Znt. J. Cancer: 50,230-237 (1992) 0 1992 Wiley-Liss, Inc.

%&

PdoIicalion of the lnlernationa J n on Aga nst Cancer PLD cat on oe I J n on lriternat onale Contre e Cancrr

A CASE-CONTROL STUDY FOR EVALUATING LUNG-CANCER SCREENING IN JAPAN Tomotaka SOBUE'.4,Takaichiro SUZUKI',Tsuguo NARUKE~and THEJAPANESE LUNG-CANCER-SCREENING RESEARCH GROUP 'Department of Field Research and 'Division of Cancer Epidemiology, Research Institute, Center for Adult Diseases, Osaka, Higashinan-ku, Osaka 537; and 'Division of Thoracic Surgery, National Cancer Center Hospital, 1-1 Tsukiji, 5-chome, Chuo-ku, Tokyo 104, Japan. In order to evaluate the efficacy of lung-cancer screening, a case-control study was conducted using the data from 50 areas where population-based lung-cancer screening programmes have been operated by local municipalities. In most areas, chest X-ray examinations for all participants and sputum cytology for high-risk participants were offered annually. Case series consisted of 273 deceased lung-cancer cases. For each case, 2 to 5 controls (a total of 1,269 controls) were collected from those who were alive at the time of diagnosis of the corresponding case, matched by sex, age, smoking status and type of health insurance. Cases and controls were limited to a high-risk group for males and a non-high-risk group for females. Screening histories, which were obtained from the list of screenees, were compared between case and matched controls for the identical calendar period before the time of diagnosis of the case. The odds ratio of dying from lung cancer for those screened within I 2 months vs. those not screened was 0.72 (95% confidence interval 0.50-1.03; p = 0.07). The odds ratios increased towards unity, as the length of time in which screening histories were compared increased. After adjusting for some other variables, which appeared to be associated with the opportunities of chest X-ray examination, the estimated odds ratio did not change. These results suggest some benefits from lung-cancer screening in terms of reduction of lung-cancer mortality and should be subject to further research.

The preferred approach to the evaluation of screening for cancer is a randomized controlled trial. So far, 3 trials in the US (Fontana et al., 1986; Melamed and Flehinger, 1987; Tockman et al., 1985) and one trial in Czechoslovakia (Kubik et al., 1990) have been conducted by this approach, to evaluate the efficacy of lung-cancer screening. None of these studies showed any evidence of benefit from lung-cancer screening in terms of mortality reduction from the disease. However, there has never been a satisfactory trial comparing a screened group with a non-intervention group (Prorok et al., 1984). This is mainly because chest X-ray examinations have already become widespread for the general population, so that it is difficult to isolate a control group not being screened by any medical resources. The case-control approach has been proposed for evaluating cancer screening (Sasco et al., 1986), especially for screening already widely carried out in the population. So far, screenings for cervical cancer (Clarke and Anderson, 1979; Aristizabal et al., 1984), breast cancer (Collette, 1985; Palli et al., 1986), and stomach cancer (Oshima et al., 1986) have been evaluated by this approach. Lung-cancer screening using miniature chest X-rays was also evaluated by a case-control study in East Berlin (Ebeling and Nischan, 1987). The results showed no benefits from screening in terms of reduction of mortality. However, since the distribution of histologic types and survival after treatment in the GDR (Ebeling et al., 1987) was quite different from the findings observed in Japan (Hanai et al., 1988; Ikeda et al., 1985), these results cannot be directly applied to our country. In Japan, lung-cancer screening programmes have bcen conducted for several years, based primarily on the screening system for tuberculosis. Since 1987, lung-cancer screening has been supported by the national government under the Health and Medical Services Law for the Aged, and 3,884,028 people

were screened by this system in 1988, although, in Japan as well, there has been no direct evidence of the benefit from lung-cancer screening in terms of reduction of mortality. This study aims to evaluate the efficacy of lung-cancer screening in terms of mortality reduction, by means of a case-control study. MATERIAL AND METHODS

