Impact of Augmented Intelligence on Utilization of Palliative Care Services in a Real-World Oncology Setting

Abstract

PURPOSE: For patients with advanced cancer, timely referral to palliative care (PC) services can ensure that end-of-life care aligns with their preferences and goals. Overestimation of life expectancy may result in underutilization of PC services, counterproductive treatment measures, and reduced quality of life for patients. We assessed the impact of a commercially available augmented intelligence (AI) tool to predict 30-day mortality risk on PC service utilization in a real-world setting. METHODS: Patients within a large hematology-oncology practice were scored weekly between June 2018 and October 2019 with an AI tool to generate insights into short-term mortality risk. Patients identified by the tool as being at high or medium risk were assessed for a supportive care visit and further referred as appropriate. Average monthly rates of PC and hospice referrals were calculated 5 months predeployment and 17 months postdeployment of the tool in the practice. RESULTS: The mean rate of PC consults increased from 17.3 to 29.1 per 1,000 patients per month (PPM) pre- and postdeployment, whereas the mean rate of hospice referrals increased from 0.2 to 1.6 per 1,000 PPM. Eliminating the first 6 months following deployment to account for user learning curve, the mean rate of PC consults nearly doubled over baseline to 33.0 and hospice referrals increased 12-fold to 2.4 PPM. CONCLUSION: Deployment of an AI tool at a hematology-oncology practice was found to be feasible for identifying patients at high or medium risk for short-term mortality. Insights generated by the tool drove clinical practice changes, resulting in significant increases in PC and hospice referrals.

Prospective Comparison of Medical Oncologists and a Machine Learning Model to Predict 3-Month Mortality in Patients With Metastatic Solid Tumors

“…Several publications evaluating SQs and even clinical ML did not report prevalence with PPV by omitting it, leaving its extrapolation to the reader or presenting results close to random guessing (ie, with PPV close to prevalence) as instances of good performance. 27 , 28 , 29 , 49 …”

“…Institutions have increasingly used machine learning (ML) to identify patients at high risk of mortality at different points from 30 days to 5 years for activating care teams to conduct goals-ofcare discussions and engage palliative care. [42][43][44][45][46][47][48][49] Although there are multiple retrospective evaluations of ML models, there remain few prospective evaluations and even fewer prospective comparisons of clinician and ML predictions. 42,43,46,47,50,51 We compared the prognostic performance of medical oncologists using an SQ with a supervised model trained to predict the risk of 3-month mortality.…”

Impact of social determinants of health on cancer care: a survey of community oncologists

“…28 Finally, in a recent pilot study, an AI decision tool that incorporated SDOH helped patients with cancer receive timely palliative care by identifying those who were at risk for short-term mortality. 29 The physicians in our study viewed the responsibility for assisting patients with social needs as belonging to government organisations, non-profit organisations, pharmaceutical companies, hospitals and commercial payers. In the past decade, these entities have made important strides towards mitigating the impact of SDOH on clinical outcomes through policy changes, commitments for community programmes and initiatives to support individual patients that go beyond screening.…”

Advanced Analytics as an Accelerator for Palliative Care and Oncology Integration

“…In the companion to this article, Gajra et al 9 report experiences using a commercially available advanced analytics approach, specifically applying augmented intelligence, to identify patients at high risk for 30-day mortality and then deploying specialty palliative care to them. The algorithm incorporates usual administrative and clinical health data (eg, electronic health record and claims) alongside socioeconomic (US Census and NOAA) and behavioral (eg, purchasing history from credit bureaus) information to create a mortality score, displayed as high (top 50 of all patients) or medium (next 100 patients).…”

“…10 Another study in the Journal of the American Medical Informatics Association showed that a machine learning model was more accurate in predicting prognosis than oncologist estimations of survival for patients with metastatic cancer. 11 What made this model described by Gajra et al 9 novel was the incorporation of electronic record note text and laboratory or vital signs data beyond the traditional performance status evaluation that dominates clinician-derived estimates. As more experiences are reported, future research will need to compare differing types of information and their presentations on provider behaviors toward referring patients to palliative care.…”