Ischemic Placental Disease, Preterm Delivery, and Their Association With Opioid Use During Pregnancy

Abstract

Opioids affect placental development and function in animal models, but human data on their association with ischemic placental disease are limited. Using a cohort of pregnant women in the US nationwide Medicaid Analytic eXtract (2000–2014), we compared women with ≥2 opioid dispensings in pregnancy with unexposed women. Given an uncertain etiologically relevant window, we assessed exposure occurring in early pregnancy, late and not early pregnancy, and both early and late pregnancy. For placental abruption, preterm delivery, small for gestational age (SGA), and preeclampsia, we estimated adjusted hazard ratios (aHRs) and 95% confidence intervals (CIs) using Cox proportional hazard models adjusting for demographic factors, indications/comorbidities, and medications. Among 1,833,871 eligible pregnancies, ≥2 opioid dispensings were filled in 6.5%. We observed an early exposure aHR of 1.34 (95% CI: 1.26, 1.43) for placental abruption, 1.21 (95% CI: 1.18, 1.23) for preterm delivery, 1.13 (95% CI: 1.09, 1.17) for SGA, and 0.95 (0.91, 0.98) for preeclampsia. Estimates for late exposure were attenuated. Early and late exposure was associated with higher aHRs for placental abruption, 1.62 (95% CI: 1.47, 1.78); preterm delivery, 1.37 (95% CI: 1.33, 1.42); and SGA, 1.26 (95% CI: 1.19, 1.33); but not preeclampsia, 0.99 (95% CI: 0.93, 1.05). Prescription opioids may modestly increase risk of placental abruption, preterm birth and SGA, but they do not appear to be associated with preeclampsia.

Abbreviation

aHR

adjusted hazard ratio

CI

confidence interval

HR

hazard ratio

IPD

ischemic placental disease

IUGR

intrauterine growth restriction

LMP

last menstrual period

MAX

Medicaid Analytic eXtract

MME

morphine milligram equivalent

SGA

small for gestational age

Editor’s note: An invited commentary on this article appears on page 769, and the authors’ response appears on page 773.

Pain complaints are common during pregnancy (1), and opioids are frequently prescribed during this time (2–5). Nationwide estimates suggest that approximately 20% of Medicaid beneficiaries and 14% of commercial insurance beneficiaries are dispensed an opioid during pregnancy (2, 6).

Opioid use has been associated with adverse pregnancy outcomes including neonatal abstinence syndrome (7) and certain congenital malformations (8). Less intensively studied in relationship to prescription opioid exposure is ischemic placental disease (IPD), including preeclampsia, placental abruption, and intrauterine growth restriction (IUGR)/being born small for gestational age (SGA), which often result in preterm birth (9).

An association between prescription opioid use and IPD is biologically plausible. Opioids can cross the placenta (4–7), and animal studies suggest that opioid use may result in placental insufficiency, whereby deterioration of placental function reduces transfer of oxygen and nutrients causing fetal hypoxemia and poor growth (10, 11). Stimulating opiate κ receptors by morphine exposure can depress acetylcholine-facilitated amino acid transport, which is essential for placental perfusion (12). Under these conditions, spiral artery remodeling expected between 10 and 20 weeks of gestation can fail, resulting in reduced trophoblast invasion and inability of the placenta to provide adequate nutrients to the developing fetus (13). This is seen in animal models, where morphine exposure results in reduced placental and fetal weights (12).

While data are few in relation to prescription opioid exposure, IPDs occur at a higher frequency in women with opioid use disorders (14–19). However, this may be explained by lifestyle factors and other exposures related to opioid use disorders, and well-controlled studies are lacking. Given considerable morbidity associated with IPD and widespread use of prescription opioids during pregnancy, a better understanding of the influence of prescription opioid use on the risk of IPD and preterm birth is needed. We sought to examine this association using a nationwide cohort of pregnant Medicaid beneficiaries, paying specific attention to confounding and other potential biases.

METHODS

Setting

This study was conducted using the Medicaid Analytic eXtract (MAX), which contains administrative billing data for Medicaid enrollees in 46 states and Washington DC from 2000 through 2014 (the most recent data available when the analyses were conducted). Pregnancies identified via delivery codes were linked to live-born infants by a family case number. Medication use and medical histories were reconstructed from billing codes for medical encounters and outpatient pharmacy prescription dispensings. The utility of MAX as a resource for assessing the impact of pregnancy exposures has been previously demonstrated (20, 21), and numerous exposure-outcome associations have been explored in the linked mother-infant cohort (7, 22–26). Use of the MAX data set for these analyses was approved by the Brigham and Women’s Hospital Institutional Review Board.

