Joint modeling approach on evaluating time-dependent cutoff values for c-reactive protein in mixed intensive care unit population

Abstract

Background: Serial C-reactive protein (CRP) biomarker values are frequently recorded from patients in adult in-tensive care units (ICU). The aim of this study was to assess the time-dependent diagnostic accuracy of repeated CRP measurements in predicting ICU mortality and determine the time-dependent cutoff values for this biomark-er in mixed ICU population. Methods: Joint modeling was performed to model repeated CRP measurements and survival data. Time-depen-dent AUC (td-AUC) values were used to assess the diagnostic performances. Maximization of the product of sensi-tivity and specificity rule was applied to determine the time-dependent cutoff values. Results: Time-dependent diagnostic performance of serial CRP values were found as moderate in overall, observ-ed to be higher in males than females, ranging from 0.603 to 0.624 in females and 0.639 to 0.690 in males. On the other hand, time-dependent cutoff values either remained constant or decreased through the 3rd day after the last measurement for both gender groups. Conclusions: Newly proposed time-dependent cutoff values for CRP biomarker are suggested to be used in clinics to discriminate subjects who are at risk and who are not during the first three days after the last measurement. Furthermore, taking serial CRP values in predicting the risk of death at ICU is highly recommended, to be able to assess the change in longitudinal profiles of subjects throughout the follow-up period.