Kidney Week

Abstract: TH-PO382

A Predictive Model for Progression of CKD to Kidney Failure Using an Administrative Claims Database

Session Information

• CKD: Risk Scores and Translational Epidemiology
  November 07, 2019 | Location: Exhibit Hall, Walter E. Washington Convention Center
  Abstract Time: 10:00 AM - 12:00 PM

Category: CKD (Non-Dialysis)

• 2101 CKD (Non-Dialysis): Epidemiology, Risk Factors, and Prevention

Authors

• Sharma, Ajay, Healthagen Outcomes, division of CVS Health, New York, New York, United States

Ajay Sharma,

• Alvarez, Paula J., Relypsa, Inc., a Vifor Pharma Group Company, Redwood City, California, United States

Paula J. Alvarez,

• Woods, Steven D., Relypsa, Inc., a Vifor Pharma Group Company, Redwood City, California, United States

Steven D. Woods,

• Fogli, Jeanene, Relypsa, Inc., a Vifor Pharma Group Company, Redwood City, California, United States

Jeanene Fogli,

• Dai, Dingwei, Healthagen Outcomes, division of CVS Health, New York, New York, United States

Dingwei Dai,

• Mehta, Rajesh, Healthagen Outcomes, division of CVS Health, New York, New York, United States

Rajesh Mehta,

Background

To develop and validate a predictive model to identify patients (pts) with chronic kidney disease (CKD) Stage 3 or 4 at high risk for progression to kidney failure (KF) over a 24-mo period.

Methods

A predictive model was developed and validated utilizing a retrospective claims database of CKD Stage 3 or 4 patients from a large US payer. The study covered 36 mo with a 12-mo (2015) baseline period and 24-mo (2016–2017) prediction period. All pts were ≥18 yrs of age without dialysis or kidney transplant and had 36 months of enrollment. KF was defined as: eGFR <15 mL/min/1.73 m²; or dialysis; or kidney transplant; or one diagnosis (ICD-10-CM: N18.5, N18.6) in prediction period. Multivariate logistic regression was used to develop a model estimating the 2-yr probability of KF as a function of baseline covariates. Area under receiver operating characteristic (ROC) curve (AUC), calibration, gain and lift charts of the validation sample were used to assess the predictive model performance.

Results

Of the 74,114 pts studied, 2476 (3.34%) had incident KF in the prediction period (figure). The predictive model included age, gender, CKD Stage, hypertension (HTN), diabetes mellitus (DM), congestive heart failure (CHF), peripheral vascular disease (PVD), anemia, hyperkalemia (HK), and poor RAAS inhibitors adherence. The strongest predictors were CKD Stage (4 vs 3), HTN, DM and HK. The ROC curve and calibration analyses in the validation sample showed good predictive accuracy (AUC=0.834) and good calibration.

Conclusion

This predictive model provides a good level of accuracy in identifying CKD pts at high risk of progressing to KF up to 2 yrs in advance in a national health plan with over 10 million lives. Early identification using this model could potentially lead to improved health outcomes and reduce health care expenditure.

Funding

• Commercial Support –