Missing Data Reflects Clinical Judgment — AITRICS Study Highlights Clinical Value of Handling Missing Data in AI Models

AITRICS (CEO Kwang joon Kim), a company specializing in artificial intelligence(AI) technology announced on the 18th that its latest study, published in the SCIE-indexed Journal of Clinical Medicine, has demonstrated that missing data in clinical records can serve as valuable predictive information when forecasting patient deterioration.

AITRICS’ solution, AITRICS-VC (VitalCare), predicts patient deterioration using a total of 19 variables derived from electronic medical records (EMR), including six vital signs, eleven blood test results, patient consciousness level, and age. VitalCare’s strength lies in its ability to analyze a wide range of clinical data with high predictive accuracy.

However, real-world EMR data often contain missing values due to differences in hospital testing protocols or clinical judgment. This study investigated how VitalCare handles such missing data to maintain predictive performance. Importantly, the study assumes that missing data may reflect clinicians’ decisions — for instance, omitting a test might indicate the clinician deemed it unnecessary — and therefore holds clinical significance.

The results showed that VitalCare’s own AI-based method for handling missing data achieved an AUROC (Area Under the Receiver Operating Characteristic curve) of 0.896. In comparison, the Mean Imputation and MICE (Multiple Imputation by Chained Equations) methods scored 0.885 and 0.827, respectively — lower than the performance of VitalCare’s approach. Researchers concluded that traditional imputation methods may overlook the “informative presence” within missing data, reducing predictive accuracy.

This study underscores that the act of not ordering a test can itself represent a form of clinical judgment — suggesting that the test was deemed unnecessary. Hence, missing data may encode meaningful clinical context.

Taeyong Sim, Chief Medical Officer of AITRICS, said, “This study shows that treating unperformed tests as within the normal range — based on clinical decision-making — can improve predictive performance. Recognizing missing data not as gaps but as part of clinical judgment is crucial for patient safety and improving medical efficiency.”