Previously submitted to: JMIR Medical Informatics (no longer under consideration since Jun 26, 2026)
Date Submitted: Jun 12, 2026
Warning: This is an author submission that is not peer-reviewed or edited. Preprints - unless they show as "accepted" - should not be relied on to guide clinical practice or health-related behavior and should not be reported in news media as established information.
Vancomycin Trough Trajectory Phenotypes for Prognostic Risk Stratification in Patients With Sepsis: Observational Cohort Study With Logistic Regression and Machine Learning
ABSTRACT
Background:
Vancomycin remains an important agent for serious suspected or confirmed methicillin-resistant Staphylococcus aureus infection in critically ill patients. However, its use in sepsis is complicated by marked pharmacokinetic variability, nephrotoxicity concerns, and substantial clinical heterogeneity. Conventional therapeutic drug monitoring often relies on isolated trough values, whereas longitudinal vancomycin concentration patterns may provide additional information about patient status, renal function, treatment response, and prognosis.
Objective:
This study aimed to determine whether longitudinal vancomycin trough trajectories could identify clinically meaningful exposure phenotypes and improve prognostic risk stratification for mortality, severe acute kidney injury, and invasive ventilation among patients with sepsis.
Methods:
We conducted an observational cohort study using a trajectory-labeled sepsis cohort with serial vancomycin therapeutic drug-monitoring data, baseline clinical variables, microbiologic information, and downstream outcome labels. Candidate latent trajectory solutions were screened using information criteria and classification quality indices. After applying prespecified eligibility criteria, the unstable intermediate latent class was excluded, and downstream analyses were restricted to Trajectory 1 and Trajectory 3. Associations with in-hospital death, 28-day mortality, 90-day mortality, severe acute kidney injury defined as KDIGO stage 2 or greater, and invasive ventilation were assessed using three nested logistic regression models. Predictive performance was evaluated using the area under the receiver operating characteristic curve. Ten machine-learning algorithms were benchmarked for in-hospital death using a 7:3 train-test split.
Results:
The final comparative cohort included 4031 admissions, with 2647 in Trajectory 1 and 1384 in Trajectory 3. Compared with Trajectory 1, Trajectory 3 showed lower crude rates of in-hospital death, 28-day mortality, 90-day mortality, and severe acute kidney injury. In the fully adjusted model, Trajectory 3 remained associated with lower odds of in-hospital death, 28-day mortality, 90-day mortality, severe acute kidney injury, and invasive ventilation. Logistic discrimination improved with increasing covariate depth, with fully adjusted area under the receiver operating characteristic curve values of 0.727 for in-hospital death, 0.727 for 28-day mortality, 0.727 for 90-day mortality, 0.708 for severe acute kidney injury, and 0.689 for invasive ventilation. In machine-learning benchmarking, CatBoost achieved the highest area under the receiver operating characteristic curve, followed closely by support vector machine with radial basis function kernel. Across models, leading predictors included shock, race, age, kidney failure, hematocrit, platelet count, hemoglobin, respiratory rate, trajectory membership, and temperature.
Conclusions:
Trajectory-based vancomycin exposure phenotyping identified clinically distinct sepsis subgroups and contributed incremental prognostic information for mortality, severe acute kidney injury, and invasive ventilation. The findings should not be interpreted as evidence that a higher or rising trough profile is intrinsically beneficial. Instead, trajectory membership appears to function as one component of a broader multivariable risk signature that may support individualized monitoring and bedside risk assessment in sepsis. Clinical Trial: Not applicable.
Citation
Request queued. Please wait while the file is being generated. It may take some time.
Copyright
© The authors. All rights reserved. This is a privileged document currently under peer-review/community review (or an accepted/rejected manuscript). Authors have provided JMIR Publications with an exclusive license to publish this preprint on it's website for review and ahead-of-print citation purposes only. While the final peer-reviewed paper may be licensed under a cc-by license on publication, at this stage authors and publisher expressively prohibit redistribution of this draft paper other than for review purposes.