Prediction of new onset of end stage renal disease in Chinese patients with type 2 diabetes mellitus – a population-based retrospective cohort study

Current prevalence of diabetes mellitus is one in 11 adults, affecting 415 million people, and it is estimated to increase to 642 million all over the world [1]. Diabetes is the leading cause of end-stage renal disease (ESRD), which is an irreversible loss of renal function and is fatal without receiving renal replacement therapy [2, 3]. The coexistence of diabetes and ESRD elevate the risk of mortality significantly [4‐6]. Although the number of patients with ESRD account for only 0.1%–0.2% of the total population in developed countries, health spending on renal failure was 2–3% of the total healthcare expenditure [7]. Given available effective treatments in reduction of development and progression of diabetic kidney disease [8], identification of diabetic patients who are at high risk of ESRD is needed to allow target delivery of proper healthcare and facilitate service policy planning.

Several national guidelines including the National Institute for Health and Care Excellence from UK and the American Diabetes Association recommend the regular screening for diabetic kidney disease in diabetic patients [8, 9]. On one hand, some prediction models established from general population like QKidney model [10‐12] included diabetes as one of the predictors only, without examining other clinical indicators such as haemoglobin A1c (HbA1c), estimated glomerular filtration rate (eGFR) and urine albumin to creatinine ratio (ACR), which are associated with the risk of developing ESRD [11, 13, 14]. Given the 3 to 5 folds higher risk of ESRD incidence in diabetic patients compared to non-diabetic patients [10], the type and magnitude of the association of the predictors for ESRD for diabetic population may differ from those for non-diabetic population. Only a few studies have established risk prediction models in ESRD such as the New Zealand Diabetes Cohort Study and the Action in Diabetes and Vascular Disease: Preterax and Diamicron-MR Controlled Evaluation (ADVANCE). These studies were done predominantly in a non-Chinese diabetic population [4, 15]. Studies have shown different ESRD incidence rates in different racial groups including Chinese populations [16‐18]. Hence, these models may not be applied to a Chinese diabetic population. While a ESRD risk prediction model was derived from Chinese diabetic patients managed in secondary care, one of the predictors in this model, haematocrit, is not routinely available in primary care and thus this model may be more suitable to be applied in secondary care than in primary care [19]. In addition, a previous study reported that there was a potential difference in predictors of ESRD incidence between male and female [10], but most of established models without stratifying gender may not obtain accurate predicted ESRD risk. Therefore, there is a need to develop a more accurate prediction model by stratifying gender for ESRD risk based on Chinese primary care patient with diabetes.

Since there were no population-based studies on ESRD risk prediction models specific to Chinese primary care diabetic population, the aim of this study was to develop a 5-year ESRD risk prediction model among Chinese patients with Type 2 DM (T2DM) in primary care.

There were a total of 149,333 Chinese T2DM patients aged between 18 and 79 years receiving care in primary care clinics of HA between 1 January 2010 and 31 December 2010. After excluding 10,789 patients had CVD at baseline, 21,426 patients had ESRD at baseline and 609 patients had no follow-up record, the remaining 116,509 diabetic patients, including 54,472 males and 62,037 females were included in the main analysis. The three lowest data completion rates for STDR, urine ACR and waist circumference were 70%, 72% and 83%, respectively, while other potential predictors were higher than 90%. During a median follow-up period of 5 years (range: 0.04 to 6.04 years), the incidence rate of ESRD per 1000 person-years was 0.41 (95% CI: 0.40–0.43), while that in male and female groups were 0.50 (95% CI: 0.47–0.52) and 0.34 (95% CI: 0.32–0.36), respectively. For each gender, Table 1 compares the baseline characteristics between derivation and validation cohorts after multiple imputation. For the derivation cohorts, the mean ages for male and female were 61.4 and 62.7 years, respectively. The two cohorts did not show any significant differences in all potential predictors for each gender.

After selecting the predictors by forward stepwise method (shown in Additional file 1: Table S1a and 1b), Table 2 and Additional file 2: Table S2 show the estimated risk prediction models for each gender by Cox proportional hazards regression. Common predictors to both male and female included older age, usages of anti-hypertensive drugs, anti-glucose oral drugs and insulin, and increased SBP, urine ACR and eGFR. Moreover, quadratic effects of HbA1c and DBP were significantly associated with increased risk of ESRD. For male T2DM subjects, additional predictors of ESRD risk were smoking and presence of STDR. The magnitudes of the association for insulin and urine ACR also decreased across age. For female T2DM subjects, additional predictors included longer duration of T2DM and the quadratic effect of BMI. Similarly, the magnitude of the association for eGFR also decreased across age. To summarise, the developed risk prediction model for male included 11 predictors: age, smoking status, presence of STDR, usages of anti-hypertensive drugs, anti-glucose oral drugs and insulin, HbA1c, SBP, DBP, urine ACR and eGFR, whereas the one for female included 12 predictors: age, duration of T2DM, usages of anti-hypertensive drugs, anti-glucose oral drugs and insulin, BMI, HbA1c, SBP, DBP, urine ACR and eGFR. The formulae for the derivation of the predicted ESRD risk were shown in Additional file 3: Table S3.

