Evaluating depressive symptoms and risk of cognitive decline using longitudinal electronic health records

Abstract

Abstract Background Depression is associated with increased risk of Alzheimer’s disease and related dementias (ADRD). Many population‐based and convenience sample studies have examined depression and depressive symptoms in this context. However, fewer studies employed electronic health records (EHR) to assess this relationship. Because depression might be under‐ or mis‐diagnosed, diagnostic codes alone may be insufficient. We aimed to assess whether prior depression diagnosis and/or depressive symptoms–derived from natural language processing (NLP) in the EHR–predicts future mild cognitive impairment (MCI)/dementia using a large EHR‐based dataset. Method The Massachusetts Electronic Health Records Aging Dataset (MEAD) is a retrospective cohort of patients who were ≥50 years old with normal cognition at their first encounter between 1/1/2000‐12/31/2005. Depression diagnoses and incident MCI/dementia were identified from structured encounter diagnoses and problem lists, and depressive symptoms from clinical notes using NLP. Follow‐up time was calculated from index date until date of cognitive diagnosis, death, loss to follow‐up or end of study (12/31/2021). We performed Cox proportional hazards models using age as the time scale to examine the association of depression and incident MCI/dementia, using the Fine‐Gray approach to account for competing risk of death. Result The MEAD cohort included 162,069 patients. Of those, 135,272 did not have depressive symptoms or a depression diagnosis at their first encounter, 14,985 had only depressive symptoms, 4,883 had only a depression diagnosis, and 6,929 had both depressive symptoms and a depression diagnosis. The incidence rate of MCI/dementia per 1,000 person‐year was 7.8 for patients without depressive symptoms or depression diagnosis, 10.0 for depressive symptoms only, 12.4 for depression diagnosis only, and 10.3 for both. Compared with those without depressive symptoms or depression diagnosis, the multivariable‐adjusted hazard ratios for incident MCI/dementia were 1.05 (95% confidence interval [CI] 0.99,1.11) for depressive symptoms only, 1.45 (95% CI 1.34,1.56) for depression diagnosis only, and 1.15 (95% CI 1.07,1.23) for both, respectively. Conclusion In a large sample, diagnosis of depression appears to be a stronger predictor than depressive symptoms of all‐cause MCI/dementia. Further research is needed to understand the utility of reported depressive and other neuropsychiatric symptoms in clinic notes for predicting cognitive decline.