Purpose/Objective(s): Management of geriatric patients that develop cancer is a challenge given associated comorbidities, toxicity, and age‐related limited physiological reserve. The purpose of this project is to study treatment terminations (patients that stopped treatment prior to completion of their original schedule) and develop predictive models using machine learning (ML) to achieve best outcomes in geriatric patients. Materials/Methods: As a part of initial assessment on treatment terminations, we evaluated patients 80+ y treated 2013‐18. Regarding ML, data that were included was patient demographics, tumor characteristics, treatment data, and toxicity. This information was quantified to form vectors and used as input features to generate predictive models. Outcomes used for prediction included mortality within 6 months, 1, 3‐ and 5 y. Data to generate the model was collected using ONCORA, for patients treated in our department 2008‐18. The data were randomly divided into training and testing sets. The training set was used to optimize the parameters used in the model. The test data were used to ensure that the model is not overfitting the data and determining the accuracy and sensitivity of the model. Support vector machine (SVM) and K nearest neighbors (KNN) algorithms were utilized. The models were compared to logistic regression and the Kaplan Meier curve. The accuracy and sensitivity of the models were analyzed using the area under ROC curve. Results: As the initial part of our evaluation regarding treatment terminations in geriatric patients, a total of 164 patients 80+ y discontinued treatment. Ninety‐one patients terminated treatment because of death or transfer to hospice. Fifty‐one (56%) patients were scheduled for RT with curative intent. Thirty‐eight (41.7%) terminations were because patients expired, while 49 (53.8%) were sent to hospice. In this group of 164 patients that terminated treatment, a logistic regression model achieved 84% accuracy with an F score of 91%, in predicting whether patients will discontinue treatment. We then extracted data of 2362 patients treated 2008â€“18 (age >75 y) to generate a model for prediction of death after RT. Our preliminary results show that the KNN algorithm achieves a 99% confidence score with in predicting outcomes for death and mortality within 6 months, 1, 3 and 5 years. Furthermore, the SVM in conjunction with a radial basis function achieves a 99% confidence score for predicting mortality within 6 months. Conclusion: Our preliminary study using Artificial intelligence (ML) offers encouraging results in predicting death after RT in this geriatric population. We believe that this work will result in further enhancing machine learning tools for clinical decisions by physicians in older population, to achieve best outcomes.
Guest, D. S., Polce, S. A., Rana, Z. H., Bloom, B. F., Cao, Y., Potters, L., & Parashar, B. (2019). Predicting Clinical Outcomes for Geriatric Patients after Radiation Therapy Treatment Using Artificial Intelligence (Machine Learning). International Journal of Radiation Oncology*Biology*Physics, 105(1), E143–E144. https://doi.org/10.1016/j.ijrobp.2019.06.2192