The search strategy in the MEDLINE database retrieved 32 publications, of which only 6 met the inclusion criteria. The remaining publications were excluded for the following reasons: eight studied outcomes other than fetal death (abortion, preterm, low birth weight, congenital disorders, or intrauterine growth retardation); one was a case report; seven were letters to editors; six were reviews; one was not related to caffeine consumption during pregnancy; and three were not related to caffeine consumption at all.
Among the six articles which met the inclusion criteria, three were included 11,13,14. Among the excluded articles, two focused their exposures on alcohol and cigarette smoking 15,16 and used the same database as in one of the included articles 13 and the other presented data already reported elsewhere 17. The "see related articles" feature in PubMed allowed us to find one more article 18. Hand-searching the references of the articles which fulfilled the eligibility criteria, no other publication was found. The search strategy in the LILACS database and in the other specialized libraries did not locate articles related to caffeine consumption and fetal mortality.
Table 1 presents a summary of the four studies displayed in chronological publication order. The first study was published in 1977 and the most recent in 2003. Two studies were from the United States, one from Denmark, and the other from Canada. In relation to the design, two were case-control studies, one was a cohort study, and the other a cross-sectional study. As for outcomes, two investigated fetal deaths with 28 complete weeks of gestational age or more, one included abortions and fetal deaths, and the other studied fetal deaths but did not define this term.
Quality ratings 12 showed that one of the studies had a very low qualification rate, reflecting poor quality and weaknesses in the study design 11. According to the criteria proposed by Downs & Black 12, this study failed to define the main outcomes, exposures, and principal confounders and did not describe the study population's characteristics or the missing cases. External and internal validity was doubtful, and the study power was not mentioned. The main limitations of the three other studies were: not providing p-values for the principal outcomes 13,18, not presenting the distribution of the main confounding factors in either the study sample or the source population 14,18, and not describing the study's power to detect significant results 13,14,18. Discussion on study design, analysis issues, and results of the studies are summarized subsequently.
Weathersbee et al. 11 pioneered the evaluation of the relationship between coffee consumption and fetal mortality. However, serious methodological flaws disallow considering their results as scientifically valid. The authors conducted a retrospective survey of women living in 800 households chosen by random sampling of medical records from former obstetric patients at the University of Utah Medical Center or at one of the six Intermountain Health Care Hospitals in Utah and Southern Idaho, in 1974 or 1975. A 52-item questionnaire was mailed to each household to obtain information on levels of beverage consumption by family members. Caffeine intake from coffee, tea, and cola was calculated using conversion factors presented by the authors, and the pregnancy outcomes were spontaneous abortion, stillbirth, and preterm birth.
The paper is far from clear, and the lack of information may have contributed to the difficulty previous reviewers also faced in correctly identifying the study design. Heller 19 called it a retrospective cohort study, but because the subjects were not followed in a forward direction from exposure to outcome and the exposure and outcome were both determined at the same point in time, this design fits in the group of cross-sectional studies 20.
Other problems of the paper were: no definition of the "random sampling" process and no description of the inclusion and exclusion criteria. In addition, the proportion of non-respondents was high (around 39.0%), affecting the sample's representativeness. Moreover, the investigators did not report any comparison between respondents and non-respondents 9.
Concerning caffeine consumption, it was not clear whether the question was asked specifically in relation to the index pregnancy, and as information was collected after the birth outcome, recall bias may also have affected the reported consumption patterns 9. When referring to categories of caffeine consumption, the authors found that 5 out of 16 pregnant women who consumed ³ 600mg/day of caffeine had stillbirths (31.3%), whereas the incidence of stillbirths among 356 women who did not consume caffeine was 10.7%. However, since they mixed men's and women's caffeine consumption, the results are very hard to interpret.
Finally, the authors failed to adjust for possible confounders like cigarette smoking or alcohol consumption, based on the belief that since the study population belonged to a preponderantly Mormon community, they would not be affected by such exposures. Their line of reasoning was that per capita cigarette and alcohol sales in Utah were considerably lower than in the rest of the United States. This ecological argument does not permit to infer what occurs at the individual level 20, and thus no valid conclusions can be drawn from their results.
