Document rej6nLN0deXz7oadZwoM3M6er

CAMBRIDGE WKIi 02236 k / 0T046939 TO: V. I3akcman/New York 0. M. F.ivorito W. J. McCaig M. Stollcr R. M. Vining J. VV. Wolter DATE: 28 July 1982 FROM: H. A. Eschenbach SUBJECT: Mortality Study -- Richard Monson Attached is a copy of the "quick-and-dirty" mortality study done by Richard Monson at the Harvard School of Public Health. Apparently, he completed it rather quickly after our meeting but sat on it while he went on a trip to China. As you will observe in Table 1, our major problem is death from respiratory cancer. This is no surprise. There are no other statistically significant aberrations from norms. While we cannot say for certain that our collection of death certificates is complete or representative, there is no reason to believe that there is any bias in our accumulation of them. There fore, one must assume that NIOSH's study will be in the same range. NIOSH would possibly not pursue the mortality portion of their study if we provided this information to them. On the other hand, I see no advantage or disadvantage in doing so. Harry Eschenbach HAE/cs attachment 00120!42 151210^1 H 0046940 CojM.1 of D;ath Among Sixty-six Libby Employees Kictiard R. Monson, H.O., Sc.D. 4/5/82 00120143 15125062 0f>04941 This is an analysis of 66 death certificate'. nr L i t*by employees of w.K. Grace & Company. HETHQO The 66 deaths occurred between 1950 and 1981, inclusive. The deatns were coded according to the 8th revision. International Classification of Diseases. The observed deaths were distributed by age at death and year of death in five-year strata. Expected deaths were computed using the USDR computer program (1). Expected numbers were based on age-time-cause specific proportional mortality ratios for U.S. white males. RESULTS As seen in Table 1, there was an excess of cancer of the respiratory system - 14 observed and 5.0 expected. Two of these cancers were meso theliomas. There was only a small excess of death from non-mal ignant res piratory disease. As seen in Table 2, most of the excess being cancer has occurred since 1975. COMMENT The data in this report are consistent with excess death due to respiratory cancer as a result of exposure tc asbestos. REFERENCE 1. Monson RR: Analysis of relative survival and proportional mortality. Computers and Biomedical Research 7: 325-332, 1974. 0012ul44 15125063 l.ihn? . ..auso. . : <:. .! .* n* i <:m i . OtJ04942 ICD No.a 140-209 150-159 160-164 185-189 390-458 460-519 800-999 - Cause of Death All All cancer Digestive Respiratory0 Genitourinarye Residual Circulatory Respiratory External causes Residual Observed 66 20 3 14 3 0 28 5 4 9 Expectod^ 66.0 14.0 3.9 5.0 1.8 3.3 35.5 4.3 4.7 7.5 Obs/Exp --.i -f.- 1.0 1.4 0.8 2.8d 1.7 0.8 1.2 0.9 - a. International Classification of Diseases, Eighth Revision. b. Expected numbers computed on the basis of age-time-cause specific proportional mortality ratios for U.S. white males. c. Lung cancer - 11, mesothelioma - 2, mediastinal cancer - 1. d. 951 confidence interval = 1.7-4.5. e. Prostate cancer - 3. 0012 0i4 5 1512506* lib i<' UD'jL'rytMl dlul i.'vj'i'i t. 11 fi-Jti*i> -r . or ro'.iii r ot death _____................ . jMLl.-| Year < 1965 1965-69 1970-74 1975-79 1980 Observed 0 3 1 7 3 Expoc tod 0.5 0.8 1.1 1.8 0.8 14 5.0 >1 ! tl-J ^ , 0S04G343 1 i i i I i<j 15125065 004tt944 ........1 h; i i i'll Analysis of Relative Survival and Proportional Mortality* Km mvp'i K Miisvis /< ,MdWi M/ ,|/ / (* . /' ^ `/'id * //.!*'I. h>nt'it\ m,,*/ I' KtxciscJ August 21. 19?* \ eo.***/ iui program oult/n.c v..ic accairw scvr.tev*S's\tK m.**i t'.its or incidence r :ct i Jw'vi-lvJ The pro^iam is i.-.-j tu determine ir.c cvjvcica mimlvr of deaths or , * ' to<wjsr n.i cohort .h:..