Tobacco Documents Online

i Revrca.d h+.o ti. Troc..map ol Ue N.noNu Ac.osrt or bciz>,ct. Vel. 4S. f:o. f, pP UF-39S. liuel. 1959. SSOME STATISTICAL OBSERVATIONS ON A COOPERATIVE STUDY OF HUMAN PULMONARY PATHOLOGY. II BT EDWIN B. M't130\ AN-D MARY H. BL'RT.E i orrlcc or NAt'AL itsLALCH, lOSt'ON, lWBACHp{TTTE, AND TO.ACC'O lNDVtTRY auuacn COIfYRTit, I(Ltv TOti, M[w' TOtIC CO7MfYiUC6W aCYA16fr t6, I958 In our first paper' we gave some general average data for the readings of eight pathologists in eight diffcnent cities on slides made from sections taken in standard positions in run-of-the-mill lungs at autopsy, using the folloning elaaai6cations: normal, byperplasia, metapLlsia, atypical, metaplasia, carcinonis-in-situ and carcilloma. As carcinoma-in-.itu was found so rarely by any of the pathologists, that classification will be combined with atypical metaplwia in this continuation of the study; there will be only five groups and their rank indices' will be 0, 1. 2,3,4. Gc,u1921 / a~~ ~` `~°' ~ 11~`~~ 0~'.~ ~ ~~ ~ =.~

a9o PATXOLOGY: R'ILSOh' AND BUItKE Paoc. \. A. S. When we became convinced that the elasaification waa being made on different baeea by the diB'erent pathologists, we asked all twelve to r+ead a selected aample of 40 slides. This they kindly did, and we reported on the oonaiderable statistical difi'erenoea in the rr,adings. Aa the nuin object in all the work haa been to obtain oomparable data in the twelve cities for the degree of pathology in the lungs ex- amined, we stated that it would be well to have a considerable sample of the alides from all cities read by several pathologists. Tbe need for this is elear from the differences shown in Table I for the peroenta;ea of their slides placed in the S groups by the pathologists in eight of the twelve cities.' I TABLE I Piacz.rT.oz DdiancnoN..oa Ji"ss, Aoz 25 uxD [,'r lt..d.. alLdr 0 1 2 3 4 Isd.. J ............. 909 28.8 63.6 11.7 4.2 1.8 0.9-1 D ............. 941 67.1 21.1 7.7 11.1 3.0 0.82 A ............. 40$ 38.7 46.1 15.0 0 0 0.2 01-1 E ............. 630 66.7 9.7 18.7 3.6 1.3 0 63 B ............. 223 65.9 9.4 21.1 2.6 0.9 0 63 L ............. 2495 76.4 6.9 11.9 3.3 1.5 0.i' I ............. 669 74.4 8.4 16.3 0.0 0 0 044 H ............. 1416 81.8 9.7 8.0 0.4 0.1 0 2: Aleu, 61.2 20.6 13.8 3.3 1.1 o.c2 We were fortunate enough to 5nd three of the pathologists who were willing to read a sample of 609 alides drascn from the different cities by random processes.' We included also tbe 40 alides previously read by all twelve. The present paper is a report on the results of the rereading. The two sets of slides uill be treated separately. The gross results are in Tables 2 e.nd 3. TABLE 2 DzsrmcnoN or Toz.u or 609 Sunrs ov Rr:cADi.c n..de+ il.d. 0 I 2 3 4 l.d<, A.... ..... 609 359 93 177 14 16 0:;; E........ .. 609 348 25 212 6 16 0 6b5 L... .... .. 604 357 88 133 7 22 0.-1 6J Toul 1827 1066 206 472 27 S6 0:96 Reader A is high in atypicals (3) and Reader E is low in byperplasia (1) and high in metaplasia (2) compared with the other two. TABLE 3 DrfT1JaVTION or TEL 40 SLIDLB oN lttal.ADIKG Rsdtr •Gde 0 1 1 3 4 lsdr. A........... 40 4 4 27 3 2 1.8;s E........... 40 5 2 28 2 3 1 900 L........... 40 b 6 25 1 3 1' S Total 120 14 12 80 6 8 1. BS0 In this small sample, distributed very differently from the large one, the difTer- ences noticeable in the latter are not in evidence; but the distribution is significantl~~ different from that previously found by all twelve pathologists, %-iz., 48, 1?0. 223, 57, 32; though it is not aignificantly different from what the three rereadcr, found, viz., 1G,• 20, 64, 10, 10. The rereadings of the 40 slides by the three readers and their original readings I OT 4., 14 `'? C11-R, VIN 003444

