One-year risk of death following postpartum opioid exposure

Abstract

Background:

Opioids are commonly prescribed to women for acute pain following childbirth. Postpartum prescription opioid exposure is associated with adverse opioid-related morbidities, but the association with all-cause mortality is not well studied.

Objective:

To examine the association between postpartum opioid prescription fills and the one-year risk of all-cause mortality among women with live births.

Methods:

In a retrospective cohort study of live births among women enrolled in Tennessee Medicaid (TennCare) between 2007 and 2015, we compared women who filled two or more postpartum outpatient opioid prescriptions (up to 41 days post-delivery discharge) to women who filled one or fewer opioid prescription. Women were followed from day 42 post-delivery discharge through 365 days of follow-up or date of death. Deaths were identified using linked death certificates (2007–2016). We used Cox proportional hazard regression and inverse probability of treatment weights to compare time to death between exposure groups while adjusting for relevant confounders. We also examined effect modification by delivery route, race, opioid use disorder, use of benzodiazepines and mental health condition diagnosis.

Results:

Among 264,135 eligible births, 216,762 (82.1%) had one or fewer maternal postpartum opioid fills and 47,373 (17.9%) had two or more fills. There were 182 deaths during follow-up. The mortality rate was higher in women with two or more fills (120.5 per 100,000 person-years) than in those with one or fewer (57.7 per 100,000 person-years). The risk of maternal death remained higher in participants exposed to two or more opioids fills after accounting for relevant covariates using inverse probability of treatment weighting (adjusted hazard ratio: 1.46 (95% confidence interval: 1.01, 2.09)). Findings from stratified analyses were consistent with main findings.

Conclusions:

Filling two or more opioid prescriptions during the postpartum period was associated with a significant increase in one-year risk of death among new mothers.

BACKGROUND

Pregnancy-associated deaths, or deaths from any cause while pregnant or within the first year after delivery, have risen in recent decades.¹ While many of these deaths are attributed to pregnancy complications and chronic health conditions exacerbated by pregnancy,² maternal deaths attributable to accidental or incidental causes during or after pregnancy, such as self-harm and violence, are increasingly recognized as being common, preventable and in urgent need of intervention.^3–7 Maternal deaths related to opioid use, in particular, have increased in recent years.^1,8–11 In Tennessee, specialized reviews and adjudication panels have determined that maternal deaths related to substance use disorders may be underreported as causes of death in death certificates. Furthermore, after comprehensive review of available subject records, substance use disorder was identified as a contributing factor in approximately one-third of all state maternal deaths in 2017.¹²

Postpartum prescribing of opioid medications to treat pain after cesarean and vaginal births may contribute to these trends.^13–18 Among opioid-naïve women, 59% of vaginal births and 91% of cesarean births in Tennessee Medicaid filled at least one postpartum prescription (2007–2014).¹⁹ By comparison, an average of approximately 28% and 39% of women of childbearing age (2005–2011),²⁰ and 14% and 22% of pregnant women (2000–2007) filled an opioid prescription annually among privately insured and Medicaid-enrolled populations, respectively.^21,22 Further, our previous work has demonstrated that the number of postpartum opioid prescriptions is associated with serious opioid-related events in women, including persistent opioid use, opioid use disorder (diagnosis or buprenorphine/methadone fills), opioid overdose, and opioid-related death.¹⁹ However, the association between postpartum prescription opioids and all-cause mortality among new mothers is not well characterized. Therefore, we investigated the association between the number of opioid prescriptions filled in the first six weeks after birth and the subsequent one-year all-cause mortality among women enrolled in Tennessee Medicaid (TennCare).

METHODS

We conducted a retrospective cohort study using administrative TennCare claims and pharmacy data linked to birth certificates, death certificates and a registry of hospital-based encounters (Hospital Discharge Database System) from the State of Tennessee for years 2007–2015. TennCare, the managed care Medicaid program in Tennessee, provides healthcare insurance to eligible Tennessee residents, including pregnant women and children. For the last four decades, Vanderbilt Investigators have worked in collaboration with the Division of TennCare to perform studies on the safety and effectiveness of medications and policies among subjects enrolled in TennCare. Data from TennCare are received and processed at Vanderbilt and supplemented with vital records and hospital-based encounter data to enable longitudinal evaluations of medications and policies. Prescribed medications are ascertained using pharmacy files, which include the date that the prescription was filled, the prescribed days of medication supply and the dispensed quantity. Computerized pharmacy records have been shown to be an accurate source of medication data²³ and have high concordance with patient self-report of medication use.^23–26 This research platform has been extensively used for pharmaco-epidemiological studies.^27–37 The study protocol was approved by the Institutional Review Boards of Vanderbilt University Medical Center and Tennessee Department of Health, and by the Division of TennCare.

