Comparing a response rate from one drug's trial against a response rate from another drug's trial is the single most common error in retail biotech analysis. The numbers look comparable. They are not, and the reasons are structural rather than subtle.
Two trials of two drugs in the same disease still differ in almost everything that determines the number at the end. They enroll different patients: one trial's "advanced melanoma" population may be younger, earlier in treatment, or selected by a biomarker the other trial ignored. They measure different endpoints, or the same endpoint defined differently: "response" can mean tumor shrinkage confirmed at one scan or at two, assessed by the investigator or by a blinded committee. They run in different eras, against different background care: a control arm from 2016 faced a different standard of treatment than one from 2024. And they make different statistical plans, sized for different effects.
Change any one of those and the same drug produces a different number. Change all of them and the comparison stops meaning anything. This is not a pedantic point; it is why regulators, when they want to know whether drug A beats drug B, require a trial that randomizes patients between them.
Whether a result comes from a randomized controlled trial or a single-arm study is not trivia; it is often the decision. Our archive documents this directly. Replimune's RP1 received a Complete Response Letter in July 2025 and a second in April 2026 before approval in August 2026, a three-pass history in which the adequacy of the supporting trial was centrally at issue, including at the July 2026 advisory committee meeting. Capricor's Deramiocel received a CRL in July 2025, and at its advisory committee meeting in July 2026 the panel voted 3-9 against. The design and strength of the evidence, not a comparison against another drug's numbers, is what these decisions turned on.
Refusing to rank two drugs' efficacy numbers does not mean nothing can be said. Several comparisons are factual, sourced, and genuinely useful.
A head-to-head result, where one exists. If a trial randomized patients between the drug and an active comparator, that comparison is a measured fact: comparator, endpoint, result, and the trial that produced it. Most trials are not designed this way, and when they are not, the honest statement is "not studied head-to-head."
Trial design itself. Randomized versus single-arm, blinded versus open-label, the size of the study, and the choice of endpoint are objective facts from the registry record, and they are exactly what the FDA weighs.
Label against label. Once two drugs are approved, their FDA labels are primary sources. If one label restricts to a narrower population, requires monitoring the other does not, or carries a boxed warning, those are factual, citable differences that matter commercially.
Structural differences. Route of administration, dosing frequency, storage requirements, and regulatory designations are label and filing facts, and they often predict adoption better than efficacy deltas do.
Trials of different drugs enroll different populations, use different endpoints, and run in different eras. Cross-trial efficacy comparison is not valid, and we do not publish one. What is comparable: trial design, the approved label, route and dosing, and any head-to-head result that actually exists.
pdufa.bio publishes dates, decisions, trial designs, and measured stock reactions, each linked to its source. We do not publish approval probabilities, efficacy rankings, or a verdict on which drug is better, because the honest answer to "which drug won?" across two separate trials is that nobody ran the trial that would tell you.