Evaluation of a Population Roster as a Source of Population Controls: The Massachusetts Resident Lists

Transcription

1 American Jurnal f Epidemilgy Cpyright O 1999 by The Jhns Hpkins University Schl f Hygiene and Public Health All rights reserved Vl. 150, N. 4 Printed In USA. Evaluatin f a Ppulatin Rster as a Surce f Ppulatin Cntrls: The Massachusetts Resident Lists Kari Bhlke, 1 ' 2 Bernard L. Harlw, 2 Daniel W. Cramer, 2 Dnna Spiegelman, 1 and Nancy E. Mueller 1 Published ppulatin rsters can serve as a cnvenient surce f ppulatin cntrls. The authrs evaluated ne such rster, the Massachusetts Resident Lists, by estimating the cmpleteness f the Lists and by describing the differences between persns included and nt included n the Lists. The subjects were cases frm three case-cntrl studies f varian cancer cnducted in eastern Massachusetts between 1978 and Fr each f the three case series, mre than 90% f the cases were lcated n the Resident Lists. Age was ne f the primary differences t emerge between cases included and nt included; in the mst recent case series, cases yunger than age 40 years were less likely than lder cases t be included n the Lists. Am J Epidemil 1999;150: case-cntrl studies; epidemilgic methds; registries Surces f ppulatin cntrls fr case-cntrl studies include randm digit dialing, US Health Care Financing Administratin lists, driver's license lists, and, where available, resident lists by twn r cuntry. The cmpleteness f each f these surces will vary and is likely t influence the extent t which each adequately represents the surce ppulatin f the cases. Therefre, when a given surce f ppulatin cntrls is used, it is imprtant t understand bth the extent and the nature f its ppulatin cverage. Thse persns wh are included n a particular rster and are available fr selectin as cntrls may differ in imprtant ways frm thse wh are nt. A related pint t cnsider when selecting a ppulatin rster frm which t draw cntrls is the need t exclude cases wh are nt n the rster (1). Fr these cases, the rster des nt represent the crrect surce ppulatin. If the rster is nt reasnably cmplete, the number f cases that will need t be excluded may prduce a ntable lss f sample size and pwer. On the ther hand, failing t exclude these cases may result in selectin bias if they differ regarding characteristics predictive f the utcme f interest. Received fr publicatin July 9, 1998, and accepted fr publicatin Nvember 20, Abbreviatin: Cl, cnfidence interval. 1 Department f Epidemilgy, Harvard Schl f Public Health, Bstn, MA. 2 Obstetrics and Gyneclgy Epidemilgy Center, Department f Obstetrics and Gyneclgy, Brigham and Wmen's Hspital, Harvard Medical Schl, Bstn, MA. Reprint requests t Dr. Kari Bhlke, Grup Health Cperative, Center fr Health Studies, 1730 Minr Avenue, Suite 1600, Seattle, WA In the present study, we examined the ppulatin cverage f a unique ppulatin rster, the Massachusetts Resident Lists. T d s, we assessed case inclusin n the Lists fr three separate casecntrl studies (2-4) f varian cancer cnducted in eastern Massachusetts between 1978 and A questin f interest was whether the prprtin f the ppulatin included n the Resident Lists, as estimated by case inclusin n the Lists, changed ver time. In 1971, an eastern Massachusetts case-cntrl study f cancer f the lwer urinary tract reprted that 96 percent f cases diagnsed in 1967 were included n the Resident Lists (5). Whether cverage had remained this high, and whether it was high fr cases with cancers at ther sites, was nt knwn. In additin t estimating the prprtin f the ppulatin represented n the Resident Lists, we identified demgraphic, reprductive, and ther differences between cases included and nt included n the Lists. We then cnsidered the implicatins f case-cntrl analyses that failed t exclude cases wh were nt n the Resident Lists when cntrls were selected frm the Lists. MATERIALS AND METHODS The Massachusetts Resident Lists are annually published ppulatin lists, by twn, f all Massachusetts residents aged 17 years r lder. Inclusin n the Resident Lists is mandated by state law (Massachusetts General Laws (Chapter 51, sectins 4, 6)) and is based n respnse t a mailing r visit by the twn registrar. The infrmatin n the Resident Lists 354

