A case-control study of lung cancer screening in Okayama ...

Lung Cancer 34 (2001) 325–332www.elsevier.com/locate/lungcanA case-control study of lung cancer screening in OkayamaPrefecture, JapanKenji Nishii a, *, Hiroshi Ueoka b , Katsuyuki Kiura b , Tsuyoshi Kodani a ,Masahiro Tabata b , Takuo Shibayama b , Kenichi Gemba a , Takushi Kitajima b ,Akio Hiraki b , Masashi Kawaraya a , Tomio Nakayama c , Mine Harada aaDepartment of Respiratory Medicine, Okayama Institute of Health and Preention, 408-1 Hirata, Okayama 700-0952, JapanbDepartment of Internal Medicine II, Okayama Uniersity Medical School, Okayama, JapancDiision of Epidemiology, Department of Field Research, Osaka Medical Center for Cancer and Cardioascular Diseases,Osaka, JapanReceived 8 March 2001; received in revised form 15 June 2001; accepted 18 June 2001AbstractThe effectiveness of lung cancer screening in reducing mortality still remains uncertain. In order to evaluate theefficacy of lung cancer screening, a case-control study was conducted in Okayama Prefecture, Japan. The study areaconsisted of 34 municipalities where a population-based lung cancer screening had been conducted. Chest X-rayexaminations for all participants and sputum cytology for high-risk participants were offered annually. The casesanalyzed in this study consisted of 412 individuals aged between 40 and 79 who died of lung cancer. A total of 3490controls, two to ten for each case matched by gender, year of birth, and living district were randomly collected.Screening histories of cases were compared with those of and matched controls for the identical calendar period prioto diagnosis of the case. Smoking adjusted odds ratio (OR) of death from lung cancer for screened individuals versusunscreened, within 12 months before diagnosis, was calculated as 0.59 (95% confidence interval: 0.46–0.74;P=0.0001). The OR for women (0.39, 95% confidence interval: 0.24–0.64) was lower than that for men (0.67, 95%confidence interval: 0.51–0.87), although both were statistically significant. These results suggest that lung cancerscreening contributes to reducing lung cancer mortality by 41%. © 2001 Elsevier Science Ireland Ltd. All rightsreserved.Keywords: Case-control study; Chest X-ray examination; Lung cancer mortality; Lung cancer screening; Odds ratio; Sputumcytology1. Introduction* Corresponding author. Tel.: +81-86-224-3258; fax: +81-86-224-3250.E-mail address: nkenji@lime.ocn.ne.jp (K. Nishii).The ideal setting to assess a cancer screeningprogram is a randomized controlled trial [1]. Forevaluation of lung cancer screening, four randomizedcontrolled trials in the 1970s–1980s failed to0169-5002/01/$ - see front matter © 2001 Elsevier Science Ireland Ltd. All rights reserved.PII: S0169-5002(01)00270-7

326K. Nishii et al. / Lung Cancer 34 (2001) 325–332demonstrate that screening reduced mortalityfrom lung cancer [2–5]. Therefore the AmericanCancer Society decided not to recommend anytest for early detection of lung cancer [6]. Furthermore,in a recent article analyzing the long-termfollow-up results of the Mayo Clinic trial, therewas no significant reduction of mortality in thegroup receiving lung cancer screening [7]. However,the results do not reflect recent advances inthe diagnostic and therapeutic modalities of lungcancer. The Memorial Sloan–Kettering andJohns–Hopkins Lung Project compared a combinationof annual chest X-ray and sputum cytologywith chest X-ray only. The Mayo LungProject and the Czechoslovak Study comparedfrequent screening with sporadic screening.Thus none of the four studies evaluated the efficacyof annual lung cancer screening in comparisonwith no screening [8]. Now, a large-scalemultiphase screening trial (PLCO Cancer ScreeningTrial) sponsored by the National Cancer Instituteis going on in the US, but the results of thestudy are not expected to be published before2010.A case-control study is another method to evaluatecancer screening [9,10]. In fact, this approachhas been applied in the evaluation of screening forcervical cancer, breast cancer, gastric cancer, andcolorectal cancer [11–14]. Screening for lung cancerhas already been evaluated using case-controlstudies in East Berlin [15] and in Japan [16,17].However, since the odds ratios of death from lungcancer for screened individuals versus unscreenedin these studies were inconsistent (0.54–1.1), conclusiveresults have not yet been obtained.In Okayama Prefecture, Japan, a lung cancerscreening program was conducted from 1976through 1986 as a pilot study [18], which