Nearly one in four indigenous adults in Odisha had impaired kidney function in a survey

The study

This is a secondary analysis of data from the Odisha Tribal Family Health Survey (OTFHS, 2022-23). It included 19,703 adults from 53 indigenous communities in Odisha. Serum creatinine was measured, and estimated glomerular filtration rate (eGFR) was calculated with the CKD-EPI equation.

Impaired kidney function (IKF) was defined as eGFR <90 mL/min/1.73m2. It was graded as mild, moderate, severe or kidney failure. The authors also built two machine learning models, a deep neural network (DNN) and eXtreme gradient boosting (XGBoost), to predict IKF early.

What they found

Overall, 24.8% of participants had IKF (95% CI: 23.8-25.9). By grade, 16.6% were mild, 6.4% moderate, 1.5% severe and 0.3% kidney failure.

Prevalence rose steeply with age: 5.6% in the 18-30 y group and 79.4% in the >70 y group. IKF was more common in men and in people with lower education. It was also more common with tobacco or alcohol use, and in people with hypertension, diabetes or underweight.

The DNN model had accuracy of 0.78 and recall of 0.69. Recall rose to 0.78 at a threshold of 0.4. Decision curve analysis suggested a possible benefit of ML-based targeted screening. SHAP analysis named age, hypertension, diabetes, tobacco use, alcohol use and low dietary diversity as the key predictors.

How much weight to give it

This is a survey analysis, so it shows association, not cause. The risk factors listed are linked to IKF in this sample, but the data cannot show that they led to it.

The IKF definition (eGFR <90) includes the mild category, which was the largest group at 16.6%. The abstract does not say whether eGFR was confirmed on repeat testing or whether albuminuria was measured, so 24.8% should not be read as a CKD prevalence. The abstract does not say whether eGFR was confirmed on repeat testing, or whether albuminuria was measured. So the figure of 24.8% should not be read as a CKD prevalence.

The large sample across many communities is a strength. For the prediction models, the abstract does not report external validation, precision, or how the models would perform in a clinic. The authors describe accuracy of 0.78 and recall of 0.69 as good predictive performance; without external validation or precision data, this is hard to judge. The benefit from decision curve analysis is described as suggestive. Predictors such as age, hypertension and diabetes are already well known, so the added value of ML over simple risk scoring is not shown in the abstract.

For your clinic

If you work with tribal or other underserved communities, this analysis supports keeping kidney health on the agenda. Older age, hypertension, diabetes, tobacco and alcohol use, and underweight were linked to lower eGFR here.

For people with diabetes or hypertension, routine kidney function testing remains sensible. Follow your usual guidelines on how to test and confirm results. Do not label a person as having CKD from a single eGFR value.

The ML models are research tools at this stage. They should not change screening practice until they are validated in other settings and compared with simple approaches. Read the full paper for model details, the handling of missing data, and how mild IKF was interpreted before drawing conclusions for your own programme.