The study population comprised 15- to 44-year-old females enrolled in TennCare who delivered a live birth in a Tennessee hospital between January 1, 2007 and August 14, 2015, as determined by linked birth certificates. Eligible women were required to be continuously enrolled in TennCare 180 days prior to delivery through 41 days post-delivery hospitalization discharge to allow for the observation and measurement of important covariates related to maternal health (i.e., diagnosis and procedure codes, pharmacy fills). We excluded women who remained hospitalized more than five days post-delivery, and those who died before the start of follow-up. The start of follow-up (t₀) was defined as day 42 post-delivery hospitalization discharge. The follow-up period continued until the woman either died or reached 365 days after t₀ (eFigure 1). Since the one year follow-up period begins at t₀, not birth, our estimated rates of death do not correspond to established definitions for maternal mortality (i.e., deaths occurring up to one year after delivery). Rather, we report the one-year risk of all-cause death following the 6 week postpartum period.

Exposure Assessment

Prescription opioid exposure was defined by the number of opioid prescriptions filled at an outpatient pharmacy in the postpartum period. The window for assessing prescription fills was three days prior to delivery (to account for early fills for planned deliveries) through 41 days post-delivery discharge. Study opioids included oral and transdermal formulations of hydrocodone, oxycodone, codeine, tramadol, propoxyphene, meperidine, morphine, oxymorphone, dihydrocodeine, fentanyl, pentazocine, and tapentadol. Other opioid formulations commonly used for cough or diarrhea were not included. As previous studies have showed that filling two or more initial opioid prescriptions may be associated with a higher risk of opioid-related events, mainly chronic opioid use,^19,38 we compared women with two or more postpartum prescriptions filled (exposure) to women with one or no postpartum prescriptions filled (comparator).

Outcome Assessment

The primary outcome was maternal death from any cause identified during the 365-days following t₀. This definition allows for a full year of follow-up after the approximate six-week postpartum exposure assessment period. Deaths were identified using linked Tennessee death certificates, which allowed for the ascertainment of date and causes of death. We classified underlying causes of death into six categories defined by International Classification of Disease (ICD-10) codes. We defined overdose deaths (any drug or intent) using previously established codes for drug poisoning (X40-X44; X60-X64; X85; Y10-Y14)³⁹ or other instances with an underlying cause related to drug use. We categorized other records, as appropriate, into the following groupings: violence (including self-harm) or other accident (e.g., X65-Y09 and other/unspecified injury), motor vehicle accident (V00-V99), cardiovascular/circulatory-related (I00-I99 and related O codes), cancer (C00-C97) and other medical causes (including but not limited to unspecified causes and miscellaneous O codes). These categorizations were based on the approach used by the 2017 Tennessee Department of Health Maternal Mortality Review Committee.¹²

We further determined whether these deaths were opioid related by evaluating contributing causes of death, as indicated in secondary cause of death fields on the death certificate. For each of the six underlying causes of death, opioid-related deaths were identified as those having contributing causes of death related to opium (T40.0), heroin (T40.1), other opioid (T40.2), methadone (T40.3) or synthetic narcotic (T40.4) exposure.^39–41

Covariates

Covariates were identified based on their potential association with the risk of death and opioid exposure. Covariates were assessed during the prenatal (180 days through 1 day before birth) and postpartum (date of birth through 41 days post-delivery discharge) periods. We defined covariates using information from birth certificates, relevant diagnostic codes, procedure codes, and prescriptions fills. Covariates included sociodemographic information (age, race, ethnicity, and rurality of residence⁴²); income (estimated as median income in census track of residence); mental health condition diagnoses (bipolar, anxiety, depression, and schizophrenia or delusional disorders); concurrent prescriptions (benzodiazepines, attention deficit hyperactivity disorder (ADHD) medications, anti-depressants, and anti-psychotics); opioid use disorder (diagnosis or filled methadone or buprenorphine prescription); tobacco use during pregnancy; and characteristics of pregnancy and delivery (birth year, parity, plurality, mode of delivery).^19,43 We characterized pain related conditions including trauma, severe maternal morbidity,⁴⁴ self-inflicted pain, and accidents. We also characterized the presence of kidney disease as a contraindication for nonsteroidal anti-inflammatory drugs. Mode of delivery was defined as vaginal or cesarean using birth certificate information.

