Study design and population
This was a nationally representative historical cohort study, not structured to determine the prevalence of CKD, and conducted by the Committee of Measures for Pediatric CKD of the Japanese Society for Pediatric Nephrology (JSPN). The first survey was sent in December 2017 to 399 general and local perinatal medical centers in Japan, inviting them to report pediatric patients over 3 years with VLBW who were managed as of April 1, 2017. The second survey was sent in June 2018 to 113 institutions that had participated in the first survey. The deadlines for returning the first and second surveys were March and September 2018, respectively, although the deadline for the second survey was postponed to February 2019.
The first questionnaire was designed to record the approximate number of patients per year with VLBW in each institution, and the willingness to complete the second questionnaire. Factors recorded for each patient during the perinatal or neonatal period included date of birth, sex, gestational age, birth weight and height, maternal age, maternal smoking, gravidity, maternal steroid administration, 5-min Apgar score, pH, respiratory distress syndrome, artificial breathing management period, patent ductus arteriosus, doses of indomethacin, late-onset circulatory collapse treated with corticosteroids, kidney dysfunction (serum creatinine ≥ 1.5 mg/dL) in neonatal period, duration of treatment with antibiotics and diuretics, and duration of enteral feeding. Factors recorded on the last day of consultation in children aged > 3 years included height, weight, blood pressure, proteinuria, serum creatinine (SCr) concentration, serum cystatin C concentration, congenital anomalies of the kidney and urinary tract, confirmed kidney disease, malformation syndrome, congenital heart disease and its severity, digestion or absorption disorder, chromosomal abnormalities, neuromuscular disease, and thyroid disease. For the purposes of this survey, height and weight were recorded within 3 months, and blood pressure and proteinuria within 1 year, after or before measurement of SCr. Age was calculated from the date of birth and the date of last encounter. The respondents were asked to search their medical records for patients with VLBW and with measurements of SCr and height over 3 years. Children with VLBW were included if they were aged over 3 years on April 1, 2017, and if height was measured within 3 months after SCr. If a patient was assessed more than once, the results obtained at the oldest age, but under 19 years, were included.
The study was conducted in accordance with the ethical principles in the Declaration of Helsinki, and with the ethical guidelines for medical and health research involving human subjects stipulated by the Ministry of Education, Culture, Sports, Science and Technology and the Ministry of Health, Labour and Welfare in Japan. The study was approved by the ethics committees of the Japanese Red Cross Toyota College of Nursing (approval number 2911), which waived the requirement for informed consent due to the retrospective nature of this study.
Data were stored in DATASELECT Inc (Aichi, Japan), a data center independent of our study group. Data on personal information were submitted by each hospital as case questionnaires, transferred to an Excel sheet, and stored in the data center. The list of patients was kept in each hospital carefully, preventing investigators from accessing any personal information. The encrypted data sheet was accessible only to a principal investigator and co-investigators.
Statistical analysis and variables
Univariable analyses were initially performed only to determine the factors to be entered into multivariable analyses, such as multiple regression analysis and structural equation modeling (SEM). SEM was chosen because it is a powerful statistical modeling technique for observational data that can be used to statistically confirm a hypothesized model by evaluating the observed covariance structure of the data. The initial model was based on the hypothesis that in VLBW infants, prematurity, IUGR, and various stressors during the neonatal period are related to CKD in childhood and adolescence.
In SEM, the opposite of an observed variable is a latent variable. Latent variables are loosely documented events that cannot be observed directly but only implied indirectly through consequences of observed variables, much representing concepts. Although observed variables are the only type of variable used in regression analyses, SEM can handle other types of variables, including latent, unobserved, and theoretical variables. Observed variables are represented by rectangular nodes in SEM, and latent variables are represented by circles or ellipses. An important difference between these two types of variables is that an observed variable usually has a measurement error associated with it, while a latent variable does not. In this study, prematurity, IUGR, and stress in the neonatal period were chosen as latent variables, and gestational age, Z-score of birth weight, maternal age, antibiotic dosing period, diuretic dosing period, maternal smoking, late-onset circulatory collapse, and kidney dysfunction in the neonatal period were selected as observed variables. The definition of a Z-score in this text is the position of a raw score in terms of its distance from the mean, measured in standard deviation units. The double-headed curved arrows connecting two variables represent covariances. We posited the unanalyzed correlations considering clinical impressions and by exploring the model like any other. This helped to improve the overall model fitting.
The primary outcome was eGFR calculated based on SCr at age > 3 years. eGFR was calculated using Uemura’s formula [
24] for Japanese children. Because urine protein to creatinine ratio (u-P/C) was available for only 75 (16.8%) patients, u-P/C could not be a study outcome. During initial univariable analyses,
t tests for independent variables and linear regression analyses for continuous independent variables were used to compare differences in eGFR. For categorized variables, extremely preterm infants were defined as those born at < 28 weeks’ gestation, and extremely low birth weight (ELBW) infants were defined as those with < 1000 g birth weight. Multivariable analyses included factors such as gestational age, weight Z-score, maternal age, antibiotic dosing period, diuretic dosing period, maternal smoking, late-onset circulatory collapse, and kidney dysfunction in the neonatal period. Logistic regression analysis was performed with the same explanatory variables and CKD (eGFR < 83.5 ml/min/1.73 m
2) as the objective variables, despite the disadvantage of only 30 events.
SEM has the ability to include latent variables, which are loosely documented events assessed by clustering of measured indicator variables. If the initial model fit was poor, the model was improved by adding or subtracting certain paths. Adequate model fit was defined as a comparative fit index (CFI) > 0.90, a root-mean-square error of approximation (RMSEA) < 0.05, and a chi-square mean/degree of freedom (CMIN/df) < 3.0. All statistical analyses were performed using SPSS Statistics 26 (IBM Corporation) and SPSS Amos 26 (IBM Corporation) software.
Complete data with no missing values is needed for many kinds of calculations such as regression analysis. We created the complete dataset by using listwise deletion which removed from our data any observation which had a missing value among one or more variables. Multiple imputation is a simulation-based statistical technique for handling missing data. After multiple imputation by SPSS, multiple regression analysis, logistic regression analysis, and SEM in the imputed dataset were performed. Little’s MCAR test showed that the null hypothesis (data missing completely at random) was rejected (chi-squared = 39.890, df = 26, p = 0.040). Multivariable analyses of the complete dataset showed that the results from the two datasets resembled each other closely.