Trajectories of Prescription Opioid Utilization During Pregnancy Among Prepregnancy Chronic Users and Risk of Neonatal Opioid Withdrawal Syndrome

Abstract

Little is known about the impact of dose, duration, and timing of prenatal prescription opioid exposure on the risk of neonatal opioid withdrawal syndrome (NOWS). Using a cohort of 18,869 prepregnancy chronic opioid users nested within the 2000–2014 Medicaid Analytic eXtract, we assessed average opioid dosage within biweekly gestational age intervals, created group-based trajectory models, and evaluated the association between trajectory groups and NOWS risk. Women were grouped into 6 distinct opioid use trajectories which, based on observed patterns, were categorized as 1) continuous very low-dose use, 2) continuous low-dose use, 3) initial moderate-dose use with a gradual decrease to very low-dose/no use, 4) initial high-dose use with a gradual decrease to very low-dose use, 5) continuous moderate-dose use, and 6) continuous high-dose use. Absolute risk of NOWS per 1,000 infants was 7.7 for group 1 (reference group), 28.8 for group 2 (relative risk (RR) = 3.7, 95% confidence interval (CI): 2.8, 5.0), 16.5 for group 3 (RR = 2.1, 95% CI: 1.5, 3.1), 64.9 for group 4 (RR = 8.4, 95% CI: 5.6, 12.6), 77.3 for group 5 (RR = 10.0, 95% CI: 7.5, 13.5), and 172.4 for group 6 (RR = 22.4, 95% CI: 16.1, 31.2). Trajectory models—which capture information on dose, duration, and timing of exposure—are useful for gaining insight into clinically relevant groupings to evaluate the risk of prenatal opioid exposure.

Abbreviations

AR

absolute risk

aveMME

average daily morphine milligram equivalent

CI

confidence interval

GA

gestational age

GBTM

group-based trajectory model

ICD-9-CM

International Classification of Diseases, Ninth Revision, Clinical Modification

MME

morphine milligram equivalent

NOWS

neonatal opioid withdrawal syndrome

As women undergo physiological changes during pregnancy, such as loosening of the ligaments and weight gain, many experience pain, ranging from slight discomfort to severe pain (1). Pregnancy can further exacerbate preexisting pain conditions (2). Opioids for analgesic treatment are among the most frequently prescribed medications in pregnancy in the United States, with previous estimates suggesting that approximately 1 in 5 publicly insured women and 1 in 7 commercially insured women have at least 1 opioid medication dispensed during pregnancy (3–6). In addition, the majority of women who are chronically exposed to prescription opioids prior to pregnancy continue using opioids during pregnancy (6). Such widespread use is concerning, given that opioids have been associated with adverse pregnancy outcomes, including congenital malformations, placenta-mediated complications, and neonatal opioid withdrawal syndrome (NOWS) (7–9).

NOWS is a group of serious conditions in the newborn caused by in utero exposure to opioid agonists (including heroin, prescription opioids, or opioid agonist therapy with methadone or buprenorphine). A study of publicly insured US women found a risk of neonatal withdrawal of 5.9 per 1,000 pregnancies involving exposure to prescription opioids at any time during pregnancy (10); there was increasing risk with higher dose, longer exposure duration, and additional risk factors.

Studies on the impact of prescription opioid use on neonatal outcomes typically focus on pregnancy exposure that is classified dichotomously, based on days of opioid supply or cumulative dose throughout pregnancy. This removes important information on dose changes, intensity of use, and gap periods. Given that NOWS is due to opioid withdrawal, it is expected that risk will be influenced by the timing of use and dose, but this has not been well characterized. Further, for women who use opioids chronically prior to pregnancy for chronic pain management—who are probably the group at highest risk for having an infant with NOWS—there is little information for guiding them on how changing their prescription opioid-use patterns (e.g., weaning to a lower opioid dose) may modulate the risk of NOWS in their offspring.

Recently, longitudinal trajectory analyses have been increasingly used to assess drug utilization in pregnancy (11–14). These methods classify individuals with similar exposure patterns over time into groups, allowing for the comparison of more homogeneous exposure groups. Opioid exposure during pregnancy, particularly among prepregnancy chronic users, is a good candidate for the application of such methods, as it is important to understand how dose, duration, and timing—measures that can be captured in trajectories—together affect the risk of NOWS. Therefore, our objectives in this study were to: 1) characterize patterns of prescription opioid use during pregnancy among prepregnancy chronic opioid users using group-based trajectory models (GBTMs); 2) evaluate the association between individual trajectory groups and the risk of NOWS; and 3) compare GBTM results with methods based on traditional exposure definitions, using nationwide data from Medicaid insurance beneficiaries.