Statistical Analysis

We calculated exposure propensity scores to account for exposure group differences in the distribution of covariates listed above. Propensity scores are useful for the study of rare study outcomes that require adjustment for numerous potential confounders.⁴⁵ Using propensity scores, we computed stabilized inverse probability of treatment weights to decrease the influence of extreme propensity score values. Standardized mean differences were calculated before and after propensity score weighting to determine the balance of covariates between exposure groups. Standardized mean differences of less than 0.10 were considered evidence of good covariate balance.⁴⁶ Inverse probability weighted Cox proportional hazard regression models were used to estimate adjusted hazard ratios (aHR) and 95% confidence intervals (95% CI) for death during the 365 day follow-up period, comparing women with one or fewer postpartum opioid prescriptions filled (comparator) to those with two or more postpartum opioid prescriptions filled (exposed). The robust variance estimator was specified to account for lack of independence introduced by women who had multiple live births during the study period.⁴⁷ Missing covariate data (<5% for all covariates) were imputed using multiple imputation chained equations and five imputed datasets were generated for subsequent analyses. We calculated E-values to estimate the sensitivity of our findings to unmeasured confounding for the main point estimate of the adjusted hazard ratio and, for the most conservative estimate, for the lower bound of its confidence interval.⁴⁸ In exploratory analyses, we examined effect modification across strata of baseline characteristics including mode of delivery, race (limited to Black and White due to small sample size in other groups), mental health conditions, opioid use disorder (diagnosis or methadone/buprenorphine prescription fills), and benzodiazepine exposure using interaction terms. These baseline characteristics were defined as any diagnoses or use of relevant medications in the prenatal or postpartum period. All statistical analyses were performed in Stata/IC 16 and R v 3.6.2.

RESULTS

Cohort characteristics

Overall, 264,135 births from 192,015 women were included in the study, including 216,762 (82.1%) births in which women filled one or fewer postpartum opioid prescriptions and 47,373 (17.9%) in which women filled two or more (Table 1, Figure 1). We excluded 184,217 births that did not meet TennCare enrollment and survival criteria (Figure 1). Compared to women having included births, women having excluded births were more likely to be White and to be smokers (eTable 1). Twenty-eight percent of births with cesarean deliveries and 13% of births with vaginal deliveries filled two or more prescriptions. The mean age of women at delivery was 24.1 ± 5.2 years, and most births were among non-Hispanic White women (62.9% and 78.4% in comparator and exposed groups, respectively). Births in the exposed group had a higher proportion of cesarean delivery (49.1% vs. 27.6%), longer total days’ supply of opioid prescriptions (median [interquartile range (IQR)]: 10 days [6, 16] vs. 2 days [0, 5]) and were more likely to be tobacco users (45.1% vs. 24.9%) compared to the comparator group. The number of exposed births decreased modestly with increasing year. Study participants from the comparator group were more likely to be from an urban setting than the exposed participants (73.7% vs. 69.2%). Most filled opioid prescriptions were oxycodone or hydrocodone (89.9% of first fills and 85.5% of second fills).

Table 1.

Baseline characteristics of live births in Tennessee Medicaid (2007–2015) by number of maternal postpartum opioid prescriptions