Study population

Within the mother-infant linked data set, we defined a cohort of women with enrollment in Medicaid, starting ≥3 months prior to the estimated date of the last menstrual period (LMP) and continuing for ≥30 days after delivery, and no supplemental insurance or restricted benefits. Given our focus on the impact of prescription opioid use, we excluded women with ≥1 pharmacy dispensing of naltrexone, naloxone, or buprenorphine, or a charge code for methadone given for opioid maintenance therapy for dependence (27), or ≥1 diagnosis of opioid use disorder or opioid poisoning from 3 months prior to the LMP through delivery. Women who received opioids during the 90 days prior to the LMP but not during the first 20 weeks of pregnancy were also excluded to minimize potential exposure misclassification if women used opioids dispensed prior to pregnancy.

Exposure

Exposure to prescription opioids was assessed using pharmacy claims for dispensings of any opioid medication (Web Table 1, available at https://doi.org/10.1093/aje/kwab132). To reduce misclassification in which women may fill but not use an opioid prescription, the main analyses compared women with ≥2 dispensings with those with 0 dispensings during the relevant exposure window. In subsequent analyses, exposure was categorized based on the number of dispensings received (0, 1, 2–4, 5–9, ≥10) and the quintile of cumulative exposure in morphine milligram equivalent (MME) units (28).

The etiologically relevant window is uncertain given possible effects on placental development early in pregnancy and placental function later. As such, we assessed early pregnancy exposure (opioids dispensed during the first 20 weeks of pregnancy regardless of later exposure), late pregnancy exposure without early pregnancy exposure (opioids dispensed after the 20th week of pregnancy among those without early exposure), and both early and late exposure (≥2 opioid dispensings both before and after the 20th week).

Pregnancies with IPD outcomes are on average shorter than other pregnancies. To avoid differential opportunity for exposure and ensure that comparisons of opioid use included women at the same point in pregnancy (and thus at the same gestational age–specific risk of each outcome), exposure status was updated weekly after the 20th week of gestational age, and analyses were conducted using an Andersen-Gill data structure (29). Patients were considered exposed from the date of the second opioid dispensing through the end of pregnancy. The population at risk included only women who remained pregnant as of the gestational age when exposure was ascertained. The reference group included individuals with no opioid exposure within 3 months prior to pregnancy and within the applicable exposure window (Web Figure 1).

Outcomes

Outcomes were identified using International Classification of Diseases, Ninth Revision, diagnosis codes recorded on the mother’s and baby’s administrative claims for the 30 days after delivery. This approach has been previously validated, with high positive predictive values for preeclampsia, placental abruption, IUGR/SGA, and preterm delivery (21, 30, 31).

Covariates

Demographic characteristics (age, race/ethnicity, US Census region, multiparity, and year) were identified based on Medicaid enrollment files and defined on the date of delivery. Other potential confounders were defined using International Classification of Diseases, Ninth Revision, codes, Current Procedural Terminology codes, Healthcare Common Procedure Coding System codes, and National Drug Codes. For all analyses, the maternal comorbidity index (32), psychiatric conditions (e.g., depression, anxiety, substance use), conditions that may increase risk of IPD (e.g., anemia, diabetes, hypertension), and prior medication use (e.g., antinausea, antibiotic, anticonvulsant, or antidepressant medications) were defined using data from 90 days prior to the LMP through the 20th week of gestation. Indications for opioid use were measured during the same time frame for early pregnancy analyses and were updated every 2 weeks after the 20th week of gestational age for late pregnancy analyses. Covariates were considered present from the date of the first claim carrying a relevant code through the end of pregnancy. Multiple gestations were identified from LMP through delivery + 60 days. A full list of covariates is shown in Web Table 2.

Analyses

Characteristics were described by exposure status and timing. Risk of each outcome was compared between women with and without exposure to opioids during each time frame using Cox proportional hazards models, and hazard ratios (HRs) were calculated with their 95% confidence intervals (CIs), both unadjusted and adjusted for the previously mentioned covariates. Gestational age was treated as the time- scale for the analyses including late pregnancy exposure, with an exposure status indicator that identified new dispensings up to 20 weeks’ gestational age as early exposure and was subsequently updated to identify additional exposures at each week between the 20th week and delivery (for analyses including late pregnancy exposure assessment).

