The ultimate goal of cancer research is to find treatments that can prolong patient survival. However, clinical trials in cancer are often evaluating “surrogate endpoints”—biological measurements, such as tumors shrinking (usually called “objective response rate”) or the time until tumors grow or multiply (called “progression-free survival”).
In theory, the effect that an experimental drug has on a surrogate endpoint is supposed to be a good indicator of whether that drug will also prolong life, and if this is true, then conducting trials using surrogate endpoints makes sense. Effects on a surrogate can be observed more quickly (and inexpensively) than effects on patient survival. So by using a surrogate endpoint that is strongly correlated with survival, cancer researchers can quickly figure out whether a treatment prolongs patient life.
Unfortunately, many of the surrogate endpoints used in cancer trials are not strongly correlated with survival. In fact, as we showed in a recent paper published in the European Journal of Cancer, the effects seen on surrogate endpoints are usually only weak predictors of prolonging patient life. The histogram below shows the number of published analyses that found weak, moderate, or strong correlation between surrogate endpoints and survival across different cancer settings (graded using criteria proposed by the Institute of Quality and Efficiency in Health Care).
Strength of surrogate endpoints across cancer settings
Although most of the correlations between surrogates and survival were weak, there were some cancer types where surrogates were strong predictors. For example, a study published in 2012 looking at colorectal cancer trials found that improving progression-free survival was a good predictor of improving overall survival. A study on pancreatic cancer from 2016 also showed a strong correlation between progression-free survival and overall survival.
However, for some types of cancer, there have now been many published papers evaluating the strength of the surrogate endpoints. When this happens, we found that the results have often been discordant. For example, the graph below shows a timeline for all the published analyses of surrogate endpoints in metastatic colorectal cancer.
Each diamond in the timeline represents a published study looking at the relationship between a surrogate outcome and patient survival in metastatic colorectal cancer. The size of the diamond corresponds to the number of trials included in the analysis. More trials included (and a larger diamond) typically means that we can get a more precise estimate of the association between surrogate outcome and survival. So a bigger diamond will generally mean more decisive evidence. The color of the diamond corresponds to the strength of the association: a strong correlation between surrogate and survival is green; moderate strength correlation is orange; weak correlation is blue.
Timeline of studies evaluating the strength of association between surrogate endpoints and overall survival in metastatic colorectal cancer
The variability in the strength of associations across these studies highlights one of the challenges for knowing how to use (or how not to use) surrogate endpoints. All of these analyses are looking at trials of colorectal cancer patients, and in some cases they are analyzing an overlapping set of data. Yet, these analyses do not all get the same answers, even about the same (or similar) questions.
So if, for example, one study finds that progression-free survival is a good predictor of overall survival in one set of colorectal cancer trials, this does not necessarily mean that progression-free survival can be treated as a good predictor of survival benefit for all colorectal cancer trials. Indeed, we consistently found that a surrogate could look like a strong or moderate predictor of survival benefit for one set of patients and be a weak predictor of survival benefit for another closely-related set of patients.
This suggest that when it comes to evidence about the strength of surrogate endpoints, we ought not extrapolate too broadly. Even when we find that a surrogate endpoint is strongly associated with survival, this may only reflect the realities from one set of trials. The graph below shows the timeline of analyses for all the cancer indications, and drives the point home: It is rare for there to be a consistently strong signal for surrogate endpoints in any cancer or setting.
The figure below is also interactive, and we invite you to explore the dataset and papers for yourself: You can mouse-over each diamond to see more information about the study. You can use the tools on the right to zoom, pan, or save the image. You can also click on the diamonds, which will link you directly to the publication.
Evidence map of all studies evaluating the strength of association between surrogate endpoints and overall survival in cancer