Identifying Language Features Associated With Needs of Ovarian Cancer Patients and Caregivers Using Social Media

Online health communities (OHCs) can be a source for clinicians to learn the needs of cancer patients and caregivers. Ovarian cancer (OvCa) patients and caregivers deal with a wide range of unmet needs, many of which are expressed in OHCs. An automated need classification model could help clinicians more easily understand and prioritize information available in the OHCs.

The aim of this study was to use initial OHC postings to develop an automated model for the classification of OvCa patient and caregiver needs.

We collected data from the OvCa OHC and analyzed the initial postings of patients and caregivers (n = 853). Two annotators coded each posting with 12 types of needs. Then, we applied the machine learning approach with bag-of-words features to build a model to classify needs. F1 score, an indicator of model accuracy, was used to evaluate the model.

The most reported needs were information, social, psychological/emotional, and physical. Thirty-nine percent of postings described information and social needs in the same posting. Our model reported a high level of accuracy for classifying those top needs. Psychological terms were important for classifying psychological/emotional and social needs. Medical terms were important for physical and information needs.

We demonstrate the potential of using OHCs to supplement traditional needs assessment. Further research would incorporate additional information (eg, trajectory, stage) for more sophisticated models.

This study shows the potential of automated classification to leverage OHCs for needs assessment. Our approach can be applied to different types of cancer and enhanced by using domain-specific information.