Blood transfusion in total knee arthroplasty and total hip arthroplasty: A nationwide study of complications, costs and predictive modelling

Abstract

Purpose

Blood transfusion during total knee and hip arthroplasty is associated with increased postoperative complications, prolonged hospital stays and greater healthcare costs. As outpatient arthroplasty expands, identifying patients at high transfusion risk is essential. This study analyses over 4 million arthroplasty procedures from the Nationwide Inpatient Sample (NIS) to assess the clinical and economic impact of transfusion and develop a machine learning-based risk prediction tool. We hypothesised that transfused patients would experience higher complication rates, longer hospital stays and increased hospitalisation costs.

Methods

We conducted a retrospective cohort study using NIS data (2016–2019) including primary total knee arthroplasty and total hip arthroplasty cases. Propensity score matching (PSM) was used to balance clinical and demographic variables. Outcomes included length of stay (LOS), total charges, complications and mortality. Logistic regression, random forest and deep neural networks (DNNs) were trained to predict transfusion using preoperative data. Validation methods included hold-out testing, class weighting and dropout.

Results

After matching, transfusion was linked to increased surgical site infection (TKA RR = 17.0; THA RR = 13.5), sepsis (TKA RR = 13.4; THA RR = 5.0) and pulmonary embolism (TKA RR = 6.0; THA RR = 3.5). Transfused patients had longer LOS (TKA: 4.2 vs. 2.7 days; THA: 4.0 vs. 2.9 days) and higher charges (TKA: $79,996 vs. $59,600; THA: $89,283 vs. $77,239). The DNN achieved the best predictive performance (area under the curve: 0.8644–0.8783). Top preoperative predictors of transfusion included chronic anaemia, chronic kidney disease, female gender and osteoporosis.

Conclusions

Blood transfusion significantly worsens clinical outcomes and increases costs in arthroplasty. Our machine learning tool, while not clinically implemented yet, shows promise in identifying high-risk patients and supporting preoperative planning.

Level of Evidence

Level III.