Register Epidemiology Studies of Recent Cancer Trends in Selected Workers HANS S. R. MALKER AND JAN A. WEINER The National Board of Occupational Safety and Health I71 84 Solna, Sweden
JOSEPH K. McLAUGHLIN National Cancer Institute Bethesda, Maryland 20892
INTRODUCTION More than 200years ago Bernadino Ramazzini asserted that “Longe praestantius est prevenire quam curare” (“it is better to prevent than to cure”). To prevent a disease, one needs to identify and evaluate risk factors. One approach to such identificationis to analyzesystematicallydatafromnationalregisters,which link data on diseases and occupations. In England,more thana lOC!yearsago,FarrandOgleusedmortalitydatatodetect high-risk groups in the work environment.’,* In 1884, vital statistics were used in Sweden to identify risk groups in the work environment for a report of the Worker Insurance Committee.3 Sweden has a tradition of population registration dating back to 1749. The personal identification number, unique for every individual, was introducedin the late 1940s. With modern computer techniques, this provides a method for easy linkage of data bases at the individual level. Due to tradition and the personal identification number, Sweden and the other Nordic countries have unique opportunities to link, nationwide, population-based registers such as the National Swedish Cancer-Registry (NSCR) or the Cause of Death Registrywith Swedish census data. Register epidemiology is an analysis based on data from such linked administrative registers like the Cancer-Environment Registry (CER) that linkseffect datafromtheNSCRwith occupationaldatafromthe censuses and also links the Death Cause Registry (DCR) data with census data. The purpose of this paper is to discuss the potentials and limitations of these statistics in identifying time trends for cancer risks among occupational groups in Sweden. 322
MALKER ct PI.:
REGISTER EPIDEMIOLOGY STUDIES
323
MATERIAL The CER” which links cancer incidence data (1961-1979, 1971-1984) with employmentdata (l960,1970),providesauniqueopportunitytosystematicallyassess the occurrence of malignant neoplasms by industry and occupation for an entire country. The NSCR wasestablished in 1958andcontainsabout 605,000cancercasesfor the period 1961-1979 and 555,000 cancer cases for the period 1971-1984.A nationallaw requires cancer cases to be reported to the NSCR. The compulsory reporting law covers all malignant tumors and also certain histologically benign tumors and precancerouslesions.’ Approximately95% of thecastsin the NSCR are microscopically verified. The 1960 and 1970National Population and Housing Censuses contain detailed informationon the Swedishpopulation. The informationon occupationand industry in the two censusesrefer to employmentduring one particular week in October 1960 and 1970,respectively,when the censuseswere performed. The occupationalclassifications used in the censuses were revised versions of the standards from the International Labour Office and the United Nations?,’ Record linkage between the NSCR and the 1960and 1970censuses was possible owing to the personal identification number, which consists of year, month, and day of birth plus a three-digit serialnumber and acheck digit. In the present CER analysis, cancer incidence data from 1961-1979 is linked with the 1960 census and cancer incidence from 1971-1984 is linked with the 1970censu~.6*’*~~ Results from this latest linkage became available in October 1989.
METHODS To be able to estimate time trends by occupation the CER has been divided into three 8-yeartime periods, 1961-1968,1969-1976with the 1960census as background population, and 1977-1984 with the 1970census as background population. A standardized incidence ratio (SIR) was used to estimate relative risks in the three time periods for the occupationalcategoriesused as examples. The SIR is the ratio between the observed number of cases in a particular occupation and the expected number of cases in the same group. The expected number of cases is obtained by applying the birth cohort- and sex-specificincidence rates for a specific neoplasia for the period under study from the total Swedish population to the birthcohort-andsex-specificdistributionofthe occupationalgroup under study. Five-year birth cohorts were used in these calculations. Because of variation in cancer incidence rates by region, SIRSwere adjusted for region, with the use of the three metropolitan areas and 24 countiesin the indirect standardization.’’ Togive examplesof time trendsin absoluterisk,datafrom theNSCRfor two time periods (1961-1970and 1971-1980)have been used with the background population from the census 1960and 1970,respectively. The examplesin this paper were chosen based on our earlier results uncovered in hypothesis-generating investigationsusing the CER for the time period 1961-1979.
ANNALS NEW YORK ACADEMY OF SCIENCES
324
RESULTS Pleural Mesotheliomas
Pleural mesothelioma is induced by asbestos exposure often as much as 40 years later and in Sweden that invariably means exposure in the work environment. As seen in FIGURE 1,the number of cases are exponentially distributed over time. The age-standardized incidence has doubled in avery short time period from 1980to 1985 from l.l/lOO,OOO to U100,OOO. This also suggests that for the same SIR the incidence rate should also rise exponentially. Among occupations with 10 or more cases for the period 1961-1979 plumbers, mechanics, painters, and electricians were found to have the highest excess risks in Sweden." As seen from FIGURE 2, plumbers and pipefitters have experienced an increasing relative risk of pleural mesothelioma over time, whereas mechanics, electricians, and also painters and paper hangers have shown no clear time trend. However, the number of cancer cases per 100,OOO for those occupations are increasing, except for painters and paper hangers, as shownin FIGURE 3. The number of cases upon which these trends are based are small (as can be seen in FIGURE 2) and estimates of age- or birth-cohort specific rates cannot be made. Those occupations with increasing risks are those that repair asbestos-containing materials.
