Beyond Time-Based Billing: An AI-Enabled Complexity Model for Secure Message Reimbursement

Abstract

Background

: Current telehealth reimbursement guidelines, based primarily on evaluation and management duration, inadequately capture the true clinical complexity and cognitive effort involved in provider responses to secure patient messages. There is a critical need for reimbursement frameworks that reflect the nuanced reality of clinical care delivered through asynchronous telehealth messaging.

Materials and Methods

: We analyzed 149,499 secure messages (2023–2024) from a large health system. Given the impracticality of manually annotating all threads, we used supervised pseudo-labeling informed by rigorous annotation guidelines (interrater reliability: Cohen’s κ = 0.816) applied to a representative subset. Clinical complexity was quantified through features of providers’ electronic health records engagement (Domain Engagement [DE]) and message linguistic complexity (Cognitive Judgment [CJ]). A gradient boosting classifier leveraging these complexity-based features was developed and validated via stratified cross-validation, temporal validation, and leave-one-provider-out testing. Interpretability was assessed using permutation importance and Shapley Additive Explanations values.

Results

: The complexity-based model achieved strong predictive performance (area under the curve [AUC] ∼0.82), substantially outperforming a time-based baseline (AUC ∼0.57). Key predictors included unique medical concepts, breadth of clinical content, and emotional urgency. Traditional metrics like message length and duration were poor predictors. Billable threads showed significantly higher medical density (∼55 vs. ∼33 concepts) and emotional intensity (∼0.6 vs. ∼0.3), supporting DE and CJ validity.

Discussion

: Complexity-based features more accurately captured providers’ cognitive effort than elapsed time alone, consistent with Cognitive Load, Complexity, and Sensemaking theories.

Conclusion

: This AI-enabled complexity-based model provides a practical, accurate, and explainable foundation for telehealth billing reform, facilitating fairer provider reimbursement and improved telehealth sustainability.