Elevated Maternal and Child Mortality among Women with Multiple DUIs Compared to Sociodemographically Matched Controls

Abstract

AIMS

To assess whether having multiple convictions for driving while under the influence of alcohol (MDUI) in women is a risk factor for maternal, infant and child mortality.

DESIGN

Retrospective cohort design using record linkage, comparing women with MDUI convictions with propensity matched women without alcohol-related driving offences ascertained through state records, on rates of maternal, infant and child mortality.

SETTING

Missouri, United States.

PARTICIPANTS

MDUI women (N=1658) and women with no alcohol-related driving convictions (CONTROL, N=184,252) who gave birth from 2000–2004.

MEASUREMENTS

Data were obtained from state administrative records and US Census data. The outcomes were maternal, infant and child mortality. The input variable was presence or absence of MDUI convictions. Propensity matching variables were maternal (smoking during pregnancy, delayed prenatal care, previous child deaths, age at birth, mother Missouri-born, education, pre-pregnancy obesity, marital status) reproductive partner (un-named partner, race/ethnicity, education, DUI status), and census tract (socioeconomic advantage, urbanicity) characteristics.

FINDINGS

Women with MDUI convictions had higher odds of maternal, infant and child mortality than propensity matched controls (OR=2.65, 95% CI=2.07–3.40 and OR=1.75, 95% CI=1.17–2.61 respectively).

CONCLUSIONS

Having multiple convictions for driving under the influence of alcohol in women appears to be a risk factor for increased maternal, infant and child mortality.

INTRODUCTION

National population-based surveys may not capture many of the most severe cases of alcohol use disorder (AUD), a chronic, relapsing disorder with low treatment rates that is associated with medical and psychiatric problems, periods of homelessness, incarceration, and loss of child custody, among other adverse consequences (1–7). The impact of these consequences is particularly harmful for the children of severely affected women, who typically are the primary caretakers (8). The children of even mildly AUD-affected women may experience negative consequences via residence in more socially deprived neighborhoods with adverse environmental exposures, and elevated likelihoods of children engaging in early onset risk behaviors (9–13). These increased environmental adversities add to the well-established contribution of genetic factors to AUD imparted by parental AUD in making the children particularly vulnerable to the development of alcohol and other substance use disorders (14, 15). Characterizing the risk-exposures and risk-mechanisms in extreme high-risk families can inform harm reduction efforts for children and mothers that may block the intergenerational transmission of risk.

The severe AUDs found in women with multiple (two or more) DUIs (MDUIs) differ markedly from the much milder and more transient episodes that dominate in general population surveys (16, 17). Women with MDUIs have rates of AUD as high as 95% (18, 19) and typically manifest severe AUD as indicated by high endorsement of unsuccessful quit attempts, alcohol-related marital and family problems due to drinking, and withdrawal syndrome (20). In a study of women with single and multiple DUI convictions, 80% of whom had children, women with recurrent DUIs had higher rates of lifetime AUD, a longer duration of AUD symptoms, higher rates of illicit drug use, more frequently reported alcohol problems in one or both parents and in their romantic partner, reported more sexual or physical assault prior to age 13 (21), and were more likely to have been accused of child abuse themselves (22). The expression of severe AUDs in women with MDUIs thus impacts many aspects of life, including partner choice and parenting behaviors, which can increase immediate and long-term risks to children.

Child environmental risk associated with maternal AUDs, including mortality risk, begins with increased maternal pregnancy risk behaviors, not only alcohol use - in U.S. population data, 8–39% of women reported moderate alcohol use and 1–17% reported binge drinking during pregnancy (23, 24) - but also tobacco use, because of the extensive comorbidity between smoking and AUDs in women as well as men (25). Alcohol and tobacco use during pregnancy are associated with low birthweight and pre-term birth, both of which increase risk for infant mortality (26–28), but literature comparing pregnancy behaviors in women with severe AUDs and appropriately matched controls appears lacking. Increased mortality in women compared to men with AUD history is well-established (e.g. (29, 30)), including in a large Finnish study of DUI offenders, in women with age-matched controls (31). However, the mortality risk to the child associated with severe maternal AUD, and the rate of child bereavement through maternal death (32), appear un-documented.

