Risk Prediction of Dementia in Middle‐aged and Older Adults: The Dementia Risk Prediction Pooling (DRPP) Consortium
Norrina Allen, Amy Krefman, John Stephan, Jingzhi Yu, Padraig Carolan, Sanaz Sedaghat, Maxwell Mansolf, Aicha Soumare, Alden L. Gross, Allison Aiello, Archana Singh‐Manoux, M. Arfan Ikram, Catherine Helmer, Christophe Tzourio, Claudia L. Satizabal, Deborah A Levine, Donald Lloyd‐Jones, Emily M Briceno, Farzaneh Sorond, Frank J. Wolters, Jayandra Jung Himali, Lenore J J. Launer, Lihui Zhao, Oscar L. Lopez, Stéphanie Debette, Sudha Seshadri, Suzanne E Judd, Tim M. Hughes, Vilmundur Gudnason, Denise Scholtens- Psychiatry and Mental health
- Cellular and Molecular Neuroscience
- Geriatrics and Gerontology
- Neurology (clinical)
- Developmental Neuroscience
- Health Policy
- Epidemiology
Abstract
Background
Several risk prediction equations exist for the incidence of Alzheimer’s Disease and related dementias (ADRD); however, current equations are based on studies of older individuals, with limited follow‐up time and covariates, and have generally not performed well in external validation. The aim of this study is to develop a novel dementia risk prediction equation based on a pooled analysis of population‐based studies that include both middle‐aged and older adults.
Methods
We analyzed data from 13 cohorts within the Dementia Risk Prediction Pooling (DRPP) Consortium, which pooled individual participant data from cohorts in the US, France, Iceland, England, and the Netherlands, with baseline exams ranging from 1948 to 2006. Participating cohorts conducted longitudinal in‐person assessments of clinical, genetic, and behavioral risk factors over at least 10 years, and assessed dementia status. We rigorously harmonized demographic, education, lifestyle, and clinical factors as well as dementia status. From these 25 harmonized variables, we used LASSO competing risks regression to identify a parsimonious set of covariates with the best predictive performance for incident dementia based on C statistics obtained from competing risks.
Results
Among the 55,614 participants included in this analysis (mean age 62.5y, 48.8% female; see Table 1), 2,523 developed incident dementia within 10 years and 4,072 within 20 years of the index exam. The resulting clinical risk prediction equation included age, gender, education, BMI, presence of depression, physical activity, smoking, and total cholesterol. The top performing equations had C statistics of 0.86‐0.89 at 10 years and 0.88‐0.92 at 20 years. Calibration plots are shown in the Figure 1.
Conclusions
Using pooled individual‐level data from a large harmonized consortium of cohorts with detailed clinical and behavioral data, we developed a novel dementia risk prediction equation with excellent discrimination and acceptable calibration across a broad range of predicted risk; these equations have better performance than published risk scores. The ability to accurately identify high‐risk individuals, even as early as middle‐age, will allow healthcare providers to target preventive efforts (e.g., statins) to those at the greatest risk to reduce the future burden of dementia.