VoL. 45, 1959 PATHOLOGY: WILSON AND BUAfiE 391 have the properties in Table 4. The first neader has not changed his mean sig- nificantly, the second has decreaeed his, and the third increased his, each signif- kantly. The meina thus have oome cloeer together. Tbe .eU-correlation ooetb- cienta vary from 0.65 to 0.86. TABLE 4 sre.r A .............. u... II 1.875 u..a I 1.800 4-0.075 t 0.114 0.65 E .............. 1.900 9.125 -0.225 t 0.103 0.76 L .............. 1.775 1.625 +0.ZS0 t 0. 09 1 0.86 In the random sample, the numbers of alides belonging to A, E, and L, respec- tively, were 73, 72, and 60. The compariaoo of the rereadings by each of his own alides is given in Table 5. It is aeen that the three patbologists are reading their own slides about as they did before and that the aeU-correlation coefficienta' are of about the same magnitude as for the 40 slides. •.rber bf.u Il TABLE 5 Nau I Hw. 11 - Yre 1 Corr.tatiea n. n A ............. 0.548 0.644 -0.0D6 f 0.089 0.60 E ............. 0.792 0.764 -F0.028 t 0.073 0.81 L ............. 1.150 1.a33 -0.183 ~ 0.142 0.65 With this background we may turn to the standardization of the percentages over elasxs which result from using the rereadin{rs of the three patbologists as a basis. The method is similar to that on etsndardizing death rates for age and eez against the age and sex distributions of a standard population. In Table 1. J put 28.8 per cent of his alides in the normals. The sample drawn for J from his 909 slides and presented to the thret pathologists among other slides, contained 32 normals, 59 byperplasias, 17 metaplasias, 5 atypicals, and 3 carcinomtu. Tbese were distributed by the three pataologists (averaged) as given in Table 6. R'e TABLE 6 R.ek NuoE.r 0 1 2 3 4 0 .............. . 32 31 '/. 0 0 1 .............. . 59 38'/s 13 7'/, 0 2 .............. . 17 3 '/$ 10'/, 2'/, 0 3 ............... 5 2 'A 1'/, 0 1 4 .............. . 3 0 0 '/' '/8 2 have to assume that all J's al.idea of each class would have been diatributed in these same proportions bad they all been reread. Thus his 28.8 per oent of normals in Table 1would have been distributed as "/b of 28.8 per eent normaLs, '/„ of 28.8 per cent byperplasia, and '/« of 28.8 per oent metsplasia. In this way one calculates Table 7. TABLE 7 r 0 1 ! 3 4 Oritinal . ?$.8 b3.6 1 1.7 4.2 1.8 0.... ...... ?~.9 0.0 0 3 0 0 0 0 1............ 34.8 11.8 . 6 7 0 3 0 0 2 ............ 2.1 04 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 Sundardued 66 6 13.1 1 6.1 2.3 2.0 CTq:r 1423 0 0 13 '~•4 `t

392 PATHOLOGY: WILSON AND BURKE paoC. N. A S. The oomparison of J'a original peroentages at the top of this table with the adjustment by the averaged readinga of the three pathologists reveals the fact that they would bave read his alidea very differently and ..otild indeed have given for them a percentage diatdbudon not very far from the mean. Thii does not mean that J w-aa wrong and they an right; it only mean that there is a diBerence. Treating all eight in the same way, Table 1 as adjusted becomea Table 8. TABLE 8 AnrorrcD Pri<czrrT.as DDrrintmoKs aoa M..t.cs, Aoc 25 .xD Ur A..der tt+dr 0 1 2 1 4 Iade. J ............. 909 66.6 13.1 16.1 2.3 2 0 0 60 D ............. 