Mortality predictions and behavioral nudges by machine learning systems are associated with increased advance care planning utilization and reductions in high-acuity care.
Researchers at the University of Pennsylvania recently studied the effectiveness of machine learning algorithms in identifying high-risk cancer patients nearing the last six months of life between June 17, 2019 and April 20, 2020. During that time period, clinicians received weekly lists of these patients generated by the system, as well as ongoing email and text message prompts sent to clinicians to initiate goals-of-care discussions.
The study’s results mark an important step in the role that artificial intelligence can play in improving end-of-life outcomes, according to researcher Dr. Ravi Parikh, oncologist and assistant professor of medical ethics and health policy and medicine at the University of Pennsylvania’s Perelman School of Medicine. Parikh is also associate director of the Penn Center for Cancer Care Innovation.
“This study demonstrates that we can use informatics to improve care at [the] end of life,” Parikh told local news. “Communicating with cancer patients about their goals and wishes is a key part of care and can reduce unnecessary or unwanted treatment at the end of life. The problem is that we don’t do it enough, and it can be hard to identify when it’s time to have that conversation with a given patient.”
These machine learning-based interventions led to increased rates of serious illness conversations among 13.5% of the 20,506 cancer patients examined, the study found. This was a “significant increase” compared to 3.4% of patients who held advance care planning conversations prior to deployment of the machine learning algorithms, researchers said.
Additionally, machine learning-based interventions were associated with reduced high-cost care utilization at the end of life among 10.4% of the cancer descendents studied. These cancer patients showed increased hospice enrollment and length of stay, fewer occurrences of inpatient deaths, less intensive care unit use within the last 30 days of life, and reduced use of systemic therapy two weeks before death (such as chemotherapy or inhibitor therapy).
The study’s results also suggest that predictive analytics could help improve health outcomes among underserved populations by increasing conversations around goals of care at the end of life, according to Parikh.
The clinician notifications led to increased advance care planning rates among 5.2% of all Black, Hispanic, American Indian, Asian and Pacific Islander Medicare beneficiaries in the study, compared to 0.9% without these interventions.
Researchers plan to dig further into the patient data to determine whether machine learning can yield similar results with palliative care referrals or impact awareness, education and communication regarding care options, according to Parikh.
“While we significantly increased the number of dialogues about serious illness taking place between patients and their clinicians, still less than half of patients had a conversation. We need to do better, because we know patients benefit when their health care clinicians understand each patient’s individual goals and priorities for care,” Parikh said.
Hospices have increasingly leveraged machine learning tools to identify patients in need of their services earlier in their illness trajectories and to help ensure patients receive appropriate levels of care.
These systems can use algorithms and statistical models to detect patterns in patient data from electronic medical records and other sources of information, helping providers predict probable changes in patient conditions.
Case in point, Palm Beach Accountable Care Organization (PBACO) recently told Healthcare Finance that machine learning and predictive analytics contributed to a 29% reduction in hospital lengths of stay among their patients. This represents a $47,000 cost savings per patient and improved, timely care transitions to hospice, according to PBACO’s COO David Klebonis.
“The industry considers both long and short stays as failed prognoses. That’s what gravitated us toward this program,” Klebonis told Healthcare Finance. “Ultimately, the goal of machine learning is to bring together components and be able to create a list for your interventions. Every time we fail on determining a prognosis on the back end, the patient is seven times more expensive than the patient you made the right decision on.”
Minnesota-based St. Croix Hospice — a portfolio company of the private equity firm H.I.G. Capital — in 2020 began using a predictive model in the Medalogix-Muse platform to analyze clinical data to predict patient mortality seven to 12 days in advance.
Medalogix merged with the former Muse Healthcare in 2021, with financial backing from the private equity firm and the home health and hospice providers LHC Group (NASDAQ: LHCG), Amedisys (NASDAQ: AMED), and Encompass Health (NYSE: EHC) as minority investors.
This initiative helped St. Croix to achieve 100% performance on quality measures for patient visits during the last days of life, according to Chief Medical Officer Dr. Andrew Mayo.
“I really view it as a sixth vital sign,” Mayo previously told Hospice News. “It provides our clinical team with additional information that helps them make decisions about care … It can trigger increased involvement at a time where patients, their families and caregivers may need increased hospice involvement and guidance.”
Companies featured in this article:
Amedisys, Encompass Health, H.I.G. Capital, Healthcare Finance, LHC Group, Medalogix Muse, Muse Healthcare, Palm Beach Accountable Care Organization, Penn Center for Cancer Care Innovation, Perelman School of Medicine, St. Croix Hospice, University of Pennsylvania