Comment on “The association between cognitive function trajectories and all-cause mortality in middle-aged and older Chinese adults with cardiovascular disease: A longitudinal study from CHARLS”

Dear Editor,

We read with interest the study by Song et al. [1]. The longitudinal analysis is valuable, linking cognitive trajectories to mortality in CVD patients and highlighting the prognostic role of dynamic assessment. However, some methodological and interpretative aspects merit discussion.

First, the trajectory modeling lacked thorough comparative validation. The use of a DTW-based K-means algorithm was not benchmarked against modern techniques (e.g., deep learning). Given DTW's sensitivity to noise and high computational cost, especially with only three time points, the robustness and optimality of the identified patterns are unclear. The choice of K = 4 over more parsimonious models lacked full statistical justification and external validation. Comparative studies show deep learning methods offer superior stability and accuracy (e.g., clustering stability +32%, accuracy decline 5.7% vs. 18.3% under noise) [2].

Second, the analysis did not adequately address time-varying confounders. While cognitive trajectories were defined using 2011-2015 data, mortality follow-up extended to 2020. Factors like medication adherence changes, incident stroke, or heart failure progression during this period were not controlled for, potentially biasing the causal inference. Studies confirm these are major confounders; controlling for them can substantially alter risk estimates (e.g., reducing a mortality risk difference from 42% to 25%) [3].

Finally, the discussion of subgroup interactions (e.g., with hypertension and BMI) lacked depth. The nonlinear relationship with BMI and the heightened risk in hypertension require more mechanistic exploration, considering factors like sarcopenia, fat distribution, specific antihypertensive drug classes, and cerebral small vessel disease, which were not measured [4,5].

Future studies would benefit from robust algorithm comparisons, methods to address time-varying confounding, and integration of multimodal data to clarify mechanisms and enhance clinical translation.

CRediT authorship contribution statement

Lipeng Zhao: Writing – original draft. Manrui Wang: Conceptualization, Methodology, Supervision, Validation.

Funding

None.

Declaration of competing interest

Greetings! We hereby formally declare that there are no conflicts of interest associated with the submission of this manuscript. The manuscript has been thoroughly reviewed and approved by all authors for publication.

On behalf of my co-authors, I affirm that the work described in this manuscript is original and has not been published previously, in whole or in part. It is also not under consideration for publication elsewhere. All the authors listed have reviewed and approved the enclosed manuscript.

We would be extremely grateful if you could consider our manuscript for publication. I am eagerly looking forward to your prompt response.