Professional Documents
Culture Documents
Effects of Medication Intake in Early Pregnancy On The Fetal-2019
Effects of Medication Intake in Early Pregnancy On The Fetal-2019
DOI: 10.1002/pd.5436
ORIGINAL ARTICLE
Victoria Petruzzi | Jeffrey Chang | Sofia Buenaventura | Paul Walden | Benjamin Gardner |
Prenatal Diagnosis. 2019;39:361–368. wileyonlinelibrary.com/journal/pd © 2019 John Wiley & Sons, Ltd. 361
362 KUHLMANN‐CAPEK ET AL.
Linear mixed‐effects model was used to describe how the mean FF gestations, and three times among 10 women, while only one woman
changes over time, as women whose initial cell cfDNA resulted in low was tested four times.
FF may have repeated the test later in pregnancy, and assessments at The indications for cfDNA testing included advanced maternal age
different time points in the same pregnancy are correlated. In a multi- (288; 27.4%), previous abnormal aneuploidy screening (115; 11%),
variate analysis, we investigated how the gestational age at time of abnormal ultrasound findings (98; 9.3%), family history of aneuploidy
cfDNA testing, maternal age, obesity (BMI ≥ 30 kg/m2), (11; 1%), previous child with aneuploidy (5; 0.5%), or others/multiple
race/ethnicity, and prescription drugs may affect mean FF. Interaction indications (534; 50.8%).
between maternal obesity and number of prescribed medications was Table 1 lists the medications patients were prescribed at the time
also tested. The fixed effects account for the mean responses of the of cfDNA testing. Two hundred sixty‐eight women were receiving
model at different time points (ie, gestational age at cfDNA), which only one medication, 132 were on two or more medications, and
are shared across all subjects. As each subject contributed only one 651 were not on any medication. The mean (±SD) age of the subjects
pregnancy, the random effects account for pregnancy‐specific (longi- enrolled in our study was 30.9 ± 6.1 years; 332 (32.2%) were nullipa-
tudinal) correlations. Ignoring the correlation within pregnancy‐ rous; 450 (42.8%) were obese with BMI ≥ 30 kg/m2. The mean FF for
specific observations can lead to incorrect conclusions, such as inaccu- the entire cohort was 9.4 ± 4.6%, while the mean gestational age at
rate standard error estimates in fixed effects and false results from cfDNA testing was 15.5 ± 5 weeks gestation. Table 2 presents the
hypothesis tests. Random intercept models were considered at the patients' characteristics according to medication exposure. Women
individual level. taking medications were more likely to be obese, Caucasian, and to
Multivariate Cox proportional hazards models were used to esti- have had cfDNA testing earlier in pregnancy. The FF was 8.8% (95%
mate the association between maternal characteristics and the hazard CI, 6.6‐12.1) among those not taking any medications compared with
to have a low FF. This type of modeling demonstrates the incidence or 8.7% (6.3‐11.6) and 7.7% (5.1‐9.3) among those taking one or two or
hazard rate, which is the number of new adverse events per popula-
tion at‐risk per unit time. It is a semiparametric method, which TABLE 1 Medications used by patients at the time of cfDNA testing
assumes a particular probability model to represent survival times Number (%) of Women
and is therefore more robust than parametric methods. It can accom- Pharmacological Agent Taking the Medication
modate both discrete and continuous measures of event times, as well ASA 23 (4.8)
as time‐dependent covariates.11 Multivariate Laplace regression was NSAIDS other than ASA 2 (0.4)
used to study the median gestational age at which cfDNA reaches
Plavix 1 (0.2)
FF ≥ 4%. Laplace regression was introduced in the biostatistical liter-
Heparin (LMWH and UGH) 16 (3.3)
ature as a method of evaluating conditional quantiles of a potentially
Metformin 25 (5.1)
censored outcome. The method has been presented in the epidemio-
logic literature as a percentile‐based approach for the analysis of time‐ Insulin 30 (6.2)
and the Laplace regression. P < 0.05 was considered statistically signif- Antibiotics 2 (0.4)
icant. Statistical analyses were performed using the Stata 14.2 Immunomodulators 2 (0.4)
(StataCorp, College Station, Texas). Chemotherapy 2 (0.4)
Antiepileptic 13 (2.7)
Antiviral 3 (0.6)
4 | RESULTS
Anticholesterol 2 (0.4)
Thyroid (levothyroxine, methimazole, PTU) 46 (9.5)
Out of 1186 women, we excluded 11 patients because their cfDNA
tests results did not include FF, 16 because of twin gestations, and Antidepressants, mood stabilizers, 75 (15.4)
antipsychotic
three subjects who had spontaneous pregnancy reduction from twins
Antihypertensive 52 (4.4)
to singleton, leaving data from 1156 patients available for analysis.
cfDNA was tested once among 1051 pregnancies, twice in 94 Abbreviation: cfDNA, cell‐free DNA.
