When Cumulative Frailty Includes Future CVD Events: Time‐Order Concerns in CKM Syndrome Research

Dear Editor,

Yang and colleagues [1] deserve commendation for their longitudinal study examining the associations between frailty and cardiovascular risk across multiple dimensions, including baseline frailty status, cumulative frailty burden, and frailty trajectories, in individuals with CKM syndrome stages 0–3. The use of group‐based trajectory modeling to identify distinct frailty progression patterns represents a methodological strength. However, two methodological aspects warrant further scrutiny. Both concern the temporal ordering of exposure and outcome. A third interpretive consideration also merits attention.

First, the cumulative frailty index (cumFI) was calculated as the sum of FI values at baseline and three subsequent follow‐up assessments. The authors defined cumFI as the sum of FI values across all available time points [1]. In Cox models, the exposure must precede the outcome in time. For participants who experienced CVD during follow‐up, the cumFI includes FI measurements taken after the CVD event. For example, a participant who had a stroke in 2015 would have their 2018 FI value included in the cumFI. This effectively uses post‐outcome data to predict the outcome. This violates the temporal ordering assumption of survival analysis. Wolfe and colleagues [2] cautioned that survival models can yield misleading conclusions if time‐dependent factors are not carefully defined. They noted that one should “almost never use a covariate that has been averaged over a patient's entire follow‐up time as a baseline covariate” [2]. Instead, the cumulative average up to each point in time should be used as a time‐dependent covariate [2]. This distinction matters: using post‐event exposure information means the hazard ratio represents a retrospective association, not a prospective prediction. This issue is especially concerning given the strong association reported (Q4 vs. Q1: HR 8.46), which the temporal mismatch may partly explain. Penning de Vries and colleagues [3] showed that covariate measurement timing can substantially bias estimates of time‐varying exposure effects. We suggest two alternatives: conduct a sensitivity analysis using only FI measurements obtained prior to the CVD event, or use a time‐dependent cumulative exposure metric. The weighted cumulative exposure method, which assigns weights to past exposures according to their recency, is one such approach [4].

Second, the frailty trajectories identified by GBTM using all four waves of FI data were then used as predictors of incident CVD. GBTM was applied to FI measurements from 2011, 2013, 2015, and 2018. The resulting trajectory assignments were entered into Cox models to predict CVD events over the same period [1]. This approach uses the entire follow‐up to classify participants, then evaluates whether these groups predict outcomes from the same period. A participant's trajectory is unknown until all waves are available. In practice, we cannot know in 2013 that a patient will follow a “high‐level increasing trajectory” by 2018. Using future information to define exposure means the Cox model no longer estimates a prospective association. Nagin and colleagues [5] emphasized that GBTM describes longitudinal heterogeneity, not prospective prediction using post hoc assignments. Mésidor and colleagues [6] showed that GBTM can generate spurious findings, especially when classification adequacy is assessed using limited criteria. Serra and colleagues [7] further highlighted key methodological pitfalls in trajectory analyses, including the sensitivity of model selection and the interpretation of trajectory groups. The HR of 9.33 for the high‐level increasing trajectory may be inflated by this circularity. We suggest that the authors explore a truly prospective application. For example, using only baseline and 2013 data to predict outcomes after 2013 would more closely approximate clinical prediction, with subsequent waves used for validation. This would strengthen causal interpretability of the findings.

Third, the consistency across sensitivity analyses, while supporting robustness, also raises a conceptual question. The association between FI and CVD risk remained significant after excluding hypertension and diabetes from the FI, the very conditions that define CKM syndrome staging. This suggests the FI captures non‐specific physiological vulnerability rather than a CKM‐specific pathway. This does not diminish frailty's clinical utility but is worth explicit acknowledgment, as it may inform how clinicians interpret frailty scores in CKM syndrome management.

We propose three refinements. First, re‐evaluate cumFI by restricting to FI measurements that precede the CVD event, or use a time‐dependent cumulative exposure metric. Second, conduct a truly prospective trajectory analysis, using early‐wave data to derive trajectories and validating their predictive performance on later outcomes. Third, explicitly acknowledge that the FI may capture non‐specific physiological vulnerability rather than CKM‐specific pathology. Our comments are intended to enhance methodological transparency of this otherwise valuable contribution rather than to diminish its overall significance.

Author Contributions

C.Q. conceived the letter concept, performed the literature analysis, and drafted the manuscript. H.D. critically revised the manuscript for important intellectual content and provided supervision. All authors approved the final version and agree to be accountable for all aspects of the work.

Funding

The authors have nothing to report.

Ethics Statement

The authors have nothing to report.

Consent

The authors have nothing to report.

Conflicts of Interest

The authors declare no conflicts of interest.

Data Availability Statement

Data sharing not applicable to this article as no datasets were generated or analyzed during the current study.