METHODS

Study cohort

We used a cohort of mothers linked to their liveborn infants, nested within the Medicaid Analytic eXtract for 2000–2014 (the most recent Medicaid data released by the Centers for Medicare and Medicaid Services at the time of study conduct), which comprises the health-care utilization records of Medicaid beneficiaries nationwide. As a joint state and federal health insurance program for low-income individuals, Medicaid covers approximately 50% of all pregnancies in the United States (15, 16). Medicaid Analytic eXtract data include information on patient demographic characteristics, insurance enrollment, diagnostic and procedure claims for all inpatient, outpatient, and emergency room visits, and outpatient prescription dispensings. The development of the mother-infant linked Medicaid cohort has been described in detail elsewhere (16), and the cohort has been used extensively for studies on medication safety in pregnancy (10, 17–22).

To ensure accurate capture of relevant diagnoses, procedures, and medication exposures, the included women (who were aged 12–55 years) were required to have continuous Medicaid coverage from 3 months before the date of the last menstrual period to 1 month after delivery. The date of the last menstrual period and gestational age (GA) at birth were estimated using a previously validated algorithm based on diagnostic codes for preterm birth combined with the delivery date (23). Deliveries without preterm codes were considered term deliveries, and a gestational length of 270 days was assigned to these pregnancies. Infants were required to have coverage for at least 1 month after birth, unless they died before then, in which case a shorter period was allowed. Mothers and infants with restricted benefits, private health insurance, or certain capitated managed-care programs that might underreport claims to the Medicaid Analytic eXtract were excluded from the cohort (16), as were beneficiaries from Texas (because of data quality issues). The cohort was further restricted to women with prepregnancy chronic opioid use, since 1) this is probably the group at highest risk for having an infant with NOWS and 2) a goal of the study was to help define patterns of management (e.g., weaning to a lower opioid dose) that may modulate risk of NOWS. Prepregnancy chronic opioid use was defined as at least 1 prescription opioid dispensation in each of the 3 months prior to pregnancy and 1 or more additional dispensings during pregnancy. In sensitivity analyses, we further restricted the cohort to women with ≥2 additional dispensings during pregnancy. Because the aim of the study was to evaluate the risk of NOWS in infants born to mothers with chronic opioid use for pain, we attempted to minimize exposure misclassification due to use of illicit opioids not captured in pharmacy records by excluding women with diagnoses indicating an opioid-use disorder or women who were treated with opioid agonist therapy at any time during the 3 months before pregnancy or during pregnancy. Lastly, to guarantee a uniform window of observations during pregnancy (i.e., the same gestational length) and because term infants are more likely to develop NOWS/more severe NOWS than preterm infants (24, 25), we limited our analysis to pregnancies ending in term births.

Exposure

Opioid agents considered in our analysis included butorphanol, codeine, dihydrocodeine, fentanyl, hydrocodone, hydromorphone, levorphanol, meperidine, methadone (by prescription, not dispensed for opioid-use disorder), morphine, oxycodone, oxymorphone, pentazocine, propoxyphene, tapentadol, and tramadol.

For the creation of GBTMs, exposure to prescription opioids during pregnancy was assessed within biweekly GA intervals (19 different intervals). For each prescription filled during pregnancy, we calculated the corresponding morphine milligram equivalents (MMEs) based on quantity, dosage dispensed, and the morphine conversion factor (as provided by the Centers for Disease Control and Prevention (26)). We then calculated the average daily MME (aveMME) dose separately for each GA interval, by dividing the sum of MMEs for each prescription active during the respective interval by the duration of the assessment period. To exclude women with implausibly high prescription opioid doses (e.g., due to data entry errors), we restricted our cohort to women with an aveMME less than or equal to 200 in each GA interval. Women were categorized into groups with similar opioid exposure patterns during pregnancy using GBTMs, as described in detail below.

In addition, to compare these trajectory groups with more “traditional” definitions of opioid exposure during pregnancy, especially in the context of NOWS, we also considered alternative exposure groups (10). First, we estimated the cumulative time of opioid exposure by summing the number of days’ supply for each filled prescription during pregnancy, which we then dichotomized into long-term use (≥30 days’ supply) versus short-term use (<30 days’ supply). Second, we categorized exposure based on the timing of opioid dispensing during pregnancy by comparing women with 1 or more prescriptions in the third trimester, irrespective of earlier exposure, to those with at least 1 prescription during the first or second trimester but not during the third trimester. Lastly, we created cumulative opioid dose tertiles (measured in MMEs) based on all opioid prescriptions filled during pregnancy (cumulative tertiles 1–3).

