Maternal Caffeine Consumption and Spontaneous Abortion

A Review of the Epidemiologic Evidence

Abstract

Since the 1980s, numerous studies have investigated whether caffeine intake during pregnancy affects the risk of spontaneous abortion. A clear consensus, however, has not been reached. Because of the public health importance of this question, we reviewed the results of 15 epidemiologic studies on this topic, with particular attention to the specific methodologic problems that would generate biased findings. These include selection and recall bias, confounding, several issues pertaining to exposure measurement, and the failure to account for fetal karyotype, caffeine metabolism, the timing of fetal demise, and the possibility that an effect of caffeine may be gestational age-specific. All of the studies reviewed suffer from important methodologic limitations that hinder both the interpretation of each study individually and the comparison of results across studies. Despite the fact that most epidemiologic studies have observed a positive association between maternal caffeine intake and the risk of spontaneous abortion, we conclude that the evidence must be considered to be equivocal, given the biases likely present and the fact that most of the potential biases would tend to overestimate any association. Until studies can overcome these limitations, evidence for a causal link between caffeine and spontaneous abortion will remain inconclusive.

ArticlePlus

Click on the links below to access all the ArticlePlus for this article.

Please note that ArticlePlus files may launch a viewer application outside of your web browser.

• https://links.lww.com/EDE/A86

A number of studies on maternal caffeine consumption and the risk of spontaneous abortion (miscarriage) have been published since the 1980s. However, a consensus on whether caffeine poses a risk is still lacking. The purpose of this review is to summarize the epidemiologic evidence on caffeine intake during pregnancy and the risk of spontaneous abortion, and to address the methodologic issues that have seriously hampered the interpretation of previous studies and the establishment of an accepted conclusion on this important topic.

METHODS

We identified articles for inclusion in this review by a MEDLINE literature search using the search terms “caffeine or coffee or tea or cola or chocolate” and “miscarriage or spontaneous abortion or fetal loss.” We included only articles dealing with human subjects published in English between 1966 and January 2003. The bibliographies of the identified articles were reviewed for any additional studies that were not discovered through the MEDLINE search. We considered for review epidemiologic studies that referred specifically to caffeine exposure that occurred during the pregnancy of interest and that had spontaneous abortion as the outcome. We identified 22 reports of original data that met these inclusion criteria. Seven studies were not reviewed for the following reasons: the articles presented secondary analyses of data already reported elsewhere^1–5; the study was restricted to spontaneous abortion among women undergoing in vitro fertilization and gamete intrafallopian transfer treatments⁶; or the participants in a prospective study were enrolled only after 24 weeks gestation.⁷

Results from the remaining 15 cross-sectional, case-control and cohort studies are summarized subsequently, grouped by study design. Within each of these 3 sections, we present specific methodologic issues germane to that type of study design. Discussion of design and analysis issues relevant to all studies of caffeine and spontaneous abortion is reserved for the latter part of the review.

RESULTS

Evidence From Cross-Sectional Studies

In the 2 cross-sectional studies reviewed (Table 1), women were surveyed about past pregnancies, their outcomes, and coffee consumption during each pregnancy. In the Swedish study by Axelsson et al.,⁸ 463 female hospital workers in 1985 reported on all pregnancies that occurred during the period 1965–1984. The authors reported an elevated risk of spontaneous abortion for consumption of more than 3 cups of coffee per day during the first trimester of pregnancy (relative risk [RR] = 1.56; 95% confidence interval [95% CI] = 1.06–2.30) (Fig. 1). Similarly, Armstrong et al.⁹ reported on 35,848 past pregnancies of an undisclosed number of women in Montreal. They observed a significant trend of increasing risk of spontaneous abortion with increasing numbers of cups of coffee per day during the first trimester (RR = 1.017; 95% CI = 1.004–1.030 for each cup per day).

TABLE 1: Epidemiologic Studies of Caffeine (or coffee) Consumption and Spontaneous Abortion (SAB)

FIGURE 1.: Relative risks reported for the association between caffeine intake and spontaneous abortion in the studies under review. Numbers above the line are relative risks, dots represent the cutoff points between caffeine exposure categories, and numbers written below the dots show the caffeine exposure cutoff point when it does not fall on an increment of 100. Ref = reference category. Unless otherwise noted, the reference for a relative risk was zero consumption. In studies^8,9,14,15 in which results were presented only in terms of cups of coffee per day, 100 mg of caffeine per cup of coffee was assumed for graphing purposes. For the study by Cnattingius et al.,¹³ results are for nonsmokers. For the study by Hansteen,¹⁴ results were calculated by the reviewers from data presented in the published article.

