AI quantifies risk of short-term death and facilitates treatment decisions for advanced cancer
An augmented intelligence (AI) tool can identify patients with advanced cancer who have a high or medium risk of short-term mortality, according to a study published in JCO oncology practice.
The tool resulted in a significant increase in palliative care and palliative care referrals in a large hematology-oncology practice, according to the researchers.
The tool, Jvion CORE, uses a “continuous learning n-dimensional spatial environment to determine the most likely trajectory for an individual,” the researchers explained. “This trajectory is used to determine an individual’s risk of mortality by understanding the probability that the trajectory will cross high risk areas in clean space.”
In a retrospective study, researchers assessed the impact of deploying the tool on palliative care and palliative care referrals at a community hematology-oncology practice in the Pacific Northwest. The practice had 6 sites and 21 providers handling an average of 4,329 unique patients per month.
Between June 2018 and October 2019, the tool was used to screen 28,246 patients. He identified 886 patients as having a medium or high risk of 30-day mortality.
Among the patients at risk, the most common cancer was lung cancer (24.9%), followed by breast cancer (22.1%) and small or colorectal bowel cancer (11.9%).
The average monthly referral rates in palliative care and palliative care were calculated 5 months before the deployment of the tool, which served as a reference, and 17 months after the deployment of the tool.
The average hospice palliative care visit rate was 17.3 per 1,000 patients per month (PPM) before deployment, and increased to 29.1 per 1,000 PPM after deployment. The average palliative care referral rate fell from 0.2 to 1.6 per 1,000 PPM between pre-deployment and post-deployment.
When researchers eliminated the first 6 months after deploying the tool to account for user learning curve, the average rate of palliative care visits rose to 33.0 per 1000 PPM after deployment, and the average rate of palliative care referrals rose to 2.4 per 1000 PPM.
The researchers found the 30-day mortality algorithm to be accurate, with area under the receiver operator characteristic curve of 0.93 at 30 days and 0.92 at 90 days. Among the patients identified as medium or high risk, 10.3% had died at 30 days and 16.4% had died at 90 days.
“The deployment of an AI tool in a hematology-oncology practice has proven feasible to identify patients at high or medium risk of short-term mortality,” the researchers concluded. “The information generated by the tool has led to changes in clinical practice, resulting in significant increases in the number of CPs [palliative care] and palliative care referrals.
Disclosures: Some study authors are employed by biotechnology, pharmaceutical and / or device companies. Please see the original reference for author affiliations.
Gajra A, Zettler ME, Miller KA. Impact of augmented intelligence on the use of palliative care services in a real oncology context. JCO Oncol Practice. Published online September 10, 2021. doi: 10.1200 / OP.21.00179