In 1993, Little & Weinberg 18 published a case-control study on risk factors for antepartum and intrapartum fetal mortality. Data were obtained from the 1980 National Natality Survey and the National Fetal Mortality Survey conducted by the National Center for Health Statistics in the United States. Multiple births and births to mothers with serious medical problems and to unmarried mothers were excluded. After a complex process of sampling 1,835 cases (women with fetal deaths with at least 28 weeks of gestation or >= 1,000g if gestational age was unknown) and 2,832 controls were included. A questionnaire was mailed after delivery to both groups requesting information on maternal demographics, reproductive history, smoking, drinking, caffeinated coffee and/or tea use, and other variables. The information was completed with data from hospital records and birth/death certificates.
The non-response rate was higher among cases (34.3%) than controls (25.8%) (p 21.
The results did not provide information about missing values for study variables. Although it is stated that the analysis was restricted to cases and controls which had valid values on all variables, the total number of cases and controls vary from table to table. In the descriptive analysis 2,565 controls were included and in the different tables showing the results of the adjusted analyses there were 2,668, 2,619, and 1,565 live births, respectively. The authors did not mention having calculated sample sizes, and in the separate analysis of antepartum and intrapartum fetal deaths there is no information regarding the study's power to detect differences between groups.
Besides the fact that the study was not primarily designed to analyze the relationship between caffeine consumption during pregnancy and fetal mortality, the measurement of caffeine intake ("cups of coffee/tea with caffeine per day during pregnancy") is far from adequate. In addition, no information was provided about how caffeine intake was ascertained, nor whether the authors considered mean coffee/tea consumption throughout pregnancy or during a specific gestational period.
The crude results showed that 20.4% of mothers of live born infants, 23.4% of those with antepartum deaths, and 19.5% of those with intrapartum deaths consumed ³ 3 cups of caffeinated coffee and/or tea daily during pregnancy. In the adjusted analysis, the highest consumption category was changed to ³ 5 cups of coffee/tea per day, and eight probable confounders (region of birth, mother's age, race, pre-pregnancy body mass index, parity, education, and cigarette and alcohol consumption) were analyzed. The highest category of caffeine consumption showed a marginally significant increase in the risk of total fetal mortality (OR = 1.37; 95%CI: 1.03-1.82), but not for antepartum or intrapartum deaths. No significance levels are provided, and in view of the marginal significance in the highest consumption category, one cannot conclude that caffeine consumption is a risk factor for fetal mortality.
Little & Weinberg 18 pioneered the separate analysis of risk factors for fetal death according to time of death (antepartum or intrapartum), which is an important contribution by the authors. It apparently makes more sense to study caffeine consumption in relation to prenatal fetal mortality, since the determinants of intrapartum deaths are much more closely related to access to quality of medical care during labor and delivery than to maternal factors 22.
Infante-Rivard et al. 14 conducted a matched case-control study planned primarily to examine the association between lupus anticoagulants, anticardiolipin antibodies, and fetal loss. Data on caffeine intake were also collected, and the association between caffeine intake before and during pregnancy and increased risk of fetal loss was investigated. Cases were women hospitalized with a medically confirmed diagnosis of spontaneous abortion or fetal death from May 1987 to November 1989 at Hospital Sainte-Justine in Montreal, Quebec. Three controls were matched to each case in the following periods of gestation: £ 16, 17-20, 21-27, and ³ 28 weeks. Controls were women in the same period of pregnancy as cases and who had not experienced a fetal loss. They were recruited from pregnant women expected to deliver at the hospital when they presented for routine blood analysis. Previous history of spontaneous abortion was an exclusion criterion for both cases and controls. A total of 331 cases and 993 controls were studied.
