Partitioning Variation in Prescribing of Statins as a Percentage of Total Prescribed Medications for the Cardiovascular System: A Spatiotemporal Ecologic Study in England (2015-2019)

Background: Statins are widely prescribed for the primary and secondary prevention of cardiovascular diseases. However, few studies have examined both time trends and sources of variation in statin prescribing.

Objectives: To describe time trends and variation in GP Practice (GPP) annual prescribing of statins as a percentage of total prescribed cardiovascular system (CVS) medications (i.e., the response) in England, accounting for spatiotemporal correlation at the Clinical Commissioning Group (CCG) level, as it was hypothesized that nearby CCGs would have similar prescribing policies.

Methods: Spatiotemporal ecologic study using prescription data in England, 2015-2019. We aggregated data by year to adjust for GPP (patient list size; GP headcount; urban; percentages of registered patients that are female, aged 65+ years, hypertension, obese, diabetic; patient population-weighted GPP area deprivation) and CCG (population density) factors measured yearly. We fitted Bayesian logistic spatiotemporal models adjusted for CCG/GPP factors to estimate variance partition coefficients (VPCs). VPCs quantify the proportion of variance in the response explained by structured and unstructured spatiotemporal variation at the CCG level and correlated non-spatial unstructured and temporal variation at the GPP level. We specified diffuse priors for fixed and non-spatial random effects and an intrinsic Conditional Autoregressive prior for spatial random effects.

Results: We included 7639 GPPs nested within 191 CCGs between 2015-2019. The prescribing of statins as a percentage of CVS medications was stable over time (median: 21, 21, 22, 22, and 22 in 2015, 2016, 2017, 2018 and 2019, respectively). CCGs and GPPs in the North, Central and London regions tended to have higher percentages of statins prescribed relative to the East and South regions of England. Nearly all the total variation in the response was estimated to be due to differences within and between GPPs (Combined VPC: 99% (95% credible interval: 99, 100); VPC between CCG: < 1% (0, < 1)). There was evidence of variability in the annual rate of change in the response between GPPs, with a moderate negative correlation of -.28 (-.31, -0.28) between the GPP random year slopes and random intercepts. Whilst between CCG variation was minimal overall, there was evidence of CCG level spatiotemporal variation, where a fair proportion of between CCG variation was spatiotemporal (VPC: 42% (31, 46); VPC non-spatial: 58% (54, 69)). All estimates were adjusted for CCG and GPP factors.

Conclusions: Estimated VPCs indicate that policies aimed at influencing statin prescribing may be more effective targeted at GPPs than CCGs (health regions). VPCs provide insight into sources of variation in prescribing that may better inform policy.