Beyond Compliance: NetraMark's Joseph Geraci, PhD, on Making AI-Driven Patient Subgroups Explainable

In this episode of Beyond Compliance, host Otis Johnson, PhD, MPA, founder and principal consultant at Vantix Operations, talks with Joseph Geraci, PhD, co-founder and chief scientific and technical officer at NetraMark, about how averaging patients into broad disease categories can obscure real treatment effects and hide the subgroups most likely to respond to a therapy.

Geraci, a mathematician whose work spans quantum computation, oncology, and neuroscience before founding NetraMark, argues that most drug failures aren't due to bad science but to trials that fail to account for biological heterogeneity, leaving real economic and clinical value on the table. He explains why black-box AI models, built on the same averaging techniques that power large language models, are poorly suited to clinical trial data and difficult to defend to biostatisticians, clinicians, and regulators. NetraMark's alternative approach, he says, is built to recognize when it doesn't have enough information, allowing it to surface model-derived patient subgroups characterized by variables clinicians can recognize and act on.

Geraci also discusses how insights from earlier-phase trials can be transported into later, more expensive phases, and how this approach could inform adaptive trial designs. He closes with a call for balancing broad, economically favorable drug labels against the biological reality of who a therapy actually helps.

Key takeaways

• Grouping patients under broad disease labels and averaging treatment effects can dilute or hide meaningful response signals in specific subgroups, potentially causing viable drugs to appear unsuccessful.
• Black-box AI models built on data-averaging techniques are difficult to defend in regulated drug development because sponsors need to understand and explain how a conclusion was reached.
• NetraMark's approach centers on a model that can acknowledge when it doesn't understand a given patient population, rather than forcing a prediction, which allows it to surface clinically recognizable subgroup characteristics.
• Insights from earlier-phase trials can potentially be transported into later, more expensive phases, helping sponsors better target enrollment and interpret placebo response before reaching costly Phase III trials.