Several sensitivity analyses were performed. Although women with documented opioid use disorders were excluded, we recognize that their ascertainment could be incomplete. To broaden this exclusion criterion, we conducted an analysis excluding individuals with potentially aberrant opioid use as suggested by >120 mg of average daily MME for ≥90 consecutive days or use of >3 pharmacies or >3 prescribers for opioid prescriptions at any point during the pregnancy (33). To determine whether study findings were robust to changes in the exposure definition, we conducted analyses excluding individuals who used opioids in the form of cough syrup. We then considered subgroups of primiparous women, multiparous women, and singleton births. We censored follow-up at preterm delivery given that this could present a competing risk for IPD outcomes. Recognizing that inclusion of data after the LMP when ascertaining baseline comorbidities risks inclusion of conditions newly developed after exposure, we modified the covariate ascertainment to include only data prior to the LMP. For placental abruption in late pregnancy, we ended accrual of new exposure a week prior to delivery to avoid capturing opioid use due to pain associated with early abruption. Finally, we assessed exposure limited to early pregnancy only to focus on the time period relevant for placental development. In this analysis, women were defined as exposed if they received opioids during the first 20 weeks of pregnancy but were censored at the first opioid dispensing after the 20th week of gestation with inverse probability of censoring weights applied.

Quantitative bias analyses were also performed. First, we assessed the strength of an unmeasured confounder that would be needed to explain study results. We then considered the plausible impact of exposure misclassification, expected to occur when women with opioid dispensings did not use them (e.g., diverted or stored them) or women without dispensings accessed opioids in a way that was not recorded in the database (e.g., obtained opioids through nonmedical sources or used medication from an old dispensing). We then considered outcome misclassification, recognizing that while positive predictive values for algorithms used in this study were high, bias is still possible. Finally, we considered residual confounding due to incomplete ascertainment of smoking (see Web Appendix 1).

RESULTS

Of 1,833,871 eligible linked pregnancies, 94,096 (5.1%) had ≥2 opioid dispensings in early pregnancy (irrespective of late exposure), 25,429 (1.6%) had ≥2 dispensings in late but not early pregnancy, and 33,372 (1.8%) had ≥2 dispensings in both early and late pregnancy (Figure 1).

Figure 1.

Formation of the cohort for a study opioid use and ischemic placental disease, Medicaid Analytic eXtract, United States, 2010–2014. LMP, last menstrual period.

Women with opioid use were more likely to be White, multiparous, reside in the southern region of the United States, and have documented tobacco exposure. Pain conditions were more prevalent in the exposed; the most common were abdominal pain and back/neck pain. Opioid-exposed women more often had psychiatric disorders such as depression, and conditions associated with IPD, such as hypertension (Table 1, Web Table 2).

Table 1.

Selected Demographic and Clinical Characteristics, Medicaid Analytic eXtract, United States, 2000–2014

