Skip to content

FDA quantitative medicine models

FDA defines quantitative medicine as follows (Federal Register 2026-15887):

Quantitative medicine approaches integrate mathematical and computational models with biological, clinical, and statistical data to inform drug development, regulatory, and clinical decisions.

FDA does not publish a catalog of the quantitative medicine models it has accepted or relied on. The closest list is CDER's Fit-for-Purpose table (current as of 10 January 2025), which holds four tools, three of them statistical methods. The CDER Quantitative Medicine Center of Excellence's Progress and Priorities Update (October 2025) says the Center "aims to develop knowledge management resources to catalog past and current QM related regulatory interactions", and that commenters on an earlier FDA request asked for "improved understanding of regulatory precedence with examples".

This report records nine models in two groups:

  • Models FDA has acted on in a specific decision, plus one academic benchmark without FDA standing (five models).
  • The four Fit-for-Purpose tools.

For each model it records what DisMech holds and where the model belongs. It also records the validation and regulatory properties that the DisMech schema has no slot for today. Curation of the models into kb/ is tracked in dismech#16321.

How to read the ratings

Validation ladder. Each level presumes the ones before it. The scores are this report's reading of the published record, not an FDA qualification.

Level Meaning
0 Context of use stated: question, decision, population, acceptable performance
1 Reproducible: code and parameters available, result can be regenerated
2 Biologically valid: outputs match independent human data for the step it represents
3 Intervention response: drug or genetic perturbations move outputs as seen clinically
4 Prospective prediction: an outcome was predicted before it was observed and acted on

FDA-side strength rates only the FDA-side record, not the model:

Rating Meaning
Strong FDA's own published words
Moderate An FDA-hosted document written by the sponsor
Developer-reported No FDA document found; acceptance is reported by the developers
None FDA has not acted on the model

Ladder marks: ● met, ◐ partly met, ○ not met, ? not determined.

Where level 1 is ? (code or parameters not public, or reproducibility not checked), the marks for levels 2–4 record this report's reading of the published evidence for each level on its own. They are not met in the cumulative sense until level 1 is determined.

Summary

Model Type Ladder 0·1·2·3·4 FDA status FDA-side strength DisMech target In DisMech
Pembrolizumab every-6-weeks PK and exposure–response Population PK, exposure–response ●?●●● Basis of 2020 dosing approval Strong Cutaneous_Melanoma No
CiPA in silico torsade model Cardiac electrophysiology (modified O'Hara-Rudy) ●●●●◐ Developed largely in FDA labs; advisory committee endorsed the approach in 2017 Strong Long_QT_Syndrome Yes, partly
Ibrutinib PBPK Physiologically based pharmacokinetic ●?●●◐ Informed drug–drug interaction labeling Moderate Chronic_Lymphocytic_Leukemia No
UVA/Padova type 1 diabetes simulator Glucose–insulin physiology, virtual population ●?●●◐ Reported accepted in place of animal studies (2008) Developer-reported Type_I_Diabetes No
Populations of human cardiac models Cardiac electrophysiology, model population ●?●●◐ None None Long_QT_Syndrome No
CAMD Alzheimer's disease progression model Disease progression (placebo, drop-out) not scored Fit-for-Purpose, 12 Jun 2013 Strong Alzheimer_Disease No
MCP-Mod Dose-finding statistical method not scored Fit-for-Purpose, 26 May 2016 Strong None Not applicable
BOIN design Dose-finding statistical method not scored Fit-for-Purpose, 10 Dec 2021 Strong None Not applicable
Empirically based Bayesian Emax models Dose-finding statistical method not scored Fit-for-Purpose, 5 Aug 2022 Strong None Not applicable

The four Fit-for-Purpose tools are rated Strong because FDA lists them on its own page. Their ladder levels were not scored for this report.

Models FDA has acted on

Pembrolizumab every-6-weeks dosing

  • Type: Population pharmacokinetic and exposure–response models.
  • Developer: Merck.
  • Context of use: Whether 400 mg every 6 weeks gives exposures comparable to the approved regimens, across all adult indications.
  • Data integrated: The model predicted exposures for 400 mg every 6 weeks. These were compared with observed exposures under 2 mg/kg every 3 weeks, 200 mg every 3 weeks and 10 mg/kg every 2 weeks, supported by exposure–response analyses across the development program. Interim pharmacokinetics and response came from KEYNOTE-555 cohort B, in melanoma.
  • Validation: ●?●●●. Interim KEYNOTE-555 data checked the predicted exposures; FDA reported an overall response rate of 39% (95% CI 24–55) in the first 44 patients.
  • Reproducibility: Not checked.
  • FDA status: Accelerated approval of the new schedule on 28 April 2020. Continued approval may depend on a confirmatory trial.
  • FDA-side document: FDA approves new dosing regimen for pembrolizumab. Strength: Strong.
  • DisMech: Not present. Pembrolizumab is mentioned in 66 entries; Cutaneous_Melanoma is the proposed home because the confirming cohort was melanoma. This is a model of a drug, not a disease mechanism; see the open question in dismech#16321.
  • References: Lala et al., Eur J Cancer 2020, PMID:32305010; KEYNOTE-555 cohort B, PLoS One 2024, PMID:39531423.

