Prevalence of and risk factors for chronic kidney disease of unknown aetiology in India: secondary data analysis of three population-based cross-sectional studies

Objectives To assess whether chronic kidney disease of unknown aetiology (CKDu) is present in India and to identify risk factors for it using population-based data and standardised methods. Design Secondary data analysis of three population-based cross-sectional studies conducted between 2010 and 2014. Setting Urban and rural areas of Northern India (states of Delhi and Haryana) and Southern India (states of Tamil Nadu and Andhra Pradesh). Participants 12 500 individuals without diabetes, hypertension or heavy proteinuria. Outcome measures Mean estimated glomerular filtration rate (eGFR) and prevalence of eGFR below 60 mL/min per 1.73 m2 (eGFR <60) in individuals without diabetes, hypertension or heavy proteinuria (proxy definition of CKDu). Results The mean eGFR was 105.0±17.8 mL/min per 1.73 m2. The prevalence of eGFR <60 was 1.6% (95% CI=1.4 to 1.7), but this figure varied markedly between areas, being highest in rural areas of Southern Indian (4.8% (3.8 to 5.8)). In Northern India, older age was the only risk factor associated with lower mean eGFR and eGFR <60 (regression coefficient (95% CI)=−0.94 (0.97 to 0.91); OR (95% CI)=1.10 (1.08 to 1.11)). In Southern India, risk factors for lower mean eGFR and eGFR <60, respectively, were residence in a rural area (−7.78 (-8.69 to –6.86); 4.95 (2.61 to 9.39)), older age (−0.90 (–0.93 to –0.86); 1.06 (1.04 to 1.08)) and less education (−0.94 (-1.32 to –0.56); 0.67 (0.50 to 0.90) for each 5 years at school). Conclusions CKDu is present in India and is not confined to Central America and Sri Lanka. Identified risk factors are consistent with risk factors previously reported for CKDu in Central America and Sri Lanka.


127
HbA1c using high-performance liquid chromatography, and the serum creatinine using the rate-blanked and 128 compensated kinetic Jaffe method, traceable to isotope dilution mass spectrometry (Nair et al. 2012).

130
We reported mean eGFR and prevalence eGFR<60 according to different characteristics of the study 131 populations. UDAY and CARRS studies did not involve fully random population samples (since sampling was 132 based on households, with one participant per household) and the proportions of study participants with 133 particular outcomes (e.g. eGFR<60), will not be exactly the same (but very similar) to what would have been 134 obtained with genuine random population samples; thus in this paper we refer to the prevalence in the study 135 participants, not overall population prevalence estimates. We used linear regression models to estimate the 136 associations between potential risk factors and eGFR and logistic regression models to estimate the associations 137 between potential risk factors and eGFR<60. We also repeated the analyses separately for males and females.

138
Variables associated with eGFR in the basic analyses (adjusted for age and sex) were considered for the multiple 139 regression analysis. In the final multiple regression model, we included all variables that were of a priori interest 140 and/or had shown independent associations with eGFR. We then checked for multicollinearity for each variable 141 in the multiple regression analyses in comparison with the basic analyses (Greenland et al. 2016). 6% of 142 participants had missing values for education, 4% for BMI and 11% for fat free mass. For BMI and fat free mass, 143 we excluded participants with missing values to compare models non-adjusted and adjusted for these variables.

177
Risk factors for lower eGFR and eGFR<60 178 As expected, age was an important risk factor for reduced eGFR: eGFR was 0.93 ml/min per 1.73 m 2 (95%CI=-179 0.95 --0.91, model adjusted for sex) lower for each additional year of age. Additionally, being male, living in a 180 rural setting, living in Southern India and consuming alcohol were associated with decreased mean eGFR (Table   181 3). Similarly, the odds of eGFR<60 also increased by each year of age [OR adjusted for sex (95%CI)=1.1 (1.1 -182 1.1)] and being male, living in a rural setting, living in Southern India and consuming alcohol were also 183 associated with eGFR<60 (Table 3). Risk factors for decreased mean eGFR and for eGFR<60 were similar for 184 men and women (supplementary material, Table S2).

