Today's Discussion of the Mancuso Study: Notes and Comment

Page 1

Today's discussion of the kn cu so study: notes and comment some results. These centered - in a single, simple study of Hnnford did these persons, suitably ClaS6ified into cohorts and sub classified by age, sex, and race, live longer (or shorter) than a suitable control group. Cohort6 were defined%ccording t o t h e year within which employt o consider cause of death, length of time of employment, o r , needles8 t o say, radiation exposure. sibs of employees and "non-starts". A Ikon-start" is a person accepted for employment at Hanford following, among other things, a medical exam but who, for one reason or another,did not in fact become an employee. 2. The " u n i t of measurement" for each age/sex/race-specific subgroup of a cohort wa8 not the number of persons surviving t o the present, nor t h e percentage ~ u rvi val ,but rather the number of man-years lived to the was actually the quantity used, but this distinction appears t o be of no importance tothe findings. as 'lyeara'l when man-years were obviously the quantities presented. A t any rate, man-years were calculated based upon the date of entry upon About half or s o of the discussion was spent on presentation of The question was asked of data from this group: There was, as was pointed out by the discussants, no attempt Two separate control groups were used: \ The complement, the number of man-years not lived to the present The tables shown were labelled, confusingly, 3.A comparison of the number of man-years not lived was made, for employees vae a control population, for each cohort subgroup. Since the number of employees i n each cohort subgroup does not i n general equal the number of (for example) sib s in the corresponding subgroup, the man-years were adjusted t o make the number of employees equal t o the number of sibs. wati referred to 88 an age-distribution adjustment. cohort subgroup comparison. employees exceeds that for the controls ; otherwise score a - The null hypothesis for the sign test is that one expects a n equal number of + ' e and -'a; one then calculates the probability of the proportions This adjustment was done for each subgroup and A formal teet of e ign if i can ce - the sign teGt - wa8 appliecbto each For each subgroup, score a + if the man-years not lived for Suppose one cohort had t e n age-specific from the earliest date onwards, and for subgroups according to increasing age a t date of employment, there appeared t o be a wide scattering of +'a and - Is, with no apparent obvious a seo cia tion with cohort (time) o r When the reGults were shown in tables arranged according t o cohorts, Most of the sign tests were non-sign=-t.

Page 2

6 . Generally speaking, the sign test is insensitive. More sensitive tests exist which can be applied when - as i n this case - not merely the sign but the magnitude of the difference i n questiop is given by the data. The matter of insensitivity was not referred %hat I heard, 7. Also, generally speaking, the suggestion occurs t o me that the data analysis was done i n haste, perhap6 t o meet some recently imposed r e - quirement t o have some results ready for today. fiom the discussion, however, there is little justification for believing that any such recent imposition has in fact adversely affected the quality of the t o be frequently imprecise, unclear, glib. There was clearly no concept of a thought-out study plan based upon clearly defined objectives "highly meritorious" without one word of ju ct if i cation. I found all three speakers (Mancuso, Sanders, Brodsky) -it was asserted that the use of sibs as a control group was -how s i b s of opposite sex were handled was not mentioned. -the importance of length of employment was not voluntarily mentioned; upon questioning, the defense given was that once a n individual is exposed t o employment he can not become unexposed. (Gofman's estimate about) the 32,000 additional cancer cases in the U. S. population - an assertion presented without an iota of evidenoe, nor even any suggestion of any work done t o date involving counting neo plastic de atb. -the assertion(in various forme) was firmly and repeatedly made u y three discussants, especially Mancuco, that the data and data analysis in hand demonstrate conclusively "the feasibility of the study", whatever that means. Previous objections by statisticians, by John Storer, and by Alpen were referred to and summarily dismissed. -it was flatly as&-ted that the data in hand 'lalready't exclude but that in the minds of the discussants the only way t o proceed i8 t o expand the study. Expaneion should occur by adding locations of AM: plants and by adding other categories of employees. For example, it was suggested that bford employees engaged i n "minor construction" I don't see how anyone listening could come t o any other conclusion Oak Ridge people present is that you're whistling in the dark t o think that suitable data for the past can be obtained o r reconstructed. By reasoning t h a t was not presented, Mancuso, while not disagreeing with the description of the situation factually, argued all the more for continuing t o search recorda and dig for data. mentioned their plans for obtaining their "real" control groups by construction of matched samples from a 'lone percent work stunplet1 taken from social security records for the U. S a population. The discuseion of what to do toget radiation exposure cbta (records) It seemed t o me that the message from both the hn ford and The discussants also

