Study design and data sources

We conducted a case-control study to quantify the association between medically assisted reproduction and the risk of multiple births. We used data built by a random selection of births with at least 1 live born from January 2006 through December 2008 in the Régie de l’Assurance Maladie du Québec (RAMQ) database using International Classification of Diseases, Ninth Revision ( ICD-9 ) and medical procedure codes related to vaginal or cesarean deliveries. Data on the mothers and children after the end of the pregnancy were obtained with the linkage of data from a maternal self-administered questionnaire with 3 administrative databases: RAMQ, Med-Echo, and Institut de la Statistique du Québec (ISQ). The RAMQ provides medical coverage to all 7.8 million Quebec residents and includes information on medical services (diagnoses and procedures) dispensed, emergency department visits, and hospitalizations (date and diagnoses coded using the ICD-9 ). The Med-Echo database records acute care hospitalizations and includes data on the primary and secondary discharge diagnoses (coded with ICD-9 until the end of 2005 or International Statistical Classification of Diseases, 10th Revision since 2006), date of admission, duration of hospitalization, and treatments received during the stay in the hospital. The ISQ database provides demographic information on the mother, father, and baby. Comparisons of ISQ data and medical charts have shown ISQ data to be complete and valid.

Additional data were collected via questionnaire administered to all selected mothers to obtain data on medically assisted reproduction utilization (ovarian stimulators, IUI, ART, and medication use during the fertility treatment) and baseline information on lifestyles, sociodemographic status, weight, and height. Information collected by the questionnaire was linked to databases using the Quebec residents’ unique identifier numéro d’assurance maladie .

Outcome definition

The births were assigned to the case group if they resulted in >1 live birth to which a validated diagnosis of multiple births (twin births or higher-order multiple births) were assigned at delivery ( ICD-9 : 6441, 6509-6519, 6580-6649, 6680-6699). Each case was matched on maternal age and year of delivery with 3 singleton live births (controls) randomly selected among all Quebec’s singleton live births. The multiplicity status (multiple birth or singleton) of each birth was first obtained from the medical records of the RAMQ and then confirmed by the delivery hospitalization found in Med-Echo database. This information was also validated using the ISQ birth registry and finally by the answer of the mother given in the questionnaire. For both cases and controls, we set the index date as the time of delivery.

By completing the questionnaire, a woman gave her consent to participate in the study.

Exposure

Exposure to all medically ART was initially defined dichotomously. This was then categorized into: (1) treatment with oral ovarian stimulators used alone (Clomid [Sanofi Aventis US, Bridgewater, NJ], Serophene [EMD Serono Inc., Rockland, MA], Nolvadex [AstraZeneca Pharmaceuticals LP, London, UK], and Femara [Novartis, Basel, Switzerland]) or injectable ovarian stimulators alone (Repronex [Ferring Pharmaceuticals Inc, Parsippany, NJ], Menopur [Ferring Pharmaceuticals Inc], Bravelle [Ferring Pharmaceuticals Inc], Gonal-F [EMD Serono Inc.], Luveris [EMD Serono Inc.], and Puregon [Organon, Merck & Co, Kenilworth, NJ]); (2) treatment with IUI with ovarian stimulators; and (3) invasive ART procedures (IVF, in vitro maturation, intracytoplasmic sperm injection, intracytoplasmic sperm injection, zygote intrafallopian transfer, and assisted hatching); and (4) women who had conceived spontaneously (reference category).

Covariates

The following potential confounding variables were assessed at the beginning of pregnancy: sociodemographic variables such as maternal age, maternal obesity (body mass index ≥30 kg/m ² ), ethnicity, education level, family income, urban residence, and marital status (living with partner vs alone); lifestyle habits in the year prior to pregnancy such as regular use of cigarettes, alcohol, and coffee; number of previous live born deliveries; subfertility (time to pregnancy >6 months [definition used in fertility clinics in Quebec] ); reduction of embryos; monitoring of the fertility treatment; diabetes mellitus diagnosis ( ICD-9 codes: 250.0-250.9, 271.4, 790.2, 648.0, and 648.8 or International Statistical Classification of Diseases, 10th Revision : E10-E14, R730, O243, and O998); metformin treatment; and folic acid supplementation.

Statistical analyses

Descriptive statistics were used to describe the study population; the proportions of categorical variables were compared between multiple and singleton births using χ ² tests. The means of the continuous variables were compared using Student t tests ( P < .05). Univariate logistic regression models were used to quantify the independent associations between each potential confounder and multiple births. The inclusion of potential confounders in the multivariate logistic regression models was based on the published literature, data availability, and univariate analysis. Results were expressed in odds ratios (ORs) with 95% corresponding confidence intervals (CIs). In primary analyses, we evaluated medically assisted reproduction use as a binary exposure (yes/no). We conducted an additional analyses evaluating medically assisted reproduction use according to specific fertility treatment of ovarian stimulator use alone, IUI use with ovarian stimulators, and invasive ART including IVF and related procedures compared to nonmedically assisted reproduction (reference category). We also quantified the risk of twin births and the risk of triplets, using ORs, associated with each type of fertility treatment. In secondary analyses, among the subgroup of medically assisted reproduction users only, we compared the risk of multiple births among IUI and invasive ART used with ovarian stimulation with the risk of ovarian stimulator use alone (reference category). We also conducted analyses according to the type of ovarian stimulators (oral or injectable). Missing values in questionnaire variables were replaced by random imputation using the distribution of the variable among patients with complete data. The missing data rate for the variables included in Table 1 was <3%. Statistical analyses were performed using software (SAS, release 9.2; SAS Institute, Cary, NC).

This study was approved by the Sainte-Justine Ethics Committee. Permission to use the data was obtained from the government ethics committee, the Commission d’Accès à l’Information of Quebec, which also approved the linkage between databases.