Neonatal Opioid Withdrawal Syndrome Is Associated With Infant Foster Care Entry At The County Level

Abstract

Although the number of infants diagnosed with neonatal opioid withdrawal syndrome (NOWS) and the number of infants entering foster care have increased substantially in the US since 2009, analyses exploring their relationship are lacking. Using data from 580 US counties in eight US states from the period 2009–17, we examined the association of county rates of NOWS and county-level characteristics with infant foster care entries. In adjusted analyses, every one diagnosis of NOWS per ten births was associated with a 41 percent higher rate of infant foster care entry, and rural county residence was associated with a 19 percent higher rate of infant foster entry. A higher employment rate was associated with lower rates of infant foster care entry both overall and in urban counties when we stratified by rurality. These findings suggest that policy makers could use information about county characteristics to better target funding to support opioid-affected families at risk for foster care involvement.

The opioid crisis is devastating individuals, families, and communities across the US. From 2010 to 2017 the number of pregnant women with an opioid-related diagnosis at delivery and the number of infants diagnosed with neonatal opioid withdrawal syndrome (NOWS), also known as neonatal abstinence syndrome, approximately doubled nationwide.¹ Nationwide, infants younger than age one make up nearly 20 percent of new foster care entries,² with notable geographic variation in rates of infant foster care entry across the US.^3,4 Infants born to mothers with opioid use disorder are commonly involved with the child welfare system, likely contributing in part to the recent surge in infant foster care entries.^5,6

Understanding the relationship between opioid-related diagnoses and child welfare outcomes is critical to designing systems that improve outcomes for the maternal-infant dyad. Some national studies found a positive association between overdose deaths and drug-related hospitalizations and foster care entries.^4,6 Other analyses focused on opioid prescribing found a positive association between county-level child removals and opioid analgesic prescribing rates in single states,^7,8 whereas a nationwide study found positive associations in some states but not others.⁹ However, comprehensive assessments of infant-specific child welfare outcomes related to maternal opioid use during pregnancy are lacking. In addition, county characteristics have been shown to influence outcomes for opioid-affected maternal-infant dyads but are understudied for child welfare outcomes. For example, a recent study found that counties with a shortage of mental health clinicians and higher long-term unemployment had significantly higher rates of NOWS than counties without those characteristics.¹⁰ Similarly, county-level rates of child poverty, single parenthood, White race, and population density have been associated with child maltreatment report rates.¹¹ Collectively, there has been a paucity of analyses in large samples focused on infant-specific child welfare outcomes related to maternal opioid use during pregnancy and how these associations vary at the county level.

We sought to address this critical knowledge gap regarding the association between opioid-affected maternal-infant dyads and the child welfare system and to enhance the ability of local child welfare systems to better serve families^4,12 by examining the relationship of NOWS and county-level characteristics with county foster care entries among infants. Informed by the existing literature,^10,13,14 we hypothesized that the availability of health care (density of obstetricians), community support (density of religious institutions), interpersonal social or financial support (rates of births to married women), and economic stability (employment) would be associated with lower rates of infant foster care entry, whereas residing in a rural county would increase risk.

Study Data And Methods

This retrospective cohort study included 2009–17 data from 580 counties in eight US states (Florida, Kentucky, Massachusetts, Michigan, North Carolina, New York, Tennessee, and Washington). Following prior work,¹⁰ we chose these states to represent the range of geographical regions and diverse patterns of population-level opioid-related complications.¹⁵ The Institutional Review Board of Vanderbilt University Medical Center deemed this study exempt from human subjects review.

COVARIATES

We estimated the risk for county-level child welfare involvement associated with county rates of maternal opioid use, maternal race and ethnicity, and community-level factors in five domains: health care availability, maternal education, rurality, community and interpersonal social or financial support, and economic stability. To measure maternal opioid use during pregnancy, we obtained county-level rates of NOWS using inpatient admissions identified with International Classification of Diseases, Ninth Revision, Clinical Modification, code 779.5 or International Statistical Classification of Diseases and Related Health Problems, Tenth Revision, code P96.1¹⁰ from the Healthcare Cost and Utilization Project’s State Inpatient Databases. We included the proportion of births among unmarried women and women with four-year degrees at the county level, gathered from natality records from the Centers for Disease Control and Prevention (in these data, maternal race and ethnicity is self-reported). We also included the proportion of births among non-Hispanic Black women, given the literature demonstrating systemic racism in the child welfare system that results in higher likelihood of referral¹⁶ and lower likelihood of reunification after foster care placement¹⁷ among non-Hispanic Black families compared with non-Hispanic White families. Obstetrician density and employment rates were obtained from Area Health Resources Files of the Health Resources and Services Administration. Religious institution density was derived from the Northeast Regional Center for Rural Development, and rurality was determined using 2013 Rural-Urban Continuum Codes from the Economic Research Service, Department of Agriculture. Counties were designated as “urban” if they were classified in this scheme as a metropolitan area, as “rural adjacent” if they were classified as being adjacent to a metropolitan area, and as “rural remote” if they were classified as not being adjacent to a metropolitan area.¹⁰