	Early Pregnancy Exposure				Late Pregnancy Exposure Without Early Pregnancy Exposure a				Both Early and Late Pregnancy Exposure a
	No Opioid Dispensings		≥2 Opioid Dispensings		No Opioid Dispensings		≥2 Opioid Dispensings		No Opioid Dispensings		≥2 Opioid Dispensings Early and Late
Characteristic	No.	%	No.	%	No.	%	No.	%	No.	%	No.	%
All	1,562,260	100	94,096	100	1,451,518	100	25,429	100	1,451,518	100	33,372	100
US region of residence
West	435,938	27.9	18,717	19.9	411,017	28.3	5,087	20.0	411,017	28.3	6,629	19.9
Central	440,843	28.2	31,134	33.1	404,224	27.8	9,104	35.8	404,224	27.8	11,506	34.5
South	382,584	24.5	34,808	37.0	345,817	23.8	8,733	34.3	345,817	23.8	12,036	36.1
Northeast	302,895	19.4	9,437	10.0	290,460	20.0	2,505	9.9	290,460	20.0	3,201	9.6
Year of delivery
2000–2004	409,962	26.2	19,483	20.7	380,218	26.2	6,756	26.6	380,218	26.2	5,981	17.9
2005–2009	627,906	40.2	38,692	41.1	581,987	40.1	10,787	42.4	581,987	40.1	13,965	41.8
2010–2014	524,392	33.6	35,921	38.2	489,313	33.7	7,886	31.0	489,313	33.7	13,426	40.2
Race/ethnicity
White	551,280	35.3	58,735	62.4	499,441	34.4	14,188	55.8	499,441	34.4	23,817	71.4
Black	512,675	32.8	21,631	23.0	476,289	32.8	7,041	27.7	476,289	32.8	5,321	15.9
Hispanic	273,802	17.5	6,940	7.4	261,333	18.0	2,052	8.1	261,333	18.0	1,955	5.9
Other	224,503	14.4	6,790	7.2	214,455	14.8	2,148	8.4	214,455	14.8	2,279	6.8
First birth	437,412	28.0	16,846	17.9	410,476	28.3	5,720	22.5	410,476	28.3	4,933	14.8
Multiple gestation	27,932	1.8	2,068	2.2	25,816	1.8	485	1.9	25,816	1.8	714	2.1
Tobacco use	89,975	5.8	24,441	26.0	77,982	5.4	3,595	14.1	77,982	5.4	10,524	31.5
Comorbidity score
0	902,354	57.8	38,886	41.3	844,878	58.2	12,578	49.5	844,878	58.2	12,084	36.2
1	341,783	21.9	23,973	25.5	314,171	21.6	6,468	25.4	314,171	21.6	8,517	25.5
2	171,210	11.0	14,682	15.6	157,633	10.9	3,324	13.1	157,633	10.9	5,671	17.0
≥3	146,913	9.4	16,555	17.6	134,836	9.3	3,059	12.0	134,836	9.3	7,100	21.3
Potential opioid indications
Abdominal pain	381,038	24.4	52,593	55.9	340,204	23.4	10,377	40.8	340,204	23.4	19,618	58.8
Arthritis	142,658	9.1	35,312	37.5	125,400	8.6	4,572	18.0	125,400	8.6	15,303	45.9
Back and neck pain	166,368	10.6	46,476	49.4	144,267	9.9	6,262	24.6	144,267	9.9	21,080	63.2
Cough	52,857	3.4	9,875	10.5	46,700	3.2	1,598	6.3	46,700	3.2	4,142	12.4
Dental pain	40,626	2.6	19,555	20.8	31,928	2.2	2,972	11.7	31,928	2.2	7,792	23.3
Joint pain	57,038	3.7	18,984	20.2	49,751	3.4	2,042	8.0	49,751	3.4	8,613	25.8
Migraine/headache	121,151	7.8	30,780	32.7	105,112	7.2	4,415	17.4	105,112	7.2	13,544	40.6
Orthopedic injury	86,197	5.5	25,372	27.0	74,944	5.2	3,022	11.9	74,944	5.2	10,602	31.8
Other neuropathies	20,761	1.3	10,579	11.2	17,609	1.2	964	3.8	17,609	1.2	5,784	17.3
Other comorbidities
Anemia	131,978	8.4	14,643	15.6	119,685	8.2	3,036	11.9	119,685	8.2	5,329	16.0
Anxiety disorder	58,727	3.8	15,159	16.1	51,962	3.6	2,055	8.1	51,962	3.6	7,347	22.0
Bipolar disorder	26,955	1.7	6,120	6.5	23,640	1.6	992	3.9	23,640	1.6	2,755	8.3
Chronic pulmonary disease	116,904	7.5	18,929	20.1	104,328	7.2	3,209	12.6	104,328	7.2	7,778	23.3
Depression	106,075	6.8	19,094	20.3	94,232	6.5	3,373	13.3	94,232	6.5	8,323	24.9
Diabetes	50,689	3.2	6,449	6.9	46,249	3.2	1,135	4.5	46,249	3.2	2,430	7.3
Hypertension	67,289	4.3	11,051	11.7	60,693	4.2	1,611	6.3	60,693	4.2	4,629	13.9
Infection	176,893	11.3	19,263	20.5	160,293	11.0	4,003	15.7	160,293	11.0	7,032	21.1
Obesity/overweight	72,084	4.6	9,705	10.3	65,398	4.5	1,497	5.9	65,398	4.5	3,668	11.0
Medication use
Antinausea	286,397	18.3	46,727	49.7	252,543	17.4	9,211	36.2	252,543	17.4	18,894	56.6
Antibiotic	812,334	52.0	81,596	86.7	738,750	50.9	17,685	69.5	738,750	50.9	29,343	87.9
Anticonvulsant	22,973	1.5	9,721	10.3	20,147	1.4	830	3.3	20,147	1.4	5,219	15.6
Antidepressant	119,111	7.6	30,556	32.5	103,666	7.1	4,723	18.6	103,666	7.1	14,239	42.7
Antihypertensive	61,897	4.0	12,983	13.8	54,862	3.8	1,923	7.6	54,862	3.8	5,820	17.4
Benzodiazepines	27,980	1.8	17,942	19.1	23,996	1.7	1,403	5.5	23,996	1.7	9,581	28.7
Corticosteroids	231,054	14.8	31,048	33.0	208,961	14.4	5,521	21.7	208,961	14.4	12,465	37.4
Other hypnotic agents	86,008	5.5	19,455	20.7	74,599	5.1	3,378	13.3	74,599	5.1	9,236	27.7
Suspected teratogens	207,572	13.3	29,133	31.0	185,859	12.8	5,495	21.6	185,859	12.8	11,239	33.7

