(175) PHARMACOM-EPI: A Framework for Integrating Pharmacometric Modelling into Pharmacoepidemiological Research Using Real-World Data

Sunday, August 28, 2022

8:00 AM – 1:45 PM CEST

Location: Hall C

Publication Number: 1220

Presenting Author(s)

Hiie Soeorg
– University of Copenhagen, Department of Drug Design and Pharmacology, København S, Hovedstaden, Denmark

Co-Author(s)

Eva Sverrisdóttir
– Pharmacometrics research group, Department of Drug Design and Pharmacology, University of Copenhagen, Denmark
Morten Andersen, MD, PhD
– Professor, University of Copenhagen, University of Copenhagen, Copenhagen, Hovedstaden, Denmark
Trine Meldgaard Lund
– Pharmacometrics research group, Department of Drug Design and Pharmacology, University of Copenhagen, Denmark
Maurizio Sessa, PhD
– Assistant Professor of Pharmacoepidemiology, University of Copenhagen, Kobenhavn O, Hovedstaden, Denmark

Background: It is often assumed that the observed association in pharmacoepidemiology is related to the pharmacological effects of the investigated drug that, in turn, are connected to its pharmacokinetic and pharmacodynamic properties, described by pharmacometric models. However, validated frameworks suggesting how to combine pharmacoepidemiology and pharmacometrics for pharmacological substantiation of observed associations are missing.

Objectives: We aimed to propose PHARMACOM-EPI, a framework for a structured approach on how to apply pharmacometric models in pharmacological substantiation of pharmacoepidemiological findings.

Methods: A step-by-step framework was developed considering all important aspects of applying pharmacometric models in pharmacoepidemiology. For proof-of-concept, we conducted a pharmacoepidemiological study in Denmark to pharmacologically substantiate death associated with valproate use in people aged 65 years or older. Additionally, using this real-world data we trained an artificial neural network (long-short term memory (LSTM)) for prediction of plasma concentrations of valproate to compare it with traditional pharmacometric models.

Results: PHARMACOM-EPI describes practical details on how to identify pharmacometric models (propose research question, conduct search, extract data about model development), characterize the models (representativeness of sample and sampling scheme used for model development, model performance), apply pharmacometric models to real-world data (choose software, format dataset, handle missing covariates/dosing data, perform external evaluation of models) and provide pharmacological substantiation of pharmacoepidemiological findings (obtain predictions and relate these to therapeutic window).
Accuracy of population predictions from eight valproate pharmacometric models to categorize concentrations into subtherapeutic ( < 50 mg/L), therapeutic (50-100 mg/L) and toxic (>100 mg/L) range was 0.07-0.50, but that of LSTM 0.64. Pharmacological substantiation of death demonstrated that in 45.6% of individuals who did not die (n=1084) individual predictions were within the subtherapeutic range compared with all individuals having subtherapeutic concentrations among them who died (n=169). Of individuals who died 66.3% (n=112) had a cause of death possibly and 33.7% (n=57) unlikely related to valproate.

Conclusions: The proposed multidisciplinary approach, PHARMACOM-EPI, has the potential to obtain real-world evidence in pharmacoepidemiology by pharmacological substantiation using pharmacometric models.