showedthe possible efficacy of mass screening for lungcancer. The screening program was supported bythe national government under the Health andMedical Services Law for the Aged since 1987,and a total of 2 323 455 individuals were screenedbetween 1987 and 1998 in Okayama Prefecture.Nine hundred and sixty cases with lungcancer were detected and 65.8% of them couldbe surgically resected. In order to evaluate theefficacy of lung cancer screening in OkayamaPrefecture, Japan, we conducted a case-controlstudy.2. Methods2.1. Mass screening for lung cancer in the studyareaThe study area consisted of 34 cities, villages,and towns in Okayama Prefecture where we havebeen conducting lung cancer screening and couldutilize the computerized records of people whowere invited to the screening. This study wasconducted between 1991 and 1996. The beginningof the study period varied according to the yearswhen the records were computerized in each area:four municipalities in 1991, 12 in 1992, 13 in 1993,three in 1994, and one each at 1995 and 1996.The Tuberculosis Control Law required allJapanese citizens aged 16 or over to have anannual chest X-ray examination since the 1950s.Lung cancer screening programs have been addedto the system since 1987, for individuals aged 40or over, under the Health and Medical ServicesLaw for the Aged. People who were invited to thescreening were all National Health Insuranceholders or family members of Employment-relatedHealth Insurance, because the EmploymentrelatedHealth Insurance holders themselves wereassigned to be screened in their companies undera different system.An annual chest X-ray examination was conductedfor all screenees using 100×100 mmminiature photofluorography with tube voltage of140 kV. Smoking habits of all screenees wererecorded. Sputum cytology with Saccomanno’s3-day pooled method was performed for the individualsin the high-risk group. The high-riskgroup was defined as individuals aged 50 or overwith a smoking index (average number ofcigarettes smoked per day multiplied by the numberof years smoked) of at least 600 and individualsaged 40 or over with a history of bloodysputum. For the individuals suspected to havelung cancer by chest X-ray or sputum cytology,further examinations were performed to confirmthe diagnosis.

K. Nishii et al. / Lung Cancer 34 (2001) 325–332 3272.2. Source populationThe source population of this study was definedas people who were resident in the study area andwere invited to the screening at least once duringthe study period. Information about all the peoplewho were invited to the screening were preservedas computerized records. A total of 247 537 people(110 035 men and 137 502 women) were thesource population. Table 1 shows the distributionof age and gender in the population. Attendancerates in 1997 are also shown on Table 1. Attendancerates for men were lower than those forwomen and attendance rates were generally higherin elderly people.2.3. Case identificationThe cases analyzed in the present study weredefined as the residents in the study area who diedof lung cancer during the study period, at an agebetween 40 and 79. Data on diagnosis were obtainedfrom the hospital records and/or from theOkayama Cancer Registry. Data on smokinghabits were obtained from the Registry, hospitalrecords, or previous screening records. When anydiscrepancies concerning smoking index were observed,the highest number was used for theanalyses. Histologic types were classified accordingto the World Health Organization histologicclassification [19].There were 1154 deaths from lung cancer in the34 municipalities during the study period, accordingto the death certificates of all residents. Ofthose, 371 had been over 79 years old and five hadbeen under 40 years old at the time of death, andthey were excluded. Fifty-eight cases had not undergonescreening during the study period andwere excluded. Data on the diagnosis of the remaining720 cases were examined at the OkayamaCancer Registry. Lung cancer had been diagnosedbefore the beginning of the study period in 223and they were excluded. Since 72 cases had beendiagnosed and had died in other prefectures,where the Okayama Cancer Registry was unableto collect appropriate information, these caseswere excluded. Eight cases were excluded becausethe dates of diagnosis were unable to be confirmeddue to the physician’s refusal or loss ofmedical records. Five cases were excluded becausethe true diagnosis was metastatic lung tumor. Nocases had moved in from another municipalityduring