Tu1030 A Novel Algorithm for Predicting Patient Absenteeism for Gastroenterology Procedures

Abstract

Background: Patient absenteeism for scheduled gastrointestinal (GI) procedures occurs frequently in the U.S. healthcare system. When patients do not arrive for procedure appointments (no-show), resources allocated to that appointment are not utilized, resulting in a financial loss. Further, the no-show patient occupies space in the patient queue and delays access to diagnostic and screening procedures. We sought to demonstrate how a predictive model can accurately identify patients at high risk for no-show. Methods: As part of a two-year prospective study that addresses high rates of patient no-show in the Veterans Administration (VA) healthcare network, we identified predictors of absenteeism and developed an algorithm that identifies patients at risk for no-show. Determinants of absenteeism were obtained from the VA's Computerized Patient Record System (CPRS), and these predictors were entered in a forward stepwise logistic regression model to measure the adjusted independent contribution of each clinical variable. From this empirical analysis, each patient was assigned a composite risk score representing the probability that the patient will noshow for a procedure. Results: Data collected on 1397 VA patients indicated that the following variables independently predicted patient no-show for a GI procedure: previous absenteeism for GI appointments; ratio of cancelled appointments to all appointments scheduled for any medical procedure; low socioeconomic status; comorbid medical conditions, as measured by the Charlson Comorbidity Index; and diagnoses of mood, anxiety, conduct, personality, or substance use disorders during the last three years. History of completed GI procedures was a positive predictor of attendance for subsequent GI appointments. Using a predicted probability of missing an appointment of 0.40 as a cutoff, the model predicted missing or making appointments with 85% accuracy. The predictive model had a sensitivity of 40%, a specificity of 96%, a positive predictive value of 67%, and a negative predictive value of 86%. Although the model was not highly sensitive, our goal was to minimize the number of patients predicted to no-show who successfully attend their appointments. Conclusion: A predictive model that uses information from electronic health records can predict whether a patient will attend a GI appointment with a high degree of accuracy. We are currently applying this model prospectively to identify patients at high risk for no-show. The no-show appointment slots are made available to patients waiting in the scheduling queue, allowing them to be seen more quickly than with standard booking. We will test whether this system can reduce the financial losses created by GI procedure absenteeism by improving resource utilization and the efficiency of patient care. (Funded by VA HSR&D Merit Award)