No Meta-Analytic Evidence for Risks due to Prenatal MRI in Animal Models

Abstract

Background:

Magnetic resonance imaging (MRI) is a powerful, non-invasive tool for both clinical practice and research. Though the safety of MRI has been endorsed by many professional societies and government bodies, some concerns have remained about potential risk from prenatal MRI. Case-control animal studies of MRI scanning during gestation and effects on offspring are the most direct test available for potential risks. Herein, we provide a meta-analysis of extant animal studies of prenatal MRI examining reproductive and offspring outcomes.

Methods:

Relevant articles were identified through PubMed search and citation searching of known articles and review papers. Eighteen relevant studies were identified with case-control designs of prenatal scanning conducted in vivo with mammalian species using MRI-relevant field strength. Standardized mean difference effect sizes were analyzed across k=81 outcomes assessed across n=649 unexposed dams, n=622 exposed dams, n=3031 unexposed offspring, n=3378 exposed offspring, using a multi-level meta-analysis approach that clustered effect sizes within publication.

Results:

The meta-analysis indicated no significant evidence for an effect of prenatal MRI (SMD=0.17, 95% CI=[−0.19,0.54], t=0.94, p=.35) across outcomes. Similarly, no effects were observed when separately examining the four most commonly assessed outcomes: birth weight, litter size, fetal viability, and physical malformations (p>.05).

Conclusions:

Case-control mammalian animal studies indicate no significant known risks of prenatal MRI to reproductive outcomes or offspring development. This is largely mirrored in human research, though the lack of randomized case-control designs limits direct comparison. The current findings provide additional support to the prevailing consensus that prenatal MRI poses no known risk to offspring.

Introduction

Since the 1980’s, magnetic resonance imaging (MRI) has become an increasingly powerful and commonly used tool both in clinical and research settings, with millions of scans performed each year (1). With advances in MRI technology, sequences, and field strength, many have sought to confirm the safety of MRI for patients and research participants. Overall, the safety of non-contrast MRI at typical field strengths (e.g., 3 Tesla) has been confirmed by a number of organizations internationally (2-7). That said, scanning with injected contrast (e.g., gadolinium) (8,9), experimental sequences, or high field strengths (e.g., 7T) may warrant further investigation. Importantly, typical MRI scanning does not utilize ionizing radiation or radioactive tracers as with X-ray, computerized tomography (CT), or positron emission tomography (PET). Thus, the main safety concerns are contraindicated metal in the scan room, acoustic noise, and tissue heat absorption–all of which can be mitigated via standard training and protocols (7,10-12).

One lingering area of concern has been conducting MRI during pregnancy. Clinically, the use of MRI during pregnancy is supported (2,3,13-15) and often preferred to X-ray, for example. Prenatal MRI can be a powerful tool for both maternal and fetal assessment in obstetrics and perinatal health practice, as a complement to ultrasound, and when other modalities are contraindicated (13,16-20). Furthermore, many institutions are actively performing human fetal MRI research (21,22), with the goal of increasing knowledge about prenatal development. Yet, at the same time, many research institutions currently require a urine pregnancy test to avoid performing MRI on a pregnant person–with the implication that this is potentially unsafe. We recently presented the argument that urine pregnancy testing as part of screening procedures for neuroimaging research can itself incur risks and discomfort for participants (23). Concerns about prenatal MRI risk in human research studies may currently be overstated, and subsequently measures taken to protect participants may expose them to unintended harm. Factors that may not be currently considered in the assessment of potential risk include that acoustic noise is naturally damped in utero (24,25), minimal tissue or amniotic heating observed with typical parameters (25-27), no negative long-term outcomes of prenatal MRI have been found, and the safety of MRI during pregnancy is supported by human studies (28-30) and numerous organizations (2-7).

The safety of prenatal MRI and requirement of pregnancy testing for neuroimaging research has become increasingly salient recently. Following the overturn of Roe v. Wade by the Supreme Court of the United States, abortion is no longer protected by federal law. Some U.S. lawmakers have made abortion illegal and further aim to criminalize abortion. As such, acquiring and maintaining pregnancy-related data as part of research is becoming ethically and legally fraught. Understanding the potential impact of MRI during pregnancy can help inform the field and shape policies related to pregnancy testing before MRI in research institutions (23). Such information can also be provided to research participants and clinical patients, who often do not have background knowledge regarding the safety of MRI, e.g., (31), to help make informed choices about participating in MRI.

