Predicting 90-Day Mortality After Total Knee Replacement: Development and Validation of a Prognostic Model

Overview

This study developed and externally validated a prognostic model to predict 90-day all-cause mortality after total knee replacement (TKR) using routinely collected pre-operative data from large UK and US healthcare databases. The model demonstrated the feasibility of predicting short-term mortality risk, potentially aiding clinical decision-making and targeting interventions for high-risk patients.

Background

Total knee replacement is generally safe with low and declining rates of 90-day post-operative mortality. However, there is limited data identifying which patients are at higher risk of death shortly after surgery. Existing prediction models for post-operative outcomes often lack external validation, use variables not routinely available pre-operatively, or do not leverage comprehensive patient data. A robust, clinically applicable mortality prediction model using routinely collected pre-operative data could inform patient decisions and improve perioperative care.

Data Highlights

Candidate predictors included 89,031 variables derived from demographics, medical events, drug prescriptions, and healthcare utilization prior to surgery. Models were trained using LASSO logistic regression with internal train-test splits and externally validated across databases.

Key Findings

• 90-day all-cause mortality after TKR is low but predictable using routinely collected pre-operative data.
• The study utilized large observational datasets from the UK (THIN) and US (Optum), harmonized via the OMOP Common Data Model to ensure consistency and replicability.
• Candidate predictors encompassed demographics, comorbidities, medication use, healthcare visits, and established risk scores, totaling over 89,000 variables.
• Models were developed using LASSO regularized logistic regression with internal validation via train-test splits and externally validated across the two databases.
• External validation demonstrated model transportability between UK and US populations, supporting generalizability.
• The model’s performance was assessed using the area under the receiver operating characteristic curve (AUC), confirming predictive ability.

Clinical Implications

Clinicians can use this validated prognostic model to estimate individual patients' 90-day mortality risk before TKR surgery, facilitating shared decision-making. High-risk patients may reconsider surgery or receive targeted perioperative interventions to mitigate risk. The use of routinely collected data ensures the model’s applicability in real-world clinical settings without requiring additional testing.

Conclusion

This study successfully developed and externally validated a robust, clinically applicable model to predict 90-day mortality after total knee replacement using routinely collected pre-operative data. The model holds promise to improve patient counseling and optimize perioperative care by identifying patients at elevated risk.

References

1. Reps et al. 2021 -- Framework for patient-level prediction model development and validation
2. THIN Database -- The Health Improvement Network
3. OMOP-CDM -- Observational Medical Outcomes Partnership Common Data Model