Re: Interpreting Cannabis Use and Postoperative Pain in Older Adults: Methodologic Considerations

To the Editor,

We thank Khan and colleagues for their interest in our study¹ and for underscoring the importance of understanding cannabis use in older surgical patients. We welcome the opportunity to clarify several aspects of our design, analysis, and interpretation.

Our objective was to quantify associations between documented preoperative cannabis use and acute postoperative pain and opioid requirements in older adults. Consistent with this aim, we presented all findings as associations and acknowledged the limits of observational data. We agree that state policy and disclosure patterns may influence identification of cannabis users; however, our cohort was not restricted to medical cannabis use, and the prevalence of current use was similar to national estimates for older adults,² suggesting broader representativeness. After propensity-score matching, chronic pain conditions, comorbidities, and baseline pain scores were well balanced, reducing but not eliminating confounding, which we reflected in our interpretation.

We agree that under-documentation can lead to exposure misclassification and noted this explicitly. Such misclassification is likely non-differential with respect to postoperative pain and would bias results toward the null. Regarding external validity, our prior NLP work emphasized that the algorithm was developed and validated within UF Health and would require local calibration in other systems.³ We concur that multisite validation of both the NLP method and the observed associations is an important next step.

We selected a 60-day window because it reflects the typical timeframe for preoperative evaluations at our institution and the manner in which “current use” is documented in perioperative practice. Although acute pharmacodynamic effects are shorter, our research questions focused on patterns of chronic or ongoing use rather than isolated recent exposure. Documented use within this window generally captures persistent or recurrent use rather than remote experimentation. As noted in the manuscript, we did not assess the timing of last use relative to surgery. Documentation in clinical notes was supplemented with ICD codes for cannabis use disorders and cannabis-related poisoning. We agree that future studies should incorporate finer temporal resolution and distinguish between chronic and intermittent use.

More granular exposure information, such as product type, potency, route, frequency, and duration, would undoubtedly strengthen future work. EHR systems rarely capture these characteristics reliably, which we noted as a limitation.

We also appreciate the reference to the target trial emulation framework. This approach is valuable for questions involving assignable interventions. In our study, cannabis use was a pre-existing exposure rather than an intervention initiated at a defined perioperative time point. Nonetheless, time zero for outcome assessment was defined as the end of surgery, with exposure and covariates ascertained in advance.

Residual confounding, under-documentation, and limited exposure detail are important considerations, and we explicitly acknowledged these issues. Within these real-world constraints, we believe our approach is appropriate to the clinical question, our conclusions are measured and associative, and the consistency with prior studies supports the robustness of our findings. We appreciate the authors’ insightful comments and agree that they highlight meaningful directions for future multi-site and mechanistic research.