^a Potential opioid indications were updated biweekly using gestational age as a timescale. Proportions in this table show covariate status at the end of pregnancy.

Early pregnancy exposure

Preterm delivery occurred in 14.6% of those with ≥2 opioid dispensings prior to the 20th week of gestation and 9.8% of those without early exposure, producing an unadjusted HR of 1.49 (95% CI: 1.46, 1.51). The HR declined to 1.21 (95% CI: 1.18, 1.23) after adjusting for covariates listed in Web Table 2. IUGR/SGA, occurring in 4.6% of those with ≥2 dispensings and 3.0% of the unexposed, had an HR of 1.50 (95% CI: 1.46, 1.55) and an adjusted HR (aHR) of 1.13 (1.09, 1.17). Preeclampsia, occurring in 3.9% of those with ≥2 opioid dispensings and 3.2% of the unexposed, had an HR of 1.23 (95% CI: 1.19, 1.27) and an aHR of 0.95 (95% CI: 0.91, 0.98). Placental abruption, occurring in 1.5% of those with ≥2 opioid dispensings and 0.9% of the unexposed, had an HR of 1.63 (95% CI: 1.54, 1.72) and an aHR of 1.34 (95% CI: 1.26, 1.43) (Table 2).

Table 2.

Associations Between Opioid Use and Ischemic Placental Disease Outcomes by Timing of Exposure, Medicaid Analytic eXtract, United States, 2000–2014

Outcome and Exposure Window	No. of Events in Unexposed		No. of Events in Exposed		Unadjusted		Adjusted
	No.	%	No.	%	HR	95% CI	HR	95% CI
Preterm delivery
Early exposurea	153,697	9.8	13,754	14.6	1.49	1.46, 1.51	1.21	1.18, 1.23
Late exposure without early exposureb	143,012	9.9	2,499	9.8	1.31	1.26, 1.37	1.18	1.14, 1.23
Both early and late exposurec	143,012	9.9	5,177	5.5	1.77	1.72, 1.82	1.37	1.33, 1.42
IUGR/SGA
Early exposure	47,568	3.0	4,303	4.6	1.50	1.46, 1.55	1.13	1.09, 1.17
Late exposure without early exposure	43,889	3.0	911	3.6	1.28	1.20, 1.37	1.11	1.04, 1.19
Both early and late exposure	43,889	3.0	1,730	5.2	1.86	1.77, 1.95	1.26	1.19, 1.33
Preeclampsia
Early exposure	49,332	3.2	3,653	3.9	1.23	1.19, 1.27	0.95	0.91, 0.98
Late exposure without early exposure	45,803	3.2	803	3.2	1.09	1.02, 1.17	0.96	0.89, 1.03
Both early and late exposure	45,803	3.2	1,391	4.2	1.44	1.36, 1.52	0.99	0.93, 1.05
Placental abruption
Early exposure	14,422	0.9	1,415	1.5	1.63	1.54, 1.72	1.34	1.26, 1.43
Late exposure without early exposure	13,310	0.9	267	1.1	1.38	1.22, 1.55	1.24	1.10, 1.41
Both early and late exposure	13,310	0.9	567	1.7	2.08	1.92, 2.27	1.62	1.47, 1.78

Abbreviations: CI, confidence interval; HR, hazard ratio; IUGR, intrauterine growth restriction; LMP, last menstrual period; SGA, small for gestational age.

^a 1,562,260 pregnancies had no opioid dispensings from 90 days prior to the LMP through 140 days after the LMP; 94,096 pregnancies had ≥2 opioid dispensings within 140 days after the LMP.

^b 1,451,518 pregnancies had no opioid dispensings from 90 days prior to the LMP through delivery; 25,429 pregnancies had no opioid dispensings prior to 140 days after the LMP and ≥2 opioid dispensings between the 140th day after LMP and delivery. Analyses were conducted using a time-varying exposure.