OUTCOMES

County-level infant foster care entry rates were obtained from the Adoption and Foster Care Analysis and Reporting System, Administration for Children and Families, Department of Health and Human Services. Infant foster care entry was defined as entry for children younger than age one.

STATISTICAL ANALYSIS

We computed descriptive statistics, including medians and interquartile ranges for continuous variables and frequencies and percentages for categorical variables. We created state figures to visualize county-level differences in rates of infant foster care entry per 1,000 infants and rates of NOWS diagnosis per 1,000 births. Next, to evaluate county factors associated with infant foster care entry, we estimated negative binomial models using xtset and xtnbreg in Stata, version 15. These models were stratified by rurality, with county random effects and year fixed effects; the number of infant births in each county was used as an offset.

We chose negative binomial models because our outcome is a count variable that is skewed and because the dispersion of foster care entry was too great to use Poisson regression. We used county random effects, rather than fixed effects, in line with other studies looking at similar research questions^6,18 and to avoid overparameterization of our model. These models assumed that the dispersion in foster care rates at the county level followed a transformed Beta distribution.^19,20 Our study was interested in within-county as well as cross-county variation—that is, the extent to which differences in NOWS rates between counties explained differences in infant foster care entries. Fixed effects models do not permit exploring between-county variation. In addition, county fixed effects do not permit independent variables that vary between counties but not over time, including rurality.^21,22

In our study, we were interested in identifying the extent to which rurality may relate to infant foster care entries, both as an independent variable and as a stratification factor. Rurality stratification was done by rural (rural remote and rural adjacent) and urban. Unadjusted and adjusted incident rate ratios were calculated. Adjusted models included covariates outlined previously. We conducted supplemental analyses to ensure robustness, including a negative binomial regression model with county fixed effects, two modified Poisson regression models with county random and fixed effects, and linear regression with county random effects. Details on the main analysis and the supplemental analyses are in the online appendix.²³ Analyses were completed using Stata, version 15, and R, version 3.6.3.

LIMITATIONS

This study had limitations that merit mentioning. First, the retrospective cohort study design did not allow us to infer causation between infant foster care entries, NOWS, and county characteristics. Second, outcome data included all infant foster care entries rather than entries related to maternal opioid use or, more generally, parental drug use, as the validity of the data collected for these entry reasons is questionable for many jurisdictions.²⁴ This resulted in wide variance in the percentage of drug-related foster care entries across states that is inconsistent with the literature. We could only identify ecological associations between incidence rates and do not know the proportion of infant foster care entries directly attributable to maternal opioid use. Third, NOWS was used as a measure of maternal opioid use during pregnancy and was identified from hospital discharge data. Although diagnosis codes have been shown to have high positive predictive value,²⁵ the possibility of misclassification bias exists.

Fourth, maternal use of medications for opioid use disorder, which has been shown to improve pregnancy outcomes,²⁶ can result in NOWS. Fifth, we likely underestimated the relationship between maternal opioid use during pregnancy and infant foster care entries, given that many infants exposed to opioids prenatally will not be diagnosed with NOWS.²⁷ Sixth, analyses in rural communities may be limited by low sample size or heterogeneity on key metrics such as density of obstetricians (see appendix tables 1 and 2).²³ Seventh, even though the study used nine years of data for 580 counties in eight diverse states, the study results might not be generalizable to other states or periods.

Study Results

During our study period, the median rate of county infant foster care entry was 11.3 per 1,000 infants (interquartile range [IQR]: 5.75, 18.60), and that of NOWS diagnoses was 8.4 per 1,000 births (IQR: 2.9, 18.4) (exhibit 1). County rates of births to Black mothers (median: 22; IQR: 0, 125) and to women with four-year degrees (median: 130; IQR: 16, 250) varied substantially by county. Counties classified as urban were most represented (45.0 percent), followed by rural adjacent (31.9 percent) and rural remote (23.1 percent).

EXHIBIT 1.