The study areas consisted of 50 local municipalities (cities, towns, and villages) where population-based lung-cancer screening programmes have been conducted for several years by one of the Research Group participants. Geographical distribution of the study areas are shown in Figure 1. Most of these are located in rural areas and the populations have been stable, especially for elderly people. The screening programmes were started in 1977 (1 area), 1978 (3 areas), 1980 (1 area), 1981 (10 areas), 1982 (7 areas), 1983 (20 areas), 1984 (4 areas), 1985 (3 areas), and 1986 (1 area). Screening was offered by local governments in co-operation with local health centers or various private sector groups concerned with screening examinations in each of the areas. Since these screening programmes were based primarily on the screening system for tuberculosis offered by local governments, the target population was defined as all residents aged 16 or over except those who had the opportunity to be screened by chest X-ray examination at their places of employment or education. For employees, annual chest-X-ray screening has to be offered by employers under the Tuberculosis Control Law, independent of the screening offered by local governments, although compliance is not high in small companies. In practice, however, people screened by this system below the age of 40 were relatively few. In 37 of the study areas, both chest X ray for all participants and sputum cytology for high-risk groups were used in screening. In 9 areas, screening by both chest X ray and sputum cytology, and screening by chest X ray alone were conducted concurrently. In 4 areas, only chest X ray was used in screening. In most areas, chest X rays were taken by miniature photofluorography with 90-140 kv, using 100 mm x 100 mm film. Saccomano's 3-day pooled method was used for sputum cytology screening. The high-risk group for sputum cytology was defined a5 those who ranked 600 or more on the Smoking Index (average number of cigarettes smoked per day times years smoked) for 34 areas, those who ranked 800 or more on the Smoking Index for 6 areas, and those who ranked 400 or more on the Smoking Index for 6 areas. Ex-smokers who had stopped smoking 5 or more years ago were considered as a non-high-risk group, regardless of their rankings on the Smoking Index. Those who, after xreening either by chest X ray or by sputum cytology, were suspected of having lung cancer were further examined, and surgical resection was undertaken if possible.

4To whom correspondence and reprint requests should be addressed. Received: September 17,1991.

EVALUATION OF LUNG-CANCER SCREENING

23 1

FIGURE 1 - Geographical distribution of study areas (each dot represents one study area).

The study population, from which both cases and controls were selected, was defined as all residents in 40 areas, those who were invited to the screening in 8 areas, and those who have National Health Insurance in 2 areas. These were determined according to the availability of lists from which control candidates were chosen. Therefore these study populations were slightly different from the population actually invited to the screening programmes, except for the 8 areas where lists of invited people were available. Lists of all residents included those who were employed and not invited to the screening programmes, although the proportion of these people did not appear to be high in these 40 areas, which were all located in rural areas. National Health Insurance is one of the health insurance systems available in Japan. I t is run by local governments and covers all residents not covered by other health-insurance policies related to employment, such as farmers, retailers, and those employed by small companies, as well as retired people. By definition, all National Health Insurance holders were invited to the screening programmes. The difference was in those who were not employed but who were covered by health-insurance policies related to employ-

ment, such as the family members of employees. These people were invited to the screening programmes but not included in the list of National Health Insurance holders. The 2 areas in which lists of National Health Insurance holders were used for selecting cases and controls were located in suburban areas where a substantial proportion of residents appeared to be employed. In all, 136,860 males and 146,481 females aged 40 to 74 years old were included in the lists at the time the screening programmes started. Attendance rates at the screenings each year for these fixed cohorts could not be obtained, As a surrogate, attendance rates for dynamic cohorts aged 40 to 74 years old were calculated from the annual reports of screening in each study area, for all residents, those who were invited to the screening, and National Health Insurance holders. Study population and screenees per year were calculated from the year when the screening programmes were started up to the end of the period designated for collecting deceased lungcancer cases. Sex- and age-specific attendance rates were then calculated on a person-year basis. In 19 areas, where sex- and

232

SOBUE ET AL.