Maternal Characteristics	0–1 prescriptions		≥2 prescriptions
Total, n (%)	216,762	(82.1)	47,373	(17.9)
Age at delivery, years, mean (SD)	24.0	(5.2)	24.7	(5.1)
Race*, n (%)
White	136,309	(62.9)	37,129	(78.4)
Black	77,620	(35.8)	9870	(20.8)
Asian	1680	(0.8)	154	(0.3)
Other	1013	(0.5)	202	(0.4)
Ethnicity*, n (%)
Hispanic	6954	(3.2)	953	(2.0)
Non-Hispanic	209,648	(96.8)	46,388	(98.0)
Rurality*, n (%)
Rural	17,991	(8.3)	4778	(10.1)
Mixed	38,895	(18.0)	9791	(20.7)
Urban	159,193	(73.7)	32,709	(69.2)
Region of Tennessee, n (%)
West	73,401	(34.0)	10,111	(21.4)
Middle	72,250	(33.4)	18,700	(39.5)
East	70,428	(32.6)	18,467	(39.1)
Delivery Route, n (%)
Vaginal	157,013	(72.4)	24,094	(50.9)
Cesarean	59,749	(27.6)	23,279	(49.1)
Birth Year, n (%)
2007	24478	(11.3)	6472	(13.7)
2008	24498	(11.3)	6793	(14.3)
2009	24933	(11.5)	6357	(13.4)
2010	25251	(11.6)	6096	(12.9)
2011	25486	(11.8)	5425	(11.5)
2012	25684	(11.8)	5225	(11.0)
2013	26219	(12.1)	4612	(9.7)
2014	25465	(11.7)	4193	(8.9)
2015	14748	(6.8)	2200	(4.6)
Parity*, n (%)
0	83754	(38.8)	15600	(33.0)
1	66553	(30.8)	15183	(32.2)
2	36859	(17.1)	9727	(20.6)
3	16373	(7.6)	4072	(8.6)
4+	12549	(5.8)	2620	(5.6)
Births per woman, n (%)
1	157314	(72.6)	34701	(73.3)
2	45961	(21.2)	10053	(21.2)
3+	13487	(6.2)	2619	(6.1)
Tobacco Use, n (%)	54,049	(24.9)	21,359	(45.1)
Severe Maternal Morbidity†, n (%)	3334	(1.5)	1004	(2.1)
Benzodiazepine Use, n (%)	7114	(3.3)	5224	(11.0)
Mental Health Condition, n (%)	30,786	(14.2)	14,729	(31.1)
Opioid Use Disorder, n (%)	1354	(0.6)	2026	(4.3)
Postnatal Opioid Prescriptions‡
Number, median (IQR)	1	(0,1)	2	(2–3)
Total days’ supply, median (IQR)	2	(0,5)	10	(6, 16)

IQR, interquartile range

^*N is the number of non-missing values. Total number of observations= 264,135. Observations available for race was 99.9% (263,977), ethnicity 99.9% (263,943), Income 96.5% (254,896), Rurality 99.7% (263,357), parity 99.7% (263,290) and region 99.7% (263,357).

^†Severe maternal morbidity was defined as the presence of any of the following during the delivery encounter: acute myocardial infarction, aneurysm, acute renal failure, adult respiratory distress syndrome, amniotic fluid embolism, liver failure, cardiac arrest/ventricular fibrillation, conversion of cardiac rhythm, disseminated intravascular coagulation, eclampsia, heart failure/arrest during procedure, puerperal cerebrovascular disorders, pulmonary edema/acute heart failure, severe anesthesia complications, sepsis, shock, sickle cell disease (includes crisis), air and thrombotic embolism, blood transfusion, hysterectomy, temporary tracheostomy, cardiomyopathy, or ventilation⁴⁴

^‡assessed three days prior to delivery through 41 days post-delivery discharge

Figure 1.

Flow chart of subject eligibility

For both prenatal and postpartum medications, women in the exposed group were more likely to have at least one prescription of benzodiazepines, ADHD medications, antidepressants, and antipsychotics relative to the comparator group (Table 2). Additionally, exposed women were more likely to have prenatal or postpartum diagnoses of depression, anxiety, injuries due to accident, and trauma (e.g., fractures, dislocations, wounds, etc.). Exposed women had higher rates of kidney disease during the postpartum period women compared to women receiving one or fewer opioid.

Table 2.

Baseline medications and comorbidities among women with live births in Tennessee Medicaid (2007–2015) by prenatal and postnatal periods and number of postpartum opioid prescriptions