185
In the multiple regression analyses, decreased mean eGFR remained associated with older age, being male and 186 living in a rural setting and alcohol consumption ( Table 4). Risk of eGFR<60 remained associated with older 187 age, being male and living in a rural setting and having no formal education remained associated with increased 188 risk of eGFR<60 (Table 4). We adjusted all the multiple regression models for fat free mass and vegetarianism 189 to assess the possibility that differences observed between urban and rural participants were due to differences in 190 diet and/or body composition. These adjustments had little effect on the results (Table 4). We performed a sensitivity analysis including those with ACR>300 (but without hypertension or diabetes, n=33) 203 as we were concerned that those with CKDu might develop proteinuria at more advanced CKD stages. However, 204 this did not alter the mean eGFR (mean eGFR among the overall study population=105.0±17.8, mean eGFR in 205 this sensitivity analysis =105.0±17.8), nor the estimated prevalence of eGFR<60 (prevalence among the overall 206 study population=1.6%; prevalence in this sensitivity analysis =1.7%). The findings on risk factors were also 207 similar to the findings from the primary analyses (supplementary material, Table S3).

208
Given concerns about potentially different thresholds to define diabetes and high blood pressure in different  Table S4). HbA1c was associated with eGFR<60 in this non 215 diabetic population but inclusion of this variable did not alter the OR for other risk factors observed in the 216 primary analysis (supplementary material, Table S4). Therefore, although the relationship between sub-clinical 217 diabetes and impaired kidney function requires further prospective investigation, there is no evidence that the 218 excess risk of low eGFR (i.e. lower mean eGFR and higher prevalence of eGFR<60) in rural Southern India is 219 associated with either impaired fasting glucose or higher blood pressure.

221
We report the distribution of eGFR in people without diabetes, hypertension or heavy proteinuria and estimate 222 the prevalence of CKDu in our study population, including participants from urban and rural settings. We found 223 that the rural population from Southern India (Vishakhapatnam district) had the highest risk of low eGFR (lower 224 mean eGFR and higher prevalence of eGFR<60). In Southern India, rural residence, older age and lower 225 education were risk factors for decreased eGFR, and there was also some evidence for higher risks in males. In 226 Northern India, older age was the only risk factor for low eGFR. This is the first population-based evidence, 227 using standardised methods, which indicates that CKDu is present in India and is not confined to Central As in Central America, the risk of low eGFR was higher in rural settings than in urban settings. This is in 230 concordance with a previous study from Hyderabad (India), that has provided evidence of a higher risk of low 231 eGFR in a rural population compared to urban-migrant and urban population (Bailey et al. 2013  India, and potentially explain the differences observed between these areas. The associations between urban/rural 237 residence and lower mean eGFR was much more marked in Southern India than in Northern India, and the 238 associations between urban/rural residence and eGFR<60 was only observed in Southern India. The higher 239 prevalence ratio (for eGFR<60) in the working age population compared to older age groups is consistent with 240 the hypothesis that deceased in eGFR could be potentially explained by occupational exposures. The suggestive 241 sex differences may also support this hypothesis. However, we did not have detailed data on occupation that 242 allowed us to explore these associations in greater detail.

243
The higher risk of low eGFR in Southern India (Chennai and Vishakhapatnam districts) observed in our study is

270
To estimate the actual prevalence of reduced eGFR, future studies should include validated methods to estimate 271 GFR in the Indian population. We were concerned that the validity of CKD-EPI among the Indian population 272 may be also compromised by differences in muscular mass and meat consumption between population groups 273 within India. We adjusted the analyses for fat free mass and vegetarianism, but this did not alter the results, 274 suggesting no confounding effect by these variables.

78
Existing reports indicate that CKDu may be common but it is difficult to be definite about this because of the 79 absence of population-based studies using standardised and comparable methods. Data from the Indian CKD

86
We conducted a secondary analysis of representative sample surveys conducted in India between 2010-2014.

87
Given the absence of a clear case definition for CKDu it is necessary to make a presumptive diagnosis based on 88 measures/estimates of GFR in the absence of known risk factors for kidney disease. The overall aim of the 89 current study was to use a methodology which is comparable to previous studies elsewhere in the world 90 (particularly in Central America) to assess the extent to which reduced kidney function is a problem in India, and 91 which areas and subpopulations are most affected. We therefore: (i) assessed the distribution eGFR and 92 prevalence of eGFR below 60ml/min per 1.73m 2 (eGFR<60) in Indian populations restricted to those without 93 known risk factors for CKD, i.e. diabetes, hypertension or heavy proteinuria; ii) compared these outcomes in 94 obtained with genuine random population samples; thus in this paper we refer to the prevalence in the study 154 participants, not overall population prevalence estimates. We used linear regression models to estimate the 155 associations between potential risk factors and eGFR and logistic regression models to estimate the associations 156 between potential risk factors and eGFR<60. We also repeated the analyses separately for males and females.

