Learning Objective 1: Describe the statistical analyses that can be used to identify systematic attrition related to an intervention
Learning Objective 2: Compare the univariate, survival analysis and Cox regression methods to assess for systematic attrition
Methods: To illustrate this, an example from a study of the effects of a 12-week muscle strengthening exercise on recovery after hospitalization for a medical event was used. The 68 men and 152 women (mean age=78 years) were randomly assigned to exercise or a control conditions. Data was collected 24 hours before discharge and 1, 2 4, 6, 8 and 12 weeks later.
Results: Attrition was 49% and 51% in the exercise and control groups, respectively. There were no significant differences in survival time (cumulative occurrence of withdrawing from the study) or hazard ratio (risk of dropping out at each time point) (p>.76 for X2s). Univariate comparisons of attrition status revealed that those who remained in the study had significantly a lower length of stay than those who dropped (5.53 and 6.61, respectively, t(218)=2.18, p=.03). There were no significant differences in attrition status for gender, treatment in ICU, and malnutrition. In contrast to these univariate differences, Cox regression revealed that age and treatment in ICU were significant contextual predictors of attrition (p<.04) while gender, malnutrition, and length of stay were not significant. Experimental status did not explain significantly more to the probability of attrition.
Conclusion: Unlike univariate comparisons of attrition, survival analysis and Cox regression provide information about systematic differences in the cumulative probability attrition over time.