Rate of infant foster care entry, rate of neonatal opioid withdrawal syndrome (NOWS), and county-level characteristics, 2009–17

	Median	IQR
Rate of infant foster care entriesa	11.3	5.75, 18.60
Rate of NOWSb	8.4	2.9, 18.4
County-level characteristics
Births to Black womenb	22	0, 125
Births to women with a 4-year degreeb	130	16, 250
Births to married womenb	566	485, 750
Employedc	420	380, 456
Religious institutionsd	8	6, 10
Obstetriciansc	0.05	0, 0.09
	Frequencye	Percent
County rurality
Rural remote	1,197	23.1
Rural adjacent	1,656	31.9
Urban	2,331	45.0

source Authors’ own analysis of data from the Adoption and Foster Care Analysis and Reporting System, the Healthcare Cost and Utilization Project’s State Inpatient Databases, Centers for Disease Control and Prevention natality records, Area Health Resources Files, the Northeast Regional Center for Rural Development, and Rural-Urban Continuum Codes. notes Rates shown are those for the given measure per specified population divided by 9 years to produce a rate per year during the study period. IQR is interquartile range.

^aRate per 1,000 infants.

^bRate per 1,000 births.

^cRate per 1,000 population.

^dRate per 10,000 population.

^eTotal number of county-years classified as rural remote, rural adjacent, and urban.

Between 2009 and 2017 the rate of infant foster care entry increased from 9.4 (IQR: 5.2, 16.5) to 13.0 (IQR: 6.8, 21.2) per 1,000 infants, and the rate of NOWS diagnoses per 1,000 births increased more than fourfold from 3.2 (IQR: 0.0, 7.7) to 14.6 (IQR: 7.0, 29.5). Rates of births to Black women and births to married women stayed fairly constant over time, whereas the rate of births per 1,000 births to women with a four-year degree more than doubled, from 62 (IQR: 0, 176) in 2009 to 157 (IQR: 62, 278) in 2017. The number of rural counties and county rates of employment, religious institutions, and obstetricians stayed constant (exhibit 2 and appendix table 3).²³ We observed within-state, county-level variation in high rates of infant foster care entry and NOWS. In Kentucky, North Carolina, and Tennessee, states chosen as a representative example of contiguous states, counties with the highest rates of infant foster care entry and NOWS diagnosis in their state represented both rural and large metro areas, but both tended to occur more commonly in rural counties (exhibits 3 and 4). Similar county-level differences were observed for the other study states (see appendix exhibits A and B).²³

EXHIBIT 2.

Temporal trends in rate of infant foster care entry, rate of neonatal opioid withdrawal syndrome (NOWS), and county-level characteristics, 2009–17

	2009	2010	2011	2012	2013	2014	2015	2016	2017
Rate of infant foster care entrya	9.4	10.5	10.1	11.3	11.8	11.5	11.6	12.6	13.0
Rate of NOWSb	3.2	4.4	6.1	7.2	9.3	10.0	13.8	14.3	14.6
County-level characteristics
Births to Black womenb	22	19	21	20	21	25	23	19	25
Births to women with a 4-year degreeb	62	84	133	138	136	148	146	158	157
Births to married womenb	554	568	561	566	570	572	568	576	567
Employedc	425	419	421	423	419	413	417	422	427
Religious institutionsd	8	8	8	8	8	8	8	8	8
Obstetriciansc	0.05	0.05	0.05	0.05	0.05	0.04	0.04	0.04	0.04
County rurality
Rural remote	133	133	133	133	133	133	133	133	133
Rural adjacent	184	184	184	184	184	184	184	184	184
Urban	259	259	259	259	259	259	259	259	259

source Authors’ own analysis of data from the Adoption and Foster Care Analysis and Reporting System, the Healthcare Cost and Utilization Project’s State Inpatient Databases, Centers for Disease Control and Prevention natality records, Area Health Resources Files, the Northeast Regional Center for Rural Development, and Rural-Urban Continuum Codes.

^aRate per 1,000 infants.

^bRate per 1,000 births.

^cRate per 1,000 population.

^dRate per 10,000 population.

EXHIBIT 3. Geographic variation in the infant foster care entry rate per 1,000 births in Kentucky, North Carolina, and Tennessee, 2009–17.

source Authors’ own analysis of data from the Adoption and Foster Care Analysis and Reporting System and Centers for Disease Control and Prevention natality records.

EXHIBIT 4. Geographic variation in the neonatal opioid withdrawal syndrome rate per 1,000 births in Kentucky, North Carolina, and Tennessee, 2009–17.

source Authors’ own analysis of data from the Healthcare Cost and Utilization Project’s State Inpatient Databases and Centers for Disease Control and Prevention natality records.