Maternal sociodemographic characteristics are key potential confounders in the relationship between severe maternal AUD and mother and child mortality risks. AUD, and particularly severe AUD, is linked to reduced employment and socioeconomic status (16, 33–35), with a disproportionate effect on women (36), including women with DUI convictions (31), and low socioeconomic status is associated with inadequate prenatal care (37, 38).

In the current study, we use propensity score methods (39, 40) to stratify MDUI and control mothers into groups matched on many of the correlated socioeconomic risk-factors and pregnancy risk-behaviors that accompany severe AUDs in women. We use state administrative data on women with MDUIs to identify a large sample of women with severe AUD history. By merging driver’s license data with vital records data (births and deaths) and census data (indices of neighborhood socioeconomic advantage and disadvantage), we focus on pregnancy and early-life risk exposures of children born to MDUI mothers compared to controls, namely pregnancy risk-exposures (delayed prenatal care; maternal smoking during pregnancy), maternal sociodemographic characteristics, neighborhood (i.e. census tract) sociodemographic indicators, in conjunction with the outcomes of maternal and child mortality. This design will help disentangle the unique risks to women and children that are associated with severe maternal AUDs from the many correlated factors that might also affect their lives, with the ultimate goal of identifying fruitful points for prevention or intervention.

METHODS

Design

Subjects were identified and characterized using state administrative data from birth, death, and driving records. Neighborhood characteristics were derived from census data linked to the census tract associated with the birth address of the child. Propensity scores predicting being born to an MDUI mother were used to stratify MDUI and control mothers into groups matched on characteristics associated with MDUIs.

Participants

State administrative data (e.g., birth and driving records) are maintained at the Family and Population Research Center (FPRC) at Washington University in St. Louis (R24AA023487, Heath, PI). The release of state records with identifiers to the FPRC is permitted under Missouri state statutes for research with relevance to public health and safety. The current study was approved by the administrative arm of the Missouri vital records division and by the state ethics board, the FPRC oversight committee, and the Washington University Institutional Review Board. FPRC data managers identified women who met study criteria (described below) and built a de-identified data set for use in these analyses.

Missouri-resident women who gave birth in the state of Missouri in the years 2000–2004 were identified using state birth records. The years 2000–2004 were chosen so that women would have an adolescent-aged child available for participation in future data collections. Of the women identified via birth records, those with 2 or more lifetime convictions for DUI were identified using state driving records (N=1316). We also included women with one DUI only (N=315) or no DUIs on record (N=27) provided they had additional alcohol-related convictions for tampering with ignition interlock device, since interlock devices are required in the vehicles of subjects with 2 or more DUIs (41), or excessive blood alcohol content (where BAC was greater than .08, the threshold for DUI in Missouri; actual range was .08 −.32, median=.16, N=334), since, with adequate resources to hire legal help, it is possible to plea bargain DUIs for a lesser conviction. Plea bargains have been offered as a potential explanation for the ineffectiveness of child endangerment laws to prevent child deaths due to riding with adult drinking drivers (42) (see Supplemental Table 1 for DUI sample composition and supporting information).

For the comparison sample, we selected women with no alcohol-related driving offenses (N=184,252). Only White non-Hispanic women were included in the current study because the number of African American women with MDUIs was too small to permit meaningful analysis (N=90).

