Serious illness conversations

Why use machine learning models ?

For patients, 80% of Americans would prefer to die at home but only 20% of them do. Many receive unnecessary treatments during the last months of life. By some estimates, of those admitted to hospitals, less than half needing palliative care actually receive it. Patients that do receive hospice care report excellent outcomes on a variety of measures which include: Better pain and symptom management; Increased patient and family satisfaction; Significant reductions in health care cost; Increased time spent at home.

Most patients who enroll in hospice are medically complex. Clinicians often rely on ‘gut’ instinct to estimate life expectancy in these patients because no easy-to-use, at-the-bedside, generalisable, risk assessment tools exist. This predictive uncertainty often delays the formulation of a timely care plan, resulting in missed opportunities when patients are relatively stable and can make informed decisions that will truly alter their outcomes and quality of life. Instead, serious illness conversations and palliative care referrals occur in last-minute settings, after patients have become critically ill, or are admitted to the hospital. Indeed, there has been considerable interest in developing institutional programs to train clinicians in identifying high risk patients and have serious illness conversations earlier.

A lack of accurate prognostic models hinders formulation of timely care plans and results in a large amount of waste. Hospice stay costs average $153/day compared to $6,200/day for in-hospital care. The consequences of misestimating the risk of death has resulted in systemic hospice underutilization in the US. Machine learning models can accurately estimate risk of death in complex multi-condition medical patients. They provide a systematic,objective measure of 6-month mortality risk and can be automated in order to streamline workflows. This can improve access to timely, data-informed serious illness conversations between patients and their providers.