Mark R. Segal

Empirical Comparison of Approaches to Forming Strata: Using Classification Trees to Adjust for Covariates

Abstract In medical studies, the computation of a meaningful incidence rate frequently demands adjustment for covariates. One means for achieving such adjustment is standardization. The (directly) standardized rate is calculated by using the appropriate weighted average of rates over strata determined by the covariates. Thus strata formation is crucial in obtaining good results. This article compares three competing methods for stratification using data from a drug study in which the incidence rate of a certain side effect is of interest. The techniques examined are (a) contingency-table cell aggregation, (b) classification trees, and (c) logistic regression. The relative merits of the three methods are discussed.

Conventional cemented total hip arthroplasty. Assessment of clinical factors associated with revision for mechanical failure.

This study investigates risk factors associated with mechanical loosening of cemented total hip arthroplasties. Mechanical failure was evaluated using survivorship analysis on all arthroplasties performed at the authors institution by the Orthopedic Service from March 1971 until September 1983. Failure was defined as the necessity for replacement of one or more components for any reason other than infection. The failure rate was approximately 1.7% per year and, at 12 years, 20% of the hips had failed. Variables evaluated as potential risk factors for arthroplasty revision included weight, gender, age, surgeon, preoperative functional status, prosthetic type, and diagnosis. A Cox proportional hazard analysis indicated that weight (p less than 0.015) and age (p = 0.087) are important determinants of hip failure. The use of regression trees identified subsets of patients at differing risks for failure. Patients who weighed less than 75.22 kg had the best outcome with a 90% survival to 12 years. Patients weighing more than 75.22 kg are at varying risk depending on their age. These data define a subset of special-risk patients not previously described. Patients weighing more than 75.22 kg who were older than 75.4 years had a revision rate of 73% by eight years.