Abstract
Globally and historically, deaths from ‘probable suicide’ or drug-related causes have carried compelling and varied emotive and societal weights. Recently, both ‘probable suicide’ and drug-related deaths (DRD) have been combined with chronic alcohol-related deaths into a mortality category of “deaths of despair”. For several decades, governments and international bodies have attempted to systemically prevent, and therefore reduce, the number of deaths. Prevention strategies for ‘probable suicide’ advocate for the expansion of psychiatric healthcare, predominantly talking therapies and (more controversially) antidepressant prescriptions, while DRD intervention strategies typically focus on prescribed opioid substitution therapy (OST). Both therapeutic approaches are associated with some improvements in patient health and small reductions in mortality, yet none of these approaches have significantly ameliorated the rising death rates. There are, not infrequently, suggestions that ineffective and inefficient healthcare services are to blame for the continuing high death rates, however, little conclusive data has been presented supporting, or indeed challenging, these claims due to a widespread reliance upon simple analyses of observational data. A large bank of previously collected healthcare data was available from a study undertaken before this doctorate began. Thus, the aim of this thesis was to investigate the healthcare usage at a variety of services of those who died from ‘probable suicide’ or DRD, contrasted with matched community controls, using this pre-existing database. Specific aims were to examine the rates of salient interventions, and provide data with which to question assumptions of poor healthcare.Using linked, administrative healthcare records for the Tayside region, held within a certified data Safe-Haven, 677 individuals classified by the National Records of Scotland as a ‘probable suicide’, DRD or both, were identified during the period 01.01.2009-31.12.2014. Each of the deceased were matched with 4 non-deceased ‘controls’ (using sex, age and estimated socio-economic status). Healthcare data, including key prescriptions, were extracted for the twelve-month prior to death. After healthcare usage comparisons, based on the total cohorts, sub-sections within these, and pivotal treatment-related measures (e.g., OST dosage per day), a clustering analysis was performed. This analysis used a partitioning around medoids (PAM) algorithm, which accounts for mixed-type data by standardising each variable into a numerical value of dissimilarity, averaging the dissimilarity value across all variables, and applying that average as a measure of distance, from which similar individuals can be grouped into clusters. The variables read into the PAM algorithm included categorical demographic variables and binary or frequency healthcare attendance measures. A goodness-of-fit measure, known as a silhouette width, was calculated for values from 2 to 10 clusters. Independent analyses were run for the ‘probable suicide’ and DRD groups, with a final iteration including both cohorts, attempting to isolate patterns of antecedent healthcare usage that were unique or shared across study samples.
Initially, 605 individuals were extracted from the National Records of Scotland (NRS) for the ‘probable suicide’ cohort, however only 586 were validated with official cause of death criteria. Correspondingly, 311 DRD individuals were extracted from the NRS, however only 288 were validated. Those who died a ‘probable suicide’ or DRD attended all healthcare services and received a higher rate of psychotropic prescriptions than matched community controls. Of those receiving salient prescriptions (antidepressant or OST, respectively), those who died attended services at higher rates than the controls, likewise considered “in treatment”. Comparisons confined to Accident and Emergency presentations showed that after possible self-harm events, those who went on to die received more psychiatric follow-up in the 21 days after, than the controls with identical presentations. Finally, within the cohorts of ‘probable suicide’ and DRD, those “in treatment” were evidently attending healthcare services and receiving prescriptions at higher rates than those not “in treatment”. The clustering analysis, performed on the ‘probable suicide’ only, DRD only and a combined cohort identified three basic patterns across all cohorts; low attendance (associated with a large group of men), attendance at specific services (antidepressant prescriptions or methadone prescriptions, generally associated with women and men respectively) and high attendance at all services (including an unforeseen group of women with antidepressant, methadone and benzodiazepine prescriptions, in all 3 analyses).
Each comparison showed both the ‘probable suicide’ and DRD cohorts to have higher historic rates of healthcare utilisation, consistent with the poorer health often associated with “deaths of despair” and their association with poverty, psychological distress and illicit drug use. Furthermore, specifically “in treatment” comparisons demonstrated similar patterns to the total cohort comparison, in that those who went on to die had higher rates of healthcare usage. Specific comparisons indicated the ‘probable suicide’ group redeemed more antidepressant prescriptions, while no difference in methadone dosage between DRD and controls in OST were found; these results seriously challenge the idea that gaps in the healthcare system are largely responsible for these types of death. Multiple clustering analyses demonstrated the difficulties of differentiating meaningfully between ‘probable suicide’ or DRD deaths, and in 10-cluster analyses, found three attendance patterns, associated with particular demographic groups, though significant heterogeneity in cluster compactness was noted. This thesis demonstrates various sub-groups can be identified in both ‘probable suicide’ and DRD cohorts, which could potentially improve the efficacy of interventions, if targeted appropriately.
Date of Award | 2023 |
---|---|
Original language | English |
Awarding Institution |
|
Supervisor | Keith Matthews (Supervisor) & Benjamin Vincent (Supervisor) |
Keywords
- Suicide
- Drug-related death
- Routine data