Right Dose, Right Now: Development of AutoKinetics for Real Time Model Informed Precision Antibiotic Dosing Decision Support at the Bedside of Critically Ill Patients

Luca F. Roggeveen, Tingjie Guo, Ronald H. Driessen, Lucas M. Fleuren, Patrick Thoral, Peter H. J. van der Voort, Armand R. J. Girbes, Rob J. Bosman, Paul ElbersFrontiers in Pharmacology 2020 · DOI

TL;DR

This paper introduces AutoKinetics, a novel clinical decision support system (CDSS) designed to provide real-time, model-informed antibiotic dosing recommendations for critically ill patients. By integrating deeply with electronic health records, AutoKinetics aims to overcome the challenges of personalized antibiotic therapy, potentially improving patient outcomes in sepsis.

What Problem Does This Paper Address?

Sepsis, a life-threatening condition, requires prompt and effective antibiotic treatment. However, in critically ill patients, their pharmacokinetics (PK) are highly variable and can change rapidly, making standard dosing regimens often inadequate. This leads to suboptimal drug exposure, increasing the risk of treatment failure or toxicity. While Therapeutic Drug Monitoring (TDM) exists, it's often limited to specific drug classes, requires manual data entry and interpretation, and delays decision-making. The authors identify a critical need for a system that provides immediate, personalized antibiotic dosing advice directly at the bedside, accessible to clinicians without extensive PK expertise.

Methods

The development of AutoKinetics followed a rigorous, iterative process:

  • Goal Directed Task Analysis (GDTA): Used to structure the system into modules and define information needs for optimal decision-making.
  • Workflow Analysis: Facilitated integration into existing Electronic Health Record (EHR) systems (Metavision, Epic) by focusing on a passive integration approach (data retrieval, advice provision, no automatic order changes).
  • Application Development Life Cycle: A four-step process (Development, Testing, Acceptance, Production) with separate environments for robust development and testing with pseudonymized data.
  • Pharmacokinetic Model Integration: A literature review identified and selected PK models for five commonly used antibiotics for sepsis (meropenem, ciprofloxacin, ceftriaxone, cefotaxime, vancomycin). New models were developed where necessary (e.g., for cefotaxime). Models were validated and calibrated using retrospective data.
  • Ordinary Differential Equation (ODE) Solver: Implemented a numerical solver (Runge-Kutta) for PK models, empirically validated against the gold standard NONMEM software.
  • Dose Calculation Algorithms: Developed methods for calculating maintenance doses (MD) and loading doses (LD) based on PK principles, aiming to achieve target exposure levels (e.g., %T > MIC, AUC/MIC).
  • Bayesian Maximum A Posteriori (MAP) Calculator: A module for applying Bayesian estimation to adjust dosing when plasma concentration data is available, crucial for antibiotics like vancomycin.
  • Graphical User Interface (GUI) Design: Focused on a minimalistic, intuitive design to present complex PK information visually and textually, balancing information density with usability for clinicians with limited PK knowledge.

Datasets

The study utilized retrospective data from patients at Amsterdam UMC, location VUmc, and OLVG Oost Hospital, Amsterdam, Netherlands. This data was used for:

  • Model Validation and Calibration: To assess and refine the performance of selected PK models.
  • Safety Analysis: To compare AutoKinetics' simulated dosing advice against current clinical practice (pharmacist-led TDM) using trough samples from 97 vancomycin patients.

No new benchmarks were explicitly introduced, but the study aimed to establish a new standard for real-time, bedside dosing support.

Key Insights and Ideas

  • Deep EHR Integration is Crucial: The success of AutoKinetics hinges on its seamless integration with EHR systems, minimizing physician workload and ensuring immediate availability of advice.
  • Passive Integration as a Safety Feature: By not automatically changing medication orders, AutoKinetics maintains a crucial "air gap," enhancing patient safety and simplifying EHR integration.
  • User-Centric Design for Adoption: The intuitive GUI, providing visual feedback and clear advice, is key to enabling adoption by clinicians who may not be PK experts.
  • Bayesian Estimation for Adaptive Dosing: The inclusion of a Bayesian calculator allows for real-time adjustments based on actual patient data, moving beyond static model predictions.
  • Addressing Regulatory Hurdles: The authors proactively addressed compliance with Medical Device Regulations (MDR) and GDPR, highlighting the importance of these considerations for clinical tools.

Key Results

  • AutoKinetics was successfully developed as a CDSS for real-time, model-informed antibiotic dosing.
  • The system was integrated into two popular EHRs (Metavision, Epic).
  • A SQL-based relational database backend and a data loader were developed for real-time data retrieval.
  • The system supports five extensively used antibiotics for sepsis.
  • Simulations indicated that AutoKinetics dosing regimens could lead to an increased percentage of time patients remain within desirable concentration ranges and a reduction in time spent below therapeutic levels, compared to physician TDM.
  • The ODE solver's performance was empirically validated against NONMEM, showing minimal differences in estimated concentrations.

Limitations and Future Work

  • Limited Scope: Currently, AutoKinetics only provides dosing advice and does not offer decision support for antibiotic selection or treatment duration.
  • PK Model Quality: The accuracy of AutoKinetics is inherently dependent on the quality and applicability of the PK models used.
  • No Closed-Loop Dosing: The system requires physician intervention to place orders, which is a deliberate safety feature but limits the speed of dose adjustments.
  • Future Work: The authors plan to expand AutoKinetics to include decision support for antibiotic choice and duration. They also aim to integrate more sophisticated PK models and explore its use in non-critical care settings and with other EHR systems. A multi-center randomized control trial (Right Dose Right Now) is underway to validate the concept in 420 patients with severe sepsis and septic shock.

Summary updated

Abstract

We successfully developed a CDSS for real time model informed precision antibiotic dosing at the bedside of the critically ill. This holds great promise for improving sepsis outcome. Therefore, we recently started the Right Dose Right Now multi-center randomized control trial to validate this concept in 420 patients with severe sepsis and septic shock.
BibTeX
@inproceedings{roggeveen2020right,
  author = {Luca F. Roggeveen and Tingjie Guo and Ronald H. Driessen and Lucas M. Fleuren and Patrick Thoral and Peter H. J. van der Voort and Armand R. J. Girbes and Rob J. Bosman and Paul Elbers},
  title = {{Right Dose, Right Now: Development of AutoKinetics for Real Time Model Informed Precision Antibiotic Dosing Decision Support at the Bedside of Critically Ill Patients}},
  booktitle = {Frontiers in Pharmacology},
  year = {2020},
  doi = {10.3389/fphar.2020.00646},
}
Figure