Statistical design considerations in the NHLI multiple risk factor intervention trial (MRFIT)

https://doi.org/10.1016/0021-9681(77)90013-3Get rights and content

Abstract

This paper restricts itself to discussion of statistical design considerations; other important design issues such as how patients are being obtained, recruitment sources, eligibility criteria, procedures for classification of causes of death, etc. will be discussed in a separate paper, now in preparation.

The specific statistical aspects of trial design which are discussed are as follows:

  • 1.

    (a) The approach of assessing overall risk for an event endpoint by fitting a multiple logistic function of risk factors to incidence data is described.

  • 2.

    (b) The results of fitting such a function (including the variables, serum cholesterol (mg per dl) diastolic blood pressure (mm Hg) and reported number of cigarettes smoked per day) to Framingham data, a priori suitable for design of MRFIT, are reported. This includes data on adequacy of fit of the model to Framingham data which indicate the model fits well except possibly for underestimating risk at extremely high risk factor levels.

  • 3.

    (c) The assumptions upon which statistical computations of sample size and power were based are specified and include such refinements as taking into account expected non-compliance as it occurs over the duration of the study, the possibility of a lengthy period of treatment before maximal risk reduction is attained and a procedure for selecting individuals into the trial in such a way as to optimize statistical power.

  • 4.

    (d) The procedure used to estimate anticipated individual risk reduction is described and the imperfections and potential pitfalls of the procedure are discussed.

  • 5.

    (e) The selection procedure qualifying individuals for entrance into the trial is described in some detail; in addition the performance of the procedure is checked both by comparing average estimated risk levels and other characteristics in Framingham with components of the Pooling Project and with MRFIT for those deemed eligible for the trial and by comparing predicted and observed incidence of coronary death for Pooling Project components. Agreement between MRFIT and Framingham characteristics was excellent except for per cent non-smokers which was about 70% greater in MRFIT; agreement on most characteristics was good among Pooling Project components including Framingham and estimated incidence was consistently lower than observed incidence.

  • 6.

    (f) The potential of the trial for isolating effects of each intervention modality is discussed.

  • 7.

    (g) Discrepancies between design assumptions and the realities of the trial are reviewed and their potential impact explored.

  • 8.

    (h) Some brief remarks are made on potential uses of the risk function approach for monitoring of the trial.

  • 9.

    (i) A brief appendix describes the methodology for adjustment of sample size computations to take into account lag in attaining maximum effect of treatment and non-compliance.

References (10)

There are more references available in the full text version of this article.

Cited by (0)

The list of investigators appears at the end of the paper.

View full text