Ethics statement

The Ethics Committee of the Finnish Institute of Occupational Health approved the study.

Study design

We used data from the Finnish Public Sector study, which includes the entire personnel in the municipal services of 10 towns and within 6 hospital districts in Finland [30]. For this study, we included 66,418 employees who responded to a survey of workplace social capital in 2000–2002 and/or 2004 (a total of 79% of the eligible employees responded at least once). In case of repeat responses, the more recent one was used in the analysis. Personal identification numbers were used to link all study participants to national prescription and health registers from 1994 through 2005.

Of the respondents, we identified those who had physician-verified hypertension during the survey year, and who were alive at least one year after the survey year (total n = 4538). We excluded all 896 employees of one town for whom prescription data were not available because their medication costs were covered by the employer, 120 participants due to missing data on workplace social capital, and 7 participants who had died before the beginning of the 1-year follow-up which ran from January 1 following the survey year. Thus, the final analytic sample comprised 3515 hypertensive employees (76% women, mean age 53.9 years, range 22–66).

Case ascertainment for hypertension

Hypertension and its duration were defined based on entitlement to special reimbursement. In Finland, the National Health Insurance administered by the Social Insurance Institution (SII) covers all permanent residents regardless of age, wealth, or employment status. Residents with chronic and severe hypertension are eligible for lower copayments for antihypertensive drugs. Currently, 72% of the costs of medicines are reimbursed after a fixed deductible by purchase with annual maximum copayments of around 600 Euros, provided the person is listed in the Drug Reimbursement Register maintained by the SII. In order to receive special reimbursement, the patient must submit a doctor's certificate to SII stating the illness, and the medication needed to treat it. For hypertension, the current criteria are stricter than the guidelines for treatment of hypertension [11], and include documentation of repeated blood pressure measurements of ≥105 mmHg diastolic, or ≥95 mmHg diastolic with signs of complications or other cardiovascular risk factors, or ≥200 mmHg systolic. Special reimbursement can be granted only after 6 months of surveillance with lifestyle counselling followed by 6 months of drug therapy.

Workplace social capital

Workplace social capital was assessed with an 8-item self-assessment scale designed to measure social capital in the workplace [31]. Using a 5-point Likert-scale, the participants assessed workplace social capital, defined as the shared values, attitudes, and norms of trust and reciprocity as well as practices of collective action in their work unit [31]. As previously [32], [33], we assessed social capital in two alternative ways: (a) using each individual's own assessment, and to minimise subjectivity bias, (b) summing up the assessment of co-workers, but excluding the individual's own assessment. We calculated the mean of the individual items to construct a scale ranging from 1 to 5 with higher score indicating high social capital. Prior psychometric evaluation in the Finnish Public Sector Study has demonstrated the scale to have high internal consistency (Cronbach's alpha 0.88) and r_wg index (0.88) indicating significant within-unit agreement [31].

Assessment of adherence to antihypertensive medication

Adherence to antihypertensive medication was assessed from filled prescriptions for antihypertensive drugs recorded in the Drug Prescription Register. The automated pharmacy dispensing data are operated under a single system and cover virtually all filled out-patient prescriptions from pharmacies in Finland. Pharmacy refill data of antihypertensive drugs were extracted for the year following the survey year, and included a listing of all antihypertensive prescriptions filled, the dispensing date, the quantity dispensed in Defined Daily Dosages (DDDs), and the drug class according to the WHO Anatomical Therapeutic Chemical (ATC) classification for codes C02 (antihypertensives), C03 (diuretics), C07 (beta blockers), C08 (calcium channel blockers), and C09 (ACE inhibitors). DDD is a globally unified measurement unit, defined as the assumed average maintenance dose per day for a drug used for its main indication in adults [34].

