Actionable Insights Newsletter

Translating research to power whole health

Issue 11 |  April 2025

Klinrisk model shows strong performance in identifying high-risk CKD patients in the US

The Klinrisk chronic kidney disease progression model has been validated in US populations, offering potential for early intervention and cost savings, and high accuracy across healthcare systems.  

Domain(s): CKD, ESRD, machine learning

Summary

Background
Chronic kidney disease (CKD) awareness is low, and the disease is often caught late due to the asymptomatic nature of earlier stages^[1],[2]. Early intervention of high-risk (CKD) can facilitate optimal medical management and improve outcomes^[3],[4]. The previously validated Klinrisk chronic kidney disease progression model accurately predicts a 40% decline in estimated glomerular filtration rate (eGFR) or kidney failure using commonly available laboratory and demographic data in two Canadian samples^[5],[6],[7]. We aimed to test the accuracy of this model in large US commercial, Medicare, and Medicaid populations. This validation is crucial for CKD patients as the model has the potential to identify high-risk individuals early, enabling timely intervention that could change the trajectory of CKD, prevent health complications and reduce associated costs.

Methods/Design
Three cohorts, consisting of insured adults enrolled in (a) commercial, (b) Medicare, and (c) Medicaid plans from the Carelon Research Healthcare Integrated Research Database (HIRD®) between 01/01/2007 and 12/31/2020 with at least 1 serum creatinine test, an eGFR between 15ml/min/1.73m² and 180ml/min/1.73m², and at least 7 of the 19 other laboratory analytes available were developed. From a payer perspective, we validated the Klinrisk model's ability to predict CKD progression over 2 years, focusing on the composite outcome of either a sustained 40% decline in eGFR or kidney failure. Model performance was evaluated using two elements: (1) how effectively to distinguish between different outcomes by area under the receiver operator characteristic curve (AUC), and (2) accuracy of a prediction by Brier scores. We also evaluated utilization and costs of care by category of risk.

Results
In all three cohorts, the Klinrisk model was effective at distinguishing between patients with worsening CKD and those without. This is shown by its high accuracy and precision score, which are summarized in Table 1. 

Table 1 Model performance for the prediction of 40% decline in eGFR or kidney failure at 2 years

*The AUC ranges from 0.5 to 1.0. An AUC of 0.5 signifies that the test is equivalent to random chance in differentiating between individuals with and without the disease. In contrast, an AUC of 1.0 represents perfect discrimination. Generally, AUC values above 0.80 are considered clinically valuable, whereas those below 0.80 have limited clinical significance.[8]
⁺ The Brier Score ranges from 0 to 1 and measures the accuracy of a predictive model. A lower score indicates higher accuracy, with 0 representing perfect accuracy.

When we grouped patients based on their risk levels, we found a clear difference between high-risk (top 10%) and low-risk (bottom 50%) individuals.

• Increasing 8-16 times more in CKD-related inpatient admissions and ER visits for high-risk patients (Figure 1)
• At least triple all-cause costs per patient per month (PPPM) and fifteen times more CKD-related PPPM costs for high-risk patients (Figure 2)

Figure 1                                                                                                         Figure 2 

                                      

Key takeaways

Publications

• "Validation of the Klinrisk machine learning model for CKD progression in a large representative US population" submitted to Kidney International (under review).

• Tangri, Navdeep¹; Ferguson, Thomas W.¹; Bamforth, Ryan J.¹; Teng, Chia-Chen²; Smith, Joseph L.^2*; Guzman, Maria²; Goss, Ashley³. External Validation of the Klinrisk Model in US Commercial, Medicare Advantage, and Medicaid Populations: SA-OR37. Journal of the American Society of Nephrology 34(11S):p 71, November 2023. | DOI: 10.1681/ASN.20233411S171a

For more information on a specific study or to connect with the Actionable Insights Committee,
contact us at [email protected].

Sponsor: This study was conducted by Carelon Research, Inc. in collaboration with Elevance Health and Boehringer Ingelheim Pharmaceuticals, Inc.