To lend further support for the safety of MRI during pregnancy, we provide a metaanalysis of extant animal studies examining prenatal MRI, focusing on non-human mammalian models. Though individual studies may suggest differences between MRI-scanned and unscanned groups of animals on certain outcomes, effect sizes have been small, likely due to random variation, and null results are often discounted. As such, we hypothesize that no significant effects of MRI during pregnancy will be found regardless of scan timing, duration, or field strength. Note that human studies also do not find strong evidence of risk (6,23). Herein, we focus on animal studies that utilize randomized case-control designs to most effectively test the potential for risk of prenatal MRI.

Methods

Study Selection (Figure S1) (32).

Our goal was to identify potential effects of in utero exposure to MRI in case-control mammalian animal studies. A PubMed search was conducted for studies published up to January 2023 and identified 516 English-language empirical articles using the search criteria: ((("pregnancy"[All Fields] OR "gestational"[All Fields] OR "prenatal"[All Fields] OR "in utero"[All Fields]) OR fetal[All Fields]) AND ("MRI"[title] OR "magnetic"[title]) NOT "review"[Publication Type]) AND ((animal[Filter]) AND (English[Filter])). An additional 32 articles were identified in review papers or cited by identified articles. The selected articles (n=18, see below) were entered into AI-based citation mapping tools, ResearchRabbit.ai (33,34) and Inciteful.xyz (35), to probe for other related articles that may have been missed (n=96). The full reference list (n=645) is available on GitHub (https://github.com/dpagliaccio/PrenatalMRI_Rodent_Meta).

The majority of identified references were excluded for not testing case-control differences (e.g., no control condition without MRI), for involving in vitro or ex vivo MRI, postnatal scanning, non-mammalian models (e.g., chicken eggs that might not involve the same mechanisms of risk), or lack of extractable numerical data for analysis (e.g., data only provided graphically in figures). Studies were also excluded from a tangential line of research probing potential effects of electrical fields or very weak, fluctuating magnetic fields (e.g., in the micro or nano Tesla range) as would be generated by computer monitors or similar electronic sources; these were excluded as they did not include a strong static MRI field.

After screening and reviewing 645 references, 18 applicable articles were selected and summarized for analysis (Table 1) (36-53). For articles testing multiple conditions, one MRI condition was selected that would maximize the potential to observe differences. Specifically, among multiple MRI conditions, one was selected that employed the most intensive scanning (longest time, highest field strength, and/or most scans). Conversely, among multiple control conditions, the one with least exposure was selected (typically, cage rearing). Conditions with exposure to other imaging, e.g., X-ray, or injection of contrast, were excluded.

Table 1:

Summary of Included Studies

Study	hours	Tesla	Timing (dpc)	Control	Dams (un)	Dams (ex)	Pups (un)	Pups (ex)	Main Outcomes
(36) McRobbie 1985	0.17	0.17	3	sham	23	23	301	299	Litter size, survival^, weight^
(37) Heinrichs 1988	16	0.35	8	cage	29	37	182	231	Litter size, survival, weight, length
(38) Tyndall 1991	0.6	1.5	7	sham	15	15	103	118	litter size, survival
(39) Murakami 1992	7	6.3	7	cage	50	84	649	1087	Litter size, survival, anomalies, weight
(40) Mevissen 1994	480	0.03	1	sham	12	12	102	62	Litter size, survival, anomalies, weight^
(41) Rofsky 1994	0.28	1.5	9	cage	15	16	161	151	Litter size, survival, anomalies, weight
(42) Carnes 1996	8	4.7	9	cage	8	10	49	56	litter size, survival^, weight^
(43) Narra 1996	0.5	1.5	2	cage	289	227	0	0	embryogenesis
(44) High 2000	30	9.4	7	cage	33	34	32	62	litter size, anomalies, weight^
(45) Magin 2000	9	4	9	cage	4	4	63	49	litter size, survival, weight^
(46) Tablado 2000	336	0.5	7	sham	10	10	100	96	litter size, survival, anomalies, weight^, testes^
(47) Gu 2001	1	0.5	8	cage	21	20	296	288	litter size, survival, anomalies, weight
(48) Lee 2001	456	0.05	0	cage	8	7	56	48	survival, neural^
(49) Okazaki 2001	48	4.7	7	cage	13	8	193	107	survival, anomalies, weight
(50) Jian 2004	4.66	0.35	12	sham	7	8	10	10	behavioral^
(51) Saito 2006	1	0.4	7	sham	80	80	714	651	litter size, anomalies, weight
(52) László 2011	40	1.47	14	cage	6	6	-	-	prematurity
(53) Hoyer 2012	22.5	7	1	sham	26	21	13	13	behavior^

