Stelios Tzellos | Oncology Forecasting Is Broken. Here's Why.

Stelios Tzellos

A major oncology drug missed its commercial forecast by 40% in its first year. Not because the science was wrong or the market wasn't there. The forecasting team built their model on assumptions borrowed from a different therapeutic area. They treated a first-in-class immunotherapy like a late-line chemotherapy and projected patient uptake accordingly. The result was a launch strategy built on fiction.

This happens more often than the industry admits. Stelios Tzellos of the UK has spent over a decade working in pharmaceutical analytics and oncology strategy, and he has seen the same pattern repeat across companies and indications. Forecasting teams rely on analogs that don't apply, ignore the biological nuances that drive physician behavior, and deliver numbers that look precise but aren't accurate.

The Analog Trap

Analogs are the backbone of most pharmaceutical forecasts. Find a similar drug that launched in a similar market, adjust for a few variables, and project forward. It works well enough for the fifth ACE inhibitor or the third statin. It fails completely in oncology, where every new mechanism of action changes the competitive field.

Tzellos recognized this early. During his time at GlobalData, he built epidemiology models and competitive assessments for oncology indications including Hodgkin's lymphoma. The models that worked were the ones built from the disease up, not from the market down. You had to understand treatment sequencing, patient segmentation by biomarker, and how oncologists actually make decisions in the clinic.

That requires a different skill set than most forecasting teams have. It requires someone who can read a phase 3 readout and understand what it means for the treatment algorithm, not just for the stock price.

Where the Models Go Wrong

Three things consistently break oncology forecasts. First, they underestimate how quickly combination regimens shift the standard of care. A drug that looks like a monotherapy competitor today becomes part of a combination backbone tomorrow, and patient flow models built on monotherapy assumptions collapse.

Second, they overestimate the speed of biomarker testing adoption. A targeted therapy might have a clear biological rationale and strong trial data, but if community oncologists aren't ordering the companion diagnostic, uptake stalls.

Third, they treat oncology like a single market when it's actually dozens of micro-markets defined by tumor type, line of therapy, and molecular subtype. Stelios Tzellos has worked across these segments at IQVIA and AstraZeneca, and the granularity matters. A forecast for all non-small cell lung cancer patients tells you almost nothing useful. A forecast for EGFR-mutated, second-line patients after progression on osimertinib tells you something you can act on.

What Better Forecasting Looks Like

At AstraZeneca, Tzellos leads cross-functional projects that connect business insights with product strategy and portfolio decisions. The forecasts that work are the ones that bring clinical, medical, and commercial perspectives together from the start, not the ones where the commercial team builds a model and then asks medical affairs to review it.

This isn't a new idea. But it's one the industry still struggles to execute. The organizational structures don't support it. The incentives don't reward it. And the talent pipeline doesn't produce enough people who can do it.

Fixing the Pipeline Starts with Fixing the People

Stelios Tzellos studied Biochemistry and Molecular Biology at Imperial College London before moving into pharmaceutical consulting and then industry. That progression gave him the ability to move between scientific and commercial conversations without losing fluency in either.

The industry needs more of that. Not because scientists make better analysts, but because oncology has become too complex for analysts who don't understand the science. The forecasts that miss by 40% aren't built by bad analysts. They're built by good analysts working without the right information.

Better oncology forecasting doesn't start with better software or bigger datasets. It starts with people who know what questions to ask before they open the spreadsheet.