We used a refill-sequence model of adherence to quantify the duration of treatment [35]. Thus, we measured the total duration of consecutive refills, using 100 DDDs as a maximum for a refill (up to 3 months supply of medications can be reimbursed during a single transaction in the Finnish reimbursement legislation) and added 50 days to all refills to prospectively fill the gaps between subsequent purchases [35], [36]. This grace period for the gap in medication implies that the patients treated with half of the defined daily dose are correctly assigned as having 100 days of treatment. We then calculated (with this model) the length of the gaps, i.e. days-not-treated, between subsequent prescriptions in the sequence of all refill dates, assuming that an individual has not discontinued drug therapy at a given point in time if his/her dispensed supply from the last redemption has not elapsed or if it elapsed within the last 50 days. We excluded periods of inpatient hospital admissions during which the medication is provided by the hospital. The outcome, days-not-treated, indicated non-adherence to drug therapy and was treated as a continuous variable in the analyses [37], [38].

Covariates

Participants' age, sex, socioeconomic status (SES), type of job contract (permanent vs. temporary), type of employer (municipality vs. hospital), and geographic area (Southern, Middle or Northern Finland, based on the location of the workplace) were derived from employers' registers. SES was divided into three categories according to the occupational classification of Statistics Finland, i.e. upper grade non-manual workers (e.g. head nurses, teachers), lower grade non-manual workers (e.g. practical nurses, technicians), and manual workers (e.g. cleaners, maintenance workers). Marital status (married or cohabiting vs. other) was derived from survey responses. Standard questionnaires were used to assess behaviour-related risk factors: obesity (body mass index≥30, calculated as weight in kilograms divided by height in meters squared), smoking (current vs. ex- or never), excess alcohol use (>210g/wk) [39], and physical inactivity (<2 MET-hours/wk, MET = metabolic equivalent tasks) [40].

The presence of co-morbid physical illnesses was based on documentation of the condition in the SII, the National Hospital Discharge Register, or the Finnish Centre for Pension for entitlement to special reimbursement for the costs of medication, sickness absence or disability pension or the main diagnoses of hospital discharge. These conditions included diabetes mellitus, heart failure, coronary heart disease, cardiac arrhythmia, and cerebrovascular diseases. Notifications of diagnosed cancer were obtained from the Finnish Cancer Registry which compiles all cancer notifications in Finland. Co-morbid conditions were encoded as ‘yes’ if any data on pre-existing disease was found during four years preceding the follow-up. Co-morbid depression was identified based on documentation of any of the following: 1) long term (>90 days) work disability (sickness absence or disability pension), 2) hospital admission due to depression, 3) pharmacy records of filled prescriptions of at least 30 defined daily dosages of antidepressants (ATC code N06) during any of the four years preceding the follow-up, or 4) if the participant responded as being diagnosed with depression in the survey.

Statistical analysis

As of April 2003, pharmacy personnel have been allowed to switch to generic drug formulations to lower total prescription drug costs in Finland, additionally leading to smaller copayments to individuals, factor suggested to influence adherence [7]. We took this into account in the analyses by adjusting the models for the survey year. We examined the age- and survey-year-adjusted differences in rates of days-not-treated as a function of the cohort characteristics using analysis of variance. We calculated rate ratios (RR) and their 95% confidence intervals (CI) for the days-not-treated using negative binomial regression models to take into account the skewness of the outcome [41], allowing for overdispersion. We conducted the analyses separately for self-reported and co-worker-assessed social capital, adjusting for age, survey year, sex, socioeconomic position, marital status, type of job contract, type of employer, area of residence, duration of hypertension, behaviour-related risk factors, and co-morbid physical illness and depression at baseline. We tested whether sex or SES modified the association by adding the interaction terms sex*workplace social capital and SES*workplace social capital in the model of main effects, but we did not find any evidence of an effect modification (all p-values >0.05). Potential predictors of non-adherence were considered statistically significant at 95% confidence interval level (two-sided p<0.05).