CiPA in silico torsade model

  • Type: Human ventricular cardiomyocyte electrophysiology model, a modified O'Hara-Rudy model with dynamic hERG (IKr) drug binding.
  • Developer: FDA, within the Comprehensive in vitro Proarrhythmia Assay (CiPA) initiative.
  • Context of use: Classify a drug's torsade de pointes risk from in vitro ion-channel data.
  • Data integrated: In vitro block data for several cardiac ion channels are the model input. Outputs are compared with reference drugs of known clinical risk. FDA's page also describes checks against human stem-cell cardiomyocyte assays and phase 1 ECG biomarkers.
  • Validation: ●●●●◐. Trained on 12 reference drugs and assessed on further drugs of known risk. As with the populations-of-models benchmark below, that assessment is retrospective on compounds whose clinical risk was already known, so level 4 is partly met.
  • Reproducibility: Code at github.com/FDA/CiPA.
  • FDA status: An FDA advisory committee endorsed the approach in March 2017, pending ongoing validation studies.
  • FDA-side document: Developing a mechanistic model-based approach to assess cardiac safety of new drugs. Strength: Strong. The page dates from 2017, so the current regulatory status should be checked against the ICH E14/S7B Q&A.
  • DisMech: On Long_QT_Syndrome as three models: the original O'Hara-Rudy model (PMID:21637795, CellML), the "CiPA IKr-Dynamic ORd Proarrhythmia Model" (PMID:28878692) and ToR-ORd (PMID:31868580). None of their mechanism links records relationship, fidelity, limitations or model scale, and neither validation paper is cited.
  • References: Dutta et al., Front Physiol 2017, PMID:28878692; Li et al., Clin Pharmacol Ther 2019, PMID:30151907; Li et al., Clin Pharmacol Ther 2020, PMID:31709525.

Ibrutinib PBPK

  • Type: Physiologically based pharmacokinetic model.
  • Developer: Janssen, built in the Simcyp simulator.
  • Context of use: Predict ibrutinib exposure with CYP3A4 inhibitors and inducers, to set dose adjustments for co-medications.
  • Data integrated: Built from in vitro data. Verified against clinical studies of a strong CYP3A4 inhibitor and a strong inducer.
  • Validation: ●?●●◐. Predictions for unstudied interactions were used on the label, but those were not then confirmed by observation. The hepatic-impairment prediction overestimated exposure against interim clinical data, and a clinical hepatic-impairment study was still required.
  • Reproducibility: Not checked; Simcyp is commercial software.
  • FDA status: Secondary sources report that 24 drug–drug interaction and dosing claims on the 2013 label came from model predictions without clinical studies. The FDA review on Drugs@FDA has not been checked against that figure.
  • FDA-side document: FDA-hosted sponsor slides on post-approval MIDD for ibrutinib. Strength: Moderate.
  • DisMech: Not present. Ibrutinib is mentioned in 6 entries; Chronic_Lymphocytic_Leukemia is the proposed home. This is a model of a drug, not a disease mechanism.
  • References: de Zwart et al., Clin Pharmacol Ther 2016, PMID:27367453; survey of PBPK in NDA reviews, PMC5414891.

UVA/Padova type 1 diabetes simulator

  • Type: Glucose–insulin physiology model with a population of virtual subjects.
  • Developer: University of Virginia and University of Padova.
  • Context of use: Preclinical testing of insulin-dosing strategies and closed-loop (artificial pancreas) control algorithms.
  • Data integrated: Meal-test data from people with and without diabetes calibrate the model. Later versions added hypoglycemia counter-regulation (S2013) and single-day scenarios (S2017).
  • Validation: ●?●●◐. Used to screen control algorithms before human trials; prospective agreement with trial outcomes is reported by the developers but not assessed here.
  • Reproducibility: Distributed commercially as T1DMS by The Epsilon Group; source not public.
  • FDA status: The developers report that FDA accepted the simulator in January 2008 as a substitute for animal trials in preclinical testing, through a master file that is not public. A 2020 preprint reports that FDA evaluated 140 candidate algorithms with it, leading to 16 investigational device exemptions.
  • FDA-side document: None found. Strength: Developer-reported.
  • DisMech: Not present. Type_I_Diabetes (MONDO:0005147, type 1 diabetes mellitus) exists and has no computational_models block, so it is the proposed home. Type 1 diabetes also appears as a subtype inside Diabetes_Mellitus, and stubs exist for monogenic and susceptibility subtypes such as Type_1_Diabetes_Mellitus_2.
  • References: Dalla Man et al., IEEE Trans Biomed Eng 2007, PMID:17926672; Dalla Man et al., J Diabetes Sci Technol 2014, PMID:24876534; Visentin et al., J Diabetes Sci Technol 2023, PMID:37743740.