100
80
60
40
20
n
1960
I
I
I
1970
1980
1990
FIGURE 1. Number of incident cases of pleural mesothelioma in Sweden. 1960-1985.
MALKER cf a/.: REGISTER EPIDEMIOLOGY STUDIES
325
20
SIR
8
lo:
8.......................Q 0
4
9'
.. - \:, ?#
3= \ 1
1:
e
12
0 0 0
.-.-.a
5
>g* 2 , :\
* 4 S 4
*.
\
'0
:
w
0 0 0 0
t
4
@
--
/ @
mm***D...plumbers
. )
-.*.
0 0
00 0 0 0 0
I
mechanics painters
electricians
0
0: b-
,1
I
i
Nasal Cancer The second example is taken from nasal adenocarcinoma,which is well known to be occupationally related to woodworking, especially furniture making. The incidence rate for nasal cancer in Sweden has been fairly constant over time about 1case/ 100,OOOfor males. In Sweden, cabinet workers and other wood-workers had the highest excess risk based on more than 10 cases in the period 1961-1979.12 FIGURES 4 and 5 indicate that in spite of improvements in the work environment since the late 1960s, it seems that the latency period may be so long that the lowered exposure levels in the 1970s and 1980s have not had any effect on the relative risk or incidence per 100,ooO.
Physicians and Squamous Cell Carcinoma The third example is skin cancer where earlier we reported more than a fivefold excess risk for squamous cell carcinoma of the upper limb among physicians associated with low doses of ionizing r a d i a t i ~ n The . ~ doses of ionizing radiation have gone down considerably over the last 30 years and the SIR has also declined. The total
ANNALS NEW YORK ACADEMY OF SCIENCES
326
11100 000
50
plumbers
40 30 mechanics
20
painters
electricians
10
0 64
66
68
70
72
74
76
FIGURE 3. Incidence rate of pleural mesothelioma for selected occupations during the periods 1961-1970 and 1971-1980.
number of cases among physiciansincreases during the period (FIG.6). However, the number of physicians increased between 1960 and 1970 and the incidence rate has slightly decreased (FIG.7). Skin cancer is also associated with solar radiation, and the incidence rate for the total Swedish population has been increasing? Thus the decreasing SIR and the slightly decreasing incidence rate suggest that the importance of occupational risk factors have decreased during the period.
DISCUSSION Incidence or Mortality Data Although the NSCR data for the period 1961-1979 has been linked to the 1960 Census and for the period 1971-1984 to the 1970 Census, the Swedish cause of death registry has been linked for a longer period using more censuses. The deaths for the period 1961-1970,1971-1980, and 1981-1986 have been linked to the Censuses 1960, 1970,and 1980,respectively. This would seem to indicate that perhaps mortality data should be used for a trend analysis of occupational cancer. There are three main reasons why incidence data were preferred to mortality data in this analysis of occupations at high risk of cancer. Firstly, the codingrulesin the Death CauseRegistrywere changed in 1981. If there is a cause of death in part Iof the death certificate (the underlying cause of death) that can lead to the death by itself, it alone will be coded with no mention of a contributing cause, even if a cancer. This change can easily be seen in cancer mortality statisticsfor
MALKER
ct
al.: REGISTER EPIDEMIOLOGY STUDIES
327
Sweden. There is not enough information in the WHO data base on which trend calculations and international comparisonsfor this monograph are based. Therefore, the WHO data base on deaths cannot be used for trend estimations on cancer mortality for Sweden covering the period before and after 1981. Secondly, the cancer incidence data in the NSCR is preferred to cancer deaths registered in the DCR,because among other things, the diagnoses are better and histologyis included. Moreover, the cancer register is also linked to the cause of death register to get the death cause, e.g., to be able to calculate survival rates. Thirdly,from an etiologic point of view, incidence data rather than mortality data provide a more accurate picture of the trend because mortality may change due to other circumstances, e.g., better treatment.
Industry versus Occupation
Time trends can be estimated for occupations but not for specific industries in Sweden using the CER for two reasons. First of all, the industrial classification and coding rules changed between the 1960and 1970 censusesand there is no translation between the 1960industrycodes and 1970industrycodes on thedetailedlevels which is a three-digit level for 1960 and four-digit level for 1970 census. Secondly, the
SIR
9
10
10
1 5
1 : 60
I
i
70
80
Ye=
FIGURE 4. SIRS and number of cases for adenocarcinoma of the nasal cavity for selected occupations, for the periods 1961-1968.1969-1976, and 1977-1984.
ANNALS NEW YORK ACADEMY OF SCIENCES
328
m o o 000
Cabinetmakers
3200 1
* Others wood
1°i
0 ; 64
I
I
66
6.8
’
1
70
’
1
72
.
