JMIR Publications

ABSTRACT

Background:

Sepsis is diagnosed in millions of people every year, resulting in high mortality rate. Although sepsis patients present multimorbid conditions, including cancers, sepsis predictions have mainly focused on patients with severe injuries.

Objective:

Here, we present a machine learning-based approach to identify sepsis risk in cancer patients using electronic health records (EHRs).

Methods:

We utilized anonymized EHRs from the Samsung Medical Center in Korea, including 8,580 cancer patient records in longitudinal manner (2014~2019). To build a prediction model based on physical status that would differ between sepsis and nonsepsis patients, we analyzed 2,462 laboratory (lab) test results and 2,266 medication prescriptions using a graph network and statistical analysis. Based on the results, the model was trained with lab tests and medication relationships.

Results:

Sepsis patients showed differential medication trajectories and physical status. For example, in the network-based analysis, narcotic analgesics were prescribed more often in the sepsis group, along with other drugs. Likewise, 35 types of lab test, including albumin, globulin, and prothrombin time, showed significantly different distributions between sepsis and nonsepsis patients (p value<0.0001). Our model outperformed the model trained using only common EHRs, showing improved accuracy, area under the receiver operating characteristic (AUROC), and F1-scores up to 11.9%, 11.3%, and 13.6%, respectively. (Accuracy: 0.692, AUROC: 0.753, and F1-score: 0.602 for the random forest-based model).

Conclusions:

We elucidated that lab tests and medication relationships can be used as efficient features for predicting sepsis in cancer patients. Consequently, identification of sepsis risk in cancer patients using EHRs and machine learning is feasible.