Study population and sample testing

This cohort study was conducted on women attending first trimester Down syndrome screening between July 2006 and June 2007 in New South Wales (NSW), Australia. Serum samples were collected by the Pacific Laboratory Medicine Services, and then archived and stored at −80°C. During this period, this was the state’s only public screening service and received samples from throughout NSW.

Serum samples for this study were thawed and serum levels of Ang-1 and Ang-2 were measured by a semiautomated enyzimed linked sorbent assay immunoassay (R & D Systems, Minneapolis, MN). Intraassay and interassay coefficient of variation were <9.5% and the reported analytic sensitivity of the immunoassay was 0.06–84.3 ng/mL for Ang-1 and 0.05–108.9 ng/mL for Ang-2.

Data sources

Maternal information for archived serum samples was derived from the laboratory database and corresponding pregnancy and birth outcomes were ascertained via record linkage to the Perinatal Data Collection (PDC) and Admitted Patient Data Collection (APDC). The PDC is a statutory surveillance system of all births in NSW of at least 400 g birthweight, or at least 20 weeks’ gestation and includes demographic, medical, and obstetric information on the mother, labor, delivery, and birth outcome. The APDC is a census of all patient hospital admissions from NSW public and private hospitals, with records for both mothers and liveborn infants. It includes demographic, clinical, and health services information for each admission and relevant diagnoses and procedures are recorded for each hospital admission. These are coded according to the International Classification of Diseases version 10–Australian Modification (ICD10-AM) and Australian Classification of Healthcare Interventions, respectively. Validation studies of the PDC and the APDC show excellent level of agreement with the medical record and low rates of missing data. Reporting in both datasets have high specificity (>99%) indicating few false positive reports. Only factors and outcomes accurately reported in birth or hospital data were included in analyses. The NSW Centre for Health Record Linkage conducted the record linkage and identifying information was removed before the release of data for analysis. The CHeReL assesses the linkage quality for each study, and for this study reported <5/1000 missed links and <2/1000 false positive links. The study was approved by the NSW Population and Health Services Research Ethics Committee.

Study outcomes and explanatory factors assessed included: small for gestational age (SGA), preterm birth, preeclampsia, gestational diabetes, miscarriage, and stillbirth. SGA was defined as birthweight <10th percentile and <3rd percentile (severe SGA) of the distribution for gestational age and infant sex. Gestational age is reported in the birth data in completed weeks of gestation and determined by the best clinical estimate including early ultrasound (>97%) and last menstrual period. Preterm birth was defined as delivery at less than 37 weeks and very preterm birth less than 34 weeks’ gestation. Information on preeclampsia was obtained from both the APDC and PDC data, to maximize ascertainment. Preeclampsia (regardless of severity) was determined either by the box being checked in the PDC record, or if any APDC record had a diagnosis in any of the 55 fields of gestational hypertension (ICD10-AM: O13 and O16), preeclampsia (O11 and O14), or eclampsia (O15). Early onset preeclampsia was defined as women with preeclampsia requiring delivery at ≤34 weeks’ gestation. Miscarriage was defined as a spontaneous pregnancy loss between 10-20 weeks’ gestation and identified from APDC data, whereas stillbirth was defined as a spontaneous pregnancy loss after 20 weeks’ gestation and was identified from PDC data. To replicate an earlier study of ours, we defined a proxy measure of IUGR using combined criteria of SGA <10th percentile and preterm birth <37 weeks.

The key explanatory variables were Ang-1, Ang-2, and the Ang-1/Ang-2 ratio levels and covariates used in this analysis included maternal age and weight (kilograms) ascertained at the time of first trimester screening, parity (nulliparous/multiparous), smoking during pregnancy, previous diagnosed hypertension, previous miscarriage, country of birth, and socioeconomic disadvantage quintile. Socioeconomic disadvantage was defined according to the Socioeconomic Indexes for Areas relative disadvantage scores developed by the Australian Bureau of Statistics. Maternal weight was missing in 831 (16.3%) of the records. Multiple imputation was used to account for the missing maternal weight, a method that predicts missing values using existing values from other variables. Other missing data were uncommon and were excluded from the analyses: smoking was missing in 55 (1.2%) and country of birth in 3 (0.1%) of the records.

Statistical analysis

Comparison of Ang-1, Ang-2, and the Ang-1/Ang-2 ratio by maternal characteristics for women with and without each clinical outcome was performed using contingency tables and Student’s t test analysis for categorical and normally distributed variables, respectively. Spearman coefficient was used to determine the correlation between Ang-1 and Ang-2. To account for differences in Ang-1 and Ang-2 values attributable to gestational week of the test, maternal age, and weight, we standardized Ang-1 and Ang-2 levels using multiple of the medians (MoM). Kruskall-Wallis test was used to compare concentrations of Ang-1, Ang-2, and the Ang-1/Ang-2 ratio between women that subsequently had adverse pregnancy outcomes with women with unaffected pregnancies. If MoM levels were either decreased or elevated, they were dichotomized by the <10th or the >90th percentile, respectively. Multivariate logistic regression was then used to assess the association between serum biomarkers with adverse pregnancy outcomes, taking into account maternal and clinical risk factors. A backward elimination method retaining only significant explanatory variables was used to fit models for each outcome.

Predictive accuracy was assessed examining the area under the receiver operating characteristics curves (AUC), derived from logistic regression analysis and using log transformed levels to achieve Gaussian distribution. AUC results were examined to determine whether models performed better than chance (0.5). Models for serum biomarkers alone, those including maternal and clinical risk factors only and with serum biomarkers and risk factors combined were compared. This approach was applied to assess whether serum levels of Ang-1 and Ang-2 provided any additional information to maternal and clinical risk factors in predicting severe adverse pregnancy outcomes by evaluating the maximum likelihood estimates using the likelihood ratio (χ ² ) test. Finally, estimates of predictive accuracy at a fixed 5% false positive rate were calculated including sensitivity, specificity, positive or negative predictive values, and positive likelihood ratio (LR) with exact binominal confidence intervals. A P value of < .05 was considered statistically significant and analyses performed using SAS software 9.3 (SAS Institute Inc., Cary, NC).