Populations of human cardiac models (academic benchmark)

  • Type: Population of 1,213 human ventricular action-potential models.
  • Developer: University of Oxford.
  • Context of use: Predict clinical pro-arrhythmic risk of drugs.
  • Data integrated: Models calibrated to human recordings; in silico trials of 62 reference compounds at several concentrations, compared with known clinical risk.
  • Validation: ●?●●◐. 89% accuracy against clinical risk; the NC3Rs reports up to 75% for animal studies on similar data sets. The benchmark is retrospective on known compounds.
  • Reproducibility: Not checked.
  • FDA status: None. Recognized by the UK NC3Rs with the 2017 International 3Rs Prize.
  • FDA-side document: None. Strength: None.
  • DisMech: Not present. Proposed home Long_QT_Syndrome, next to the CiPA models.
  • References: Passini et al., Front Physiol 2017, PMID:28955244; NC3Rs announcement.

Fit-for-Purpose tools

FDA designates a tool fit-for-purpose when formal qualification is not available, "based on the acceptance of the proposed tool following a thorough evaluation". Each row on FDA's table links a determination letter.

CAMD Alzheimer's disease placebo/disease progression model

  • Type: Disease progression model of cognitive decline (ADAS-cog) with placebo, demographic and drop-out components, for clinical trial simulation.
  • Submitter: Coalition Against Major Diseases (CAMD), Critical Path Institute.
  • Context of use: Simulating mild-to-moderate Alzheimer's disease trials for design decisions.
  • Data integrated: Patient-level data from trials and the Alzheimer's Disease Neuroimaging Initiative, combined with summary-level literature data in a beta-regression meta-analysis.
  • Validation: Not scored for this report.
  • Reproducibility: Not checked.
  • FDA status: Fit-for-Purpose, issued 12 June 2013. The EMA later qualified the same tool.
  • FDA-side document: FDA Fit-for-Purpose table. Strength: Strong.
  • DisMech: Not present. Alzheimer_Disease carries only the DADD digital twin.
  • References: Rogers et al., J Pharmacokinet Pharmacodyn 2012, PMID:22821139; Romero et al., Clin Pharmacol Ther 2015, PMID:25669145.

MCP-Mod

  • Type: Statistical method for phase II dose finding under model uncertainty. It tests for a dose–response signal, then fits candidate models to estimate target doses.
  • Submitter: Janssen and Novartis.
  • FDA status: Fit-for-Purpose, issued 26 May 2016. Strength: Strong.
  • DisMech: No disease home; recorded here only.
  • References: Bretz et al., Biometrics 2005, PMID:16135025.

Bayesian optimal interval (BOIN) design

  • Type: Statistical design for phase I dose escalation.
  • Submitter: Ying Yuan, MD Anderson Cancer Center.
  • FDA status: Fit-for-Purpose, issued 10 December 2021. Strength: Strong.
  • DisMech: No disease home; recorded here only.
  • References: Yuan et al., Clin Cancer Res 2016, PMID:27407096.

Empirically based Bayesian Emax models

  • Type: Statistical method for dose–response estimation using Emax models with priors derived from historical dose–response data.
  • Submitter: Pfizer.
  • FDA status: Fit-for-Purpose, issued 5 August 2022. Strength: Strong.
  • DisMech: No disease home; recorded here only.
  • References: Primary publication not identified for this report.

What the DisMech schema cannot hold today

computational_models records a model's type, repository, publication, and its links to mechanism nodes with relationship, fidelity, limitations, model scale and typed divergences. It has no slot for:

  • the model's stated context of use;
  • the validation level reached against that context of use;
  • the regulatory status and the agency document behind it;
  • models of a drug rather than a disease, such as PBPK and exposure–response models. The one precedent is "PBPK Model for GLP-1 Receptor Agonists" on Type_2_Diabetes_Mellitus.

Until the schema has these, this report is where they are kept.