The Health Survey in Scania 2000 (HSS-2000) was a postal self-administered questionnaire sent out to a random sample of 23 437 individuals born from 1919 to 1981 living in Scania. The purpose of the HSS-2000 was to obtain information about health conditions and different types of health hazards among the inhabitants of Scania 22. The province of Scania in southern Sweden has a population of about 1.2 million inhabitants and is divided into 33 municipalities. In total, 59% participated, of which 98% had complete information about medication use. The present study focused on those 9.6% who indicated use of antihypertensives during the last year and who had complete information about social participation (n = 1288).

The Ethical Committee at the Medical Faculty of Lund University approved the study proposal of The HSS-2000, and all of the participants received written information about the survey.

Use of antihypertensives was based on an affirmative answer to the question "Have you during the last year used medicine, which was bought at the pharmacy...?" and indicating "Medication for high blood pressure"

Low adherence with antihypertensives during the last two weeks (dichotomised) was based on the question "Have you used (this) medicine during the last year, but not during the last 2 weeks?" Those participants who answered yes were considered to have low adherence.

Age was categorised in five groups: <35 (reference), 35–44, 45–54, 55–64 and ≥ 65 years.

Low social participation (dichotomised) was assessed after the respondent stated involvement in three or fewer activities (lowest quartile) of 13 formal or informal activities (study circle/course at work place, other study circle/course, union meeting, meeting of other organisations, theatre/cinema, arts exhibition, church, sports event, letter to editor of a newspaper/journal, demonstration, night club/entertainment, large gathering of relatives, private party), which the respondent might have participated in during the previous 12 months 23.

Low educational level (dichotomised) was defined as having nine years of education or less.

Poor self-rated health (dichotomised) was defined as a value of ≤ 3 on an ordinal self-rated health scale ranging from 1 ("Very bad") to 7 ("Very good") 24.

Poor psychological health (dichotomised) was determined by giving three or more affirmative answers to the 12 items composing the Standardised General Health Questionnaire (GHQ-12) 25.

Because of the hierarchy in the data, with individuals nested in municipalities, we used multilevel logistic regression 26 with individuals at the first level and municipalities at the second level. The area of residence might affect a person's social participation 27, and, consequently, it may be possible that the influence of low social participation on low adherence with antihypertensives during the last two weeks may vary between municipalities. Therefore, we let the slopes of the association between low social participation and low adherence vary at the municipality level. This random slopes analysis gives information about whether the association between low social participation and low adherence is different in different municipalities.

The first model i was created to study the influence of low social participation on low adherence with antihypertensives during the last two weeks, adjusting for age and sex. The second model ii was extended to also include low educational level, because low educational level could be a confounder in the association between low social participation and low adherence with antihypertensives during the last two weeks. The third model iii additionally contained poor self-rated health and poor psychological health. Impaired health may affect both low social participation and low adherence with antihypertensives.

Participants with low social participation had on average a more than twofold higher probability of reporting low adherence with antihypertensives during the last two weeks than those who did not have low social participation (OR = 2.28, 95% CI: 1.16–4.49) (Table 2). This association between low social participation and low adherence with antihypertensives during the last two weeks persisted after adjusting for low educational level (OR = 2.05, 95% CI: 1.05–3.99). However, after additional adjustment for poor self-rated health and poor psychological health, the association between low social participation and low adherence with antihypertensives during the last two weeks was not conclusive using a 95% confidence interval (OR = 1.80, 95% CI: 0.90–3.61).

We found a variance between the municipalities in both low adherence with antihypertensives during the last two weeks (intercept variance) and in the association between low social participation and low adherence with antihypertensives during the last two weeks (slope variance) (Table 2 and Figure 1). The negative covariance between intercepts and slopes (Table 2) suggested that the associations between low social participation and low adherence with antihypertensives during the last two weeks (slopes) in the 33 municipalities depended on different prevalence of low adherence with antihypertensives during the last two weeks in the different municipalities. The association between low social participation and low adherence with antihypertensives during the last two weeks (slope) was weaker in municipalities with higher prevalence than in municipalities with lower prevalence of low adherence with antihypertensives during the last two weeks (i.e., cross-level interaction).

