The Science

From population biology to trial probability.

QuantHealth's virtual patient engine uses mechanistic disease models, not statistical pattern matching based on past trial data.

Bayesian hierarchical

population PK/PD modeling

Scientific Foundation

Mechanistic disease modeling, not historical curve-fitting

Most clinical development prediction tools learn from historical trial datasets. They carry forward biases in earlier design choices and offer limited insight into populations or indications with little prior data. QuantHealth builds a physiological disease model from first principles, parameterizes it with published literature, and simulates individuals rather than interpolating cohort averages.

The result is interpretable, adjustable, and useful for novel mechanisms where historical data is sparse by definition.

  • Mechanistic physiological disease models
  • Literature-calibrated parameter distributions
  • Individual simulation, not cohort averaging
  • For novel mechanisms with sparse history
Virtual patient network showing population distributions and response clusters
Virtual Patient Model

Each virtual patient is a parameterized individual

In QuantHealth's system, each virtual patient is a parameterized instance of the disease model. Age, sex, baseline biomarker distributions, comorbidity prevalence, and responder phenotype are sampled from calibrated distributions. The resulting population matches observed epidemiology in aggregate, while individual responses capture realistic pharmacological heterogeneity.

A typical simulation run produces between 1,000 and 10,000 virtual patients. At each dose level, outputs show expected responders, partial responders, and non-responders across the full range.

Model Validation

How we validate before predicting

Each disease module follows a defined validation cycle. A model is not shipped merely because it fits calibration data. It is shipped when it predicts outcomes in held-out study data.

01
Literature corpus assembly. We collect relevant published trial data on endpoint definitions, baseline demographics, response distributions, and dose levels for parameter calibration and validation.
02
Mechanistic model construction. The disease pathway is represented as a system of equations. Target engagement, downstream signaling, and clinical endpoint mapping come from pathway biology, not the trial data itself.
03
Parameter calibration. Model parameters are calibrated against a training subset of the literature corpus. Bayesian inference yields posterior distributions over parameters rather than point estimates.
04
Held-out validation. The calibrated model is tested against studies excluded from calibration. Predicted response rate distributions are compared with observed outcomes. Models are accepted only when validation error falls within defined thresholds.
Regulatory Transparency

Every output is documentable for regulatory context

QuantHealth models are mechanistic rather than black-box statistical, so each input assumption and parameter choice can be recorded in a model description document for regulatory briefing packages. Sensitivity analyses are available on request to show how key assumptions affect the output range.

Every engagement also includes methodology documentation: a plain-language disease module summary, compound-specific inputs, the simulation protocol, and interpretation guidance written for clinical reviewers as well as computational biologists.

  • Model Description Document covering parameterization and assumptions for regulatory review.
  • Sensitivity Analysis Package with tornado diagrams, scenario analyses for each key parameter, on request.
  • Literature Reference Index tracing every parameter source to published literature and calibration data.
Methodology Documentation

Documented approach, not proprietary black-box

We publish technical notes on modeling methods, validation results, and disease module architecture. Clients receive relevant documentation with every engagement.

Technical Note, 2025 Mechanistic virtual patient simulation for dose-response prediction in early oncology trials
QuantHealth Methodology Series

Technical Note, 2024 Bayesian calibration of physiological disease models for clinical trial simulation: validation framework
QuantHealth Methodology Series

Technical Note, 2024 Virtual patient stratification for responder subgroup identification in Phase 2 immuno-oncology design
QuantHealth Methodology Series

Browse research notes
Methodological Deep Dive

Request a technical overview session.

Review the model architecture with a QuantHealth scientist. Bring questions about your compound's mechanism.

Schedule a Session Browse research