- >:j group of per^on^ know r. i. he dead. the number t ivu . h ic J v*l o. .Jvh cause can he calculated i.vi. propcruon.il mortalities. Mortal :. ra'o pjhJ oh Ur.iteJ St.icsv ual staiitics have Iven assembled lor 57 causes of d.ath r r:u!cs. while fviu.dcs and non*whitc males foi each lAe-vcjr ape and time ixr.oj li>-.'.i l'J2* tl.toitph |W*i Pioporuonal inortalaics aiul propomonal cancer mo.tal *.ivt hoc tv;n calculate*!. Incidence rates from the Conruviicut Cancer Rejivtry are I he computer program i> pcneral and can he minified io use other sets of r.iitf* in dcicrmimnp the espeeted r.urr.NT of Jeaths in am vt.:dv population. i -' i dissimilar epidemiologic techniques. the estimation of relative survival and p:.'p uuo:t;tl r.'onahiv. share j common ground: each is dependent upon a standard ! deaths In relative survival, si.mdard death rates are calculated and are used to c%vnpi:te the cv peeled urn*! ef ol lota! deaths *n a cotton In proportional n.. r: ::.t\ the number of deaths d;:e to a specific cause is dtv :de.! h\ the number of l* :.d .te.iths aiul this proportion is useJ tv' compute the evpccu\i number of cause*po*.::e de ohs .ii:,.v>mi! a croup o!'ik\c tsod persons l?. s paper describes a computer program developed to an.dv/c cohort studies u* *ie tel.i'. e siifvival and its analogues and to analv/e stu ! u*me proportion.^ rv. :i :!u\ I hoe technique* have been luliy described ar.d illustrated hv other .vat!.. :' i / .*i The standard death r.ite* ar\! proportional h:-m t.il,nes arc based on c.aise ... e ::.ne-scv-face->pccitic data tor the population ot tin* l aued States from i->:4 ' **< Cai 11 t v11*n in Si vsn wp K mo ;'/ i *u. l or e.twh ve.ir iron !'7?s ** I'M'". >\^^ .-numerated or '! .t . Vvme popu! uon the \ mted >; iu - de. *. e.* *tr.uion area was *S .v : * .i rf.v1.1-17:.* ..: i uC IvM "<*t t* if."-' ti, n . .... i b'siauic anl ii :* . . r !*|. I soM; f . V. ."Ml lihii'i.ii I mr.*- . . . , v. j , I* f I.-. VM i !l.', :!i Vu'nu*< 15125066 00120 1 4 06046945 3> MIC IIAMI K. MIINMIN uii.u.nu.l i.\ > I'or (lie\c;kn 1*125 .14. the population wa\ grouped in 10-year age group-in 4. 5 M........8St) I or succeeding >c.ifv the population was grouped in live-scar age groups iO 4. 5- 9....... R5f). I'or c;ich age group. (he average population during c.idi luc sear period (I92V-29, 19.10-34.........1965-67) was calculated for uluic in.ties, while females and non-white males. SiamLnl ,1,-ailis. For each scar from 1925 to 1967. the numbers of persons dying from seheted causes of death were abstracted from United Stales Vital Statistics (5). spccdic for age. sev and race. To date, information on 23 categories of cancer and 34 categories of other causes have been abstracted. Prior to 1933 the data are derived from the states included in the death registration area. The age-time breakdown corresponds to that for the standard population. The average number of age-specific deaths per >car within each five-year time period was calculated. For each cause a causc-of-dcath code was assigned according to the Seventh Revision of the Inter national Classification of Diseases (6). Since codes and coding procedures chance w ith time, causes from different times were grouped according to title of cause of death rather than according to 1CD code. Siamlanl rates. For each age-time period for each race and sex the number of deaths per 1000 population per year was calculated. For some five-year periods the TABLE I MotTALirr Ratu ro all CaL'su roii Whit* Fimalxj is Ukitid statu Accoumno to Aci aso Vi ah oi UrATii. Exmismo as Nuuuu or Dlatiis m 1000 Poruumm rta Yr** Ape at dc^ih 1925 19 JO- 1935- Year of death 1940- 1945- I9JO- 1953- 1960- 1965- . 