Measures

Maternal sociodemographics, pregnancy behaviors and partner characteristics were derived from the birth record. Maternal sociodemographics included geolocation of maternal residential address, maternal age, race and ethnicity, marital status, education, height, and pre-pregnancy weight. Pregnancy characteristics included timing of first prenatal care and number of prenatal care visits, smoking during pregnancy, and length of pregnancy. Delayed prenatal care was defined as no prenatal care visits in the first trimester. Although alcohol use during pregnancy for these birth years was supposed to be recorded on the birth record, it was almost never endorsed, and thus was not included in analyses. Pre-pregnancy body mass index (BMI) was estimated from the mother’s height and pre-pregnancy weight on the birth certificate and categorized as underweight (BMI<18.5), normal (BMI 18.5–24.99), overweight (BMI=25–29.99) and obese (BMI≥30)(43). Paternal demographics were available on the birth certificate for 90.6% of the sample. Information on paternal DUI history for 77.1% of the sample was obtained by linking birth record data about the reproductive partner with driving records. Paternal DUIs were coded as 1 DUI, 2 or more DUIs, no DUI but other alcohol-related driving conviction, or no DUIs nor other alcohol-related driving conviction. Missing paternal data on demographics (9.4%) and DUI (22.9%) were accounted for with dummy variables in the propensity score analysis.

Selected state census tract measures from the year 2000 were submitted to a principal factor analysis which identified 2 factors, the first representing socioeconomic advantage/disadvantage and the second representing race/urbanicity (see Supplemental Table 2 for specific measures included in analysis and factor loadings). Factor scores rather than individual census measures were used to make it easier for other groups to reproduce findings with summary measures, and to avoid violation of state regulations regarding protection of confidentiality.

These 2 factors and 7 individual census measures (see Table 1) were included in the propensity score analysis described below. State census tracts were ranked on the 2 factors and 7 individual census measures and ranks were categorized into a 7-point scale with frequency distribution 10%/10%/20%/20%/20%/10%/10%, to ensure compliance with state regulations about protection of confidentiality, a particular issue when using vital records data, since severe legal sanctions could be levied against researchers for any violation of confidentiality. The exception was urban/rural status, where a 3-level variable was sufficient to summarize the observed distribution with frequencies 40%/20%/40%. Dummy variables representing the categories were used in regression analyses, with the middle category as referent.

Table 1.

Individual and neighborhood census tract characteristics utilized in propensity score in full sample (N=185910), for CONTROL and MDUI mothers, and bivariate associations with MDUI status