^c 1,451,518 pregnancies had no opioid dispensings from 90 days prior to the LMP through delivery; 33,372 pregnancies had ≥2 opioid dispensings within 140 days after the LMP and ≥2 opioid dispensings between the 140th day after LMP and delivery. Analyses were conducted using a time-varying exposure.

Estimates increased with the total number of dispensings or MMEs for preterm, IUGR/SGA, and placental abruption (Web Figure 2, Table 3). The increase was most marked for placental abruption (for 1 dispensing, aHR = 1.14, 95% CI: 1.08, 1.20; for 2–4 dispensings, aHR = 1.30, 95% CI: 1.21, 1.39; for 5–9 dispensings, aHR = 1.48, 95% CI: 1.30, 1.69; and for ≥10 dispensings, aHR = 2.10, 95% CI: 1.70, 2.59). There was no trend across categories of cumulative exposure or dispensing count for preeclampsia after adjustment (Web Figure 2, Table 3).

Table 3.

Associations Between Opioid Use and Ischemic Placental Disease Outcomes by Timing and Cumulative Exposure Quartile, Medicaid Analytic eXtract, United States, 2000–2014

Cumulative Opioid Exposure by Timing of Use	Preterm Delivery		IUGR/SGA		Preeclampsia		Placental Abruption
	aHR	95% CI	aHR	95% CI	aHR	95% CI	aHR	95% CI
Early exposure
<180 vs. 0 MME	1.11	1.09, 1.12	1.06	1.03, 1.09	0.97	0.95, 1.00	1.12	1.06, 1.18
180–269 vs. 0 MME	1.14	1.11, 1.18	1.10	1.04, 1.17	0.96	0.90, 1.02	1.29	1.17, 1.43
270–443 vs. 0 MME	1.14	1.10, 1.18	1.04	0.97, 1.11	0.90	0.84, 0.97	1.29	1.15, 1.44
444–1,044 vs. 0 MME	1.19	1.14, 1.23	1.10	1.03, 1.18	1.01	0.94, 1.08	1.40	1.25, 1.57
≥1,045 vs. 0 MME	1.29	1.25, 1.34	1.29	1.21, 1.37	0.96	0.89, 1.02	1.47	1.31, 1.64
Late exposure without early exposure
<75 vs. 0 MME	1.05	0.80, 1.38	1.02	0.65, 1.61	1.36	0.94, 1.95	0.61	0.20, 1.90
75–99 vs. 0 MME	1.05	0.81, 1.36	0.94	0.61, 1.46	0.83	0.52, 1.32	1.10	0.49, 2.45
100–149 vs. 0 MME	1.06	0.94, 1.19	1.00	0.82, 1.21	0.88	0.72, 1.08	1.16	0.82, 1.65
150–329 vs. 0 MME	1.10	1.04, 1.16	1.09	0.99, 1.19	0.90	0.81, 1.00	1.22	1.02, 1.44
≥330 vs. 0 MME	1.35	1.27, 1.45	1.13	1.00, 1.26	1.03	0.91, 1.16	1.34	1.09, 1.65
Both early and late exposure
<180 vs. 0 MME	1.03	0.70, 1.50	0.82	0.39, 1.73	1.35	0.80, 2.29	0.44	0.06, 3.14
180–271.24 vs. 0 MME	0.86	0.62, 1.20	0.85	0.50, 1.43	0.85	0.49, 1.47	1.28	0.53, 3.08
271.25–449 vs. 0 MME	1.17	1.04, 1.31	0.99	0.81, 1.22	0.92	0.73, 1.15	1.63	1.18, 2.23
450–1,204 vs. 0 MME	1.33	1.27, 1.41	1.16	1.05, 1.27	1.04	0.94, 1.15	1.59	1.35, 1.87
≥1,205 vs. 0 MME	1.43	1.37, 1.49	1.35	1.25, 1.45	0.99	0.92, 1.08	1.63	1.43, 1.85

Abbreviations: aHR, adjusted hazard ratio; CI, confidence interval; IUGR, intrauterine growth restriction; MME, morphine milligram equivalents; SGA, small for gestational age.

Sensitivity analyses excluding those with aberrant opioid use produced lower aHR estimates for preterm delivery, IUGR/SGA, and preeclampsia but did not meaningfully change the estimate for placental abruption. Across all outcomes, there was no meaningful difference in the estimated HR when excluding women who used cough syrup, or in singleton, primiparous, or multiparous subgroups. Revising the covariate ascertainment window to include only data captured prior to the LMP and censoring at preterm delivery also had only minimal impact. Results of analyses that considered the impact of early pregnancy exposure without late pregnancy exposure by censoring at the first opioid dispensing after 20 weeks of gestation were similar to analyses that ignored late pregnancy exposure (Web Figure 2).

