Amanda E. Paluch

AI decision-support tools typically offer a fixed type of assistance, like AI recommendations and explanations, regardless of the specific decision, individual, or broader context. This fixed design has been shown to hinder both human-AI decision accuracy and human skill improvement in the task. We posit that AI assistance needs to be dynamic, changing in response to contextual factors (e.g., AI uncertainty, task difficulty), individual differences, and specified objectives (e.g., decision accuracy, skill improvement). To enable such adaptive support, we propose reinforcement learning (RL) as a general approach for modeling human-AI decision-making to optimize human-AI interaction for diverse objectives. RL enables optimizing various objectives in AI-assisted decision-making by tailoring and adaptively providing decision support to humans — the right type of assistance, to the right person, at the right time. We instantiated our approach with two objectives: human-AI accuracy on the decision-making task and human skill improvement (i.e., learning about the task) and learned decision support policies from previous human-AI interaction data. We compared the optimized policies against several baselines in AI-assisted decision-making. Across two experiments (N = 316 and N = 964), our results consistently demonstrated that people interacting with policies optimized for accuracy achieve significantly higher accuracy — and even human-AI complementarity — compared to those interacting with any other type of AI support. Our results further indicated that human learning was more difficult to optimize than accuracy. While the policies learned the best available actions to optimize learning, participants who interacted with learning-optimized policies showed significant learning improvement only at times. Our research (1) demonstrates offline RL to be a promising approach to model the dynamics of human-AI decision-making, leading to policies that may optimize various objectives and provide novel insights about the AI-assisted decision-making space, and (2) emphasizes the importance of considering skill improvement and other human-centric objectives beyond accuracy in AI-assisted decision-making, opening up the novel research challenge of optimizing human-AI interaction for such objectives.

Wearable devices offer the ability to objectively characterize free-living physical activity; however, raw step-count data generated by commercial devices require systematic processing before they can support rigorous inference. We describe a transparent, reproducible standard operating procedure (SOP) for transforming epoch-level step-count data from commercial Garmin devices into participant-level analytic variables and demonstrate its application in the STRRIDE-PD Reunion study: a long-term follow-up of older adults originally enrolled in a supervised exercise intervention trial. This data pipeline standardizes timestamps, reconstructs daily epoch grids, infers wear time from observed step patterns, and applies a prespecified valid-day threshold (≥10 hours inferred wear time) to generate participant-level summaries. Among 67 participants (mean age 71.4 years, 65.7% women), the median valid-day count was 10 days, median average daily steps were 5,794, and participant-level estimates were identical across ≥10-hour and ≥6-hour valid-day thresholds. Wearable-derived step counts were significantly associated with 9 of 16 cardiometabolic and fitness outcomes, including cardiorespiratory fitness, body composition, and lipid profiles. By contrast, self-reported exercise – assessed via a frequency-by-duration composite ranked into deciles – was not significantly associated with any outcome. A regression calibration framework applied to the full sample quantified the attenuation underlying this discrepancy: the naive self-report model systematically underestimated associations relative to both the observed Garmin model and calibration-corrected estimates. These findings demonstrate that measurement approach is a determinant of scientific conclusions in physical activity research, and that reproducible wearable data pipelines are essential infrastructure for aging epidemiology. Highlights: A reproducible standard operating procedure processed Garmin step-count data without wear-time indicators. Wearable steps predicted 9 of 16 outcomes; self-reported exercise predicted none. Regression calibration revealed that self-report systematically underestimated physical activity-health associations. Measurement approach determines physical activity-health conclusions in aging cohort research.

Background/Purpose: Frailty is a significant health challenge for the aging population, increasing vulnerability to adverse health outcomes such as chronic disease, falls, disability, and mortality. Early detection and monitoring of frailty are critical for effectively managing frailty. Daily steps, as a measure of physical activity, hold valuable information about health status and may serve as an indicator of frailty. This study aimed to examine the association between daily steps and frailty in older adults. Method: PubMed, SPORTDiscus, and Web of Science databases were searched for published studies up to June 2023. The search terms were ("daily step" or "steps per day" or "step count" or "number of steps" or "step volume") AND ("frailty" or "frail" or "pre-frailty" or "pre-frail" or "prefrailty" or "prefrail"). The inclusion criteria were peer-reviewed articles in English involving older adults aged ≥ 65 years, device-measured daily steps, and reported frailty status. Pooled estimates of mean differences with 95% confidence intervals (CI) between frailty groups (non-frail, prefrail, frail) were obtained using random-effects models. The prefrail and frail groups were combined in the primary analysis. Results: Thirteen articles comprising 3,383 participants (71.21 ± 6.69 years, 75.58% female, 42.45% prefrail and frail) were included in the analysis. Daily steps were significantly lower in the prefrail and frail combined group compared to the non-frail group (MD = 2,008, 95% CI: 1,003, 3,013, I ² = 86%). This association was consistent across subgroups stratified by health conditions, regions, wearables placement, and frailty measurement. Further analyses within the three frailty groups revealed that the prefrail group accumulated 995 steps/day (95% CI: 407, 1,582, I ² = 73%) less than the non-frail group, but 947 steps/day (95% CI: 567, 1,327, I ² = 32%) more than the frail group. Conclusion/Discussion: Current evidence suggested that older adults with frailty tend to have lower daily steps compared to non-frail older adults. Daily steps may be an indicator of frailty status in older adults.

