Study of Patient and Physician Attitudes toward an Automated Prognostic Model for Patients with Metastatic Cancer

Abstract

Listen

Translate article

<h3>Purpose/Objective(s)</h3> For patients with metastatic cancer and their doctors, prognosis is important for choosing treatments and supportive care. Oncologists' life expectancy estimates are often inaccurate, and many patients are not aware of their general prognosis. Machine learning (ML) survival models have become accurate and could be useful in the clinic, but there are potential concerns involving accuracy, provider training, and patient involvement. We conducted a qualitative study to learn about patients' and oncologists' views on these issues. <h3>Materials/Methods</h3> In this Institutional Review Board-approved study, patients with metastatic cancer and their family members, radiation oncologists, and medical oncologists were recruited from a single academic health system. Semi-structured interviews were conducted using a script. Participants were shown an anonymized report from a validated ML survival model that included a predicted Kaplan-Meier survival curve and a list of the top features increasing and decreasing predicted survival. <h3>Results</h3> Fifteen patients, five radiation oncologists, and five medical oncologists were enrolled. Patients ranged in age from 54 to 87, with head and neck being the most common primary site (n=8). Physicians practiced in both hospital-based and community outreach clinics, were either generalists or specialized in a specific disease site, and ranged in years of practice from 5 to 29. Six patients (40%) had discussed their prognosis with a physician. Six (60%) of the 9 patients who had not discussed their prognosis with a physician stated they wanted to have this conversation. Of the 14 patients who completed the full interview, all wanted the option to have the model used in their care, and 11 (79%) believed they had the right to know if the physician used it. 11 (79%) stated they would want to see the report if the model was run on them. All 5 medical oncologists, and 3 (60%) radiation oncologists said they typically inform patients if their cancer is incurable, usually at the first visit. Four of ten physicians (40%) typically discuss detailed prognosis. All physicians interviewed said they thought the ML model was valuable and would want the option to use it in clinic. Only 1 (20%) radiation oncologist believed patients should be notified if their physicians used the model, compared to 4 (80%) medical oncologists. Detailed thematic analysis will be presented at the meeting. <h3>Conclusion</h3> Patients and physicians were supportive of the use of an automated prognostic model in their care. Most patients wished to have conversation about prognosis, although only a minority of oncologists routinely discuss prognosis. These data demonstrate a need for oncologists to be more willing to have these conversations and suggest a role for ML models to help facilitate them.