	Prenatal*				Postpartum†
	0–1 prescriptions (n = 216,762)		≥2 prescriptions (n = 47,373)		0–1 Prescriptions (n = 216,762)		≥2 prescriptions (n = 47,373)
	n	%	n	%	n	%	n	%
Medications
Benzodiazepines
None	211913	97.8	44441	93.8	213846	98.7	44268	93.4
One	4088	1.9	2122	4.5	2828	1.3	2937	6.2
Two or more	761	0.4	810	1.7	88	<0.1	168	0.4
ADHD medications
None	216030	99.7	46939	99.1	215825	99.6	46700	98.6
One	303	0.1	144	0.3	907	0.4	645	1.4
Two or more	429	0.2	290	0.6	30	<0.1	28	0.1
Antidepressants
None	207145	95.6	42557	89.8	201972	93.2	40035	84.5
One	5815	2.7	2899	6.1	14224	6.6	6958	14.7
Two or more	3802	1.8	1917	4.0	566	0.3	380	0.8
Antipsychotic medications
None	215424	99.4	46777	98.7	215545	99.4	46828	98.8
One	685	0.3	320	0.7	1112	0.5	482	1.0
Two or more	653	0.3	276	0.6	105	<0.1	63	0.1
Comorbidities
Depression	2806	1.3	1057	2.2	2507	1.2	966	2.0
Anxiety	1865	0.9	975	2.1	1191	0.5	604	1.3
Schizophrenia‡	255	0.1	43	0.1	187	0.1	39	0.1
Bipolar disorder	1678	0.8	717	1.5	1439	0.7	583	1.2
Acute/Chronic Renal Failure	113	0.1	36	0.1	96	<0.1	51	0.1
Injury/Accident	4068	1.9	1591	3.4	68	<0.1	164	0.3
Self-Inflicted, suicide attempt	46	<0.1	14	<0.1	≤10	<0.1	≤10	<0.1
Trauma	6194	2.9	2583	5.5	902	0.4	842	1.8
Malignancy	2431	1.1	580	1.2	281	0.1	83	0.2

^*Prenatal = 180 days through 1 day before birth.

^†Postpartum = day of birth through 41 days post-delivery.

^‡Schizophrenia and delusional disorders

Maternal deaths

Starting at day 42 post-delivery discharge (t₀) and following for 365 days, we observed 182 maternal deaths during years 2007 through 2015 (57 and 125 among exposed and comparator groups, respectively). In the exposed group, motor vehicle accidents were the most frequent cause of death (n = 16, 28.1%), followed by overdose (n = 12, 21.1%) and other causes (n = 11, 19.3%) (Figure 2). In the comparator group, the most common causes of death were motor vehicle accidents (n = 33, 26.4%), violence (n = 28, 22.4%), and other causes (n = 28, 22.4%) (Figure 2). Overdose accounted for 15.2% of deaths (n = 19) in this group. Of 31 total overdose deaths, 18 (58.1%) were determined to be opioid related according to the death certificate information. Overdose deaths were more likely to be opioid related in the exposed group than in the comparator group (proportions suppressed due to small counts). Opioids were not identified as a contributory cause in any other underlying cause of death category.

Figure 2.

Proportion of underlying cause of death among exposed (n = 57) and comparator (n = 125) maternal deaths. *Violence, cardiac and cancer deaths combined due to small counts (n ≤ 10).

Postpartum opioid exposure and one-year mortality

The one-year death rate was higher in the exposed group (120.5 deaths per 100,000 person-years) than in the comparator group (57.7 deaths per 100,000 person-years), yielding a crude hazard ratio for death of 2.09 (95% CI: 1.53, 2.86) (eTable 2; Figure 3). Within the comparator group, the death rate was similar for women receiving no prescription (56.1 deaths per 100,000 person-years) and those receiving one (58.7 deaths per 100,000 person-years). Among the exposed, the death rate increased to 107.5 deaths per 100,000 person-years for women receiving two prescriptions and 146.6 deaths per 100,000 person-years for those receiving three or more.

Figure 3.

Kaplan-Meier survival curve for the 365-day follow-up, beginning at day of post-delivery discharge, comparing women who filled 2+ postpartum opioid prescriptions and those who filled ≤ 1

In the weighted proportional hazards regression, the adjusted hazard ratio was 1.46 (95% CI: 1.01, 2.09). Examination of standardized mean differences demonstrated good balance in the distribution of covariates between exposure groups (eFigure 2). Findings from exploratory assessments for effect modification were largely consistent with those from the primary analysis (Figure 4, eTable 2), although some estimates had limited precision. In stratified analyses, hazard ratios were slightly higher among White women as compared to Black, among those with vaginal births, and among those without mental health conditions. However, estimates were not distinguishable across strata due to substantial confidence interval overlap and interactions did not reach statistical significance (p_interaction = 0.27, 0.66, 0.53, 0.33, and 0.25 for maternal race, delivery route, mental health conditions, benzodiazepine use and opioid use disorder, respectively). A sensitivity analysis to address the unmeasured confounding that could explain away our findings yielded e-values of 2.28 and 1.11 for the HR and its lower confidence limit, respectively.