Note. A summary of the studies included in the meta-analysis is provided here. The length (hours) of scanning, field strength (Tesla), and timing of scan during pregnancy (days postconception [dpc]) are presented for the MRI condition. The control condition selected is noted: cage-control or sham-control. The number of unexposed (un) and exposed (ex) dams and pups is noted for each study. A brief summary of the main outcomes examined is provided; note that this may include weight at multiple time points or varying metrics of survival or anomalies.

^{^}included neonatal–adult outcome(s)

The full dataset is available on Github.

The most common study outcomes were birth weight, litter size, fetal viability, and physical malformations. Other outcomes included a variety of measures, such as body length, adult weight, and anxiety-related behavior. Where possible, similar binary outcomes were combined to reduce the number of outcomes per study (e.g., summing counts of resorptions, fetal death, inviable fetuses, etc.). Independent variables that might relate to the magnitude of observed effect sizes were extracted: year of publication, time in scanner, field strength (highest, if multiple), scan timing (days post-conception, earliest if multiple), control condition (cage or sham), timing of outcome assessment (days postpartum, latest if multiple), and outcome type. For analysis, time in the scanner was binarized as low (<2 hours) similar to typical human MRI vs. high (often multiple days), timing of MRI was binarized as early or later in pregnancy based on the distribution of data (before vs. after post-conception day 5), and timing of outcome assessment was binarized as peripartum (pre- to 1-day postpartum) vs. neonatal–adulthood (range=13–150 days postpartum in these data).

Analysis.

All data analysis was performed in R v4.0.3. Data and code are available on www.github.com/dpagliaccio/PrenatalMRI_Rodent_Meta. Meta-analyses were performed using functions from the metafor, meta, and dmetar packages (54-56). Standardized mean differences (SMD; mean difference / pooled SD; equivalent to Hedges’ g (57,58)) were calculated as a measure of effect size for each outcome along with their corresponding sampling variance (metafor::escalc function); for binary outcomes, a transformed odds ratio was calculated as an estimate of the standardized mean difference (assuming logistic distributions; measure="OR2DL"). All effect sizes were calculated such that negative values indicated worse outcomes in the MRI group, e.g., lower birthweight and higher death rate among MRI scanned animals would both be indicated by a negative value.

Multi-level meta-analysis (meta::metagen) was conducted using the calculated SMD and sampling variance, with random effects nesting/clustering effect sizes within study (~1∣study/es.id), using restricted maximum likelihood (REML) and t-test. For ease of visualization, effect sizes were also aggregated within study (metafor::aggregate) and metaanalyzed.

Potential publication bias was assessed using contour-enhanced funnel plots (meta::funnel.meta) and Egger’s linear regression test of funnel plot asymmetry (meta::metabias; dmetar::eggers.test) (59). Influence statistics were examined (dmetar::InfluenceAnalysis) to identify potential outcomes that exerted outsized effects on the meta-analysis. Studies with a high DFFITS were excluded in a follow-up test (based on a rule of thumb > 3 * sqrt(1/k-1) SD change in pooled effect when a study is removed (60)).

Meta-regression (meta::metareg) was used to probe potential predictors of effect size. Specifically, publication year, field strength (Teslas), MRI length (realistic for humans vs. long), MRI timing (earlier vs. later), age of outcome assessment (perinatal vs. neonatal–adulthood), and control condition (cage vs. sham) were entered simultaneously as independent variables (results were confirmed by entering each in a separate regression as well).

Results

Eighteen publications examining prenatal MRI among rodents were meta-analyzed. This included k=81 outcomes assessed across n=649 unexposed dams, n=622 exposed dams, n=3031 unexposed offspring, n=3378 exposed offspring. A multi-level model, which clustered effect sizes for multiple outcomes within publication, found no evidence for an effect of prenatal MRI exposure (SMD=0.17, 95%CI=[−0.19,0.54], t=0.94, p=.35; Figure S2), i.e., a non-significant and small effect size (0.17SD difference in means) in the direction of better outcomes for the MRI group. Figure 1 presents effect sizes aggregated within each study for visual clarity. There was substantial heterogeneity in effect size (I²=100%; between-studies τ²= 0.34 95%CI=[0.06, 1.02]; within-studies τ²= 0.88 95%CI=[0.61,1.31]) (61).

