The study was conducted in Uppsala County, Sweden, from 1996 through 1998. Cases of spontaneous abortion were identified at the Department of Obstetrics and Gynecology of Uppsala University Hospital, which is the only place in the county for the care of women with spontaneous abortions. During this period, we identified as potential case patients 652 women with spontaneous abortions who presented at the department at 6 to 12 completed weeks of gestation and whose pregnancies had been confirmed by a positive test.15 Of these women, 562 (86 percent) agreed to participate. Among the 293 women in whom chorionic villi were identified in tissue obtained at curettage, karyotyping was successful in 258 (88 percent). Chromosomes were studied with the use of G-banding, and 11 cells in metaphase were routinely analyzed16; karyotyping was considered unsuccessful if fewer than 3 cells in metaphase were obtained. Karyotype analysis revealed that 101 fetuses (58 male and 43 female) were chromosomally normal and 157 (72 male, 63 female, 16 with triploidy, and 6 with tetraploidy) were abnormal.
The control subjects were selected primarily from the antenatal care clinics in Uppsala County. They were frequency-matched to the women who had had spontaneous abortions with regard to duration of gestation (in completed weeks) and area of residence (one of the five municipalities in the county). Of the 1037 women who were seeking antenatal care and were asked to participate, 953 (92 percent) agreed to do so. All potential control subjects underwent vaginal ultrasonography before the interview. If a nonviable intrauterine pregnancy was detected, the woman was recruited as a member of the group with spontaneous abortion (this occurred in 53 of the 562 case patients).
In Uppsala County, there are approximately 3 legally induced abortions for every 10 completed pregnancies, and some of these terminated pregnancies would have resulted in spontaneous abortion if the pregnancy had continued. To limit bias in the selection of control subjects, women with induced abortions were added to the control group. In total, 310 women who had undergone induced abortions were asked to participate, and 273 (88 percent) agreed to do so. In these supplementary analyses, women with induced abortions were added to the control group according to the distribution of the length of gestation of induced abortions in Uppsala County during the study period.
Collection of Data
Three midwives conducted in-person interviews with the women with spontaneous abortion and the control subjects recruited among patients receiving antenatal care, using a structured questionnaire, and two doctors conducted interviews with the control subjects in whom abortions had been induced. Ninety percent of the case patients were interviewed within two weeks after the diagnosis of spontaneous abortion, and all were interviewed within seven weeks. All control subjects were interviewed in early pregnancy, within six days after their completed week of gestation used in matching. To avoid delay and to limit nonparticipation, 50 women who had had spontaneous abortion and 5 control subjects were interviewed by telephone.
All the women were asked to report specific sources of caffeine ingested daily on a week-by-week basis, starting four weeks before the last menstrual period and ending in the most recently completed week of gestation. Sources of caffeine included coffee (brewed, boiled, instant, and decaffeinated), tea (loose tea, tea bags, and herbal tea), cocoa, chocolate, soft drinks, and caffeine-containing medications. Respondents were offered four cup sizes from which to choose (1.0 dl, 1.5 dl, 2.0 dl, and 3.0 dl). Weekly consumption of soft drinks was estimated by the women in centiliters. We estimated the intake of caffeine using the following conversion factors: for 150 ml of coffee, 115 mg of caffeine if it was brewed, 90 mg if boiled, and 60 mg if instant; for 150 ml of tea, 39 mg if it was loose tea or a tea bag and 0 mg if herbal tea; for 150 ml of soft drinks (cola), 15 mg; for 150 ml of cocoa, 4 mg; and for 1 g of chocolate (bar), 0.3 mg. A few medications included 50 to 100 mg of caffeine per tablet.17 Of all caffeine ingested, coffee accounted for 76 percent, tea for 23 percent, and other sources for 1 percent. None of the women ingested decaffeinated coffee predominantly. The mean daily amount of caffeine ingested was calculated from the time of estimated conception (two weeks after the last menstrual period) through the most recently completed week of gestation.
Plasma cotinine was measured by gas chromatography with use of N-ethylnorcotinine as an internal standard.18 Blood samples for the measurement of cotinine were obtained from the case patients at the time of spontaneous abortion and from the control subjects at the time they were interviewed. We defined smokers as women who had a plasma cotinine concentration of more than 15 ng per milliliter19; for 23 women whose plasma cotinine values were missing, we used self-reported daily smoking during all weeks of pregnancy.
We determined scores for symptoms related to pregnancy for each week of gestation by assigning a score for nausea (0, never; 1, sometimes but not daily; 2, daily but not all day; 3, daily all day), vomiting (0, never; 1, sometimes but not daily; 2, daily), and fatigue (0, no; 1, yes but with unchanged sleeping habits; 2, yes with slightly changed sleeping habits; 3, yes with pronounced change in sleeping habits). We then calculated the average weekly score for each symptom. We also collected data on other potential risk factors.
Oral informed consent was obtained from all the women, and the study was approved by the ethics committee of the medical faculty at Uppsala University.
Data were analyzed with the use of conditional logistic-regression analysis, matched for the week of gestation. Since the study was frequency-matched, all controls were considered in the subanalyses of risks of spontaneous abortion according to fetal karyotype. Variables were included in the multivariate analyses if they were judged, a priori, to be potential confounders or if they changed the estimates of the effect of caffeine by more than 5 percent. Whenever we assessed the odds ratios for categories of caffeine intake, we performed a test for trend with the categories of caffeine intake as an ordinal scale.