the study period. The remaining 412 individualswere finally enrolled in the present studyas the appropriate cases.2.4. Control identificationControl candidacy was limited to individualswho had been living in the same area as thecorresponding case from the beginning of thestudy period until the diagnosis of the case, andTable 1Distribution of age and sex in the study population and attendance rate of screening in 1997Number of study population Screened in 1997Attendance rate in 1997 (%)Male Female MaleFemale Male Female40–44 11 90814 6078804318 7.429.645–49 15 090 18 091 13846190 9.2 34.250–54 15 283 18 686 148862399.733.455–59 13 287 17 069 16466617 12.438.860–64 14 331 17 6354252 9065 29.7 51.465–69 16 144 19 2886316 10 333 39.1 53.670–74 14 57617 6335844839540.147.675–79 941614 493 3305 5268 35.1 36.380– 14 110 23 7852829390020.016.4Total 110 035137 502 25 11556 425

328K. Nishii et al. / Lung Cancer 34 (2001) 325–332had undergone screening. Six to ten individualsfor each case, who were matched by gender, yearof birth (within 2 years), and living district, wereselected as controls from the computerizedrecords. Then, a total of 3756 people (2996 menand 760 women) were selected as controlcandidates.Of those, 17 were excluded because of movinginto the district after the diagnosis of the correspondingcase, and another 17 were excludedbecause they had died or migrated to other areasbefore the case was diagnosed. For the remaining3722 individuals (2966 men and 756 women),smoking habits were surveyed using previousscreening records or mailed questionnaires. Whenthere was no response to a mailed questionnaire, apublic health nurse conducted an interviewby telephone or home visit. If the control candidatewas deceased or ill, a family member wasinterviewed. Regardless of the interview method,informed consent was obtained before the interviewwith promised nondisclosure of private information.Information on smoking habits wasobtained for 3490 individuals (93.7%). The smokingindex of the control was calculated for theperiod up to the diagnosis of the correspondingcase.Finally, 412 cases and 3490 controls were selectedfor the analysis. The case-control sets includedone set with two controls, six sets withfour controls, ten sets with five controls, 46 setswith six controls, 35 sets with seven controls, 64sets with eight controls, 119 sets with nine controls,and 131 sets with ten controls.2.5. Screening historyScreening histories of the cases and controlswere obtained from the lists of screenees.Screening histories were traced back for the identicalcalendar period 36 months before the diagnosisof lung cancer in the case. The screening,which led to the diagnosis of lung cancer, wasincluded in the screening history. Since we had noinformation about the chances of taking chestX-rays outside the lung cancer screening, onlyscreening history in the mass screening wascounted.2.6. AnalysesOdds ratios of death from lung cancer for thescreened individuals versus unscreened were analyzed.Conditional logistic regression models wereused to estimate odds ratios and their 95% confidenceintervals (95% CIs). Smoking adjusted oddsratios were calculated including eight variables(1–399, 400–599, 600–799, 800–999, 1000–1199,1200–1399, 1400–1599, 1600-) on smoking index.In order to evaluate the effect of the screeninginterval, the main exposure variable was determinedto be a visit in the last 12 months. Statisticalsignificance was assessed at the 5% level, and a95% CI for odds ratio was presented. ThePHREG Procedure of the SAS program was usedfor this analysis.3. ResultsTable 2 shows the characteristics of the cases.Three-quarters of the total cases were men andalmost 90% of the cases were 60 years old or over.The appearance of symptoms was the most frequentcause of detection for deceased cases. Adenocarcinomawas the most common histologictype in both men and women, although the histologictype could not be confirmed in 24.3% of menand 27.5% of women.Table 3 shows the distribution of smoking indexfor cases and controls. Smoking indexes ofthe male cases were significantly higher (P=0.001) than those of the controls.These results are similar to those in the clinicbasedcase-control study in Japan [16].Table 4 shows the odds ratios of death fromlung cancer within 12 months before diagnosis.Since there was deviation of the smoking indexbetween cases and controls, the odds ratios wereadjusted by smoking indexes with a logistic regressionmodel. The smoking-adjusted odds ratioof death from lung cancer within 12 months beforediagnosis for those screened versus unscreenedwas 0.59 (95% CI: 0.46–0.74), whichdemonstrated a statistical significance (P=0.0001). The odds ratio for women (0.39, 95% CI:0.24–0.64) was lower than that for men (0.67,

