Advanced epidemiology analysis & advanced statistical methods

Janine Beezer, Newcastle University

Aim – to characterise polypharmacy in different heart failure (HF) populations and to understand and explore the factors that influence polypharmacy within these populations. To understand the impact polypharmacy has on patient outcomes. 

Research question what is polypharmacy in heart failure and impact of polypharmacy on patient outcomes? 

Design – single centre retrospective longitudinal cohort analysis. Patients will be identified from the National Heart Failure Audit (NHFA) database. It is then necessary to access hospital electronic medical records to obtain additional data, for this, and subsequent admissions to complete a comprehensive longitudinal dataset. 

Dataset data will be collected, from the first admission to hospital with HF through to death or end of the data collection period, whichever occurs first, capturing all admissions and information on medication in between, extracted into REDCaP and stored on a password protected secure hospital computer and anonymised, once hospital records data collection is complete.  

Retrospective data from hospital electronic records will include: Clinical Frailty Scale (CFS), comorbidities, life status, date of death and information within records will be used to calculate Charlson co-morbidity index and medication regimen complexity index for the first and all subsequent hospital admissions, in addition to standard NHFA data fields. Electronic documentation containing necessary data has been available since 2013, eliminating the need to look through large amounts of paper notes.  Sample size calculated based on estimating the prevalence of polypharmacy in HF populations,(1) the first stage of the characterisation process. Average prevalence of polypharmacy was taken from a recent systematic review,(2) using a definition <5 medicines, which was 73%. Using the method by Daniel 1999(1), a sample of 323 for each HF subtype is sufficient to estimate the prevalence of polypharmacy with 10% power and 5% significance level. The novelty of this research means there is a dearth of information on the relative hazard of mortality between those with and without polypharmacy in HF patients, so it is difficult to calculate the sample size for survival analyses. However, the current database has approximately 3000 records, which is a significantly inflated sample (compared to the prevalence estimation sample size) to provide confidence to conduct these analyses.  Statistical analysis – descriptive statistics (means, medians and proportions) will be used to describe the differences in patient characteristics.  

Characteristics will be compared using Chi2 for categorical data, means with t-tests and medians using appropriate non-parametric techniques. Univariate ordinal logistic regression will be used to assess if polypharmacy differs conditioned on baseline characteristics. Mixed effect models will explore changes in polypharmacy over time. 

The effect of polypharmacy on all-cause re-admission will be analysed using standard and time-toevent modelling techniques (logistic regression, Cox Proportional Hazards, Kaplan Meier, log-rank test). Continuous outcomes (e.g. length of stay) will be investigated using generalised linear models.  All models will adjust for known confounders in the literature e.g. age, renal function and co-morbidity. Outcomes The characterisation of polypharmacy within the different HF diagnoses. Including; numerical counts for polypharmacy and calculated medication complexity, enabling the detailed description and comparison of contributary factors. The identification of influencing factors of polypharmacy and the impact on patient outcomes.  

The final report is available to view here.