Machine learning helps identify older adults with cancer at high risk for adverse outcomes
Click Here to Manage Email Alerts
Key takeaways:
- The k-means unsupervised machine learning algorithm clustered patients based on symptom severity.
- Moderate- and high-severity clusters appeared associated with a higher risk for death.
Unsupervised machine learning helped identify older adults with advanced cancer who appeared more likely to experience unplanned hospitalization and death, according to study results published in JAMA Network Open.
Additional research is needed to validate the machine learning algorithm and assess the clinical utility of the algorithm in oncology practice, researchers concluded.
Rationale and methodology
“Older adults with advanced cancer often suffer from multiple symptoms due to cancer and other aging-related comorbidities,” Huiwen Xu, PhD, MHA, assistant professor in the department of population health and health disparities at University of Texas Medical Branch, told Healio. “The NCI Patient-Reported Outcomes version of the Common Terminology Criteria for Adverse Events [PRO-CTCAE] captures a wide range of patient-reported symptoms, but lacks a way to extract the summary information from the individual items. We believe unsupervised machine learning might be a solution to this issue.”
As Healio previously reported, the GAP70+ trial included 718 patients with advanced cancers who received care across 40 U.S. community oncology practice clusters.
Xu and colleagues used an unsupervised machine learning algorithm to cluster trial participants based on severity items of 24 PRO-CTCAE symptoms that ranged from zero to four and corresponded to none, mild, moderate, severe and very severe. They then calculated the total severity score as the sum of 24 items and examined whether the clusters were associated with unplanned hospitalization, death and toxic effects.
The analysis included 706 patients (mean age, 77.2 years; 56.8% men; 87.8% white) with baseline PRO-CTCAE data. The most common cancer types included gastrointestinal (34.7%), lung (24.8%) and breast (15.4%).
Unplanned hospitalization over 3 months served as the primary outcome, with all-cause mortality over 1 year and any clinician-rated grade 3 to grade 5 adverse events over 3 months as secondary outcomes.
Findings
The machine learning algorithm, k-means with Euclidean distance, classified 310 patients (43.9%) as low-severity, 295 (41.8%) as medium-severity and 101 (14.3%) as high-severity, which corresponded to within-cluster mean severity scores of 6.3 (low), 16.6 (moderate) and 29.8 (high; P < .001).
After controlling for sociodemographic variables, clinical factors, study group and practice site, researchers found patients in the moderate-severity cluster appeared more likely to experience hospitalizations (risk ratio = 1.36; 95% CI, 1.01-1.84) than patients in the low-severity cluster.
Moreover, researchers found associations of the moderate- (HR = 1.31; 95% CI, 1.01-1.69) and high-severity clusters (HR = 2; 95% CI, 1.43-2.78) with a higher risk for death but similar risk for toxic effects.
Implications
“Machine learning may be used to guide the development of risk stratification tools with the potential to assist clinicians in identifying older adults with a high risk for adverse outcomes,” Xu told Healio. “Our machine learning algorithm now needs to be externally validated and we need to assess the clinical useability of integrating the algorithm into practices. In the long run, we hope to examine whether providing a machine learning-based tool can modify clinical decisions and eventually improve patient outcomes for this growing vulnerable population with cancer.”
Reducing both undertreatment and overtreatment is critical among older adults with advanced cancer who want to receive treatment, according to an accompanying editorial by Carolyn J. Presley, MD, MHS, researcher in the department of internal medicine at The Ohio State University Comprehensive Cancer Center.
“This requires cancer physicians to understand patient symptoms as [patient-reported outcomes] and monitor them over time so that they can intervene in a meaningful way,” Presley wrote. “Older adults have been telling us the secret of how to treat them for their advanced cancer the entire time. We are continually learning how to listen and use this information in our treatment decision-making. Listening to patients and understanding [patient-reported outcomes] to bolster supportive care early on in the disease course is the path forward to improve cancer care for a rapidly growing population of older adults worldwide.”
References:
- Presley CJ, et al. JAMA Netw Open. 2023;doi:10.1001/jamanetworkopen.2023.4208.
- Xu H, et al. JAMA Netw Open. 2023;doi:10.1001/jamanetworkopen.2023.4198.
For more information:
Huiwen Xu, PhD, MHA, can be reached at huxu@utmb.edu.
Collapse
Read more about
Click Here to Manage Email Alerts