Page 3

limits f o r precision that could be obtained on present and hypothetical future(larger) populations for the type of llactuariall'study discussed above seem fishy. Sanders gets very high precision by using post-death-(rnan-)years as his units. But one gets different precision by using the complementary units, life-man-years, indicating something is very wrong. Walter Weyzen points out that Sanders' estimates of confidence

Page 4

analysis,etc., there seelned t o be little interest shown i n simple exploration of the data - atask that many stat iG tic ian s and epidemiologists would, I ' m sure, put ahead of the task of significance testing. For example, while employees-with-sibs were compared with s i b s , and while all-employees were compared with sibs, a6 discussed above, there was no comparison that might have shown greater,or smdJ.er, differences than Besides glibness of statement, a seeming lack of depth inthe data independent annlysis carried out as a kind of check, was carried out by Brodsky and presented by him. ohi-square t e s t s of significance, subgroup by subgroup and in aggregate. Right now, I question the validity of all of this analysis. in private,to etate specifically how the teat statistic(to be compared with theoretical obi-square values) was constructed, Brodsky could not give a coherent or precise answer, but s a i d he would mil me something, although he would be away from the office for ten days. When I asked whether he had constructed a new teat statistic presumably t o be reported to the literature, or whether he had used a standard one, he wa8 not sure of the answer t o that, either. Brodsky, like Sanders, seemed content to rest on alpii ~ ~of results of significance test a which ic ally ; ordinarily,the investigator wante t o be sure that the data L ~ Z ~ A A Juo behave i n such a manner and that such findings aren ' t merely a result of the u8e of i he ens it ive test e. All of the data analysis referred to above wa6 carried out by Another analysis, which was det-icribed by Mancueo as a n This was an analysis involving This is a curious procedure, philosoph12. Ina general way, I was impressed with the down-to-earth, workmanlike approach of the b ford and Oak Ridge participants present. These included computer people (inthe case of bford, Computer Sciences Corp,). The new advisory committee being assembled by kn cu so appears stacked8 bio statistician ol ht on ? ) , and another statistician to be namedo out t o be recommendations; a policy that glJ of the detailed data - coded only to prevent disclosure of identifiable personal information and security information - shall be published a6 soon as assembled for the entire epidemiological and scientific world to examine. In a situation where other other scientists The following suggestions might well, upon further thought, t u r n they are not intended t o be consistent as a - The (Atomic Energy) Commission adopts, as soon as poseible,

Page 5

c a n ' t simply go out and "repent the experiment", the Commission readily senGes that it can not afford,scientifically or politically, to let the data analysis be captured by one group and thus make itself subject t o possible blackmail ( i f only for more money). -The AM= splits the present project into two, very unequally funding, etc., are different for the two. is the operational/health protection project to collect and assemble the (right kind of) data for orderly storage and rapid retrieval, project is funded by AEC, but not by DEM. an internal AE;c operation. several major plant sites s groupa Second, the smaller project is the scientifc/epidemiological project to examine and analyze the data, t o develop scientific results i f possible, and t o guide the other project to the right types of data and record6 needed f o r scientifc/epidemiological purposes. formal provioion i8 made for scientific/~pidemiological lias on from the second pro j oct to the first, not certainly, funded by AEC, and, if so, by DBM. a8 a matter of policy includes atleast two separate principal combinatione of these actions. The Mancuso group may o r may not be a part of the two projects, but t o be a part of the second project, t h e Mancuso group wina on merit i n the usual competition. rn ntr ihi y tion is primarily i n epidemiology and biometrics. The purposes, AJS administration, project direction, first, the larger project Tn is project is essentially Day t o day direction is carried out at the Coordination is through asite - steering The second project is probably, though The second project An in vee ti gator is sought at one of the National - If the Mancuso internal advisory committee continues ( somewhat inconsistently with the above), persons such as Robert Miller, Nathan Mantel, and Marvin Kastenbaum are on It.