Figure 4.

Adjusted hazard ratios and 95% confidence intervals for the association between two or more postnatal opioid prescriptions, versus one or fewer, and maternal death, overall and stratified by maternal characteristics. Dashed line denotes the point estimate for the overall association. *Stratified estimates for race not calculated for Asian and Other race due to small counts. † Lower 95% confidence limit for Opioid Use Disorder, yes = 0.05

COMMENT

Principal findings

In this large retrospective cohort study, we observed that the one-year rate of death from all causes, including opioid and non-opioid related deaths, was approximately twice as high in women who filled two or more opioid prescriptions during the postpartum period than in women who filled one or fewer prescriptions. After accounting for maternal comorbidities and pain conditions, women who filled two or more postpartum opioid prescriptions experienced a 46% increased risk of death. Based on death certificate coding, motor vehicle accidents were the leading cause of death in both exposure groups. Opioid use was not coded as an underlying or contributing cause in motor vehicle or other non-overdose deaths, however the true role of opioids in these deaths remains unclear. We acknowledge that the assigned cause of death on the death certificate used in our study may not adequately capture opioid or other substance use. According to the 2017 Tennessee Maternal Mortality Review, which conducted a comprehensive evaluation of causes of death using death certificates, medical and mental health histories, social and environmental profiles, and case narratives, approximately one-third of pregnancy-associated deaths had substance use disorder as a contributing factor.¹²

Pertaining to deaths that were identified as overdose-related, overdose was the second-leading cause of death behind motor vehicle accidents among women with multiple postpartum opioid prescriptions. Many of these overdose deaths were attributed to opioids. By comparison, overdose was the fourth leading cause of death among women in the comparator group. These findings suggest that overdose is a prevalent cause of postpartum death regardless of postpartum opioid exposure, and that postpartum opioid prescribing may be an influential contributor to this association.

Postpartum opioid prescribing is common and highly variable across the U.S. Depending on the population studied, approximately 10–50% of women with vaginal births^14–17,43 and over 80% of those with cesareans receive at least one opioid prescription following delivery.^15,18 The postpartum period is often a woman’s first exposure to opioids, and compounded with factors associated with postpartum lifestyle adjustment, including sleep deprivation and depression, women may be at particularly high risk for opioid misuse.^49,50 Previous studies have reported associations between opioid prescribing during and after pregnancy and adverse opioid-related events, such as persistent opioid use.^43,51,52 The risk for serious opioid-related events (a composite of persistent opioid use, opioid use disorder [diagnosis or buprenorphine/methadone fills], opioid overdose, and opioid-related death) has been shown to increase with increasing number of postpartum opioid prescriptions.¹⁹ Similarly, in a non-pregnant population, the number of opioid prescriptions filled in the month of opioid initiation has been associated with long term use.³⁸

Overdose is a common cause of pregnancy-associated death, but the role of postpartum opioid prescribing patterns in maternal death has not been well studied. Goldman-Mellor et al. identified drug-related deaths as the second leading cause of maternal death in California for the 12 months following live birth, with an incidence of 3.68 per 100,000 person years (2010–2012).¹¹ A Philadelphia-based study reported that 21% of deaths during or within one year of pregnancy were due to drug overdose (2010–2014).⁵³ Between 2007 and 2016, opioid-related mortality increased from 4% to 10% of all pregnancy-associated deaths nationwide, most of which occurred prior to 42 days postpartum,¹ a time frame not captured in our study.

Our findings demonstrate that women who fill multiple postpartum opioid prescriptions experience an increased risk of one-year all-cause mortality. Additional research is needed to better understand the relationship between opioid prescribing and non-overdose causes of death, such as motor vehicle accident or violence-related deaths. Regardless of underlying cause, interventions related to opioid prescribing may be impactful in reducing maternal deaths. These interventions may include efforts to reduce opioid prescribing through patient education on pain management, use of non-opioid analgesics, appropriate opioid use when needed, and appropriate disposal of unused medication.⁴⁹ Among women who develop opioid use disorder, accessibility to treatment is needed to prevent more severe outcomes, such as overdose and death. Opioid overdose rates have been shown to be significantly lower among women with opioid use disorder on pharmacotherapy with methadone or buprenorphine, compared to those who are not.⁹ However, access to pharmacotherapy can be challenging. Women experience multiple barriers to obtaining care for opioid use disorder, including high out-of-pocket costs and limited insurance acceptance.⁵⁴

