Non-exercise estimated cardiorespiratory fitness and progressive frailty trajectories: a longitudinal cohort study

Technical Summary: Non-exercise Estimated Cardiorespiratory Fitness and Progressive Frailty Trajectories

Problem Statement
Frailty is a dynamic geriatric syndrome characterized by reduced physiological reserve and increased vulnerability to stressors, leading to adverse outcomes such as disability, hospitalization, and death. While the Frailty Index (FI) is a robust tool for assessing vulnerability, previous research has largely treated frailty as a static state rather than a dynamic trajectory of health-deficit accumulation. Furthermore, identifying individuals at high risk for accelerated frailty progression remains challenging. Although Cardiorespiratory Fitness (CRF) is a powerful predictor of mortality and health, direct measurement requires exercise testing, which is often infeasible in large-scale surveys and primary care settings. Consequently, there is a need for scalable, non-invasive indicators like Non-exercise Estimated Cardiorespiratory Fitness (NEE-CRF) to predict longitudinal frailty progression, particularly among individuals who are currently non-frail.

Methodology
This study utilized a longitudinal cohort design based on four waves of data (2011–2018) from the China Health and Retirement Longitudinal Study (CHARLS).

• Participants: The analytical sample included 4,497 adults aged 45 and older who were non-frail (FI < 0.25) at baseline and had complete data for NEE-CRF calculation and valid FI measurements in at least three survey waves.
• Exposure (NEE-CRF): Baseline NEE-CRF was calculated using validated sex-specific regression formulas incorporating age, BMI, waist circumference, resting heart rate, physical activity level, and smoking status. It was analyzed as both a continuous variable and in quartiles.
• Outcome (Frailty Trajectories): Frailty was assessed using a 32-item Frailty Index (FI). Group-based trajectory modeling (GBTM) was employed to identify distinct longitudinal patterns of frailty progression over the 7-year period.
• Statistical Analysis:
  • Regression: Multivariable logistic regression (unadjusted, demographic-adjusted, and fully adjusted for lifestyle and biomarkers) was used to estimate the association between NEE-CRF and trajectory membership. Restricted cubic splines (RCS) were used to assess dose-response relationships.
  • Machine Learning: Seven algorithms (Decision Tree, KNN, Random Forest, Elastic Net, Ridge Regression, XGBoost, SVM) were trained to predict trajectory class membership. Feature selection was performed using the Boruta algorithm.
  • Interpretability: The best-performing model was interpreted using SHapley Additive exPlanations (SHAP) to quantify feature importance and directionality.
  • Validation: Sensitivity analyses were conducted by relaxing inclusion criteria (requiring only two valid waves), and model performance was evaluated using AUC, calibration curves, and decision curve analysis (DCA).

Key Results

• Trajectory Identification: GBTM identified two distinct frailty trajectories: a "low-risk" trajectory (84.7% of participants) characterized by stable, low levels of health deficits, and a "progressive" trajectory (15.3%) characterized by rapid accumulation of deficits.
• Association with NEE-CRF: Higher baseline NEE-CRF was significantly associated with a lower likelihood of belonging to the progressive trajectory.
  • In the fully adjusted model, each 1-unit increase in NEE-CRF reduced the odds of progressive frailty by 27.6% (OR = 0.724, 95% CI: 0.680–0.770).
  • Participants in the highest NEE-CRF quartile (Q4) had substantially lower odds of progressive frailty compared to the lowest quartile (Q1) (OR = 0.243, 95% CI: 0.171–0.345).
  • RCS analysis indicated an approximately linear inverse association, with no significant nonlinearity.
• Machine Learning Performance: Ridge regression achieved the highest discriminative ability (AUC = 0.711) among the seven models. SHAP analysis identified NEE-CRF as the most influential predictor, surpassing sociodemographic variables and biomarkers.
• Robustness: Sensitivity analyses confirmed the robustness of the findings, with similar effect sizes observed when including participants with valid data in only two waves.

Significance and Claims
The authors claim that this study provides critical evidence that NEE-CRF is an independent protective factor against accelerated frailty progression in middle-aged and older Chinese adults, even among those who are initially non-frail.

• Clinical Utility: The study posits that NEE-CRF offers a feasible, low-resource strategy for early risk stratification. Because it relies on routinely collected variables (anthropometrics, vital signs, and self-reported activity) rather than exercise testing, it is particularly suitable for community health services and primary care settings where resources are limited.
• Predictive Value: By integrating NEE-CRF into interpretable machine learning frameworks, the study demonstrates that fitness-related indicators can meaningfully contribute to algorithmic risk assessment, potentially allowing for targeted prevention strategies (e.g., physical activity programs, metabolic management) for high-risk individuals before frailty becomes clinically apparent.
• Methodological Contribution: The research advances the field by moving beyond cross-sectional frailty classification to characterize dynamic trajectories and by combining traditional epidemiological methods with interpretable machine learning to enhance the translational value of fitness metrics.

The authors maintain a modest tone regarding causal inference, acknowledging the observational nature of the study and the need for external validation in independent cohorts before clinical deployment. They emphasize that NEE-CRF should complement, rather than replace, comprehensive geriatric assessments.