364 KUHLMANN‐CAPEK ET AL.
Data presented as median (IQ range) or frequency (%); obesity defined as BMI ≥ kg/m2. Abbreviation: FF, fetal fraction.
a
One‐way Analysis of Variance on Ranks;
b
Chi Square.
more medications, respectively (p < 0.01; Table 2, Figure 1). among those with BMI greater than or equal to 30 (p < 0.01). FF
Multivariate linear mixed effects model revealed that mean FF was increased by 0.3% (95% CI, 0.2‐0.35) with every 1‐week increase in
directly correlated with gestational age at the time of cfDNA testing the gestational age when cfDNA was tested (p < 0.01). When com-
and inversely correlated with maternal obesity (Table 3). Mean FF pared with the nonexposed group, the mean FF was significantly
was 2.7% (95% CI, 2.3‐3.2) lower among obese when compared with lower among pregnant mothers taking two or more prescription drugs
non‐obese women, as it averaged 6% (95% CI, 5.4‐6.7) among (3.3%; 95% CI, 2.5‐4.2% vs 5.4%; 95% CI, 4.6‐6.3; p = 0.02). No signif-
mothers with BMI < 30 as opposed to 3.3% (95% CI, 2.5%‐4.2%) icant interaction was found between maternal obesity and prescribed
TABLE 3 Multivariate linear mixed effect modes investigating factors affecting mean FF (%)
Obesity −2.7 (−3.2‐2.3) <0.01 −3.1 (−3.7‐2.4) <0.01 −3 (−3.7‐2.3) <0.01 −3 (−3.7‐2.4) <0.01
Gestational age 0.3 (0.2‐0.35) <0.01 0.3 (0.2‐0.3) <0.01 0.3 (0.2‐0.3) <0.01 0.3 (0.2‐0.3) <0.01
Medications
0 §
1 0.1 (−0.2‐0.5) 0.5
≥2 −0.6 (−1.2‐0.1) 0.02
ASA −0.6 (−2.4‐1.1) 0.5
Heparina −0.4 (−1.8‐0.9) 0.5
Metformin −1.8 (−3.4‐0.2) 0.02
Linear mixed effects models with random intercept at the individual level: Fixed effects coefficients indicate the increase (positive value) or decrease (neg-
ative value) in the mean FF value for a unit change in continuous covariates or when compared with the reference group (§) in case of categorical covariates.
The random effects covariance matrix was assessed for each model, and it was found to be 16.4 when the entire cohort was studied, while 17.7, 17.9, and
18 when the model respectively investigated aspirin, heparin, or metformin. Abbreviation: FF, fetal fraction.
a
Heparin includes women exposed to either UFH or LMWH.
medications. When pharmacological agents were investigated individ- model also revealed that women taking two or more prescription
ually, exposure to metformin was associated with 1.8% (95% CI, 0.2‐ drugs developed FF ≥ 4% 1.6 weeks (95% CI, 0.7‐2.4) earlier than
3.4; p = 0.02) lower mean FF when compared with subjects not taking women taking only prenatal vitamins (p < 0.01; Table 4).
any prescription drugs. Treatment with aspirin or heparin (either
unfractionated or low molecular weight) did not affect FF.
