Background
Chronic kidney disease (CKD) and decreased glomerular filtration rate (GFR) are both associated with elevated risks for end-stage renal disease (ESRD), cardiovascular disease, and death [
1,
2]. The incidence of CKD has increased worldwide with important public health and economic implications especially in developing countries where resources are limited. Despite the high prevalence rate, CKD awareness rates are often very low in the community as early CKD is usually asymptomatic [
3,
4]. As a consequence, CKD is not frequently detected until it has already advanced, and opportunities for intervention are lost. A risk score that identifies those at higher risk for future CKD has been proposed as a stratification and prediction device [
5]. Cardiovascular risk scores, such as the Framingham score, [
6] have influenced public health policy in the primary prevention of cardiovascular diseases and have been tested in many populations [
7]. A simple and accurate renal risk score would lead to targeted medical management at the primary care level to individuals at the highest risk for future CKD.
Asians represent 60% of the global population and have among the highest prevalence of CKD in the world [
8]. Asian countries include developed and low to middle-income countries with different risk factors for CKD development [
9,
10]. In low to middle-income countries, the burden of disease is changing from infections towards chronic lifestyle-related diseases as a result of demographic transition and urbanization. In 2011, Thailand was reclassified by the World Bank from a lower-middle income to a higher-middle-income country. Over the last decade, the numbers of patients with ESRD have increased by an order of magnitude. Data from the Thailand Renal Replacement Therapy registry reported that the numbers of patients on renal replacement increased from 30 per million people in 1997 to 1199 per million people in 2014 (file:///E:/TRT%202007–2017/1.TRT-report−2014-_3–11-59_-final_pdf). This staggering increase is both due to public health policy changes as well as due to higher rates of CKD.
Risk scores for incident CKD have been developed in the US general populations [
11,
12], but there are limited information on risk models that allow long term predictions of incident CKD in Asians. Since risks factors for CKD and decreased GFR in populations from Asian countries including Thailand may differ from Caucasian populations, risk scores developed in an Asian community may be more appropriate to evaluate risks in Asian populations. In this study, we aimed to evaluate risk predictors and develop risk models and risk scores for developing decreased GFR at 10 years follow-up in a Thai general population cohort. We hypothesized that new cases of decreased GFR may be predicted by a risk score containing a subset of clinical variables. Because subjects from low to middle-income Asian countries often have limited resource and less access to routine medical checkups than their Western counterparts, we have developed a risk score based on clinical and simple laboratory parameters easily assessed in the primary care setting and compared them to a score based on more expanded, but still standard laboratory work up.
Discussion
We developed risk prediction models for developing decreased eGFR at 10 years in a middle-age to older Thai general population using standard clinical parameters and routine laboratory tests. The predictors for the clinical model were: age, sex, systolic blood pressure, waist circumference or body mass index, and history of diabetes. The predictors for clinical and limited laboratory tests comprised of age, sex, systolic blood pressure, diabetes mellitus and baseline eGFR. The risk models demonstrated good discrimination and calibration with good internal validation. The addition of more laboratory tests of hemoglogin concentration, uric acid, HDL did not increase the performance of the clinical and limited laboratory tests significantly. Based on these results, we developed 2 simplified risk scores: a clinical risk score and a combined clinical and limited laboratory risk score. External validation using a separate cohort confirmed good performances of these scores. The parameters used in the scores are readily available for self–testing or evaluation by medical personnel in the primary care settings appropriate for a resource-limited setting such as Thailand or other parts of Asia.