1
74
i
Yew
76
FIGURE 5. Incidence rate for adenocarcinomas of the nasal cavity for selected occupations during the periods 1961-1970 and 1971-1980.
industry codingisaccordingto “the main activity” forthe firm, which may change over a ten-year period. Fortunately, the occupational codes to date have been comparable over time and translations exist between different revisions. However, the forthcoming 1988 ISCO revision is completely different, so future trend analysis by occupation may also be difficult to perform.
The Quality in the CER The quality of the CER data depends on that of the constituent registries. Studies
on the completeness of the NSCR have estimated the deficits in reporting of cancer cases to be 4% in total, and only Y YO when notification was made by at least two independent sources (clinicians and pathologists). Completeness varied by region and age, with significant deficits in certain counties and among the elderly. Multiple myeloma and leukemia had the largest deficits reaching 18% .13 A n independent validation of industrial and occupational data by a reinterview of a random sample of the 1960census revealed good agreement at the broad one-digit level (95%) for occupation and somewhat less at the two-digit level (90%), and even less for some occupations at the three-digit level. The main areas of discrepancyin the employment codings were in administrative work. The same study also revealed that the number of gainfully employedis seriously underestimated among mamed women,
MALKER
et
al.: REGISTER EPIDEMIOLOGY STUDIES
329
especially in agric~lture.'~ To date, no comparative evaluation has been performed on the 1970 census. In thelinkages, only about 1% of thecancer cases could not be traced in the census. False matches between thecensus 1960andthe NSCR 1961-1979 have been estimated to be less than 0.5% .15 In the linkage between NSCR 1971-1984 and the census 1970 only 0.1% of the cancer cases could not be traced.' Several factors affect the quality of data in the CER. The employment status of the Swedish population is defined only for the week of the census, and then followed over time for cancer experience within a specified occupational group. The occupation and industry codes very often include a heterogeneous group of workers. The variation in intensity and type of exposure within these groups can be substantial. Sometimes this problem can be reduced by evaluating occupation within industry, or possiblyutilizingotherinformation, to increase the focuson the observed population: If only a fraction of the individuals included in an occupation or industry classification is exposed, the results will underestimate the real risk, thus reducing the likelihood of reporting false excess risks.'6 There are also misclassificationsin the coding of the Census information that may differ between the censuses. Such differences need to be evaluated to be able to interpret the results.
10
-
4
phyddanr
1
I
1
Year
FIGURE 6. SIRS and number of cases for non-melanotic skin cancer of the upper limb among physicians, for the periods, 1961-1968.1969-1976, and 1977-1984.
ANNALS NEW YORK ACADEMY OF SCIENCES
330
The Methods The denominator in the estimation of expected numbers is based on the population in 1960 and 1970, respectively. To use the census population instead of personyears-at-risk implies that the calculated expected number will be incorrect if the intercurrent deaths differ between the studied group and the total population. Even when considering intercurrent deaths for extreme categories with respect to life expectancies, the influence on the SIR has been demonstrated to be minimal.17
1/100 000 110 7 100
-
80 70 60 90
50 40
physicians
-
-
20 10 30
0 ; 64
’
1
I
66
68
.
1
70
.
1
72
’
1
74
,
year
76
FIGURE 7. Incidence rate for non-melanoticskin c a n e r of the upper limh for physicians, during the periods 1961-1970 and 1971-1980.
Evenwhen usingdatafromanentirecountry, witha population of about 8 million, the number of cancer cases can be very small when examined by time periods and occupational and industry classifications. Another problem is selective migration between occupations o r industries, either by individualsor by changesincodingrules. With diseases of long latencies like cancer the use of different background populations that are not linked, may impose serious errors when estimating time trends, especially when a selective “migration” between occupations occurs.
CONCLUDING REMARKS To our knowledge, this is the first attempt to use linked registers to estimate time trends for cancer risks by occupation using different censuses as background popula-
MALKER er al.: REGISTER EPIDEMIOLOGY STUDIES
331
tions. Pleural mesothelioma cases doubled in a very short time period, when ascertainment for this rare site remained consistently high. The time period studied is fairly short, 24 years, and may not be long enough to detect substantial changes in the work environment. More sophisticated methods for analysis of time trends need to be used and more studiesshould be done toevaluate if this typeof analysis is a useful way to develop further register epidemiology and to refine the identification of highrisk groups in the workplace.
SUMMARY The Swedish Cancer Environment Register (CER) was used t o analyze occupational risks of cancer over time. Using expanded time coverage (1961-1984) and two census-based sources in employment information, risk in the work environment was evaluated for a 24-year period for pleural mesothelioma, nasal adenocarcinoma, and non-melanotic skin cancer. Standardized incidence ratios (SIRs) for pleural mesotheliomas were found to increase among plumbers and pipefitters over this period, whereas those for mechanics, electricians, painters, and paperhangers remained relatively stable. Risk for nasal adenocarcinoma among cabinet makers remained exceedingly high over the 24-year period and SIRs for other kinds of woodworkers moderately increased during the same time. Excess skin cancer risk among physicians declined over the period. Reasons for these trends are discussed as well as the limitations and potential of register epidemiology in the evaluation of cancer trends in the work environment.
