Validation of non-invasive risk models for predicting prevalent undiagnosed chronic kidney disease (CKD) in diabetic patients receiving care at the Douala General Hospital

CaHReF 20116, Yaoundé Conges hall, 23 – 26 August 2016 , PL105

Health and Human Development (2HD) Research Group, Douala, Cameroon

Although CKD is a well-documented chronic complication of diabetes, it still remains a silent disease that presents with irreversible damage at later stages. There exist risk models for predicting prevalent undiagnosed CKD but so far only two of these models have been validated in Africa.

In a poor resource setting like ours, the use of non-invasive risk models in diabetic patients (a population already receiving medical care) will help to identify individuals with higher risk who require early referral and specialist care and other therapeutic options.

We set out to validate the Korean and Thailand CKD risk models by measuring discrimination, calibration assessing overall performance and effects of simple recalibration on their performance.
Data was gotten from medical records of diabetic patients receiving routine care at the Douala General Hospital. The Korean and the Thailand CKD risk prediction models were identified from a systematic review. The MDRD and CKD-EPI equations were used to estimate glomerular filtration rate. Discrimination was measured and compared using c-statistics, and other measures; calibration was assessed by calibration curves before and after intercept adjustment

733 participants were included in our study. Of this number 421 (57.4%) were females. The mean age was 57.0 (SD ±10.4) in the overall population. The MDRD equation diagnosed 223participants with CKD whereas the CKD-EPI equation diagnosed 194(eGFR<60ml/min/1.73 m2). The Korean model had the highest c-statistics-0.696 (0.654-0.739). After intercept adjustment, the Korean model still performed better (0.696 (0.654-0.739). There was variable but good performance across subgroups. Although we had some over and underestimation in different cases, the calibration ranged from acceptable to good. Both models performed better after recalibration.

we conclude that the Korean and Thailand models have good calibration and discrimination with modest adjustment in predicting undiagnosed CKD in diabetic patients receiving care at the Douala General Hospital.

chronic kidney disease, calibration, discrimination, validation, risk model