Kidney Week

Abstract: PO2342

Developing a Prediction Model for Incidence of Newly Detected CKD Among US Veterans, 2009-2018

Session Information

• Reassessing Race in Predicting Progression
  November 04, 2021 | Location: On-Demand, Virtual Only
  Abstract Time: 10:00 AM - 12:00 PM

Category: CKD (Non-Dialysis)

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

Authors

• Bragg-Gresham, Jennifer L., University of Michigan Medical School, Ann Arbor, Michigan, United States

Jennifer L. Bragg-Gresham, PhD

• Gillespie, Brenda W., University of Michigan School of Public Health, Ann Arbor, Michigan, United States

Brenda W. Gillespie,

• Singh, Karandeep, University of Michigan Medical School, Ann Arbor, Michigan, United States

Karandeep Singh,

• Zhang, Xiaosong, University of Michigan Medical School, Ann Arbor, Michigan, United States

Xiaosong Zhang,

• Han, Yun, University of Michigan Medical School, Ann Arbor, Michigan, United States

Yun Han,

• Steffick, Diane, University of Michigan Medical School, Ann Arbor, Michigan, United States

Diane Steffick,

• Shahinian, Vahakn, University of Michigan Medical School, Ann Arbor, Michigan, United States

Vahakn Shahinian,

• Weitzel, William, Veterans Health Administration, Ann Arbor, Michigan, United States

William Weitzel,

• Vydiswaran, V. G. Vinod, University of Michigan Medical School, Ann Arbor, Michigan, United States

V. G. Vinod Vydiswaran,

• Veinot, Tiffany C., University of Michigan School of Public Health, Ann Arbor, Michigan, United States

Tiffany C. Veinot,

• Crowley, Susan T., Veterans Health Administration, New Haven, Connecticut, United States

Susan T. Crowley,

• Saran, Rajiv, University of Michigan Medical School, Ann Arbor, Michigan, United States

Rajiv Saran,

Background

Both screening and awareness of CKD remain low in the US. We sought to develop a tool to aid physicians and health systems in identifying patients most likely to develop CKD, using a large national sample of patients in the Veterans Health Administration (VHA).

Methods

Using 29,524,195 observations from Veterans, aged 18+ with outpatient s. creatinine data (2006-2018), we modeled the probability of newly detected CKD using discreet survival methods. Veterans were screened for 2-3 years to ensure no pre-existing CKD. Newly detected CKD was defined as a diagnosis or by laboratory measurement (eGFR <60 ml/min/1.73m² or UACR 30+ mg/g). Predictors included demographics, comorbidities, nephrotoxic medications, and laboratory values updated each year. Model fit assessed by the c-statistic.

Results

The cohort had a mean age of 59 years with 89% males and 15% Black race. The average eGFR was 87 ml/min/1.73m² and median UACR was 8 mg/g, with an average of 3.9 years follow-up. The largest predictors of incident CKD were diabetes, kidney stones, urinary tract infections, sickle cell anemia, and an eGFR between 60-69 ml/min/1.73m². Concordance was high (c-statistic=0.84, Fig: ROC curve). Using a threshold of 3% risk for screening would require testing ~1/3 of Veterans (~1 million per year), yielding an 83% true positive and a 17% false negative rate.

Conclusion

We are able to accurately predict the probability of incident CKD in the VHA . This predictive model has the potential for improving targeted screening efforts for CKD, facilitating its earlier detection, raising awareness, and reducing disparities. If externally validated, the impact of these findings would be generalizable to populations/health systems beyond the VHA.

Funding

• Veterans Affairs Support