REFERENCES 1. F m , W. 1864. Letter to the Registrar General. In Supplement to the 25th Annual Report of the Registrar General of births, deaths and marriages in England. HMSO. London. 2. OGLE,W. 1885. Supplement to the 45th Annual Report of the Registrar General of births,
deaths and marriages in England. HMSO. London. 3. 1884 &rs arbetaref6rsakringskommittesbetankande. Stockholm 1888 (Swedish only). 4. WIKLUND, K., J. EINHORN, G. WENNSTR~M & E. RAPAPORT. 1981. A Swedish cancer environment register available for research. Scand. J. Work Environ. Hlth. % 64-67. 5. MALKER, H. S. R. 1988. Register-Epidemiology in the identification of cancer risks. Arbete och Halsa, National Institute of Occupational Health. Solna, Sweden. 6. Socialstyrelsen. 1989. Cancermiljoregister -70. En samkorning av cancerregistret 1971-84 och Folk- och bostadsrakningen 1970. Socialstyrelsen. Stockholm, Sweden (Swedish only). 7. THECANCER REGISIRY. 1989. Cancer Incidence in Sweden-1985. National Board of Health and Welfare, Stockholm, Sweden. 8. INTERNATIONAL LABOUROmm. 1958. International Standard Classification of Occupations. International Labour Office. Geneva. 9. UNITEDNATIONS STATISTICAL Omce. 1958. International Standard Industrial Classification of all Economic Activities. United Nations Statistical Office.New York. 10. Socialstyrelsen. 1980. Cancerrisk och miljofaktorer. Rapport till Arbetarskyddsfonden. Stockholm, Sweden (Swedish only). 11. MALKER. H. S. R., J. K. MCLAUGHLIN, B. K. MALKER, B. J. STONE,J. A. WEINER, J. L. E. ERICSSON & W.J. BLOT.1985. Occupational risks for pleural mesothelioma in Sweden 1961-79. J. Natl. Cancer Inst. 74: 61-66.
-.
ANNALS NEW YORK ACADEMY OF SCIENCES
332
12. MALKER, H. S. R., J. MCLAUGHUN, W. J. BLOT,J. A. WEINER,B. MALKER, J. L. E. ERICSON &B. J. STONE. 1986.NasalcancerandoccupationinSweden1961-1979.Am.J.Epidemiol. R 477-485.
13. MAITSSON, B. 1984.Cancer registration in Sweden. Studies on completeness and validity of
incidence and mortality registers. Thesis. Karolinska Hospital. Stockholm, Sweden. 14. BRIVKALNE, M. 1964. The control study made in connection with the 1960 census of population. Statistical reports B. National Central Bureau of Statistics. Stockholm, Sweden (Swedish only). 15. WIKLUND, K. & G. EKLUND. 1986. Reliability of record linkage in the Swedish cancerenvironment register. Acta Radiol. UC 198-201. 16. AXELSON, 0.1986.Linking of register information on morbidity and occupation, especially with regard to quality aspects. 27-40.WHO ICP/OCH 002.Copenhagen. 17. MALKER,H. & J. A. WEINER.1984. The Cancer-Environment Registry 1961-1973. Examples of the use of register epidemiology in studies of the work environment. Arbete och Halsa, National Board of Occupational Safety and Health. Solna, Sweden (summary in English).
DISCUSSION
JOHN BAILAR(McCill University, Montreal, P.Q.): I think there are some even broader lessons that might be drawn here. One is the legitimate concern about reporting bias. I recall many years ago discussions about theincreasingratesrecorded for mesothelioma in the United States asked how much of this was real and whether all of the recorded increase might reflect an artefact. Theincrease was noted and was initially attributed just to changes in diagnosis; that is, bias. On the other side, we have the pressing need for hypothesis generation. But that also can fail sometimes. We have as evidence the frequency of reports of geographic clusters of cancer, virtually all of which on further investigation prove to be not real. Somehow we have to learn to draw a fine line between the need to avoid bias and over-detection of alot ofthingsthat aren’t thereand theneedtogenerateappropriate hypotheses from the clinical and registry experience. I don’t think we draw the line at the optimum place at present. There may be an over-concern with bias. We ought to be a bit more liberal about this. But we can’t go too far in either direction without getting in a lot of trouble. D. GOLDSTEIN (Robert WoodJohnsonMedical School, Piscataway, NJ): BERNARD I want to take issue with Dr. Malker’s statement. I have a minority view perhaps, but I think it is a misconception of the work that Selikoff has done. Selikoff has clearly shown us that if you get exposed to asbestos for afew years earlier in life then you’ve got perhaps 30 years of waiting for the potential lag time of the mesothelioma. Dr. Malker has told us that for nasal cancer and adenoid carcinomas, maybe you have not seen a decrease because there is some lag time involved from the problems that wood workers are being exposed to. I really take issue with that. I think we have to keep in mind that asbestos stays in the body during those many years. Most of the other chemical exposures are transient. I think that you’re much more likely to expect the kind of curve we see when you stop smokingcigarettes-avery quick decrease in risk. If you do not see that decrease in risk after the cessation of exposure and we’re studying something that remains in the body a long time, then we must consider that
JOSEPH K. McLAUGHLIN National Cancer Institute Bethesda, Maryland 20892