The design of these studies makes them of limited value in evaluating a causal relationship between coffee and spontaneous abortion. The accuracy of recalling coffee consumption for a particular 3-month period many years in the past is questionable. Recall may also be affected by the miscarriage event itself (recall bias). Moreover, confounding factors that were similarly retrospectively measured (eg, smoking, alcohol drinking) would also be subject to the same recall issues, with residual confounding by mismeasured factors contributing to the observed results.

Evidence From Case-control Studies

The findings from 8 case-control studies conducted in 6 countries (United States,^10,11 Canada,¹² Sweden,¹³ Norway,¹⁴ Italy,^15,16 and Saudi Arabia¹⁷) are presented in Figure 1. These studies identified miscarriage cases from one or more hospitals, and chose controls from the same hospitals (or county); controls were either still pregnant and attending prenatal care^11–14 or had already delivered.^10,15–17 Each of these studies reported odds ratios (ORs) above 1.0. The findings ranged from a modest increase in risk associated with first trimester caffeine intake more than 300 mg per day (OR = 1.22; 95% CI = 0.80–1.87)¹⁰ to a quadrupling of risk with first trimester intake of 4 or more cups of coffee per day (OR = 3.98; 95% CI = 2.55–6.21).¹⁵

The possibility of selection bias exists for many of these studies. For example, some studies defined their control group as women who gave birth “at term” to “healthy infants” or who were admitted to the hospital for “normal delivery.”^15–17 If caffeine or coffee consumption is related to an increased risk of lower birthweight or preterm infants, which some^18–21 but not all^22–25 previous studies have reported, then choosing controls with full-term normal deliveries may be selectively choosing women with lower caffeine consumption. Furthermore, several studies^11–14 imposed a selection criterion for controls (not present for cases) that the women had to have been receiving prenatal care. Pregnant women are typically counseled during prenatal care to avoid caffeine, so selecting from this population may have created a control group with lower caffeine intake. Both types of selection bias mentioned could spuriously overestimate an association between caffeine intake and spontaneous abortion. Also, recall bias can affect the reporting of exposures in case-control studies, with the cases potentially having a stimulus (the miscarriage) to more thoroughly report caffeine intake.

Evidence From Cohort Studies

We reviewed 5 prospective cohort studies,^26–30 all conducted in the United States. In the largest of these, Fenster et al.²⁷ observed 499 miscarriage events in a cohort of over 5000 women from a managed care population in California. They found no convincing evidence that caffeine intake during pregnancy increased the risk of spontaneous abortion (RR = 1.29; 95% CI = 0.80–2.06) for >300 mg per day vs. none. Dlugosz et al.²⁸ followed a cohort of nearly 3000 women planning on giving birth at Yale New Haven Hospital, and reported a positive association at caffeine levels above 300 mg per day during the first month of pregnancy (RR = 1.75; 95% CI = 0.88–3.47). Mills et al.²⁹ followed 431 women who were enrolled within 21 days after conception (76% enrolled before conception). They found that women who consumed any caffeine during the first trimester had only a slight increase in risk of spontaneous abortion (RR = 1.15; 95% CI = 0.89–1.49), and a test for trend indicated no significant increase in risk with increasing caffeine intake. In their hospital-based study, Srisuphan and Bracken³⁰ reported a 73% increase in risk among women consuming >150 mg per day compared with ≤150 mg per day (no confidence interval provided, P = 0.03). Finally, Wen et al.²⁶ observed a dose-response with increasing first-trimester caffeine intake in an HMO-based population of women in Minnesota (RR = 1.5; 95% CI = 0.8–2.7 for 20–99 mg per day; RR = 2.0; 95% CI = 1.0–4.1 for 100–299 mg per day; and RR = 2.5; 95% CI = 1.0–6.4 for ≥300 mg per day, all compared with <20 mg per day).

The rates of spontaneous abortion observed in the 5 cohorts were quite variable (2.2%,³⁰ 4.5%,²⁸ 9.7%,²⁷ 11.5%,²⁶ 13.7%²⁹) and reflect the different gestational enrollment periods of the studies. Only 2 studies enrolled women into the cohort near the time of conception,^26,29 and thus show rates closer to what would generally be expected among clinically recognized pregnancies (approximately 15%^31,32). The remaining 3 studies enrolled pregnant women seeking prenatal care by the 13th week,²⁷ 16th week,²⁸ and 28th week³⁰ of gestation—too late in pregnancy to capture the true incidence of spontaneous abortion. These studies have excluded the spontaneous abortions that occurred earlier in the gestational period. If caffeine intake proves to be more strongly related to earlier spontaneous abortions, then these studies would have underestimated the effect of caffeine.

