Company Context

You’re the PM for OncoLens Evidence Cloud, a B2B real‑world evidence (RWE) analytics platform used by 25 of the top 30 pharma companies and 60 academic cancer centers. OncoLens ingests de‑identified oncology data from EHRs, claims, and tumor registries across ~18M longitudinal patient records in the US and EU. The product is subscription-based ($250K–$2M/year) with usage-based add-ons for advanced analytics.

OncoLens competes with Flatiron-like RWE vendors and internal pharma data science teams. Your differentiator is a self-serve cohort builder plus auditable analysis templates that can be exported into partner slide decks and regulatory appendices.

User / Market Scenario

A strategic pharma partner, AsterBio, is preparing for a medical affairs advisory board and wants to answer a high-stakes question quickly:

“Does Drug A work better than Drug B for EGFR exon 20 insertion NSCLC patients in 2L+?”

They are considering reallocating field medical resources and funding a prospective study. If your platform can provide credible directional evidence in 6 weeks, it increases renewal likelihood and expands your footprint into their HEOR org.

Personas

What You Have in the Database

OncoLens includes:

• Structured: demographics, diagnoses (ICD), procedures, medications (NDC/J-codes), lab values, encounter dates, death indicators.
• Oncology-specific abstractions: line of therapy (LOT), progression events (imperfect), biomarker results (EGFR variants), ECOG (sparse), smoking status (sparse).
• Unstructured-derived: pathology and biomarker extraction with ~92% precision / 85% recall.

Known limitations:

• Progression dates are noisy; overall survival (OS) is more reliable than PFS.
• Treatment exposure is captured well for infused drugs and moderately for oral drugs (adherence uncertainty).
• Some sites have delayed data feeds (up to 60 days).

Problem Statement

AsterBio wants a defensible comparison of outcomes for Drug A vs Drug B in a narrow subtype. Your platform currently supports cohort creation and descriptive outcomes, but does not have a standardized “comparative effectiveness” workflow that guides users through:

• defining index date and eligibility,
• handling confounding and selection bias,
• choosing endpoints appropriate for data quality,
• running sensitivity analyses,
• producing an exportable, auditable report.

If you ship the wrong thing, you risk:

• partners drawing incorrect conclusions,
• reputational damage with KOLs,
• compliance issues if outputs are misused as promotional claims.

Your Task (Deliverables)

1. Clarify the user’s job-to-be-done and translate the partner’s question into an analysis-ready product requirement (what decision will it inform, what level of evidence is acceptable?).
2. Design the analysis workflow using OncoLens data: cohort definitions, index date, endpoints, confounders, and comparison strategy.
3. Propose the product experience (UI + outputs): what the user configures, what is automated, what is explained, and what is exportable.
4. Prioritize MVP vs later iterations using a structured prioritization method (e.g., RICE/Kano). Explain trade-offs.
5. Define success metrics and a validation plan (how you’ll know the workflow is both useful and methodologically safe).

Constraints

• Timeline: MVP must be usable by AsterBio in 6 weeks.
• Team: 2 backend engineers, 1 data scientist, 1 analytics engineer, 1 designer (50%).
• Compliance: Outputs must include disclaimers; no “promotional” language; audit logs required.
• Data: EGFR exon 20 insertion is rare; expect small sample sizes. You must plan for low power and missingness.
• Scalability: Customer Success cannot hand-hold every analysis; the workflow must be repeatable across partners.

What Good Looks Like

AsterBio should be able to generate a report that a HEOR lead and a skeptical data scientist both consider credible, including cohort definitions, baseline balance checks, primary and sensitivity analyses, and clear limitations—without requiring custom SQL or bespoke analyst work.