Abstract: TH-PO469
CALIBRA: A Cardiovascular, Literature-Based Risk Algorithm for CKD Patients
Session Information
- CKD: Epidemiology, Outcomes - Cardiovascular - I
November 02, 2017 | Location: Hall H, Morial Convention Center
Abstract Time: 10:00 AM - 10:00 AM
Category: Chronic Kidney Disease (Non-Dialysis)
- 303 CKD: Epidemiology, Outcomes - Cardiovascular
Authors
- Neri, Luca, Fresenius Medical Care, Bad Homburg, Germany
- Bellocchio, Francesco, Fresenius Medical Care, Bad Homburg, Germany
- Barbieri, Carlo, Fresenius Medical Care, Bad Homburg, Germany
- Mari, Flavio, Fresenius Medical Care, Bad Homburg, Germany
- Tschulena, Ulrich, Fresenius Medical Care, Bad Homburg, Germany
- Stuard, Stefano, Fresenius Medical Care, Bad Homburg, Germany
Background
Current cardiovascular (CV) risk scores may have limited generizability; do not implement patient-specific prognostic reasoning; require large datasets for derivation and cannot be easily updated as new evidence emerge. CALIBRA is a Knowledge-Driven Bayesian Network risk algorithm overcoming such limitations. We compared CALIBRA predictions against established risk scores and a data-driven model.
Methods
CALIBRA is a Bayesian Network (BN) based on several meta-analyses of original cohort studies on CV risk among CKD patients. For each potential risk factor, all effect sizes were pooled with a fixed effect method and then converted to BN weights using epidemiologic data (incidence of CV events in the population; prevalence of risk factors). CALIBRA implements personalized prognostic reasoning by yielding a summary risk score, by ranking the most impacting risk factors for each patient, by suggesting further diagnostic testing maximizing prognostic accuracy. We evaluated CALIBRA accuracy (AUC) in the EuCliD CKD stage 3-5 cohort (2011-2015) and compared it to a data-driven model (logistic regression derived in the same data).
Results
We included 32 variables from high quality studies in the final model (e.g.: socio-demographic, life-style, antropometry, comorbidities, bipochemistry tests). We used the knowledge-driven model to predict CV occurrence in the validation cohort (n=6239). There were 153 CV hospitalizations in 12 months (0,025 events/100 person*year) in the validation sample, Due to incomplete clinical records, predictions were based on 16.9±2.0 variables per patient. CALIBRA accuracy was similar to the data-driven model (AUC=0.77 for both) (fig.1)
Conclusion
CALIBRA provides unique features of prognostic reasoning along with accurate risk prediction. It represents a valid tool for risk stratification and clinical evaluation. The tool is easily and promptly updatable with the most recent scientific knowledge.