METHODOLOGIC ISSUES

Measuring Exposure—How and When, and What Is the Reference Period?

Women can change their caffeine intake considerably over the course of pregnancy.^{11–13,33–36} In particular, pregnant women generally decrease their caffeine intake markedly during the first trimester. Intake may then remain low throughout pregnancy or fluctuate in response to pregnancy symptoms such as nausea, indigestion, food aversions, or increased olfactory sensitivity. This poses considerable challenges to collecting data on caffeine intake during pregnancy. It is not enough to elicit a single measurement of caffeine intake as many studies have done.^{8,9,11,12,14–17,28,30} Approximately half of the reviewed studies used some version of the question “What was your usual or average frequency of consumption during pregnancy (or during the first trimester, or during the first month of pregnancy),” asked retrospectively.^{8,9,11,12,15,16,28} In view of the dynamics of caffeine intake during pregnancy, it is unclear how well study participants can respond to this question with a single assessment. Even in prospective cohort studies, caffeine intake was generally assessed at only one point in time,^27,28,30 and the timing and reference periods have varied from interviews at 4–13 weeks gestation with reference to intake “last week,”²⁷ to interviews at 5–16 weeks with reference to the “first month of pregnancy,”²⁸ to interviews at any time before 28 weeks with an unspecified reference period.³⁰

A few studies have attempted to incorporate temporal changes in caffeine intake into their exposure measurement. In both their case-control and cohort studies, Fenster et al.^10,27 took note of whether the reporting of caffeine intake before pregnancy and during pregnancy was different and, if so, asked at what point consumption had changed. With this information, the authors computed an average consumption that took into account the change. However, it is arguable how well this approach could capture multiple changes in caffeine intake over time, particularly because in one of the studies,¹⁰ women were recalling their intake an average of 16.3 months (cases) and 18.5 months (controls) after the last menstrual period. In the study by Kline et al.,¹¹ the authors used information from women who volunteered that they changed their intake during pregnancy to compute an average consumption. However, participants cannot be relied on to voluntarily report such changes when they are specifically being asked about their “usual frequency of consumption.”

Three studies made specific attempts to collect serial measurements of caffeine intake.^13,26,29 Cnattingius et al.¹³ elicited a weekly report of caffeine consumption during the first trimester, but in the context of a case-control study, this reporting was necessarily retrospective. Within their prospective study, Wen et al.²⁶ collected data on caffeine consumption in a monthly food frequency questionnaire, and then averaged monthly values to obtain first-trimester exposure. This method improves over the single measurement, but still may not be sensitive enough to capture true fluctuations in intake. Mills et al.²⁹ used the soundest methodology, assessing caffeine intake 5 weeks after the start of the last menstrual period, and then again at weeks 6, 8, 10, and 12, averaged to arrive at an estimate of first-trimester intake.

In case-control studies, differential timing of interviews between cases and controls may lead to intractable bias. This situation can arise when case and controls are not well-matched on gestational age, or gestational age cannot be appropriately accounted for in analysis.^{11,12,14,15,17} For example, in the study by Kline et al.,¹¹ a substantial proportion of the cases (41%) had spontaneous abortions before 12 weeks gestation, and the mean gestational age at interview for all cases was approximately 14 weeks. However, less than 1% of controls were identified before 12 weeks, and the mean gestational age at interview for controls was approximately 25 weeks. Cases were therefore interviewed on average much earlier in pregnancy than controls, and gestational age could not be controlled for in the analysis because of the dearth of controls at early gestational ages. Because of the expected changes in caffeine consumption over time, the usual or average consumption for controls (interviewed later) could be lower than for cases (interviewed earlier). Also, when asked to report on “usual” consumption, women may simply report on current or recent habits,¹¹ and these may also reflect lower consumption for controls than cases simply as a result of the timing of interview. Thus, the positive association observed in this study could have been simply the result of the differential timing of interviews, as the authors acknowledge.¹¹ In another study,¹⁵ cases were interviewed shortly after the spontaneous abortion and controls were interviewed after their delivery, all with reference to first-trimester intake. Thus, although cases and controls reported on a comparable reference period, the difference in the burden of recall was substantial. Other studies also had unequal recall periods for cases and controls, to varying extents.^10,12,16,17 The only case-control study under review that closely accounted for the gestational age of cases and controls and the associated timing of interviews was that by Cnattingius et al., which matched cases and controls on exact week of gestation.¹³

Quantifying Caffeine Exposure

Although in most studies, caffeine intake is dominated by coffee consumption, a complete accounting of caffeine exposure requires information on other sources of caffeine such as tea, soft drinks, chocolate drinks, solid chocolate, and caffeine-containing drugs. Only one study under review¹³ accounted for all these sources. Furthermore, for 2 primary sources of caffeine intake, coffee and tea, the amount of caffeine in a given volume depends on a number of factors, including the method of preparation.³⁷ Only 3 of the reviewed studies^11,13,17 took coffee preparation method into account.

