Interpreting Multiple Logistic Regression Coefficients in Prospective Observational Studies

Multiple logistic models are frequently used in observational studies to assess the contribution of risk factors to disease. In the presence of correlation among risk factors, the estimated magnitude of a multiple logistic coefficient can become uncertain or meaningless. This paper highlights the problem of interpreting a multiple logistic coefficient and suggests a procedure for examining the total contribution of a risk factor to disease that includes a direct association and associations that exist through relationships with other antecedent characteristics. Examples are given, along with results that are not immediately obvious when considering the multiple logistic coefficient alone. Conclusions that are presented are important in biological studies if isolating the effect of an antecedent characteristic is unreasonable in the presence of confounding influences.