In:
Journal of Clinical Oncology, American Society of Clinical Oncology (ASCO), Vol. 36, No. 34_suppl ( 2018-12-01), p. 144-144
Abstract:
144 Background: Acute care accounts for half of cancer expenditures and is a measure of poor quality care. Identifying patients at high risk for ED visits enables institutions to target symptom management resources to those most likely to benefit. Risk stratification models developed to date have not been meaningfully employed in oncology, and there is a need for clinically relevant models to improve patient care. Methods: We established a predictive analytics framework for clinical use with attention to the modeling technique, clinician feedback, and application metrics. The model employs EHR data from initial visit to first antineoplastic administration for new patients at our institution from January 2014 to June 2017. The binary dependent variable is occurrence of an ED visit within the first 6 months of treatment. From over 1,400 data features, the model was refined to include 400 clinically relevant ones spanning demographics, pathology, clinician notes, labs, medications, and psychosocial information. Clinician review was performed to confirm EHR data input validity. The final regularized multivariate logistic regression model was chosen based on clinical and statistical significance. Parameter selection and model evaluation utilized the positive predictive value for the top 25% of observations ranked by model-determined risk. The final model was evaluated using a test set containing 20% of randomly held out data. The model was calibrated based on a 5-fold cross-validation scheme over the training set. Results: There are 5,752 antineoplastic starts in our training set, and 1,457 in our test set. The positive predictive value of this model for the top 25% riskiest new start antineoplastic patients is 0.53. The 400 clinically relevant features draw from multiple areas in the EHR. For example, those features found to increase risk include: combination chemotherapy, low albumin, social work needs, and opioid use, whereas those found to decrease risk include stage 1 disease, never smoker status, and oral antineoplastic therapy. Conclusions: We have constructed a framework to build a clinically relevant model. We are now piloting it to identify those likely to benefit from a home-based, digital symptom management intervention.
Type of Medium:
Online Resource
ISSN:
0732-183X
,
1527-7755
DOI:
10.1200/JCO.2018.36.34_suppl.144
Language:
English
Publisher:
American Society of Clinical Oncology (ASCO)
Publication Date:
2018
detail.hit.zdb_id:
2005181-5
Bookmarklink