ASSOCIATION OF NOWS, FOSTER CARE, AND COUNTY CHARACTERISTICS

In adjusted models, an increase of one NOWS diagnosis per ten births in a county was associated with a 41 percent higher rate (adjusted incident rate ratio: 1.41; 95% confidence interval: 1.26, 1.57) of infant foster care entry, and rural counties were associated with a 19 percent higher rate (aIRR: 1.19; 95% CI: 1.08, 1.31) compared with urban counties (exhibit 5). One additional person employed per 10 population was associated with a 14 percent lower rate (aIRR: 0.86; 95% CI: 0.82, 0.91) of infant foster care entries, whereas one additional obstetrician per 1,000 population was associated with a 72 percent lower rate (aIRR: 0.28; 95% CI: 0.16, 0.48). Other county-level characteristics associated with lower rates of infant foster care entry included rate of births to Black women per 10 population (aIRR: 0.96; 95% CI: 0.94, 0.98) and rate of births to married women per 10 population (aIRR: 0.98; 95% CI: 0.97, 0.99). The rate of births to women with a four-year degree (versus women without a four-year degree) was not associated with infant foster care entry. In models stratified by rurality, urban counties showed associations with infant foster care entry that were similar to those seen in the overall model. In addition, one religious institution per 1,000 population was associated with a 48 percent higher rate (aIRR: 1.48; 95% CI: 1.18, 1.86) of infant foster care entry in urban counties. Among rural counties, only the association with births to Black women remained, and religious institution density was associated with a 19 percent lower rate (aIRR: 0.81; 95% CI: 0.68, 0.97) of infant foster care entry (exhibit 5 and appendix table 4).²³ In sensitivity analyses, associations of NOWS and the county employment rate with infant foster care entry remained significant across all models, whereas other county-level characteristics varied (appendix tables 5–8).²³

EXHIBIT 5.

Association of rates of infant foster care entry with neonatal opioid withdrawal syndrome (NOWS) and county-level characteristics, by rurality, 2009–17

	Overall		Urban		Rural
	Unadjusted IRR	Adjusted IRRa	Unadjusted IRR	Adjusted IRRa	Unadjusted IRR	Adjusted IRRa
Rate of NOWSb	1.67**	1.41**	1.87**	1.50**	1.50**	1.20**
County-level characteristics
Births to Black womenb	0.95**	0.96**	0.94**	0.95**	0.97**	0.97**
Births to women with a 4-year degreeb	1.00	1.01	1.00	1.02**	1.01	1.00
Births to married womenb	0.99	0.98**	0.98**	0.97**	1.00	0.99
Employedc	0.87**	0.86**	0.87**	0.83**	0.90**	0.96
Religious institutionsd	1.21**	1.12	1.89**	1.48**	0.78**	0.81**
Obstetriciansd	0.11**	0.28**	0.08**	0.18**	0.55	0.76
County rurality
Urban	Ref	Ref	—f	—f	—f	—f
Rurale	1.44**	1.19**	—f	—f	—f	—f

source Authors’ own analysis of data from the Adoption and Foster Care Analysis and Reporting System, the Healthcare Cost and Utilization Project’s State Inpatient Databases, Centers for Disease Control and Prevention natality records, Area Health Resources Files, the Northeast Regional Center for Rural Development, and Rural-Urban Continuum Codes. notes This exhibit shows the effect of an increase of one of the specified measure on the rate of infant foster care entry; see the text for more explanation. Rates shown are those for the given measure per specified population divided by 9 years to produce a rate per year during the study period. IRR is incident rate ratio.

^aAdjusted models accounted for year fixed effects.

^bRate per 10 births.

^cRate per 10 population.

^dRate per 1,000 population.

^eRural remote and rural adjacent counties combined.

^fNot applicable.

^**p < 0.05

Discussion

In this study of 580 counties in eight US states during the period 2009–17, county rates of neonatal opioid withdrawal syndrome and rurality were associated with higher rates of infant foster care entry, whereas higher employment was associated with lower rates. With these findings taken together, this study provides further evidence that the opioid crisis is taking a toll on maternal and child well-being beyond clinical settings. Further, these findings highlight the potential role of economic development and local community support in mitigating infant foster care entry with equitable, prevention-focused substance use and child welfare policies.

