More than a million and a half Americans are diagnosed with cancer every year. And while doctors and scientists have made significant strides in treatment, 600,000 people in the U.S. still die from the disease annually.
For many patients, their families and oncologists, cancer can seem (and is described) like a “fight” to win. However, research shows that this fight against cancer can also result in patients close to end of life being referred to palliative care far too late. Further, patients with advanced cancer who are not receiving palliative care measures tend to have more visits to the ER and hospital admission for cancer and therapy related complications. This can lead to additional risks for such patients as noted during the ongoing Covid-19 pandemic.
Delays in incorporating palliative care for patients with advanced cancer and late referral to hospice services not only diminished the quality of life, but also adds to the cost of care to the health care system without any improvement in overall survival. Acute care utilization close to end of life can limit the ability of any healthcare system to provide emergency room, inpatient or ICU care to those with reversible illnesses as also noted during the pandemic. Thus, it is critical to identify and those patients with advanced cancer who are at greatest risk for deterioration in the short-term so as to institute appropriate palliative care referrals.
Understanding Mortality Risks
Part of the challenge with palliative care is how doctors assess risk of mortality in patients with cancer. Research shows that, in general, oncologists patients with cancer, and regularly overestimate actual survival, which hurts chances of getting palliative care sooner.
While some causes of inaccurate mortality estimations come from following clinical trial data (which typically includes healthier patients who tend to have better survivals than patients in the real world), other reasons can include a mission to not give up, or simply shying away from the prospect of carrying out a discussion regarding palliative care or a transition away from cancer fighting therapy. Another key area rarely examined could be non-clinical factors such as social determinants of health which are typically not incorporated into the clinical decision-making.
So, if oncologists cannot accurately assess the mortality of patients with cancer soon enough for better end-of-life care, the question becomes, how do we better solve for this?
An AI Solution
Understanding the importance of end-of-life care for patients and their families, JVION, a developer of technology in the clinical artificial intelligence arena, and Cardinal Health, developed an AI/machine learning tool that examines clinical and nonclinical data for cancer patients.
The tool, which assigns a risk of low, medium or high for mortality within 30 days, examines clinical information from electronic health records (EHRs) including cancer type, tumor staging, therapies, as well associo economic information including income, household size, employment and behavioral data. The socioeconomic information is particularly important in this tool because many oncologists self-report that they do not have the time to assess social determinants of health with patients due to time constraints.
Built as an objective decision support tool, it saves oncologists time by parsing through data, and derives data on SDOH from sources that the oncologists may not have access to, thus allowing them to get a full understanding of patient risks and decide whether to intervene or modify care for patients at medium to high-risk of mortality. Once a patient has been identified as a high risk for mortality, the clinical team can look for reversible or treatable causes such as subclinical infection, lack of social support for transportation or homecare, but if the deterioration is truly due to the progressive malignancy then a palliative/supportive care workstream can be initiated. This latter workstream can be customized at the specific practice based on access to palliative and supportive care resources.
Real-World Case Study
In order to test the real-world application of the tool, JVION and Cardinal Health partnered with a large community oncology practice with 21 providers managing an average of more than 4,000 unique patients per month.
Over a 17-month period following the integration of the Jvion AI tool, (from June 2018 to October 2019), palliative care consults increased by 68% while the average monthly rate of hospice referrals increased by 8-fold . More so, the tool identified 886 at-risk patients with 50% of those identified high-risk patients having died within the first 180 days of initial identification. As an EHR agnostic tool, it can be utilized with any HER allowing an easier way for the insights generated to assist practices identify the highest risk patients. Further, the insights are dynamic and hence change from week to week depending on the evolution of the patient’s clinical and SDOH. In this era of value-based care, the early identification and timely initiation of palliative care can limit acute care ER visits and hospital admissions helping practices enhance their delivery of quality care as well as improving patient journey and quality of life.
Other Use Cases + Conclusion
Beyond mortality, AI and machine learning tools like this can also assess oncology patients’ mental health and pain control as well as develop predictive models for a variety of issues specific to patients with cancer such as risk of neutropenia and infection following chemotherapy, allowing oncologists to assess risks or change the program of treatment.
Overall, these findings showcase the valuable implications for the use AI and machine learning as oncology tools to guide treatment discussions, prevent acute HCU, and to plan for end-of-life care in patients with cancer.