Data on gynaecologist visits, screening participation, cancer outcomes and abortions were collected by municipality in 5-year age groups. The number of visits to private gynaecologists, based on reimbursements, was obtained from the Social Insurance Institution of Finland for 1999. There is no formal study on the quality of the register, but it is believed to be relatively complete because it is based on money transfers. The number of visits to public gynaecologists in hospital outpatient clinics by municipality were obtained from a previous study in STAKES (Hospital Benchmarking) [8], based on the routine data collection of hospital districts. Visits due to pregnancy and delivery were excluded. These visits were distinguished by an administrative classification, because the registers do not included the reasons for visits
Data on the numbers of women invited and participating in organised cancer screening in 1998 were obtained from the files of the Mass Screening Registry at the Finnish Cancer Registry [9, 10]. Only those age groups in which screening was obligatory for the municipalities were included in this study. Because the data were organised in 5-year-age-groups, 59 years rather than 60 years was used for cervical cancer screening.
The numbers of women diagnosed in 1996–98 to have cervical cancer (C53 in ICD10), uterine cancer (C54), ovarian cancer (C56), and breast cancer (C50) were obtained by stage from the Finnish Cancer Registry. Hospitals, physicians and pathology laboratories are required by law to notify on all cases of cancer. The completeness and accuracy of the register is good [11]. The register classifies cancers into five categories according to stages at diagnosis: localised, regional metastasis, distant metastases, non-localised, NOS (not known whether regional or distant), no information of the stage.
Numbers of induced abortions by municipality, 5-year age groups, and social class in 1998 were obtained from the Finnish Abortion Register [12]. Social class was the one used in the Abortion Register, based on woman's own occupation: upper white collar, lower white collar, workers, enterprisers, students, other. In Finland, STAKES has to be notified of all induced abortions by law. The completeness and data quality in regard to variables used in this study is good [13]. The number of births was obtained from the Finnish birth register [12].
Denominators (the number of women in each age group, by social class in each municipality) were obtained from another project (Jansson, personal communication, STAKES, 2001). Originally the data are from the SOTKA 1999 database, describing the population in December 1998 [14]; SOTKA gets its data from the National Population Register and from Statistics Finland. The social class index was formed by dividing the number of women in the upper and lower white collar classes by the number of women in the class of workers in each age-municipality group; the higher the index, the higher the social class of the population group. The social class in SOTKA is based on various aspects of social class, but women's own occupation is the main determinant. Municipality urbanity (grouped into urban or suburban and rural) was obtained from Statistics Finland.
Analysis
In the first analysis approach, the municipalities were classified into three groups by the level of use of private gynaecologists: low, middle and high use. The age-adjusted (direct standardisation) rate of private gynaecologist use was calculated separately for each municipality, using the age-distribution of all municipalities as the reference. Municipalities with over 1000 women aged 20–64 years were divided into three groups, each having about same number of women. The limits for age-adjusted rates (visits per woman) turned to be < 0.242, 0.242–0.333 and > 0.333. Municipalities with less than 1000 women aged 20–64 years were then added to the groups having the corresponding rate of visits to the gynaecologist. Similar procedures were used to create three groups by the level of all gynaecologist visits, both private and public. The limits were < 0.579, 0.580–0.687, > 0.687. The women in each group were pooled over the municipalities to avoid the problem of small numbers in small municipalities.
Abortion rates were age- and social class adjusted by the direct standardisation method using the age and social class distributions of all municipalities as the reference. Social class was not available for women with cancer or cancer screening participation, and these outcomes were only age-adjusted.
The correlation between the size of the municipality and the use of private gynaecologist was calculated by Pearson correlation coefficient including all municipalities and by scattergrams between the two, separately for municipalities less then 5000 20–64 women (n = 397 municipalities), 5000–10000 (n = 30), and over 10000 (n = 25). The main results according to municipality size [<5000 (n = 397), 5000–30000 (n = 48) and over 30000 women (n = 7)] were calculated.
In the second analysis approach, each 5-year-interval age-group in each municipality was the unit of analysis. These units were classified into three groups, each having about an equal number of units, by the level of private gynaecologist use. Because different age groups were used for different outcomes, for some municipalities data were lacking and only units having outcome events were included, the number of women as well as the limits of the low, middle and high gynaecologist use vary (see table 3 in Results).
The statistical significance of differences between municipality groups was tested using tests for relative proportions and rate differences by z-test. Odds ratios were calculated by logistic regression and rate ratios by Poisson regression using the lowest private gynaecologist use group as the reference. Furthermore, age-adjusted (5-year interval groups) and age- and social-class-adjusted (index grouped into four classes and as a continuous variable within each 5-year age group) odds ratios were calculated. Similar analyses were made separately for urban/suburban and rural municipalities. SAS GENMOD procedure was used to fit the logistic regression model to allow events per trials responses and to fit Poisson regression model to compare rates.