K. Nishii et al. / Lung Cancer 34 (2001) 325–332 329Table 2Characteristics of the casesMale (%) Female (%)Age40–44 3 (0.9) 2 (2.2)45–4910 (3.1) 3(3.3)50–54 5(1.6) 4(4.4)55–5916 (5.0) 5(5.5)60–64 36 (11.2)11(12.1)65–6986 (26.8) 19(20.9)70–74 85(26.5) 27 (29.7)75–79 80(24.9) 20 (22.0)Cause of detectionScreening 26(8.1) 7 (7.7)Other health checkup2 (0.6) 0(0.0)Symptom 93(29.0) 32(35.2)Other disease44 (13.7) 4(4.4)Other reason 1 (0.3)2 (2.2)UnknownHistologic typeSquamous cell carcinoma15587(48.3)(27.1)465(50.6)(5.5)Adenocarcinoma105 (32.2) 47(51.7)Small cell carcinoma 43(13.4) 9 (9.9)Large cell carcinoma7 (2.2) 5(5.5)Other 1(0.3) 0(0.0)Unknown78 (24.3) 25(27.5)Total 3219195% CI: 0.51–0.87), although both had statisticalsignificance.Since the control candidates who did not respondto the smoking survey had a low attendancerate of screening (3.8%), probably due to alow level of concern with health, exclusion ofthese from the analysis might lead to overestimationof the efficacy of the screening. Thereforesmoking index derived from the median of othercontrols (660 for men, 0 for women) were assignedto the 232 nonrespondents, and analyseswere performed for a total of 3722 controls (Table5). The smoking-adjusted odds ratio of deathfrom lung cancer, 0.64 (95% CI: 0.51–0.81, P=0.0002), was somewhat closer to 1.0, but still hadstatistical significance.4. DiscussionIn a population based case-control study oflung cancer screening conducted in 50 municipalitiesin Japan, Sobue et al. suggested that annualscreening could reduce the risk of death from lungcancer by a maximum of 28% (odds ratio 0.72,95% CI: 0.50–1.03) [16]. However, their resultshad only marginal significance (P=0.07). On theother hand, in this study, we demonstrated thestatistically significant efficacy of populationbasedannual screening to reduce the risk of deathfrom lung cancer.The odds ratio in this study was lower than thatin Sobue’s report [16], which may be explained inpart by improvement in the sensitivity of chestX-ray examinations. In Sobue’s report, somechest X-ray examinations were taken with lowvoltageexposure (90 kV tube voltage), whereasonly high-voltage exposure (140 kV) were used inthe present study. Although the calculating methodswere different, the sensitivity of chest X-raywith high-voltage exposure was reported as 70%(Nagasaki Study [20], conducted in 1987–1989),while that in a previous study including lowvoltageexposure, was 56.8% (Osaka Study [21],

330K. Nishii et al. / Lung Cancer 34 (2001) 325–332conducted in 1981–1985). This difference in sensitivitymay have affected the odds ratio.The efficacy to reduce the risk of death wasprominent in women compared with that in men,which was consistent with the results previouslyreported by Sobue et al [16]. These results mayreflect that a majority of lung cancer in women isperipheral type adenocarcinoma, which are easilydetectable by chest X-ray, when good qualitycontrol was achieved. In the present study, theincidence of screening as a cause of detection wasmarkedly low, which may suggest the favorableresults for screening (low risk of death in thepatients detected by mass screening).There are several methodological issues involvedin this study, which need to be discussed[22]. First, the time of beginning of the studydiffered from place to place due to variation in theyears when records of the screening were availablein each area. However, since we rigorously selectedthe matched controls in each locality, theyhad exactly equal opportunities for screening.Second, since we excluded case candidateswhose diagnosis preceded the beginning of thestudy period, cases were restricted to those with arelatively short interval between diagnosis anddeath. For that reason, our evaluation of screeningefficacy might be overestimated [23].Third, several investigators recommended thatthe screening visits with symptoms should not becounted as screened [24]. In the present study,although screenees had been asked at the screeningvisit whether he/she had a history of bloodysputum or not, the information had not beenappropriately preserved. Other information onsymptoms was also unavailable. Therefore, all ofthe screening visits were counted as screened inthis study, and the results might beunderestimated.Fourth, different