The authors excluded patients admitted at night and discharged before the next morning, as well as those admitted on weekends or legal holidays, a methodological issue that generated criticism 9,21,23. However, the fact that cases were not representative of all cases in the target population would not necessarily lead to bias in the estimate of caffeine consumption/ fetal loss association. On the other hand, the high percentage of refusals among cases (30.0%) and the impossibility of determining whether cases who agreed to participate in the study were more or less likely to have a history of caffeine consumption (as compared to those in the target population) may could have produced a bias. In addition, controls were recruited among women attending prenatal care while cases were recruited upon their hospitalization and no information was provided as to whether cases had been receiving prenatal care. Moreover, prenatal care was not included among the potential confounders presented by the authors. This imposed selection criterion only for controls may be another source of selection bias in this study 9,23.
Another debatable methodological aspect of the study was that the authors mixed abortions and third-trimester fetal deaths (10.0% of cases). As already pointed out by Levinton & Cowan 9 these two different outcomes may have different risk profiles, and caffeine exposure may not affect them in the same way.
An interview covered mother's age, race, education, obstetric history, smoking and alcohol use during pregnancy, occupational exposures, and medical conditions. Regarding caffeine consumption, women were asked about the intake of beverages containing caffeine such as coffee, tea, and cola before pregnancy (the month preceding conception) and during pregnancy (up to the time of study enrollment). Although cases and controls reported their caffeine intake during a relatively comparable reference period, since the investigators obtained information on both current and past caffeine intake, differential recall bias may have affected the study if cases were more likely to remember the exposure than controls. In addition, since the control group was selected among women attending prenatal care, where counseling about avoiding caffeine consumption may have occurred, the control group may have had lower caffeine intake which could lead to an overestimated association between caffeine intake and fetal loss in the study 7,21.
Quartiles for the distribution of caffeine consumption ( 321mg/day) were used as cutoffs for caffeine intake, and the category of 1,24,25. Neither a "cup" of coffee nor a "cup" of tea is a precise measure of coffee or tea intake and hence, the dose of caffeine may have been incorrectly calculated, leading to exposure misclassification. Even though this kind of error would be non-differential between cases and controls, differences in measurement methods hinder comparison across studies 7,23.
After adjusting for maternal age, education, smoking, and alcohol use during pregnancy, uterine abnormalities, and work schedules, caffeine intake during pregnancy was statistically and linearly associated with fetal loss (p 321mg/day, OR = 2.62; 95%CI: 1.38-5.01). However, it is not clear what the category of
Choice of controls is a persistently thorny methodological issue in case-control studies 26. According to the investigator's sampling approach for controls, case-control designs can be "traditional", "concurrent", or "inclusive". In "traditional" designs, controls are sampled from the population still at risk at the end of the study period. In "concurrent" designs, controls can be selected concurrently from those still at risk when a new case is diagnosed and a person originally selected as a control can therefore be classified as a case at a later date. Finally, in "inclusive" designs, controls are chosen from among all individuals in the population regardless of whether they have already had the condition under study. The latter two choices of controls allow to obtain direct estimates of relative risk and relative rate, respectively, instead of OR, an indirect estimate 27. When studying fetal death as outcome, many investigators select live births as controls. They compare their cases with "the best possible controls", those who survived the entire gestational period and were born alive. When the primary objective is to identify an association, then such case-control studies have the greatest power to find a statistically significant result. In the study by Infante-Rivard et al. 14, the fact that controls were recruited at the same time in pregnancy as the cases suggests a "concurrent" design. Controls were women at risk of experiencing a fetal loss because at the time of recruitment their fetuses were alive. As pregnancy advanced, if a woman previously selected as a control suffered a fetal loss, ideally she would have had the opportunity to be included as a case as well. In this type of design the control group represents the person-years-at-risk experience, and an analysis matched on time of selection will yield an unbiased estimate of the relative rate (incidence density ratio) instead of OR, which overestimates the real effect 27. Since with such a design the authors found a statistical association between caffeine consumption during pregnancy and fetal death, it would be expected that using a traditional design the magnitude of the observed association would have been even greater.