1W 62. 7 13.3 18 0 0 8 6.1 0.72 A ............. 408 63.9 1813 19.9 0 0 0.9 0 59 E ............. 630 68.5 6.5 22.5 1.8 0.8 0.Sb t B ............. 223 74.3 7.6 14 0 1 0 3.0 0 51 - L ............. 2495 60.9 12.2 21.0 0 9 6 0 0. I8 I ............. 669 71.1 6 2 20.3 0 5 3.0 0 58 H............ 1418 68.8 13.3 13.6 0 6 3.7 0.5: f.f.an..... ... 67.1 10.8 19.2 1 0 2 9 0 62 When one eoulpares Tables 1 and 8, bearing in mind that, had an.• three other patbologiata reread the glidea, the adjustments would have been different,' and further bearing in mind that the adjustments have been made by scaling up samples in the different cities of from 60 to 75 with one exceptionally iarge one of 116, it is obvious that most of the differenoes between the eight cities have disappeared and that it would be very difficult to aeparate out from the adjusted percentages items which proved that the pathological conditions of the lungs in the difierent cities were in fact different.t Even when comparisons-of general morbidity or mortalit}• conditions in di5erent places are dubious because of difterences in reporting, the analysis of local report~ by those familiar with local conditions has value. We hope that the inditiduol pathologists who have been good enough to engage in this stln-ey tt-ill work up their data in any way they please. We shall be glad if our study furnishes them some- thing of value for &irs. I Tbex Paocuotnos, 43, 1073-1M, 10.57. ' TLie rtll cuhe the mean iodicre, standard deviationa, and eorreLation eoc(5cienta of the prevwus paper pot stnctly eomparable with those here, but the eompariwn aill not have to be ms. de. e lt lu, betD Decey.ary to omit four of tbe trelve eities. One of tbe co-opcrating potboloG:Fta failed to send in the data from bis city; ooe 1ud so few cases that it seemed better not to include pi" eitv in tbe rereading; ooe s.nt in Do alidea to be reread; one had used gbe 6.+w roll inrtead of the standard satioD., and we f.arsd this mi8ht introduce Doncomparability. e Tbe 609 alidee art not strictly nDdom because a few more had beeo drawn randomly, of which some had to be discarded because at kart two or the three rereaders felt that thev were not good enough to be rud at all. It 6, boweTer, our belief that this loss did Dot seriouely disturb the ru+dotancn. eTEe self-correlation eoeffieieots 6a.i long been used by psyebometriata, educational tettere, and otben to 8ive one estitnate of tbe reprodueibiiity of the data. Sce, (or ecampk, C. 6pearrn3n, TAe Abililiea of Ilan, TAeir Nslure and lleanurnnenl; T. L. Rellc~., Crowroode in Ae Mind o,l alJon- A Slmdy of DifJe.enliaUe Xewtal Abililiea; J. P. Guilford, Peydwmetnc .11elAode pn pores 411 ff. of thc lsst u`iveu a brief grneral discuaaion of various concepts relatcd to relubil t> . Our index is a rank index, uu inJez of onlinwl position So arr nun%', if not most, of the gr:,,ler or marks which teachen ure. It ttu,v be quettionable whether oue should trrat nuk. ae cnrJuul numbers, but that is videh• done as rc arc doing it. 1424 CTR,1 I ~~~~~~ r11q ~~~ ~44Gj

