Predicting patient returns due to complications and recommending follow-up appointments after a dental extraction using machine learning

After a dental extraction, knowing which patients are at higher risk of return due to complications may help plan a better treatment approach and lead to targeted follow-up appointments. The objective of this study was to predict which patients are at higher risk of return due to complications from a dental extraction using data from electronic dental records (EDRs). Structured and unstructured data in EDRs of 14,541 patients who had a dental extraction were used to train three types of machine learning models. Methods to address an anticipated class imbalance were also investigated. The primary evaluation measure was recall. Other measures included area under the receiver operating characteristic curve (AUC), precision, and F1-score. Follow-up appointment recommendations were compared to the clinical judgment of dental practitioners. Of the 14,541 patients who had a dental extraction, 488 patients (3.4%) returned due to complications from the dental extraction visit. The best performing classifier identified patients at risk of complications and recommended to return for a follow-up appointment with a recall of 0.53, compared to dental practitioners who scheduled follow-up appointments with a recall of 0.26. The overall performance was low indicating the presence of unmeasured confounders. The model demonstrated limited predictive power, suggesting that additional variables are necessary to determine if it is feasible to reliably predict which patients will return due to complications after a dental extraction. A possible solution may be to find ways to efficiently record standardised information about relevant risk factors in EDRs, supporting more targeted scheduling of follow-up appointments.