Main Findings
We have demonstrated an approach to monitoring change in population mental health using antidepressant prescriptions. We found the strength of the association between demographic, living arrangement, socioeconomic factors and antidepressant use varies by the pattern of prescription use over a six-year period. Whilst some factors such as sex, social grade and neighbourhood deprivation have similar associations with all the prescription patterns; age, living alone, ethnicity, marital status and employment status had unique relationships with the antidepressant prescription trend groups. Young, white, female participants, of low social grade, living in deprived neighbourhoods, living alone, being separated/divorced or out of the labour force, were more likely to have started using antidepressants during the study period. This may be because of the social and economic climate at this time, which included a range of austerity-related measures that had a disproportionate impact on these women. Research has estimated that 85% of tax and benefit changes have impacted on women’s incomes – particularly on low-income women living in deprived areas [26]. Further, public sector service cuts (such as to libraries, children’s centres, community centres, advice services etc.) have adversely affected women – because women are higher users of these local resources and are more likely to be employed in the public and voluntary sector than men are (ibid.). They are also more likely to be engaged in low-wage work, and have more sustained engagement with the benefit system. This has led to discussions about how women have experienced a ‘triple jeopardy’ of public sector service reductions, job losses and welfare reform [27].
Strengths
The use of six years of monthly prescription data in a large representative sample of the Scottish population is a major strength of the study. The declining stigma surrounding treatment and increasing awareness of mental health may strengthen this indicator of mental illness in the future [28]. The auxiliary measures of mental illness history (admissions) and subjective mental illness status are also major strengths. This is one of the few large studies that has linked multiple indicators of mental illness, which is key to understanding a complete picture of population mental health [29]. Administrative data does not suffer from loss to follow up to the same extent as cohort studies, although there is some attrition due to death and migration. We have applied a novel technique to classify six-month prescription status. Previously this method has had limited application in health studies, with patterns of health care access a recent exception [30]. This technique has the advantage of uncovering the complexity of being prescribed medication and the pattern of relapse or remission that corresponds to disease progression, which is obscured by more commonly, applied dichotomous measures.
Weaknesses
The prescriptions data were not available before 2009, so we were unable to compare with usage during this time, which might have meant that some of the participants classified as having become prescribed had been prescribed previously. We did not have the medical diagnosis so a small number of the prescriptions may have been used to treat conditions other than depression. We also assumed that treatment dispensing was synonymous with antidepressant use, but we did not have information on actual consumption to verify this. Covariate information was limited to measures collected in the census, therefore we were unable to understand effect of other lifetime factors [31] shown to be important for predicting antidepressant prescriptions (e.g. tobacco consumption ). Although mental health service use stigma may be declining, it still exists, the current results are limited in that they underestimate certain populations at risk (e.g. young males) [32].
Comparison with existing literature
Similar methods have been used in previous studies to classify trajectories of annual antidepressant dose over time [33]. The author’s of this study use latent class model, which is shown provide similar groupings to the sequence analysis used in the current study [34]. The sample used in their study was very specific – patients before and after being granted disability pension due to common mental disorders [33]. These individuals would have formed part of our ‘out of labour force’ group, which had the strongest relationship (OR 4.06; 95% CI 4.01 to 4.11) with having been prescribed most of the time, which might explain why they found homogeneity in the pattern of the Daily Defined Dose (DDD) (i.e. 89% of the sample varied very little). We have shown how the current method could provide a scalable multi-national way to monitor medication use in the general population.
The relationship between socioeconomic variables and antidepressant pattern indicates that the greatest inequality exists for long-term prescription use. Previously it was found that there was little socioeconomic patterning in antidepressant review consultations in Scotland [35], which suggests that rather than differences in healthcare provision, the difference is due to disease severity. We found that unemployment was associated with decreasing use of antidepressants similar to other studies, which have found that unemployment status correlates with decreasing antidepressant use [36]. This effect is thought to be driven by health selection, whereby mental health status deteriorates before unemployment, and then improves during unemployment [37, 38]. Living alone had stronger associations with antidepressants previously (OR 1.81; 95%CI 1.46-2.23)[24] than in the current study (OR ranged from 1.10 to 1.28 for decreasing, increasing and most of the time groups), which may be explained by the differences in sample (they only included those in employment in the last 12 months). The association between living alone and common mental illness is mostly (84%) due to higher levels of loneliness [39]. Previous work estimated that psychotropic medication peaks 6-9 months before divorce and declined for 18 months thereafter [40], however we found that separation (which often precedes divorce) had a stronger relationship with a reduction in antidepressants, indicating that separation may provide a buffer to mental health distress between marriage and divorce.
The pattern of antidepressant prescriptions gives a good indication of mental illness; being on prescriptions most of the time is strongly and positively associated with self-reported mental illness and previous hospital admissions; the inverse is true for no prescriptions or occasional prescriptions. Increasing prescriptions had a weaker relationship with self-report mental illness than decreasing prescriptions, which shows that there might be a lag between starting medication and identifying oneself as having a mental illness. No association exists between previous psychiatric hospital admissions and the increasing prescriptions group, which may indicate that these patients have had a new episode of depression following the Recession. The negative relationship between self-reporting mental illness and previous hospital admissions, and the occasional prescription group confirmed that this group is unlikely to be suffering from persistent depression. Significant polypharmacy existed with antidepressant, anxiolytics and antipsychotics prescriptions. Antidepressants combined with anxiolytics were prescribed together particularly for those who had increased their antidepressant prescriptions. Antidepressants combined with antipsychotics were prescribed especially for those that have been prescribed antidepressants continuously over the study period. Polypharmacy has been advocated as a way to treat severe and treatment-resistant depression [41], however concerns have been raised especially for antidepressant-antipsychotic combinations with benefits outweighed by the increased risk of adverse effects (e.g. suicide) [42].
Implications for public health and research
Public health organisations could utilise the methods outlined in this paper to continuously monitor population mental health. The current application has shown the national trends and groupings for 2009-2014, but it could also be useful for a number of spatiotemporal configurations. Further drilling down of groupings may also be useful. Recommendations can be made to reduce the burden of mental illness and by linking these data on auxiliary measures of mental illness and key individual, area and macro level determinants. Future research could usefully develop this approach to examine measures of mental illness across the life course to understand continuation, relapse and remission, in combination with personal experiences by patients [43]. In particular, a high-risk change in living arrangements - going from marriage to separation and divorce and how that can lead to loneliness associated with living alone, warrants further investigation.