age-specific numbers were not available for a particular year, these numbers were estimated by applying sex- and agedistribution observed for the nearest year available in the same area. In 1981-88, all deaths due to lung cancer in the study populations were identified by death certificates for 1 year (1 area), 2 years (1 area), 3 years (19 areas), 4 years (4 areas), 5 years (14 areas), 6 years (2 areas), and 8 years (9 areas), after the screening programmes were started. O n average, the collection period of deceased lung-cancer cases in each area started 1.5 years after the year screening started. These collection periods were determined according to the availability of death certificates and the possibility of conducting additional interview surveys, described later. The case series used for the case-control comparison was defined as those who died of lung cancer between the ages of 40 and 74, were diagnosed as having lung cancer after the screening programme was started, and had been living in the same area since the screening programme was started. Information at diagnosis was collected from the medical records of the hospitals where the lung cancer was diagnosed or from cancer registries, if available. Migration status was investigated from records of residence obtained from local governments. For cases satisfying the above definition, an additional interview survey with the families of the deceased cases was conducted to investigate various conditions of the deceased cases before the time of diagnosis, namely, smoking habits, type of health insurance, frequency of health check-ups other than the lung-cancer screening offered by local governments, and routine consultations at local medical facilities. Since interview surveys could not be conducted with cases themselves, detailed questions to determine whether health checkups or medical consultation included chest X-ray examination or not were not asked. These interviews were conducted by well-trained public health nurses who belong to local health centers or local governments, and who visited the homes of cases using a uniform questionnaire. In order to increase the efficiency of identifying controls matched by smoking habits, cases were limited to a high-risk group for males, and a non-high-risk group for females. This limitation was added because more effort had to be made to find non-high-risk males and high-risk females as controls, due to the high smoking prevalence in males (65.5% in 1984) and low smoking prevalence in females (14.0% in 1984) (Tominaga, 1986). The definition of the high-risk group was the same as that used for screening programmes in 40 areas, where 400 or 600 on the Smoking Index was used as the cut-off value. In 6 areas, where 800 on the Smoking Index was used as the cut-off value, and in 4 areas, where no sputum cytology was conducted, 600 or more on the Smoking Index was used to define the high-risk group. Ex-smokers who had stopped smoking 5 or more years before the date of diagnosis of the cases were not included in the study. Information on smoking habits for the cases were obtained from medical records, screening records, and interview surveys. When there were discrepancies in information, priority was set to the higher Smoking Index. There were 861 lung-cancer cases identified by death certificates. Of these, 355 and 13 cases were excluded because they were either over 75 years old or less than 40 years old at the time of death. Eighteen cases were excluded because it was found through the review of medical records that there was clear evidence indicating other causes of death. Most of these cases had pulmonary metastases from other sites of cancer. Sixty cases were excluded because they had been diagnosed as having lung cancer before the screening programme was started. Fifteen cases were excluded because they moved into the municipality after the screening programme started. Nine cases were excluded because the date of diagnosis could not be identified. For males, 52 cases were excluded because they

were classified as belonging to a non-high-risk group, and for females, 17 cases because they were classified as belonging to a high-risk group. Finally, 28 cases were excluded because interview with the family was impossible, due to refusal, migration of the family to another area or because no family member remained. Response to the interview survey was made mainly by the spouse (63%), otherwise by a son or daughter (20%) or a daughter-in-law (11%). In 3 areas, no interview survey was conducted because records of smoking habits which covered the entire target population were available and no agreement to conduct interview surveys was obtained from local health centers. Also, in 6 areas, interview surveys were conducted only for smoking status. As a result, a total of 294 cases were collected. For each case, 20 candidate controls were selected from the list which covered the entire study population as defined above for the year the screening programme started. In each study area, all cases were identified in the list, and control candidates were selected from those located near the case, matched by sex and by year of birth ( ? 2 years). In most areas, the list was sorted by residence, so that potential controls were mostly matched by residence. In 18 areas, the list of residents was not available at the time when the screening programmes started, although figures on age and sex distribution were available from other statistics. In these areas, the oldest available list was used adding all deceased residents between the year screening started and the year the list was made. The average difference between these years was 3.2 years. For these control candidates, the same questions were asked as were asked for the cases. Interviews were conducted by public health nurses by home visits alone (12 areas), telephone and home visits (25 areas), telephone alone (2 areas), or mail and telephone (8 areas). Interviews were continued until 5 controls satisfied the following conditions, namely, that they had been living in the same municipality since the screening programme was started and were alive at the time of diagnosis of the case, and that they also belonged to the high-risk group for males or to the non-high-risk group for females. When there were discrepancies in information on smoking habits obtained from screening records and interview surveys, priority was set to the higher Smoking Index. During these processes, 34 control candidates were excluded because of migration to another area and 53 because of death before the diagnosis of the matched case. Another 451 control candidates were excluded because they were classified in the non-high-risk group for males, and 6 control candidates because they were classified in the high-risk group for females. In addition, 133 control candidates were skipped because contact with them was not possible. Response to the survey was made mainly by the control (67%), otherwise by the spouse (22%), a son or daughter (6%) and the daughter-in-law (5%). Again, no interview surveys were conducted in 3 areas, and interview surveys only for smoking status were conducted in 6 areas. As a result, a total of 1,470 controls were collected. According to the results of the interview surveys, 21 cases and 123 controls were covered by health-insurance policies related to their own employment in the areas where lists of all residents were used for choosing cases and controls. Since these cases and controls were originally not invited to the screening offered by local governments and had a higher chance of being screened outside the screening programme, they were excluded from the analysis. In addition, 78 controls whose matched case was excluded were excluded in order to hold matched sets (one case and a matched control). Finally, therefore, 273 cases and 1,269 controls were used for the analysis. Screening histories both for cases and for controls were obtained from the lists of screenees kept at local government offices or local health centers. In each area, screening histories

