The study population were participants aged 40–69 years taking part in two population-based surveys, the Know Your Heart (KYH) study, Russia (2015–2018), and the seventh wave of the Tromsø Study (Tromsø7), Norway (2015–2016). These studies were conducted in parallel as part of the Heart to Heart project aimed at understanding the reasons for Russia having much higher rates of CVD mortality than Norway. Several aspects of data collection between the studies have been harmonized (including calibration of laboratory tests of blood samples) [20] providing a unique opportunity to compare levels of CKD in the general population of both countries.
Know your heart (Russia)
KYH is a cross-sectional study including 4607 men and women aged 35–69 years recruited from the general population in the Russian cities of Arkhangelsk and Novosibirsk, described in detail previously [21]. The response rate in the study was higher for the city of Arkhangelsk (43.9% of participants 35–69 out of all addresses sampled completed a baseline interview compared to 20.9% from Novosibirsk) [21].
In brief, a random sample of addresses where a person aged 35–69 years was living, stratified by age, sex and district was selected from a population register. Addresses were visited by trained interviewers who recruited participants to take part in the study. Participants who agreed to take part completed an interview about their health (stage 1) and were invited to attend a health check at a polyclinic (stage 2). At stage 2 all participants were asked to provide a blood sample and a spot urine sample. Participants had the choice to provide a urine sample during the health check or to do this at home and return it to the clinic. Participation at each stage of the study is shown in Fig. 1a.
Blood samples were collected in SST II vacutainers BD® (Beckton, Dickinson and Company, Preanalytical Systems, US). Samples were left at room temperature for 30 min and then stored at 4 °C before processing. Samples were centrifuged in cooled centrifuges at 4 °C at 2100-2200 g for 15 min. Urine sample were collected using Beckton Dickinson urine collection pots. If participants did this at home they were instructed to store the sample at 4 °C and return to the clinic within 18 h in order to meet target of freezing samples within 24 h. Blood samples were frozen within 2 h and urine samples within 24 h at -20 °C, and transferred to a − 80 °C freezer within 3 weeks. All samples were analysed centrally in one batch at the end of the study at the same central laboratory in Moscow. Serum creatinine was measured using an uncompensated kinetic Jaffe reaction on a Beckman Coulter machine. Serum cystatin C was assessed with an enzymatic method using Particle enhanced Immunoturbidimetric test. Urinary albumin was measured using Immuno-turbidimetric test and urine creatinine was traceable to the IDMS method using a compensated kinetic Jaffe reaction on a Beckman Coulter machine. Coefficient of analytical variation (CV) For serum creatinine CV was < 7%, for serum cystatin C CV was < 4%, for urinary albumin CV was < 3% and for urinary creatinine CV was < 9%.
Tromsø7 (Norway)
The Tromsø Study [22] is an ongoing population-based study of residents of the Norwegian municipality Tromsø. In Tromsø7, all residents aged 40 years and older were invited, of which 21,083 men and women participated (65%), and 17,646 were aged 40–69 years.
All participants were invited to a basic examination including questionnaires, biological sampling (including serum creatinine) and medical examinations. A random subset was invited to an extended clinical examination with additional biological sampling (including serum cystatin C and urine samples). Participation at each stage of the study is shown in Fig. 1b.
Blood samples were collected in SST tubes and left for 30 min at room temperature and then centrifuged within 1 h for 10 min at 2000 g. Analyses were done within the same day as sample collection at the University Hospital of North Northway laboratory that is accredited according to the ISO 15189 standard.
Urine samples were first morning void samples taken on three consecutive days. Participants returned the samples when they attended the second part of the examination and analyses were done on the same. For comparison with KYH the first day sample was used.
Serum creatinine was measured using enzymatic colorimetric method traceable to isotope dilution mass spectrometry, and serum cystatin C using the immunoturbimetric method. Urinary albumin was measured using colorimetric method and urinary creatinine using the enzymatic colorimetric method. All analyses were done using Cobas 8000 Roche devices. For serum creatinine CV was < 2%, for serum cystatin C CV was < 3%, for urinary albumin CV was < 5% and for urinary creatinine CV was < 2%.
Calibration of laboratory analyses
A calibration study of laboratory analyses on blood samples between the studies was conducted [20]. Regression equations derived from this study (Calibration plots Supplementary Fig. 1a and b) were used to correct for differences in methodology between the laboratory by adjusting the KYH serum creatinine measures by − 29.42 + 1.21*(original KYH measurement in μmol/L) and KYH cystatin c measures by 0.06 + 0.98*(original KYH measure in mg/L).
Outcome variables
Serum creatinine was used to calculate eGFR using the creatinine based CKD-EPI eGFR formula to estimate renal function [23]. Serum creatinine alone was used as the primary measure of eGFR as this was available for the full subset of participants at Tromsø7.
A composite outcome labelled CKD was defined as reduced eGFR (< 60 ml/min/1.73 m2) and/or elevated albuminuria defined as ≥30 mg/g albumin/creatinine in urine. Repeat measures were not available in this study therefore the CKD outcome was based on single measures of eGFR and urinary albumin and creatine.
Differences in prevalence of reduced eGFR using serum creatinine were compared with those from using combined creatine and cystatin C in sensitivity analyses defined using the CKD-EPI combined creatinine-cystatin C equation [23].