Outcome

The outcome of interest was NOWS, which was identified on the basis of at least 1 instance of International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM) diagnostic code 779.5 (“Drug withdrawal syndrome in newborn”) in the infant or maternal records (because infant conditions are sometimes recorded in the maternal claims before the infant’s eligibility is processed) within the first 30 days after birth. While ICD-9-CM code 779.5 does not provide sufficient information to distinguish NOWS from neonatal withdrawal syndrome due to nonopioid substance use, the assumption was that most children in our cohort with such a code would have NOWS, as they were all born to mothers with prepregnancy chronic opioid use and opioid exposure during pregnancy and because neonatal withdrawal syndrome is most often caused by prenatal opioid exposure. The identification of neonatal withdrawal syndrome using administrative data has previously been validated, with high positive predictive values of 91%–100% (27, 28) and sensitivities of 84%–95% (29, 30). As a secondary outcome, we assessed severe cases of NOWS, defined on the basis of ≥1 code for NOWS and the presence of at least 1 of the following complications, which may be markers for NOWS severity: admission to the neonatal intensive care unit, feeding difficulties (ICD-9-CM code 779.3x), respiratory symptoms (ICD-9-CM codes 769–770.xx), or signs of seizures (ICD-9-CM codes 779.0x and 780.3x) within the first 30 days after birth (10, 28).

Covariates

We assessed several patient characteristics which are potential predictors of opioid utilization during pregnancy and which may increase the likelihood or severity of NOWS (or can directly lead to withdrawal symptoms in the newborn even in the absence of opioid exposure). These included maternal demographic characteristics (age, race/ethnicity, and US Census region), select acute and chronic pain conditions, mental-health–related conditions such as anxiety and depression, nonopioid substance use/misuse (including tobacco use and alcohol dependence, which could worsen NOWS-related symptoms (31–35)), and third-trimester exposure to other psychotropic medications such as selective serotonin reuptake inhibitors, benzodiazepines, and anticonvulsants, which are considered independent potential risk factors for neonatal drug withdrawal syndrome and its severity (36–40). While third-trimester medication use represents postbaseline characteristics, these covariates may serve as proxies for preexisting conditions. For details on the individual characteristics and definitions used, see Table 1 and Web Table 1 (available at https://doi.org/10.1093/aje/kwab249).

Table 1.

Characteristics of 18,869 Women Chronically Exposed to Opioids Prior to Pregnancy Who Had at Least 1 Dispensed Opioid Prescription During Pregnancy, by Pregnancy Opioid-Use Trajectory Group, Medicaid Analytic eXtract, 2000–2014^a

	Trajectory Group
Maternal Covariate	Full Cohort (n = 18,869)		Group 1 (n = 7,660) (Continuously Very Low Dose)		Group 2 (n = 5,379) (Continuously Low Dose)		Group 3 (n = 2,903) (Initially Moderate Dose With Decreaseto Very Low Dose/No Use)		Group 4 (n = 555) (Initially High Dose With Decrease to Very Low Dose)		Group 5 (n = 1,966) (Continuously Moderate Dose)		Group 6 (n = 406) (Continuously High Dose)
	No.	%	No.	%	No.	%	No.	%	No.	%	No.	%	No.	%
Maternal age at delivery, yearsb	27.3 (5.4)		26.0 (5.2)		27.7 (5.2)		27.9 (5.5)		29.0 (5.1)		29.0 (5.3)		30.3 (5.1)
Maternal race/ethnicity
White	13,680	72.5	5,422	70.8	3,959	73.6	2,102	72.4	423	76.2	1,460	74.3	314	77.3
Black/African-American	2,823	15.0	1,272	16.6	775	14.4	429	14.8	61	11.0	246	12.5	40	9.9
Hispanic/Latino	1,105	5.9	462	6.0	286	5.3	195	6.7	39	7.0	105	5.3	18	4.4
Other/unknown	1,261	6.7	504	6.6	359	6.7	177	6.1	32	5.8	155	7.9	34	8.4
US Census region
Midwest	6,224	33.0	2,521	32.9	1,813	33.7	875	30.1	180	32.4	682	34.7	153	37.7
Northeast	2,136	11.3	767	10.0	499	9.3	400	13.8	89	16.0	295	15.0	86	21.2
South	7,234	38.3	3,031	39.6	2,146	39.9	1,130	38.9	193	34.8	630	32.0	104	25.6
West	3,275	17.4	1,341	17.5	921	17.1	498	17.2	93	16.8	359	18.3	63	15.5
Indication for opioid usec
Abdominal pain	8,963	47.5	3,911	51.1	2,589	48.1	1,253	43.2	222	40.0	821	41.8	167	41.1
Arthritis/arthropathy, musculoskeletal pain	6,579	34.9	2,360	30.8	2,128	39.6	942	32.4	195	35.1	789	40.1	165	40.6
Back/neck pain	10,578	56.1	3,428	44.8	3,497	65.0	1,649	56.8	364	65.6	1,359	69.1	281	69.2
Dental problems	2,680	14.2	1,095	14.3	970	18.0	277	9.5	66	11.9	230	11.7	42	10.3
Infection	2,856	15.1	1,211	15.8	807	15.0	424	14.6	77	13.9	290	14.8	47	11.6
Joint pain	3,354	17.8	1,128	14.7	1,145	21.3	473	16.3	115	20.7	406	20.7	87	21.4
Migraine/headache	5,297	28.1	2,043	26.7	1,844	34.3	662	22.8	137	24.7	516	26.2	95	23.4
Orthopedic injury	3,955	21.0	1,573	20.5	1,319	24.5	492	16.9	100	18.0	402	20.4	69	17.0
Other indicationsd	7,117	37.7	2,219	29.0	2,379	44.2	1,023	35.2	266	47.9	999	50.8	231	56.9
Maternal mental-health–related conditionse
Anxiety	4,303	22.8	1,528	19.9	1,424	26.5	646	22.3	122	22.0	494	25.1	89	21.9
Depression	4,522	24.0	1,753	22.9	1,404	26.1	646	22.3	129	23.2	497	25.3	93	22.9
Bipolar disorder	1,525	8.1	630	8.2	457	8.5	188	6.5	56	10.1	159	8.1	35	8.6
Sleep disorder	1,132	6.0	390	5.1	362	6.7	185	6.4	42	7.6	129	6.6	24	5.9
Other mental-health–related variablesf	357	1.9	149	1.9	115	2.1	51	1.8	<11g		30	1.5	<11g
Nonopioid substance abuse or dependencec
Alcohol drinking	225	1.2	86	1.1	71	1.3	34	1.2	<11g		23	1.2	<11g
Tobacco use	4,517	23.9	1,693	22.1	1,429	26.6	691	23.8	125	22.5	472	24.0	107	26.4
Other substance abuseh	286	1.5	93	1.2	88	1.6	52	1.8	<11g		35	1.8	<11g
Use of nonopioid psychotropic prescription medicationi
Selective serotonin reuptake inhibitors	1,946	10.3	633	8.3	706	13.1	232	8.0	58	10.5	265	13.5	52	12.8
Benzodiazepines	1,712	9.1	260	3.4	762	14.2	114	3.9	57	10.3	426	21.7	93	22.9
Anticonvulsants	689	3.7	148	1.9	267	5.0	67	2.3	21	3.8	151	7.7	35	8.6
Barbiturates	743	3.9	261	3.4	276	5.1	97	3.3	14	2.5	80	4.1	15	3.7
Other psychotropic medicationsj	3,188	16.9	1,071	14.0	1,160	21.6	337	11.6	89	16.0	435	22.1	96	23.6

