Informing depression-specific dementia risk models: An evidence-based analysis of moderators of the depression-dementia association.

Depression is a major modifiable risk factor for dementia, yet most prediction models treat it as a homogeneous exposure, despite evidence that risk varies among people with depression. This study aimed to identify key modifiers of the depression-dementia association to inform the development of tailored prediction models. A narrative synthesis was conducted, incorporating (1) an umbrella review of nine meta-analyses examining the depression-dementia association; (2) a systematic review of depression-related medication use on dementia risk; and (3) findings from three Lancet Commission reports on dementia (2017, 2020, and 2024). Seven key modifiers were identified that influenced the reliability and direction of risk estimates: demographic factors, assessment methods, depression severity, follow-up duration, depression timing and trajectory, the outcome predicted (e.g., all-cause vs dementia subtypes), and antidepressant use. Late-life and severe depression conferred the highest risk, with associations being stronger for vascular dementia than for Alzheimer's disease. Clinical diagnoses yielded higher risk estimates compared to symptomatic rating scales. Duration of follow-up was associated with contradictory directional effects. Antidepressant use was associated with increased dementia risk. However, class-specific analyses were inconclusive due to extreme heterogeneity. The Lancet Commission emphasized late-life and mid-life depression as key modifiable risk factors. Multiple clinical, methodological, and temporal factors influence dementia risk estimates in individuals with depression. The findings support developing depression-specific dementia risk models that prioritize high-risk subgroups. Recommendations include distinguishing between symptom-based and clinical diagnostic approaches, addressing heterogeneity in timing and severity, modeling reverse causation, and validating models across diverse populations.