Human studies examining the cellular mechanisms behind sarcopenia, or age-related loss of skeletal muscle mass and function, have produced inconsistent results. A systematic review and meta-analysis were performed to determine the aging effects on protein expression, size, and distribution of fibers with various myosin heavy chain (MyHC) isoforms. Study eligibility included MyHC comparisons between young (18–49 yr) and older (≥60 yr) adults, with 27 studies identified. Relative protein expression was higher with age for the slow-contracting MyHC I fibers, with correspondingly lower fast-contracting MyHC II and IIA values. Fiber sizes were similar with age for MyHC I, but smaller for MyHC II and IIA. Fiber distributions were similar with age. When separated by sex, the few studies that examined females showed atrophy of MyHC II and IIA fibers with age, but no change in MyHC protein expression. Additional analyses by measurement technique, physical activity, and muscle biopsies provided important insights. In summary, age-related atrophy in fast-contracting fibers lead to more of the slow-contracting, lower force-producing isoform in older male muscles, which helps explain their age-related loss in whole muscle force, velocity, and power. Exercise or pharmacological interventions that shift MyHC expression toward faster isoforms and/or increase fast-contracting fiber size should decrease the prevalence of sarcopenia. Our findings also indicate that future studies need to include or focus solely on females, measure MyHC IIA and IIX isoforms separately, examine fiber type distribution, sample additional muscles to the vastus lateralis (VL), and incorporate an objective measurement of physical activity.

The National Heart, Lung, and Blood Institute convened a virtual workshop in September 2022 to discuss “Optimal Instruments for Measurement of Diet, Physical Activity, and Sleep.” This report summarizes the proceedings, identifying current research gaps and future directions for measuring different lifestyle behaviors in adult population‐based studies. Key discussions centered on integrating report‐based methods, like questionnaires, with device‐based assessments, including wearables and physiological measures such as biomarkers and omics to enhance self‐reported metrics and better understand the underlying biologic mechanisms of chronic diseases. Emphasis was placed on the need for data harmonization, including the adoption of standard terminology, reproducible metrics, and accessible raw data, to enhance the analysis through artificial intelligence and machine learning techniques. The workshop highlighted the importance of standardizing procedures for integrated behavioral phenotypes using time‐series data. These efforts aim to refine data accuracy and comparability across studies and populations, thereby advancing our understanding of lifestyle behaviors and their impact on chronic disease outcomes over the life course.

INPLASY202460109 Effects of age on human skeletal muscle: A systematic review and meta-analysis of myosin heavy chain isoform protein expression, fiber size and distribution Keywords: aging, sarcop…

Imagine if AI decision-support tools not only complemented our ability to make accurate decisions, but also improved our skills, boosted collaboration, and elevated the joy we derive from our tasks. Despite the potential to optimize a broad spectrum of such human-centric objectives, the design of current AI tools remains focused on decision accuracy alone. We propose offline reinforcement learning (RL) as a general approach for modeling human-AI decision-making to optimize human-AI interaction for diverse objectives. RL can optimize such objectives by tailoring decision support, providing the right type of assistance to the right person at the right time. We instantiated our approach with two objectives: human-AI accuracy on the decision-making task and human learning about the task and learned decision support policies from previous human-AI interaction data. We compared the optimized policies against several baselines in AI-assisted decision-making. Across two experiments (N=316 and N=964), our results demonstrated that people interacting with policies optimized for accuracy achieve significantly better accuracy -- and even human-AI complementarity -- compared to those interacting with any other type of AI support. Our results further indicated that human learning was more difficult to optimize than accuracy, with participants who interacted with learning-optimized policies showing significant learning improvement only at times. Our research (1) demonstrates offline RL to be a promising approach to model human-AI decision-making, leading to policies that may optimize human-centric objectives and provide novel insights about the AI-assisted decision-making space, and (2) emphasizes the importance of considering human-centric objectives beyond decision accuracy in AI-assisted decision-making, opening up the novel research challenge of optimizing human-AI interaction for such objectives.

Background: Taking fewer than the widely promoted “10 000 steps per day” has recently been associated with lower risk of all-cause mortality. The relationship of steps and cardiovascular disease (CVD) risk remains poorly described. A meta-analysis examining the dose–response relationship between steps per day and CVD can help inform clinical and public health guidelines. Methods: Eight prospective studies (20 152 adults [ie, ≥18 years of age]) were included with device-measured steps and participants followed for CVD events. Studies quantified steps per day and CVD events were defined as fatal and nonfatal coronary heart disease, stroke, and heart failure. Cox proportional hazards regression analyses were completed using study-specific quartiles and hazard ratios (HR) and 95% CI were meta-analyzed with inverse-variance–weighted random effects models. Results: The mean age of participants was 63.2±12.4 years and 52% were women. The mean follow-up was 6.2 years (123 209 person-years), with a total of 1523 CVD events (12.4 per 1000 participant-years) reported. There was a significant difference in the association of steps per day and CVD between older (ie, ≥60 years of age) and younger adults (ie, <60 years of age). For older adults, the HR for quartile 2 was 0.80 (95% CI, 0.69 to 0.93), 0.62 for quartile 3 (95% CI, 0.52 to 0.74), and 0.51 for quartile 4 (95% CI, 0.41 to 0.63) compared with the lowest quartile. For younger adults, the HR for quartile 2 was 0.79 (95% CI, 0.46 to 1.35), 0.90 for quartile 3 (95% CI, 0.64 to 1.25), and 0.95 for quartile 4 (95% CI, 0.61 to 1.48) compared with the lowest quartile. Restricted cubic splines demonstrated a nonlinear association whereby more steps were associated with decreased risk of CVD among older adults. Conclusions: For older adults, taking more daily steps was associated with a progressively decreased risk of CVD. Monitoring and promoting steps per day is a simple metric for clinician–patient communication and population health to reduce the risk of CVD.