INTRODUCTION More than 200years ago Bernadino Ramazzini asserted that “Longe praestantius est prevenire quam curare” (“it is better to prevent than to cure”). To prevent a disease, one needs to identify and evaluate risk factors. One approach to such identificationis to analyzesystematicallydatafromnationalregisters,which link data on diseases and occupations. In England,more thana lOC!yearsago,FarrandOgleusedmortalitydatatodetect high-risk groups in the work environment.’,* In 1884, vital statistics were used in Sweden to identify risk groups in the work environment for a report of the Worker Insurance Committee.3 Sweden has a tradition of population registration dating back to 1749. The personal identification number, unique for every individual, was introducedin the late 1940s. With modern computer techniques, this provides a method for easy linkage of data bases at the individual level. Due to tradition and the personal identification number, Sweden and the other Nordic countries have unique opportunities to link, nationwide, population-based registers such as the National Swedish Cancer-Registry (NSCR) or the Cause of Death Registrywith Swedish census data. Register epidemiology is an analysis based on data from such linked administrative registers like the Cancer-Environment Registry (CER) that linkseffect datafromtheNSCRwith occupationaldatafromthe censuses and also links the Death Cause Registry (DCR) data with census data. The purpose of this paper is to discuss the potentials and limitations of these statistics in identifying time trends for cancer risks among occupational groups in Sweden. 322
MALKER ct PI.:
REGISTER EPIDEMIOLOGY STUDIES
323
MATERIAL The CER” which links cancer incidence data (1961-1979, 1971-1984) with employmentdata (l960,1970),providesauniqueopportunitytosystematicallyassess the occurrence of malignant neoplasms by industry and occupation for an entire country. The NSCR wasestablished in 1958andcontainsabout 605,000cancercasesfor the period 1961-1979 and 555,000 cancer cases for the period 1971-1984.A nationallaw requires cancer cases to be reported to the NSCR. The compulsory reporting law covers all malignant tumors and also certain histologically benign tumors and precancerouslesions.’ Approximately95% of thecastsin the NSCR are microscopically verified. The 1960 and 1970National Population and Housing Censuses contain detailed informationon the Swedishpopulation. The informationon occupationand industry in the two censusesrefer to employmentduring one particular week in October 1960 and 1970,respectively,when the censuseswere performed. The occupationalclassifications used in the censuses were revised versions of the standards from the International Labour Office and the United Nations?,’ Record linkage between the NSCR and the 1960and 1970censuses was possible owing to the personal identification number, which consists of year, month, and day of birth plus a three-digit serialnumber and acheck digit. In the present CER analysis, cancer incidence data from 1961-1979 is linked with the 1960 census and cancer incidence from 1971-1984 is linked with the 1970censu~.6*’*~~ Results from this latest linkage became available in October 1989.
METHODS To be able to estimate time trends by occupation the CER has been divided into three 8-yeartime periods, 1961-1968,1969-1976with the 1960census as background population, and 1977-1984 with the 1970census as background population. A standardized incidence ratio (SIR) was used to estimate relative risks in the three time periods for the occupationalcategoriesused as examples. The SIR is the ratio between the observed number of cases in a particular occupation and the expected number of cases in the same group. The expected number of cases is obtained by applying the birth cohort- and sex-specificincidence rates for a specific neoplasia for the period under study from the total Swedish population to the birthcohort-andsex-specificdistributionofthe occupationalgroup under study. Five-year birth cohorts were used in these calculations. Because of variation in cancer incidence rates by region, SIRSwere adjusted for region, with the use of the three metropolitan areas and 24 countiesin the indirect standardization.’’ Togive examplesof time trendsin absoluterisk,datafrom theNSCRfor two time periods (1961-1970and 1971-1980)have been used with the background population from the census 1960and 1970,respectively. The examplesin this paper were chosen based on our earlier results uncovered in hypothesis-generating investigationsusing the CER for the time period 1961-1979.
ANNALS NEW YORK ACADEMY OF SCIENCES
324
RESULTS Pleural Mesotheliomas
Pleural mesothelioma is induced by asbestos exposure often as much as 40 years later and in Sweden that invariably means exposure in the work environment. As seen in FIGURE 1,the number of cases are exponentially distributed over time. The age-standardized incidence has doubled in avery short time period from 1980to 1985 from l.l/lOO,OOO to U100,OOO. This also suggests that for the same SIR the incidence rate should also rise exponentially. Among occupations with 10 or more cases for the period 1961-1979 plumbers, mechanics, painters, and electricians were found to have the highest excess risks in Sweden." As seen from FIGURE 2, plumbers and pipefitters have experienced an increasing relative risk of pleural mesothelioma over time, whereas mechanics, electricians, and also painters and paper hangers have shown no clear time trend. However, the number of cancer cases per 100,OOO for those occupations are increasing, except for painters and paper hangers, as shownin FIGURE 3. The number of cases upon which these trends are based are small (as can be seen in FIGURE 2) and estimates of age- or birth-cohort specific rates cannot be made. Those occupations with increasing risks are those that repair asbestos-containing materials.
