not ready for this meeting may later make it possible for us or others to extend this statistical discusaion to some such matters; but in the short time available for the presentation of th,is paper we have been unable to exhaust even thox m:.terials and calculations which we consider germane to the topics here discu.%bed. t S. P. Reimann, "Patholo&-anatomic Study of Human Lunp: Tobacco lndustny Resrorch Committee Communication, June, 1957" (to be revised for publication). Thc arrrrmcnt was tn use approzi=tely 15 sections in eertain de6oite paru of the trocheobronchW trcr. At kist l1atf of these .ectiona "re to pave au much aa 50 per ecot of the muco.ae intact. Thc wcrate num- ~ i 41:3 © ~ CTR Ntq 002314,3'

Vut.. 43, 1957 PATHOLOGY: WILSON AND BURFE 1077 ber of 6uch aections proved to be about 12. Tbi. was not w detaikd a.tudy as that of Oscar Aucrbach (Nae Bwpl. J. X.d., 21%. 97-104, January 17, 1967), who tooh a great many sections; but hie study was at a sintle plaoe and haa l.poctad only aliBDtly osr l00 eaaa, w6er.as tLis rwd,v tnvol.a orer 1,000. I Btudia of "aron of ob.er.ation" sa.a 1o0[ b..n made in tL6 mct ac'seooem oo the aa+umption that tbere was aoa+etding under observation which was far lr variable than werc the ob.ervnr'e po.aibilitir. of ma.wlrin8 it. la tbe medica! and ot6.r koai..ct aciano.e the word "arror" lta. aome- times b«a used wLere dlfferweam of Enformatioo or of Judgment ia complicated aituations are involvrd to wuL a d.p« that to attribute erroe to those who differ is uade.irsble. A few of the .arlier studies of reiiability an: lt. C. Cabot, "Diavnostic Pitfalla ldeatibed during a Study of Three Thousand Autopsiee", J. AwL Hd. Aa.ec., 69, 2295-4798, 1912; E. E. Soutbud and A. W. Stearn., "?Le Margin of Error kl Pqe.II[atfic Boapital Di.ptoaa.," Bastow Ned. Surp. J., 171, 895-900, 1916; Edwin B. W'll.on aad Julia Deming, •'Statirtical Comparison of Prvchiatric Di.plo.ee in Some Maa.acbu.eu. State Sorpitale during 1925 and 1926," Qunr1. Buff. dfa... Dcyf. Jdentaf Di.wsna, Vol. 11 (repcinted in 6d[soDJtrnLia: &atiarioaf Studiu Jrom Gk Bo.ton P.t,rhoya(Aie B'oapieal, 19ZS-34, eollected for pri.ate diren'butioo by Dr. C. *,iacfie Campbell); C. R Doering and A. F. Raymond, "Rsliability of Obaer.atioo. in Pqchlatric and Related Char- acteriaic.," Arw. J. Ai.Lepryd:nby, 4, 249-257, 1?34 (tzpcinted in Satisoylrewia). E. E. SieBkr, "liicradiapocs of Carcinom.-in-Siw of the Uterine Cervix: A Comparative Study of PatLol- ogi.ta' Diagoo.ea," Contv, 9, 463-469, 1958; and C. C. Birkelo, W. E- Chamerlain, P. S. Ph,elpe, P. E. Schools, D. Zaeki, and J. Yeruahalmy, 'Tuberettloe:u Ca+e Fendin`: Comparison of Eder- tiveneae of Variotu RoentgtnoRapbic and PhotoEuorpapbk Metbod.," J. Am. M.d. Aaaoc., 133, 359-,366, 1947. e lf one doea aot wish to have so amall a theoretical .alue u 7.4 in LSe ehi-.quare teat, one may make two diviaion., ri:., 0 and I and 2, 3, 4, and 5 with the rerult that ehi-quare u about 30 for 11 deRea of freedom, wherw the 1 per eent point is at 25. c This 6hows that actually in praetice the epread in the judpaentL stated is not jwt a borduline atf a'v. Another way to show the di.4erencee is by oomparing the ka.t severe and the most severe read- inp of the individual alides: l.asr ljo.r tr.us- te+ta. 0 7 2 1 4 ! Tae.w 0 .. '1 6 30 10 3 .. .. ~ 6 1 1 I? 2 2 2 2 6 3 .. .. .. .. .. 