We then investigated the determinants of FF above 4%. Multivar- 5 | DISCUSSION
iate Cox proportional hazard model revealed a significantly higher
Hazard ratio (HR) for FF less than 4% among obese women (HR 9.7; Noninvasive prenatal testing using cfDNA is a widely used aneuploidy
95% CI, 4.1‐22.6; p < 0.01), and women taking two or more prescrip- screening test in pregnancy. However, despite its widespread use,
tion drugs (HR 2.3; 95% CI, 1.2‐4.4; p < 0.01; Table 4). As FF is gesta- there are limited data about conditions, which may affect the reliability
tional age‐dependent, we investigated the potential factors that of results, particularly the FF. Our study provides evidence that sup-
could affect the median gestational age at which cfDNA collection would ports previously reported data that gestational age and obesity signif-
lead to FF values greater than or equal to 4%. Multivariate Laplace icantly impact the FF. We also found that metformin may negatively
regression suggested that maternal obesity delays by 5 weeks (95% CI, impact the FF. In addition to metformin, our research shows trends,
3.6‐6.3), the gestational age at which such cfDNA threshold is reached: which imply that other medications, such as aspirin and heparin, may
The median gestational for cfDNA with FF ≥ 4% was 13 weeks (95% also impact FF, but the low numbers of patients in our study who were
CI, 12.3‐13.6) for women with BMI < 30 as opposed to 18 weeks (95% taking these medications of interest may have limited our ability to
CI, 16.6‐19.3) for obese women (p < 0.01; Table 4, Figure 2). The same show significance. This lack of statistical significance, presumably
TABLE 4 Multivariate Cox regression investigating factors associated with FF < 4% and multivariate Laplace regression determining covariates
associated with median gestational age at cfDNA testing leading to FF ≥ 4%
Multivariate Cox regression: HR > 1 indicate an increased likelihood, while HR < 1 indicated a decreased likelihood of achieving FF < 4% at cfDNA testing,
when compared with the reference group (§) for categorical covariates. Multivariate Laplace Regression: Coefficients indicate an increase (positive value) or
a decrease (negative value) in the median gestational age when cfDNA reaches values ≥4% for a unit change when compared with the reference group (§)
in case of categorical covariates.
366 KUHLMANN‐CAPEK ET AL.
secondary to limited sample size, is in contrast with previously pub- biguanide, which works by decreasing gluconeogenesis and increasing
lished literature, which demonstrated a significant negative effect with peripheral insulin sensitivity. This drug is useful in the treatment of
low‐molecular‐weight heparin. diabetes and polycystic ovarian syndrome. Insulin is a naturally occur-
A recently published case report showed that a patient had a false‐ ring substance in the human body, used to control the level of glucose
negative cfDNA testing for trisomy 21 (with FF 1.6%) while she was in the blood, but there are many different forms of synthetic insulin,
receiving low‐molecular‐weight heparin. A repeat cfDNA while patient which, when administered to diabetic patients, may affect the body
was off low‐molecular‐weight heparin was high risk for trisomy 21 differently than that which is produced by the pancreas. Steroids are
with FF 5.76%.14 The authors concluded that the reason for the used as anti‐inflammatory agents in a wide variety of conditions,
false‐negative result was the use of low‐molecular‐weight heparin, including asthma exacerbations and lupus.
which is hypothesized to reduce trophoblast apoptosis and enhance Our study has both strengths and limitations. The sample size is one
survival. Other cases report low‐FF and cfDNA failures in patients tak- of the largest to date on this subject; however, it may not have been
ing aspirin, prednisolone, and low‐molecular‐weight heparin.15,16 large enough to identify differences with individual medications. All of
Women using low‐molecular‐weight heparin have smaller DNA frag- the cfDNA analyzed and included in this study were done using the Pan-
ments and higher proportion of guanine‐cytosine base pairs. This cor- orama test by Natera, which allows consistency in sample handling,
relates with higher content of chromosome 18, which may lead to since the sampling methods and procedures can differ between individ-
false‐positive testing, and lower content of chromosomes 21 and 13, ual tests. However, this also means that these results may or may not be
which may lead to false‐negative testing. Low‐molecular‐weight hepa- applicable to testing done by other platforms. Additional strengths
rin is used for prophylaxis and treatment of venous thromboembolism include a diverse patient demographic, which enhances the generaliz-
and conditions, which may predispose an individual to developing ability of our results, and the availability of data on most of the babies
venous thromboembolism in pregnancy, such as antiphospholipid anti- at time of delivery, which allowed us to unequivocally verify the results
body syndrome and inherited thrombophilias. We hypothesized that of the screening test. While we reported on the rates of re‐draws in our
other medications may cause similar changes to the ones observed study, we did not analyze the effect of time between re‐draws or FF at
with low‐molecular‐weight heparin. However, for most of these, there each interval. We did not perform a formal power analysis or included
are little to no data, which address the effects of these medications on the effect of coefficient of variation of the assay into that analysis.