^a Percentages may not total 100 because of rounding.

^b Values are expressed as mean (standard deviation).

^c Assessed throughout pregnancy.

^d Other indications included other neuropathies, cough, pain not elsewhere classified, fibromyalgia, surgery, renal calculus, sciatica, osteoarthritis, generalized pain, rheumatoid arthritis, malignancy, sickle cell disease, pain related to psychological factors, chronic pancreatitis, gout, complex regional pain syndrome, trigeminal neuralgia, sarcoidosis, peripheral neuropathy, postherpetic neuralgia, and central pain.

^e Assessed from 3 months before the date of the last menstrual period to the end of pregnancy.

^f Other mental-health–related variables included psychosis, schizophrenia, and personality disorder.

^g Numbers were suppressed in accord with the Centers for Medicare and Medicaid Services cell size suppression policy.

^h Other substance abuse included cocaine, marijuana, hallucinogens, amphetamines, and sedative hypnotics.

ⁱ Assessed during the third trimester of pregnancy.

^j Other psychotropic medications included tricyclic antidepressants, serotonin-norepinephrine reuptake inhibitors, anxiolytics, amphetamines, other hypnotics, and atypical/typical antipsychotics.

Statistical analysis

Using GBTMs (created with the add-on package “proc traj” (41) in SAS, version 9.4 (SAS Institute, Inc., Cary, North Carolina))—which allow for simultaneous estimation of group-assignment probabilities and longitudinal trajectories (42)—women were categorized into groups of similar individual trajectories based on their aveMME within biweekly GA intervals. Group trajectories were modeled using the censored normal distribution, and models were created separately for 2–8 groups. For each model, time was modeled with third-order (cubic) polynomials in each of the groups, the assumption being that with 19 time intervals, an order of 3 should be sufficiently flexible to capture meaningful changes in opioid exposure over time. Evidence for polynomial overfitting was visually assessed by comparing estimated and observed group trajectories.

There is no formal test for determining the optimal number of groups. Therefore, selection of the final number was based on combined assessment of 1) the Bayesian information criterion (with least negative values indicating better model fit (43)), 2) the number of pregnancies included in the smallest group (44), and 3) the clinical meaningfulness of the trajectories. The fitness of the final selected model was evaluated using the diagnostic criteria described by Nagin (45). For more details on the selection of the final model and assessment of model fitness, see Web Appendices 1 and 2, Web Tables 2 and 3, and Web Figure 1.

