Why Pediatric CVICU Patients Require Specialized Dosing Models
Selecting a suitably predictive pharmacokinetic (PK) model for model-informed precision dosing (MIPD) of vancomycin is uniquely challenging in special subpopulations like pediatric cardiovascular intensive care unit (CVICU) patients. In addition to challenges associated with age-dependent maturation seen across all pediatric populations, CVICU patients often display significantly altered PK compared to the general pediatric population. Volume of distribution (Vd) and drug clearance (CL) may be altered due to post-surgical fluid shifts, the use of co-medications or co-therapies, and other intensive interventions.
To address these complexities and improve vancomycin dose-optimization, researchers have developed several specialized PK models that aim to capture the altered PK of pediatric CVICU patients better than general population models.
Our Approach: Evaluating PK Models in Real-World Pediatric CVICU Data
At InsightRX, we continuously review, validate, and improve the models used in InsightRX Nova to reflect the diversity and complexity of your patient populations. In this project, our Data Science team:
- Analyzed 219 pediatric CVICU patients receiving vancomycin across 11 clinical sites
- Evaluated and compared the predictive performance of three specialized CVICU models and two frequently well-performing general population models
- Used root mean square error (RMSE), mean percentage error (MPE), and accuracy as primary metrics
- Applied our continuous learning framework to refit the best-performing models’ parameter estimates on real-world data—to see if predictive performance could be further improved
Which Models Performed Best?
Among the published population PK models, we found that:
- All three CVICU-supporting models (Kamp 2024, Moffett 2019, Shimamoto 2023) outperformed general population models (Colin 2019, Le 2014)
- The Moffett 2019 and Shimamoto 2023 models performed best overall
After applying our continuous learning framework:
- We were able to further improve both models
- The refit Shimamoto 2023 model delivered the best overall performance across a priori predictions (where predictions are made using population PK parameters) and a posteriori predictions (where PK parameter estimates are tailored to an individual patient using therapeutic drug monitoring samples and Bayesian estimation)
This well-performing, refit model is now available in InsightRX Nova for all customers using the pediatric vancomycin module.
Predictive Performance of Published and Refit Vancomycin PK Models in Pediatric CVICU Patients
Above: Predictive performance of published and refit pharmacokinetic models for pediatric cardiovascular intensive care unit (CVICU) patients receiving vancomycin. Assessed by root mean square error (RMSE), mean percentage error (MPE), and prediction accuracy. Error bars represent the point and 95% confidence interval estimate for each model. For RMSE and accuracy, the dotted lines represent the upper and lower limits of the clinically acceptable range, respectively. For MPE, the dotted lines represent the lower and upper limits of the clinically acceptable range, and the solid line represents a target value of zero bias.
Why This Matters: Equity in Precision Dosing
This work is part of our broader effort to improve healthcare equity in special patient populations receiving vancomycin (and other drugs), including:
- Neonatal and pediatric populations
- Critically ill adult patients on continuous renal replacement therapy (CRRT)
- Pediatric oncology patients (coming in Part II of this series)
By refining and validating models for subpopulations like pediatric CVICU patients, we aim to bring safer, more effective dosing to patients who need it most.
Take a look at Part II, where we explore vancomycin model performance in pediatric oncology patients, another population with unique PK challenges.
Want to Dive Deeper? Learn more about our model selection, evaluation, and improvement process by reading the full analysis below.