Few studies have collected information on beverage cup size^11,13,29; most refer only to “servings” or “cups” of coffee or tea. Without more information, 6-oz cups, 10-oz mugs, and the 16-oz or greater cup that is standard at take-away coffee establishments are assumed to be the same, contributing to exposure misclassification. In a recent study that assessed heterogeneity in self-reported caffeine intake among nearly 2500 pregnant women, the range of actual portion sizes for coffee was from 2–32 oz. The 8-oz cup size commonly assumed in epidemiologic studies was in fact used by only 31% of women.³⁸

A range of caffeine conversion factors (ie, how much caffeine is assumed to be contained in one typical serving of coffee or tea) has been used in previous studies. Studies even within the same country (U.S.) have used figures for noninstant coffee ranging from 107 mg^10,27,28,30 to 139 mg,²⁶ a difference of 30%. For tea, the range is 34 mg^10,27,28,30 to 64 mg,²⁶ a difference of 88%. The ideal would be to assay the caffeine content of beverages provided by study participants. This is, however, beyond the budgetary scope of most epidemiologic studies. Bracken et al.³⁸ recently undertook such an analysis of coffee samples prepared by a sample of pregnant women and demonstrated a remarkably high degree of variability: 93 samples representing 22 coffee brands—all prepared by automatic drip—had a caffeine content of 3 to 742 μg/mL. Even samples limited to the single most common brand of coffee showed a range of 16 to 696 μg/mL.

A substantial amount of measurement error is therefore present in published studies. The absolute caffeine intake reported in individual studies is thus difficult to gauge (for example, is a reported category of 300 mg per day really 450 mg per day because study subjects on average were using 12-oz cups instead of the presumed 8-oz cups?). This leads to difficulty in offering public health advice about a threshold exposure level. Furthermore, differences in measurement methods hinder comparison across studies. For example, 2 studies reviewed here^12,26 reported a similar increase in risk (approximately 150%) for a similar level of caffeine intake (approximately >300 mg per day). However, one study assumed that a cup of coffee contained 107 mg¹² and the other assumed nearly 140 mg.²⁶ Thus, the reported risk in one study²⁶ actually occurred at an exposure level approximately one -third less than the other¹² (in terms of “cups of coffee,” which was the original metric), although the caffeine-converted results at first glance seem consistent to the reader.

Range of Exposure

Most studies have included very few women who consume moderate to high levels of caffeine, with less than 10% of their study populations reporting intake of more than 3 cups of coffee per day or more than 300 mg of caffeine per day.^{10–12,15,26–30} This narrow range of exposure may reflect the effect of medical advice to avoid caffeine during pregnancy, the effect of pregnancy symptoms such as nausea or food aversions, or measurement error stemming from crude methods of calculating caffeine exposure. Whatever its explanation, the narrow range seriously limits the power of epidemiologic analyses. This has prevented investigators from generating precise results for moderate- and high-intake categories, because those calculations are based on small numbers of women if they are undertaken at all. One recent study was conducted in Sweden because of the high intake of coffee in that country.¹³ In that study, 33% of the women consumed ≥300 mg of caffeine per day and 12% consumed ≥500 mg per day during the first trimester.¹³ Despite the large size of this study, it was still difficult to calculate precise effect estimates in the high-exposure categories because of the number of factors that needed to be incorporated into the analyses (through stratification and statistical adjustment). As medical advice to avoid caffeine intake during pregnancy becomes more widespread, it will be increasingly difficult to conduct epidemiologic studies that can address the possible risk associated with higher exposure levels.

Control of Confounding

Many potentially confounding factors have either been matched on or statistically adjusted for in the studies under review. These include maternal age,^{8–16,27–30} gestational age at enrollment or interview,^{12,13,27,28,30} cigarette smoking,^{9,10,12,15,27–29} alcohol consumption,^{9,10,12,13,15,27–29} pregnancy symptoms,^13,15 parity or gravidity,^9,13,15,29 history of spontaneous abortion,^{9,10,13,15,27,29,30} history of therapeutic abortion,¹⁰ prior gynecologic surgery,³⁰ uterine abnormality,¹² race or ethnic group,^9,10,27,30 marital status,^10,27 insurance coverage or payment group,^10,11 education,^9,12,15,29 socioeconomic status or income,^27,29 and employment status or work schedule.^9,12,27 Only crude associations were presented in other papers.^17,26 High caffeine intake is often correlated with higher maternal age, higher gravidity or parity, cigarette smoking, alcohol intake, and a history of spontaneous abortion,^27–30 all variables that are themselves risk factors for spontaneous abortion. Therefore, it is critical that studies be able to separate the effect of caffeine from these and other factors.