	CONTROL N=184252	MDUI N=1658	Odds Ratio (95% CI)
Maternal individual characteristics
Smoking during pregnancy, %	18.06	59.23	6.59 (5.97–7.27)
Delayed prenatal care 1, %	8.14	18.25	2.52 (2.22–2.86)
Previously born children have died 2, %	2.67	5.02	1.93 (1.45–2.55)
Maternal age at birth index child, %
< 20	9.8	14.0	1.80 (1.53–2.13)
20–24	25.4	31.6	1.57 (1.38–1.80)
25–29	28.5	22.6	1.00
≥ 30	36.3	31.8	1.11 (0.97–1.27)
Mother born in Missouri, %	64.64	68.55	1.19 (1.07–1.32)
Highest level education, %
Less than high school	13.78	35.21	1.88 (1.68–2.11)
High school only	29.85	40.48	1.00
Some college/college grad/post-grad	56.37	24.31	0.32 (0.28–0.36)
Pre-pregnancy overweight/obesity, %
Underweight (<18.5)	5.35	9.58	1.51 (1.28–1.80)
Normal (18.5–24.99)	51.81	61.29	1.00
Over weight (25–29.99)	22.34	17.80	0.67 (0.59–0.77)
Obese 1–2 (≥ 30)	20.50	11.33	0.47 (0.40–0.55)
Partnering characteristics 3
Unmarried, %	26.74	63.51	4.77 (4.31–5.27)
No reproductive partner named, %	9.28	26.84	3.59 (3.21–4.00)
Reproductive partner race/ethnicity, %
White	95.10	90.94	1.00
African American	2.24	5.54	2.59 (2.00–3.35)
Other race	2.66	3.52	1.38 (1.01–1.90)
Partner age at birth index child, %
< 20	3.48	5.66	1.83 (1.40–2.38)
20–24	17.62	26.01	1.66 (1.41–1.95)
25–29	27.63	24.58	1.00
≥ 30	51.26	43.75	0.96 (0.83–1.11)
Highest level education, %
Less than high school	11.08	23.37	1.35 (1.16–1.56)
High school only	34.20	53.54	1.00
Some college/college grad/post-grad	54.72	23.10	0.27 (0.23–0.31)
Partner DUI status, %
No DUI	85.01	55.61	1.00
1 DUI	6.50	19.56	4.60 (3.92–5.39)
2 or more DUIs	3.78	18.07	7.30 (6.20–8.60)
Other alcohol-related conviction	4.71	6.77	2.20 (1.72–2.80)
Neighborhood characteristics from census data 4
Socioeconomic advantage	4.43 (1.58)	3.68 (1.48)	0.74 (0.72–0.76)
Urbanicity	3.87 (1.46)	3.49 (1.45)	0.83 (0.81–0.86)
Educational level of residents aged 25 and older, M(SD)	4.28 (1.62)	3.64 (1.55)	0.79 (0.76–0.81)
Total household income, M(SD)	4.41 (1.62)	3.58 (1.53)	0.73 (0.71–0.76)
% Children aged 0–17 in married 2-parent households, M(SD)	4.26 (1.48)	3.90 (1.38)	0.84 (0.81–0.87)
% Managerial/professional employment, M(SD)	3.95 (1.67)	3.66 (1.66)	0.90 (0.87–0.93)
% in labor force who are unemployed, M(SD)	3.61 (1.56)	4.12 (1.49)	1.24 (1.20–1.28)
% Occupied housing units with no vehicle, M(SD)	3.57 (1.55)	4.01 (1.46)	1.20 (1.16–1.24)
% Rural population, M(SD)	1.94 (0.91)	2.19 (0.88)	1.34 (1.27–1.42)
Model fit statistics	-2LL (df)	Change in fit p-value	C Statistic
Intercept only	18951.53 (1)
Intercept + main effects	16077.09 (93)	< .001	.836
Intercept + main effects + interactions	15096.91 (297)	< .001	.868

¹Delayed prenatal care=no care in first 3 months of pregnancy

²conditional on more than 1 child

³Partnering characteristics conditional on having partner data. Ns range from 143385 for partner DUI status to 167322 for partner age.

⁴Neighborhood characteristics: means based on 7-point scale for all variables except rural which has a 3-point scale (see Methods for details).

Child and maternal mortality indicators (yes/no) and dates of death were obtained by matching birth records to state death records for all deaths through April of 2019.

Analysis

Propensity score analysis (39) was used to stratify CONTROL and MDUI mothers into groups matched on factors likely to confound the association of MDUI with mortality. We estimated a logistic regression model predicting MDUI status with measures from the birth record, reproductive partner DUI status, and neighborhood characteristics derived from the census data (see Table 1 for all variables used).