Results of bias analysis show that potentially differential exposure misclassification could impart considerable uncertainty in the magnitude of association, with an expected bias away from the null. Small increased risks are still plausible for preterm birth, IUGR/SGA, and placental abruption. Outcome misclassification had only a small impact, creating the expectation of bias toward the null. The impact of misclassified smoking information depends on the expected relationship between smoking and each outcome; however, corrected estimates were closer to the null than observed estimates for preterm delivery and placental abruption. Residual confounding due to smoking may completely explain the observed association between opioid use and IUGR/SGA but is not likely to have masked an increased risk of preeclampsia (Web Appendix 1, Web Table 3, and Web Figures 3 and 4).

Late pregnancy exposure without early pregnancy exposure

Compared with estimates for early exposure, aHRs were similar when considering late pregnancy exposure among those with no early pregnancy exposure. There were slight increases in the aHR for preterm delivery (aHR = 1.18, 95% CI: 1.14, 1.23), IUGR (aHR = 1.11, 95% CI: 1.04, 1.19), and placental abruption (aHR = 1.24, 95% CI: 1.10, 1.41). No association was observed for preeclampsia (aHR = 0.96, 95% CI: 0.89, 1.03) (Table 2). The association for placental abruption was slightly stronger (aHR = 1.30, 95% CI: 1.50, 1.47) when stopping exposure accrual 1 week prior to delivery. Patterns observed in sensitivity analyses mirrored those seen for early pregnancy exposure, but few mothers accrued ≥10 dispensings of opioids over the second half of pregnancy. Exclusion of women with aberrant opioid use patterns resulted in minimal changes (Web Figure 2). Residual confounding due to smoking may plausibly explain the observed association for IUGR but would neither explain the increased risk of preterm delivery and placental abruption nor mask an association with preeclampsia (Web Figure 4).

Early and late pregnancy exposure

Women using opioids during both early and late pregnancy were at higher risk for preterm delivery (aHR = 1.37, 95% CI: 1.33, 1.42), IUGR (aHR = 1.26, 95% CI: 1.19, 1.33), and placental abruption (aHR = 1.62, 95% CI: 1.47, 1.78). Again, no association was observed for preeclampsia (aHR = 0.99, 95% CI: 0.93, 1.05) (Table 2). Increases across categories of MME and opioid dispensings were more pronounced (Web Figure 1, Table 3). Patterns in sensitivity analyses were generally similar to those observed for early pregnancy exposure, noting that exclusion of those with aberrant opioid use patterns produced a higher estimate for placental abruption (aHR = 1.74, 95% CI: 1.24, 2.44) rather than the lower estimates seen elsewhere (Web Figure 2). The impact of misclassification and residual confounding was similar to what was observed for late pregnancy exposure (Web Figure 4).

DISCUSSION

Results of this study suggest that prescription opioid use in pregnancy modestly increases the risk of placental abruption, especially at higher levels of cumulative exposure. Although the direction of the association was the same for all exposure windows considered, exposure both early and late in pregnancy had the strongest association. Smaller increases in risk were observed for preterm delivery and potentially IUGR/SGA; however, no positive association was observed between opioid use and preeclampsia.

Complications in pregnant women with opioid use disorders include IPD among multiple adverse effects (14–19). However, it is not clear whether observed associations are explained by lifestyle factors and related exposures; few well-controlled studies are available. This study provides information on therapeutic opioid use outside the context of opioid use disorders, a large number of pregnancies to obtain precise estimates for rare outcomes, and improved control of many key confounders.

A recent Swedish register study found similar associations between opioid use and preterm birth and SGA, with a small increased risk for preterm delivery (adjusted odds ratio = 1.38, 95% CI: 1.31, 1.45) and no association with SGA (1.02, 95% CI: 0.93, 1.10). Risks increased for both outcomes with exposure across multiple trimesters (34). Our findings support this work, and we expand it to include additional outcomes and subgroup analyses.