There is a distinct possibility of residual confounding by smoking—the effect of which would be to exaggerate any association between caffeine and spontaneous abortion. Many studies did not adjust for smoking, because this variable did not meet a significance criterion,³⁰ because adjustment for smoking was found to have little effect on the relative risk for caffeine,^11,16,26 or for unstated reasons.^8,14,17 However, smoking should be a confounder, because smoking behavior is linked to caffeine intake,^27–30 and smoking is an accepted risk factor for spontaneous abortion.³⁹ The fact that adjustment for smoking was found not to affect the caffeine–miscarriage relative risk in some studies implies that the adjustment failed (eg, as a result of the use of crude categories of exposure, failure to account for changes in smoking behavior during pregnancy, and so on) or that study participants misclassified their true smoking behavior. In studies of pregnant women, there is equivocal evidence of agreement between self-reported (yes/no) and biomarker determination of smoking behavior (both poor^40,41 and good^42–44 agreement have been observed). However, there is somewhat more solid evidence that pregnant women who self-report smoking underreport the amount smoked.^41,42,45 Only one of the reviewed studies of caffeine and spontaneous abortion used plasma cotinine levels, an objective measure of exposure to cigarette smoke.¹³ In this study, smoking was found to be an independent risk factor for spontaneous abortion, and confounding by smoking was avoided by reporting separate relative risks for smokers and nonsmokers.¹³

Perhaps the most significant issue of confounding related to studies of caffeine and spontaneous abortion involves pregnancy symptoms. It is known that pregnancy symptoms like nausea (and aversions to tastes or smells) often influence the amount of caffeine consumption in early pregnancy. In addition, nausea is more frequent and/or severe early in pregnancies that are eventually carried to term than in those that miscarry.^{10,13,15,26,27,46,47} Therefore, to what extent does caffeine intake in early pregnancy only reflect the underlying viability of the fetus? Women with nonviable pregnancies may continue to drink coffee unabated because pregnancy symptoms do not discourage it. As a result, their higher caffeine intake could be interpreted as causally related to the spontaneous abortion, when caffeine in fact was simply a marker for nonviability.

For many reasons, this complex issue has been addressed in just a few studies and only in a rudimentary manner. First, pregnancy symptoms can be difficult to characterize and quantify accurately. Second, caffeine intake and pregnancy symptoms in early pregnancy have a dynamic nature. To carry out a proper evaluation of these factors, and the relationship between them, both caffeine intake and pregnancy symptoms should ideally be measured daily or, at most, weekly. This is impractical in most longitudinal studies and a near impossibility in retrospective studies. Third, even if such detailed data were collected, it is inherently difficult to isolate the independent effects of such interrelated factors.

Most studies have not reported collecting any information on pregnancy symptoms.^{8,9,12,14,16,17,28,30} Some collected data but did not incorporate them into their analysis,^11,27,29 noting that doing so did not alter their results.^11,27 Some studies that collected pregnancy symptom data used an “any versus none” approach,^10,27 which does not take into account the severity, duration, or timing of the symptoms. Parazzini et al.¹⁵ collected information on the occurrence of nausea during the first trimester of pregnancy and also on the intensity (none, light, moderate, serious). However, because the collection of this variable started midway through the study, they had data on nausea for less than half of their study population. Cnattingius et al.¹³ asked women to assign weekly scores for nausea, vomiting, and fatigue. The average weekly scores for each variable were calculated for the first trimester and then adjusted for in analysis. After these variables were added to the regression model, the increase in risk associated with caffeine at the highest level (500 mg per day or more) was nearly halved.¹³

Two studies^10,26 did not adjust for nausea, but instead used it as a stratification factor to assess interaction. Wen et al.²⁶ noted that the association between first-trimester caffeine intake and spontaneous abortion seemed to be restricted to the period after nausea started. The risk elevation observed for postnausea caffeine intake was substantial (RR = 5.4; 95% CI = 2.0–14.6 for ≥300 mg per day vs. <20 mg per day), but the finding was based only on 4 cases of spontaneous abortion. Similarly, Fenster et al.¹⁰ observed an association with caffeine among women who reported “any” nausea during pregnancy (RR = 2.10; 95% CI = 1.20–3.70 for >300 mg per day vs. none), but the opposite effect among women who reported “no” nausea (RR = 0.53; 95% CI = 0.27–1.04 for >300 mg per day vs. none). A possible explanation for these findings is that the absence of nausea is linked to a higher risk of spontaneous abortion; thus, in a population of women already at an elevated risk of spontaneous abortion, it would be harder to detect any additional risk contributed by caffeine.¹⁰

