I, Virtual Patient

---

By Mark D. Uehling, Bio-IT World

Roche Diagnostics had an idea: Why not model diabetic patients in silico, and try to find a simple biological indicator to predict their sensitivity to insulin? Roche — a division of F. Hoffmann-La Roche Ltd. (Basel, Switzerland) — took the idea to Entelos Inc. (Foster City, CA), a company specializing in the computer modelling of diseases. Entelos finished the heart of the job — finding a new biomarker — in three months.

Entelos scientists were hardly newcomers to Type II diabetes. The company had assembled a mathematical model of the condition in its Metabolism PhysioLab, a platform for modelling disease. Parts of the company’s diabetes model had come from medical literature; parts of it had come from partners running animal studies or clinical trials. Additional pieces of the model came from knowing that testing for insulin sensitivity requires intravenous insulin and glucose infusions over the course of several hours.

With the new biomarker, insulin sensitivity could theoretically be reduced to a simple blood test. The companies say they have no immediate plans to commercialize the technology, but they don’t rule out a medical device or laboratory test incorporating the new biomarker.

Computer-generated Patients
The work began almost two years ago. Roche scientists wanted to see whether multiple metabolites and hormones from a single blood sample could be used as a single, reliable indicator of insulin sensitivity. Such biomarkers, if validated by additional research, can serve as indicators of a disease state.

The Roche-Entelos analysis was performed using 62 diabetic virtual patients (after excluding 29 computer-generated patients) who represented diverse phenotypes and pathophysiologies. The virtual patients, according to Entelos, had behaviours comparable to those observed in clinical diabetic populations for which fasted biomarkers and insulin sensitivity data were known. What’s more, the patients’ (virtual) physiological responses to a number of experimental protocols were consistent with what is observed in diabetic patients in the clinic.

The Entelos software, the company says, is capable of simulating human in vivo plasma levels for more than 20 metabolites and hormones. But to provide statistical comparisons between this research and published biomarkers for predicting insulin sensitivity, a novel “prevalence weighting” statistical scheme was developed. That math let Roche and Entelos assign a probability of occurrence for each virtual patient based on characteristics of a published clinical population.

One early question was whether Entelos’ digital patients really behaved just like living ones. The Entelos model turned out to be quite easy to validate, according to Roche, with Roche’s in-house expertise in the disease.

Problems with Real Patients
But another part of the appeal of the Entelos project, for Roche, was simply to participate in systems biology — a field that had long intrigued the company. “Systems biology” may sound futuristic, but for Roche, it meant that Entelos could simulate not one dietary state in patients, but three — something that might be impossible or impractical to do with real human volunteers. Actual patients, after all, might not comply with detailed instructions about exactly what to eat and precisely when to draw their own blood.

For Entelos, the project was an extension of other projects, with a twist. The company usually helps customers discover drugs. But Entelos believes its suite of tools and approach could be just as applicable in the diagnostics field.

As one of a handful of diseases the company has studied intensively, diabetes was a natural project for collaboration on a lab test. Ironically, the simulation of digital patients forced Entelos scientists to consider the mundane real-world questions that arise in trials with people. What if the clinical trial patients lied about fasting? What if they ate a much different meal from what they reported?

The best biomarker, Entelos and Roche agreed, should be responsive to the patient’s condition after a variety of meals. The Entelos technology allowed Roche to see that its biomarker would be responsive in a variety of nutritional scenarios.

As Roche moves forward to use Entelos’ in silico biosimulation technology for predicting other clinically important measures, the company hopes to reap significant time savings from using a single platform that can recalculate or re-fit all assay data, and also allow easier data sharing across all experiments.