Related Experiment Videos
Limitations of current risk-adjustment models in the era of coronary stenting
Jorge R Kizer1, Jesse A Berlin, Warren K Laskey
1Division of Cardiology, Department of Medicine, Weill Medical College, Cornell University, New York, NY, USA. jok2007@med.cornell.edu
Insights
Existing risk-adjustment models for coronary angioplasty are inaccurate for current stenting practices. Recalibration improves fit but leads to inconsistent performance assessments, necessitating new models.
Area of Science:
- Cardiology
- Health Services Research
Background:
- Risk-adjustment models are crucial for comparing coronary angioplasty outcomes.
- Contemporary interventional cardiology utilizes stents and new pharmacotherapies, potentially impacting model accuracy.
Purpose of the Study:
- To validate existing risk-adjustment models in current interventional cardiology practice.
- To assess model accuracy in predicting in-hospital mortality after coronary angioplasty.
Main Methods:
- Validated 3 principal predictive models using data from 11,681 stenting and 6,475 balloon-only angioplasty patients (July 1996-Dec 1998).
- Assessed model accuracy using discrimination (c-index) and calibration (Hosmer-Lemeshow statistic).
Main Results:
- Models retained discriminative ability (c-indexes 0.85-0.89) but showed poor calibration, significantly overestimating adverse outcomes.
- Recalibration improved model fit, but different recalibrated models yielded inconsistent laboratory performance ratings.
Conclusions:
- Established risk-adjustment models are not directly applicable to current coronary angioplasty practices involving stents.
- New, generalizable risk-adjustment models are needed and require ongoing reassessment due to evolving technology and practice patterns.
Background:
Several risk-adjustment models have been developed to compare outcomes of conventional coronary angioplasty across physicians and institutions. Yet the accuracy of these models in contemporary interventional practice--characterized by the widespread use of stents and novel adjuvant pharmacotherapies--has not been sufficiently studied.
Methods:
The principal published predictive models for inhospital mortality after angioplasty were validated in 11,681 patients undergoing coronary stenting and 6475 patients undergoing balloon-only procedures in the Society for Cardiac Angiography and Interventions registry from July 1996 to December 1998. We examined the 2 components of model accuracy: discrimination, as determined by the c-index; and calibration, as measured by the Hosmer-Lemeshow statistic and predicted-versus-observed probability plots.
Results:
The discriminative properties of the models were preserved in the validation cohort and did not differ statistically from one another (c-indexes 0.85-0.89). Hosmer-Lemeshow statistics, however, showed poor fit (P <.001), with all 3 models substantially overestimating the risk of adverse outcomes. Although recalibration of the models achieved satisfactory goodness of fit, laboratory-specific ratings differed depending on the model applied.
Conclusions:
Predictive models developed in the era of conventional angioplasty cannot be applied directly to current interventional practice. Although recalibration restores model fit, application of different recalibrated models yields inconsistent assessment of laboratory performance. Development of new, widely generalizable models is warranted, but such models will require continued reassessment as medical technology evolves and practice patterns change.