Accounting for Metabolism

When studying caffeine in relation to spontaneous abortion, it may be necessary to go beyond measuring just the amount of caffeine consumption, and to measure the speed at which caffeine is cleared from the body. Aside from the fact that pregnancy itself affects the metabolism of caffeine,⁴⁸ caffeine clearance rates also differ among individuals through genetic propensity and in response to environmental factors.⁴⁹ The metabolic rate could play a role in how or whether caffeine influences the risk of spontaneous abortion because it determines overall exposure to caffeine, and also exposure to caffeine metabolites, which themselves may be relevant etiologic factors.

Only 2 studies have simultaneously evaluated caffeine intake and markers of caffeine metabolism in relation to spontaneous abortion.^2,3 However, both of these studies included small numbers of cases and yielded inconclusive results. One, a study based on 73 cases and 141 controls, found no interaction between cytochrome P4501A2 (CYP1A2, the enzyme primarily responsible for caffeine metabolism) phenotype and caffeine intake with respect to spontaneous abortion.³ However, the phenotyping was performed after pregnancy, and therefore may not have accurately reflected the true enzyme activity during pregnancy. The other study of 101 normal karyotype miscarriage cases and 953 controls found, contrary to the a priori hypothesis, that caffeine intake was a risk factor for spontaneous abortion among women with high and not low CYP1A2 activity (determined by phenotyping).² Both of these studies found that women with slow N-acetyltransferase 2 (NAT2) activity, determined by genotype² and phenotype,³ were at modestly increased risk for recurrent spontaneous abortion, although these studies had limited power.

The cost of collecting and processing biologic samples is only one hindrance to incorporating markers of caffeine metabolism into epidemiologic studies of spontaneous abortion. Using a standard “caffeine challenge” protocol for phenotyping is not practical in a population of pregnant women, so investigators would have to rely on spot urine testing, resulting in missing data for women who did not consume caffeine recently. Also, there is some evidence that caffeine intake can affect the rate of caffeine metabolism (and therefore the measured phenotype) in a dose–response manner,^50,51 interfering with the interpretation of results.

Identifying the Time of Fetal Demise

No study of caffeine intake and spontaneous abortion to date has been able to determine the time of fetal demise. Because a clinical spontaneous abortion may occur weeks after actual fetal demise,⁵² it must be assumed that epidemiologic studies have counted some caffeine exposures occurring after this event as being relevant to that event, when in fact they are not. This issue is of critical importance given the link between pregnancy symptoms and caffeine intake. Assuming that certain pregnancy symptoms are hormone-induced, and that these symptoms are often the reason for discontinuing the use of caffeine (especially coffee) during early pregnancy, it is likely that once these symptoms abate with the failure of the pregnancy, caffeine intake will resume or increase. It may not be for days or weeks that the failed pregnancy is expelled and a spontaneous abortion is recognized. When data on caffeine exposure are collected, it may appear that such a spike in caffeine intake preceded the spontaneous abortion, raising the question of whether caffeine intake caused the spontaneous abortion. At the very least, this caffeine intake is being factored into the reporting of usual or average pregnancy intake during data collection, when it should be ignored. This “epiphenomenon of pregnancy,” despite being described by Stein and Susser³³ more than a decade ago, remains a serious methodologic issue that has not been adequately addressed in epidemiologic studies.

Only one study to date attempted to discount caffeine intake occurring after fetal demise.¹³ In this study of first-trimester spontaneous abortion, the investigators excluded caffeine intake during the 2 weeks (an arbitrary but nevertheless reasonable assumption of the time period between fetal demise and expulsion) before the clinical spontaneous abortion and a similar time period for the controls. The relative risk for caffeine intake of 500 mg per day or more fell from 2.2 (95%CI = 0.8–6.4) to 1.4 (95% CI = 0.5–3.7). Although neither effect is precise, this analysis suggests that inadvertent inclusion of postfetal-demise caffeine intake may account for much of the positive association seen in other studies. For the Cnattingius et al. study,¹³ this issue could be addressed through a retrospective weekly report of caffeine intake very close to the relevant time period (90% of the cases were interviewed within 2 weeks of spontaneous abortion, with controls matched on the same week of gestation). However, for nearly all of the other reviewed studies, the cruder estimation of caffeine intake makes such a detailed approach impossible.