2 Cntrl Selectin and the Massachusetts Resident Lists 355 varies frm twn t twn but at a minimum includes name and street address. In mst twns, date f birth is prvided. Many but nt all twns include infrmatin abut ccupatin. The case-cntrl studies that generated the three case series included nly English-speaking female residents. Cases in the first tw studies were identified frm 10 majr hspitals in the Bstn area, and cases in the mst recent study were identified thrugh the Massachusetts Cancer Registry. The 215 cases frm the first study (2) were White, aged years, and diagnsed with epithelial varian cancer between Nvember 1978 and September The 235 cases frm the secnd study (3) were als White but were aged years and diagnsed between July 1984 and September The 407 cases frm the third study (4) were aged years and diagnsed between January 1992 and February 1996; 95 percent were White. Subject enrllment fr the third study cntinued beynd February 1996, but these subjects were nt included in the present analysis. T determine whether a case was included n the Resident List f a particular twn, we reviewed all Lists published within 1 year f the date f the case interview (i.e., the year f the interview, the year befre the interview, and the year after the interview) until the case was fund r until the third List was checked, whichever ccurred first. In a few instances, when a twn was t far away t visit easily and Bstn-area library cllectins f Resident Lists were incmplete, we used Lists frm nly 1 r 2 f the 3 pssible years. We were able t btain sme but nt all f the missing Lists frm twns, and cases were classified by using the Lists available. Cases were lcated n the basis f their street addresses. If a case was nt fund at the given address, and if the List fr that twn included an alphabetic listing f residents in additin t the primary listing by street address, in sme instances we were able t identify cases wh had mved t a different address in the same twn. We accepted a persn at a different address as the case f interest if bth the name and date f birth matched. T assess differences between subjects included and nt included n the Resident Lists, we cnsidered thse variables cmmn t all three studies: age, place f birth, educatin, religin, marital status, parity, ral cntraceptive use, smking, bdy mass index, and family histry f breast r varian cancer in a firstdegree relative. The assciatin between each f these variables and inclusin n the Lists was assessed with lgistic regressin by using the SAS statistical sftware package (6). We calculated bth age- and multivariate-adjusted dds ratis and 95 percent cnfidence intervals fr the prbability f being included n the Resident Lists. Tests fr hetergeneity amng the studies were cnducted by using the likelihd rati test. T evaluate the imprtance f excluding cases nt included n the Resident Lists when selecting cntrls frm the Resident Lists, we cnsidered data frm each f the three studies separately. By using a mdel cntaining several risk factrs fr varian cancer, such as parity, ral cntraceptive use, talc use, and family histry f breast r varian cancer, we calculated effect estimates befre and after exclusin f cases wh were nt n the Lists. The specific type f mdel used was cnsistent with the riginal analyses frm each study; data frm the first study were analyzed by using cnditinal lgistic regressin, whereas data frm the secnd tw studies were analyzed by using uncnditinal lgistic regressin. The first tw studies drew cntrls exclusively frm the Resident Lists, and the third study selected cntrls by randm digit dialing fr cases yunger than age 60 years and frm the Resident Lists fr cases aged 60 years r lder. Therefre, fr the third study, we restricted ur analysis t the 115 cases aged 60 years r lder and the 73 Resident List cntrls. RESULTS Five cases frm the first series and ne frm the secnd were excluded because f missing address infrmatin. A ttal f 21 subjects were excluded because f address prblems, including street names nt fund n the Resident List f the indicated twn and pst ffice bxes. The numbers and percentages f subjects with such address prblems, frm case series 1, 2, and 3, were 7 (3.3 percent), 4 (1.7 percent), and 10 (2.5 percent), respectively. Finally, an additinal three subjects, ne frm the first case series and tw frm the secnd, were excluded because n Resident Lists frm the necessary years were readily available. After these subjects were excluded, the respective numbers f subjects frm the three case series available fr analysis were 202, 228, and 397. The prprtins f subjects frm each case series fund n the Resident Lists, verall and by the level f each predictr cnsidered, are shwn in table 1. Fr each f the three case series, mre than 90 percent f the subjects were lcated n these Lists. Of thse subjects nt, 5 frm the first case series, 2 frm the secnd, and 11 frm the third were classified n the basis f nly ne r tw available Lists (data nt shwn). If all f these subjects were lcated n the remaining unchecked List(s), the maximum percentages f cases that culd be included n the Lists were 94.6, 93.0, and 93.0 fr case series 1, 2, and 3, respectively. Am J Epidemil Vl. 150, N. 4, 1999