Sensitivity analyses

We conducted three sets of sensitivity analyses. First, as 80% of days treated is commonly used as a cut-point in adherence studies [3], [7], we tested the robustness of the findings by dichotomizing the outcome as less than 80% of days treated versus other. Using this binary outcome, we then conducted logistic regression models to calculate odds ratios (OR) for the risk of non-adherence associated with self-assessed and co-workers' assessment of workplace social capital. Second, because the effects of workplace social capital on health may vary by dimension [42], we examined the vertical (i.e. employees' relations with their employers and supervisors) and horizontal (i.e. social contacts, cooperation and trust in relation to co-workers) components of workplace social capital separately. We determined the vertical or horizontal component of workplace social capital as in our previous study [43], divided the co-worker assessed and self-reported ratings into quartiles, and examined whether they were associated with non-adherence with negative binomial regression models. Third, we restricted the sample to 907 (77% women) new users of antihypertensive medication because long-term users may differ from new users in terms of baseline risk; there is also evidence to suggest that adherence reduces over time [4], [6]. We traced those who were newly granted special reimbursement for hypertension after responding to a survey in 2000–02 or 2004, followed their adherence to antihypertensive medication during the first year of treatment starting from the date of granting special reimbursement, and repeated the main analysis using negative binomial regression models adjusting for age and year of treatment, and then for all covariates.

All statistical analyses were conducted with SAS 9.13 statistical software (SAS Institute, Inc., Cary, NC, USA).

Strengths and limitations

We employed data from a large cohort of workers with physician-verified hypertension and need for continuous antihypertensive medication, successfully linked to a comprehensive population level prescription register to enable a 12-month prospective follow-up. All participants were covered by the Finnish National Health Insurance, thus avoiding selection bias due to insurance coverage or varying time of insurance coverage which is possible in commercial health insurance plans. We investigated both prevalent and new users because barriers to good adherence may differ for newly-diagnosed patients versus those with long-standing treatment for hypertension [6], [7]. We used a psychometrically validated multi-item measure to assess workplace social capital. We measured medication adherence objectively from pharmacy refill data. Rates of refilled prescriptions are considered valid as measures of gaps in medication supply within a closed pharmacy system provided that the refills are measured at several points in time [3], [49].

However, the results need to be interpreted in the light of several limitations. First, pharmacy dispensing records provide an indirect measure of adherence and we could not ascertain whether the medication was actually taken. Second, the participants were covered by the Finnish National Health Insurance which may compromise generalisability of the findings to uninsured workers. Third, qualifying for special reimbursement required blood pressure levels of Stage II hypertension or Stage I with other cardiovascular risk factors or co-morbidities [11]. Hence, we do not know whether workplace social capital contributes to adherence in milder cases of hypertension. Fourth, we assessed workplace social capital at baseline only, and we did not assess changes in social capital in relation to changes in adherence behaviour. Fifth, when prescription registries are used to define episodes of medication use, apparent gaps in treatment can occur. By filling the gaps, we avoided bias from embedded gaps due to a lower dose than DDD or tablet splitting; however, we could still have misclassified participants with terminal gaps, i.e. discontinuing the medication without finishing the supply [50]. If the 50 days we used for filling the gaps is too long a period, we have counted those non-adherent as adherent; conversely, if the period is too short, we have missed individuals who continued taking the medication, i.e. who were adherent. Non-differential misclassification of the outcome, which can be assumed to be the case with the Prescription Register, is likely to bias the measures of associations towards the null. Finally, there may be a tendency of those employees whose adherence is poor to develop complications of high blood pressure and to leave employment [10], [50]. However, our follow-up covered both employed and unemployed population, reducing the healthy worker effect related to occupational cohorts [51].

Conclusions

In this paper, we report null results of the relationship between workplace social capital and adherence to antihypertensive medication. Reporting null results may provide balance for the social capital research area not to produce publication bias and not to show social capital in a too favourable light. We wish not to guard research away from studying social capital in workplaces, but instead, encourage comparative work. To our knowledge, this is the first and only investigation on this topic in the work context to date. As the meaning of workplace social capital may differ by cultural background or branch of industries, further studies in other cohorts and contexts are needed to refute the overall hypothesis that social capital is associated with adherence to antihypertensive medication.