Reading an oncology dataset from the patient upward

Summary tables tell you what happened on average. Patient-level review tells you why. Reading an oncology dataset well means moving between the two and being clear about what each can support.

Begin with the patients who drive the story

Look first at the responders, the patients who stayed on treatment longest, the early discontinuations and anyone with a serious or unexpected event. Reconstruct each one’s chronology: dosing, adverse events, laboratory changes, concomitant medication and tumor assessments in date order. Contradictions and implausible sequences often appear here before they are visible anywhere else.

Move to the study-level picture

Interpret response and time-to-event results against the protocol and analysis plan: who was evaluable, how follow-up is maturing, how response was assessed and how the numbers behave across subgroups. FDA’s guidance on cancer trial endpoints is a useful reference for what different endpoints can and cannot support.

Read safety and efficacy together

Responses achieved at doses that patients cannot stay on are a different finding from responses at doses that are well tolerated. Look at dose reductions, interruptions and discontinuations next to response and exposure, not in a separate safety chapter.

Make the benefit–risk judgment explicit

State who the intended patients are, what they have available now, what the agent offers and what it costs them in toxicity, with the uncertainties named. In early development this judgment shapes dose and go/no-go decisions; later it frames the regulatory conversation.

Then decide how to say it

Choose the key messages the data can bear, and the claims they cannot. A board, a regulator and a scientific audience need different emphasis, but all of them should receive the same conclusion. Prepare for the questions the data invite before someone else asks them.