157
Variables associated with eGFR in the basic analyses (adjusted for age and sex) were considered for the multiple 158 regression analysis. In the final multiple regression model, we included all variables that were of a priori interest 159 and/or had shown independent associations with eGFR. We then checked for multicollinearity for each variable

174
Patients were not involved in the design of this analysis.

176
Characteristics of study participants 177 12,500 people were eligible for the current analyses ( Figure 2).   208 As expected, age was an important risk factor for reduced eGFR: eGFR was 9.30 ml/min per 1.73 m 2 (95%CI=-209 9.51, -9.09, model adjusted for sex) lower for each additional 10 years of age. Additionally, being male, living in 210 a rural setting, and consuming alcohol were associated with decreased mean eGFR (Table 3). Similarly, the odds 211 of eGFR<60 also increased with age [OR per 10 years, adjusted for sex (95%CI)=2.34 (2.12, 2.59)] and being 212 male, living in a rural setting, living in Southern India and consuming alcohol were also associated with 213 eGFR<60 (Table 3). In general, risk factors for decreased mean eGFR and for eGFR<60 were similar for men 214 and women (supplementary material, Table S2), but few differences were observed. Regarding mean eGFR, 215 living in Southern India was associated with decreased mean eGFR in men and with increased mean eGFR in 216 women; tobacco consumption was associated with increased mean eGFR in men and with decreased mean eGFR 217 in women; vegetarianism was associated with decreased mean eGFR in women but not in men; and being 218 overweight was associated with decreased mean eGFR but in men but not in women. Regarding risk of 219 eGFR<60, living in Southern India was associated with increased risk of eGFR<60 in men but not in women.   (Table 4). We adjusted all the multiple regression models for fat free mass and vegetarianism 228 to assess the possibility that differences observed between urban and rural participants were due to differences in 229 diet and/or body composition. These adjustments had little effect on the results (Table 4).

238
However, this decrease was much more marked in Southern India. In Northern India, rural residence,

239
formal education (and duration) and age were the only other risk factor associated with reduced eGFR. In 240 Southern India, being male was also a risk factor for reduced eGFR, whereas formal education was only a 241 risk factor for reduced eGFR among those with more than 10 years of schooling (Table 5). We also 242 observed an interaction between the effects of latitude (North/South) and urban/rural residence in 243 association with eGFR<60 (p-value likelihood-ratio test for interaction<0.001). In Northern India, 244 eGFR<60 was not associated with urban/rural residence, and older age was the only factor associated 245 with eGFR<60. In Southern India, rural residence was the strongest risk factor for eGFR<60 but older age 246 and lower years of formal education also increased the risk of eGFR<60 (

287
As in Central America, the risk of low eGFR was higher in rural settings than in urban settings. This is in 288 concordance with a previous study from Hyderabad (India), that has provided evidence of a higher risk of 299 support this hypothesis. However, we did not have detailed data on occupation that allowed us to explore 300 these associations in greater detail.

301
The  Indian population. We were concerned that the validity of CKD-EPI among the Indian population may be 328 also compromised by differences in muscular mass and meat consumption between population groups 329 within India. We adjusted the analyses for fat free mass and vegetarianism, but this did not alter the 330 results, suggesting no confounding effect by these variables.

331
Our study has at least three potential limitations. First, we only had one measure of eGFR, and therefore 332 we could not differentiate acute kidney injury (AKI) from CKD. This is a common limitation in 333 epidemiological studies, as it is challenging to obtain more than one measure of eGFR at least 3 months 334 apart in large population-based investigations. Therefore, we may have misclassified some cases of AKI 335 as reduced eGFR, and therefore overestimate the prevalence of this condition. Nevertheless, there is no a 336 priori reason to think that potential misclassification was different according to the evaluated risks factors.