techniques were used to obtainthe smoking histories of the cases and controls.For the cases, data were collected from theregistry, hospital records, or previous screeningrecords, while they were obtained for the controlsfrom previous screening records or interviews(mail, telephone or home visit), which may havecaused some bias in the results. However, thequestionnaires for both cases and controls werethe same, consisting of smoking status, the age ofstarting smoking, number of cigarettes smoked ina day, and the age of stopping smoking. Furthermore,it was reported previously in Japan that thedifference in the method used to obtain smokinghistories did not significantly cause bias in theresults [25]. The Japanese prevalences of smokingin 1998 were 50.8% in male and 10.9% in femaleaccording to the homepage of the Japanese Ministryof Health, Labour and Welfare. The prevalencesin the present study were lower in malesand higher in females, which may be attributableto exclusion of young adults (less than 40 yearsold) in the present study.Table 3Smoking index of cases and controlsSmoking indexCasesControlsMale a (%) Female (%) Male a (%) Female (%)045 (14.0) 81(89.0) 708 (25.4) 681 (97.2)1–399 8 (2.5) 2 (2.2) 237 (8.5) 7 (1.0)400–599 37 (11.5) 2 (2.2) 320 (11.5) 6(0.9)600–799 31 (9.7) 2 (2.2) 298 (10.7) 2(0.3)800–999 59 (18.4) 0 (0.0) 413 (14.8) 2 (0.3)1000–1199 80 (24.9) 2 (2.2) 408 (14.6) 1 (0.1)1200–1399 27 (8.4) 1(1.1) 175 (6.3) 0(0.0)1400–1599 9 (2.8) 1 (1.1) 69 (2.5) 0 (0.0)1600- 25 (7.8) 0(0.0) 161 (5.8) 2(0.3)Total321 (100.0) 91(100.0) 2789 (100.0) 701 (100.0)a Smoking index of male cases are significantly higher (P=0.001) than those of male controls.

K. Nishii et al. / Lung Cancer 34 (2001) 325–332 331Table 4Odds ratios of death from lung cancer for individuals screened within 12 monthsGenderNumber ofScreened (%)OR a95% CI bCasesControls Cases ControlsTotal 412 3490 32.3 43.9 0.59 0.46–0.74Male 3212789 32.4 42.1 0.67 0.51–0.87Female91 701 31.9 51.2 0.390.24–0.64a OR: Odds ratio.b95% CI: 95% confidence interval.Fifth, in this study, smoking history was notmatched in cases with controls. Since this mightlead to lower precision in case-control comparison,we used multivariate analysis with a smokingindex to compensate bias with respect to smoking.Furthermore, we also performed additional smoking-matchedanalyses by excluding smoking-unmatchedcontrols (eg excluding high-risk malecontrols for a non-high-risk male case, and so on).Of 3490 controls, 1603 were excluded becausesmoking habits were not matched, and 11 of 412cases were also excluded because they did nothave any appropriate controls. For 401 cases and1887 controls, odds ratio of the smoking-matchedanalysis was 0.55 (95% CI: 0.43–0.70). This resultrevealed that statistical significance was consistentregardless of the matching of smoking habits.The 232 nonrespondents to the smoking surveyof 3722 control candidates were excluded from themain analyses of this study. Since exclusion of thenonrespondents creates the possibility of overestimatingthe efficacy of screening, those 232 controlcandidates were included in another analysis byassigning their smoking index as the median ofthe controls responding to smoking survey (660for males, 0 for females). As shown in Table 5, theresults revealed that the odds ratio came a bitcloser to 1.0, but statistical significance was stillobtained.In conclusion, this case-control study suggestedthat annual lung cancer screening in OkayamaPrefecture, Japan, consisting of chest X-ray andsputum cytology, could reduce the lung cancermortality by 41% (95% confidence interval: 26–54%). The significant reduction of the risk forlung cancer death was observed both in males andin females.AcknowledgementsWe would like to thank Dr Suzuki of OsakaMedical Center for Cancer and CardiovascularDiseases, and Dr Kawamura of the OkayamaCancer Registry for data collection. We alsothank Ms Masakage and Ms Tanaka for theirassistance with collection and analysis of data.This study was supported by Grant-in-Aid forTable 5Odds ratios of death from lung cancer for individuals screened within 12 months: analysis including non-respondentsNumber ofScreened (%)OR b95% CI cCases Controls a CasesControlsTotal 412 372232.341.8 0.640.51–0.81Male 321 296632.1 40.0 0.71 0.55–0.93Female 91 756 31.9 48.8 0.450.28–0.74aAnalysis was performed assigning the median of smoking index (660 for men and 0 for women) to 232 nonresponding controlsto smoking survey.b OR: Odds ratio.c 95% CI: 95% confidence interval.