Vot 45, 1050 PATXOtAGY: WlLSON AND BURRP 393 t The mea.n .alue of the three aalf eorrelationa oo tLe forty alidat tr 0.76 f 0.06, and of those on their o.n alideft is 0.6S * 0.09. We have six mutual eocralatioeu of the three pathologists in p.ir. on the forty slide. and dx on theL own; the ralua of the meana an evpaetisly 0.60 f 0.03 and 0.68 t 0.04. Owing to the amaL oumbcrs in the asmplel tbaee maan6 Eave bttk atatirtioa) su- billty; but ao far aa tbe .ridenoe goea. it tndicatea that tLe a.U eoenlatiooa an not much krger than the mutual eoerelationa. Or in other wnrda, the tLr.e patboiopeu r.produoe oex anotber'a r.adinp about as well as they reproduoe their owa--aa oaawuad by tLeae eorr.iationa. Tbe tsatural interyrntatioo i1 that their eemainiag diHaraooea an ebitEy foriuitoua or randam, due to lack of da6nition and poaIbly to lack of aomplete deBnabAity of the patSolosical mattrial. 8ome alide6 may be far from ole.r; ahould they be di.oarded? Some may have part of the mucowe Lek- ing; wbat about tbem' If w, A, m, a, c be the fraetiool (pcobab0ities) of alidea of a oertain area on which the worst oonditioo is oormal. byperplaria, mstaplsia, at7Pica1 metapLsia, and eucinoma, wsat would be the fractions on alidea which had t..ioe that area? • Tltis queatioo exnnot be anrwered witb any informatioa we pave; but it is interesting to eoo.3der and may auaeet iater.ating research. If the condition re.ealed by the alide were so widespread that it would ooctv oo both Ealvea of the alide of double araa, there would be oo didereneew in the probabilitia. At the other ertreme where the (worrt) oooditioo is so asarply bcali.ed that the eooditioa oo tbe slide bad uo relation to that on an adlaeent .qual ara., the fractioaa for alidea oost•itK a doubled area oould be ob_ t.ined from oombinatiooa of terms in the expanaioo of (n + A + trt + a + c)'. A+ an illurtra- tion, if for alidee covering a given area, the fnctiool(probabaitie6) are n- .70, A - .10, m- .15, a~.03, co- .02, then the results for the alide6 oo.eriat twice tbe area would be 0.49, 0.15, 0:6, 0.06, 0.04, respectively. If the work were to be done over, it might be well to record enougF, about the eooditlotv appearing on the alidea to learn aometLins about tDeir oor7ehtion6. Sucb a nudy might revea) evidence bearing on the question w6etber in truth the five oondition6 are in fact tucoamve. I The two eitiea, J and H, top and bottom of Table 1, which showed tse big~>elt and the lowest values of the index aod tbe lowest and higbeet percentarea of normalr, were each first adjueted by uaing the rertadinp of each of the three patbolopata, and the rerulu were in fact different. z s must be expected; but a study of tLe'v aimilaritiea indicat.ed that an averigir.g of the 6ndir,g-. of the three paDoioou 6hould give oot only a etabler result but on< wtich would give a rund- ardisatioa worth ' Conaider, for example, what the rereading by A, E, L has done to tbeir own previou6 dir- tributiona: A Old A He. L Old L Ne. L Oi1L\.. Ld.. .. .. ..... 0.77 0 69 0 63 0 a6 0 47 0 71 Iu seai eermu ...... 3i.7 a3.9 a6.7 a1.6 76 4 60 9 For these three the raean index was 0.6: and has become 0.65--an inaigni5cnnt change. Tnc dilfenncet from the old mean index were +0.15, +0.01, -0.15; from the neR -0.06, -0 0", ;0.13 De.criptively the eorrelatioo is oeptive, though not aigni5e.ant The diderentts frorrm the rerpective means of per oent normal were -21.9, +6.1, +15$ and become -0S, +4.1, -3.5, and again the correlation is oeptive. This is simply an indic3tion of the diAereoees inhcrent in paasing judgmenu on the alide6. If ws oorrelated the two .eu of per cent noraal in Tables I and 8. we would 6nd r- 0.24, and Lt we correlated the two sets of indices, r - 0.05 The atriking phenomenon to notice is how much the standardization has reduced scatter. CTRJ 1425 I ' CT~~'; 001344cc