Background The aim of the study was to identify demographic and socioeconomic characteristics of participants and non-participants in a Swedish population-based case-control study on brain tumours and to analyse the association between socioeconomic factors and glioma and meningioma risk.
Methods Record linkage was made to an official register to gather information on socioeconomic status, income, education and demography for all participating and non-participating cases and controls.
Results 494 glioma cases, 321 meningioma cases and 955 controls were eligible and 74%, 85% and 70%, respectively, participated. Working status and income level were positively associated with participation among cases and controls. Among both cases and controls, being married, and having a high education were also associated with participation. Having a family income level in the highest quartile was associated with an increased glioma risk (OR 1.5, 95% CI 1.1 to 2.1). This risk increase diminished when only participating individuals were included in the analysis. Socioeconomic factors were not associated with meningioma risk.
Conclusions Non-participation, related to socioeconomic factors, is a potential source of bias in case-control studies that should be acknowledged; however, the effect was not large in the present study due to the fact that the level of participation was comparable between cases and controls and participation was similarly influenced by socioeconomic factors among cases and controls. The association between a high income level and an increased glioma risk and possible underlying factors needs to be explored further.
- selection bias
- socioeconomic factors
- cancer epidemiology
- case control me
Statistics from Altmetric.com
- selection bias
- socioeconomic factors
- cancer epidemiology
- case control me
A decline in participation in epidemiological studies has been noted in recent years.1 2 Low participation reduces precision, and may introduce bias and thereby compromise validity. A low participation does not, however, necessarily mean that results are biased. To cause a bias, non-participation must be related to both exposure and outcome (case-control status) either directly or indirectly, for example through socioeconomic status.
Characteristics of non-participants have previously been described, mostly in cross-sectional studies (surveys), but also in cohort studies and a few case-control studies. The knowledge of non-participation characteristics among cases is, therefore, limited. Non-participation has been reported related to young and/or old age,3–10 male sex,3 5 9–11 being single,5 7 9 12 13 and to low educational level, low socioeconomic status or low income.4–7 9 10 12–16
Previously published studies on socioeconomic factors and glioma risk have reported an increased risk associated with higher socioeconomic status, which is in contrast to studies of meningioma where results are conflicting.17–21 Interpretation of the reported associations and possible underlying factors has been difficult, as issues such as non-participation possibly related to socioeconomic status, study design (type of control group), availability of care and health-seeking patterns might have influenced the observed associations.
The aim of the present study was to compare demographic and socioeconomic characteristics of participants and non-participants in a Swedish population-based case-control study of brain tumours. The mode of participation and reason for non-participation are described in relation to socioeconomic factors. Record linkage to an official registry was used to gather information on socioeconomic factors for participants and non-participants. The possible influence of non-participation on the association between socioeconomic factors and glioma and meningioma risk was also analysed.
Study population and period
As part of the previously described INTERPHONE study,22 a population-based case-control study on brain tumours was conducted in Sweden from September 2000 to August 2002. The study population consisted of all individuals aged 20–69 years in the geographical areas covered by the regional cancer registers in Stockholm, Göteborg, Umeå and Lund.
Individuals who did not possess intellectual or language skills necessary to complete an interview or who were completely deaf prior to the disease were excluded.
Eligible cases were all individuals diagnosed during the study period with intracranial meningioma (International Classification of Diseases tenth revision (ICD-10), code D32, C70; International Classification of Diseases for Oncology second edition (ICD-O-2), code 9530-9539) or intracranial glioma (ICD-10, code C71; ICD-O-2, code 9380-9384, 9390-9394, 9400-9401, 9410-9411, 9420-9424, 9430, 9440-9443, 9450-9451, 9460, 9480-9481, 9505). Cases were identified continuously during the study period through collaboration with all departments of neurosurgery, oncology and neurology within the study areas. The regional cancer registries were searched approximately every third month for additional case identification.
Approval was sought from the attending physician to approach each case. If a case had died or was too sick to participate the closest relative was contacted as a proxy-respondent where possible.
Controls were randomly selected from the study population and frequency matched on sex, age (in 5-year groups) and residential region. They were selected continuously throughout the study period from the registry of the Swedish population.
