Materials and Methods

This is a secondary analysis of women enrolled from September 1999 through February 2002 in a multicenter, prospective, double-blind, randomized controlled trial of 17-OHPC vs placebo, conducted by the Eunice Kennedy Shriver National Institute of Child Health and Human Development Maternal-Fetal Medicine Units Network. The trial enrolled 463 women with a singleton gestation who had a history of spontaneous PTB and randomized them to receive either weekly injections of 17-OHPC (n = 310) or placebo (n = 153), beginning at 16-20 ^3/7 weeks’ gestation and continuing until 36 ^6/7 weeks’ gestation or delivery. The trial demonstrated a reduction in the rate of recurrent PTB from 54.9% in the placebo group to 36.3% in the treatment group ( P < .001).

Institutional review board approval and subject consent for the original study, as well as future analyses such as this study, were obtained at each of the 19 participating network sites by trained research nurses. This secondary analysis was reviewed by the University of Utah Institutional Review Board and determined to be exempt secondary to deidentification of data and study samples prior to this analysis. As a part of the original trial protocol, maternal saliva samples were collected for future analyses. Saliva samples were frozen at –20°C. DNA was extracted and amplified from saliva samples using established methods (Puregene; Qiagen Systems, Valencia, CA) per manufacturer’s instructions in July and August 2008.

Individuals were genotyped with SNPs in the PGR gene using TaqMan chemistry (Applied Biosystems, Foster City, CA) with established primers according to kit protocols. Tagging SNPs were selected to encompass the large PGR haplotype block, and are listed, along with known function or previously published associations in Table 1 . SDS 2.3 software (Applied Biosystems) was used to automatically determine sample genotypes (autocaller) and generate cluster plots. Genotypes were subsequently manually verified, and SNPs were evaluated for deviation from Hardy-Weinberg equilibrium using the exact test, as previously described. Samples were labeled only with a unique bar-coded study identification number, thus, researchers and laboratory personnel were blinded to all clinical data, including pregnancy outcome and treatment group assignment of the biologic samples. Only personnel at the statistical coordinating center had access to the key linking the study identification number with clinical data.

Because allele frequencies vary between races, women were stratified by self-reported race into 2 groups: African American and Caucasian/Hispanic. Women of other self-reported races were excluded. Allele and genotype frequencies were calculated for each SNP. Logistic regression was performed with PTB <37 weeks’ gestation and very PTB <32 weeks’ gestation as dependent variables. The interaction between PGR genotype and 17-OHPC therapy was evaluated by including an interaction term for these variables in the logistic regression model. Those SNPs identified to have significant treatment-genotype interactions were considered for a limited haplotype analysis. Haplotype phase was estimated using R-package haplo.stats version 1.4.4 (Mayo Clinic, Rochester, MN). Potential confounders (factors known to be associated with PTB), including prepregnancy body mass index (BMI), smoking status, and number of prior preterm deliveries, were controlled for in adjusted models.

Additive, codominant, dominant, and recessive inheritance models were considered. The additive model assumes that having 2 copies of minor allele has twice the effect of having 1 copy, the codominant model assumes that heterozygotes have an increased risk of disease over both homozygote groups, the dominant model assumes that having at least 1 copy of the minor allele is sufficient for disease, and the recessive model assumes that 2 copies of the minor allele are needed for disease. The treatment-genotype interaction was considered significant at P < .05. The inheritance model with the best P value was considered to be the best-fitting model for the respective SNP and was used to calculate odds ratios (ORs).

As this was an exploratory study, no adjustment to the alpha level was made for multiple comparisons, and all comparisons are reported. However, the false discovery rate (FDR) was calculated to evaluate the proportion of false positives among the identified positives. We used the statistical framework proposed by Tusher et al to calculate FDRs for these identified SNPs. SAS (SAS Institute, Cary, NC) and R ( www.r-project.org ) were used for the statistical analysis.