Radiology No-Show Calculator: A Social Determinants of Health-Enriched Machine Learning Prediction Model with Financial Analysis.

OBJECTIVE: Outpatient radiology appointment no-shows delay timely diagnosis and exacerbate healthcare disparities, while also resulting in a significant financial and operational burden on healthcare systems. Machine learning (ML) models offer a promising tool for identifying at-risk patients before no-shows occur. We sought to develop an ML model enriched with social determinants of health (SDOH) factors to predict outpatient radiology no-shows and evaluate the health equity and financial implications of targeted interventions. METHODS: In this IRB-approved retrospective case-control study at a large academic health system, we analyzed 32,776 outpatient radiology appointments from 9,998 patients (5,000 with at least one no-show during the study period, 4,998 without) between January 2023 and December 2024. The 1:1 case-control design was chosen to enrich the outcome for stable model training (population no-show prevalence 3.9%). A Random Forest model with 16 predictors, including the Area Deprivation Index (ADI), was developed and internally validated. Predicted probabilities were recalibrated to the population prevalence to form a deployable risk calculator. Generalized estimating equations (GEE) logistic regression quantified adjusted associations between SDOH factors and no-show risk. RESULTS: The Random Forest achieved an area under the receiver operating characteristic (AUROC) of 0.760. No-show rates demonstrated a clear Area Deprivation Index (ADI) gradient, ranging from 15.0% in the least deprived neighborhoods to 31.8% in the most deprived. The top 10% of highest-risk appointments captured 25.7% of all no-shows, with a number needed to intervene (NNI) of 1.8. The estimated yearly cost burden of missed appointments was $57.7 million. CONCLUSION: The radiology no-show calculator is a prediction model that uses SDOH factors to estimate the likelihood of missed appointments. The calculator demonstrates moderate discrimination for predicting no-shows and enables prospective targeted outreach. Future deployment of this model holds promise for improving health outcomes among socioeconomically vulnerable patients while lowering institutional cost burden.