0- 15.9 U.O 12.1 9.7 7.4 5.6 5.0 4.4 38 5- 1.6 1.3 1.2 O.S 0.6 0.5 0.4 0.4 0.3 10- 1.6 I.J 1.0 0.7 0.3 0.4 0.3 0.3 0.3 IS :.9 2.1 1.5 1.0 0.9 0.6 0.5 0.3 0.5 20- :.9 2.1 2.1 1.5 1.1 0.7 0.6 0.6 0.6 :s- 4.0 J.2 2 6 1.7 1.3 0.9 0.7 0.7 0.7 J0- 4.0 j.: 3.0 2.2 1.6 1.2 1.0 1.0 1.0 JS- it 4.8 3.7 2.9 2.J 1.7 1.5 I.J 1.3 40- JS 4.S 5.0 J.9 3.2 2.7 2.4 2.3 2.4 IS- 10.1 9.0 7.0 3.7 4.S 4.1 3.7 J.6 3.7 SO- 10.1 9.0 9.9 S.6 7.4 6.4 5.7 3.5 3 6 JS- 21 0 19.6 14.7 12.9 11.0 9.7 8.1 1.1 S3 to- 21.0 19.6 22.4 19.6 17.2 15.2 13.7 12.5 12.4 65- 49 2 43.4 34 9 30.7 26.7 24.2 22.6 20.9 22.3 70- 49.2 45.4 54.2 31.1 43.J 40.4 36.6 34.0 J3 4 75- 118.2 107.9 92 5 81.2 74.6 66.8 62.6 37.3 55.0 SO- 118.2 107.9 141.7 131.9 114 J 109.0 105.4 IC0.4 94 4 *5* 281 1 244.7 220.5 208 9 191.4 196.0 193.8 200 i. 15125067 0012 u140 0ft434fc HI I AlIXT St MXIVAl AM) I'lllil-UHIIOSAI MlllUMIIV 327 number' nf deaths for certain cause* are not published: the rate* during these periods arc assumed in be those of preecJme or succeeding |H.-nvd* 1 he rates for 1965-6*i arc based mi deaths from 19f5 67 only. As an example, total death rates for all causes for wlmc females arc prcscincd in Table I. (The rates have been truncated for purposes of presentation.) /,nifuirtit>-i,il niortaliiicj. The denominator of a proportional mortality is the number of deaths due to all causes The numerator is the number of deaths due to a specific cause. The proportional mortalities for each cause were calculated, accord ing to the age-time distribution presented in Table I. I'/oporiwnal ftiticcf mortality. The denominator of a proportional cancer mortality is the number of deaths due to cancer. The numerator is the number of deaths due to a specific cancer. Proportional cancer mortalities were calculated for each cancer, according to the age-time distribution presented in Table I. Data Analysis Using Relative Survival For each member of the study cohort, the following information is needed: sea. race, age at entry, year of entry, years of survival, status at end of study (live, deceased or lost to follow-up) and cause of death, where appropriate. The steps in the computer program are described belosv. Except for step 8. each table has the format of Table 2. The data may or may not be calculated, depending upon the program option selected. Also, options arc available as to the number of tables to be printed. Step 1'. The death rate matrix is read. Step 2. The age-year-of-entrv distribution of the study cohort is determined in ftxe-ycar groupings. Step J. The age-time-spccifie person-years of the study cohort are calculated. A person is assumed to hn\c entered on July I of the scar of entry and at the midpoint of the age of entry. If a person is lost to follow-up. withdrawal is assumed to have occurred on July 1 of the year of withdrawal. (Options are available to consider lost persons to have died at the lime of loss or to be atix-e until the common closing data of the study.) A person entering and leaving in the same year contributes 0.25 person-years. A person who is a live withdrawal is withdrawn at the common closing date, usually January l or July I. As an example, consider the person described by Hill (7). He entered the study in 1952 at age 23 and left in 1962. His person-years of survival would be distributed as shown in Fig. I. The method used in this program is less precise for an individual than Hill's method, since the exact days of survival arc not computed. For the total population, however, essentially no error should be introduced, since persons who are given too many partial jears of survival are counterbalanced b\ those with too few. OGtiilfcMS MIC'll Alii> K. MM.NMIN 1 lie Ii-i.iI in J study of survival following carunoin.i of the endometrium < S1 .ire distributed as illustrated in Table 2. i'/r/> 4 Hie age-time distribution of the observed deaths is determined. The observed rjtes are calculated by dividing the observed deaths by ihc person- years of survival. Sup .'