The study took place in Bidbid, a city located about 30 kilometres west of the capital, Muscat. This study was conducted in collaboration with an ongoing randomized community trial named: "Delaying the Development of Diabetes Mellitus Type 2 in Oman", also called the "AMAL study". This project was launched in 2004, and it aims to estimate the prevalence of pre-diabetes among an Omani population and apply appropriate interventions to prevent the occurrence of diabetes. The AMAL study enrolled a total of 1313 subjects, 824 of whom were women. Among the female enrolees, 283 were nulliparous women and the remaining 541 were parous.
Our target was to enroll the 541 parous women and to collect relevant information about their pregnancies. Out of the 541 parous women, 532 (98.3%) agreed to participate after reviewing an informed consent. The study was approved by the Medical Research and Ethics Committee at Sultan Qaboos University.
The participants were asked to fill out a maternal health card (MHC) with details of all their pregnancies. These cards were our primary source of information for antenatal and clinical details pertaining to pregnancies. MHCs are registry cards that document all the events that occurred to the mother throughout pregnancy and after delivery. The cards contain the following sections: socio-demographics, pre-pregnancy risk factors, past medical history, obstetric & gynecological history, clinical findings at each visit, investigations, details of delivery, and post-natal findings.
The participants provided a list of all their pregnancies and the MHCs. After the exclusion of miscarriages, twin-pregnancies, and pregnancies < 20 weeks of gestation, the study's final population included 1939 singleton pregnancies with available MHCs among 479 women.
An incident case of AIP was defined as an episode of plasma hemoglobin level less than 11.0 g/dl first diagnosed in the second trimester or later, i.e. from 12 weeks onwards. The cutoff point was designated as 11.0 g/dl in accordance with the WHO recommendation and the local practice in Oman. The 12th week was specified as the starting point of the eligible time frame for incident cases of AIP because it is during the beginning of the second trimester that pregnancy usually causes the steepest reduction in Hb level [22]. If a case of anaemia was diagnosed at booking or during the first trimester, it was thus considered to be a prevalent case for the purpose of this study and the pregnancy was excluded in order to limit the study population to those at risk of developing AIP.
Initial calculations of the cumulative incidence (risk) and the average hemoglobin level of occurrence of AIP were made for each level of parity, every single unit being treated as a level. The crude and adjusted measures of the effect of parity on the occurrence of AIP were obtained by using multi-level logistic regression (MLLR) analysis [23]. MLLR was preferred for analysis because it accounts for the dependency that exists among pregnancies that belong to the same woman.
The MLLR models were developed for AIP as an outcome using the hierarchical (PROC NLMIXED) regression modelling of SAS software (with a binomial distribution and logit link function). Two-level models were constructed which allowed for the grouping of pregnancy outcomes within women in order to include residuals for each pregnancy and for each woman. Thus the residual variance was partitioned into two components for each level, one showing the variance of residuals between different women and the second showing the variance of residuals between pregnancies in the same woman. This bi-level analysis revealed unobserved characteristics that affect pregnancy outcomes for the same woman, and it was these unobserved variables which showed the correlation between outcomes for pregnancies in the same woman. Variables deemed significant at p < 0.20 in a bivariate model were used in a multivariate model. A backward-selection procedure was then carried out, and variables meeting the p < 0.10 significance level were included in the final model. The odds ratios that were produced by the MLLR approximate the risk ratios which measure of effect of the relation between parity and AIP. In all MLLR models, goodness-of-fit was checked by via examining maximum likelihood estimates. The level of statistical significance was set at 0.05.
Two series of logistic regression models were conducted with different categorizations of parity. The first series treated parity as a dichotomous variable: LP (< 5) and HP (≥ 5). For the second series, parity was included as a categorical variable with the following categories: 1, 2-3, 4-5, 6-7, 8-9, and ≥ 10. With this categorization, we were able to evaluate if there was a dose-response relation between parity and risk of AIP.
Each series of analysis also included two sets of sub-analyses: a crude model and an adjusted model. In the crude model, parity was the only predictor of the occurrence of AIP. In the adjusted model, the following significant confounders were adjusted for: maternal age, maternal educational status, family income, past history of AIP, year of delivery, and inter-pregnancy time.
In order to explore the effect of changing the definition of AIP on the measures of effect of the relation between parity and AIP, two secondary analyses were conducted. First, the relationships were re-examined resetting the cutoff values for anaemia at hemoglobin levels < 10.5 g/dl and < 10.0 g/dl. Second, the study population was expanded to include all pregnancies with available MHCs (1939) and re-examined including these additional prevalent and incident cases of anaemia. All statistical analyses were performed using SAS software version 9.1 (SAS Foundation, Cary, NC).
