Human-Machine Collaborations to Improve Prognosis and Clinical Decision-Making in Advanced Cancer

PROJECT SUMMARY/ABSTRACT Advance care planning and palliative care represent evidence-based, high-quality care for patients with advanced cancer. Early identification of patients at risk of short-term mortality is a promising strategy to increase advance care planning and palliative care. However, this is limited by prognostic inaccuracy among

oncology clinicians, who overestimate prognosis for 70% of their patients with advanced cancer. While recent advances in electronic health record (EHR) infrastructure and machine learning (ML) have allowed accurate identification of patient' mortality risk, there is a fundamental gap in understanding how to integrate ML

prognostic algorithms alongside clinician intuition (“human-machine collaborations”) in the routine care of patients with cancer. Dr. Parikh's research objective is to develop and test human-machine collaborative systems that leverage ML algorithms to improve clinicians' prognostic accuracy in order to prompt earlier

advance care planning and palliative care among patients with advanced cancer. In prior work, Dr. Parikh has prospectively validated and embedded into the EHR an automated ML algorithm to predict short-term mortality risk among patients with cancer. In this application, Dr. Parikh proposes to take a fundamental next step in this

work by exploring strategies to improve prognostic accuracy and decision-making among oncologists treating patients with advanced cancer. In Aim 1, Dr. Parikh will retrain and validate the existing ML mortality risk prediction algorithm by integrating recently-available patient-generated health data. In Aim 2, Dr.

develop prognostic that Parikh will a vignette-based survey to assess optimal strategies of presenting ML predictions to improve accuracy. He will administer this survey to a large national sample of medical oncologists to ensure clinician perspectives are incorporated into interventions.In Aim 3, Dr. Parikh will develop two models of

human-machine collaborative systems to generate real-time mortality estimates that integrate clinician and algorithm predictions. In a pragmatic multi-institutional clinical trial among patients with advanced cancer, Dr. Parikh will test the impact of human-machine collaborations on prognostic accuracy and rates of advanced

care planning and palliative care referral. These findings will have important implications for patients with cancer, their caregivers, oncology clinicians, and health systems. More broadly, the methods proposed may serve as a blueprint to develop and evaluate human-machine collaborations in oncology. This

facilitate judgment highly-qualified Dr. development improving research will t raining in areas vital to Dr. Parikh's career goals: dvanced predictive modeling, survey methods and and decision-making, human-machine interfaces, and pragmatic clinical trials. Dr. Parikh has two and committed mentors: Dr. Justin Bekelman, an expert i n cancer care delivery r esearch, and

Jinbo Chen, an expert in EHR-based predictive model development. The proposed research and career plan will enable Dr. Parikh to transition to an independent physician-scientist devoted to the quality and applicability of predictive analytics in the care of patients with cancer. a