A Readmission Risk Score for Transcatheter Aortic Valve Replacement: An Analysis of 200,000 Patients.

OBJECTIVE

The objective of this study was to leverage a national database of TAVR procedures to create a risk model for 30-day readmissions.

METHODS

The National Readmissions Database was reviewed for all TAVR procedures from 2011 to 2018. Previous ICD coding paradigms created comorbidity and complication variables from the index admission. Univariate analysis included any variables with a P-value of ≤0.2. A bootstrapped mixed-effects logistic regression was run using the hospital ID as a random effect variable. By bootstrapping, a more robust estimate of the variables' effect can be generated, reducing the risk of model overfitting. The odds ratio of variables with a P-value <0.1 was turned into a risk score following the Johnson scoring method. A mixed-effect logistic regression was run using the total risk score, and a calibration plot of the observed to expected readmission was generated.

RESULTS

A total of 237,507 TAVRs were identified, with an in-hospital mortality of 2.2 %. A total of 17.4 % % of TAVR patients were readmitted within 30 days. The median age was 82 with 46 % of the population being women. The risk score values ranged from -3 to 37 corresponding to a predicted readmission risk between 4.6 % and 80.4 %, respectively. Discharge to a short-term facility and being a resident of the hospital state were the most significant predictors of readmission. The calibration plot shows good agreement between the observed and expected readmission rates with an underestimation at higher probabilities.

CONCLUSION

The readmission risk model agrees with the observed readmissions throughout the study period. The most significant risk factors were being a resident of the hospital state and discharge to a short-term facility. This suggests that using this risk score in conjunction with enhanced post-operative care in these patients could reduce readmissions and associated hospital costs, improving outcomes.