Click here to


Are you sure ?

Yes, do it No, cancel

Identifying Predictors of Unplanned Hospitalizations After Radiotherapy Using Regularized Survival Models

C Ahern1*, T Pheiffer1 , C Berlind1 , W Lindsay1 , Y Xiao2 , C Simone3 , (1) Oncora Medical, Inc., Philadelphia, PA, (2) University of Pennsylvania, Philadelphia, PA, (3) University of Maryland School of Medicine, Baltimore, Maryland


(Tuesday, 7/31/2018) 3:45 PM - 4:15 PM

Room: Exhibit Hall | Forum 9

Purpose: Unplanned hospitalizations due to radiotherapy toxicities can have a substantial impact on patient quality of life and are a major driver of the cost of cancer care. We use regularized survival models to identify and quantify potential factors associated with unplanned hospitalizations.

Methods: In this IRB-approved analysis, a dataset of 16,689 radiotherapy courses performed at one institution from 2008-2015 with over 230 variables per course was created from EMRs and treatment planning systems using automated extraction software. Unplanned hospitalizations were defined as urgent or emergency inpatient encounters. Time to hospitalization was defined as the number of days from first fraction to the first unplanned hospitalization after treatment start. Patients not hospitalized were treated as censored. A Cox proportional hazards (CPH) model with elastic net penalty was trained using 5-fold cross-validation on a 75-25 train-test split of the data, with performance measured by the concordance index.

Results: The model achieved a concordance index of 0.768 on the test set. 22.7% of the encoded features were eliminated from the model with coefficients of zero. Clinical features most associated with unplanned hospitalization were notable symptoms prior to start of treatment (hazard ratio 1.188), prior chemotherapy (1.173), black race (1.141), a malignant neoplasm of unspecified site (1.125), lung cancer (1.101), and prior chemotherapy (1.091). Features associated with lack of hospitalization were breast (0.774) and prostate (0.871) as treatment site, modality of SBRT (0.866) and proton therapy (0.868), and total delivered dose (0.870).

Conclusion: The regularized model identified prior chemotherapy and treatment disease site, as well as treatment modality as important factors predicting for unplanned hospitalization. This model can be used to provide pre-emptive interventions to avoid hospitalizations. Multi-center validation of these findings is needed.

Funding Support, Disclosures, and Conflict of Interest: CA, TP, CG, and WL are employed by and own equity in Oncora Medical, Inc.


Statistical Analysis, Feature Extraction


TH- Dataset analysis/biomathematics: Machine learning techniques

Contact Email