Methodological Issues Associated with Using Different Cut-off Points to Categorize Outcome Variables

Abstract

Knowledge of the factors that contribute to delay in seeking medical treatment for acute myocardial infarction (AMI) provides the basis for interventions that are intended to facilitate prompt care-seeking behaviour. However, operational definitions of delay time vary across research studies. The use of inconsistent cutoff times to distinguish between delayers and non-delayers is likely to compromise comparability and generalizability of the findings across studies. The purpose of this paper is to examine the impact of inconsistent operationalization of delay, in terms of cut-off times, on the validity of research findings pertaining to identifying its predictors. Secondary data analysis was performed using a sample of 73 patients who had recently experienced out-of-hospital AMI and concluded that their symptoms were related to the heart. Several regression models were built to examine the influence of using different cut-off times (1, 2, 3, 6, and 12 hours, median delay) on the number and nature of predictors of AMI care-seeking delay. The impact of varying cut-off times on the explained variance, sensitivity, specificity, and predictive values associated with each regression model was examined. The use of different cut-off times produced different sets of independent predictors, which varied in number and nature. The variance explained by the different regression models as well as their classification indices varied. Use of different cut-off times for the definition of delay time led to inconsistent results. Thus, it is recommended that criteria be established among clinicians and researchers with regard to operationally defining care-seeking delay for AMI.