Canonical text (the exact bytes fingerprinted)

TRUVACE RECORD VERSION
record: TRV-2026-2061
version: 1
kind: certified
reason: Certified into the record
timestamp: 2026-10-21T01:09:50.161903Z
status: published
lens: g_space
sector: health
headline: Application of Machine Learning Models to Predict Tigecycline Exposure in Critically Ill Patients Using Different Limited Sampling Strategies
dek: 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…
gain_title: 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.
problem_title: (none)
trace_subject: (none)
gain_reading: 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.
gain_evidence: The LSTM model established in this study enables an accurate estimate of tigecycline AUC 0-24h | C 0 alone reduced the sampling burden
problem_reading: (none)
problem_evidence: (none)
quick_read: 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.
limitation: Results come from internal validation on a single pharmacokinetic dataset of 98 patients, and the single-sample model was less accurate than the three-point model.
tag: Evidence-backed gain
key_points: LASSO screened seven covariates: total bilirubin, creatinine clearance, SOFA score, body mass index, serum creatinine, age and prothrombin time. | The LSTM was the best of the four ML models in internal validation. | Three-point sampling (C0, C0.5, C6) with covariates gave R2 0.887, RMSE 2.496 mg·h/L and F20 of 75%. | The trough-only (C0) model gave R2 0.818, RMSE 3.164 mg·h/L and F20 of 60%, comparable to some PopPK models.
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:
- peer_reviewed | Drug Design, Development and Therapy | https://doi.org/10.2147/dddt.s630414 | 2026-10-14
prev: 0000000000000000000000000000000000000000000000000000000000000000

sha256

76144ef81fa5774f2ea66e5bcab901493bc84a9f4bd68db7aa21a0c81d8e06b4

previous

0000000000000000000000000000000000000000000000000000000000000000