Once main effects were identified, all possible interactions of factor scores and census variables with individual-level variables, and interactions among individual-level variables, were tested using p<.10 for inclusion (39). Interactions that significantly improved the fit of the model (N=204 interactions) were retained in the final prediction model. Fit statistics of the null, main effects, and final models are displayed in Table 1. The C-statistic from the final model (.868) indicated excellent model fit (44). Next, the logistic regression score (“propensity score”) from the final model was categorized into quintiles, and observations with no counterfactuals, i.e., MDUI mothers with scores higher than all CONTROL mothers (N=4) and CONTROL mothers with scores lower than all MDUI mothers (N=6294) were removed from the sample. After this initial trimming, CONTROL mothers were evenly distributed in the quintiles, but MDUI mothers were concentrated in the highest-risk quintiles (top 40% of the distribution), with only 2.4% in the bottom 40% of the distribution, where scores were close to 0. We therefore (39, 45) removed subjects in the bottom 40% of the distribution (2.4% of MDUIs, 40.3% of CONTROLs), re-estimated a second propensity score in the trimmed sample, and divided it into quintiles. The final sample contained 109928 CONTROL mothers and 1614 MDUI mothers. The small numbers of child and maternal deaths within each quintile prohibited analysis by quintile; we therefore combined the bottom 3 quintiles (60% of the propensity score distribution with lower levels of socioeconomic adversity) and the top 2 quintiles (40% of the propensity score distribution with higher levels of socioeconomic adversity). Within these quantiles, MDUI and CONTROL mothers were well matched for almost all of the confounding characteristics associated with MDUI (see Table 2), with the exception of smoking during pregnancy and partner DUI history in the top 40% of the propensity score distribution. These latter covariates were included in the logistic regression testing the association of MDUI status with child and maternal mortality in the top quantile of the propensity score distribution (see Supplemental Table 4 for standardized bias estimates of covariates before and after propensity score estimation and trimming, (45, 46)). Analyses were performed using SAS statistical software (47, 48). To assess the effect of data missingness on observed results, we used Markov-chain Monte Carlo multiple imputation (Stata (48), mi impute) on the data, then estimated the propensity score with imputed data. Results were substantially similar (results available upon request). Thus, the non-imputed results are presented in this manuscript.

Table 2.

Individual and neighborhood census tract characteristics utilized in propensity score in trimmed sample (N=111542), for CONTROL and MDUI mothers, after propensity score matching, by quantile. Bold indicates standardized bias remains after matching.

	Quantiles after propensity score matching
	Bottom 60% (lower risk)		Top 40% (higher risk)
	CONTROL N=66640	MDUI N=285	CONTROL N=43288	MDUI N=1329
Maternal individual characteristics
Smoking during pregnancy, %	11.50	12.28	56.17	70.81
Delayed prenatal care 1, %	8.30	9.61	16.91	20.65
Previously born children have died 2, %	2.80	3.28	4.53	5.25
Maternal age at birth index child, %
< 20	11.81	18.25	19.43	13.47
20–24	30.60	30.53	36.07	32.51
25–29	26.93	24.56	21.21	21.97
≥ 30	30.66	26.67	23.30	32.05
Mother born in Missouri, %	69.36	66.90	68.64	68.92
Highest level education, %
Less than high school	11.60	16.61	36.27	40.17
High school only	40.88	40.99	41.44	40.86
Some college/college grad/post-grad	47.52	42.40	22.29	18.97
Pre-pregnancy overweight/obesity, %
Underweight (<18.5)	5.07	8.54	9.79	10.05
Normal (18.5–24.99)	52.79	53.02	59.82	63.68
Over weight (25–29.99)	24.03	24.91	18.77	15.82
Obese 1–2 (≥ 30)	18.11	13.52	11.62	10.44
Partnering characteristics 3
Unmarried, %	25.24	33.33	67.49	71.86
No reproductive partner named, %	6.90	10.18	27.05	31.08
Reproductive partner race/ethnicity, %
White	95.40	93.90	91.37	89.78
African American	1.99	2.85	5.27	6.58
Other race	2.61	3.25	3.36	3.64
Partner age at birth index child, %
< 20	3.92	5.95	7.87	5.84
20–24	23.06	26.59	28.83	26.71
25–29	27.45	26.19	26.37	24.47
≥ 30	45.57	41.27	36.92	42.99
Highest level education, %
Less than high school	12.40	15.10	25.94	26.56
High school only	45.55	46.53	55.20	57.21
Some college/college grad/post-grad	42.05	38.37	18.86	16.23
Partner DUI status, %
No DUI	86.19	81.82	59.86	47.22
1 DUI	5.25	9.55	19.46	22.88
2 or more DUIs	2.37	2.73	13.46	22.76
Other alcohol-related conviction	6.19	5.91	7.22	7.14
Neighborhood characteristics from census data 4
Socioeconomic advantage	4.20 (1.43)	4.11 (1.44)	3.71 (1.43)	3.56 (1.45)
Urbanicity	3.72 (1.42)	3.83 (1.37)	3.52 (1.43)	3.40 (1.45)
Educational level of residents aged 25 and older, M(SD)	4.03 (1.50)	4.07 (1.48)	3.64 (1.47)	3.52 (1.54)
Total household income, M(SD)	4.16 (1.47)	4.09(1.45)	3.63 (1.49)	3.44(1.50)
% Children aged 0–17 in married 2-parent households, M(SD)	4.11(1.38)	3.94 (1.31)	3.88(1.35)	3.88 (1.38)
% in labor force who are unemployed, M(SD)	3.74 (1.47)	3.86 (1.50)	4.11 (1.47)	4.19 (1.46)
% Occupied housing units with no vehicle, M(SD)	3.71 (1.48)	3.79 (1.59)	4.01 (1.43)	4.08 (1.41)
% Rural population, M(SD)	2.03 (0.90)	1.97 (0.89)	2.18 (0.89)	2.24 (0.87)

