Descriptive results are presented in Table IV. Of the 260 525 live births within 232 hospitals that met study criteria, 2.6% (n = 6756) had a diagnosis of prenatal substance use. Roughly half of IPSE (n = 3070; 45.4%) were reported to CPS within 14 days of delivery. Diagnosed births that were reported to CPS were more likely to have had prenatal care initiated late or not at all, to be low birth weight, to have missing paternity, and to be the third or greater order birth. Diagnosed IPSE births to Black mothers had lower rates of CPS reports (37.3%) than those to Hispanic mothers (47.0%) and White mothers (47.6%). However, differences by race and ethnicity were eliminated when stratified by substance type. Among those with only cannabis diagnoses, 25.0% of Black mothers, 24.2% of Hispanic mothers, and 29.6% of White mothers were reported to CPS in the first 2 weeks. Among women with diagnoses related to substances other than cannabis, 59.9% of Black mothers, 63.7% of Hispanic mothers, and 59.7% of White mothers were reported in the first 14 days.

Prevalence distribution of substance-specific diagnostic codes by race and ethnicity is presented in Table V: 5.2% of births to Black mothers had a relevant diagnosis, compared with 2.1% of those born to Hispanic mothers and 2.3% born to White mothers. Among those with diagnoses, 54.6% had a code related to cannabis, 29.7% to amphetamines and other stimulants, 12.6% to opioids, and 6.8% to alcohol use. In addition to substance-specific codes, almost all (96.5%) had an ICD-10-CM code related to pregnancy or delivery complicated by alcohol or drug use. Nearly half (45.6%) of IPSE births had a code related to cannabis, but no other specific substance. Births to Black mothers were more likely to have no substance other than cannabis (64.7%) compared with Hispanic (42.2%) and White (40.2%) mothers.

An unconditional random-intercept model yielded an interclass coefficient of .24, signifying that 24% of variance in CPS reporting was between-hospitals. Individual- and hospital-level covariates were added to estimate a conditional random intercept model with interaction terms. The fully adjusted model identified an association between CPS reporting and public health insurance (relative risk: 1.95; 95% CI: 1.65, 2.30) and between CPS reporting and delivering in a hospital serving majority publicly insured maternal patients (relative risk: 1.77; 95% CI: 1.41, 2.21) (see Table VI).

Predicted probabilities show the differential association between hospital-level patient demographics and CPS reporting for individual racial and economic categories (Figure 1). For White mothers, the predicted probability of reporting in a minority-White hospital (0.49; 95% CI: 0.46, 0.53) trended higher than in a majority-White hospital (0.43; 95% CI: 0.39, 0.48). The same was true for Hispanic mothers (0.45; 95% CI: 0.43, 0.48 vs 0.36; 95% CI: 0.30, 0.42). Yet this pattern is not observed among Black mothers, whose likelihood of reporting was not significantly different in a minority-White hospital than in a hospital with greater than 50% of births occurring to White patients.

For those with private insurance, delivering in a minority-public health insurance hospital had a lower predicted probability of reporting (0.24; 95% CI: 0.20, 0.28) relative to a majority-public hospital (0.42; 95% CI: 0.36, 0.48). For those with public insurance, however, the predicted probability of reporting was not statistically significantly different in a majority-public vs a minority-public hospital. Within minority-public hospitals, births covered by public health insurance were nearly twice as likely to be reported than births covered by private insurance (0.46; 95% CI: 0.42, 0.51 vs 0.24; 95% CI: 0.20, 0.28).

In this cohort of Black, White, and US-born Hispanic births with evidence of substance exposure in the maternal record, racial differences existed in both substance use and substance type, with more births to Black mothers having documented cannabis exposure. Among diagnosed IPSE, we found that the delivering hospital explained 24% of variance in CPS reporting. In a fully adjusted hierarchical model, findings revealed race and insurance-type differences in the likelihood of a CPS report. Births to White and Hispanic mothers were less likely to be reported in White-dominated hospital settings compared with less White hospitals, while this was not true for births to Black mothers. Publicly insured births in hospitals where more than half the births were covered by private insurance had nearly twice the probability of being reported compared with those with private health insurance.

In line with earlier research that used California hospital discharge records of Black, White, and Hispanic mothers from 2006, we found that less than 3% of the birth cohort was IPSE, as documented in diagnostic coding.¹⁸ This is substantially lower than self-reported prenatal substance use prevalence, which was 13% in 2018.³⁰ Our finding that Black mothers were disproportionately likely to be diagnosed with cannabis use is also consistent with the prior analysis, though cannabis prevalence among all those with a substance use diagnosis in our 2018 cohort (74.3%) was higher than that found in 2006 (46.6%). Although differential rates of cannabis use by race and ethnicity is consistent with California-based analyses that relied on universal toxicology results³¹ or banked biospecimens from a prenatal screening program,³² prevalence of prenatal cannabis use diagnoses is far lower than positive toxicology tests. It is worth noting that the biological samples used in those studies were taken in the first and second trimesters of pregnancy, while the diagnoses used in this analysis were in the hospital discharge record at the time of a live birth event.

