Comparing Ordered Logistic Regression and Random Forest for Predicting Lacosamide Trough Concentration Tiers in a Real-World Cohort: A Methodological Cautionary Study
Year: 2026
Type: preprint
Study design: Observational Study
Abstract: Background and objective : Lacosamide (LCM) concentrations differ widely between patients, and it remains unclear whether routine covariates can anticipate those differences. We therefore compared ordered logistic regression (OLR) with random forest (RF) for assigning LCM trough measurements to concentration tiers, and we quantified how enzyme-inducing antiseizure medications (EIASMs) affect exposure when exposure is matched to the time of sampling. Methods : Data came from 1,655 LCM trough measurements taken from 952 patients (age 2–92 years) at a single centre. Trough concentrations were divided into low (<2.5), therapeutic (2.5–6.5), and high (>6.5 μg/mL) tiers. An OLR model and a class-weighted RF were fitted to 916 complete-case records, without imputing missing covariates; predictors were age, sex, body weight, weight-adjusted dose, EIASM use within ±7 days, and the number of concomitant AEDs. Discrimination and calibration were then evaluated. Results : Prediction was poor for both models. OLR gave a weighted AUC of 0.542, a macro-F1 of 0.329, and 56.4% accuracy; the corresponding values for RF were 0.614, 0.411, and 50.9%. The proportional-odds assumption was not violated (P = 0.536). EIASM use within ±7 days lowered the odds of a higher tier (odds ratio 0.518; 95% CI 0.291–0.923; P = 0.026), while discrimination stayed poor in all sensitivity analyses (weighted AUC 0.51–0.66). Conclusions : Neither model could predict an LCM concentration tier from a single TDM reading plus routine covariates. The only clear pharmacological signal was that temporally matched EIASM exposure lowered the odds of a higher tier by a clinically meaningful margin. Prospective longitudinal designs are needed.
Source: Research Square
Institution: Zhejiang University
Language: English
Subfield: Clinical Neurology
Field: Medicine
Domain: Health Sciences
Open Access status: green
OpenAlex ID: W7204887643