100
80
60
40
20
n
1960
I
I
I
1970
1980
1990
FIGURE 1. Number of incident cases of pleural mesothelioma in Sweden. 1960-1985.
MALKER cf a/.: REGISTER EPIDEMIOLOGY STUDIES
325
20
SIR
8
lo:
8.......................Q 0
4
9'
.. - \:, ?#
3= \ 1
1:
e
12
0 0 0
.-.-.a
5
>g* 2 , :\
* 4 S 4
*.
\
'0
:
w
0 0 0 0
t
4
@
--
/ @
mm***D...plumbers
. )
-.*.
0 0
00 0 0 0 0
I
mechanics painters
electricians
0
0: b-
,1
I
i
Nasal Cancer The second example is taken from nasal adenocarcinoma,which is well known to be occupationally related to woodworking, especially furniture making. The incidence rate for nasal cancer in Sweden has been fairly constant over time about 1case/ 100,OOOfor males. In Sweden, cabinet workers and other wood-workers had the highest excess risk based on more than 10 cases in the period 1961-1979.12 FIGURES 4 and 5 indicate that in spite of improvements in the work environment since the late 1960s, it seems that the latency period may be so long that the lowered exposure levels in the 1970s and 1980s have not had any effect on the relative risk or incidence per 100,ooO.
Physicians and Squamous Cell Carcinoma The third example is skin cancer where earlier we reported more than a fivefold excess risk for squamous cell carcinoma of the upper limb among physicians associated with low doses of ionizing r a d i a t i ~ n The . ~ doses of ionizing radiation have gone down considerably over the last 30 years and the SIR has also declined. The total
ANNALS NEW YORK ACADEMY OF SCIENCES
326
11100 000
50
plumbers
40 30 mechanics
20
painters
electricians
10
0 64
66
68
70
72
74
76
FIGURE 3. Incidence rate of pleural mesothelioma for selected occupations during the periods 1961-1970 and 1971-1980.
number of cases among physiciansincreases during the period (FIG.6). However, the number of physicians increased between 1960 and 1970 and the incidence rate has slightly decreased (FIG.7). Skin cancer is also associated with solar radiation, and the incidence rate for the total Swedish population has been increasing? Thus the decreasing SIR and the slightly decreasing incidence rate suggest that the importance of occupational risk factors have decreased during the period.
DISCUSSION Incidence or Mortality Data Although the NSCR data for the period 1961-1979 has been linked to the 1960 Census and for the period 1971-1984 to the 1970 Census, the Swedish cause of death registry has been linked for a longer period using more censuses. The deaths for the period 1961-1970,1971-1980, and 1981-1986 have been linked to the Censuses 1960, 1970,and 1980,respectively. This would seem to indicate that perhaps mortality data should be used for a trend analysis of occupational cancer. There are three main reasons why incidence data were preferred to mortality data in this analysis of occupations at high risk of cancer. Firstly, the codingrulesin the Death CauseRegistrywere changed in 1981. If there is a cause of death in part Iof the death certificate (the underlying cause of death) that can lead to the death by itself, it alone will be coded with no mention of a contributing cause, even if a cancer. This change can easily be seen in cancer mortality statisticsfor
MALKER
ct
al.: REGISTER EPIDEMIOLOGY STUDIES
327
Sweden. There is not enough information in the WHO data base on which trend calculations and international comparisonsfor this monograph are based. Therefore, the WHO data base on deaths cannot be used for trend estimations on cancer mortality for Sweden covering the period before and after 1981. Secondly, the cancer incidence data in the NSCR is preferred to cancer deaths registered in the DCR,because among other things, the diagnoses are better and histologyis included. Moreover, the cancer register is also linked to the cause of death register to get the death cause, e.g., to be able to calculate survival rates. Thirdly,from an etiologic point of view, incidence data rather than mortality data provide a more accurate picture of the trend because mortality may change due to other circumstances, e.g., better treatment.
Industry versus Occupation
Time trends can be estimated for occupations but not for specific industries in Sweden using the CER for two reasons. First of all, the industrial classification and coding rules changed between the 1960and 1970 censusesand there is no translation between the 1960industrycodes and 1970industrycodes on thedetailedlevels which is a three-digit level for 1960 and four-digit level for 1970 census. Secondly, the
SIR
9
10
10
1 5
1 : 60
I
i
70
80
Ye=
FIGURE 4. SIRS and number of cases for adenocarcinoma of the nasal cavity for selected occupations, for the periods 1961-1968.1969-1976, and 1977-1984.
ANNALS NEW YORK ACADEMY OF SCIENCES
328
m o o 000
Cabinetmakers
3200 1
* Others wood
1°i
0 ; 64
I
I
66
6.8
’
1
70
’
1
72
.