233

EVALUATION OF LUNG-CANCER SCREENING

of cases and matched controls were reviewed within the same calendar period, which was 60 months before the time the case had been diagnosed as having lung cancer. The screening test which led to the diagnosis of the case was included in the screening history. Since screening programmes were started fairly recently in some areas, and also for some sets, when the case had been diagnosed shortly after the screening started, the number of months in which the screening histories could be compared was 12 months for 273 sets, 24 months for 245 sets, 36 months for 210 sets, 48 months for 155 sets and 60 months for 109 sets. Screening histories were compared between the case and the controls during the 12 months before the date of diagnosis of the case. The duration during which screening histories were compared was then extended to 24, 36,48 and 60 months. Odds ratios of dying from lung cancer for screened vs. non-screened people were calculated using logistic regression analysis with a conditional likelihood function. Smokingadjusted odds ratios were calculated by including 8 dichotomous variables on the Smoking Index (200 to 399, 400 to 599, 600 to 799,800 to 999,1000 to 1199,1200 to 1399,1400 to 1599, and 1600 or more) in the model. These variables on the Smoking Index were excluded from the model when no subject was present in the category either for cases o r controls. Screening histories were categorized as dichotomous whether the study subjects were screened or not within 12, 24, 36, 48 and 60 months before the time when the case had been diagnosed as having lung cancer. Data on attendance to other health check-ups and the frequency of routine medical consultation, which appeared to be associated with the opportunities of examination by chest X ray outside the lung-cancer screening, were available from the interview surveys only for 201 cases and 875 controls. In order to estimate the confounding effects, these variables were included in the model, together with the variables on lungcancer screening and Smoking Index. The variable for other health check-ups was defined as those who had any kind of health check-up other than the lung cancer screening offered by local governments at least annually around the time when the matched case was diagnosed, and treated as dichotomous. The variable for routine medical consultations was defined as those who had had a routine medical consultation at least every 3 months for any reason around the time of diagnosis of the matched case, and treated as dichotomous. Subgroup analyses for the odds ratios according to sex, age, histologic type and location of tumor were also conducted by conditional logistic regression model, holding each matched set. In the analysis for estimating the effects of X ray alone and both X ray and sputum cytology separately, these 2 variables were included in the model as well as variables for the Smoking Indexes. In these analyses, variables for attendance at other health check-ups and the frequency of routine medical consultations were not included in the model. Statistical significance was evaluated with a 5 % level, and 95% confidence intervals (CI) were presented for estimated odds ratios. Analysis was conducted with the PROC MCSTRAT in the SAS computer programme (Naessens et al., 1986). RESULTS

Table I shows sex and age distribution of the study population at the time screening programmes were started. For all areas combined, 136,860 males and 146,481 females were included in the list at the time screening programmes were started, from which both cases and controls were selected. Sexand age-distribution was quite similar to that observed over the entire Japanese population. Attendance rates to lung-cancer screening programmes, from the time screening programmes