FF levels and the potential mechanisms by which they may affect Other weaknesses of this study include the possibility of confounding
these levels. Aspirin is a nonsteroidal anti‐inflammatory drug, which data, low numbers of patients using the medications of interest, and
is used clinically for treatment of pain and fever, as well as for its anti- an inability to separate the effect of the medication from that of the dis-
platelet properties for women with cardiovascular disease. In addition, ease process itself. The patients in our study who were taking metfor-
many pregnant women with risk factors for preeclampsia take aspirin min were diabetics. Therefore, it is plausible that the effects of
to reduce their risk of developing this disease. Metformin is a metformin on FF may have been secondary to the disease itself. This
KUHLMANN‐CAPEK ET AL. 367
is an important distinction, and until there is a plausible proposed mech- gestational age is different from other published data, which show a
anism by which medication, rather than disease process, mechanistically median FF of approximately 10% by 10 to 13 weeks' gestation in all
impacts FF, care must be taken when discussing the effects of either women.20,21 Along with consideration of other factors, which may
factor. While metformin has been shown to reduce the secretion of sol- affect FF, providers can begin to understand the likelihood of achiev-
uble fms‐like tyrosine kinase 1 and soluble endoglin secretion from vil- ing an interpretable result and counsel patients accordingly. This data
lous cytotrophoblast cells, and preterm preeclamptic placental villous suggest that if mothers are obese, NIPT may need to be ordered at a
explants,17 we cannot speculate based on our data the exact molecular later gestational age to decrease the chance of low FF.
mechanisms by which metformin (or other medications) may affect FF, In conclusion, our study confirms the evidence that gestational age
as we did not perform corresponding lab work to inform such assertions. and obesity correlate with significant changes in FF. We also found
Moreover, while Natera states that results of the Panorama cfDNA test that metformin intake may be associated with lower FF. It is impera-
are validated at FF ≥ 2.8%,13 Cuckle2 reports that 1.6% to 6.4% repre- tive that, as the scope of testing utilizing cfDNA continues to expand,
sent the low FF range depending on the company/platform used, and we conduct further research to understand how different factors in
Revello1 states that up to 8% is considered low FF. For this study, we the prenatal condition might impact results. This also further highlights
used 4% as this the most commonly used cutoff in other publications, the importance of considering testing using cfDNA to be a screening
which provides a basis for comparison. Of note, we studied FF not only test only and that women should have definitive testing to confirm
as a categorical/binary variable but also as a continuous variable. any suspected anomalies.
The relationship between maternal and fetal DNA in a given sample
can be expressed as either 3FF + 2(1‐FF), or as 2 + FF, per genome.2 The CONFLIC T OF INT E RE ST
lower the FF, the harder it is to discriminate aneuploidies. The specific The authors report no conflict of interest and no sources of support.
cfDNA test utilized in this analysis (Panorama by Natera) analyzes
single‐nucleotide polymorphisms. Of note, this specific methodology ORCID
for estimating FF is associated with a relatively high rate of uninterpret-
Maggie Kuhlmann‐Capek https://orcid.org/0000-0002-7632-121X
able results, compared with other methods.18 Y chromosome‐based
methods for determining FF are useful if the fetus is male, as theY chro-
RE FE RE NC ES
mosome is an unambiguous indication of fetal DNA. There are also
1. Revello R, Sarno L, Ispas A, Akolekar R, Nicolaides KH. Screening for
methods, which have been validated independent of fetal gender, such trisomies by cell‐free DNA testing of maternal blood: consequences
as the SeqFF and SANEFALCON methods. SeqFF is a multivariate of a failed result. Ultrasound Obstet Gynecol. 2016;47(6):698‐704.
model, which contains two regression models. SANEFALCON deter- 2. Cuckle H. cfDNA screening performance: accounting for and reducing
mines FF through the distribution of reads, which have been mapped test failures. Ultrasound Obstet Gynecol. 2017;49(6):689‐692.
around nucleosome positions on autosomal chromosomes.19 3. Pergament E, Cuckle H, Zimmermann B, et al. Single‐nucleotide
polymorphism‐based noninvasive prenatal screening in a high‐risk
An interesting outcome of our study was the Laplace regression,
and low‐risk cohort. Obstet Gynecol. 2014;124(2, PART 1):210‐218.
which actually projected that women exposed to two or more medica-
4. Ashoor G, Syngelaki A, Poon LCY, Rezende JC, Nicolaides KH. Fetal
tions would achieve a higher FF at an earlier time point, which fraction in maternal plasma cell‐free DNA at 11–13 weeks' gestation:
contradicted what was determined with previous models. We believe relation to maternal and fetal characteristics. Ultrasound Obstet
this can be explained by recognizing that the previous models investi- Gynecol. 2013;41(1):26‐32.
gated mean FF and the hazard ratio of having a low FF, while Laplace 5. Yared E, Dinsmoor MJ, Endres LK, et al. Obesity increases the risk of
regression investigates time to event. We do not investigate factors failure of noninvasive prenatal screening regardless of gestational
age. Am J Obstet Gynecol. 2016;215(370):e1‐e6.
