Written by Peter Liu
Spoon Feed
The end-of-life Index (EOLI), an electronic medical record (EMR)-based predictor of 6-month mortality in patients with cancer, demonstrated only moderate accuracy in predicting 6-month mortality.
EMR tools only moderately accurate at predicting 6 month mortality
For clinicians managing high-acuity disease, the ability to anticipate a poor prognosis allows for timely advanced care planning, goals-of-care discussions, and specialty involvement such as palliative care involvement. Today’s featured study evaluated the performance of the End-of-life Index (EOLI) on predicting 6 month mortality for 871 inpatients with cancer. The Index had a moderate accuracy in predicting 6-month mortality (c-statistic = 0.71, 95%CI 0.61-0.72), and patients with EOLI score >40 had higher mortality (45.9% vs 16.3%, p<0.001), higher ICU utilization (12.4% vs 6.5%, p=0.02), and were more likely to be discharged to a location other than home (13.5% vs 5.3%, p<0.001). This demonstrates that readily-available EMR-based models are available that give helpful predictive information to identify patients with poor prognosis.
There are several considerations to using such models. While many models exist, performance varies, including other machine-learning models trained with the same dataset which predicted 6 month mortality more accurately than the EOLI. When applied outside of their testing environments, clinicians should expect significant variance. The EOLI was developed and trained on just three healthcare systems. If applied to a fourth, dissimilar healthcare system, with different demographics, the validity of its predictive capabilities is likely to be lower. Lastly, the interpretation of this EOLI should be nuanced. While mortality was higher in the EOLI and patients were less likely to be discharged home, the EOLI >40 group were more likely to be discharged to a skilled nursing facility or inpatient rehab.
How does this change my practice?
In my current inpatient practice, EMR tools such as the EOLI are not yet readily viewable by clinicians. However, if they become available, I will have to interpret the prognostic information cautiously. Just because a predictive tool may recommend advanced care planning, goals-of-care discussions, and palliative care involvement doesn’t mean a patient or their family will be ready or willing.
Source
Performance of electronic medical record tool in predicting 6-month mortality in hospitalized patients with cancer. J Hosp Med. 2025 Feb 17. doi: 10.1002/jhm.70012. Epub ahead of print. PMID: 39961789
