Early-phase trial efficacy and safety effect estimates regress upon testing in phase 3 trials

Many drug development and care decisions rest on phase Ib or II trial evidence, which may overestimate efficacy due to regression to the mean or bias. The objective of the study was to assess concordance between efficacy and safety estimates from phase Ib/II (P1b/2) and subsequent phase III (P3) solid tumor therapy trials.

We identified P3 solid tumor trials completed 2014-2020, and then matched each to supporting phase P1b/2 trial(s) testing the same drug-indication pairing. Correlation was assessed for objective response rate (ORR), median overall survival (mOS), and serious adverse events (SAEs). We also assessed whether P3 effect sizes fell within 95% CIs of P1b/2 results and whether trial design features influenced regression.

In 87 P1b/2-P3 identified pairings, correlations were positive and significant for ORR (β = 0.78, adj. R² = 0.549, P < .001), mOS (β = 0.67, adj. R² = 0.523, P < .001), and SAE rates (β = 0.54, adj. R² = 0.285, P < .001). Regression was greater when P3 had a low risk of bias (RoB) for ORR (β = 0.660, adj. R2 = 0.445) than when RoB was high (β = 1.00, R² = 0.704). P3 estimates were at least twice as likely to fall below P1b/2 estimates than above (ORR: 21% vs 10%; mOS: 25% vs 6%; SAE: 22% vs 4%).

Efficacy and toxicity estimates from P1b/2 trials of solid tumor drugs often regress in P3. The variability between early and confirmatory trials underscores the need for caution when interpreting early-phase estimates for clinical decisions.

Phase Ib and II clinical trials play an important role in the cancer drug development. They are often used to decide whether a treatment should move forward to larger P3 trials and, in some cases, to support regulatory approval or clinical recommendations. However, results from early-phase trials may overestimate a drug's effects. We wanted to know how closely early trial results match what is later seen in P3 trials. We compared early and later trials that tested the same drug for the same type of solid tumor, looking at how often patients' tumors shrink, how long they live, and how often severe adverse events occur. We found that results from early and later trials were related, but early trials usually had better efficacy results. On average, early trials overstated tumor response rates by about one-fifth. However, SAEs were often less common than early trials suggested. This matters because early-phase trial results are widely used to guide drug approvals, clinical guidelines, and health care spending. In cancer care, treatments are sometimes adopted based mainly on early results, and many early trials are never followed by definitive P3 studies. Our findings suggest that early cancer trials are useful for spotting promising treatments, but they are not a reliable guide to how well those treatments will work in routine care. Careful P3 trials remain essential for making decisions about cancer treatment.