Maximizing the efficiency of a study on alcohol-related birth defects by means of data collection and analytic strategy dissociation.

Abstract

Application of the terms "prospective" and "retrospective" to quasi/experimental research designs used in clinical investigations is often ambiguous; they often lack specification as to whether they apply to data acquisition, analytic strategy, or both. Data may be collected prospectively by preplanned protocol or retrospectively by chart review. In addition, the direction of analysis may be prospective--comparing differential outcomes for at-risk and not at-risk (exposed/not exposed) groups--or retrospective--comparing differential preceding risks for affected and nonaffected groups. Prospective data collection is advantageous in assuring the availability of crucial variables for an entire sample, but prospective analytic strategies may be inefficient, underestimating effect sizes, particularly when the outcome of interest is rare or the effect, even if large, is seen in only a small proportion of cases. In a prospective multilinear regression model of risks for lowered birthweight for gestational age, 2% of the variance was explained by alcohol variables after adjustment for confounding. In comparison, a similar multilinear regression of case-control sample, in which the frequency of the abnormal outcome is raised, showed 18% of the variance explained by alcohol variables. Data collection and analytic strategies may be dissociated. Although some statistical power may be lost because of reduced sample size, synthetic case-control analysis is efficient in examining small effects. It also has the advantage of decreasing the size of the database to be managed and the time required for analysis. Using this approach, it is feasible to perform virtually any analysis that can be done on a mainframe on medium-sized departmental computer system.