Overall survival (OS) is required to assess the cost-effectiveness of new oncology treatments. Trials often prioritize immediate outcomes like progression-free survival (PFS) and response to treatment and have incomplete data on OS. This is relevant for early-stage disease or treatments inducing sustained response where patients may have extended survival, limiting the number of deaths observed in the study. Modelers will sometimes rely on surrogacy analyses using aggregate data to correlate the benefits of treatment on surrogate outcomes, like progression or response, and longer-term outcomes like OS. For example, using an equation relating the magnitude of hazard ratios (HRs) for PFS to those of OS from past trials in the same type of cancer and at similar stage of disease to estimate for OS HR based on the observed HR for PFS in a new study. We caution that this approach may be problematic in two specific situations: in early stages of cancer and for treatments with a novel mechanisms of action (MOA). With early-stage cancer, where mortality isn’t yet aggressive, survival observed in trials may be highly affected by treatments assigned following discontinuation of the study treatments. Depending on the timing and location of the trials, these can vary considerably between the trials included in the surrogacy analyses and specifically from the trial for which a prediction is needed. In particular, patients in more recent trials may have access to newer, more effective later-line treatments. Thus, correlations between surrogate and OS may not be applicable to the current study. Similar issues arise with treatments with novel MOAs such as immunotherapies; these treatments can produce long sustained response, and hence prolonged survival. While limited mortality events during a trial can require relying on surrogate date, using surrogacy analyses from historical trials risks underestimating the benefits on OS in these cases.