Inclusion of 'ICU-Day' in a Logistic Scoring System Improves Mortality Prediction in Cardiac Surgery
Fabian Doerr1, Matthias B Heldwein1, Ole Bayer2
1Department of Cardiothoracic Surgery, University of Cologne, Cologne, Germany.
Insights
Integrating intensive care unit (ICU) length of stay into scoring systems significantly improves mortality prediction. A logistic model incorporating an
Area of Science:
- Critical Care Medicine
- Health Services Research
- Biostatistics
Background:
- Prolonged intensive care unit (ICU) stay is a known predictor of patient mortality.
- The length of ICU stay has not been previously incorporated into additive scoring systems for mortality prediction.
- The optimal method for integrating ICU length of stay into predictive models remains unclear.
Purpose of the Study:
- To investigate the impact of integrating ICU length of stay into a cardiac surgery scoring system (CASUS).
- To compare the predictive performance of additive, modified, and logistic CASUS models.
- To determine if including ICU length of stay improves mortality prediction accuracy.
Main Methods:
- Development of a 'modified CArdiac SUrgery Score' (CASUS) by adding length of stay to the additive CASUS.
- Comparison of additive, modified, and logistic CASUS models using data from 5207 cardiac surgery patients.
- Statistical analysis included discrimination (area under the curve) and calibration (observed/expected ratio) from days 1-13.
Main Results:
- All CASUS models demonstrated good discrimination, with the logistic model showing slight superiority from day 5.
- The logistic CASUS model exhibited good calibration.
- The modified CASUS showed accurate calibration, while the additive CASUS tended to underestimate mortality risk in later days.
Conclusions:
- The integration of ICU length of stay as a variable significantly enhances mortality prediction accuracy.
- An 'ICU-day' variable is best incorporated into a logistic regression model, not an additive one.
- This finding has implications for refining risk stratification in cardiac surgery patients.
Abstract:
BACKGROUND Prolonged intensive care unit (ICU) stay is a predictor of mortality. The length of ICU stay has never been considered as a variable in an additive scoring system. How could this variable be integrated into a scoring system? Does this integration improve mortality prediction? MATERIAL AND METHODS The 'modified CArdiac SUrgery Score' (CASUS) was generated by implementing the length of stay as a new variable to the 'additive CASUS'. The 'logistic CASUS' already considers this variable. We defined outcome as ICU mortality and statistically compared the three CASUS models. Discrimination, comparison of receiver operating characteristic curves (DeLong's method), and calibration (observed/expected ratio) were analyzed on days 1-13. RESULTS Between 2007 and 2010, we included 5207 cardiac surgery patients in this prospective study. The mean age was 67.2 ± 10.9 years. The mean length of ICU stay was 4.6 ± 7.0 days and ICU mortality was 5.9%. All scores had good discrimination, with a mean area under the curve of 0.883 for the additive and modified, and 0.895 for the 'logistic CASUS'. DeLong analysis showed superiority in favor of the logistic model as from day 5. The calibration of the logistic model was good. We identified overestimation (days 1-5) and accurate (days 6-9) calibration for the additive and 'modified CASUS'. The 'modified CASUS' remained accurate but the 'additive CASUS' tended to underestimate the risk of mortality (days 10-13). CONCLUSIONS The integration of length of ICU stay as a variable improves mortality prediction significantly. An 'ICU-day' variable should be included into a logistic but not an additive model.
Related Concept Videos
01:15Receiver Operating Characteristic Plot
01:20Actuarial Approach
Consider the example of a high-risk surgical procedure with significant early-stage mortality. A two-year clinical study is conducted,...