Construction of a surrogate variable for impotence in the Massachusetts Male Aging Study

Uncollected data must be filled in after the fact in many epidemiological studies. We compared several methods for constructing a surrogate variable for erectile potency, based on responses to related questions, in the Massachusetts Male Aging Study (MMAS). Quadratic discriminant analysis performed best among the procedures tested, achieving high rates of correct classification in cross-validation of training data and (according to posterior estimates) in classification of MMAS subjects. Two versions of the surrogate variable were constructed, one involving a discrete classification and the other involving a set of continuous probabilities. The two versions were used to investigate medical correlates of impotence in MMAS and produced similar results, although the continuous version appeared to provide higher statistical power. We conclude that construction of a surrogate variable is a viable and potentially beneficial option in epidemiological investigations.