Abstract:
|
Confounding is common yet challenging in retrospective cohort studies. To minimize the potential for bias between treatment groups and to enhance causal inference, various propensity score (PS) approaches have been explored. Very few comparative studies have been conducted in regards to primary Total Knee Arthroplasty (TKA) surgery, where controlling confounding factors on the causal pathway are critical. This retrospective study focused on the comparisons of propensity score approaches with the joint registry database maintained by MARCQI. A total of 96,250 TKA procedures were identified in the MARCQI database and 46,709 cases met the inclusion criteria for the case study. The aim was to determine if patients having one-day hospital stay had increased odds of 90-day readmission after primary TKA surgery compared to a 2-day hospital stay. The compared PS matching approaches included nearest-neighbor matching, caliper matching, stratification, optimal matching, and matching with Gower's distance. Hierarchical logistic regression outcome models were evaluated by adjusting for PS, stratification, inverse probability of treatment weighting (IPTW), and model fitting using matched data.
|