¹Delayed prenatal care=no care in first 3 months of pregnancy

²conditional on more than 1 child

³Partnering characteristics conditional on having partner data. Ns range from 143385 for partner DUI status to 167322 for partner age.

⁴Neighborhood characteristics: means based on 7-point scale for all variables except rural which has a 3-point scale (see Methods for details).

RESULTS

Table 1 shows the prevalence of individual measures from the birth record and the mean rank of the factor scores and census tract characteristics that were included in the propensity score in the full sample, separately for MDUI and unmatched CONTROL mothers. In these unmatched analyses, MDUI compared to CONTROL mothers had a greater lack of early prenatal care, a higher prevalence of smoking during pregnancy, and a higher prevalence of the death of a previously-born child. MDUI mothers were younger than CONTROL mothers at the index birth, and compared to CONTROLs were more likely to be underweight before pregnancy, to have less education, to be Missouri-born, to be unmarried, to omit their partner’s name on the birth record, and to have a partner of African American or other racial/ethnic descent. Partners of MDUI mothers were younger and had less education than partners of CONTROL mothers and were 2 to 7 times more likely to have DUIs or other alcohol-related driving convictions. The residential census tracts of MDUI mothers, compared to those of CONTROLs, were of lower socioeconomic status, less urban and more rural, with lower levels of education, household income, and professional employment, and with a smaller proportion of children residing with married parents.

Shown in Table 2 are the frequencies of the same measures after propensity score matching in the trimmed sample, within the quantiles used in the mortality analyses. Matching evened out the differences between MDUI and CONTROL groups for all but 2 variables in the top 40% of the propensity score distribution: smoking during pregnancy and partner DUI status (Table 2 and Supplemental Table 4).

Rates of child mortality in the trimmed sample were 1.9% for MDUIs and 0.9% for CONTROLS. Average age at child death was 2.3 (SD=4.4, min-max 0–17.6), with no significant difference between MDUIs and CONTROLs. A majority of child deaths occurred within the first 12 months of life (72.3%). Rates of maternal mortality for MDUIs and CONTROLs were 4.8% and 1.3%, respectively, with average age at death of 36.8 (SD=8.4, min-max: 15.7–61.1), when average offspring age was 9.8 (SD=4.5, min-max: 0–17.7). A majority (83%) of the MDUI sample had a DUI conviction after the birth of their child. Table 3 shows within-quantile associations of index child and maternal mortality with MDUI status. In the lower 60% of the distribution, there was no significant association of MDUI with either child or maternal mortality, although there was a trend for a positive association. In the higher-risk top 40% of the distribution, however, risk for child mortality was increased 75% for children of MDUI versus CONTROL mothers, and mortality of MDUI mothers was more than twice that of CONTROL mothers.

Table 3.

