Pharmacometric covariate modeling using symbolic regression networks

Wahlquist, Ylva; Morin, Martin; Soltesz, Kristian

doi:10.1109/ccta49430.2022.9966112

Cited by 3 publications

(4 citation statements)

References 16 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Pharmacometric covariate modeling is a branch of pharmacology aimed at obtaining dynamical models that capture the response dynamics to a drug while accounting for interindividual variability [13], [16]. Commonly, a Bayesian framework is used, within which the inter-individual variability in the model parameters (V 1 , k 10 , .…”

Section: B Modeling Inter-patient Variabilitymentioning

confidence: 99%

Exploring the Influence of Patient Variability on Propofol Target-Controlled Infusion Performance

Wahlquist,

Gustafsson,

Soltesz

2024

2024 European Control Conference (ECC)

View full text Add to dashboard Cite

Target-controlled infusion (TCI) constitutes a clinically available alternative to manually administering the infusion rate of the anesthetic drug propofol. In TCI, a drug infusion profile is optimized to track a reference trajectory of blood plasma or effect site (brain cortex) drug concentration, or a corresponding clinical effect. TCI is a pure feed-forward openloop strategy, fully reliant on an underlying dynamic patient model. We show how TCI dosing of propofol-to achieve a desired depth of hypnosis-can be posed as a QP problem. We design this QP problem based on a nominal pharmacological propofol model by Eleveld et al. Then, we investigate how interpatient variability, described as mixed effects of a particular distribution within the Eleveld model, affects TCI performance. Based on the Mahalanobis distance, we sample from probability quantiles of the mixed-effect model and evaluate the TCI designed for the nominal patient across these samples. The main outcome is that performance, in terms of achieved hypnotic depth, deteriorates to what is at the limit of clinical acceptance already when considering only 1 % of the most likely patients drawn from the uncertainty model. This is under the-for the TCI system unrealistically favorable-assumption that there is no uncertainty in the relation between effect site concentration, and clinical effect. The conclusion from the arising results is that the benefit of propofol TCI over manual dosing is unclear, even within the model that the TCI system was designed for.

show abstract

Section: B Modeling Inter-patient Variabilitymentioning

confidence: 99%

Exploring the Influence of Patient Variability on Propofol Target-Controlled Infusion Performance

Wahlquist,

Gustafsson,

Soltesz

2024

2024 European Control Conference (ECC)

View full text Add to dashboard Cite

show abstract

“…Our implementation, using the neural network package Flux [19], relies on the differential programming capabilities of the Julia language [20]. A full disclosure of our implementation is found in the GitHub repository [21].…”

Section: Trainingmentioning

confidence: 99%

“…Next, we remove the N = 10 least sensitive parameters in the first parameter pruning iteration, and then one parameter per subsequent pruning iteration until only twelve network parameters are left. More details of the pruning and training can be found in [21].…”

Section: Compute the Salience Of Each (Remaining) Trainable Network P...mentioning

confidence: 99%

See 1 more Smart Citation

Learning pharmacometric covariate model structures with symbolic regression networks

Wahlquist

Sundell

Soltesz

2023

Preprint

Self Cite

View full text Add to dashboard Cite

Efficiently finding covariate model structures that minimize the need for random effects to describe pharmacological data is challenging. The standard approach focuses on identification of relevant covariates, and present methodology lacks tools for automatic identification of covariate model structures. Although neural networks could potentially be used to approximate covariate-parameter relationships, such approximations are not human-readable and come at the risk of poor generalizability due to high model complexity. In the present study, a novel methodology for simultaneous selection of covariate model structure and optimization of its parameters is proposed. It is based on symbolic regression, posed as an optimization problem with smooth loss function. This enables training the model through back-propagation using efficient gradient computations. Feasibility and effectiveness is demonstrated by application to a clinical pharmacokinetic data set for propofol, containing infusion and blood sample time series from 1,031 individuals. The resulting model is compared to a published state-of-the-art model for the same data set. Our methodology finds a covariate model structure and corresponding parameter values with a slightly better fit, while relying on notably fewer covariates than the state-of the-art model. Unlike contemporary practice, finding the covariate model structure is achieved without an iterative procedure involving manual interactions.

show abstract