94
Study population

141
We reported mean eGFR and prevalence of eGFR<60 according to different characteristics of the study populations.

142
UDAY and CARRS studies did not involve fully random population samples (since sampling was based on 143 households, with one participant per household) and the proportions of study participants with particular outcomes 144 (e.g. eGFR<60), will not be exactly the same (but very similar) to what would have been obtained with genuine random 145 population samples; thus in this paper we refer to the prevalence in the study participants, not overall population 146 prevalence estimates. We used linear regression models to estimate the associations between potential risk factors and 147 eGFR and logistic regression models to estimate the associations between potential risk factors and eGFR<60. We also 148 repeated the analyses separately for males and females. Variables associated with eGFR in the basic analyses (adjusted 149 for age and sex) were considered for the multiple regression analysis. In the final multiple regression model, we 150 included all variables that were of a priori interest and/or had shown independent associations with eGFR. We then 151 checked for multicollinearity for each variable in the multiple regression analyses in comparison with the basic 152 analyses [27]. 6% of participants had missing values for basic co-variables (i.e. education) and were excluded from 153 the analysis. 5% and 9% of participants had missing values for BMI and for fat free mass respectively. These 154 participants were included in the main analysis, but we excluded them to compare models non-adjusted and adjusted 155 for these variables. We calculated prevalence ratios of eGFR<60 for rural versus urban areas in different age groups.

156
Urban areas were defined as "all places with a municipality, corporation, cantonment board or notified town area

201
-9.09, model adjusted for sex) lower for each additional 10 years of age. Additionally, being male, living in a rural 202 setting, and consuming alcohol were associated with decreased mean eGFR (Table 3). Similarly, the odds of eGFR<60 203 also increased with age [OR per 10 years, adjusted for sex (95%CI)=2.34 (2.12, 2.59)] and being male, living in a rural 204 setting, living in Southern India and consuming alcohol were also associated with eGFR<60 (Table 3). In general, risk 205 factors for decreased mean eGFR and for eGFR<60 were similar for men and women (supplementary material, Table   206 S2), but few differences were observed. Regarding mean eGFR, living in Southern India was associated with decreased 207 mean eGFR in men and with increased mean eGFR in women; tobacco consumption was associated with increased 208 mean eGFR in men and with decreased mean eGFR in women; vegetarianism was associated with decreased mean 209 eGFR in women but not in men; and being overweight was associated with decreased mean eGFR but in men but not 210 in women. Regarding risk of eGFR<60, living in Southern India was associated with increased risk of eGFR<60 in 211 men but not in women.  In the multiple regression analyses, decreased mean eGFR remained associated with older age, being male, living in a 217 rural setting, and alcohol consumption ( Table 4). Risk of eGFR<60 remained associated with older age, being male 218 and living in a rural setting, and having no formal education (Table 4). We adjusted all the multiple regression models 219 for fat free mass and vegetarianism to assess the possibility that differences observed between urban and rural 220 participants were due to differences in diet and/or body composition. These adjustments had little effect on the results 221 (Table 4).   We observed an interaction between the effects of latitude (North/South) and urban/rural residence in 228 association with reduced eGFR (p-value for interaction<0.001). The mean eGFR was lower in rural settings in 229 both Northern and Southern India (controlling for age, sex, education and alcohol intake). However, this 230 decrease was much more marked in Southern India. In Northern India, rural residence, formal education (and 231 duration) and age were the only other risk factor associated with reduced eGFR. In Southern India, being male 232 was also a risk factor for reduced eGFR, whereas formal education was only a risk factor for reduced eGFR 233 among those with more than 10 years of schooling (Table 5). We also observed an interaction between the 234 effects of latitude (North/South) and urban/rural residence in association with eGFR<60 (p-value likelihood-235 ratio test for interaction<0.001). In Northern India, eGFR<60 was not associated with urban/rural residence,

246
We performed a sensitivity analysis including those with ACR>300 (but without hypertension or diabetes, 247 n=33) as we were concerned that those with CKDu might develop proteinuria at more advanced CKD stages.

277
As in Central America, the risk of low eGFR was higher in rural settings than in urban settings. This is in 278 concordance with a previous study from Hyderabad (India), that has provided evidence of a higher risk of low  higher prevalence ratio (for eGFR<60) in the working age population compared to older age groups is consistent 287 with the hypothesis that deceased in eGFR could be potentially explained by occupational exposures. The

288
suggestive sex differences may also support this hypothesis. However, we did not have detailed data on 289 occupation that allowed us to explore these associations in greater detail.

315
population may be also compromised by differences in muscular mass and meat consumption between 316 population groups within India. We adjusted the analyses for fat free mass and vegetarianism, but this did not 317 alter the results, suggesting no confounding effect by these variables.