2 2 ~ .. .. .. .. .. 1 1 a .. .. .. .. .. .. 0 TeW 0 1 14 18 7 4 40 . \ote the peruliar di.tributioo of the frequenciee of the didercnces in elaae between the kaut and mo.t aevere readinp, namely, 2, 8, 18, 11, and 1, re.pecti.ely, for didereneee of 0, 1, 2, 3, and 4 eLa.ea. Still another way is to tabulate the 40 slides according to the number of different elaa.a used in elaa+ifyin8 each. Tbe frepueacie. art 2, 11, 17, and 10, trrpectivtly, for the use of 1, 2, 3, and 4 elaa.e.. Tbere i6 0o 61ide for whieh all reeden asreed that it was normal or that it was earcinomz. I It would be a separate and probably invol.ed atudy to d.tarmine Eow an index of patho- genicity abould be made up. Naturnlly,'bormal" would be taken a. 0, and byperplaaia might be aaeigned the fisur. 1, but, apart from aome tatricate eoosiderstioaa of pathology and rtati.tice, one could only ari8n classes 2, 3,.4, and bgemal values n,.,., aad c to rtand for wbaLever might be the pathoteaicitiea of those ela..es ralative to byperplaaia. That 1< m< a<. < e would be aaumed becauae of the badc aarumptioa that tba elaa.e" are in inu.asin8 drpee of patboteaicity. Con.ider a.imple eoerclatioo tabk between two r..dern of the 40 alide..uch as: 1tUtl A Ita.)p .t. ^ 0 3 3 3 4 a 0 , 1 2 .. .. .. 2 1 1 3~ ~~ .. .. 3 ~ .. .. .. .. 2 .. 1 .. .. .. .. .. 0 . I CTR, 1414 i C`I- ° ~`~~`~ C~03~-~~'~

1078 PATXOLOCY: WILSON AND BURKE p1a N-AS. It ie eku that tbe.e readere ." in 23 of 40 cue. aad din4ru In 17. Tbe tuer of Reader B ovsr Rwder A on the iade: would be E- 1+2m+4{rn-1)-6-n-(*i-1)-4(a--.n)- -!+ b+n-?o, aad any r.lua a..igaed to n> 1 and a > w wbieE mak. E positl.r will make Rrder B.evexr tLaa A. U we eet wi - 1- n' > 0 aod a-.w - a' > 0, ws hare E- -{ + 4e+' - Zo' as tbe azar. If we p3ot a' a. ordinate and .n' a. ab.eina, we eaa ay thai, for .ay yoint (w', a') !n the Errt qoadraat which lim above the linc -{ + Iwi' - 24, - 0, Raader B will be 1er.eran than A, w6erau, for any point below It, B will be mrx..e.are. II wa tah. +w - 2, a- 3, wti' - 1, a' - 1 ae wm did, tbea E- -2 and B u tbe ler rerne. e Tbe.e matters 6..e been di.cuarod by P. I:. 1See1il !n his book Cliwiool .aew 3tntialionl Pre- lidien (Mianeapolie: Dni.ersity of Miane.ota pr.r, 10.54) .ad befon the 1456 convention of the American PrycEolopcal As.xutioo tmder the t3tk "Wben 6'haI1 We Use Our Beada Instead ol . FormuL'" and in Sia artick "Congturetipaal 6corinj," J. C.nrullinp PtycAe2opy, 14, 16&- 171, 1450. 0 I %-d*TR• 1415 ~` f R MH 0~'.~ ~34:1 ~•''