Age

Number of study population at the time the screening programmes started

X ray (%)

Cytology (%)

45,503 45,369 32,758 13,230 136,860

23.2 33.0 52.5 50.1 36.2

2.7 6.6 12.9 11.9 7.4

43,467 45,729 40,762 16,523 146,481

43.6 51.7 55.7 49.4 50.2

0.5 0.8 1.0 1.o 0.8

Males 40 to 49 50 to 59 60 to 69 70 to 74 Total Females 40 to 49 50 to 59 60 to 69 70 to 74 Total

Attendance rate during the study period

'Based o n the dynamic population after the screening programmes were started.

TABLE I1 - DISTRIBUTION OF HISTOLOGIC TYPE, LOCATION AND CLINICAL STAGE AT DIAGNOSIS FOR DECEASED LUNG-CANCER CASES Males Number

Histologic type Squamous-cell Adeno Small-cell Large-cell Other Unknown Location' Hilar Peripheral Other Unknown Clinical stage* Occult, I I1 111

IV Unknown Total

Females

% __ __ Number

%

79 59 26 15 3 26

38.0 28.4 12.5 7.2 1.4 12.5

11 32 6 3 0 13

16.9 49.2 9.2 4.6 0.0 20.0

71 94 7 36

34.1 45.2 3.4 17.3

12 31 5 17

18.5 47.7 7.7 26.2

18 8 67 77 38 208

8.7 3.8 32.2 37.0 18.3 100.0

6 2 16 27 14 65

9.2 3.1 24.6 41.5 21.5 100.0

'Determined by chest X ray examination.-2Based o n UICC, 1978.

were started until the end of period for collecting deceased lung-cancer cases, were also shown for all study areas combined according to sex and age categories. The mean duration for reviewing screening histories in each area was 6.1 years. Attendance rates at the screenings were calculated for the dynamic population rather than for fixed cohorts, as defined by the list. For X-ray examination, the screening rate was higher in females (50.2%) than in males (36.2%), but for sputum cytology, the rate was higher in males (7.4%) than in females (0.8%), reflecting the larger high-risk group in males. These attendance rates in each study area did not change substantially each year from the time the screening programmes were started, mainly because of the long history of screening for tuberculosis before the screening programmes were started. Table I1 shows the distribution of some characteristics at diagnosis for deceased lung-cancer cases which were used for case-control comparison. Out of 273 cases, 234 cases (86%) were diagnosed with histological or cytological evidence of lung cancer, while the remaining 39 cases were diagnosed by chest X-ray examinations and/or clinical findings alone. The predominant histologic type was squamous-cell carcinoma (38.0%) for males, and adenocarcinoma (49.2%) for females. About half of the cases were located peripherally on chest

234

SOBUE ET AL. TABLE I11 -DISTRIBUTION OF AGE AND SMOKING INDEX AT DIAGNOSIS FOR CASES AND CONTROLS Males

Females

Cases

Controls

Numher

Number

70

-

Cases

Controls

Numher

56

Numher

7 P

To

Age at diagnosis 40 to 49 50 to 59 60 to 69 70 to 74 Smoking index 0 1 to 399 400 to 599 600 to 1199 1200 to

5 34 99 70

2.4 16.3 47.6 33.7

17 142 45 1 354

1.8 14.7 46.8 36.7

5 20 31 9

7.7 30.8 47.7 13.8

20 91 148 46

6.6 29.8 48.5 15.1

0 0 10 118 80

Total

208

0.0 0.0 4.8 56.7 38.5 1oo.n

0 0 67 730 167 964

0.0 0.0 7.0 75.7 17.3 100.0

57 7 1 0 0 65

87.7 10.8 1.5 0.0 0.0 100.0

282 22 1 0 0 305

92.5 7.2 0.3 0.0 0.0 100.0

TABLE V - ODDS RATIOS OF DYING FROM LUNG CANCER ACCORDING TO MONTHS INCLUDED IN THE COMPARISON OF SCREENING HISTORY BEFORE DIAGNOSIS OF THE CASE Case: control = 1:2

