Cardiac arrest risk standardization using administrative data compared to registry data.

Area of Science:

• Healthcare analytics
• Cardiovascular research
• Health services research

Background:

• Cardiac arrest (CA) outcome comparisons are crucial for improving resuscitation.
• Current methods rely on CA registries, which lack comprehensive hospital participation.
• Most CA patients are treated at non-registry hospitals, limiting outcome comparisons.

Purpose of the Study:

• To evaluate the performance of risk standardization modeling using administrative data versus registry data for CA outcomes.
• To determine if administrative data can achieve comparable accuracy to registry data in predicting CA survival.

Main Methods:

• Developed two logistic regression models for risk standardization: one using registry data (gold standard) and one using administrative data.
• Utilized data from 2453 patients treated between 2000-2015 across three hospitals.
• Compared model performance using c-statistics, calibration curves, and Bland-Altman plots.

Main Results:

• The administrative model achieved a c-statistic of 0.891, closely matching the registry model's 0.907.
• When using non-modifiable factors, the administrative model's c-statistic was 0.818, comparable to the registry model's 0.810.
• All developed models demonstrated good calibration, indicating reliable risk prediction.

Conclusions:

• Risk standardization using administrative data is a viable alternative to registry data, performing comparably.
• This administrative data approach provides a novel tool for comparing hospital performance in CA survival.
• The findings support broader application of administrative data for healthcare quality assessment and system-level comparisons.