Prevalence of child and maternal mortality by predicted probability of MDUI

	Child Mortality		Maternal mortality
Quartile	N (%)	Odds Ratio (95% CI)	N (%)	Odds Ratio (95% CI)
1. Lowest probability MDUI (bottom .60 of distribution)
CONTROL	492 (0.74)		609 (0.91)
MDUI	4 (1.48)	2.01 (0.75–5.43)	5 (1.85)	2.04 (0.84–4.95)
2. Highest probability MDUI (top .40 of distribution)
CONTROL	470 (1.09)		826 (1.91)
MDUI	26 (1.94)	1.75 (1.17–2.61)	73 (5.44)	2.65 (2.07–3.40)

NOTE: regression in top 40% of propensity score distribution includes covariates for smoking during pregnancy and partner DUI status, due to incomplete matching of MDUI and CONTROL mothers on these variables

DISCUSSION

In this study of MDUI mothers, a number of factors related to risk for child and maternal health were identified, including lack of prenatal health care, smoking during pregnancy, single parenthood, a strong tendency to partner with men with probable alcohol problems, and poorer residential neighborhoods. Children of MDUI mothers with the highest prevalence of these factors were 75% more likely to die in infancy and by the time they would have been teenagers than were children of matched CONTROL mothers with similar risk profiles. Additionally, the MDUI mothers were themselves twice as likely as CONTROL mothers to die at a young age, after accounting for the multiple confounding factors associated with MDUI status. These striking findings confirm that severe maternal AUDs, as represented by multiple DUI convictions, affect the health and mortality of women and children above and beyond the many negative behavioral and environmental risks associated with severe maternal AUDs. MDUIs, as markers of risk for women and children, are readily identifiable, and represent an opportunity for targeted intervention efforts to block the intergenerational transmission of risk for AUDs and associated problems.

This is the first study to our knowledge to examine mortality in the children of MDUI mothers, apart from child deaths in automobile crashes due to alcohol-impaired drivers (49). In the highest-risk 40% of the propensity score distribution, we found an increased risk of death from any cause in childhood or adolescence among children born to MDUI mothers compared to children of CONTROLs, with a similar non-significant trend for the remainder of the distribution, suggesting that the severe AUDs with which MDUIs are associated play a prominent role in child mortality. The small number of child deaths in the lower-risk 60% of the distribution gives limited power to detect a significant effect of MDUI. It is also possible that the deaths of children born to MDUI mothers in the lower risk quantiles are accounted for by the measures included in the propensity score, so that MDUI contributes no additional risk, or that the greater socioeconomic advantage and education, and unmeasured variables such as much higher rates of health insurance coverage, represented in the lower risk quantiles diminish the relative influence of MDUI and severe maternal AUD on child risk of death.

Factors that affect the prenatal environment of the child were also included as control variables in this study. More than half of MDUI mothers reported smoking during pregnancy, and nearly 20% reported no prenatal care in the first trimester, both notable risks to the prenatal environment. In a meta-analysis of studies about smoking during pregnancy, smoking cessation during pregnancy was more likely among women who lived with a partner, were married, did not use alcohol before or during pregnancy, and who had adequate prenatal care (50), all factors lacking among MDUI mothers in the current study. Given the many factors affecting the prenatal and postnatal environments of children born to MDUI mothers, perhaps DUI history in pregnant women should be used as a marker of potential child risk, with mothers referred to programs to reduce pre- and postnatal drinking (51, 52) and to address factors such as partner drinking and psychiatric problems (53) which, in turn, can reduce the possibility of child harm.

The increase in maternal mortality in this study is consistent with registry-based data showing elevated death among women with DUIs compared to matched controls (31, 54). The current study increases confidence in a link between MDUI and early mortality since the propensity score methods adjusted for many correlated factors not included in earlier studies, including direct measurement of socioeconomic status via maternal sociodemographics from the birth record, linkage of birth address with census data and, importantly, partnering characteristics and pregnancy risk-behaviors.

