Response to Xu et al: Time-to-event analysis,data variability,and consideration of model selection: Considerations in relation to stroke-heart syndrome

We appreciate the interest of Xu et al. in our recently published article on “stroke-heart syndrome, focused on the incidence and clinical outcomes of cardiac complications following intracerebral haemorrhage.” ¹

We thank them for their thoughtful comments and for raising important considerations regarding the use of Cox proportional hazards (CoxPH) model in our analysis.

Xu et al. suggest that our study may involve mixed censoring outcomes and that the assumption of constant hazard ratios over time may not hold, potentially affecting the validity of the CoxPH model. We acknowledge that the CoxPH model has limitations when the data exhibit significant deviations from the proportional hazards assumption. However, we carefully evaluated the model’s appropriateness for our dataset, particularly given its widespread use in similar clinical research contexts to handle time-to-event data.

In their letter, Xu et al. suggest alternative approaches, such as the Accelerated Failure Time (AFT) model, to handle different types of censored data (right, left, and interval-censored data). While we agree that the AFT model can be beneficial in certain scenarios, our dataset primarily included right-censored observations, where the time to event was either observed or right-censored. Further, the proportional hazards assumption can be easily tested in R using the “survival” package and “coxph” function. Here, generalized Schoenfeld approach can be used, which showed no significant violations. Therefore, we believe the CoxPH model remains a valid approach for this dataset. We appreciate the suggestion to evaluate alternative models if the proportional hazards assumption is violated. Indeed, where necessary, stratification of the cohort and use of time-varying covariates can be beneficial, yet more difficult to collect and handle statistically.

We agree that future studies may benefit from incorporating more flexible modeling techniques, such as the AFT model or other parametric survival models, especially in the presence of more complex censoring or non-proportional hazards. As suggested, authors may consider these models in future research to provide a comprehensive understanding of the relationship between exposures and outcomes, especially with long-term follow-up and covariates that likely change over time.

In conclusion, we appreciate the constructive feedback from Dr. Xu and colleagues and agree that ongoing research should explore alternative methods to validate and expand upon our findings. Their comments highlight the importance of careful model selection and validation in time-to-event analyses, particularly in clinical settings where the nature of the data may vary. The bottom line remains that stroke-heart-syndrome is an important clinical entity that merits holistic or integrated care management approaches, ² to mitigate the adverse outcomes associated with this under-diagnosed condition.^3,4