3 TABLE 1. Differences In selected variables between varian cancer cases and nt n the Massachusetts Resident Lists, and age-adjusted dds ratis () and 95% cnfidence Intervals (CI) fr the likelihd f being, * Variable Case series 1: (n = 202) %nt Case series 2: (n = 228) %nt Case series 3: (n = 397) %nt Ttal Age (years) < , , , , , , Place f birth Massachusetts Elsewhere , , , 0.90 Educatin SHIgh schl >High schl , , , 3.99 Religin Cathlic Jewish Prtestant Other , , , , , , , , 1.51 Ever married N , , , 3.91 Parity Nulliparus Parus , , , N Oral cntraceptives, ever use N Smking Never Ever Bdy mass Index S23 >23 Family histry f breast r varian cancer N * Sme percentages may nt ttal 100% because f runding. 0.20, , , , , , , , , , , , 2.15

4 Cntrl Selectin and the Massachusetts Resident Lists 357 Of the variables cnsidered in table 1, age was assciated with inclusin n the Resident Lists in each f the three case series and was statistically significant in case series 3. In case series 3, cases aged years were nearly fur times mre likely t be included n the Lists than were cases yunger than age 40 years (95 percent cnfidence interval (CI): 1.87, 8.48). Cases aged 60 years r lder were 4.6 times mre likely t be included than were cases yunger than age 40 years (95 percent CI: 1.80, 11.51). Place f birth was als significant in case series 3. Cases brn utside f Massachusetts were less likely t be included n the Resident Lists than were cases bm in Massachusetts (dds rati = 0.45, 95 percent CI: 0.23, 0.90). The effect f marital status appeared t differ between the earliest and the mst recent case series. Ever-married cases in case series 1 were less likely t be included n the Resident Lists; in case series 3, they were mre likely t be included. The test fr hetergeneity between the tw studies was statistically significant (p < 0.05). In all three case series, parus wmen were mre likely than nulliparus wmen t be included n the Resident Lists, althugh these assciatins were nt significant. There was sme suggestin in case series 3 that smkers and ever users f ral cntraceptives were less likely t be included, but these assciatins were nt apparent in case series 1 and 2. In a multivariate analysis f all the variables in table 1, age and place f birth remained significant in case series 3, and educatinal level became significant. Thse persns with mre than a high schl educatin were 2.4 times mre likely t be included n the Resident Lists than were thse with less educatin (95 percent CI: 1.09, 5.38) (data nt shwn). The effect f failing t exclude frm analyses thse cases nt included n the Resident Lists is shwn in table 2. Results fr age are nt given because age was a matching factr in the first study. In each study, the effect estimates frm the mdel that included all cases were generally similar t thse frm the mdel that included nly thse cases n the Resident Lists. u "5 j a c a I is j i n s N OJ CJ O d T^ d d T^ CD CO CO d ^ c <> " ^ CJ" CJ CJ T-; d d T^ *-' CJ i- a> i^ in eg i- m Q cj in p d d 1 cs? d*- S c 0) CJ i- * > a> c\4 ^ c d d ais n s <; d d ^ cj CJ" d 12?5S d T- cvj c CO" T-" r-" 1-" r;no 00 d d T^ d t- i-?85 3 d T^ c\i c ddr: 3 d ills fill DISCUSSION The present study fund a high prprtin f varian cancer cases included n the Massachusetts Resident Lists, suggesting fairly high ppulatin cverage by these Lists f wmen in this age and ethnic grup. Furthermre, it is likely that what we bserved was an underestimate f the true cverage, since we were nt able t check all three ptential Lists fr all subjects. The extent f ppulatin cverage appeared t remain fairly cnstant acrss the three time perids studied n s n OJ TJ- cp c «5 d d T^ ^ i I e s 8 H> Isc C CO l list a S c n- * a> Am J Epidemil Vl. 150, N. 4, 1999

