Missing lactate values in critical care registries: a multiple imputation framework for cardiogenic shock research

Lactate is a critical prognostic biomarker in observational studies of cardiac intensive care. In research settings, lactate is an integral component of prognostic models such as the IABP-SHOCK II risk score. However, lactate measurements are frequently missing due to heterogeneous clinical and logistical factors, potentially introducing systematic bias and compromising the validity of epidemiological inferences. In practice, missing lactate values are often replaced by indicator variables or excluded entirely, potentially hampering the analytical validity of research results. We illustrate a methodological framework to address missing lactate values in clinical research settings.

We used data from 17,993 admissions recorded between 2017 and 2021 across 35 cardiac intensive care units participating in the Critical Care Cardiology Trials Network registry to characterize the mechanism of lactate missingness and develop a structured model for multiple imputation (MI). To assess the practical implications of this imputation procedure, we evaluated risk reclassification when using imputed lactate for computing the IABP-SHOCK II risk score.

Missingness was associated with patient severity, confirming that the data were not missing completely at random. Identified predictors of lactate missingness were included in the MI model. The inclusion of imputed lactate in the IABP-SHOCK II risk score resulted in significant reclassification of risk, demonstrating the magnitude of bias inherent in naïve imputation (event-specific net reclassification ranging from -0.13% to 2.70% across shock populations).

Missing lactate measurements are a source of selection and measurement bias in critical care registries. A structured imputation framework mitigates bias and improves the reliability of real-world evidence in cardiovascular epidemiology.

Lactate is a blood marker that helps clinicians assess how sick a patient is and is commonly used in risk prediction tools for patients with cardiogenic shock (CS). However, lactate measurements are often missing in clinical registries because testing practices vary across hospitals and patient groups. When lactate values are missing, researchers often either exclude those patients from analyses or assume that the missing value is low. Both approaches can introduce bias and may lead to inaccurate estimates of patient risk. In this study, we developed and evaluated a structured approach for handling missing lactate values using MI, a statistical method that estimates likely values based on other available clinical information. Using data from the Critical Care Cardiology Trials Network registry, we assessed how accounting for missing lactate affected risk classification in patients with CS. We found that replacing missing lactate values using a principled imputation approach led to meaningful changes in risk classification and identified additional patients at higher risk of death who would not have been recognized using standard approaches. The results were robust across multiple sensitivity analyses. These findings suggest that appropriate handling of missing biomarker data can improve the validity of registry-based research and may lead to more accurate assessment of patient risk when complete data are unavailable.