Data collection
From January 2010 to December 2020, relevant data were collected from the electronic medical records and laboratory information systems at Wonju Severance Christian Hospital (WSCH), and all data were decoded automatically. Women who had singleton or multiple births at > 38 weeks of gestation and were aged 16–50 years at the time of delivery were included in the analysis; the SCr concentration was measured at least once during the pregnancy period. Pregnant women with adverse pregnancy outcomes such as PE, gestational diabetes, gestational hypertension, and other underlying medical and surgical problems were excluded. If the SCr test was performed two or more times a week, only one of the initially tested SCr results was included. This study was approved by the Institutional Review Board (IRB) of WSCH (CR321084). This study was conducted in accordance with the principles of the Declaration of Helsinki. This was an observational study without medical interventions; therefore, the need to obtain informed consent from patients was waived. The waiver of informed consent was confirmed by the IRB of the WSCH (CR321084). The SCr concentrations were measured using the Cobas® 8000 system (Cobas® c 702 and e 801 module; Roche Diagnostics, Switzerland), Modular DPE analyzer (Roche Diagnostics), Vitros FS 5.1 (Ortho Clinical Diagnostics, Raritan, NJ, USA), Vista 1500 (Siemens Healthineers, Erlangen, Germany), and Atellica CH 930 analyzer (Siemens Healthineers). Data of the SCr concentrations were analyzed by converting mg/dL to µmol/L.
Analysis Of Original Data
For the enrolled normal pregnant women, the general characteristics, number of cases examined, and distribution of SCr concentrations according to GW were analyzed. The total GWs (0–40) was divided into three trimesters, and each trimester was split into four periods (Supplementary Table 2). The Gaussian distribution of SCr concentrations was defined according to the following three criteria: met35, kurtosis 3 (2.5–3.5), skewness 0 (− 0.5 to 0.5), and a p-value of > 0.05 in normality tests. Normality tests were performed using the Shapiro–Wilk test. Computational statistics and graphics were performed using R language and environment for statistical computing version 4.2.0 (R Foundation for Statistical Computing, Vienna, Austria).
Resampling Process
Three hyperparameters were required for the resampling process: DV as new data, RS, and RTs. DV type was set as the mean value in this study.
Step 1) A resampled dataset was extracted from the original data by setting n as the RS without replacement (Fig. 4).
Step 2) The mean value (defined as DV in this study) was obtained for a resampled dataset.
Step 3) Steps 1–2 m (referred to as RTs) were repeated. Thus, the new dataset included m DVs of SCr. According to the central limit theorem, when sufficient numbers of m and n are set, the values in the new dataset follow a Gaussian distribution, although the SCr concentrations in the original data do not show normality.
Step 4) RS (n) was increased by 1 (from 2 to 20), while RT (m) was increased by 50 increments (from 50 to 1,000). As a result, the total number of newly created datasets was 380.
Step 5) The creation of 380 new datasets (steps 1–4) was performed in each of the 12 GPs. After all the 4,560 new datasets were obtained, all distributions were evaluated to determine whether the new dataset satisfied the criteria of an RI. The first criterion of RI was that results falling between the 2.5th and 97.5th percentiles of the new dataset should be included within the 90% confidence interval of the original data. The second criterion was that the p-value of the normality test should be > 0.05. The final criterion was whether the distribution showed kurtosis (2.5 to 3.5) and skewness (− 0.5 to 0.5) (Fig. 4).
Establishment Of Gp-specific Scr Ri From New Datasets
Obstetric experts manually categorized GWs (0–40 GWs) into 12 GPs. As each GP had three or four GWs, the estimated RIs of the total GWs in each period resembled the angular cliffs. Polynomial linear regression equations were applied to develop the smooth-curved RIs. Because implantation was not completed at 0–3 GWs, 1 GP (containing 0–3 GWs) was excluded. The independent variable was the middle GW of each of the 12 GP, while the dependent variables were resampled values of the 2.5th and 97.5th percentiles in each GP. In order to identify the best degree of polynomial linear regression, the MSE was calculated based on the following degrees: 1 to 5. As k-fold cross-validation was chosen as the methodology, the degree with the smallest MSE was defined as the best degree. This cross-validation was repeated 100 times for every polynomial regression model.
Establishment Of Gp-specific Egfr Ri From Gp-specific Scr Ri
Our research team focused on the value of hyperfiltrated SCr concentration as a marker of gestational GFR. To calculate the “hyper” filtrated rate, we defined the BSC as the median of 4 GW SCr RI (56.4 µmol/L). Similarly, the first GP (0–3 GWs) was excluded.
Step 1) The gaps between BSC and the median of 2–12 GPs were calculated.
Step 2) As the proportion of each gap from step 1 was calculated based on the BSC, these rates were set in hyperfiltrations.
Step 3) The gaps between the median and the 2.5th percentile, 25th percentile, 75th percentile, and 97.5th percentile were calculated.
Step 4) The gaps from step 3 were calculated to identify the proportion of each gap using the median value in each GP.
Step 5) Boxplots were constructed according to the proportions from step 4. However, hyperfiltration was expressed in percentage; the eGFR of 120.1 mL/min/1.73 m2, as the normal value for non-pregnant women15, was multiplied to generate the eGFR in mL/min. These boxplots were used as the source for deriving the regression equations for the lower limits, medians, and upper limits. In the same way that the RI of GW-specific SCr concentration was established, the MSEs for lower limits, medians, and upper limits were calculated 100 times.
Development Of Gw-specific Scr And Egfr Ris
The RI was constructed according to the 2.5th and 97.5th percentiles, which were the lower and upper limits, respectively. As GWs that required RIs were 4–40, the GW-specific RIs were derived using the best polynomial regression equations of the upper and lower limits.
Creation Of The Gef
The median regression equation from the GP-specific RI of eGFR could help obtain the median gestational eGFR for any GW. As every SCr concentration was not the same as the median SCr concentration, extra adjustment was necessary to determine a more precise eGFR.
Step 1) We calculated the proportion of maternal SCr concentration relative to the median value of the GW-specific SCr RI.
Step 2) Because gestational eGFR was expressed in mL/min, this proportion was transformed into mL/min. Hence, 120.1 mL/min, as the normal eGFR value of non-pregnant women15, was multiplied.
Step 3) The formula was created based on the sum of the median regression equation of the GP-specific eGFR RI and the calculation process for steps 1–2.