(AL) Classifying Polypharmacy: A Latent Class Analysis in Older Adults in Quebec, Canada

Publication Number: 654

Background: The imprecise definition of polypharmacy, often designated as the concomitant use of five or more medications, does not distinguish between appropriate and inappropriate polypharmacy. Other characteristics could be used to classify polypharmacy into subgroups with varying levels of risk.

Objectives: To characterize different types of polypharmacy among older adults in the province of Quebec, Canada.

Methods: Using healthcare databases from the Quebec Integrated Chronic Disease Surveillance System, we selected a community-based random sample of 110,000 Quebecers aged 66 years and older on March 31st, 2017. Only those who were covered by the Public Drug Insurance Plan were included. Categorical indicators used to describe polypharmacy were evaluated over a 6-month period (October 1st, 2016 to March 31st, 2017) and consisted of: (1) number of medications, (2) potentially inappropriate medications (PIM), (3) drug interactions, (4) high-alert medications, (5) anticholinergic burden score (ABS), (6) use of blister cards, and (7) administration complexity (i.e., different routes of administration). Latent class analysis (LCA) was used to subdivide polypharmacy into groups. Models ranging from 1 to 7 classes were tested. Choice of the final model was based on fit statistics, parsimony, and interpretability.

Results: In total, 93,516 individuals were included. Participants were on average 75.5 (±7.1) years old and took 6.4 (±4.7) medications. Nearly half (52.7%) had at least one PIM, one drug interaction or one high-alert medication. Mean ABS was 0.9 (±1.5). One third (33.6%) had at least one complex route of administration and 15.7% used blister cards. A 4-class model was selected, based on statistical criteria and clinical interpretability. Entropy was 0.77. The groups can be described as: (1) High-medium number of medication, low risk (33.5% of population share), (2) Medium number of medication, PIM use ± high ABS (8.1%), (3) No polypharmacy (45.7%) and (4) Hyperpolypharmacy, complex use, high risk (12.8%). However, tested models had continuously decreasing fit indices (BIC, AIC, SABIC) and relatively low entropy ( < 0.80), making the choice of model more arbitrary and compromising classification’s validity.

Conclusions: LCA suggests 4 classes of polypharmacy which differ according to the number of medications and level/nature of associated risks. LCA models could be refined by including new indicators or by integrating both categorical and continuous variables as indicators.