Healthcare analytics · R · Complex surveys
An interpretable statistical modeling study of demographic and clinical factors associated with one-time laboratory evidence consistent with CKD in NHANES August 2021–August 2023.
5,343
Biomarker-complete adults
14.0%
Weighted prevalence (95% CI 12.6%–15.4%)
0.789
Core model apparent weighted AUC
1,318
Core validation target
The question
Chronic kidney disease is clinically defined by abnormalities that persist for at least three months. NHANES offers nationally representative biomarker data, but each participant is measured at one point in time.
This project therefore asks which factors are associated with cross-sectional evidence consistent with CKD. It pairs a clinically precise outcome definition with complex-survey methods, a compact primary model, and explicit limits on causal and diagnostic interpretation.
Approach
Eight official CDC NHANES August 2021–August 2023 files were acquired, checked, and merged one-to-one by participant identifier. The derived data retain all survey participants so the design is established before restricting to the adult analysis domain.
Evidence consistent with CKD was defined as eGFR below 60 mL/min/1.73 m² using the race-free 2021 CKD-EPI creatinine equation or uACR of at least 30 mg/g. The wording deliberately avoids claiming a clinical diagnosis from one-time measurements.
A six-coefficient core survey-weighted logistic model estimated adjusted associations for age, sex, BMI, current smoking, hypertension, and diagnosed diabetes. A secondary model added race and ethnicity, education, and income-to-poverty ratio.
The analysis assessed missingness, nonlinearity, discrimination, calibration, alternate survey weights, and low-eGFR and albuminuria outcomes separately. Transparent event and precision criteria then informed future-study sample-size targets.
What the analysis found
Approximately one in seven U.S. adults had CKD-consistent evidence: 14.0% after applying the NHANES survey design, with a 95% confidence interval from 12.6% to 15.4%.
In the core model, each additional decade of age was associated with 64% higher adjusted odds. Hypertension was associated with 60% higher adjusted odds, while diagnosed diabetes was associated with more than threefold higher adjusted odds.
The compact core model reached an apparent weighted AUC of 0.789. The extended model reached 0.796, a small apparent difference that does not establish improved generalizable performance.
Responsible interpretation
One-time creatinine and uACR measurements cannot establish the persistence required for clinical CKD. The cross-sectional design also cannot establish temporal or causal relationships.
The discrimination and calibration estimates are apparent and may be optimistic. The extended model is secondary because its 16 coefficients exceed the survey design's 15 degrees of freedom.
For future validation, transparent precision and event criteria suggest approximately 1,318 participants for the core analysis and 2,537 if all extended-model coefficients are retained.
Explore the complete work
Review the reproducible R pipeline, analysis plan, model tables, figures, sample-size calculations, and final report.