Improved clinical prediction is an essential component of personalized medicine. Clinical prediction tools such as the Framingham cardiovascular risk score [
6] have helped shape public health policy in the primary prevention of cardiovascular disease in many countries. However, despite the identification of several key renal risk factors, [
5] similarly useful risk scores for predicting long term risk of new CKD has not been developed in an Asian general population. We are aware of 3 prior published risk prediction scores for incident CKD from community-based cohorts. The first study was derived and validated using data from middle-aged and older adults in the US community [
12]. From this study, the final model included 8 variables: age, sex, anemia, hypertension, diabetes mellitus, cardiovascular disease, history of heart failure, and peripheral vascular disease. This risk score had moderate discriminatory power (c-statistic 0.70) and did not contain data on baseline GFR or proteinuria. The second study evaluated Taiwanese subjects, but was compromised by poor discriminatory power (c-statistic 0.67) and short follow-up (median 2.2 years) [
28]. Because the follow-up of this study was very short, only those subjects with very rapid decline in GFR would be detected and the cumulative effects of risk factors on CKD development would be underestimated. The most recent risk score was derived from Caucasian subjects from the Framingham cohort [
11]. This study shared several elements to our study including a similar follow-up period of 10 years with similar, but not identical risk predictors. In the Framingham study, age, diabetes and hypertension were significant predictors for the clinical score, and age diabetes hypertension, GFR, proteinuria were predictors in the combined clinical and laboratory model. Both the clinical and the combined clinical and laboratory tests had a high degree of accuracy and discriminatory power in US Caucasian or Black subjects (AUC 0.78–0.83). However, the Framingham risk score had low discriminatory power and accuracy when tested in our cohort (χ2 = 256.5,
p < 0.001 and AUC 0.63 for model 2). Compared to our scoring system, age was the most significant contributor to the combined risk score in the Framingham study with diabetes, and hypertension only contributing in a minor role. In our score, both diabetes, hypertension and baseline eGFR were more important contributors to the score. In addition, being overweight was an important clinical predictor of CKD whereas this was not included in the Framingham score. The importance of obesity and diabetes highlighted in our score is especially striking given the rise in obesity and diabetes across low to middle-income populations in Asia with increasing globalization [
8,
10].
Previous prediction scores for incident CKD were derived using the modification of diet in renal disease study (MDRD) equation [
21]. The MDRD formula was first developed in US patients with established CKD (6). CKD-EPI equation, which was derived from both CKD and normal subject cohorts has been shown to be more accurate than MDRD especially in subjects with preserved GFR [
20] Furthermore, CKD-EPI has been shown to be superior to MDRD at predicting adverse outcomes and improved the accuracy in outcome prediction in Caucasian and Black US subjects [
29] There is considerable controversy on the optimum eGFR equation in Asian populations [
30,
31]. There are as much as 20–30 ml/min/1.73m
2 differences in GFR estimates between various Asian formulae. These discrepancies results in as much as 10 fold variations in CKD prevalence rate, and alter the prognostic significance attributable to the presence of CKD [
32]. Differences in the reference GFR methods, and the proportion of non-CKD subjects in the development cohort likely account for these discrepancies as much as any biological differences between Asian subjects of various ethnicities [
30,
31]. A Thai eGFR equation has been developed with Thai patients with established CKD [
33] using a short plasma clearance of 99
Tc DTPA as the GFR measurement method. Given the lack of inclusion of normal subjects in the development cohort and methodological issues used to develop the Thai eGFR equation [
33], we elected to use CKD-EPI for the sake of generalizability of the score to other Asian cohorts and for comparisons with other global populations [
34,
35]. The rationale of our choice is supported by the fact that the CKD-EPI equation-based CKD staging has been shown to result in similar risk predictions for adverse outcomes in Asians, Whites and Blacks in a large meta-analysis [
36]. In addition, we also tested the performance of our risk scores using the Asian coefficient of the four-level race variable CKD-EPI equation to calculate eGFR. Although this equation was developed in Asian populations, its role remain uncertain as the accuracy can vary in different Asian populations [
26] Changing to the CKD-EPI Asian equation resulted in lower incident cases with
Deceased GFR, but the performance of risk models were largely similar to using the original CKD-EPI equation. This suggests that the risk scores may be used to predict
Decreased GFR when the Asian coefficient for CKD-EPI is used to calculate eGFR but the performance might be slightly reduced.
There are several potential implications of this work. First, by allowing physicians to determine an individual’s estimated risk for
Decreased GFR, the score may inform clinical decision-making, for example to modify treatment, frequency of follow-up or institute renal primary prevention measures in high risk patients. Secondly, the use of the score may raise the profile of kidney disease among the general population, a key goal as the current CKD awareness rates is only 1.9% in Thailand [
4]. Thirdly, it is noteworthy that the discrimination of the clinical risk score is already fairly high and this score could be used for focused renal screening, identifying individuals in whom creatinine measurement would be most cost-effective. Of note, although proteinuria tended to improve the discrimination of the clinical model, the improvement was slight and not statistically significant. Given the increased cost, our study suggests that routine population-based dipstick testing proteinuria may not be worthwhile. Of course, the risk score should not be used as a substitute for established urinalysis-screening intervals in people with diabetes or have other high risks. Finally, our score may be useful in estimating the individual risk and future prevalence of CKD in middle age to older subjects from other Asian general population [
8,
10]. The risk factors used in our score such as older age, diabetes and hypertension are universal risk factors for CKD [
10]. Many low to middle-income Asian countries are exposed to similar health impact of globalization as Thailand, and share similar prevalence of many CKD risk factors that may be considerably different from the West [
37]. Combined with a closer genetic background, risk scores developed in one Asian population may be more accurate at predicting CKD in another Asian population. Although our scores have been validated externally using a separate cohort, the cohort used for validation consisted of younger Thai subjects from a similar employment background as the Derivation cohort. Participants also had shorter follow-up period of five years and a lower incidence of
Decreased GFR. Further studies in diverse Asian cohorts with longer follow-up duration are necessary to confirm the usefulness of our score in predicting the long term risk of CKD in other Asian populations.
