Multidisciplinary clinician perceptions on utility of a machine learning tool (<scp>ALERT</scp>) to predict 6‐month mortality and improve end‐of‐life outcomes for advanced cancer patients

Abstract

BACKGROUND: There are significant disparities in outcomes at the end-of-life (EOL) for minoritized patients with advanced cancer, with most dying without a documented serious illness conversation (SIC). This study aims to assess clinician perceptions of the utility and challenges of implementing a machine learning model (ALERT) to predict 6-month mortality among patients with advanced solid cancers to prompt timely SIC. METHODS: One-on-one semi-structured interviews were conducted with oncology physicians, advanced practice providers, registered nurses, and social workers until knowledge saturation was reached (N = 19). Thematic analysis was conducted on the transcribed interviews, which were reviewed and coded by a team of interdisciplinary investigators. RESULTS: Clinician-perceived benefits were (1) guiding prognostication and the objectivity of the prediction easing clinician distress with EOL treatment planning; (2) standardizing prognosis discussions across specialties, limiting aggressive EOL procedures; (3) respecting patient values by providing them time to get affairs in order and plan for cultural EOL rituals; and (4) facilitating earlier SIC and palliative care referrals. Challenges identified were (1) integration of predictions with clinical expertise; (2) balancing the reliability and accuracy of the model with a rapidly evolving therapeutic landscape; and (3) concern about patient distress due to poor communication. CONCLUSIONS: Clinicians expressed widespread acceptability of ALERT and identified clear benefits, particularly in triggering earlier SIC and standardizing prognosis discussions across care teams to avoid aggressive hospital interventions at EOL. Challenges identified, including concerns regarding communication of the prediction and integration with clinical expertise and new research, will guide refinement of the ALERT model.

Ähnliche Arbeiten