More than 60% of drugs approvals in the United States over the past twenty years have been based on indirect indicators, known as surrogate endpoints: measures that do not directly assess clinical benefit for patients but are expected to predict it.
A classic example is LDL cholesterol. Lowering LDL is seen as a positive signal because we know it is associated with a reduced risk of cardiovascular events and mortality. For a parameter to be truly reliable, like LDL cholesterol, it must be backed by robust data verifying that when it improves, consistent and tangible long-term clinical benefits result. But demonstrating this degree of predictive value is not easy, which is why many of these biomarkers have not been fully validated.
In such cases, the risk is that regulatory and reimbursement-related decisions involving billions of dollars are made on shaky ground. To address this dilemma, Oriana Ciani coordinated the international task force of the International Society for Pharmacoeconomics and Outcomes Research (ISPOR), which published a report on the subject. In it, the authors propose a rigorous framework for determining when these indicators are reliable and when instead they may mislead public and private decision-makers in the pharmaceutical sector. The report is consequential because it systematically links the statistical validation of surrogate endpoints with the economic models used to quantify a drug’s value.
Cutting time and costs
In clinical research, surrogate endpoints have become commonplace because they cut the time and cost of studies. In fact, instead of waiting years to observe a final outcome (such as survival), researchers use intermediate indicators such as certain biomarkers or measures of disease progression.
This practice has also proliferated due to accelerated approval programs adopted by regulatory authorities. However, while the issue is already widely debated among regulators, the world of economic evaluation (HTA, Health Technology Assessment) still lacks a shared approach. Agencies diverge considerably in the criteria they apply, creating uncertainty for companies and public decision-makers.
The literature identifies three levels of evidence for establishing whether a surrogate endpoint is valid. We’ll illustrate them using LDL cholesterol as an indicator of cardiovascular mortality.
- Relationship between treatment effect on the surrogate and on the final outcome (Level 1). This is the strongest level of evidence. Multiple clinical studies must demonstrate that treatments that lower LDL cholesterol (such as statins) also decrease mortality rates. Only then does the indicator become genuinely predictive of a patient-relevant outcome.
- Relationship between the surrogate and the final outcome at the individual level (Level 2). At this level, researchers observe that, among individual patients, lower LDL is associated with a reduced risk of cardiovascular events or death. This is what we might call a prognostic indicator.
- Biological plausibility (Level 3). Here, theory comes into play. We know that cholesterol contributes to the formation of atherosclerotic plaques, which in turn increase the risk of heart attacks and strokes.
These three levels are not equivalent; biological plausibility and individual-level correlations are not enough. Without robust evidence at the first level (that is, proof of the ability to predict the clinical effects of treatments), even an apparently intuitive parameter may prove insufficient for guiding policy or investment decisions.
This leads to the central research question: How can these indicators be rigorously evaluated and credibly incorporated into economic and healthcare decision-making?
An international task force
The report mentioned above was compiled by an international task force bringing together statisticians, health economists, regulators, and industry representatives. The objective was to define good practices for using surrogate endpoints in HTA decision-making.
From a methodological standpoint, the main contribution of the report concerns statistical techniques for validating these indicators. The researchers recommend multivariate meta-analytic approaches, which combine data from multiple clinical studies, accounting for correlations among variables. They also advise making preferred use of individual patient data whenever available, and applying Bayesian approaches capable of integrating evidence from different contexts (drug classes, clinical indications), particularly when data are limited.
The key, according to the authors, is integration with economic models. In fact, cost-effectiveness models must be consistent with the evidence supporting surrogacy. What’s more, uncertainty in the relationship between the surrogate and final outcomes must be explicitly modeled, for example through sensitivity analyses and alternative scenarios.
The report also highlights the increasingly vital role of real-world data, in other words, information collected outside randomized clinical trials, in routine clinical practice. Although more susceptible to bias, these data are especially useful when clinical trials are incomplete or limited in number.
Identifying, accepting, and managing uncertainty
For surrogate outcomes to be useful tools, they must be used with methodological rigor and transparency. But in the absence of solid validation, surrogates amplify uncertainty and the risk of ineffective or inefficient decisions.
When making investment and pricing decisions, companies and public authorities should explicitly factor in the degree of validity of the endpoints they are using. By integrating advanced statistical expertise into evaluation processes, economic models can be designed to incorporate uncertainty and make it visible.
Surrogate endpoints can accelerate innovation, but only if the uncertainty surrounding them is acknowledged and properly managed.
Sylwia Bujkiewicz , Oriana Ciani , Bart Heeg et Al. “ Methods for Evaluation of Surrogate Endpoints for Health Technology Assessment Decision Making: A Good Practices Report of an ISPOR Task Force .” Value in Health , 2026, Published online. DOI: https://doi.org/ 10.1016/j.jval.2026.01.020 .
