Oncology practice uses AI to significantly improve end-of-life care
Northwest Medical Specialties is a physician-owned, community oncology practice with five sites in the Tacoma, Washington, area.
NWMS is dedicated to providing compassionate, high-quality cancer care, and takes part in several cancer care quality initiatives and value-based care programs, such as the Center for Medicare and Medicaid Innovation’s Oncology Care Model, American Society of Clinical Oncology’s Quality Oncology Practice Initiative, and the National Committee for Quality Assurance.
“Like a lot of oncology practices, one area we struggled with was identifying patients nearing the end of their lives,” said Dr. Sibel Blau, medical director, hematology and oncology, at NWMS. “As oncologists, we tend to overestimate the survival of our cancer patients, which results in delays in initiating palliative care conversations with patients, which then leads to more intensive and costly care at the end of their life.”
Too often, oncologists miss the chance to provide these patients with the comfort and peace they deserve in their final days, simply because often oncologists cannot predict who may die in the next few weeks or months, she added. One study found 68% of cancer patients are never referred for palliative care.
“We have 24 providers managing approximately 4,300 unique patients per month, so it’s not easy for us to catch every instance where a patient is close to the end; we simply don’t have the staff to review the charts to manually look for high-risk patients,” Blau explained.
"We’ve been able to better target pain management agents, resulting in a 33% reduction in moderate or severe pain with oncology patients."
Amy Ellis, Northwest Medical Specialties
“If we can catch these patients earlier, we can refer them for palliative hospice care, and help them cope with the mental health struggles that come with reaching the end of one’s life, improving their end-of-life experience.”
Having more of these important end-of-life conversations with patients also was a key part of reaching value-based care goals, she noted.
“For example, hospice referrals are a key quality metric under the OCM, which directly impacts our reimbursement,” she said. “We also wanted to do a better job addressing depression, which is really common with cancer patients. Left untreated, depression can have a negative impact on patient outcomes and their end-of-life experience, so one of our value-based care goals was to increase our depression screenings.”
Another strategic initiative NWMS has for value-based care is treating the whole patient with a multidisciplinary team that includes clinicians, social workers and care coordinators. A patient may have cancer, but they may also have unmet psychosocial and mental health needs or social determinants of health challenges with finances, transportation, nutrition or housing that impact their overall health.
“Managing these external factors takes a lot of care coordination with other community providers, whose data may not be included in our EHR,” she said. “To get a better understanding of patient risk, we needed a better way to bring these external factors into the view of our care teams.”
Technology vendor Jvion’s AI CORE was made available to NWMS through a pilot study with Cardinal Health Specialty Solutions.
“The AI solution would integrate with our EHR system and provide lists of patients at risk of clinical deterioration across seven vectors, including 30-day mortality and avoidable hospital admissions,” said Amy Ellis, director, quality and value-based care, at Northwest Medical Specialties.
“In addition to showing the patients at risk, the AI tool also would detail the clinical and nonclinical factors – including socioeconomic and behavioral factors – driving each patient’s risk, and recommend personalized, clinically validated actions to mitigate each patient’s unique set of risk factors.”
"Care teams need to understand that the role of AI is to augment, not replace, the intelligence of clinicians by providing information that would otherwise be out of view."
Dr. Sibel Blau, Northwest Medical Specialties
The initial phase of the project would focus on putting into action the 30-day mortality insights to refer the right patients for palliative care or hospice care. NWMS also would use the six-month depression-risk insights to screen high-risk patients for depression, thus enabling the provider organization to better treat the whole patient.
Other goals included improving pain management and reducing rates of avoidable admissions, readmissions and ED visits through the proactive interventions recommended by the CORE.
MEETING THE CHALLENGE
The CORE is integrated with the organization’s EHR, where it merges the clinical data in the EHR with external data on social determinants of health from public and private sources, such as data from the U.S. Census, Department of Housing and Urban Development, the EPA and USDA.
All told, the CORE analyzes around 4,500 clinical, social, environmental and behavioral data points per patient for a comprehensive understanding of patient risk, Blau explained.
“In this way, the CORE enabled us to account for the external factors that are often invisible to our care teams, improving our care coordination and enabling us to bring in external resources to treat patients holistically,” she said. “For example, by revealing that a patient may not have access to transportation to their appointment, we can adapt the care plan to include transportation assistance, after-hours appointments or telehealth.”
NWMS care coordinators get a list of high-risk patients along with the recommended actions to reduce their risk. They then pass this information on to the clinicians on the care team so they can put into action the CORE’s insights.
The care coordinators work with the patients flagged by the CORE to schedule appropriate services, such as evaluations and treatment for pain, distress or depression; completing advance directives; or referrals to hospice care.
“The results we’ve seen with the Jvion CORE have exceeded our expectations,” Ellis stated. “By using the AI to predict which patients were at risk of mortality and acting on that information, we’ve been able to improve the end-of-life experiences for a lot of our patients. We really started to see results after about six months, when our care teams had adjusted to the learning curve.”
In months seven to 17, NWMS’s palliative care consults nearly doubled, from 17.3 to 33.0 per 1,000 patients per month. In the same period, hospice referrals increased by twelvefold, from 0.2 to 2.4 per 1,000 PPM. The timely integration of palliative care for patients with advanced cancer helped NWMS reach its quality benchmarks, Ellis noted.
“The CORE also enabled us to increase depression screenings by 171%, depression diagnoses by 22% and case management evaluations by 84%,” she said. “We’ve also been able to better target pain-management agents, resulting in a 33% reduction in moderate or severe pain with oncology patients. Finally, we’ve been able to better prevent patient deterioration, as shown by an up to 30% reduction in loss of function/ADLs.”
Overall, she added, the AI technology has enabled NWMS to achieve the Quadruple Aim of healthcare and deliver high-quality cancer care that is patient-centered, outcomes-focused and cost-effective, while reducing the cognitive burden on care teams.
But more importantly, she said, it has allowed NWMS to have more meaningful end-of-life conversations with patients, giving terminal patients agency over their final days.
ADVICE FOR OTHERS
“With clinical AI, it’s really important to get buy-in from the physicians and care team members using the tool,” Blau advised.
“We were all a little skeptical about it at first, but once we started to manually validate the data, it became clear how powerful the tool could be for value-based care,” she explained. “Having champions of AI across physician, nursing and care coordination teams who can advocate for the technology and its proper use goes a long way toward building trust in the system and ensuring a successful AI implementation.”
Another important thing to keep in mind is that AI is not intended to replace the judgement of clinicians, and having that expectation can actually foster resistance and hold back success, she added.
“Care teams need to understand that the role of AI is to augment, not replace, the intelligence of clinicians by providing information that would otherwise be out of view,” she concluded. “Ultimately, physicians don’t have as much time to spend with patients as we’d like to, and that makes it hard to get a complete picture of the patient. AI is really helpful for filling in the gaps, but it takes a cultural shift in the practice to accept that.”