5 358 Bhlke et al. Restrictin f the study ppulatin t varian cancer cases was bth a strength and a limitatin f the present analysis. Selecting a similar ppulatin frm each f the three time perids was necessary fr inferences regarding changes ver time in ppulatin cverage by the Resident Lists. Hwever, the generalizability f these results may be limited. When we evaluated differences between cases included and nt included n the Resident Lists, significant differences emerged mre frequently in case series 3 than in case series 1 r 2. This finding may be due t the larger sample size f case series 3, but it may als reflect changes in the nature f the cverage ver time. Given the Lists' high verall cverage, it may nt be surprising that failing t exclude cases wh were nt n the Resident Lists did nt ntably affect the results f case-cntrl analyses f the three studies. The lack f an effect may be specific t the particular variables in the mdel and may nt be generalizable t studies f ther cancers. The relatively lw cverage f wmen yunger than age 40 years in case series 3 may be a cause fr cncern in case-cntrl studies f yung wmen. The Resident Lists may nt adequately represent the surce ppulatin f these cases. This differential cverage by age is cnsistent with a previus study f ppulatin cverage by the 1980 US Decennial Census (7). That study nted that yunger residents were less likely t be captured by the Census, as were members f several racial r ethnic minrities, the unmarried, and thse persns with a lw level f incme r educatin. We were unable t assess the basis fr the differential cverage by age bserved in ur study, but higher residential mbility amng the yung is ne ptential explanatin. Practical cnsideratins related t use f the Lists include accessibility and the nature f the infrmatin they prvide. The Resident Lists are public recrds and therefre are readily accessible. The Lists prvide residential address rather than mailing address, and persns wh have pst ffice bxes may be difficult t cntact by mail. Hwever, in the present analysis, relatively few f the subjects had pst ffice bxes. The 1971 case-cntrl study cited previusly (5) is nt the nly ne t cnsider case inclusin in ppulatin rsters. In a case-cntrl study f bladder cancer that selected cntrls yunger than age 65 years frm Iwa driver's license lists, the investigatrs nted that an analysis restricting cases yunger than age 65 years t Iwa driver's license hlders prduced findings similar t thse frm the verall study (8). Of the cases in the study wh were aged years, 92 percent f the wmen and 97 percent f the men reprted hlding Iwa driver's licenses (9). Furthermre, a cmparisn f the Iwa driver's license lists with the US Decennial Census suggested that the lists adequately represented residents aged years (9). Hwever, as nted previusly, the US Decennial Census is nt a perfect gld standard. In a large, multicenter case-cntrl study f cln cancer that enrlled cntrls frm a number f surces, Slattery et al. (10) assessed case inclusin n Health Care Financing Administratin lists and driver's license lists. Cverage f varius case grups by the Health Care Financing Administratin was high, ranging frm 94.4 t 96.6 percent. Driver's license lists included between 89.7 and 99 percent f cases, depending n age, sex, and study site. T cnclude, the Massachusetts Resident Lists appear reasnably cmplete fr the segment f the ppulatin evaluated. It wuld be interesting fr investigatrs wh are studying ther types f cases, particularly thse cases wh are male r nn-white, t cnduct a similar analysis t expand ur understanding f this and ther registries. ACKNOWLEDGMENTS Dr. Bhlke was supprted by a training award in cancer epidemilgy frm the Natinal Institutes f Health, Natinal Cancer Institute, Bethesda, Maryland (T32 CA ). The authrs thank Melissa Jhnsn fr assistance with data cllectin. REFERENCES 1. Wachlder S, Silverman DT, McLaughlin JK, et al. Selectin f cntrls in case-cntrl studies, n. Types f cntrls. Am J Epidemil 1992;135: Cramer DW, Hutchisn GB, Welch WR, et al. Determinants f varian cancer risk. I. Reprductive experiences and family histry. J Nad Cancer Inst 1983;71: Cramer DW, Harlw BL, Willed: WC, et al. Galactse cnsumptin and metablism in relatin t the risk f varian cancer. Lancet 1989;2: Cramer DW, Harlw BL, Titus-Emstff L, et al. Over-thecunter analgesics and risk f varian cancer. Lancet 1998; 351: Cle P, Mnsn RR, Haning H, et al. Smking and cancer f the lwer urinary tract. N Engl J Med 1971;284: SAS Institute, Inc. SAS/STAT user's guide, release 6.12 ed. Carey, NC: SAS Institute, Inc, Fein DJ. Racial and ethnic differences in U.S. census missin rates. Demgraphy 1990,27: Cantr KP, Lynch CF, Jhnsn D. Bladder cancer, parity, and age at first birth. Cancer Causes Cntrl 1992;3: Lynch CF, Lgsden-Sackett N, Edwards SL, et al. The driver's license list as a ppulatin-based sampling frame in Iwa. Am J Public Health 1994;84: Slattery ML, Edwards SL, Caan BJ, et al. Respnse rates amng cntrl subjects in case-cntrls studies. Ann Epidemil 1995;5: Am J Epidemil Vl. 150, N. 4, 1999