Similar to the 2006 California-based study, nearly half of the sample was reported to CPS (in 2006, 53.4% were reported in the first 28 days; in 2018, 45.4% were reported in the first 14 days).¹⁸ The current study extends that work by incorporating hospital-level data to examine associations with CPS reporting, which may contribute to variations in reporting prevalence. In line with an analysis from Washington, we found that clustering the data by birth hospital explained 24% of variance in the likelihood of CPS reporting and that hospitals with a majority of publicly insured births were more likely to report IPSE to CPS than hospitals with a minority of publicly insured births.¹⁹ The inclusion of interactions between individual- and hospital-level characteristics allowed us to demonstrate differential associations between maternal race and ethnicity and economic position and CPS reporting dependent on the predominant patient demographics of the birthing hospital.

The Child Abuse Prevention and Treatment Act requires states to quantify substance-exposed births, yet many states lack mechanisms to accomplish this outside of CPS reporting.¹⁰ Prevalence estimates of IPSE involved in CPS reveal the burden of parental substance use in those reported for maltreatment but do not offer insights about a broader population of hospital-identified IPSE who may not be reported.^2,33 Hospital diagnostic codes may be a valuable way to estimate the prevalence of IPSE in policy environments where universal CPS reporting is not required. However, in this analysis, as in previous similar studies, diagnostic codes demonstrated low sensitivity as a measure of prenatal use, though likely high specificity.^34,35

To maximize the utility of diagnostic codes, institutions must address inconsistencies in verbal screening, toxicology testing, and diagnostic documentation.^36,37 For diagnoses to serve as a measurement tool for IPSE prevalence, there must be no institutional pressure to file maltreatment reports on all identified IPSE.^11,38 Mechanisms for public health surveillance, such as those that allow for notification of substance-exposed births outside of maltreatment reporting, would improve quantification of prenatal substance use in a way that is not punitive or stigmatizing for a family.^39,40 The low prevalence of IPSE in a large and diverse birth cohort found in this study contributes to an understanding of the magnitude of bias at play in the screening and identification of IPSE across institutions and providers and is a useful starting point from which to focus improved medical coding or other documentative mechanisms.

Efforts at the federal level, including proposed amendments to the Child Abuse Prevention and Treatment Act, indicate an intended shift away from universal reporting and toward situating responses to IPSE in a public health framework.^7,10,41 Instrumental to this movement is a reassessment of the role of CPS and an improved ability of hospital teams to make decisions about maltreatment risk, including refinement of reporting policies and provider behaviors that engage CPS only when necessary to prevent child harm.^42,43

For decades, evidence has shown that bias is at play in CPS reporting, even in settings with universal screening or identification.^44–46 Our findings of reporting variance by birth hospital cluster and differential associations dependent on hospital-level patient demographics confirm this. Efforts to combat biased reporting should be informed by perspectives and input from providers, clinicians, and affected families.^11,47,48 Institutions should emphasize clinical education that acknowledges the role of provider bias⁴⁹ while pursuing comprehensive and systemic solutions to structural racism.⁵⁰ Individualized and team-based approaches that have an evidence-based tolerance for risk show the most promise to reduce discrimination and disproportionality in CPS reporting.⁵¹ In cases where CPS reporting can be avoided, warm referrals to effective family support services are essential components to a comprehensive public health response to prenatal substance exposure.⁵² Financial assistance to families, in particular, may protect against future CPS intervention.^53,54

This study develops upon existing examinations of discretionary reporting by updating prevalence statistics and including interaction terms between individual and hospital-level characteristics that reveal bias in reporting. Despite its usefulness, this analysis was affected by several limitations. First, we used diagnostic codes to identify a sample of IPSE, which allowed an examination of discretionary reporting. However, the sensitivity of this inclusion criterion was low, meaning we omitted an unknown number of substance-exposed births from this analysis.³⁴

We should note that this study does not include data on prenatal substance use screening or testing. Research has consistently shown disparities in toxicology testing during pregnancy and at delivery.^37,55–58 There is likely wide variation in how California birth hospitals identify IPSE: some with universal screening policies, others with risk-based screening protocols, and others without any standardized procedures in place. For this reason, our sample of births identified by the hospital as IPSE may be considered a high surveillance or high acuity subset of all substance exposed births.⁵⁹ The current analytic sample of diagnosed IPSE provides an opportunity to examine variance in reporting conditional on diagnostic identification.

Secondly, we were unable to link infant discharge records, meaning that we misclassified some number of identified substance-exposed birth dyads. In the similar California analysis from 2006, while 10% of substance-exposed dyads had this status documented only in the infant record, only 1.6% of the birth cohort had any indication of substance exposure.¹⁸ Changes in documentation in addition to growing incidence may have contributed to these discrepancies.

Third, our analysis had the possibility of omitted variable bias. We lacked data on other potentially important individual-level factors, including substance use extent and chronicity, patient presentation, treatment engagement, and parental readiness. We also lacked data on hospital- and provider-level characteristics, including screening and testing protocols, hospital teaming approaches, the presence of perinatal social workers, etc. However, our use of hospital as a clustering variable helped to capture unobserved differences at the institutional level.

Lastly, the dichotomization of the hospital-level variables we assessed, although allowing model convergence, resulted in information loss. A follow-up analysis that incorporates clinical documentation and qualitative inquiry would allow for a richer examination of CPS reporting decisions. These limitations mean that caution should guide the extrapolation of these findings to other samples. Even still, the variance we identified provides a useful quantitative indicator of the differential experiences birthing people may have dependent on their institution of care.