32.9%), and the remaining 130 of 194 (67.0%) were subsequent uses of the surrogate. The number of unique surrogate measure-cancer combinations accepted for drug approval has increased over time between 1992 and 2019 (Figure), to as many as 70 (36.1%) from 2016 through 2019 (Table ). Most of these combinations had unknown postmarketing OS data (Table ).When examining the strength of correlation between the surrogate end point and OS among the 64 first-surrogate-based approvals, we found that 39 (61%) had no documented correlation, 10 (16%) had a poor correlation (r ≤ 0.7), 1 (2%) had a medium correlation (0.70 < r <0.85), and only 3 (5%) had a high correlation (r ≥ 0.85). Eleven approvals (17%) had varied levels of correlation across multiple validation studies. Among the 49 approvals with low or unknown correlation, 23 (47%) were accelerated approvals, and 26 (53%) were regular approvals.Discussion | The FDA has used surrogate end points approximately 194 times to approve cancer drugs since 1992, and about 1 in 3 times, a surrogate was used for the first time in a particular type of cancer. When this is the case, the strength of association between the surrogate end point and OS is often absent or weak. This means that the FDA's use of these surrogate measures is justified neither by strength of association (ie, ability to predict gains in OS) nor previous first use, since 1 in 3 approvals constitute the use of a surrogate end point for the first time in the treament of a specific cancer type. Our study is limited by some missing FDA label updates before 2006 that are not publicly accessible.Surrogate end points can expedite trial completion compared with OS, 4 but they add substantial uncertainty regarding whether the drugs involved improve quantity or quality of life. 6 Moreover, the FDA rarely demands stringent confirmation of clinical benefit following market approval. 3 We find that the FDA is steadily accepting more surrogate measures over time, which are not justified by scientific validity or adherence to regulatory precedent. This reflects a greater tolerance of risk, and if postmarketing studies are slow, incomplete, or demonstrate negative results, then patients experience harm and cost without the intended benefit.