1
74
i
Yew
76
FIGURE 5. Incidence rate for adenocarcinomas of the nasal cavity for selected occupations during the periods 1961-1970 and 1971-1980.
industry codingisaccordingto “the main activity” forthe firm, which may change over a ten-year period. Fortunately, the occupational codes to date have been comparable over time and translations exist between different revisions. However, the forthcoming 1988 ISCO revision is completely different, so future trend analysis by occupation may also be difficult to perform.
The Quality in the CER The quality of the CER data depends on that of the constituent registries. Studies
on the completeness of the NSCR have estimated the deficits in reporting of cancer cases to be 4% in total, and only Y YO when notification was made by at least two independent sources (clinicians and pathologists). Completeness varied by region and age, with significant deficits in certain counties and among the elderly. Multiple myeloma and leukemia had the largest deficits reaching 18% .13 A n independent validation of industrial and occupational data by a reinterview of a random sample of the 1960census revealed good agreement at the broad one-digit level (95%) for occupation and somewhat less at the two-digit level (90%), and even less for some occupations at the three-digit level. The main areas of discrepancyin the employment codings were in administrative work. The same study also revealed that the number of gainfully employedis seriously underestimated among mamed women,
MALKER
et
al.: REGISTER EPIDEMIOLOGY STUDIES
329
especially in agric~lture.'~ To date, no comparative evaluation has been performed on the 1970 census. In thelinkages, only about 1% of thecancer cases could not be traced in the census. False matches between thecensus 1960andthe NSCR 1961-1979 have been estimated to be less than 0.5% .15 In the linkage between NSCR 1971-1984 and the census 1970 only 0.1% of the cancer cases could not be traced.' Several factors affect the quality of data in the CER. The employment status of the Swedish population is defined only for the week of the census, and then followed over time for cancer experience within a specified occupational group. The occupation and industry codes very often include a heterogeneous group of workers. The variation in intensity and type of exposure within these groups can be substantial. Sometimes this problem can be reduced by evaluating occupation within industry, or possiblyutilizingotherinformation, to increase the focuson the observed population: If only a fraction of the individuals included in an occupation or industry classification is exposed, the results will underestimate the real risk, thus reducing the likelihood of reporting false excess risks.'6 There are also misclassificationsin the coding of the Census information that may differ between the censuses. Such differences need to be evaluated to be able to interpret the results.
10
-
4
phyddanr
1
I
1
Year
FIGURE 6. SIRS and number of cases for non-melanotic skin cancer of the upper limb among physicians, for the periods, 1961-1968.1969-1976, and 1977-1984.
ANNALS NEW YORK ACADEMY OF SCIENCES
330
The Methods The denominator in the estimation of expected numbers is based on the population in 1960 and 1970, respectively. To use the census population instead of personyears-at-risk implies that the calculated expected number will be incorrect if the intercurrent deaths differ between the studied group and the total population. Even when considering intercurrent deaths for extreme categories with respect to life expectancies, the influence on the SIR has been demonstrated to be minimal.17
1/100 000 110 7 100
-
80 70 60 90
50 40
physicians
-
-
20 10 30
0 ; 64
’
1
I
66
68
.
1
70
.
1
72
’
1
74
,
year
76
FIGURE 7. Incidence rate for non-melanoticskin c a n e r of the upper limh for physicians, during the periods 1961-1970 and 1971-1980.
Evenwhen usingdatafromanentirecountry, witha population of about 8 million, the number of cancer cases can be very small when examined by time periods and occupational and industry classifications. Another problem is selective migration between occupations o r industries, either by individualsor by changesincodingrules. With diseases of long latencies like cancer the use of different background populations that are not linked, may impose serious errors when estimating time trends, especially when a selective “migration” between occupations occurs.
CONCLUDING REMARKS To our knowledge, this is the first attempt to use linked registers to estimate time trends for cancer risks by occupation using different censuses as background popula-
MALKER er al.: REGISTER EPIDEMIOLOGY STUDIES
331
tions. Pleural mesothelioma cases doubled in a very short time period, when ascertainment for this rare site remained consistently high. The time period studied is fairly short, 24 years, and may not be long enough to detect substantial changes in the work environment. More sophisticated methods for analysis of time trends need to be used and more studiesshould be done toevaluate if this typeof analysis is a useful way to develop further register epidemiology and to refine the identification of highrisk groups in the workplace.
SUMMARY The Swedish Cancer Environment Register (CER) was used t o analyze occupational risks of cancer over time. Using expanded time coverage (1961-1984) and two census-based sources in employment information, risk in the work environment was evaluated for a 24-year period for pleural mesothelioma, nasal adenocarcinoma, and non-melanotic skin cancer. Standardized incidence ratios (SIRs) for pleural mesotheliomas were found to increase among plumbers and pipefitters over this period, whereas those for mechanics, electricians, painters, and paperhangers remained relatively stable. Risk for nasal adenocarcinoma among cabinet makers remained exceedingly high over the 24-year period and SIRs for other kinds of woodworkers moderately increased during the same time. Excess skin cancer risk among physicians declined over the period. Reasons for these trends are discussed as well as the limitations and potential of register epidemiology in the evaluation of cancer trends in the work environment.
