Determinants of infant mortality for children of women prisoners: a longitudinal linked data study

We have used data from a large data linkage project to explore infant mortality within Indigenous and non-Indigenous children of mothers who have been exposed to the corrections system at different times in relation to their pregnancy. We compared infant mortality outcomes for children whose mothers had; (i) any period of imprisonment during pregnancy, (ii) imprisonment before (but not during) pregnancy, (iii) their first period of imprisonment after birth, (iv) community-based correctional orders (but no imprisonment), or (v) no corrections record at any time over the study period.

Mosley and Chen’s [12] analytical framework for the study of child survival was adapted for the present study. The basis of this framework is that broader determinants necessarily act through biological pathways, or mechanisms, which impact before, during and after pregnancy on the healthy development of the fetus and infant, and ultimately on infant mortality. There is evidence that adverse events experienced before and during pregnancy can impact on fetal development and result in increased risk of poor infant and childhood health outcomes [13‐17].

Our adaptation of this framework first groups together key demographic factors, including birth year, sex, Indigenous status, socioeconomic status, and geographical remoteness. The second grouping includes baseline pregnancy risk factors, such as multiple gestation, birth spacing < 18 months, maternal age, and parity, which are largely unmodifiable from the commencement of pregnancy [18]. The third grouping includes key pregnancy complications that might be a precursor of infant mortality, such as nutritional deficiencies, placental disorders, prematurity, and infection [19]. The last group includes other maternal factors and exposures which are known risk factors for infant mortality or may indicate maternal vulnerability or household dysfunction, such as substance use or mental health related service contacts, external causes of injury, and having other children in contact with the child protection system.

Data were obtained through the Western Australian Data Linkage System (WADLS). The WADLS uses highly-accurate computerised, probabilistic matching with clerical review to create linkages within and between administrative data collections across a range of Western Australian government agencies [20]. The Western Australian Data Linkage Branch conducted the linkage and provided de-identified data extracts. Records were extracted from the Midwives Notifications System, Birth Registrations, Death Registrations, Department of Justice, Hospital Morbidity Data System Collection (HMDC), Mental Health Information System (MHIS), and Department of Communities: Child Protection and Family Support (CPFS) data collections. These are all statutory State-wide data collections with complete coverage.

The Birth Registration and Midwives Notifications System data provided social and demographic characteristics of mothers and children at time of birth. Mortality data include all deaths registered in Western Australia. The Department of Justice data collection includes all custodial records for offenders held in Western Australian prisons and records for offenders on community-based correctional orders. Data excluded unsentenced individuals detained in police stations and courts, immigration detention centres, and mental health facilities. The HMDC includes all inpatient records for Western Australian public and private hospitals and day surgeries. The MHIS includes presentations to all inpatient and public community mental health services. The CPFS data include all reports of concerns for child welfare made to the child protection system and the details of investigations, protection applications and orders as well as placements in out-of-home care.

The Death Registrations, HMDC, and MHIS use the International Classification of Diseases (ICD), to classify cause of death, diagnosis, or reason for health service contact, respectively. For HMDC records, we obtained one code for the principal diagnosis of the episode of care, and up to four codes for external causes of episodes of care. Over the study period, the ICD 9th Revision with Clinical Modification (ICD-9-CM) [21] related to services contacts before July 1999, and service contacts from that date used ICD 10th Revision with Australian Modification (ICD-10-AM) [22].

The study population was drawn from a retrospective longitudinal cohort study of all liveborn children born in Western Australia from 1985 to 2011 whose biological mother was imprisoned at least once within 18-years after their birth. The cohort study population included a comparison group of children whose mother had no record of imprisonment from their date of birth to their 18th birthday, which was identified through the same data sources as the cohort and matched 3:1 to cohort children on Indigenous status, age and sex. Data on second-generation children, born between 1998 and 2014 to the female members of the cohort and comparison group, were also obtained.

Stillbirths (second-generation only) and infants with chromosomal abnormalities, identified through HMDC and death records, were excluded (Fig. 1). Erroneous records with multiple mothers or missing key information such as birthdate were removed. The final study population was restricted to children born from October 1985 to June 2013 (inclusive) to ensure pregnancy exposure and death data was available for all infants.