Although preterm delivery, IUGR/SGA, preeclampsia, and placental abruption have all been considered in this analysis, their risk factors and associated morbidity vary. Each condition is multifactorial in etiology (9), and the impact of opioid use could potentially affect some but not all pathways leading to their development. Placental abruption is often symptomatic of chronic placental dysfunction. The resulting decrease in surface area for oxygen exchange and nutrient supply can lead to decreased fetal growth, and most studies show 40%–60% of placental abruption cases delivering preterm, linking the outcomes in many instances (35). The shared relationship of these outcomes with placental malperfusion can also be seen in changes in placental infarction and tissue architecture as a response to chronic insufficiency, with accelerated villous maturation acting as a histological marker of insufficiency seen across cases of preeclampsia, preterm birth, and IUGR (36). Preeclampsia appears to have distinct subtypes with variable levels of apparent placental dysfunction; however, placental bed disorders are seen in the vast majority of cases. Whereas incomplete or absent conversion of spiral arteries has also been described for other outcomes of IPD, it has been accompanied by the additional presence of obstructive lesions in IUGR and placental abruption cases (37), suggesting differences in the mechanism leading to these conditions. Future analyses should consider both interconnection of IPD outcomes and mediation of the association between opioid use in pregnancy and preterm delivery via IPD.

The stronger aHRs for exposure both early and late in pregnancy versus early pregnancy alone could be interpreted in different ways. One possibility is that different mechanisms leading to placental insufficiency are influenced by accumulated exposure across early and late pregnancy. However, sustained use may reflect chronic pain, with estimates of the HR in this population inflated by residual confounding by indication or pain-associated lifestyle factors. One would not, however, expect to find a null association for preeclampsia if the observed increases in aHR for placental abruption, preterm delivery, and IUGR/SGA were only attributable to residual confounding. While it is possible that residual confounding could apply only to outcomes other than preeclampsia, this would be unusual given the similarity of the magnitude and direction of confounding seen in models that adjusted for the 83 covariates shown in Web Table 2.

These findings should be viewed in light of several key limitations. Claims-based studies infer absence of evidence as evidence of absence with respect to chronic conditions. For example, a patient’s preexisting diabetes may be undocumented if the patient does not seek care during the baseline period. This was addressed by extending the baseline period through early pregnancy (when one would expect a first prenatal visit to occur) rather than ending baseline covariate ascertainment at the LMP. Although this could risk identification of conditions that develop following early pregnancy exposure as baseline covariates, this is unlikely given the chronic nature of these conditions and sensitivity analysis results.

Areas where missed diagnoses are particularly problematic include cocaine and methamphetamine use, opioid use disorders, and smoking status. These conditions may be actively hidden from health-care providers, and health-care providers may not document opioid use disorders or substance use to avoid potential legal consequences for the mother. Underascertainment of smoking, a potentially important confounder, is likely associated with misclassification of exposure via use of illicit opioids. This should be partially offset by adjustment for a large number of proxy variables (e.g., chronic conditions related to smoking and/or opioid use disorders). Even with the relatively extreme differences in smoking by exposure group that were included in bias analyses, some increase in the risk of preterm delivery and placental abruption is expected to remain. Unmeasured confounding by parameters not associated with those included in our models may also present a source of bias; however, the prevalence and strength of association of these unknown factors would need to exceed the extreme assumptions that we defined for smoking to meaningfully alter the interpretation of our results.

Misclassification of exposure could be introduced through several routes. Claims data do not capture LMP, and we assume that medications dispensed from a pharmacy are used rather than stored, lost, or sold. Requiring ≥2 opioid dispensings increases the likelihood that medication is used, but medication could be taken later than expected. Further, we cannot exclude use of illicit opioids in the unexposed. While the impact of misclassification was explored, these analyses rely on strong assumptions. Wide simulation intervals in these analyses show the extent to which results are uncertain but suggest that misclassification of exposure is unlikely to fully explain observed associations.

Within Medicaid patients, requiring eligibility prior to the LMP may limit the population to those with more consistent access to medical care and/or prior pregnancies. However, past assessments comparing characteristics of patients meeting and not meeting enrollment criteria have not suggested that the populations differ (38).

These limitations to the data source must also be weighed against key advantages. The MAX data set includes a population large enough to support appropriately powered analyses of the rare outcomes under study, and nearly 50% of pregnancies in the United States are insured by Medicaid (39). Given common use of opioids in this group, a better understanding of risks is warranted.

In summary, this study suggests that opioids have little role in development of preeclampsia but may slightly increase risk for preterm birth and IUGR/SGA. Placental abruption was the most strongly associated outcome, especially with exposure in both early and late pregnancy. Benefits of pain management during pregnancy should be weighed against these considerations.