Truace
Subscribe

Application of Machine Learning Models to Predict Tigecycline Exposure in Critically Ill Patients Using Different Limited Sampling Strategies

Background The high variability in the pharmacokinetics of tigecycline may impact therapeutic efficacy, particularly in critically ill patients. This study introduces machine learning (ML) methods, aiming to clarify their effectiveness in individualized drug exposure prediction and determine the optimal modeling strategy. Methods The least absolute shrinkage and selection operator (LASSO) method was used to screen the covariates. Four ML models, random forest, extreme gradient boosting, multilayer perceptron, an…

By the Truvace desk2 min read1 sourcePeer-reviewedTRV-2026-2061· cite & verify
Application of Machine Learning Models to Predict Tigecycline Exposure in Critically Ill Patients Using Different Limited Sampling Strategies
US Navy 030415-N-4182M-001 Intensive Care Unit (ICU) nurse Lt. Cmdr. Mary Ann Brantley, Nurse Crops, works on a patient in the ICU aboard the Military Sealift Command (MSC) hospital ship USNS Comfort (T-AH 20) by U.S. Navy photo by Journalist Seaman Erica Mater.. Public domain

In brief

Researchers trained four machine learning models, random forest, extreme gradient boosting, a multilayer perceptron and an LSTM network, on 751 drug concentrations from 98 patients. The aim was to predict each patient's tigecycline exposure over 24 hours at steady state, using covariates chosen by LASSO selection.

Tigecycline levels vary widely between critically ill patients, which can affect how well treatment works, so individualized exposure estimates could support dosing decisions. The evidence is internal validation on one dataset, so how the approach performs in other hospitals or in routine care is not shown by the text.

Main points

  1. LASSO screened seven covariates: total bilirubin, creatinine clearance, SOFA score, body mass index, serum creatinine, age and prothrombin time.
  2. The LSTM was the best of the four ML models in internal validation.
  3. Three-point sampling (C0, C0.5, C6) with covariates gave R2 0.887, RMSE 2.496 mg·h/L and F20 of 75%.
  4. The trough-only (C0) model gave R2 0.818, RMSE 3.164 mg·h/L and F20 of 60%, comparable to some PopPK models.

The gain

An LSTM model trained on 98 critically ill patients estimated tigecycline 24-hour drug exposure accurately, and a single trough-sample version cut the sampling burden for therapeutic drug monitoring.

The rundown

The dataset contained 751 concentration measurements from 98 patients. Models were trained and tested to predict AUC 0-24h, the area under the steady-state plasma concentration-time curve over 24 hours.

The authors also evaluated published tigecycline population pharmacokinetic (PopPK) models on the internal test set, which allowed comparison with the ML approach.

The C0-only LSTM was described as the most concise option and better suited to practical computational conditions, with fit and systematic bias control comparable to some PopPK models, though its error was higher than the three-point model.

Sources

  1. Peer-reviewedDrug Design, Development and Therapy2026-10-14

The debate