Even with technological advancements, provider involvement and oversight remain essential.
When it was introduced in 2016, the Oncology Care Model (OCM) incentivized patient-centric cancer care to improve care quality and avoid unnecessary costs. However, the OCM reimbursement scheme posed a new challenge for community oncology practices because they were tasked with improving quality and reducing costs through the elimination of unnecessary acute care use (ACU) such as emergency department (ED) visits and hospitalizations related to cancer or associated treatment complications.
Previously, data indicated that 53% of cancer-related ED visits were avoidable, resulting in unnecessary expenses, poor patient experiences, and suboptimal care or outcomes.1 Because of this, the OCM looked to address the occurrence of ACU at the cancer practice level.
However, the task of predicting cancer-related ACU is difficult for oncology practices and particularly for community oncology practices. Before this OCM incentive, community practices had already been struggling to manage provider burnout partially caused by increasingly onerous administrativetasks involved in electronic medical record maintenance.
Considering these challenges, the adoption of predictive risk modeling—and prescriptive analytics—based augmented intelligence to help address these issues came to the fore. In particular, augmented intelligence combined with nurse outreach may help decrease avoidable ED visits and hospitalizations at large OCM practices.
To investigate this further, our research team examined how augmented intelli-gence in combination with nurse outreach could impact practices participating in the Enhancing Oncology Model (EOM), which is set to launch in July 2023 for a 5-year testing period. Our findings were published in March 2023 in JCO Oncology Practice.2
To predict risk of avoidable ACU, we implemented a care optimization and recommendation enhancement (CORE) augmented intelligence tool at a large community practice where providers treated an average of 2000 OCM patients per month. The tool used augmented intelligence and machine learning to analyze a variety of data sources, including the practice’s electronic medical record, Centers for Medicare & Medicaid Services claims, and third-party patient behavioral and experiential data. The augmented intelligence tool then analyzed the data and scored patients weekly as having low, medium, or high risk of an ED visit or hospital admission within the next 30 days.
The augmented intelligence tool sent medium- or high-risk patient names to their respective nurse case managers, along with patient-specific interventional recommendations, prompting the case managers to review the patient’s CORE-generated risk level and top 5 recommendations, audit the patient’s chart for additional risk factors, and contact the patient to recommend interventions that case managers selected based on the top 5 recommendations or at their own discretion. Further, the augmented intelligence recommendations could consist of medication or dosage changes, laboratory tests or imaging, referrals (eg, psychologic therapy, palliative care, or hospice), and surveillance or observation.
Integrating human decision-making with augmented intelligence–powered patient-centric risk identification and prescriptive analytics enabled a feasible, effective workflow that resulted in significant clinical practice and outcome improvements during a 53-month period. Per 100 unique OCM patients, the practice realized an 18% decline in monthly ED visits, a 13% decline in quarterly hospital admissions, and a potential annual savings of $2.8 million on avoidable cancer-related ACU.2
Notably, many oncology practices participating in the OCM also looked to reduce unplanned ED visits or hospitalizations by expanding clinic hours. Implementing this tool, however, allowed the practice to identify at-risk patients and intervene while keeping clinic hours the same.2
Looking Ahead to the EOM
Although this quality improvement project and data analysis occurred contemporaneously with the OCM program, the findings showcase a compelling opportunity for community oncology practices to implement technology that will promote meaningful patient engagement in the EOM program without requiring additional staff.
Building upon learnings from the OCM, the EOM will focus on making value-based, patient-centric care more accessible and equitable for patients with cancer undergoing systemic chemotherapy in 6-month episodes of care to improve care quality and experience.3 The EOM encourages multipayer participation (eg, commercial insurers and Medicaid) and a health equity focus on how practices approach providing affordable, accountable care for all patients; both of these will be vital to the success of the EOM program.
We found that implementing tools such as CORE—which integrate human and artificial intelligence—can contribute greatly to meaningful patient engagement and can reduce health disparities by catching symptoms of chronic and infectious diseases earlier, which is paramount for improving outcomes. Given the expanding body of research on racial disparities between Black individuals and White individuals in preventable hospitalizations and ED visits, it is increasingly apparent that although cancer spares no population demographic, it does not affect all patient populations equally.
Exploring Augmented Intelligence Use in Health Care
As augmented intelligence continues to gain ground in health care, it is important to understand the roles of augmented intelligence and health care providers in collaboration with one another. Our study, which looked at how nurses and augmented intelligence could collaborate to reduce ACU in patients with cancer, showed that monthly ED visits dropped from 13.7 to 11.5 (16%) per 100 unique OCM patients, demonstrating a sustained month-over-month improvement. Additionally, quarterly admissions dropped from 19.5 to 17.1 (12%), which is a sustained quarter-over-quarter improvement. As previously noted, overall, the practice realized potential annual savings of $2.8 million on avoidable ACU.2
The augmented intelligence tool was able to predict patient risk and prescribe interventions that were then reviewed by nurse case managers to identify and resolve critical clinical issues and reduce avoidable ACU. This filtration of information by the augmented intel-ligence tool was game-changing for oncology practices looking to identify high-risk patients sooner without adding staff or increasing administrative workload.2
We welcome the advancement of technology and augmented intelligence in health care, but we believe that provider involvement and oversight are essential in oncologic care. Accordingly, our study focused on augmented intelligence–powered predictive modeling and prescriptive analytics for informing patient care—that is, offering direction but ultimately leaving the decisions to the provider.
About the Authors
Yolaine Jeune-Smith, PhD, is the director of scientific writing and strategic research at Cardinal Health.
Alexandrina Balanean, MPh, is the lead publications scientist in strategic research at Cardinal Health.