There are several strengths to this study. To our knowledge, this is the first prospective risk score to predict incident cases of
Decreased GFR with follow-up of up to 10 years in an Asian population. We employed a community-based cohort with detailed assessment of risk factors and standardized calibrated enzymatic creatinine measurements. Several limitations also should be acknowledged. Baseline and follow-up creatinine were measured on a single occasion. According to KDIGO guidelines, the diagnosis of CKD requires two estimates of GFR separated by 3 months [
14]. As such, the outcome we evaluated in this study does not fulfill the criteria of incident CKD, but rather, the outcome represented incident cases with decreased eGFR (eGFR < 60). This is a limitation our study shares with most published studies involved in developing incident CKD score including the Framingham heart study [
11]. Multiple measurements in cohort studies are costly to perform especially in a resource-limited setting such as ours. By measuring the follow-up eGFR only once, we cannot exclude the fact that some subjects may have reversible acute kidney injury rather than persistent CKD. In addition, some subjects with a rather low borderline eGFR may have a follow-up eGFR slightly lower than 60 just due to random variation of serum creatinine. In practice, the EGAT subjects who attended the follow-up examination were not acutely ill and significant acute kidney injury was probably not frequent. Exclusion of subjects with less than 5 ml/min/1.73 m2 change in eGFR in our sensitivity analysis did not alter the performance of the risk score. Thus we expect that the risk factors identified in our study and our risk score should be valuable in identifying subjects at risk of developing incident CKD in the general population. Nonetheless, a single measurement of eGFR may lead to an overestimation of incident CKD. Future studies with repeated creatinine measurements that can confirm the presence of decreased eGFR after 3 months should provide a more accurate prediction of risks of developing incident CKD, although such a study would be more expensive to perform.
We used dipstick proteinuria rather than urine albumin creatinine ratio
. It is possible that quantitative proteinuria measurement might have provided better prediction for
Decreased GFR and improved our prediction models. The aim of this study was to devise scores to screen subjects at risk of
Decreased GFR in a resource-limited setting and urine albumin creatinine ratio was not performed at baseline because of the higher expense. The KDIGO 2012 guidelines [
14] suggested that urine dipstick might be substituted for albumin creatinine ratio when the latter is not available. In other scores e.g. Framingham, substitution of urine dipstick for urine albumin creatinine ratio did not alter the results [
11].
A number of the participants who attended the visit in 2002–2003 did not attend the follow-up visit in 2012–13. It is not surprising that the health risk profile of those who attended both visits were statistically better than the total study population at baseline since death or retirement were common reasons for non-attendance. Although the differences in these risk factors were statistically significant, the actual differences were clinically quite small for most variables. Nonetheless, it is possible that the subjects who did not attend the follow-up examination were sicker and the true incidence of
Decreased GFR might have been underestimated. Although the EGAT cohort is a community-based cohort, there may be some differences in the participant profiles from the Thai population as a whole. All participants were Thais and Thai-Chinese who represent the vast majority (over 95%) of the Thai Census population. Our study included only middle-age to older subjects and had a higher percentage of males than females compared to the total Thai population. EGAT employees come from all regions of Thailand and cover a wide-range of sociodemographic backgrounds [
13]. Nonetheless, the socio-economic status of EGAT employees is probably better than some of the most severely economically disadvantaged Thais, and the study did not include the severely ill or disabled subjects excluded from employment. The prevalence of subjects with decreased eGFR and the CKD risk factors in our study are comparable to other a representative cross-sectional population surveys from Thailand [
4]. Therefore, our risk score should applicable in assessing the risks of developing decreased GFR in community-based Thai subjects, although caution may be necessary in extrapolating findings to groups not represented in our study (for example younger or very old subjects or those who are institutionalized).