. The apc-timc->pccilic expected deaths arc calculated by multiplying the ngc-limc-spccilic standard death rales by the agc-iimc-xpccilic person- years of follow-up. YEAR lac. I. IVron-vc.m experienced bye person entering at age 2.1 in 19)2 ar.J leaving in 1962. Stt'p 6. Mortality raiie>s are calculated by dividing each observed number of deaths by each expected number of deaths. Slip 7. The probability that the observed number of deaths differs significantly from the expected numbers of deaths is calculated using a modification of the Mantcl-Hac.iszcl procedure (9). A Manicl-Haenszel ,Y; is calculated for each marginal ratio as well as for the total observed,expected ratio. The Poisson approximation is made in determining the variance of the expected deaths, that is. the variance is assumed to be equal to the mean of the expected deaths. The resultant ,VJ is conservative, that is, it is less than the corre x:sponding calculated using the binomial variance. Sh-pS. The relative survival life tabic is calculated (/). The standard errors are estimated according to the method of Greenwood {In). For each year of survival, the following numbers arc calculated, as illustrated in Table .1 (the data in Table .1 arc limited to the first ten years of follow-up): l.X. number starting each interval: WX, number of live withdrawals; UX. number lost to follow-up: OllStDX), observed deaths; EXP(DX). expected deaths; OHS I XI'. oberveJ. expected ratio; SURV, crude interval survival (i to 15125069 004Gt9 KIlAllVI Sl'HVIV \| VMJ l*M (111 MU II >v VI. HK I VI II V 329 /4 !). RSR, relative interval vurvival; SCRSR. standard error of RSR; CSL'KV. crude cumulative survival (0 to i + I): C'RSR. relative cumulative survival: SI.CRSR. standard error of CRSR. TADLE 2 I'l KVIN'-Yl ARS Ol SURVIVAL IN A t Of 76t Willi l Kl MAI I * ri>Lt<>VVINC TALATSItNT IO Carcinoma or mi Esuosuimiusi Accorijing to Aca and Yiar Year Ape 1-920 t.925- 1930- 1935- -1940- 1945- J950-. 1955- I960- 1965- Sum 253035<0 45505500657075F0 65a Sum 0.5 9.5 3.5 1.5 4.5 It.3 14.0 6.3 10.5 40.1 47.1 14.0 53.3 113 5.* 51.3 96.0 4.0 26.3 63.1 o.s I9J 45J 2.3 19.0 1.5 1.0 1.5 49* 210.0 377.3 25 4.5 17.0 15.1 31.3 I0J.I 123.* 112.5 52.* 31.3 12.3 0.5 3.5 0.3 6.0 It 5 54.0 *2.3 95.1 166.* 117.* 100.3 37.5 17.5 1.5 7.5 9.0 20.0 71.* 132.0 171.5 159.0 204.3 161.5 90.0 21.5 7.0 3.5 5.5 19.5 370 69.3 120.5 256.0 295 * 20*.* 209.0 141.5 63 S 12 5 4.0 13.0 10.5 3*.5 900 129.5 237.* 346.0 343.3 203.5 1*5.* 95.5 43 0 1.0 10.0 19.0 14.0 55.5 112.0 165.0 263.5 347.0 329.5 158.0 1060 65.0 0.5 3.5 4.5 3.5 15-3 27J 36.0 49.0 64J 37.0 14.5 11.5 13.5 39.8 (1.5 159.0 391.3 729.0 1214.* 1543.8 1536.5 11*6.3 702.3 332.0 142J 515* 760.0 1063.5 1442.5 1740. V 1645.5 267.5 *072.0 TABLE 3 Rti aiivi Survival Ratios for 761 Wumin Following Tri aivii si ior Carcinoma or Tut Ekdomctrivm (lit Tint ior OcscRirnos or Data) Year LX WX UX 0DS(DX1 EXIXDX) OBS EXTSURV RSR SCRSR CSURV CRSR SECRSR 0- 761 1 1- 669 0 2- (07 1 3- 56S . 1 4. 339 0 5- 521 2 6- 497 5 7- 4*1 12 F- 459 24 9- 424 20 0 0 0 0 0 0 0 0 0 0 91 60 40 2* II 22 It 10 11 11 14.4 6.3 .88 .90 .01 SB .90 .01 12.* 4.7 .91 .93 .01 .80 .83 .02 12.7 3.2 .93 .95 .01 .75 .to .02 ll.t 2.4 .95 .97 .01 .71 .77 .02 11.5 1.6 .97 .99 .01 .69 .76 .02 11.5 1.9 .96 .98 01 .66 .73 .02 11.7 0.9 .98 1 00 01 .64 .75 .02 12.7 0.8 .98 1 01 01 .63 .75 .02 12.7 09 .9! 1 00 .01 .62 .75 .02 12.0 09 .97 1 00 .01 .60 .76 .02 Sum _ 296 0 465 263.6 18 _ _ _ m-m 00i20151 06046350 uo K1C IIAkt) R. MONVIS Slept 9-11. These steps arc similar to steps J-6. However, Ihe person-years for each person arc noi distributed diagonally, as in Table 2. but horizontally according to age at entry into the cohort. Steps 12-14. These steps arc similar to steps O il. The person years arc distributed horizontally according lo.tror of entry into the cohort. Modifications of Rixaiiyl Survival The numbers of cause-specific deaths expected in a cohort can be obtained if cause-specific mortality rates are