A.D. Corlan > clinical prediction

Alexandru Dan Corlan MD PhD

Research Program on Clinical Prediction Models

Purpose

Methods

In principle, all credible observations should be explained by the StoMP, not unlike a meta-analysis in which a common model fits all available studies. In StoMPs, unlike in a classical meta-analysis, the model is, however, fit by different types of studies, using different observation clusters each, while in the meta-analysis, the model and each of the fit studies are the same. The StoMP is also very different from that underlying each observation study, and may include parameters and stochastic functions that are not directly measurable. A simple StoMP example is a farmacometric model.

Developing StoMPs involves the following broad steps:

• Identify and document variables, observables and stochastic functions that connect them--such as given by pharmacokinetic/equations;
• Identify and extract available observations and their context; this needs to be done in standardized computer usable format, which is being defined in the SOLON project.
• Develop mechanistic (PKPD, physiologic, empirical) and population models, pre and post processing;
• Tune models with available observations; prove the systems are overdetermined and the models are properly tuned.
• Design and construct user interfaces;
• Ensure that model implementations and the extracted observations are correct---posibly by dual implementatation/extraction;
• Document and publish models, manuals and examples;
• Design new observational studies to test predictions;
• Manage databases of models, runs and tunnings.

Online applications and resources

• Clinical event risk estimates from literature
• Stochastic calculator for the glomerular filtration rate and the probability of renal failure given serum creatinine, using the CKD-EPI formula.
• Pharmacokinetic/pharmacodynamic model of coumarinic anticoagulants
• 2D tissue simulation of AP propagation and ECG

Publications