Figure 1: Forest Plot of Aggregated Effect Sizes.

Note. A forest plot with effect sizes aggregated by study is plotted here. A forest plot with all individual effect sizes is plotted in Figure S1. The number of outcomes summarized per study is noted as well as the aggregated standardized mean difference (SMD) effect size and corresponding 95% confidence interval (CI). A negative SMD indicates worse outcomes in the exposed group. The blue diamond indicates the overall estimated SMD and the red bar indicates the estimated prediction interval.

We did not see clear evidence of publication bias; effect sizes were not significantly related to the size of their standard error (Figure S3; Egger’s test: t(79) = −1.40, p=.16). Examining influence statistics and omitting n=12 outcomes with high influence (DFFITS>0.33) did not change the results meaningfully (SMD=−0.12, 95%CI=[−0.39,0.14], t=−0.94, p=.35).

In a meta-regression, only the age at which outcomes were assessed was a significant predictor (b=−0.76, 95%CI=[−1.38,−0.14], t(75)=−2.43, p=.02; Figure S4); such that perinatal outcomes had a SMD closer to zero (k=16, 1209 dams, 6325 offspring; SMD=−0.02, 95%CI=[−0.38,0.35], t=−0.1, p=.92) compared to a higher, but still nonsignificant, SMD among adult outcomes (k=9, 260 dams, 1225 offspring; SMD=0.72, 95%CI=[−0.15,1.58], t=1.91, p=.09). Thus, there was a significant assessment age effect, yet no significant meta-analytic effects were found in perinatal nor neonatal–adult outcomes. No significant effects were found for publication year, scan length, scan timing, field strength, or control condition (all ∣t∣<1.27, p>.20; Figure S5). Effects were similarly non-significant if tested as individual predictors in separate models or using continuous rather than binarized predictors.

In further post-hoc analyses, no significant meta-analytic effects were observed when separately analyzing the four most commonly reported outcomes: birth weight (n = 13, SMD=−0.07, 95%CI=[−0.80,0.66]), litter size (n=11, SMD=−0.34, 95%CI=[−0.69,0.02]), fetal inviability (n=8, SMD=0.19, 95%CI=[−0.13,0.50]), or physical malformations (n=8, SMD=0.76, 95%CI=[−0.10,1.62]).

Discussion

The current meta-analysis finds no evidence for risk to offspring following prenatal MRI scanning in case-control non-human mammalian studies. Across 18 studies from 1985–2012, over 600 pregnant rodents were scanned using MRI at varying stages of gestation and a variety of reproductive and birth outcomes (e.g., resorptions, litter size, birth weight) as well as later developmental outcomes (e.g., adult weight, anxiety-related behavior) were examined. Overall, results indicate non-significant differences between MRI-scanned and unscanned offspring across a variety of outcomes. In addition to a non-significant p-value, the meta-analytic estimate trended positive; though better outcomes among the scanned offspring do not likely represent a true beneficial effect, the lack of deleterious effects was not likely due to insufficient power, for example. Furthermore, several additional studies that were excluded because data could not be extracted for meta-analysis also support null effects of prenatal MRI (62-65). Similarly, studies have shown null effects of MRI on related outcomes, e.g., no influence on viability or functioning of ovary cells (66-68).

Human Literature.

Results from this meta-analysis are consistent with the human literature supporting the lack of risk related to non-contrast prenatal MRI with typical field strength (e.g., 1.5T or 3T) and acquisition protocols (6,23). Across extant human studies, no increased risk for negative outcomes has been associated with prenatal MRI, e.g., hearing or vision problems, reduced birthweight, increases in perinatal death rate, tumors, or other functional impairment (28-30,69-76). The largest study to-date finds no evidence of risk in a population-based cohort study of over 1.4 million pregnancies, isolating 1737 cases with an MRI during the first trimester (28). As randomized case-control examination of MRI is limited in human research, studies generally examine outcome among individuals already receiving medically indicated MRI through case studies or chart review. Variability in methodology and frequent lack of control group inhibits meta-analysis. Also note that these studies may entail biased sampling, i.e., individuals not randomly assigned to MRI, as those indicated for fetal MRI (often as follow-up to prenatal ultrasound) may be at increased medical risk. Despite this, studies do not indicate any substantive evidence of risk.