Our results suggest that low social participation is associated with low adherence with antihypertensives during the last two weeks, independently of low educational level. In other words, the association between low social participation and low adherence withstood adjustment for socio-economic position (expressed by educational level) in our analyses. Social participation might therefore be considered as a real construct, and not only a proxy for socio-economic position. However, the association between low social participation and low adherence with antihypertensives during the last two weeks was weakened after additional adjustment for poor self-rated health and poor psychological health. Furthermore, the association between low social participation and low adherence with antihypertensives during the last two weeks may vary between municipalities in Scania.

The weakening of the association between low social participation and low adherence with antihypertensives during the last two weeks after we adjusted for poor self-rated health and poor psychological health may be an expression for confounding. Impaired health may negatively affect both social participation and adherence with antihypertensives. On the other hand, the observed reduction of the association may instead be telling us that physical and mental health are in the pathway between low social participation and low adherence with antihypertensives during the last two weeks 29. Social participation may be considered as an early factor in the causal pathway that determines individual health-related behaviour, such as low adherence with medication. Social networks, which are connected to social participation, may promote shared norms around health behaviours, as treatment adherence, which could explain the pathway between social networks and impaired health 16.

Our present finding that the association between low social participation and low adherence with antihypertensives during the last two weeks may vary between municipalities in Scania gives empirical support to the existence of cross-level interactions (i.e., between municipality and individual) associated with health-related behaviours, such as low adherence with medication. These behaviours may be a result of the interaction between a person and his or her area of residence 18. In a previous analysis, we observed that both individual and neighbourhood social participation are associated with individual impaired health and with use of hormone replacement therapy in women 18.

The present study shows that multilevel regression analysis can be used for investigation of geographical disparities in health and health-related behaviour (e.g, adherence with medication), without analysing any specific area characteristic 30. In multilevel analysis, area effects can be investigated by measures of variance and by examining how area boundaries modify individual level associations 31.

The rather low participation rate (59%) may increase the risk of selection bias and reduce the ability to generalise the results and compare them to other populations. Nevertheless, the participation rate for participants aged 51–80 was about 65% and the participants using antihypertensives had a mean age of 63 years.

The low adherence question we used was stated in a non-threatening manner, which might facilitate for participants to give an honest response and not underreport low adherence 32. Self-reported adherence has been reported to correlate with clinical measures of disease activity and control 4. Moreover, self-report offers a convenient and non-invasive estimate of adherence behaviour. Nevertheless, the procedure of measuring adherence is controversial. Self-report can be subject to self-presentational and recall biases. People may overestimate their adherence and their memory may be inaccurate 32. We might have reduced memory bias in this study by restricting the recall time to two weeks. However, the prevalence of current low adherence (11%) in this study is lower than low adherence reported in a longer period of time, which may be as high as 50% 14. Therefore, our results may underestimate the association between social participation and low adherence with antihypertensives. It is possible that some participants with high adherence in the last two weeks had low adherence in other periods of the year. If this kind of misclassification would be more frequent among participants with low social participation, there would be differential misclassification, and the association between low social participation and adherence with antihypertensives could be underestimated. Non-differential misclassification would also underestimate the association between low social participation and low adherence with antihypertensives. Other ways of measuring adherence may be more appropriate, such as Morisky's four-item scale 33, which will be used in the Health Survey for Scania 2004.

People with low social participation and low adherence with antihypertensives during the last two weeks may have been less inclined to respond to the HSS-2000 questionnaire. This possible selection bias could lead to an underestimation of the association between low social participation and low adherence with antihypertensives during the last two weeks.