To visually assess how well opioid exposure patterns of individual pregnancies matched the group’s average regression line, we created spaghetti plots for 200 randomly selected pregnancies separately for each group. If most of the randomly selected pregnancies showed patterns close to the average, this would suggest that the patterns of most pregnancies could be explained well by the groups’ average (43).

Patient characteristics were compared between each of the final selected trajectory groups and alternative exposure groups (≥30 days’ supply vs. <30 days’ supply; third-trimester exposure vs. first- or second-trimester exposure; cumulative tertiles 1–3).

Absolute risks (ARs) of NOWS and severe NOWS were calculated separately for each trajectory group and each alternative exposure definition, and unadjusted relative risks and 95% confidence intervals (CIs) were determined (with the reference groups being the lowest dose trajectory group, <30 days’ supply, first- or second-trimester exposure, and tertile 1, respectively). Using logistic regression models for each of these exposure-outcome contrasts, we assessed and compared corresponding receiver operating characteristic curves and concordance statistics.

To further evaluate the potential benefits of using clustering techniques like GBTMs over more traditional approaches with regard to assessing the associations between prescription opioid use and NOWS, we additionally fitted a logistic regression model including all alternative exposure groups described above as independent variables, and compared the receiver operating characteristic curves and concordance statistic between this model and the trajectory group model. To ensure a fair comparison between models with the same number of degrees of freedom, we defined a trajectory group model based on 5 groups for this purpose; this resulted in 4 degrees of freedom for both models.

To account for potential predictors of opioid utilization during pregnancy and other risk factors for neonatal drug withdrawal, we fitted logistic regression models (including all trajectory groups as independent variables), adjusting for variables in a stepwise manner: 1) exposure to risk factors for NOWS/neonatal drug withdrawal syndrome, including use of psychotropic medication, mental-health–related conditions, and nonopioid substance use and abuse; 2) indications for prescription opioid use; and 3) demographic variables.

Lastly, we repeated analyses using the final GBTM model and calculation of the AR and relative risk of NOWS and severe NOWS after restricting the cohort to women with ≥2 dispensings of prescription opioids during pregnancy.

RESULTS

Of the initial mother-infant linked cohort, consisting of 1,948,427 pregnancies with maternal eligibility from 3 months before pregnancy to 1 month after delivery and infant eligibility for ≥1 month after birth, 18,869 mothers were chronically exposed to opioids prior to pregnancy, filled at least 1 additional opioid prescription during pregnancy, and fulfilled the additional inclusion requirements described above (see Figure 1 for details on cohort creation). Of these women, 13,759 (72.9%) were dispensed opioids with ≥30 days’ supply during pregnancy, and 11,312 (60.0%) were exposed to opioids in the third trimester. Cumulative opioid dose when considering all opioid prescriptions dispensed during pregnancy ranged from 4.5 MMEs to 47,235 MMEs, corresponding to average daily doses between 0.02 MMEs and 174.9 MMEs. Respective tertile cutpoints were at 945 MMEs and 3,480 MMEs (corresponding to average daily doses of 3.5 MMEs and 12.9 MMEs).

Figure 1.

Creation of a linked mother-child cohort of 18,869 pregnancies with maternal prepregnancy chronic opioid use nested within the Medicaid Analytic eXtract (MAX), 2000–2014. Prepregnancy chronic opioid exposure was defined as 1 or more dispensed prescriptions for opioids during each of the 3 months prior to pregnancy. In order to accurately calculate morphine milligram equivalents (MMEs) for each opioid prescription, we excluded women who did not fill a prescription for oral opioids or fentanyl patches during pregnancy from the cohort. GA, gestational age; ORT, opioid replacement therapy.

Group-based trajectory models

The 6-group model was chosen as the final model (Figure 2 and Table 2; for details on final model selection, see Web Appendix 1, Web Table 2, and Web Figure 1), with the 6 identified trajectory groups corresponding to patterns showing: 1) continuous use at very low doses throughout pregnancy (aveMME throughout pregnancy = 2.3; n = 7,660), 2) continuous low-dose use (aveMME throughout pregnancy = 14.0; n = 5,379), 3) initial moderate-dose use with considerable reduction or discontinuation during the second half of pregnancy (aveMME during the first 3 months of gestation = 23.0 vs. aveMME during the last 3 months of gestation = 0.6; n = 2,903), 4) initial high-dose use which decreased to very low-dose use in late pregnancy (aveMME in the first 3 months of gestation = 73.6 vs. aveMME in the last 3 months of gestation = 6.3; n = 555), 5) continuous moderate-dose use (aveMME throughout pregnancy = 40.1; n = 1,966), and 6) continuous high-dose use (aveMME throughout pregnancy = 92.4; n = 406). There was no evidence of polynomial overfitting, and the 6-group model appeared to fit the data very well based on the diagnostic criteria previously described by Nagin (45) (Web Appendix 2, Web Table 3).

Figure 2.