Risk factors
Medication classes were defined using the Anatomical Therapeutic Chemical (ATC) classification system [24]. For antihypertensive medication use, we combined answering “yes” to the questions “Do you use blood pressure lowering drugs and information from a self-reported list of brand names of regularly used medication as antihypertensives (ATC codes C02, C03, C07, C08 and C09). Hypertension was defined as ≥140 mmHg systolic and/or ≥90 mmHg diastolic at examination (mean of 2nd and 3rd measurement) or use of antihypertensives (ATC codes C02, C03, C07, C08 or C09 and/or yes to the question “Do you use of blood pressure lowering drugs?”). Hypertension was classified as a) uncontrolled untreated hypertension (raised blood pressure at examination/no use of antihypertensives); b) treated uncontrolled hypertension (raised blood pressure at examination/use of antihypertensives); c) controlled and treated (raised blood pressure at examination /use of antihypertensives) and d) normotensive (no raised blood pressure at examination/no use of antihypertensives). Diabetes was defined as HbA1c ≥ 6.5% [20] and/or self-report of diagnosis and/or self-reported use of diabetes medication (ATC codes A10B, A10A).
Information on education (lower, middle and higher coded based on education system in each country) and smoking status was collected from questionnaires. Education was defined as lower (incomplete secondary and vocational no secondary), middle (complete secondary, vocational and secondary, specialised secondary) and higher (incomplete higher, higher) education for KYH and lower (primary) middle (upper secondary) and higher (university/university college) for Tromsø7. Body mass index (BMI) was calculated from measured weight and height (kilograms/meter2 [kg/m2]). Abdominal obesity was defined as waist:hip ratio > 0.9. Due to differences in measurement protocols for waist circumference between the two studies (minimal waist for KYH, level of umbilicus for Tromsø7) waist circumference from Tromsø7 was converted to minimal waist using an established equation for the conversion [25]. All measurements were conducted by trained personnel.
Statistical methods
Objective 1: comparison of prevalence of CKD between the two studies
The age- and sex- standardised prevalence of CKD, reduced eGFR and albuminuria were calculated with direct standardisation to the European 2013 standard population using 5 year age bands. Between-study differences in prevalence were investigated by fitting logistic regression models for each outcome with study as the exposure adjusted for age and sex. Models with and without interaction terms between study and a) age and b) sex were compared using likelihood ratio tests.
Objective 2: comparison of association with risk factors between studies
Risk factors for reduced eGFR and albuminuria were investigated for each study. Separate logistic regression analyses were run for reduced eGFR and albuminuria. In both models associations with age, sex, reduced eGFR (in the model for albuminuria) or albuminuria (in the model for reduced eGFR), education, smoking, BMI, waist: hip ratio, diabetes and hypertension were investigated. Models were adjusted 1) for age and sex (and city for KYH) and 2) mutually adjusted for all risk factors.
Objective 3: extent to which between-study differences in CKD were explained by established risk factors
The extent to which between-study differences in reduced eGFR and albuminuria were explained by established risk factors were investigated by fitting separate logistic regression models for each outcome with study as the main exposure and then adjusting systematically for risk factor variables one at a time and then all together. Here systolic and diastolic blood pressure and use of antihypertensive medication were adjusted for separately to disentangle the effects of the components of high blood pressure compared to possible effects from taking medications given medication use is related both to clinical indication and other wider health systems factors such as levels of awareness and treatment practices.
Accounting for missing data
Objective 1
For KYH study missing data in the outcome was investigated by calculating inverse probability weights from age, age squared and sex taken from the sampling frame according to participation in the first stage separately for each city and then using a multiple imputation model [26] to impute missing outcomes using the factors associated with missingness at stages 1,2 and 3 shown in Fig. 1a. For Tromsø7 a multiple imputation model was used for outcomes which were measured only at the second visit (cystatin C, albuminuria). The predictors of missingness used were sex, age, age square, education (Fig. 1b) and reduced eGFR measured using serum creatinine. Individual level data on predictors of attendance at the first study visit were not available so this was not accounted for in the analysis. For both studies eGFR was used a priori in imputation models for albuminuria and CKD even if not associated with missingness in the outcomes.
Prevalence estimates for CKD overall (albuminuria and/or reduced eGFR) as well as prevalence estimates for reduced eGFR (using creatinine only and joint creatinine –cystatin C) and albuminuria separately were computed for both studies by sex and 10-year age bands. Analyses using survey weights were conducted using the survey-set suite in Stata, by reweighting the sample using the derived probability weights for each using robust variance estimators.
Objective 2
Chained multiple imputation [27] was used to take account of missingness in both the outcome and in the risk factors. Separate imputation models were used for the two outcomes and studies. For KYH data on reduced eGFR were imputed for all participants using both the risk factors in the model and predictors of missingness from stage 1 (Fig. 1a). For albuminuria this model would not converge therefore a simpler model using low eGFR and the measured risk factors (but not the additional predictors of missingness) from stage 1 was used to impute data for all participants who attended stage 2 of the study. For Tromsø7 models for both outcomes were imputed from measured risk factors for participants attending the basic examination of the study. The model for albuminuria additionally included low eGFR.
Objective 3
Chained multiple imputation (MI) models including both outcomes and all risk factors in the models were used for main analyses. These models were fitted separately for each study using dataset of all participants attending the health check for KYH and all participants attending the basic examination in Tromsø7 to create a consistent dataset for reduced eGFR and albuminuria. The MI datasets were then combined to conduct the analyses.
For all three objectives sensitivity analyses were conducted comparing findings with complete case analysis.
All analyses were conducted using Stata 16 [28].
The STROBE (Strengthening the reporting of observational studies in epidemiology) guidelines for the reporting of cross-sectional studies were followed in reporting the study findings.