Distinguishing Karyotype

Only 3 of the reviewed studies have performed fetal tissue karyotyping to distinguish normal from abnormal karyotype abortions.^11,13,14 Because caffeine is not known to cause chromosomal aberrations in humans, it is reasonable to expect that caffeine intake during pregnancy should affect only the risk of the former type. Thus, including abnormal karyotypes in analyses would dilute the case group with fetal losses for which the association with caffeine intake is null. In the absence of other biases, this would attenuate the association between caffeine intake and spontaneous abortion. However, because analyses involving abnormal karyotype miscarriages might be subject to similar biases as normal karyotype miscarriages (in terms of confounding by pregnancy symptoms, recall bias, selection bias, and incorporating postfetal-demise exposure), one might also expect to observe spurious positive associations between caffeine and spontaneous abortions of abnormal karyotype.

Two of the studies that performed fetal karyotyping indeed reported positive associations with both normal and abnormal karyotype abortions (RR = 1.9 and 1.6 for >224 mg per day,¹¹ and RR = 2.2 and 1.8 for ≥500 mg per day¹³ for normal and abnormal karyotype, respectively, see Fig. 1). These results suggest noncausal mechanisms, and also imply that women with abnormal karyotype abortions may serve as an informative control group, neutralizing the differences that typically exist between miscarriage cases and control groups of women with viable pregnancies. Hansteen,¹⁴ in contrast to these 2 studies, found that drinking more than 4 cups of coffee per day during pregnancy increased the risk for normal (RR = 2.1; 95% CI = 1.1–3.9) but not abnormal karyotype spontaneous abortion (RR = 1.2; 95% CI = 0.6–2.6) (results calculated by the reviewers from data presented in the published article). These results are more suggestive of a causal relationship between caffeine and normal karyotype spontaneous abortion.

Establishing a Time Window of Susceptibility

It is biologically plausible that caffeine may affect the risk of spontaneous abortion preferentially during a particular period of gestation. On the one hand, caffeine may be more detrimental earlier than later in gestation, because the exposure of the (smaller) fetus may be assumed to be relatively greater.¹³ On the other hand, maternal metabolism of caffeine remains relatively unaltered during the first trimester and does not appreciably diminish until the second trimester.⁴⁸ Thus, in the second trimester, caffeine may linger in the maternal system for a longer period of time, resulting in increased exposure to the fetus. Whether the effect of caffeine on spontaneous abortion, if one exists, depends on the time period in which the fetus is exposed is a question of great public health importance. However, the design of prior studies has prevented any thorough assessment of this question.

Previous studies have not adhered to any standard criterion for the timing of fetal loss (preventing meaningful comparison across studies) and generally have not included sufficient numbers of women from different gestational periods (preventing the evaluation of effect modification by gestational age in individual studies). Table 1 shows the various inclusion criteria used by past studies. Some placed no defined limits on the range of gestational age.^8,9,12,26 Five studies had all or the majority of spontaneous abortions occurring during the first trimester,^{13,15–17,29} but most of the reviewed studies included fetal losses spanning 2 trimesters,^{8–11,14,17,26–30} and in one case 3 trimesters.¹² If there is an effect of caffeine on spontaneous abortion that is restricted to a certain time window (a reasonable assumption given the well-established time sensitivity of other external influences on the fetus⁵³), this specificity of effect is likely to be overlooked in studies that group cases from all weeks of gestation.

We are aware of only 2 studies^11,13 that evaluated whether an effect of caffeine was gestational age-specific. Cnattingius et al.,¹³ after excluding abnormal karyotype abortions and smokers, found that an association between caffeine intake and spontaneous abortion was strongest for gestational weeks 6–8. The association was attenuated but still present at weeks 9–10, and disappeared for weeks 11–12. Kline et al.,¹¹ however, found little evidence for a differential effect of caffeine on what they termed prefetal and fetal losses.

The differing gestational age distributions among the miscarriage cases in prior studies will also translate into different proportions of normal and abnormal karyotypes across the studies because later miscarriages typically consist of relatively more normal karyotype fetuses.⁵⁴ As discussed previously, the inclusion of abnormal karyotype abortions may affect the observed association between caffeine intake and spontaneous abortion. The magnitude of any resulting bias would thus be expected to vary across studies, making the comparison of study results problematic.

