A data-based approach to assessing clinical interventions in the setting of chronic disease.

Accurately assessing the effect of a clinical intervention is a complex task based on observing patients and their outcomes. Two strategies, the randomized trial and the observational data bank, represent different approaches toward detecting an outcome gradient across treatments. The observational data base is rooted on prospective data collected on an unselected group of patients with continuous surveillance of outcome events. As such, this approach captures a broad spectrum of patients within which prognostically important factors can be identified. In addition, the ongoing nature of the observational approach provides a means for capturing the dynamics of the disease process as well as the dynamics of the surrounding technology. These factors are particularly useful in the study of chronic illness where time frames of years to decades are encountered. In assessing treatment effects, both the randomized trial and the observational data base are subject to inferential errors due to multiple comparisons. In this paper, we describe some of the complexity associated with the study of chronic illness and demonstrate the multiple comparison problem. We have found multivariate models helpful in attenuating multiple comparison problems, but it still seems that independent confirmatory studies are required to establish confidence in a questionable intervention.

Duke Scholars

Author Charles Franklin Starmer Computer Science

Author Kerry L. Lee Biostatistics & Bioinformatics, Division of Biostatistics