<< Back to Agenda

Slide 1: Surrogate Endpoint Definition and Application: Aspects of the Observational Study and Clinical Trial Interface

Ross Prentice

Fred Hutchinson Cancer Research Center

• Surrogate outcome definition and concepts
• Role of intermediate outcome trials in hypothesis generation and development
• A lesson from observational studies and clinical trials of postmenopausal hormone therapy and cardiovascular disease

Slide 2: Surrogate Endpoint Definition and Concepts

T - time to response 'true' endpoint

{S(t); t > 0} - S(t) is history prior to follow-up time t of surrogate outcome process

x - treatment indicator vector

When can S replace T for evaluation of effects of x on T?

Define S to be a surrogate for T in respect to x if:

T independent of x <=> S independent of x

(Prentice, 1989, Statist in Med)

Slide 3: (Page of equations)

Slide 4:

• Principal point of developing these criteria is to document the very restrictive conditions under which study of the relationship between S and x provides valid information on association between T and x.
• Criterion 1 is highly restrictive, and may require S to be of high dimension to be plausible.
• Criterion 1 can never be fully established empirically, but rather requires justification on biological or mechanistic grounds.
• Even if one argues that (i) - (iii) are plausible, a short-term study with S as principal outcome variable may be poorly suited to this assessment of the overall benefits versus risks of x.

Slide 5: Meta analytic Approach

Another approach to the use of a short-term S in place of a 'true' outcome T involves exploiting a correlation between T versus x treatment effect parameters with S versus x treatment effect parameters in prior studies of similar treatments in similar populations

• Doesn't require the strong condition (i) above to hold
• Involves joint modeling of the joint distribution of T and S, given x
• Involves decisions about similarity of treatment and study populations
• May be few situations where prior studies meeting reasonable criteria exist, while relationship between T and x is in doubt?

Slide 6: Role of 'Intermediate' Outcome Trials in Hypothesis Generation and Development

• Even though intermediate outcomes S may rarely be able to replace a study of T in assessing the effects of a treatment or intervention, trials of well-selected outcomes S in relation to X are Fundamental to the development and screening of preventive interventions.
• Preventive hypotheses mainly arise from observational studies (specificity? bias?); or from therapeutic trials (timing? relevance?)
• New technologies (e.g., genomics, proteomics) have potential to allow intermediate outcome trials (e.g., human feeding trials, exercise intervention trials) to be increasingly comprehensive and informative.
• Need for trans-NIH forum to encourage such trials and to identify/prioritize interventions that may be appropriate for full-scale trial evaluations.

Slide 7: A Lesson from Combined Postmenopausal Hormone Therapy and Cardiovascular Disease

Women's Health Initiative study of estrogen plus progestin among postmenopausal women in the age range 50-79 at baseline (table)

Slide 8: Summary

• Surrogate outcomes rarely available that can provide definitive information about 'true' endpoints of interest in respect to treatment effects
• Intermediate outcome trials becoming practical that can greatly invigorate preventive intervention research agenda
• Situations where both RCT and observational data are available provide excellent opportunities to identify and address study design and analysis issues, and avoid the promulgation of inaccurate public health information.

<< Back to Agenda