Another issue connected to possible time-related susceptibility to caffeine is the fact that, among the reviewed studies, only Mills et al.²⁹ included subclinical losses, and even they could not include losses that occurred before 21 days postconception. Thus, it is unclear whether caffeine intake at the beginning of pregnancy can result in a failure of pregnancy so early that it either goes unnoticed or does not require medical attention. It is exceptionally difficult to conduct an epidemiologic study, even a cohort study, which will capture subclinical spontaneous abortions. This requires the enrollment of women as they are attempting to conceive, as well as repeated biologic measurements to identify conception and then the failure of the pregnancy. Because of such practical difficulties, this may remain an unexplored issue.

CONCLUSIONS

Most of the studies under review reported positive associations between caffeine intake during pregnancy and the risk of spontaneous abortion. They are, however, inconsistent in the magnitude of effect at different levels of intake, and major differences in study design and statistical analysis preclude direct comparison of their results. Because of this, and because of the difficulty in quantifying the net effect of likely biases in each of the individual studies, it would be imprudent to pool the results of these studies to arrive at an average effect of caffeine, as has previously been attempted.⁵⁵

Although the direction of bias in certain instances may be debated, the reasonable expectation for most in the studies reviewed is to overestimate the association between caffeine intake and spontaneous abortion. For example, the fundamental bias that arises because of the interrelationship among nausea, caffeine consumption, and fetal viability, which no prior study can claim to have overcome, would create or inflate an association between caffeine and spontaneous abortion. This bias may be the most difficult to address in any future research endeavors. Other biases that would have tended to overestimate the effect of caffeine in prior studies are recall bias, selection bias, differential timing of data collection, confounding by smoking, and the inclusion of caffeine intake after fetal demise. However, the inclusion of abnormal karyotype abortions could theoretically have the opposite effect, and random measurement error in quantifying caffeine intake would bias results toward the null.

The studies to date have also failed to shed light on issues of clear public health importance. For example, can a threshold exposure be identified below which caffeine is thought not to affect the risk of spontaneous abortion? Is an effect of caffeine, if one exists, restricted to a certain time window during gestation or to women with particular patterns of caffeine metabolism?

An “optimal” study would need to identify and assemble a large cohort of women attempting to become pregnant and willing to complete detailed daily diaries of caffeine intake (sources, portion sizes, preparation details, reasons for changes in consumption if they occur), cigarette smoking, alcohol use, and pregnancy symptoms. Diaries started while attempting to conceive would ensure that these factors would be measured from conception onward. Human chorionic gonadotropin (hCG) levels in blood or urine would have to be closely monitored to identify when pregnancy occurs, and soon thereafter both blood and urine samples would be taken for estimating caffeine metabolism characteristics. Samples of usually consumed coffee and tea would be sent for laboratory analysis of caffeine content, and a personal interview would collect data on confounders not covered by the diaries (eg, pregnancy history, demographic data). A feature of this optimal study would be the collection of repeated urine samples (for quantitative hCG testing) and later frequent ultrasound examinations (useful starting at approximately 6 weeks gestation when a fetal heartbeat can be detected), to monitor the continuation of pregnancy and identify fetal loss as soon as possible before the clinical spontaneous abortion. The urine samples would also serve to measure cotinine for an objective evaluation of exposure of cigarette smoke. At the time of clinical miscarriage, fetal tissue would be collected for karyotyping. Investigators could then perform a detailed analysis of the effect of caffeine intake on the risk of spontaneous abortion, stratified by karyotype, gestational week, and by markers of caffeine metabolism, controlling for all necessary confounders.

There are obvious issues of feasibility and fundability for the “optimal” study. Particularly if conducted in the United States, the greatest challenge to feasibility would be the fact that the message that caffeine is dangerous to pregnancy is already so widespread that the enrolled study population would likely have little exposure. Typical of a complex prospective cohort study, the cost estimate would be on the order of tens of millions of dollars. However, smaller and less complex studies may not move our knowledge beyond the current point.

In summary, the methodologic limitations of past studies, combined with the inherent difficulties of addressing the question, make it difficult to draw firm conclusions about the possible effects of caffeine on spontaneous abortion. We come to this conclusion despite the fact that most studies have suggested positive associations. Any future study should incorporate more detailed data collection and monitoring of pregnancy, and depart from the more standard methods of previous studies. Standard methods are not well-suited to the biology of spontaneous abortion, in which the entire time period in question is on the scale of weeks, exposure changes considerably during that time period in tandem with other etiologic factors, and the time of the actual outcome (fetal demise) is both obscure and of critical importance because it can affect subsequent caffeine exposure. Until investigators adopt methods more suitable to the specific difficulties of studying caffeine exposure and spontaneous abortion, the evidence regarding the relation between the two will likely remain equivocal.

ACKNOWLEDGMENTS

We gratefully acknowledge the contributions of Robert Tarone during the development of the manuscript.