REFERENCES 1. F m , W. 1864. Letter to the Registrar General. In Supplement to the 25th Annual Report of the Registrar General of births, deaths and marriages in England. HMSO. London. 2. OGLE,W. 1885. Supplement to the 45th Annual Report of the Registrar General of births,
deaths and marriages in England. HMSO. London. 3. 1884 &rs arbetaref6rsakringskommittesbetankande. Stockholm 1888 (Swedish only). 4. WIKLUND, K., J. EINHORN, G. WENNSTR~M & E. RAPAPORT. 1981. A Swedish cancer environment register available for research. Scand. J. Work Environ. Hlth. % 64-67. 5. MALKER, H. S. R. 1988. Register-Epidemiology in the identification of cancer risks. Arbete och Halsa, National Institute of Occupational Health. Solna, Sweden. 6. Socialstyrelsen. 1989. Cancermiljoregister -70. En samkorning av cancerregistret 1971-84 och Folk- och bostadsrakningen 1970. Socialstyrelsen. Stockholm, Sweden (Swedish only). 7. THECANCER REGISIRY. 1989. Cancer Incidence in Sweden-1985. National Board of Health and Welfare, Stockholm, Sweden. 8. INTERNATIONAL LABOUROmm. 1958. International Standard Classification of Occupations. International Labour Office. Geneva. 9. UNITEDNATIONS STATISTICAL Omce. 1958. International Standard Industrial Classification of all Economic Activities. United Nations Statistical Office.New York. 10. Socialstyrelsen. 1980. Cancerrisk och miljofaktorer. Rapport till Arbetarskyddsfonden. Stockholm, Sweden (Swedish only). 11. MALKER. H. S. R., J. K. MCLAUGHLIN, B. K. MALKER, B. J. STONE,J. A. WEINER, J. L. E. ERICSSON & W.J. BLOT.1985. Occupational risks for pleural mesothelioma in Sweden 1961-79. J. Natl. Cancer Inst. 74: 61-66.
-.
ANNALS NEW YORK ACADEMY OF SCIENCES
332
12. MALKER, H. S. R., J. MCLAUGHUN, W. J. BLOT,J. A. WEINER,B. MALKER, J. L. E. ERICSON &B. J. STONE. 1986.NasalcancerandoccupationinSweden1961-1979.Am.J.Epidemiol. R 477-485.
13. MAITSSON, B. 1984.Cancer registration in Sweden. Studies on completeness and validity of
incidence and mortality registers. Thesis. Karolinska Hospital. Stockholm, Sweden. 14. BRIVKALNE, M. 1964. The control study made in connection with the 1960 census of population. Statistical reports B. National Central Bureau of Statistics. Stockholm, Sweden (Swedish only). 15. WIKLUND, K. & G. EKLUND. 1986. Reliability of record linkage in the Swedish cancerenvironment register. Acta Radiol. UC 198-201. 16. AXELSON, 0.1986.Linking of register information on morbidity and occupation, especially with regard to quality aspects. 27-40.WHO ICP/OCH 002.Copenhagen. 17. MALKER,H. & J. A. WEINER.1984. The Cancer-Environment Registry 1961-1973. Examples of the use of register epidemiology in studies of the work environment. Arbete och Halsa, National Board of Occupational Safety and Health. Solna, Sweden (summary in English).
DISCUSSION
JOHN BAILAR(McCill University, Montreal, P.Q.): I think there are some even broader lessons that might be drawn here. One is the legitimate concern about reporting bias. I recall many years ago discussions about theincreasingratesrecorded for mesothelioma in the United States asked how much of this was real and whether all of the recorded increase might reflect an artefact. Theincrease was noted and was initially attributed just to changes in diagnosis; that is, bias. On the other side, we have the pressing need for hypothesis generation. But that also can fail sometimes. We have as evidence the frequency of reports of geographic clusters of cancer, virtually all of which on further investigation prove to be not real. Somehow we have to learn to draw a fine line between the need to avoid bias and over-detection of alot ofthingsthat aren’t thereand theneedtogenerateappropriate hypotheses from the clinical and registry experience. I don’t think we draw the line at the optimum place at present. There may be an over-concern with bias. We ought to be a bit more liberal about this. But we can’t go too far in either direction without getting in a lot of trouble. D. GOLDSTEIN (Robert WoodJohnsonMedical School, Piscataway, NJ): BERNARD I want to take issue with Dr. Malker’s statement. I have a minority view perhaps, but I think it is a misconception of the work that Selikoff has done. Selikoff has clearly shown us that if you get exposed to asbestos for afew years earlier in life then you’ve got perhaps 30 years of waiting for the potential lag time of the mesothelioma. Dr. Malker has told us that for nasal cancer and adenoid carcinomas, maybe you have not seen a decrease because there is some lag time involved from the problems that wood workers are being exposed to. I really take issue with that. I think we have to keep in mind that asbestos stays in the body during those many years. Most of the other chemical exposures are transient. I think that you’re much more likely to expect the kind of curve we see when you stop smokingcigarettes-avery quick decrease in risk. If you do not see that decrease in risk after the cessation of exposure and we’re studying something that remains in the body a long time, then we must consider that
