How Can AI Be Used in Palliative Care?

Researchers from Brigham and Women's Hospital and Massachusetts General Hospital say AI technology could be used in a hybrid platform that improves how clinicians communicate with patients and their families about serious illnesses and palliative care.

AI technology is starting to show up in care management and coordination platforms as an avenue for interpreting data and communicating with patients, usually through chatbots, but is it appropriate for dealing with patients in palliative care?

In a recent article in NPJ/Digital Medicine, researchers from Brigham and Women’s Hospital and Massachusetts General Hospital say the platform could help care providers with serious illness communication (SIC) by smoothing over what is often a difficult process. But the technology needs to be integrated carefully in a hybrid platform.

Patients with serious illness often experience delayed SIC because clinicians are poor at prognosticating life expectancy for terminally ill patients, usually erring on the side of optimism,” says the January 27 article, authored by Isaac S. Chua and David W. Bates of Brigham and Women’s and Christine S. Ritchie at Mass General. “Moreover, systematic methods to identify patients with palliative care needs are lacking.”

SIC is a both complicated and delicate process. Providers first have to determine whether a patient is in need of palliative care services, then talk to that patient and his or her family about everything from life expectancy to end-of-life care.

The traditional SIC delivery process consists of a series of conversations where gathering, interpreting, and integrating SIC data occur within a clinical encounter followed by manual clinician documentation in the electronic health record (EHR) post-visit,” Chua, Bates and Ritchie write. “This process can be broken down into the following steps: determining patient eligibility for SIC; gathering and interpreting information (e.g., eliciting and clarifying the patient’s illness understanding, hopes, and worries); conducting a therapeutic conversation (e.g., counseling and supporting the patient on coping with life-threatening illness) with the goal of shared decision-making; documenting the conversation; and making SIC documentation accessible to others in the HER. However, each step is a potential bottleneck because the ability to initiate SIC or make forward progress depends heavily on the clinician’s ability, skill, and judgement.”

The researchers point out that many clinicians lack SIC training, and there are no clear standards to document how the process should be conducted or how the EHR facilitates documentation. This might lead to inaccurate or uncertain diagnoses and timelines, awkward and infrequent conversations and more anguish for patients and their caregivers.

“In addition to training more clinicians to be competent in SIC, a novel workflow that addresses these barriers will be necessary to ensure that all seriously ill patients receive timely and effective SIC that informs their care in real time and naturally results in documentation of patients’ goals and preferences that is visible to others,” the article suggests.

That workflow, Chua and his colleagues write, should by a hybrid strategy that combines AI tools in the background with in-person services that should always be the backbone of SIC. The technology would be used to gather and interpret data to ensure and accurate diagnosis and timeline, and to give clinicians the information they need to have those conversations with patients.

“AI can also streamline the SIC documentation process and potentially improve the quality of SIC documentation via natural language processing (NLP)—a form of machine learning designed to understand, interpret, or manipulate human language,” the article continues. “Missing or incomplete documentation in the EHR regarding patient preferences for life-sustaining treatment is common and contributes to medical errors related to end-of-life care.”

“NLP also has the potential to address barriers resulting from poor EHR design that prevent or inhibit the extraction and flow of meaningful advanced care planning information across the care continuum,” Chua and his colleagues continue. “In its current state, identifying SIC documentation in the EHR typically involves a manual chart review that possibly includes a keyword search or utilization of note filters. NLP-enabled software that identifies free text SIC documentation would likely reduce the time and effort clinicians spend looking for this information and prevent inadvertent oversight of patient preferences leading to goal-discordant care. AI-assisted chart reviews have demonstrated higher accuracy and shorter time for extracting relevant patient information compared with standard chart reviews.”

Finally, the AI platform could also be helpful in giving clinicians feedback on their communication skills, a critical component of discussing SIC with distraught patients and family members.

Chua, Ritchie and Bates conclude by noting AI technology can greatly benefit SIC, but some of those benefits aren’t there quite yet. A hybrid approach that integrates data analysis and NLP with in-person services would be an ideal platform, improving accuracy and eliminating gaps in care while giving clinicians more information and guidance to handle challenging and often delicate conversations. But the technology hasn’t been tested enough or isn’t developed far enough to be put to use in clinical situations.

“This proposed paradigm still requires that clinicians undergo some SIC training to capitalize on the assistance provided by AI, as well as additional research to avoid unintended consequences of AI implementation,” they write. “That said, a semi-automated approach to SIC delivery holds tremendous promise and would likely improve current SIC workflow by optimizing clinical manpower and efficiency while increasing the likelihood that these critically important conversations will occur effectively and in a timely fashion.”