substituted for total mortality rates in step I. This technique has been used by others to determine whether excess mortality due to cancer has occurred in an industrial population (//. Steps 1-14 are followed, with the only modification being in step 8. Persons dying from causes other than that being investigated arc treated as lise withdrawals in the year of their doth. Thus, the total person-years of survival by the cohort is unchanged. However, only persons dying from the cause of interest are included in the observed deaths. The relative survival therefore is cause-specific. If incidence rates are substituted for death rates in step I. the expected number of incident cases of disease is obtained. Currently, the incidence rates of the Connecticut Cancer Registry from 1925-1962 are available (IJ). Data Analysis Using Proporiion.\l Mortality For each member of the study group, the following information is needed: sex. race, ace. year and cause of death. The steps in Ihe computer program follow. Step /. The proportional mortality data matrix is read. Step 2. The age-year-of-dcath distribution of all deaths in the study population is determined in five-year groupings. Step J. The age-lime distribution of the observed dcjths is determined. Step 4. The agc-timc-specific expected deaths for each cause arc calculated bv multiplying the proportional mortality matrix times the age-time distribution of total deaths. Step 5. Proportional mortality ratios arc calculated by dividing the observed ratios by the expected ratios. Step 6. The Mantel-Haenszel A'1 is calculated as dc>cribcd.in step 7 above. This procedure has been used in a study of the cause of death among alcoholics (14). Sovic Technical Fi'au. his During one computer run. many causes of death may be ev aluated in one cohort, one cause of death may he evaluated in many cohorts, or both. Several options 04 __y Kl LAllVl SIKVIV *1 AM) I'KIII'OKlHISAl. SlltHtAlllY 331 exist as to the detail of calculation 3ruJ output desired. When a single cohort is .screened against all causes of death, the cohort data is read once and the pcrson*yeur* of survival are determined. Then each death rate maim i> multiplied by the personyears m.iiriv to obtain the espcctcd numbers of cause-specific deaths. Also, the observed deaths according to each cause are counted. When a single cause, usually all deaths, is screened agjiitst many cohorts, only the relative survival table is calculate I and primed. The program is written in Fortran IV and currently is in operation on the IBM 370/165 at the MlT-llarvard Computing Center. Most of the program has been compiled and is called as n subroutine. A short program (5-20 lines) must bessritten Tor each run to describe the input characteristics of the study cohort and any selections to be made. In addition, several control cards are needed to describe further characteristics of the cohort and to select the various options described a bos c. Comments The major ndsantage of using rates from a standard population is that no formal comparison group need be assembled. In the case of retrospective cohort studies extending back many years, this is difficult. Further, since the standard rates are based on a large number of deaths, the rates are relatively stable. The use of standard population rales has several disadvantages: characteristics of the study population which are related to death (or disease) may be different from those of the standard population The procedures described in this paper take account only of age, lime, sex and race differences. No account can be made, for example, of the cfTcct of cigarette smoking on differences in observed and expected rates. The efTccis of this problem may be studied in sescral ways. If available, a standard population which is relatively similar to the study population may be selected (J). An estimate of the maximum impact of confounding due to confounding variables can be made (/, IS). The best way to ev.ilu.nc suspected causal relationships is to study them in diverse groups using diverse standard populations, or, if possible, to conduct cohort studies using a formal comparison group. Another disadvantage arises if two or more study cohorts arc compared to the standard population. If relevant demographic characteristics differ between the two study groups, inferences comparing the two groups may not necessarily be marie by comparing their espcrience relative to the standard population (16. 17). For example, if relative survival is a function of ace, and if two study cohorts have diITcrcnt age distributions, age-adjusted relative survival rates must be calculated. A second possibility is to compare the two groups directly, by using one group as the standard population anil the other as the study population. This option is available in the program. The primary disadvantages of studies using proportional mortality is that disease rates are not obtained. A proportional mortality greater than one may result from a ^125072 0012 015 3 1 004952 ./ .132 R1CIIAIU) K. MUNSON high divcaNC-vpccilic dc:ith r:ite or :i low death rate from other causes. However, a proportional mortality study i> very economical in terms of time and money and may serve as a preliminary test of a causal hypothesis (J. 14}. If suggestive results arc obtained, a cohort study (retrospective or prospective) may be designed to prov ide a more definitive test of the hypothesis. REFERENCES /. Cam, R. A. M. Cohort analysis of mortality rates as an historical or narrative technique. Brit. J.Prcr. Sot. Sled. 10. 139(1956). 2. Em ere. F., Avtill. L. M,, a\o Cciue. S. i. The relative survival rate: a statistical method* ology. In "Nat. Cancer Institute Monogr. 61," p. tOI. 1961. J. Lt, F. P.. FeACviiM. 1. F.. Je.. Mantii. N.. and Milum. R. W. Cancer mortality among chemists. J. A or. Cancer Inn. 43. 1139 (1969). 4. Geovi. P.. D.. and llti/tL. A. M. Vital statistic) rates in live United Stales, 1940-1960. pp. 772-765. National Center for Vital Statistics, U.S. Dept. Ill W, PIIS. Washington. 1966. J. Vital Statistics of the United States. National Center for Health Statistics, Washington, 19231967. 6. Manual of the International Statistical Classification of Disease. Injuries and Causes of Death. WHO. Geneva, 1937. 7. Hiil.I. D. Computing man years at risk. Brit.J.Prcr.Soc. Med.26.132(1972). J. Monson, R. R . MacMaiios, 0.. and Austin, 3. H. When rr.ay endometrial cancer be con sidered cured? Conrrr 30,419 (1972). 9. Mamil^N.. anu lUiMiii. W. Statistical aspects of the anal>sis of data from retrospective studies of disease. J. Sat. Cancer hut. 22. 719 (1959). 10. Gateswoon, M. A report on the natural duration of cancer. Reports on Public Health and Medical Subjects. No. 33. Minivtry of Health. London, 1926. 11. Lca, A. J., and Cast. R. A. M. Muvtard gas poisoning, chronic bronchitis, and lung cancer. firit.J. Prer. Sec. Met. 9. 62 (1935). 12. Liovd, J. W,, and Ciocco. A. Long-term mortality study of steelworkers. J. Ore. Med. II, 299(1969). 1J. Etscsatao, M. Cancer in Connecticut, incidence and rates. 19)3-1962. Conn. State Dept. Health, Hartford, 1966. 14. Monson, R. R., Lyon, J. L., and MacMaiion, D. Causes of death among alcoholics. In preparation. 15. Rothman, K. and Monson, R. R. Survival in trigeminal neuralgia. J. Chronic DIj. 26, 303 (1973). 16. Kahn, H. A. The Dorn study of smoking and mortality among U S. veterans: report on eight and one-half years of observation. In "Epidemiological approaches to the study of cancer and other chronic diseases" (W. Ilarnsrcl. FJ.). Sal. Canerr lu. Mount' 19,1 (1966). 17. MitniNtv. O. S. Standardisation of risk ratios. Am. J. Lpiil. 96. 363 (1972). 1512507.. 001 2 \j I j h