A Data-Driven Algorithm to Recommend Initial Clinical Workup for Outpatient Specialty Referral: Algorithm Development...

Many patients lack timely specialty care due to incomplete preliminary evaluations. Current methods rely on manual consensus-building or specialist availability, which are inefficient. Prior research has shown that incomplete diagnostic workups before referral can delay diagnosis and treatment. However, no prior work had resolved how to automate this process using data-driven approaches. This gap motivated the development of a system that could anticipate diagnostic needs without requiring specialist input at referral time. It was already known that pediatric endocrinology referrals often lack baseline testing. No existing tools had been validated for predicting initial workup orders in this context. This study addresses the need for scalable solutions in outpatient specialty care. The focus is on improving diagnostic readiness through automated recommendations.

The aim of this study was to develop and validate a data-driven algorithm to recommend initial diagnostic workups for outpatient specialty referrals. The specific problem is the inefficiency of current referral processes, which depend on specialist availability or consensus. The motivation is to reduce diagnostic delays by proactively suggesting appropriate tests. The study targets pediatric endocrinology referrals as a model system. The algorithm was designed to predict orders specialists would likely request. The goal is to improve the completeness of prereferral evaluations. This approach could streamline specialty care access. The study also evaluates the clinical appropriateness of the algorithm's output.

The researchers used electronic health record data from 3424 pediatric patients with new endocrinology referrals. They extracted data from 2015 to 2020 at an academic institution. Item co-occurrence statistics were applied to identify patterns in diagnostic orders. A holdout dataset was used to assess the algorithm's performance against actual specialist orders. Endocrinologists were surveyed to evaluate the clinical appropriateness of the predicted orders. The algorithm's recommendations were compared to a benchmark of the most common orders. Precision and recall metrics were calculated to measure performance. The study combined data analysis with expert feedback to refine the algorithm's output.

The algorithm achieved an area under the ROC curve of 0.95 (95% CI 0.95-0.96). Precision improved from 37% to 48% (P<.001) for the top 4 recommendations. Recall increased from 27% to 39% (P<.001) for the same set. Specialists surveyed indicated that less than 50% of referrals arrive with complete initial workup. The top 4 recommendations for common conditions like obesity and amenorrhea were deemed clinically appropriate by most specialists. The algorithm outperformed the reference benchmark in both precision and recall. Predicted orders aligned closely with practice guidelines. The results suggest the algorithm can effectively anticipate diagnostic needs.

The authors propose that an item association-based algorithm can predict specialists' diagnostic orders with high accuracy. The results suggest that this approach could support clinical decision support tools. The algorithm's performance exceeded a reference benchmark using common orders. The study demonstrates the feasibility of data-driven recommendations for initial workup. The authors suggest that this could improve the effectiveness of specialty referrals. The findings imply that automated systems can reduce delays in diagnostic readiness. The results align with practice guidelines for pediatric endocrinology. The authors conclude that this is a promising step toward a new paradigm in outpatient specialty care.