Limitations & Strengths.

It is important to note that the current findings are limited by the available studies. Though the meta-analysis indicated no effect, studies did vary widely in field strengths (<1T to 9.4T) and length of exposure (up to multiple days of scanning). Several of these studies employed field strengths and exposure lengths that far exceed those used in humans. Nonetheless, no effects of field strength nor scan length were found in metaregression. Though we also do not find effects of MRI timing during pregnancy, this was limited by the available data. Particularly, note that rodent gestation is roughly equivalent to the first two trimesters of human pregnancy (77,78). Most studies (16/18) included MRI in the first 10 days post-conception, and many of these included long scanning (e.g., multiple full days of scanning (40,46,48). Thus, meta-analytic results reported here provide no evidence of risk to fetuses due to MRI scanning during the equivalent of the first trimester of human pregnancy. This is a strength of the current study, as the first trimester is underrepresented in human studies examining outcomes related to MRI during pregnancy. Studies typically use cage- or sham-control conditions; when possible, cage-control results were examined to maximize the likelihood of finding an effect of MRI, which could be due to the MR field as well as acoustic noise, for example. Also note that certain outcomes were isolated to minimize redundancy within a study, e.g., when weight was tested at multiple outcome points only the earliest and latest assessments were used. Finally, there are likely additional null differences that have not been reported in studies that would further support the lack of risk, e.g., when no gross malformations or deaths were observed in either group, this datapoint is often discounted or not reported in papers.

Political Context.

These results can help individuals make informed choices about participating in research and medical MRI. Furthermore, the lack of risk associated with prenatal MRI in this meta-analysis (and extant human studies) can help research institutions. Particularly, many institutions require urine pregnancy testing for participation in MRI research (e.g., neuroimaging studies) to mitigate perceived risks of incidental prenatal MRI. As we have previously discussed (23), the risk-benefit ratio of such urine pregnancy testing has shifted with changing political environments in the U.S., where abortion is no longer protected by federal law. As of January 2023, abortion has been banned in 13 US states, with many of these states providing no exceptions for rape or incest (79). Furthermore, 18 states currently have laws that criminalize abortions with maximum incarceration sentences up to 99 years (80). These punishments largely impact abortion providers (80), although there are states considering laws to punish women who have abortions. For example, in January 2023, the Attorney General of Alabama said that women could be prosecuted for taking abortion pills (81,82), impacting potential research participants. In this context, the requirement to complete a urine pregnancy test may itself deter potential participants from engaging in MRI research. Furthermore, research institutions should consider potential long-term risks of collecting and storing information related to participants’ pregnancy status, as this information eventually could be used as evidence in states that criminalize abortion. Overall, given no known risks of prenatal MRI (at common field strengths and without contrast), the risk-benefit ratio of pregnancy testing has shifted for research.

Future Directions.

Given difficulties inherent in examining potential risk to humans associated with prenatal MRI, meta-analysis of non-human mammalian animal studies provides evidence relevant to imaging protocols. That said, prenatal MRI already has been used in a wealth of human research studies and clinical practice–future work could aim to systematically log pregnancy and offspring outcomes in such work to further support safety. Human clinical studies could also extract additional outcomes in ongoing case-control MRI studies, for example, when comparing prenatal MRI to ultrasound in detecting fetal risk factors. Future work can also systematically test potential effects of higher-field prenatal scanning (e.g., >3T) in case-control studies in animals and humans. In addition, animal studies including larger sample sizes, non-human primates, and measurements of postnatal outcomes across infancy and childhood could provide additional informative evidence supporting lack of risks related to prenatal MRI.

Conclusions.

Results presented here–indicating no evidence for risk to offspring following prenatal MRI scanning in non-human mammalian studies–are in line with the American College of Radiology’s guidance on MR safety practices, which states that “data have not conclusively documented any deleterious effects of MR imaging exposure on the developing fetus. Therefore, no special consideration is recommended for the first, versus any other, trimester in pregnancy” (4). Furthermore, the results presented here lend support to the idea that urine pregnancy testing as part of screening procedures for neuroimaging research may not be necessary (23).