By Stacey Kusterbeck
Patients who engage in advance care planning (ACP) conversations are more likely to receive end-of-life care consistent with their wishes. “An abundance of evidence has shown this. However, many patients are not engaged in these conversations early enough, if at all,” says Jessica Ma, MD, MHSc, an assistant professor at Duke University School of Medicine and physician in Geriatrics and Extended Care at the Durham VA Health Care System.
A major challenge, especially in oncology, is accurately predicting when a patient is near the end of life. This is an area where machine learning models can help, by integrating many clinical data points to identify patients at high risk for death in the near future. “The question is whether we can use a machine learning model to identify these patients, spur advance care planning conversations, and consequently improve end-of-life care outcomes,” says Ma.
Previously, Ma and colleagues developed a machine learning model to identify patients admitted from the emergency department with high near-term risk of death.1 The team implemented the same model on an inpatient unit for patients with solid cancers to explore whether it would improve advance care planning conversation rates and end-of-life care outcomes.2 The mortality prediction model substantially improved documented advance care planning conversation rates. “It is encouraging to see that we can use a machine learning model to drive changes in clinician behavior,” says Mihir Patel, BS, lead author of the study and a medical student at Duke University School of Medicine.
For the preintervention cohort of 88 patients, advance care planning conversations were documented for 2.3% of hospitalizations, compared to 80% for the post-intervention cohort of 77 patients. If the attending physician was a palliative care specialist, advanced planning conversations increased from 4.1% to 84.6% vs. 0% to 76.3% for oncologists.
However, there was no effect on end-of-life care, such as code status changes or hospice referrals. Thus, the question of how clinicians can translate advance care planning conversations to improved end-of-life care outcomes remains unanswered. “While there were a lot more ACP conversations, additional work is needed to understand whether there may be suboptimal quality of these conversations, or their documentation, or other factors responsible for this gap,” says Ma.
Appropriate implementation of a machine learning model in any clinical setting requires multidisciplinary collaboration between developers, clinicians, and ethicists. “This collaboration should continue throughout the life cycle of a model,” says Ma. Ethicists can help guide ethical development, integration, and monitoring of machine learning models, including those used for ACP, in these ways, says Ma:
- Explore whether models have potential for bias.
- Brainstorm with clinicians on how to best integrate models into clinical workflows while ensuring data privacy.
- Query clinicians on what they know about existing models and work with developers to ensure transparency of model information to clinicians.
For clinicians, it is important to know which situations the model is intended for, and situations where it should not be applied. “For example, our model was developed and validated in the inpatient population. It would not be appropriate to predict mortality risk in outpatient populations,” explains Ma.
References
- Walter J, Ma J, Platt A, et al. Quality improvement study using a machine learning mortality risk prediction model notification system on advance care planning in high-risk patients. Brown Hospital Medicine. 2024;3(3). https://doi.org/10.56305/001c.120907
- Patel MN, Mara A, Acker Y, et al. Machine learning for targeted advance care planning in cancer patients: A quality improvement study. J Pain Symptom Manage. 2024;68(6):539-547.e3.
Patients who engage in advance care planning conversations are more likely to receive end-of-life care consistent with their wishes. A major challenge is accurately predicting when a patient is near the end of life. This is an area where machine learning models can help.
You have reached your article limit for the month. Subscribe now to access this article plus other member-only content.
- Award-winning Medical Content
- Latest Advances & Development in Medicine
- Unbiased Content