Colombian Study Locates Medication Gaps in Older Adults; AI Link Is Context

The study, published in Frontiers in pharmacology in 2025, is an observational, cross-sectional analysis of 42,601 records of people over 65. The records came from a drug safety program at a pharmaceutical healthcare provider in Colombia between 2019 and 2025. The authors say the work is not designed to establish causality, so its findings are associations only.

The study did not test AI. It looked at one narrow measure: primary medication nonadherence, meaning a new medication was prescribed but the patient did not get it, or an appropriate alternative, within 30 days. Overall, 94.1% of patients (40,086) were adherent. Non-adherence was 5.9%.

The authors report that pharmacological factors were the most strongly associated with non-adherence, mainly problems of inappropriate use. They also name having several chronic diseases with several drug regimens, and administrative inefficiencies. They define inappropriate use as errors by patients or caregivers, such as missed doses, wrong dosage or frequency, or poor administration technique. Administrative inefficiencies are authorization or dispensing difficulties.

Social and coverage factors also showed up. Compared with people aged 90 and older, those aged 65 to 69 had a 32% higher probability of non-adherence (PR 1.32; 95% CI 1.07–1.64). Older adults with only primary education had a higher probability (PR 1.34; 95% CI 1.17–1.54). Living in the lowest-income housing strata raised it (PR 1.94; 95% CI 1.13–3.32), though more than half of the records (55.7%) lacked socioeconomic data. Being in the subsidized health regime raised it (PR 1.54; 95% CI 1.29–1.84). Patients without complementary insurance, 91.6% of the sample, also had a higher probability (PR 1.4).

Multimorbidity was common, at 53.2% (22,658 people), but showed no statistically significant difference in adherence (p = 0.21). The data come from one provider's program in one country, and they are not publicly available.

The link to the AI debate is context, not a finding. According to the Bipartisan Policy Center issue brief, direct care workers help with activities such as medication management. The brief says there are 5.8 million of them in the US, with jobs projected to grow 14% by 2035. It cites turnover of 94% for nursing home nursing staff in 2017 and 80% in home care in 2024. It also reports that just one-fifth of nursing and residential care respondents in a federal business survey said they use AI for any function.

The brief argues that AI could reduce administrative burden and improve care coordination. It also warns of privacy, bias and errors. It names five policy areas: oversight, public narratives, AI literacy, economic mobility and workforce supports. The Colombian study does not show that AI improves adherence. It does suggest where such tools could be judged: on use mistakes, prescribing and dispensing problems, in settings where patients also face cost and coverage barriers. This is the weakest part of the argument: the connection between the two sources is our own, and nothing in either tests it.

For vendors and operators, the strategic point is measurement. A tool that cuts paperwork but does not move outcomes like adherence may be hard to defend to regulators or payers. Safeguards the brief lists, including human oversight and continuous monitoring of outcomes, would favor products that can show results. That is interpretation, not evidence from the study.

Watch for whether any senior care AI pilots report medication outcomes, not just time saved, and whether policymakers act on the brief's five areas.

NIH Brain-Scan Study Tests a New Tracer for a Suspected Inflammation Enzyme Does Heavy Collagen Production Starve the Enzymes That Keep DNA Methylation in Check? A Testable Idea, With an Earlier Clue Do GLP-1 Drugs Help Most Where Inflammation Is High? A New SELECT Analysis Already Pushes Back Nanoscope's Light-Sensing Eye Gene Therapy: Randomized Macular Degeneration Data Are Promised, Not Yet Seen