Case Screened Not screened Case: control

=

Case: control

=

0 0 7

1:3

Case Screened Not screened

0 0 5

Total 1 8

2 1 0

1 0 0

2 1 2

Total 3

3 2 2

0 1 6

1 4 2

2 7 3

Total 29 16

3 4 7 1 0 3 2

Numher of matched controls screened

0 9 47

1 5 11

2 7 15

3 9 15

12 24 36 48 60

Cases

Controls

Cases

Controls

Smoking-adjusted odds ratio (')5%C')

273 246 210 155 109

1.269 11143 973 723 519

45 54 57 65 71

52 58 60 68 72

0.72 (0.50-1.03) 0.83 0.91 0.91 (0.54-1.55 1.13 (0.59-2.16/

Number

Screened

Of

(96)

9

Number of matched controls screened

Case: control = 1 5

Case Screened Not screened

1 0 1

Number of matched controls screened

1:4

Case Screened Not screened

Months before diagnosis

Number of matched controls screened

4 28 19

5 33 9

Total 91 116

Smoking-adjusted odds ratio = 0.72 (95% C1 = 0.50 - 1.03)

X-ray film for both sexes. Over two thirds of the cases had stage-111 or -1V cancers at diagnosis for both sexes. All these figures, however, should be interpreted with caution, because the proportions of unknown cases for these characteristics were not negligibly low. Table 111 shows the distribution of age and Smoking Index at diagnosis for cases and controls. The Smoking Index for controls was calculated using the years smoked up to the time the matched case was diagnosed. Age distribution shifted to higher categories in males than in females. Regarding Smoking Index in males, cases tended to have higher values than controls, although smoking status was matched according to high-risk or non-high-risk group. Table IV shows the distribution of matched sets according to screening histories during the 12 months before diagnosis of the case. The odds ratio of dying from lung cancer for those who were screened within 12 months compared with those not screened was 0.12 with a 0.50-1.03 95% confidence interval, after adjustment by Smoking Index. Table V shows the odds ratios when the duration for comparing screening histories was extended to 24, 36, 48 and 60 months before the year of diagnosis of the case. T h e odds ratios increased towards unity as the duration for comparing screening histories was extended.

TABLE VI - ODDS RATIO O F DYING FROM LUNG CANCER BASED ON THE CONDITIONAL LOGISTIC MODEL IN WHICH VARIABLES ASSOCIATED WITH THE CHANCES OF EXAMINATION BY CHEST X RAY WERE INCLUDED' Variables

@) Cases Controls

Smoking-adjusted Odd5 ratio (95% CI)

Lung-cancer screening' Other health check-ups' Routine medical consultation4 'Analyses were limited to 201 cases and 875 controls for which information o n additional variables was available.-'Attendance at screening offered by local government within 12 months.-'Attcndance at any kind of health check-up except the above screening at least ann~ally.-~Routine consultation at least every 3 months for any reason.

Table VI shows the odds ratios when the 2 variables for other health check-ups and routine medical consultations, which appeared to be associated with chances of examination by chest X ray outside the lung-cancer screening, were added in the model. All variables were dichotomous and the nonexposed group was used as a reference. For 201 cases and 875 controls used for this analysis, the odds ratio of dying from lung cancer for those screened within 12 months was estimated to b e 0.66 (95% CI 0.43-1.03) without adjustment for the above 2 variables. Distribution according to sex, age, Smoking Index, histologic type, location and clinical stage among these cases and controls was quite similar to those of the entire 273 cases and 1,269 controls. There was no change in the value of the odds ratio for lung-cancer screening after including the above 2 variables in the model. The variables for attendance at other health check-ups and the frequency of routine medical consultation showed the odds ratio close to unity. When the 201 cases and 875 controls were combined, 36% and 58% of those who had lung-cancer screening had other health check-ups and routine medical consultations, whereas 30% and 62% of those who did not have lung cancer screening had other health

235

EVALUATION OF LUNG-CANCER SCREENING T A R 1 F VTl - nnn

A case-control study for evaluating lung-cancer screening in Japan. Japanese Lung-Cancer-Screening Research Group.

In order to evaluate the efficacy of lung-cancer screening, a case-control study was conducted using the data from 50 areas where population-based lun...
987KB Sizes 0 Downloads 0 Views