332K. Nishii et al. / Lung Cancer 34 (2001) 325–332Cancer Research from the Ministry of Health andWelfare, Japan (9–7).References[1] US Preventive Services Task Force. Methodology forreviewing evidence: Guide to clinical preventive services,2nd edn. Williams and Wilkins, Inc., Baltimore, 1996,47-55[2] Fontana RS, Sanderson DR, Woolner LB, et al. Screeningfor lung cancer: a critique of the Mayo Lung Project.Cancer 1991;67:1155–64.[3] Melamed MR, Flehinger BJ. Detection of lung cancer:Highlights of the Memorial Sloan–Kettering Study inNew York City. Schweits med Wschr 1987;117:1457–63.[4] Tockman MS. Survival and mortality from lung cancer inscreened population: The Johns Hopkins Study. Chest1986;89:324–5.[5] Kubik A, Parkin DM, Khlat M, et al. Lack of benefitfrom semi-annual screening for cancer of the lung: followupreport of a randomized controlled trial on a populationof high-risk males in Czechoslovakia. Int J Cancer1990;45:26–33.[6] American Cancer Society. Report on the cancer relatedhealth checkup: cancer of the lung, CA Cancer J. Clin. 30(1980) 199-207.[7] Marcus PM, Bergstralh EJ, Fagerstrom RM, et al. Lungcancer mortality in the Mayo Lung Project: impact ofextended follow-up. JNCI 2000;16:1308–16.[8] Strauss GM, Gleason RE, Sugarbaker DJ. Screening forlung cancer. Chest 1997;111:754–68.[9] Prorok PC, Chamberlain J, Day NE, et al. UICC workshopon the evaluation of screening programs for cancer.Int J Cancer 1984;34:1–4.[10] Sasco AJ, Day NE, Walter SD. Case-control studies forthe evaluation of screening. J Chron Dis 1986;39:399–405.[11] Clark EA, Anderson TW. Does screening by ‘PAP’smears help prevent cervical cancer? A case-control study.Lancet 1979;II:1–4.[12] Palli D, Roselli DTM, Buiatti E, et al. A case-controlstudy of the efficacy of a non-randomized breast cancerscreening program in Florence (Italy). Int J Cancer1986;38:501–4.[13] Oshima A, Hirata N, Ubukata T, et al. Evaluation of amass screening program for stomach cancer with a casecontrolstudy design. Int J Cancer 1986;38:829–33.[14] Selby JV, Friedman GD, Quesenberry CP, et al. A casecontrolstudy of screening sigmoidoscopy and mortalityfrom colorectal cancer. New Eng J Med 1992;326:653–7.[15] Berndt R, Nischan P, Ebeling K. Screening for lungcancer in the middle-aged. Int J Cancer 1990;45:229–30.[16] Sobue T, Suzuki T, Naruke T, et al. A case-control studyfor evaluating lung-cancer screening in Japan. Int J Cancer1992;50:230–7.[17] Okamoto N, Suzuki T, Hasegawa H, et al. Evaluation ofa clinic-based screening program for lung cancer with acase-control design in Kanagawa, Japan. Lung Cancer1999;25:77–85.[18] Nishii K. The effectiveness of mass screening for lungcancer. Okayama-Igakukai-Zashi 1990;102:603–12.[19] World Health Organization. Histological Typing of LungTumors, 2nd edn. World Health Organization, Geneva,1981.[20] Soda H, Tomita H, Kohno S, et al. Limitation of annualscreening chest radiography for the diagnosis of lungcancer: a retrospective study. Cancer 1993;72:2341–6.[21] Sobue T, Suzuki T, Matsuda Mm, et al. Sensitivity andspecificity of lung-cancer screening in Osaka, Japan. Jpn JCancer Res 1991;82:1069–76.[22] Cronin KA, Weed DL, Conner RJ, et al. Case-controlstudies of cancer screening: theory and practice. J NatlCancer Inst 1998;90:498–504.[23] Weiss NS. Application of the case-control method in theevaluation of screening. Epidemiol Rev 1994;16:102–8.[24] Weiss NS. Analysis of case-control studies of the efficacyof screening for cancer: how should we deal with testsdone in persons with symptoms? Am J Epidemiol1998;147:1099–102.[25] Shimizu H, Hisamichi S. Consistency between the answersfrom husband and wife on questionnaire surveys.Kousei-no-shihyou 1987;34:3–5.