Table 3 compares the performance of the newly developed prediction models with ADVANCE and New Zealand ESRD risk scores using validation cohort. In terms of prediction power, both the new and the New Zealand models performed better than the ADVANCE model for female while there was no significant difference among the three models for male. Figure 1 displays the calibration plot on the predicted risk and the observed ESRD risk at 5 years in each gender. The new model demonstrated better calibration than the other models.

This is the first study to develop the prediction model for ESRD risk among Chinese population-based primary care patient with diabetes. Our findings showed that current ESRD risk prediction models such as ADVANCE and New Zealand had poor calibration power, which substantially underestimated the actual risk of ESRD in Chinese diabetic population. Our results showed that the prediction model for ESRD should be ethnic-specific and gender-specific. Moreover, we confirmed the importance of eGFR and urine ACR in predicting ESRD and found that the renal impairments interacted with age on the risk of ESRD for Chinese diabetic patients. Our model could identify diabetic patients who are at high risk of ESRD in order to counsel patients, allow target delivery of proper healthcare and facilitate service policy planning.

In comparison to our model, the ADVANCE and New Zealand models had comparable discriminatory power but lack of calibration power, indicating the apparent discrepancy between observed and predicted risks. As a consequence, the ultimate goal of the earlier identification for patients at high risk of ESRD may be defeated by the inaccurate predicted risk. Current results demonstrated that the ADVANCE and New Zealand models underestimated risk among Chinese diabetic population. Ethnic disparities in the risk of ESRD may be a potential explanation for this discrepancy. Although the ADVANCE model showed no difference in the risk of ESRD between Asian and non-Asian, numerous multi-ethnic studies showed that Asian diabetic patients had doubled the risk of albuminuria compared to Caucasian diabetic patients [18, 32, 33]. Meanwhile, the New Zealand model included ethnicity as a predictor to allow the modification for the various risks of ESRD in different ethnic groups. Nevertheless, only a small proportion of East Asian subjects (717 out of 25,736) was included in their sample and it may not be fully representative of a Chinese diabetic population. Moreover, a multinational study conducted by the World Health Organisation found that Chinese patients had a higher prevalence of proteinuria compared to other populations [34, 35]. A recent epidemiological study also illustrated that the risk of ESRD in Asian diabetic population varied widely between Chinese, Malay and Asian Indian [36]. The ethnic differences in disease profile may be attributable to genetic variants and the environmental factors such as health care policy and cultural behaviours between diabetic populations [34‐37], and thus the prediction model for ESRD should be ethnic-specific. The external validation should be conducted in the future study in order to validate our model in other Chinese populations.

The key novelty of our study was to extend on the previous models for the risk of ESRD by stratifying gender and including age effect on the renal impairment measured by urine ACR and eGFR. This study supports prior observations that higher risk of renal impairment in male compared to female [23‐28] and the current results prolonged this manifestation that the predictors for the risk of ESRD between genders were different. The reasons may be related to the genetic diversity between genders in kidney structure and function, receptor mediated influence of sex steroids on glomerular structure, as well as response on the synthesis and discharge of cytokines and growth factors [26, 27, 38]. The different modifiable risk factors for male and female may also suggest different approaches and therapies for the prevention of ESRD. Meanwhile, a meta-analysis involving more than 1.5 million patients in different countries from 45 cohorts conducted by the Kidney Disease, Improving Global Outcomes (KDIGO) showed that urine ACR and eGFR were independent risk factors for both progression of chronic kidney disease and ESRD [39]. Through the stringent evaluation with an almost inexhaustible list of clinical variables, our finding confirmed that the effects of urine ACR and eGFR were not diminished by other variables. In addition, the results of the present study illustrated the interaction effect between age and renal impairment on the incidence of ESRD, which indicated that the impacts of urinary albumin and kidney function were diluted by older age. This phenomenon may be understandable because the severity of renal impairment in elderly may be certainly high, and thus there is a relatively small room to decline as a result of lower changes in urine ACR and eGFR among elder patients compared to younger patients. An aforementioned observational study conducted in the United States also displayed that there is a continuous trend with a slow progressive eGFR decline observed among the general population after age of 40 [40]. Other predictors in our model were well discussed in the literatures [26, 27, 41]. A further study should be warranted to confirm the interaction effect on the incidence of ESRD between age and renal impairments, and the curvilinear relationship between risk of ESRD and the predictors including HbA1c and BP.

Our newly developed gender-specific models provide a more accurate and valid 5-year ESRD risk predictions for Chinese diabetic primary care patients than other currently existing models. We confirmed the importance of eGFR and urine ACR in predicting ESRD and found the renal impairments interacted with age on the risk of ESRD for Chinese diabetic patients. Our model could identify diabetic patients who are at high risk of ESRD and included several modifiable risk factors such as smoking and blood pressure in order to counsel patients, allow target delivery of proper healthcare and facilitate service policy planning.