Sharma A1, Alvarez P2, Woods SD3, Fogli J3, Dai D4, Mehta RR1
1Healthagen LLC, New York, NY, USA, 2Relypsa, Inc., a Vifor Pharma Group Company, Miami, FL, USA, 3Relypsa, Inc.,a Vifor Pharma Group Company, Redwood City, CA, USA, 4Healthagen LLC, Wynnewood, PA, USA

OBJECTIVES: To develop and validate a predictive model to identify patients at higher risk of developing hyperkalemia (HK) over a 12-month period.

METHODS: We utilized an administrative claims database from a US healthcare payer with 5M covered lives to select the model population. Inclusion criteria: 24 months of enrollment, ≥18 years of age, at least one estimated glomerular filtration rate (eGFR) or one chronic kidney disease (CKD) ICD-9/10-CM (i.e.: 585.x, N18.x) diagnosis code. Patients with HK or dialysis in the baseline period were excluded to predict incident HK (new cases only). The study covered 24 months with a 12-month (2016) baseline (CKD identification) period and 12-month (2017) prediction period. Models were developed using multivariate logistic regression. Patient variables included: demographics, CKD stages, comorbidities, labs, drugs known to cause HK, and healthcare resource utilization (HCRU). Predictive model performance measures included: area under receiver operating characteristic (ROC) curve (AUC), calibration, gain and lift charts; data partitions included: training, validation, and test.

RESULTS: Of 435,512 patients with CKD, 6,235 (1.43%) had incident HK in the prediction period. Patients with HK had higher comorbidity burden, use of renin-angiotensin-aldosterone system inhibitors and HCRU than those without HK. The model included 24 significant predictors. Increasing CKD stage and incremental increases in normal serum potassium levels were the strongest predictors (adjusted odds ratio [aOR] for stage 5 vs 1 = 11.62, 95% CI: 7.90–17.07; aOR for potassium per 0.1 mEq/L higher = 8.59, 95% CI: 7.85–9.40). The ROC curve and calibration analyses showed good predictive accuracy (AUC=0.836) and calibration

CONCLUSIONS: Using a single large administrative claims database, patients at high risk of HK could be identified up to 1 year in advance. Although external validation is needed, the findings support the use of this predictive model to better target at-risk patients for intervention and management of developing HK.

Conference/Value in Health Info

2019-05, ISPOR 2019, New Orleans, LA, USA

Value in Health, Volume 22, Issue S1 (2019 May)




Clinical Outcomes, Methodological & Statistical Research, Real World Data & Information Systems

Topic Subcategory

Artificial Intelligence, Machine Learning, Predictive Analytics, Clinical Outcomes Assessment, Health & Insurance Records Systems, Modeling & Simulation


Urinary/Kidney Disorders

Your browser is out-of-date

ISPOR recommends that you update your browser for more security, speed and the best experience on ispor.org. Update my browser now