Study design
From 19 September to 20 November 2016, a nationwide cross-sectional survey was conducted at 24 hospitals from 16 provinces across China. The hospitals were selected according to the multi-stage sampling method (detailed below). Pregnant women, who attended these hospitals for antenatal care, were consecutively recruited into the study regardless of their gestational age.
The study was approved by the Research Ethics Committee of West China Second Hospital, Sichuan University (2016-009), and registered at ClinicalTrials.gov (ID number: NCT02887963). All study participants signed an informed consent form. The Department of Obstetrics at West China Second University Hospital and the Chinese Evidence-based Medicine Center at West China Hospital, both from Sichuan University, took the primary responsibilities for study planning and design, project management, data collection, statistical analysis, and reporting.
Sampling methods
We used a multi-stage sampling method for selecting hospitals. According to the classification of the National Bureau of Statistics, the country consisted of six geographic regions (Supplementary Figure 1). First, from each region, we selected one teaching hospital as the regional coordinating center. Subsequently, three cities were randomly selected from that region. Then, from each selected city, we chose one tertiary hospital as a survey site based on our capacity and available resources. In total, 24 tertiary hospitals were included as survey sites (Supplementary Table 1).
The participating hospitals should meet the following criteria: 1) serum ferritin and hemoglobin were routinely tested at the local laboratory; 2) the laboratory facilities were examined by a quality control team comprising senior technicians, and 3) more than 400 pregnant women attending the hospital antenatal clinic during the survey period.
Patient eligibility
Our clinical investigators consecutively enrolled eligible pregnant women, regardless of their gestational age, who were visiting the selected hospitals for antenatal care. Eligibility criteria were: completed an antenatal visit between 19 September and 20 November 2016; agreed to take a blood test and complete the questionnaire survey, by signing an informed consent form. Those pregnant women who had been participating in another clinical study during the same period were excluded.
Outcome measures
The primary outcomes were the presence of anemia and IDA. According to the Chinese Guideline for Diagnosis and Treatment of IDA during Pregnancy [13], anemia was diagnosed by hemoglobin less than 110 g/l, and IDA was diagnosed by serum ferritin concentration less than 20μg/l and hemoglobin less than 110 g/l.
All eligible participants were subjected to hemoglobin and serum ferritin examination during their antenatal visit. Blood samples were tested at the local laboratories using standard qualified methods.
We also documented previously clinician-diagnosed IDA during this pregnancy (i.e. first, second, or third trimester), which have been diagnosed by clinicians and recorded in medical charts.
Data collection
We used two structured questionnaires which had been pilot tested to collect information. The first questionnaire consisted of items regarding participant demographics, history of pregnancy and childbirth, gestational comorbidities, behavior and lifestyle, previously clinician-diagnosed IDA, and history of other gestational comorbidities. The second clinician- reported questionnaire was completed by manually reviewing electronic medical records (EMRs), including anthropometrics (e.g. height and weight), physical examinations, laboratory results (e.g. hemoglobin, serum ferritin, red blood cell count and leucocyte count), treatment of IDA, and maternal complications during pregnancy.
A centralized online electronic data capturing system was set up to collect, review, and store data. Participants either completed the questionnaire through WeChat or webpages, or a hardcopy version which was subsequently uploaded to the central system by trained investigators. Logic verification rules and reference ranges were implemented for data verification.
From each participating woman, we thoroughly collected the following information: demographic characteristics (maternal age, ethnicity, education level, local resident status, registered residential place (urban or rural) and annual family income), gestational characteristics (gestational weeks of pregnancy, multiple gestations, parity, use of assisted reproductive technology (ART), and nausea or vomiting during the first trimester (NVP)), anthropometrics (height, pre-pregnancy weight and current weight), behavior and habits (e.g. active or passive smoking, meat intake), nutrients supplementation (intake of folic acid, multivitamins and calcium supplements), and gestational co-morbidities (e.g. hematological diseases, HBV infection, hypertension, gynecological diseases, diabetes (Type I or Type II), thyroid and autoimmune diseases). Gestational co-morbidities were defined as those conditions diagnosed before conception by clinicians, or diagnoses at the first antenatal visit.
Sample size estimation
Prevalence data reported from a retrospective study was used for the sample size calculation [14]. Assuming a conservative prevalence estimate of maternal IDA being 3.96% (i.e. ≈4%), the corresponding d (tolerance error) was taken to be 0.004 (one-tenth of ), with =1.96 and 2-tailed α=0.05, the sample size required was n=9,220.
Taking the available study resources and logistics into consideration, the investigators decided to increase the sample size to n=12,000. Each regional coordinating center was supposed to enroll at least 800 pregnant women, and the other 18 hospitals were expected to consecutively recruit over 400 participants at each site.
Statistical analysis
We summarized and compared baseline characteristics of pregnant women according to trimesters. Pearson’s chi-square or Fisher’s exact test were applied to categorical variables, and one-way ANOVA or Kruskal-Wallis test to continuous variables. Pre-pregnancy BMI was categorized into underweight (BMI <18.5 kg/m2), normal (BMI ≥18.5 kg/m2 and <25 kg/m2), overweight and obese (BMI ≥ 25 kg/m2). We included women with complete data in our analyses.
We compared the observed hemoglobin and serum ferritin levels by study subgroup (singletons; multiple gestations) and by trimesters (first; second; third trimester), using means or medians as appropriate. Similarly, the frequencies and prevalence of anemia and IDA were compared with respect to trimesters and subgroups of interest.
To determine the risk factors associated with anemia and IDA, we considered all plausible factors from the literature, namely, demographic characteristics, gestational characteristics, anthropometry indicators, behavior and habits, nutrients supplementation, and gestational co-morbidities; details of these variables were given in the data collection section. Initially, associations with potential factors were explored by fitting univariate logistic regression models accounting for trimesters, with the significance level set at p = 0.1. Next, we fitted mixed-effects logistic regression models for anemia and IDA separately. In view of the inherent correlation of observations due to the clustering effects of different hospitals, a mixed-effects logistic regression model with (24) random hospital effects was considered appropriate for our study setting.
Sensitivity analysis
In our study, we developed two sensitivity analyses. Firstly, to compare the prevalence of anemia during pregnancy with dissimilar races and settings, we used the different hemoglobin concentration thresholds for anemia diagnosis from other guidelines, involving UK guideline (i.e., anemia was defined as hemoglobin concentration <110 g/l in the first trimester, <105 g/l in second and third trimesters and <100 g/l postpartum and serum ferritin concentration < 30μg/l)[15] and WHO guideline (i.e., anemia was diagnosed as hemoglobin concentration <110 g/l in the first and third trimesters, and <105 g/l in the second trimester)[16], and reported the corresponding prevalence. Secondly, we conducted a mixed-effects logistic regression model with random hospital effects by removing pregnant women with a history of hematological diseases before pregnancy.