Skip to main content
An official website of the United States government
Grant Details

Grant Number: 1R01CA285737-01A1 Interpret this number
Primary Investigator: Jarad, Samah
Organization: Yale University
Project Title: Quantifying Patient-Provider Communication Using Machine Learning to Assess Its Impact on Metastatic Cancer Patients' Outcomes
Fiscal Year: 2024


Abstract

About two million persons are diagnosed with cancer every year in the US. The Effective clinical management of cancer patients including the use of oral therapies can improve outcomes and reduce costs. However, adherence rates to oral therapies in cancer treatment have been reported to be 20%, depending on the drug.2 Poor Patient-provider communication (PCC) is one of the barriers to adherence to oral chemotherapies in addition to adverse drug events and lack of knowledge about adherence. Timely and effective PPC with high levels of empathy shown by providers to address patients’ emotional concerns, wants, and needs can enhance patients’ trust in healthcare providers, and their clinical outcomes such as adherence and emergency services utilization. New digital health platforms such as secure messaging (SM) through patient portals can provide an effective and timely channel of Electronic Patient-Provider Communication (EPPC). Patients with and without cancer are increasingly using secure messaging to communicate their needs. As of May 2019, out of 1M patients that visited Yale New-Haven Health System (YNHHS), about 436,000 patients used the patient portal resulting in more than 2.7M messages. With the identification and quantification of EPPC in SM contents, we can measure associations and impact on patient-centered outcomes. However, existing studies to identify EPPC patterns in large scale SM data are limited as they focused only on patients’ messages and minimally included providers’ messages. Some studies are not scalable as they manually coded EPPC patterns for a small set of SM. We will fill this gap using natural language processing and machine learning approaches that will utilize big SM data. In our previous work we mapped expressions in SM into EPPC codes of communication using the Roter Interaction Analysis System (RIAS); a method to code medical interactions and developed the Electronic Patient- Provider Communication miner (EPPCminer) tool. It can detect three EPPC codes: information seeking and giving, socio-emotional behavior. In this study, we will (1) refine EPPCminer to more effectively extract more granular EPPC codes of information seeking (e.g., medication-, lab-, imaging-related), information giving of social determinants of health (e.g., transportation, economic concerns, food), socio-emotional behavior, partnership building, and shared decision-making. We will refine EPPCminer using SM from YNHHS and Cleveland Clinic and evaluate generalizability using Veterans Administration (VA) data. partnership building, and shared decision-making. We will also assess and score quality of bi-directional communication by examining providers’ responses to patients’ requests in SM. (2) We will apply EPPCminer to SM of a cohort of patients with different types of metastatic. We will then extract and characterize EPPC codes from 1 year worth of prospective patients’ and providers’ SM and examine their associations with scores of patient-reported communication assessments. (3) We will assess the impact of EPPC codes on patients’ outcomes: adherence using pharmacy data and patient-reported adherence data, emergency room (ER) visits, and hospitalizations.



Publications


None


Back to Top