In this video interview, Saamir Pasha and Carole Berini, PhD, of Ontada, describe what operationalizing regulatory-grade quality frameworks requires—and why fit-for-purpose evidence generation is a transparency and documentation discipline, not just a data cleanliness exercise.
In a recent video interview with Applied Clinical Trials, Saamir Pasha, senior biostatistician, and Carole Berini, PhD, research scientist in oncology real-world evidence, both with Ontada, discussed what it takes to generate reliable, high-quality real-world evidence from fragmented oncology data—and where the field still has significant work to do. They opened by framing the core challenge: health systems collect data to deliver care, not to support research, and the result is a fragmented landscape of EHR records, lab feeds, and genomic reports that requires substantial infrastructure to reconcile. Pasha described ON.Genuity, Ontada's common data model platform, as the foundation for that work—preserving source lineage, standardizing terminology and units, and reconciling overlapping records into an analytic-ready layer that is traceable back to its origin.
Berini addressed the practical implications of that fragmentation for research, focusing on ECOG performance status as a paradigmatic example of a clinically critical variable that is inconsistently recorded in the EHR. She described how routine lab markers could serve as validated proxies for missing functional status data—not just to recover sample size, but to recover the representativeness of the population being studied. The patients lost to missing ECOG are not random, she noted, which means the consequences of excluding them extend well beyond a statistical inconvenience.
On the regulatory quality question, both speakers emphasized that fit-for-purpose evidence generation is not a data cleanliness exercise but a documentation and transparency discipline—one that requires versioning every change, establishing benchmarks at every step, and producing a traceable record that allows others to replicate the work. Pasha framed this as building trust into the data at every stage of the model preparation process.
They closed by identifying three persistent barriers to consistent high-quality RWE generation in oncology: the foundational challenge of inconsistent data capture for key clinical concepts; structural data limitations that sophisticated modeling alone cannot overcome, as illustrated by their study's low sensitivity for poor ECOG status; and the need for earned generalizability through deliberate multi-site validation rather than direct export of models across settings.
