Investigating the associations of trajectories of depressive symptoms and self‐perceived health and incident dementia: An unsupervised machine learning approach. (31st December 2021)

• Record Type: Journal Article
• Authors: Klee, Matthias; Leist, Anja
• Is Part Of: Alzheimer's & dementia. Volume 17(2021)Supplement 10
• Page Start: n/a
• Page End: n/a
• Publication Date: 2021-12-31
• Subjects: Alzheimer's disease -- Periodicals; Alzheimer Disease -- Periodicals; Dementia -- Periodicals; Démence; Maladie d'Alzheimer; Périodique électronique (Descripteur de forme); Ressource Internet (Descripteur de forme); 616.83
• DOI: 10.1002/alz.055450
• Languages: English
• ISSNs: 1552-5260

Abstract

Abstract: Background: Risk factors for dementia show inter‐individually varying trajectories over the lifespan. However, risk factors have been mainly investigated with one time‐point assessments. New research suggests that certain risk factor trajectories are associated with increased risk of adverse cognitive outcomes (Demnitz et al., 2020 [https://doi.org/10.1101/2020.05.20.20106963]; Singh‐Manoux et al., 2018 [https://doi.org/10.1016/j.jalz.2017.06.2637]). However, it remains unclear how sequential and simultaneous changes of risk factors alter the individual risk for developing dementia. Testing the joint contribution of trajectories of depressive symptoms and self‐perceived health on incidence of dementia, we hypothesized that consistently poor as well as deteriorating trajectories increase the risk for incident dementia, and explored possible interactions of the trajectories. Method: A total of 5, 326 respondents to the SHARE survey, mean age 73.9 years, and 6 complete follow‐ups spanning ∼13 years, answered the EURO‐D depression scale, self‐perceived health (SPH) (t1‐t5), and self‐reported dementia diagnosis at last follow‐up (t6). To investigate the predictive ability of distinct longitudinal trajectories, we applied unsupervised statistical learning methods (K‐means cluster modelling). Clusters indicated distinct risk factor trajectories, which were used as exposures in stepwise logistic regressions to predict incident dementia, controlling for age, gender, education, and country. Result: Cluster analysis revealed five distinct trajectories each for SPH and EURO‐D, with varying dementia incidence. In stepwise logistic regressions, respondents with trajectories "consistently poor health" and "consistently high depression" showed elevated risk of dementia ( OR = 4.02 [1.39, 14.75] and OR = 2.26 [1.03, 4.95], respectively) compared to the combined baseline risk for "consistently low depression" and "consistently good health". Interactions were not significant. However, respondents with the combination of "consistently high depression" and "consistently poor health" showed increased risk ( N = 246; 6.1% dementia). Conclusion: Applying unsupervised machine learning is helpful to incorporate longitudinal information on depressive symptoms and self‐perceived health and model these risk factors longitudinally to test their contribution to explain incidence of dementia. The predictive ability of the trajectories of depressive symptoms and self‐perceived health for dementia indicates the potential for improving the identification of people at risk for developing dementia in late life by exploiting trajectory information readily accessible through regular medical check‐ups in old age.