Mean average daily opioid dose (in morphine milligram equivalents (MMEs)), assessed in biweekly gestational age intervals during pregnancy, among 18,869 women chronically exposed to opioids prior to pregnancy who had at least 1 prescription for opioids dispensed during pregnancy, Medicaid Analytic eXtract, 2000–2014. Solid lines represent the predicted average daily opioid dose in each pregnancy opioid-use trajectory group; dashed lines represent the observed average daily opioid dose in each group.

Table 2.

Number of Pregnant Women Included in Each of 6 Pregnancy Opioid-Use Trajectory Groups and Mean Daily Opioid Dose Among 18,869 Women Chronically Exposed to Opioids Prior to Pregnancy Who Had at Least 1 Dispensed Opioid Prescription During Pregnancy, Medicaid Analytic eXtract, 2000–2014

				Mean Daily Opioid Dose, MMEs
Trajectory Group	Opioid Use Pattern During Pregnancy	No.	%	Total		First 12 Weeks of Gestation		Last 12 Weeks of Gestation
				Dose	95% CI	Dose	95% CI	Dose	95% CI
Total		18,869	100.0	13.6	13.3, 13.8	18.4	18.1, 18.8	10.5	10.2, 10.7
1	Continuously very low dose	7,660	40.6	2.3	2.3, 2.4	4.6	4.5, 4.7	1.5	1.4, 1.5
2	Continuously low dose	5,379	28.5	14.0	13.9, 14.2	16.1	15.8, 16.3	12.6	12.3, 12.8
3	Initially moderate dose with decrease to very low dose/no use	2,903	15.4	9.1	8.9, 9.3	23.0	22.6, 23.5	0.6	0.5, 0.8
4	Initially high dose with decrease to very low dose	555	2.9	35.7	34.6, 36.8	73.6	71.3, 75.9	6.3	5.3, 7.2
5	Continuously moderate dose	1,966	10.4	40.1	39.6, 40.6	40.4	39.7, 41.1	40.0	39.3, 40.8
6	Continuously high dose	406	2.2	92.4	90.0, 94.7	97.3	94.1, 100.5	85.7	82.3, 89.1

Abbreviations: CI, confidence interval; MME, morphine milligram equivalents.

While spaghetti plots of 200 randomly selected patients for each group showed some between-individual variability in opioid use patterns, most of the individual trajectories were concentrated close to the group’s average (Web Figure 2).

Cohort characteristics

When assessing characteristics across the different trajectory groups and alternative opioid exposure definitions, women in the higher trajectory groups (especially those with moderate- to high-dose exposure corresponding to groups 4–6), as compared with those with very low-dose exposure (trajectory group 1), were generally older and more likely to have specific pain indications, particularly arthritis/arthropathy, back/neck pain and joint pain, and prescriptions for other psychotropic medications, but were less likely to have abdominal pain and dental problems. They were also more likely to be from the Northeast and to be White and less likely to be from the South and to be Black (Table 1).

Similar yet less pronounced patterns were observed when comparing women based on the traditional exposure definitions (Web Table 1).

Association with NOWS

Among all 18,869 infants born to mothers with prepregnancy chronic opioid use and at least 1 opioid dispensation during pregnancy, 520 (27.6 per 1,000 infants) were diagnosed with NOWS, with more than half of them having severe symptoms (n = 284). The risk strongly varied across trajectory groups: The lowest risk was observed among infants prenatally exposed (on average) to constantly very low opioid doses (group 1) (NOWS: AR = 7.7 per 1,000 (95% CI: 5.7, 9.7); severe NOWS: AR = 3.7 per 1,000 (95% CI: 2.3, 5.0)), and the highest risk was seen among those with continuous exposure to high doses (group 6) (NOWS: AR = 172.4 per 1,000 (95% CI: 135.5, 209.3); severe NOWS: AR = 108.4 per 1,000 (95% CI: 78.0, 138.7)). This resulted in an unadjusted relative risk of 22.4 (95% CI: 16.1, 31.2) for NOWS and 29.6 (95% CI: 18.7, 47.1) for severe NOWS when comparing group 6 with group 1. Infants with (on average) high-dose exposure at the beginning of pregnancy which decreased to very low-dose exposure in late pregnancy (group 4) had a more than 8-fold higher risk of NOWS (AR = 64.9 per 1,000, 95% CI: 44.3, 85.4) and an 11-fold higher risk of severe NOWS (AR = 39.6 per 1,000, 95% CI: 23.4, 55.9) as compared with those with (on average) constantly very low-dose exposure (group 1) (Table 3).

Table 3.

