Comparison of Algorithms to Triage Patients to Express Care in a Sexually Transmitted Disease Clinic

Abstract

The ideal approach to triaging sexually transmitted disease (STD) clinic patients between testing-only express visits and standard visits with clinician evaluation is uncertain. In this cross-sectional study, we used classification and regression tree analysis to develop and validate the optimal algorithm for predicting which patients need a standard visit with clinician assessment (i.e., to maximize correct triage). Using electronic medical record data, we defined patients as needing a standard visit if they reported STD symptoms, received any empiric treatment, or were diagnosed as having an infection or syndrome at the same visit. We considered 11 potential predictors for requiring medical evaluation collected via computer-assisted self-interview when constructing the optimized algorithm. We compared test characteristics of the optimized algorithm, the Public Health-Seattle and King County STD Clinic's current 13-component algorithm, and a simple 2-component algorithm including only presence of symptoms and contact to STD. From October 2010 to June 2015, 18,653 unique patients completed a computer-assisted self-interview. In the validation samples, the optimized, current, and simple algorithms appropriately triaged 90%, 85%, and 89% of patients, respectively. The optimized algorithm had lower sensitivity for identifying patients needing standard visits (men, 94%; women, 93%) compared with the current algorithm (men, 95%; women, 98%), as did the simple algorithm (men, 91%; women, 93%). The optimized, current, and simple algorithms triaged 31%, 23%, and 33% of patients to express visits, respectively. The overall performance of the statistically optimized algorithm did not differ meaningfully from a simple 2-component algorithm. In contrast, the current algorithm had the highest sensitivity but lowest overall performance.