ACT Statistical Guidance

Our Work

Here, we list publications that illustrate some of the recommended practices in our guidance documents. We also highlight recent scientific contributions from ACT statisticians that have applied innovative statistical methods to address some of the analytical challenges in the ACT data.

Application of augmented inverse probability weighting (AIPW) to efficiently analyze the associations between dementia-related neuropathology outcomes and Angiotensin II stimulating vs inhibiting antihypertensive medications.

In this paper, we defined the primary exposure of interest as cumulative person-years of exposure to each of the two classes of antihypertensive medications (angiotensin II stimulating vs inhibiting). The outcomes were dementia-related neuropathology in ACT participants who had consented and underwent autopsy. We used generalized raking to leverage information we observed on the eligible ACT cohort in order to improve the efficiency of inverse-probability weighted analysis of the complete case subset from the autopsy cohort. Gray S, Yu O, Gatto NM, Marcum ZA, Latimer CS, Postupna N, Su Y, Barthold D, van Dalen JW, Richard E, Keene CD, Shaw PA, McEvoy LK, Larson EB, Crane PK. Angiotensin II stimulating antihypertensives and dementia-related neuropathology. JAMA Network Open. 9(2): e2559113. doi:10.1001/jamanetworkopen.2025.59113 (2026).

Association between 24-hour activity behavior compositional exposure and 10-year physical function trajectories

Here, we analyzed 10-year trajectories of physical function in the 10 years prior to a week-long accelerometer-based assessment of 24-hour activity cycle (24HAC) behaviors using linear mixed effects modeling. We then examined the association between the person-specific slope and intercept of the trajectories with time in bed, sedentatry behavior and physical activity. A bootstrap variance estimate was used to account for the uncertainty in the trajetory-based exposures (intercept and slope). Greenwood-Hickman MA, Zhu W, Idu A, Harrington LB, McCurry SM, LaCroix AZ, Shaw PA, Rosenberg DE. Associations Between 10-Year Physical Performance and Activities of Daily Living Trajectories and Physical Behaviors in Older Adults. International Journal of Environmental Research and Public Health. 2025 Apr 29;22(5):704.

Association between 24-hour activity behavior compositional exposure and 10-year cognitive trajectories

We examined the relationsip between 10-year trajectories of cogntive function prior to a week-long accelerometer-based assessment of 24-hour activity cycle (24HAC) behaviors using growth mixture modeling. Rosenberg DE, Wu Y, Idu A, Greenwood-Hickman MA, McCurry SM, LaCroix AZ, Shaw PA. Historic cognitive function trajectories as predictors of sedentary behavior and physical activity in older adults. The Journals of Gerontology, Series A: Biological Sciences and Medical Sciences. 2024 Jul;79(7):glae125.

Application of weighted cumulated exposure to flexibly analyzed the 10-year anticholinergic medication use and its association with change in physical function.

In this work, we defined the exposure of 10-year total standard daily dose of anticholinergic medications in two ways: 1) a conventional exposure assuming constant effect over time, and 2) a flexible exposure assuming time-varying effect over time by applying a weighted function over time. The outcomes were changes in gait speed and grip strength that were measured during ACT biennial visits. Gray SL, Su Y, Eshetie TC, LaCroix AZ, Marcum ZA, Yu O. Cumulative anticholinergic exposure and change in gait speed and grip strength in older adults. JAMA Network Open. 8(7):e2519819. doi:10.1001/jamanetworkopen.2025.19819 (2025).

Development of a feature-based eye disease severity scoring systems for aging eyes using item-response theory models.

As a part of Eye ACT, an ancillary ACT study focusing on aging eyes, we developed two item response theory (IRT) models for severity scores of exudative (wet) and non-exudative (dry) age-related macular degeneration (AMD) based on 7 relevant sub-clinical eye features from free-text clinical notes. These IRT models were developed using full information likelihood approach, allowing partially observed features when deriving AMD severity scores for an individual eye. The models were developed using data from an external clinical trial on AMD and evaluated using longitudinal eye clinical data in Eye ACT. We demonstrated that the IRT-based severity scores align with ADM-related meaningful clinical events and conventional severity score for AMD and provide more granular stratification on AMD severity level compared to the conventional scores. Lee CS, Su YR, Walker RL, Krakauer C, Blazes M, Johnson EA, Cronkite D, Bowers W, Hess C, Arterburn D, Agrón E, Chew EY, Crane PK. A Data-driven Age-related Macular Degeneration Severity Scoring System Leveraging the AREDS Studies and Clinical Electronic Medical Records. Ophthalmology. 2025 Oct;132(10):1076-1087.

Generalizing findings from the community-based ACT study to its source population

Data from Behavioral Risk Factor Surveillance System (BRFSS) survey were used to estimate sociodemographic characteristics of the Seattle Metropolitan Region population over the age of 65 and compared to those in ACT. Participation weights were then constructed from the estimated probability someone in the Seattle Metropolitan Region would be included in ACT. This work then shows how to use these weights to transport study estimates from ACT to represent a population from the surrounding metropolitan area, using the example of prevalence of common eye conditions and the associations between eye conditions and the risk of Alzheimer’s disease. Gibbons LE, Mobley T, Mayeda ER, Lee CS, Gatto NM, LaCroix AZ, McEvoy LK, Crane PK, Hayes-Larson E. How generalizable are findings from a community-based prospective cohort study? Extending estimates from the Adult Changes in Thought study to its source population. Journal of Alzheimer’s Disease. 2024 Jun 25;100(1):163-74.

Integrating error-prone EHR outcomes with validated outcome data from the ACT Autopsy cohort

In this work, a novel statistical method is proposed for an inverse-probablity weighted estimator for the average treatment effect (ATE) is proposed that optimally combined an estimate of the ATE based on the validated outcomes derived from the ACT autopsy cohort and an estimate based on error-prone EHR data that have been corrected for misclassification. In numerical studies, authors found that the proposed work improved the overall efficiency of the ATE by leveraging all available data on the ACT cohort. Shen J, Isenberg D, Linn KA, Hubbard RA. Integrating Misclassified EHR Outcomes With Validated Outcomes From a Non‐Probability Sample. Statistics in Medicine. 2025 Jul;44(15-17):e70127.

Evaluating the contribution of diabetic retinopathy to risk of dementia not attributable to diabetes severity

Here, we examine whether the association between diabetic retinopathy and risk of dementia is attributable entirely to severity of diabetes. We controlled for three components of diabetes severity using data from the electronic health record (EHR): maximum 5-year albumin-to-creatinine ratio, multiple summaries of glomerular filtration rate, and long-term glycemia. The latter was quantified using the average glucose value over the prior 5 years, approximated using a Bayesian modeling approach developed in a previous ACT paper. Multiple imputation via chained equations was implemented to account for missing covariate data. We also used sum of Akaike weights to evaluate the relative importance of different components of diabetes severity and other known risk factors for explaining dementia risk. Lee CS, Krakauer C, Su YR, Walker RL, Blazes M, McCurry SM, Bowen JD, McCormick WC, Lee AY, Boyko EJ, O'Hare AM. Diabetic retinopathy and dementia association, beyond diabetes severity. American journal of ophthalmology. 2023 May 1;249:90-8.