Interviews and questionnaires
All interviews and contacts were made by interviewers employed for this purpose. All cases and controls were contacted as soon as possible after identification. Each case and control received an invitation letter and a factsheet with information about the study. Some cases were contacted at the treating clinic, but normally the first contact was made by mail. Cases and controls then were approached over the phone to arrange a time for a personal interview. The interviewers had no restrictions in their working hours and the time for an interview was thereby chosen when convenient for the cases and controls. Individuals who did not want to participate in a personal interview were offered a telephone interview instead. Those who did not want to participate in any kind of interview were asked if they could fill out a questionnaire with selected parts of the interview questions.
If no contact could be made by phone a new letter was sent 2–4 weeks after the first letter, asking for additional telephone numbers and the best time for contact. If still no contact was made an additional letter was sent 6–8 weeks after the first invitation. This letter included also the written questionnaire.
Individuals who were interviewed or answered the questionnaire were defined as participants, those who did not as non-participants. Participants were further subdivided into personal interview, telephone interview and written questionnaire participants. Non-participants were subdivided depending on reason for non-participation: those who refused to participate (‘refusals’), those with whom no contact was established (‘failure of contact’) and a third group who were too sick to participate or diseased and had no suitable proxy to contact. Finally, among the cases there was a group whom the attending physician did not give permission to contact.
Through the personal identification number assigned to every Swedish resident a linkage was made to a database at Statistics Sweden, covering the entire Swedish population. Data from the years 1995 and 2000 on marital status, highest attained educational level, disposable income and working status were linked to participants and non-participants. Information on socioeconomic status, based on occupation and type of employment among the working population in the census of 1990, was also attached.
The register linkage was made at Statistics Sweden. Before the dataset was sent to Karolinska Institutet for analyses the personal identification numbers were deleted.
ORs and 95% CIs were estimated for cases and controls separately using unconditional logistic regression as a measure of association between demographic and socioeconomic factors and participation.
Socioeconomic differences among the controls for mode of participation among participants and for reason for not participating among non-participants were estimated in contingency tables by χ2 tests, or Fisher's exact test. Among glioma and meningioma cases the numbers of phone interviews and written questionnaires were not large enough to allow separate analyses.
Glioma and meningioma risks associated with socioeconomic factors were estimated by unconditional logistic regression with adjustments for sex, age and region. Income categories were chosen based on the distribution among controls.
All analyses were performed using SAS statistical software version 9.1.23
During the study period 494 eligible glioma cases, 321 meningioma cases and 955 controls were identified. Three-hundred and sixty-six (74%) glioma cases, 274 (85%) meningioma cases and 673 (70%) controls participated in an interview or filled out a written questionnaire. Nine per cent of glioma interviews and 3% of meningioma interviews were performed with proxy respondents. Among non-responding controls 59% refused to participate, 36% were in the ‘failure of contact’ group and 5% were too sick to participate. Refusal was also the most common reason for non-participation among meningioma cases (45%), but only 21% of non-responding glioma cases refused to participate. Nine per cent of glioma and meningioma cases were in the ‘failure of contact’ group (table 1).
Information from the record linkage was completely missing for two glioma cases (0.4%) and 10 controls (1%) due to incorrectly registered personal identification numbers. Data on socioeconomic status were missing for 18% of glioma cases, 15% of meningioma cases and 21% of controls, as no information was available for individuals not working in 1990, or who were not included in the census, or did not answer the questions on type of occupation and employment in the census. For the other registry-based variables, data were missing for 3% or fewer of eligible individuals.
Sex and age did not influence participation among cases or controls. Being married was associated with participation among controls (OR 1.7, 95% CI 1.2 to 2.3) compared with being single. Any educational level beyond compulsory schooling increased the odds of participation among controls and a similar tendency was also seen among the cases. Being a non-manual employee was associated with increased participation compared with being a manual worker (OR 1.8, 95% CI 1.3 to 2.6) among controls. Working status and level of disposable income were positively associated with participation among glioma and meningioma cases and controls (table 2). In a multivariate model, adjustment for all included variables did not change the results although this created wider CIs, except for the result for non-manual workers among controls where the OR changed to 1.1 (95% CI 0.7 to 1.7) (not shown in table).