Our finding of the association of county-level NOWS diagnoses and infant foster placement expands previous research that found associations between foster entry and opioid analgesic prescribing rates^7,8 and between overdose deaths and drug-related hospitalizations.⁶ Collectively, these findings highlight the ongoing effect that the opioid crisis is having on families and communities. Between 2011 and 2017 the number of infants in the US foster care system increased from 40,000 to 50,000, with states that had the highest rate of NOWS having the highest rates of placement.³ However, this burden is not evenly distributed within states, as we observed county-level variation in rates of NOWS and infant foster care entry in this study.

Although the US child welfare system was initially designed to respond to child physical and sexual abuse, its scope has expanded to include parental substance use and primary prevention to address underlying socioeconomic concerns. Receiving services from the child welfare system is not a panacea: There is evidence that child welfare systems alone may be insufficient in supporting caregivers with substance use disorder, as evidenced by low engagement in or completion of recommended substance use treatments.^28,29 Perhaps because of these limitations, the Child Abuse Prevention and Treatment Act of 1974,³⁰ the nation’s foundational child welfare legislation, has been modified by Congress twice since 2016 to include a focus on connecting substance-affected caregivers with treatment and wraparound services via Plans of Safe Care.^31,32 Federal policy related to child welfare and substance-exposed infants has evolved since the passage of the Child Abuse Prevention and Treatment Act in the early 1970s. Although many of the recent changes to this legislation, including Plans of Safe Care, aim to address social determinants of health and connect substance-affected caregivers to treatment, analyses including a 2018 Government Accountability Office report³³ suggest that states have struggled to implement many of these changes. These challenges are particularly salient as Congress is again working to reauthorize the Child Abuse Prevention and Treatment Act and could implement provisions that address the county-level variability in foster care entry that we observed.

In this study, the association between NOWS and foster care placement was influenced by county-level factors including employment and rurality. Opioid use disorder disproportionately affects lower-income people, and some evidence supports the notion that employment opportunities improve outcomes for people with substance use disorder.³⁴ Our findings suggest that the broader employment environment (that is, the proportion of adults employed in a county) may also play a role. Although these findings imply that community economic development may attenuate foster care placement, the effect of individual and community employment opportunities in mitigating foster care placement merits further study.

In addition to employment levels, we found other county characteristics to be associated with foster care placement, although these findings were inconsistent in our supplemental analyses. In our primary model, we found that county density of obstetricians was associated with lower rates of infant foster placement and that density of religious institutions was associated with higher rates in urban but not rural counties. Given the inconsistency of these findings by rurality and in supplemental analyses, the role of provider supply and community as represented by religious institutions requires additional evaluation. Even with these limitations, the role of county-level factors in influencing the association between NOWS and foster care placement further emphasizes the complex relationships among substance use disorder, its complications, and social determinants of health, which innovations such as the Family First Prevention Services Act of 2018 are working to collectively address.³⁵

The Family First Prevention Services Act allows states to use federal foster care funds (Title IV-E) for prevention services that can decrease risk for foster care entry.^35,36 The act has the potential to be transformative, allowing foster care funds to be used directly to fund, for example, treatment of opioid use disorder in pregnancy. However, its implementation has been restrictive, perhaps limiting its potential. Candidate programs and services for Family First Prevention Services Act reimbursement are identified for review by a clearinghouse through public comment and suggestions from local, state, and federal administrators, followed by a systematic evidence review of the available literature, which rates each program and service to determine eligibility.³⁷ This process has resulted in limits on evidence-based therapies. For example, despite clinical trial evidence of buprenorphine’s efficacy³⁸ and guidelines endorsing its use for opioid use disorder treatment during pregnancy,^39,40 it is not a service approved in the Family First Prevention Services Act. Furthermore, the act states that as of late 2021, half of all expenditures must go toward “well-supported” services,⁴¹ further limiting the scope of reimbursable services including methadone programs (a mainstay of treatment) that are rated as “promising” by the clearinghouse. Increasing flexibility in the scope of reimbursable services and extending the act’s deadline for states to meet the well-supported program requirement, such as was recently called for by the American Academy of Pediatrics,⁴² may encourage communities to pilot programs tailored to their needs.

Conclusion

The opioid crisis is increasingly affecting pregnant women and infants, influencing outcomes well after delivery. Even after county factors were accounted for, neonatal opioid withdrawal syndrome was associated with higher rates of infant foster care entry; however, county-level factors including employment rate were associated with lower rates of entry. As the US grapples with the opioid crisis and Congress considers additional child welfare policy changes, improving the ability of policies to support stakeholders in tailoring holistic approaches that focus on primary prevention, treatment, and community strengths and needs is a promising approach to mitigating the effects of the crisis on the child welfare system.