OBJECTIVES: Sedative burden have been shown to increase the risk of frailty. There is a limited understanding of the co-medication patterns of sedatives associated with frailty. Using network analysis, we aimed to describe the co-medication patterns in frailty states and identify potential medication drivers of frailty.

METHODS: In this cross-sectional study, we created a sedative co-medication network (SCN) using medication data from the Positive Brain Health Now(+BHN) cohort that included 824 people living with HIV recruited across Canada (mean age = 53 years). Medications with sedative activity were identified using the Sedative Load Model (SLM). Each sedative represented a node, and an edge represented a pair of sedatives used together. Physical frailty was defined using a modified Fried frailty criteria based on self-report items. Co-medication networks of frailty states (robust, prefrail, and frail) were analyzed for network parameters such as topological features, modularity, and centrality. Network comparison was conducted by comparing graph properties, significant node changes and tracking changes in the community structures across networks using Neighbor Shift (NESH) score and ∆Betweenness. R, NodeXL, and NetShift were utilized.

RESULTS: The SCN showed a sparse network (graph density = 0.20) of 51 sedatives and 254 unique combinations, with the quetiapine-bupropion combination as the most frequent co-medication. Bupropion, clonazepam, lorazepam, zopiclone, and amitriptyline were the most important drugs in the network based on centrality metrics. The network comparisons between frailty states identified the key sedative medication drivers: Frail vs Robust - gabapentin, mirtazapine, and amitriptyline; Prefrail vs Robust states - baclofen, mirtazapine, amitriptyline, and lamotrigine.

CONCLUSIONS: Our results identify antidepressants and antiepileptics as potential critical sedative node drivers between frailty states. Future studies should investigate if these findings are driven by comorbidities such as peripheral neuropathy and mood disorders. Our findings demonstrate that co-medication network analysis could complement traditional pharmacoepidemiologic methods as a tool for hypothesis generation.