Absolute and Relative Risks of Neonatal Opioid Withdrawal Syndrome in Each of 6 Pregnancy Opioid-Use Trajectory Groups Among 18,869 Women Chronically Exposed to Opioids Prior to Pregnancy Who Had at Least 1 Dispensed Opioid Prescription During Pregnancy, Medicaid Analytic eXtract, 2000–2014

				NOWS					Severe NOWS a
Trajectory Group	Opioid Use Pattern During Pregnancy	No.	%	No. of Cases	AR b	95% CI	RR	95% CI	No. of Cases	AR b	95% CI	RR	95% CI
Total		18,869	100.0	520	27.6	25.2, 29.9			284	15.1	13.3, 16.8
1	Continuously very low dose	7,660	40.6	59	7.7	5.7, 9.7	1.0	Referent	28	3.7	2.3, 5.0	1.0	Referent
2	Continuously low dose	5,379	28.5	155	28.8	24.3, 33.3	3.7	2.8, 5.0	83	15.4	12.1, 18.7	4.2	2.8, 6.5
3	Initially moderate dose with decrease to very low dose/no use	2,903	15.4	48	16.5	11.9, 21.2	2.1	1.5, 3.1	27	9.3	5.8, 12.8	2.5	1.5, 4.3
4	Initially high dose with decrease to very low dose	555	2.9	36	64.9	44.3, 85.4	8.4	5.6, 12.6	22	39.6	23.4, 55.9	10.8	6.2, 18.8
5	Continuously moderate dose	1,966	10.4	152	77.3	65.5, 89.1	10.0	7.5, 13.5	80	40.7	32.0, 49.4	11.1	7.3, 17.1
6	Continuously high dose	406	2.2	70	172.4	135.5, 209.3	22.4	16.1, 31.2	44	108.4	78.0, 138.7	29.6	18.7, 47.1

Abbreviations: AR, absolute risk; CI, confidence interval; ICD-9-CM, International Classification of Diseases, Ninth Revision, Clinical Modification; NOWS, neonatal opioid withdrawal syndrome; RR, relative risk.

^a Severe NOWS was defined as ≥1 ICD-9-CM code for NOWS and the presence of at least 1 of the following complications, which may be markers for NOWS severity: admission to the neonatal intensive care unit, feeding difficulties (ICD-9-CM code 779.3x), respiratory symptoms (ICD-9-CM codes 769–770.xx), or signs of seizures (ICD-9-CM codes 779.0x and 780.3x) within the first 30 days after birth (10, 28).

^b Per 1,000 infants.

Similar yet less pronounced patterns of increasing risk with longer exposure duration or higher cumulative dose were observed for alternative exposure groups. Of these alternative groups, infants in the highest cumulative dose tertile had the highest risk (NOWS: AR per 1,000 = 59.4 (95% CI: 53.6, 65.3); severe NOWS: AR per 1,000 = 32.6 (95% CI: 28.2, 37.0)), which was lower than the risk observed for trajectory groups 4–6, particularly group 6 (Web Table 4).

When comparing a logistic regression model that included all of the above-mentioned alternative exposure groups as independent variables (≥30 days’ supply vs. <30 days’ supply, third-trimester vs. first- or second-trimester exposure, cumulative tertiles) with the 5-trajectory group model, receiver operating characteristic curves were fairly similar and concordance (c) statistics were almost identical (NOWS: c = 0.73 vs. c = 0.75; severe NOWS: c = 0.74 vs. c = 0.76) (Web Table 5, Web Figure 3). This finding indicates that—when used in combination to capture different aspects of exposure in terms of duration, timing, and dose—traditional exposure measures perform almost as good as trajectory models for predicting the risk of NOWS.

Stepwise adjustment for potential predictors of opioid use and risk factors for neonatal drug withdrawal only slightly attenuated the association between the respective trajectory groups and NOWS (Web Table 6).

When further restricting the cohort to women with 2 or more opioid dispensings during pregnancy (n = 16,696 pregnancies), the patterns of the trajectory groups and respective risks of NOWS/severe NOWS hardly changed (Web Figure 4, Web Table 7).

DISCUSSION

Key findings

Using a nationwide cohort of publicly insured pregnant women with chronic opioid exposure prior to pregnancy, we found strong variability in opioid dose, duration, and timing of use during pregnancy. Women in the highest trajectory group were, on average, exposed to a 40-fold higher mean daily opioid dose than those in the lowest trajectory group. They were 22 times more likely to have an infant diagnosed with NOWS, with an absolute risk of 17% as compared with 0.8% for those with very low-dose exposure during pregnancy. Whereas the risk of NOWS generally increased with longer duration and the degree of exposure in late pregnancy, we also found a considerably high risk of developing NOWS (6.5%) among infants exposed (on average) to initially high doses which gradually decreased to very low doses toward the end of pregnancy. While this would contradict the notion that it is mostly late exposure that triggers NOWS, these results need to be interpreted with caution. Women with high-dose opioid prescriptions early in pregnancy might not have used the opioids as prescribed but rather used them at a reduced dose and therefore for a longer period, which would explain the observed lower dose/no medication dispensings toward the end of pregnancy. Alternatively, it is possible that these women were more likely to take nonprescription opioids, especially since overly aggressive tapering of opioid prescriptions might cause some patients to seek out illicit sources of opioids (46).

