IQWiG correlation-based surrogate validation with surrogate threshold effect (STE)
This method checks if a surrogate endpoint (like a lab result) can reliably predict the effect of a treatment on a true clinical outcome (like survival or quality of life). It uses data from multiple studies and calculates a threshold value (the Surrogate Threshold Effect, or STE) that indicates how strong the effect on the surrogate must be to expect a real benefit. The method warns against using simplified statistical techniques when full patient-level data are not available, recommending more conservative approaches instead.
At a glance
Use when
Validating surrogate endpoints in meta-analyses, especially when assessing treatment effects in chronic diseases where long-term outcomes are hard to measure; when planning trials using surrogates; when only aggregate data are available but surrogacy must still be assessed conservatively.
Avoid when
Only one or very few trials are available; when endpoints are not continuous or normally distributed without transformation; when no correlation structure can be established between surrogate and true endpoints; when the goal is purely predictive modeling rather than causal surrogacy assessment.
Inputs
Individual patient data (IPD) or aggregate trial-level data, including treatment effects on both surrogate and true endpoints, with associated variances; multiple randomized trials are required.
Outputs
Surrogate Threshold Effect (STE) estimate; R² values at individual and trial levels; assessment of surrogacy validity; comparison of bias in STE estimates across modeling approaches.
How it works
The method implements a meta-analytical, correlation-based approach for surrogate endpoint validation, building on Buyse et al. (2000) and Burzykowski & Buyse (2006). It estimates the Surrogate Threshold Effect (STE), defined as the minimum effect on a surrogate endpoint required to predict a non-zero effect on the true clinical endpoint. Validation requires strong correlations at both individual and trial levels. The IQWiG variant evaluates biases in STE estimates when using aggregate data with simplified regression models (e.g., ordinary weighted linear regression or meta-regression) versus full models based on individual patient data (IPD). Simulation results show that ordinary regression underestimates STE, while meta-regression provides conservative overestimates. Thus, meta-regression is recommended when IPD are unavailable.
- HTA domains
- Clinical Effectiveness, Safety, Patient and Social Aspects
- Assumptions
- A linear relationship exists between treatment effects on surrogate and true endpoints; the surrogate and true endpoint are strongly correlated across trials and individuals; the effect on the true endpoint can be predicted from the effect on the surrogate; trial-level and individual-level associations are aligned.
- Strengths
- Provides a quantitative threshold (STE) for meaningful surrogacy,Distinguishes between individual- and trial-level validation,Evaluates bias in commonly used simplified methods,Recommends conservative alternatives when IPD are unavailable,Based on rigorous simulation and meta-analytic principles
- Limitations
- Requires data from multiple clinical trials,Relies on availability of individual patient data for gold-standard analysis,Assumes normality and linearity in treatment effects,May not generalize to non-continuous endpoints without adaptation,Performance depends on number and heterogeneity of included studies
- Also known as
- IQWiG STE method, correlation-based surrogate validation (IQWiG), surrogate threshold effect method (IQWiG)
Questions this answers
- › Can a surrogate endpoint reliably predict clinical benefit in a new trial?
- › What is the minimum effect on a surrogate needed to expect a true treatment effect?
- › How biased are STE estimates when using aggregate data instead of individual patient data?
- › Which statistical approach should be used for STE estimation when IPD are unavailable?
- › Is the relationship between surrogate and true endpoints consistent across trials?
- › What are the risks of using simplified regression models for surrogate validation?
References & sources
Similar by meaning
Beta record. Based on the original catalogue summary; primary-source enrichment pending.