The choice of reproductive partner, a variable not considered in most previous studies in DUI samples, can carry risks for children and mothers. MDUI mothers in this study were more likely to be unmarried and to leave their reproductive partner unnamed than were CONTROL mothers. MDUI mothers who did name a reproductive partner were 2 to 7 times as likely as CONTROL mothers to have a partner with one or more DUIs or alcohol-related driving convictions, exhibiting a strong tendency to assortative mating for AUDs. Potential risks to children associated with this partner choice include increased genetic liability to AUDs (55), increased risk of childhood abuse (56–58), unstable partnerships and concomitant lower socioeconomic status and associated risks such as unsafe neighborhoods and schools (59, 60). Risks to the mother include increased likelihood of continued problem substance use, exacerbation of correlated psychiatric disorders, and partner violence (61, 62).

This study contributes to the literature a focus specifically on women and their children. The few studies in DUI samples that have considered women separately from men have not considered the risks to children posed by maternal DUIs. The histories of men and women with DUIs, however, do suggest that their own childhood experiences were characterized by higher-than-average rates of parental substance use problems and separation, childhood physical and sexual abuse, and low socioeconomic status (21, 63–65). By sampling only women with children in the current study, and using birth records and census data to approximate the environments into which their children were born, we discovered that these environments were very similar to the childhood environments of DUI offenders that have been reported in the literature, whether assessed via self-report (21) or by use of extensive registry data (63–65). This raises the possibility of the intergenerational transmission not only of alcohol problems (66, 67) but of intra-family conflict and child maltreatment as well (68, 69). As treatment for women with DUIs is evaluated and improved (70, 71), the development of methods to lessen the impact of these intergenerational influences on their children is warranted.

Limitations

These results must be interpreted with certain limitations in mind. First, these data are cross-sectional, whereas the registry-based studies from European nations accessed decades of records about the driving histories of DUI subjects as well as their medical and family histories. Nonetheless, findings from these cross-sectional data are consistent with findings from those studies. Importantly, too, this study adds census data to measure birth neighborhood characteristics and socioeconomic status, whereas the European studies used only measures of marital and employment status to approximate social status. We lacked information about causes of death, and therefore cannot identify deaths directly related to drunk driving, but note that we view DUI convictions as an index of the severity of maternal alcohol problems, rather than as a risk factor distinct from the context of severe alcohol problems. Additionally, the concentration of deaths in the first year of life implies deficiencies in prenatal care and excess pregnancy risk behaviors may be key factors in child mortality. These results must be interpreted with the caution that administrative data do not include individual-level factors such as trauma and psychiatric histories that might confound the association between MDUIs and mortality. The data come from a single state in the US Midwest and may not reflect data from a broader geographic area. This study also had substantial missing data on the reproductive partners of MDUI mothers. The preponderance of missing data about partners in MDUI women itself is valuable information, and was accounted for in the propensity score analysis, with similar results in an analysis using imputed data. The linking of mother data to partner driving data has not to our knowledge been done in previous DUI studies, and the knowledge gained from this linkage, namely strong evidence for assortative mating, offsets concerns about missing partner data. Also, the sample includes white mothers only, and cannot be generalized to mothers of other race/ethnicities, although in Missouri the majority of women with MDUIs is white. By contrast, DUI studies in the Southwest contain substantial proportions of Hispanic participants due to the demographic composition of the state (18). Until similar studies are done in DUI women in other parts of the country, we cannot speculate on how race/ethnicity relates to child risk in DUI samples. The neighborhood characteristics included in the propensity analyses reflect characteristics at the child’s birth only, although future plans call for assessing neighborhoods later in life to understand the degree to which adverse environments may persist, worsen, or – in some cases – improve.

CONCLUSIONS

Women with multiple DUI convictions embody many risks to themselves and their children that are associated with elevated risk of their children’s mortality and their own. MDUIs are readily identifiable and represent an opportunity for targeted prevention and intervention efforts to block the intergenerational transmission of risk for AUDs and related adverse exposures.