Findings from other studies

While most cases of NOWS occur in infants with prenatal exposure to legally obtained opioids, given their widespread use (30), previous studies have focused mainly on exposure to illicit opioids or opioid agonist therapy (33, 47–51). Therefore, literature on the impact of prescription opioid analgesics is still limited (10, 30, 52).

In a previous study, we examined the association between prescription opioid use during pregnancy and NOWS and observed a 2-fold increase in risk for long-term (≥30 days’ supply) versus short-term exposure and a 24% increase for late (last 90 days of pregnancy) versus early exposure (10). This study extended that work by 1) identifying more granular opioid use patterns during pregnancy; 2) considering women with prepregnancy chronic opioid use, who are more likely to continue opioid use throughout pregnancy; and 3) evaluating how timing, dose, and duration of exposure together affect the risk of NOWS.

Limitations and strengths

This study had several strengths, in addition to its large cohort size and the availability of detailed patient-level information on clinical and demographic variables. While interest in using trajectory models to summarize medication use during pregnancy is increasing, only a few studies have used these methods so far (11–14), and none have looked at opioid exposure. We found that logistic regression models designed to evaluate the association between opioids and NOWS performed better when we included trajectory groups, as compared with more traditional exposure definitions capturing data on either dose, duration, or timing of exposure, but not necessarily when comparing trajectory groups with a combination of these measures. However, trajectory models do provide a more intuitive representation of opioid utilization patterns during pregnancy: Instead of using single exposure measures or arbitrary cutpoints, the GBTM approach naturally incorporates both quantity and timing of medication availability and allows for a visual representation of complex exposure patterns. As such, trajectory models can be useful for gaining insight into common exposure patterns and can inform the exposure definition to be considered in causal association studies, regardless of whether the definition will ultimately be based on the trajectory groups.

This study also had some limitations to consider. There was the possibility of exposure misclassification, since filling a prescription does not necessarily imply that the medication was actually taken as prescribed. This should be of less concern for our definition of prepregnancy chronic users and for women with multiple dispensings throughout pregnancy, assuming that if women go through the effort to fill a prescription multiple times, they are more likely to actually be taking the medication. However, it could have led to an overestimate of exposure among women with only 1 or a few dispensing(s) throughout pregnancy, as well as to misspecification of timing of exposure, particularly among those with high-dose use at the beginning but low-dose/no use at the end of pregnancy, since they might have used the medication consistently at a lower dose, as described above. Exposure could have further been overestimated if some women diverted their opioids or underestimated if some women also took opioids illicitly.

Additionally, relying on recorded diagnostic codes to determine the presence of NOWS might have led to underestimation of the true absolute risk of NOWS, as those infants with milder symptoms might not always carry such a code. However, previous studies have suggested that the sensitivity of ICD-9-CM diagnostic code 779.5 to identify neonatal drug withdrawal is relatively high (84%–95%) (29, 30). Likewise, severity of NOWS had to be estimated using indirect proxies such as signs of seizure, rather than more quantitative measures like the Finnegan score (53). As noted above, we could not distinguish between NOWS and other drug withdrawal syndromes. However, given that opioids are the most common cause of neonatal withdrawal and all children in our cohort were born to mothers who had opioid exposure both before and during pregnancy, children with ICD-9-CM diagnostic code 779.5 in our cohort were most likely to have NOWS.

Lastly, this cohort reflected the Medicaid population, which overrepresents socioeconomically disadvantaged women and their liveborn infants. While the biological associations studied should be generalizable to other populations, if there is more illicit substance use among Medicaid beneficiaries and if the observed effect of prescription opioids on NOWS is partially attributable to simultaneous exposure to illicit substances, our findings may not be fully generalizable to the broader population of obstetrical patients.

Implications

In addition to showing how prescription opioids are utilized in pregnancy among women who used opioids chronically prior to pregnancy, our findings reveal how timing, dose, and duration of use together affect the risk of NOWS. These results can help clinicians understand how changes in opioid utilization during pregnancy can modulate the risk of NOWS in the offspring. Nearly 1 out of 5 term-born infants who were prenatally exposed to persistently high (on average) doses of opioids developed NOWS. Further, whereas the assumption has been that NOWS is mainly triggered by high-dose exposure close to birth, we also found a considerably higher risk among infants born to mothers with (on average) high-dose exposure at the beginning of pregnancy which gradually decreased thereafter; the question of whether this is a true effect or is partly due to women saving some of their prescribed doses for use later in pregnancy needs to be explored in future studies. These findings underscore the importance of minimizing opioid use during pregnancy as much as possible (preferably throughout the entire pregnancy, not only during the last few months prior to birth) to decrease risk of NOWS and of balancing this concern with the need to treat maternal pain conditions and the risks associated with decreasing or discontinuing opioid use (e.g., prompting women to use illicit opioids).