Background: The treatment profiles associated with polypharmacy among non-elderly patients is not well understood. Given the vast number of treatment permutations observed in real-world data, unsupervised machine learning may be an efficient tool for identifying distinct prescription groupings.
Objectives: To identify and describe treatment combinations and conditions commonly associated with outpatient multi-medication use in middle aged US adults.
Methods: A sample of patients 50-64 years old with two full years of eligibility from 01/01/2018 to 12/31/2019 were identified using the Optum's de-identified Clinformatics® Data Mart Database. Patients that died or had any evidence of malignancy/pregnancy during the study period were excluded. Polypharmacy was defined as ≥ 5 unique medications with ≥ 60 days of concurrent treatment. To identify prevalent treatment combinations, all medications prescribed during the study period among patients with evidence of polypharmacy were captured and used as features in a k-means clustering algorithm. Identified treatment combinations of interest were descriptively analyzed among the full sample.
Results: 40,544 patients met study inclusion criteria (mean age 58.19; 48.79% female). Common treatments included statins (76.02%), narcotic analgesics (54.02%), glucocorticoids (49.56%) nonsteroidal anti-inflammatory agents (39.91%), and biguanides (38.46%). Common diagnoses included hypertension (87.83%), hyperlipidemia (84.41%) and diabetes (65.65%). The optimal clustering of treatment profiles yielded two groups, with the first defined by concurrent statins and biguanides and the second by the presence of concurrent narcotic analgesics and benzodiazepines. Among the full sample, over one third (N=14,152) had concurrent treatment with statins and biguanides and 22.53% (N=9,133) with narcotic analgesics and benzodiazepines (average concurrent treatment 98.82 ±174.35 days). The most common diagnosis during a period with concurrent narcotic analgesics and benzodiazepines use was spondylosis/chronic low back pain (48.26%) with additional statin (33.48%), muscle relaxant (29.35%), and/or antidepressant (22.83%) treatment during the same period.
Conclusions: In addition to chronic conditions prevalent among older adults, such as hypertension and diabetes, unsupervised machine learning identified pain management as a common determinant of polypharmacy among a middle-aged population. Understanding polypharmacy care among middle-aged adults with chronic pain and/or metabolic conditions may help to improve outcomes in this population.