Analyses of the mode of participation among participating controls showed that proportionally more women were interviewed over the telephone or answered the questionnaire. Telephone interviews were more common among divorced and widowed individuals and less common among married participants, p=0.007. Seventy-four per cent of those personally interviewed were working, compared with 62% of those who answered the written questionnaire and 50% of those interviewed over the phone, p=0.003. Having a disposable income in the highest quartile was more common among those who had a personal interview compared with telephone interview or questionnaire (table 3).
Evaluation of reasons for non-participation among controls only included the ‘refusal’ group and the ‘failure of contact’ group; only 14 controls did not participate because they were too sick. Young age, being single or divorced, and having a low level of income were more common in the ‘failure of contact’ group (table 3). The number of non-responding meningioma cases was low (47), and in analyses of reasons for non-participation no relation could be seen with demographic or socioeconomic factors. For glioma cases the only association found with reason for non-participation was age. Old age was more common among those who were too sick to participate and in the group where the attending physician did not give permission to contact the patient compared with the ‘refusal’ group, p=0.003 (not shown in table).
Socioeconomic factors and glioma and meningioma risk
The individual level of disposable income (year 2000) was associated with a slightly increased glioma risk for the highest quartile of income compared with the lowest in the overall result including all eligible individuals (OR 1.3, 95% CI 1.0 to 1.8), (table 4).
The disposable income of the family was associated with the risk of glioma analysing all eligible individuals (OR 1.5, 95% CI 1.1 to 2.1) comparing the highest quartile of income with the lowest. Analyses restricted to participants only (regardless of participation mode) gave an OR 1.2 (95% CI 0.8 to 1.8) for the highest income level. Including only individuals participating in the personal interviews changed the result to OR 1.1 (95% CI 0.7 to 1.6). Adjustment for marital status in analyses of family income did not change the results. Using data from year 1995 did not change the results substantially (results not shown).
Socioeconomic factors were not associated with meningioma risk, and analyses restricted to participants only or personally interviewed only marginally changed the results (table 5).
Non-participation can bias results and may be of special concern in case-control studies. The difference in health status can be suspected to influence the willingness to participate and the reason for non-participation, where socioeconomic factors are assumed to be more influential for controls than for cases. The present analyses of participants and non-participants in a Swedish population-based brain tumour study, however, showed participation to be similarly associated with a high income level and working status for cases and controls and the impact of selection bias is thereby reduced.
The strengths of the present study are the population-based study design and the ability to crosslink the dataset to a registry of socioeconomic factors for all eligible individuals. It was not necessary to rely on extra interviews of subsets of non-participants to be able to characterise this group, which is common in other studies.4 13–15
Comparing results from different studies of factors influencing non-participation can be difficult as the type of study, mode of participation, exposures of interest and geographical setting might influence the results. Many studies have reported non-participation to be related to young and/or old age,3–10 male sex,3 5 9–11 being single,5 7 9 12 13 low educational level, low socioeconomic status or low income.4–7 9 10 12–16 The results in the present study were similar, although sex and age were not found to be associated with participation.
Working status and income level were equally related to participation among cases and controls. Selection bias might still exist if the level of participation differs between cases and controls. The level of participation was 85% for meningioma cases, 74% for glioma cases and 70% for controls, so the differences were not large. When participation related to socioeconomic factors and participation rates are comparable among cases and controls non-participation only causes loss in precision. Other studies that have analysed characteristics of non-participants separately among cases and controls have also found socioeconomic status to be associated with participation in both cases and controls.10 14 15 Richiardi et al found that socioeconomic level of non-participants differed between cases and controls in a study of lung cancer.7 Although the reasons for non-participation differed in the present study between cases and controls, cases might sometimes report being too sick to participate as an indirect refusal that may be influenced by socioeconomic factors in the same manner as direct refusal.
The primary mode of data collection in the present study was personal interviews, but a telephone interview or a questionnaire was offered if the individual did not want to participate in a face-to-face interview. Among the controls the participation mode differed according to sex, working status, marital status and income. Non-married and not working controls participated to a higher extent in telephone interviews. Having a disposable income in the highest quartile was more common among those participating in personal interviews. Although the information collected with personal interviews, telephone interviews or self-assessed questionnaires may not be of the same quality, the use of different modes for collection of data may decrease the risk for selection bias, and should be considered when planning a study.