associated with a specific event but with the time in which the said
6. Benn P, Valenti E, Shah S, Martin K, Demko Z. Factors associated with
event occurs. The univariate analysis in Table 2 shows that women
informative redraw after an initial no result in noninvasive prenatal
exposed to two or more medications had Non‐invasive prenatal test- testing. Obstet Gynecol. 2018;132(2):428‐435.
ing (NIPT) significantly earlier that all the others. This may be the rea- 7. Grömminger S, Erkan S, Schöck U, et al. The influence of low molecular
son why, in the Laplace model, which studies the factors associated weight heparin medication on plasma DNA in pregnant women. Prenat
with the time when FF becomes greater than or equal to 4%, exposure Diagn. 2015;35(11):1155‐1157.
to two or more medication is associated with earlier gestational ages— 8. Mitchell A, Gilboa S, Werler M, Kelley K, Louik C, Hernandez‐Diaz S.
all the women taking two or more medications had early testing. Medication use during pregnancy, with particular focus on prescription
drugs: 1976‐2008. Am J Obstet Gynecol. 2011;205:51.
The Kaplan‐Mayer estimator (Figure 1) provides a way in which
9. Ayad M, Costantine M. Epidemiology of medications use in pregnancy.
providers might apply an understanding of the effects of maternal
Semin Perinatol. 2015;39(7):508‐511.
obesity on FF. This model suggests that maternal obesity delays the
10. Martin J, Hamilton B, Osterman M. Births: final data for 2011. Natl
gestational age at which the threshold FF to achieve an interpretable Vital Stat Rep. 2013;62:5‐6.
result is reached. According to this model, the median gestational 11. Lin DY, Wei LJ. The robust inference for the Cox proportional hazard
age to achieve FF greater than or equal to 4% in obese women was model. J Am Stat Assoc. 1989;408:1074‐1078.
18 weeks. In non‐obese women, the median gestational age to 12. Orsini N, Wolk A, Bottai M. Evaluating percentiles of survival. Epidemi-
achieve FF greater than or equal to 4% was 13 weeks. Of note, this ology. 2012;235:770‐771.
368 KUHLMANN‐CAPEK ET AL.
13. Bottai M, Zhang J. Laplace regression with censored data. Biom J. 19. Van Beek DM, Straver R, Weiss MM, et al. Comparing methods for
2010;524:487‐503. fetal fraction determination and quality control of NIPT samples.
14. Ma G, Wu W, Lee M, Lin Y, Chen M. The use of low molecular weight Prenat Diagn. 2017;37(8):769‐773.
heparin reduced the fetal fraction and rendered the cell‐free DNA test- 20. Wang E, Batey A, Struble C, Musci T, Song K, Oliphant A. Gestational
ing for trisomy 21 false negative. Ultrasound Obstet Gynecol. age and maternal weight effects on fetal cell‐free DNA in maternal
2017;51:276‐277. plasma. Prenat Diagn. 2013;33(7):662‐666.
15. Hui CY, Tan WC, Tan EL, Tan LK. Repeated failed non‐invasive prena- 21. Ashoor G, Poon L, Syngelaki A, Mosimann B, Nicolaides K. H: fetal
tal testing in a woman with immune thrombocytopenia and fraction in maternal plasma cell‐free DNA at 11–13 Weeks' gestation:
antiphospholipid syndrome: lessons learnt. BMJ Case Rep. effect of maternal and fetal factors. Fetal Diagn Ther.
2016;2016:1. 2012;31(4):237‐243.
16. Burns W, Koelper N, Barberio A, et al. The association between
anticoagulation therapy, maternal characteristics, and a failed cfDNA
test due to a low fetal fraction. Prenat Diagn. 2017;37(11):1125‐1129. How to cite this article: Kuhlmann‐Capek M, Chiossi G, Singh
17. Brownfoot FC, Hastie R, Hannan NJ, et al. Metformin as a prevention P, et al. Effects of medication intake in early pregnancy on the
and treatment for preeclampsia: effects on soluble fms‐like tyrosine fetal fraction of cell‐free DNA testing. Prenatal Diagnosis.
kinase 1 and soluble endoglin secretion and endothelial dysfunction. 2019;39:361–368. https://doi.org/10.1002/pd.5436
Am J Obstet Gynecol. 2016;214(3):356e1‐356e15.
18. Palomaki GE, Kloza EM, Lambert‐Messerlian GM, et al. DNA sequenc-
ing of maternal plasma to detect Down syndrome: an international
clinical validation study. Genet Med. 2011;13(11):913‐920.