Our study has several important strengths. By linking our cohort with death certificate data, we had a virtually complete ascertainment of maternal death in our study, regardless of TennCare enrollment in the follow up period. We acknowledge that there is some potential for missed deaths if a woman permanently relocated outside of Tennessee during follow up, however we expect these instances to be rare.^55,56 Similarly, the use of birth certificate information provided a detailed characterization of pregnancy and delivery events. Mortality studies that rely on death certificates to document pregnancy are prone to misclassification of pregnancy status.⁵⁷ Through the use of linked databases, we were able to efficiently and reliably identify eligible women and live births, prescription fills, important covariates, and timing and cause of death. Finally, nearly 18% of women in our study, and 13% of vaginal births, filled more than one postpartum opioid prescription. By comparison, a large MarketScan-based study of vaginal births reported 8.5% of women received postpartum opioid refills,¹⁶ and among opioid-naive Medicaid-enrolled women in Pennsylvania, only 1.6% of women with vaginal deliveries filled a second postpartum opioid prescription.¹⁷ As such, this large, highly exposed study population is well suited to address questions related to postpartum opioid prescribing and mortality.

Our study also has several limitations. First, although we controlled for numerous potential confounders, we were not able to measure the severity of clinical conditions and risk factors and therefore cannot rule out possible residual confounding. We conservatively estimated that an unmeasured confounder independently associated with both postpartum opioid prescribing and maternal death at a magnitude of 1.11-fold could explain away the lower bound of our hazard ratio confidence interval. However, we expect that any such unmeasured factor would be at least partially correlated with, and thus, partially controlled for by our measured covariates. Second, because we defined exposures by prescription fills covered by Medicaid, we were not able to determine how much of the filled prescription was consumed, nor could we capture opioid prescriptions covered by another source, paid for out-of-pocket or illicit opioids. Additionally, we did not characterize exposure outside of the postpartum period and the association between maternal death and opioid prescribing before or during pregnancy, or later in the first postnatal year was not determined. Third, given that maternal death is a rare outcome, we identified a small number of deaths over the study period. Although we used propensity scores to generate reliable estimates for our primary analysis, our exploratory subgroup analyses had limited precision. We cautiously note that, while not statistically different, the association between postnatal opioid prescribing and maternal death was stronger in White women than in Black women. While this may appear to contradict historically reported racial disparities for maternal mortality,⁵⁸ recent pregnancy-associated mortality estimates from Tennessee did not significantly differ for White and Black women (103 and 110 per 100,000 births in White and Black women respectively, 2017),¹² potentially an artifact of increased mortality among White women attributable to higher opioid prescribing,^20,59 misuse⁶⁰ or use disorder⁶¹ in White versus Black women. Additional research is needed to explore this phenomenon. Next, we relied on death certificate coding to identify opioid-related deaths. We note that opioid-related deaths may be underreported given that toxicology and autopsy reports are subjective, and that social or environmental contributors to death may not be fully evaluated or reported. Finally, because our study focused on a Medicaid population within Tennessee, our findings may not be directly applicable to women in other geographic locations or women with private insurance. However, our findings are still widely generalizable, as approximately 50% of all births in Tennessee and similar proportions in other states, are covered by Medicaid.

CONCLUSIONS

Postpartum opioid prescribing is common after childbirth, and for many, the postpartum period is the first time a woman is exposed to an opioid. Previous studies have demonstrated a strong association between opioid exposure during the postpartum period and opioid-related morbidity.^19,43 This study demonstrates an association between prescription opioid exposure during the postpartum period and maternal death, where deaths were attributable to a range of events including accidents and violence, in addition to overdose. These findings, along with others, underscore the need for risk awareness among patients and clinicians, coupled with judicious opioid prescribing practices. Women should be counseled on the risks and benefits of opioid medications and the availability of non-narcotic options. As interventions to reduce postpartum opioid prescribing and persistent use are developed, impacts on pregnancy-associated mortality will be important to evaluate.

Key Points.

• Opioid prescribing in the postpartum period is common.

• Prior studies show that >1 postnatal opioid fill is associated with adverse opioid-related events.

• >1 opioid fill within 42 days of delivery was associated with an increase in 1-year risk of death.