Non-participation reasons among controls were influenced by demographic and socioeconomic factors. Failure of contact was more common with individuals, who were younger, single and had a lower income level compared with those who refused to participate, and thus the ‘failure of contact’ group differed more from the participants than the ‘refusal’ group. Among glioma cases it was the older patients who were too sick to participate. This should not influence the results if adjustments for age are performed, but might compromise the possibility of identifying an association if the exposure under study is associated with the severity of the disease.
In studies of associations between socioeconomic factors and risk of disease it is essential for the interpretation of results to consider the possible influence of non-participation as it might be related to socioeconomic factors. The present study provided an opportunity to analyse this association with and without non-participants. Including all eligible cases and controls, an association was found between a high level of disposable income and an increased glioma risk. Non-participation caused these associations to disappear. Estimates of glioma risk and income level could have been underestimated if cases had had their income reduced due to early symptoms of disease. Analysing income data from 1995 (5 years or more before diagnosis) did not, however, change the results.
Previously published studies on socioeconomic factors and glioma risk have like this study found positive associations with a high socioeconomic level,17–21 although some have reported the opposite.24–26 In studies of malignant brain tumours or tumours of the brain and nervous system, the majority of which are gliomas, associations with higher socioeconomic status have been found.27–34
Even though most studies have found associations between glioma risk and high socioeconomic status, results can be questioned. In hospital-based studies results can be biased if the diagnoses of hospital controls are related to socioeconomic factors. Non-participation in population-based studies might bias results if non-participation is jointly related to socioeconomic and case-control status. Studies using data from official cancer registers, comparing incidence in different socioeconomic strata can be biased if the propensity to seek care and get a correct diagnosis is related to socioeconomic factors.
In a study by Inskip et al, where glioma was divided into high-grade and low-grade glioma, positive associations were seen with household income and education for low-grade glioma but not for high-grade, and they argue that this can be due to difference and timing of diagnosis in certain subgroups of the population and that this influence is largest for slowly growing tumours.20 Chakrabati et al, on the other hand, looking at glioblastoma multiforme (high grade glioma) found a positive association with socioeconomic status.21 The healthcare system in Sweden is accessible to all, even though the propensity to seek care has been shown to differ among groups.35 It is, however, unlikely that there should be large differences in healthcare utilisation among different socioeconomic groups for symptoms associated with brain tumours.
The only unequivocally identified risk factor for glioma is ionising radiation,36 37 but it is not likely that there should be differential use of radiation treatment in Sweden related to income being able to explain found associations, and such relation would also be expected to increase meningioma risk as ionising radiation also is a risk factor for meningioma.38
An association was found between a high level of income and an increased glioma risk. Other studies have found association with place of residence,17 21 income and education,18 20 and occupational sectors.19 The underlying risk factors explaining these associations are unknown, but can be expected to be related to many aspects of lifestyle and to occupational exposures.
Participation in the present study was similarly associated with socioeconomic factors among cases and controls. Non-participation associated with socioeconomic factors is a potential source of bias in case-control studies; however, the estimated influence was not extensive in the present study, but may still influence the possibility to find small risk increases. An association was observed between having a high income level and an increased glioma risk. Possible underlying factors need to be further explored.
What is already known on this subject
Non-participation associated with socioeconomic factors is a potential source of bias in epidemiological studies.
Glioma risk has in previously published studies been related to a high socioeconomic status, but the results have been questioned due to choice of control group or low participation.
What this study adds
The present study uses a population-based design with sociodemographic information from registries on all eligible cases and controls, and interview data from participating subjects.
Participation was positively associated with income level and working status among both cases and controls.
An increased risk of glioma associated with a high income level was found in the complete study population, but not when analyses were restricted to interviewed subjects.
Funding This work was supported by the European Union Fifth Framework Program, ‘Quality of Life and Management of living Resources’ (contract QLK4-CT-1999-01563), International Union against Cancer (UICC), the Swedish Research Council, and the Swedish Cancer Society. The UICC received funds for this purpose from the Mobile Manufacturers' Forum and GSM Association. Provision of funds to the INTERPHONE study investigators via the UICC was governed by agreements that guaranteed INTERPHONE's complete scientific independence. These agreements are publicly available at http://www.iarc.fr/ENG/Units/RCAd.html.
Competing interests None.
Ethics approval The study was approved by the ethics committee at Karolinska Institutet in Stockholm.
Provenance and peer review Not commissioned; externally peer reviewed.
If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.