Wisborg et al. 13 studied the association between coffee consumption during pregnancy and the risk of stillbirth and infant death in the first year of life in a prospective follow-up study. From 1989 to 1996 all pregnant women admitted for delivery at the Aarthus University Hospital in Denmark were invited to participate in the study. The study was restricted to singleton pregnancies among Danish-speaking women who filled in the first questionnaire and delivered after 28 complete weeks of gestation (n = 25,395). Further restriction was made to women who had valid information about caffeine intake during pregnancy (n = 18,478). Information about caffeine intake was obtained from a self-administered questionnaire at about 16 weeks of gestation, before the first prenatal visit. The authors restricted the analysis of caffeine to coffee intake measured as number of cups per day (0, 1-3, 4-7, and ³ 8 cups/day).
This was the first study in which the association between coffee intake and fetal death was studied in a cohort design, thus constituting its main strength.
The authors obtained information on current intake of caffeine at about 16 weeks of gestation. Due to the study design and the timing of data collection, this information was not biased by women's knowledge of pregnancy outcome. However, several investigators demonstrated that women can change their pattern of caffeine intake during the course of pregnancy 28. Even though caffeine consumption is more likely to change in the first trimester of gestation, particularly among women suffering morning sickness 29, since caffeine intake in this study was assessed at only one point in time, it may not precisely reflect the caffeine intake pattern throughout pregnancy. It would have been more appropriate to perform a subsequent assessment of caffeine intake near the end of the pregnancy to decrease the risk that changes in caffeine consumption were not taken into account.
The authors obtained information about various caffeine sources, but they only analyzed coffee intake because "only few women were exposed to high doses of caffeine from tea and hardly any from drinking chocolate or cola". However, this reason for restricting the analysis to coffee intake is not sound, because to correctly classify the study population in terms of exposure it does not matter whether women reach high caffeine levels from different sources. An extensive accounting of all different sources of caffeine exposure would have allowed the authors to study "caffeine consumption", a more comprehensive exposure.
Concerning coffee quantification, the authors measured coffee intake by "cups" per day, and as we previously mentioned, a cup is not a precise measure of coffee intake. The authors assumed that one cup of coffee contains approximately 100mg of caffeine, but they did not collect information on beverage cup size, type of coffee, or method of preparation, so the study was subject to exposure misclassification, as already mentioned in the comments on the Infante-Rivard et al. 14 study.
Regarding the results, in the crude analysis maternal consumption of ³ 8 cups of coffee/day during pregnancy was associated with increased risk of stillbirth (OR = 3.0; 95%CI: 1.5-5.9). After adjusting for smoking and alcohol intake during pregnancy, parity, maternal age, marital status, years of education, employment status during pregnancy, and maternal pre-pregnancy body mass index, the ingestion of 1-3 cups/day (OR = 0.6; 95%CI: 0.3-1.1) and 4-7 cups/day (OR = 1.4; 95%CI: 0.8-2.5) were not significantly associated with fetal mortality, but the highest category of coffee consumption was marginally significant (OR = 2.2; 95%CI: 1.0-4.7).
Helm 30 criticized the apparent lack of consistency in the category of 1-3 cups of coffee/ day, stating that there is no chance that drinking 1-3 cups/coffee produces a protective effect whereas drinking more coffee leads to a negative effect. However, the association in that category was not significant. Jacobs 31 commented that since the authors do not present the results of an overall test for the entire variable, it was impossible to determine whether, after adjustment, caffeine consumption was still significantly associated with stillbirth.
Cohort studies have several major advantages over other types of observational studies to study the relationship between caffeine intake and fetal death, but very large cohorts are required to ensure adequate numbers of outcome events to yield statistically significant results 26. In Wisborg et al. 13, the number of fetal deaths in each category of coffee consumption was small, and the risk estimate in women with the highest coffee intake was based on only 11 fetal deaths. Their results, although not definitive, suggest a trend of increasing risk of stillbirths as the number of cups of coffee consumed per day during pregnancy increases.