In total, there were 42,674 infants in the final study population, 37,469 from the first-generation (original cohort and comparison group) and 5205 from the second-generation (children of the original cohort and comparison group). Data from the birth and death registrations, midwives notifications records, and CPFS data were available for first- and second-generation children. Only first-generation children had HMDC record data. Mothers had corrections, HMDC, and MHIS data available.

The study population was categorised into: a) infants whose mothers had a record of imprisonment at any time (n = 7317 Indigenous; n = 3504 non-Indigenous); b) infants whose mothers had community-based correctional orders but no record of imprisonment (n = 5828 Indigenous; n = 653 non-Indigenous); and c) infants whose mothers had no record within any Department of Justice database (n = 12,817 Indigenous; n = 12,555 non-Indigenous).

Imprisonment records covered prison stays of any length of time, and included unsentenced remandees detained before trial as well as sentenced prisoners. Infants of mothers who had a prison record at any time over the study period were further categorised based on the timing of their mother’s imprisonment in relation to their pregnancy. The first group included infants whose mothers had any record of imprisonment during pregnancy. The second group included infants whose mothers had imprisonment records in the period before, but not during, pregnancy. The third group included infants whose mothers first record of imprisonment only occurred after the child’s birth. These groupings are shown in Fig. 1. Classifications were based on mother’s prison reception dates and the infant’s birth date.

Community-based sentences may involve treatment or vocational programs, community service, and place restrictions on offenders. Breach of conditions while on community orders may result in imprisonment. Accordingly, women with community-based correctional orders are sentenced offenders, but may differ to women given custodial prison sentences in terms of severity or frequency of their offending and other individual factors. They are not exposed to the prison environment which generally places more stringent conditions on offenders and has different implications for them and their families.

The proportion of infants in the various maternal corrections history sub-groups (Fig. 1) relate only to the study sample and do not reflect the prevalence of these groups across the whole Western Australian population.

Child month and year of birth was provided by the Midwives Notification System, or if missing from the Birth Registration data. As gestational age was not available, pregnancy start date was calculated as being nine months before the first day of the child’s birth-month.

Infant mortality was defined as the death of a live born child under one year of age [23]. Full date of death was provided in the death registration data, however, birth data were available only for month and year of birth. Accordingly, infant mortality was defined as death within 12-full months after birth. For example, for a child born in January 2000, death on or before 31 January 2001 would be determined within the category of infant mortality.

The Birth Registration and Midwives Notifications System data provides social and demographic characteristics of mothers and children at time of birth, including sex, socioeconomic status and geographical remoteness. Sex was taken primarily from Midwives Data, or if missing from Birth Data.

Area-based socio-economic status of infant’s place of residence at time of birth was assigned using the Socio-economic Indexes for Areas (SEIFA) Index of Relative Socio-economic Disadvantage [24]. The smallest area of SEIFA reporting is Collectors District (CD) level, which is approximately 250 households or less in rural areas. Missing CD scores were imputed with mean CD-score by postcode before using broader area scores of SEFIA available for Statistical Local Area or Local Government Area [25].

The Accessibility/Remoteness Index of Australia (ARIA) [26] was used to classify the geographical remoteness of infant’s place of residence at time of birth. ARIA is derived from the measure of place of residence to populated locations and key services and classified as major cities, inner regional, outer regional, remote, and very remote locations. For the current study major cities and inner regional areas were combined given both have greater accessibility of relevant services [26].

Indigenous status for infants and mothers in the study populations was ascertained from the Derived Indigenous Status Flag variable generated by the WADLS using best-practice algorithms, which assess individuals’ Indigenous status across multiple data collections to enhance accuracy [27].

Maternal birth date was determined using all available data sources. Maternal age was calculated as the age of mother at time of birth. Birth date was available for all siblings which enabled parity and duration of birth spacing to be determined. Multiple gestation pregnancies were also derived based on siblings having shared birth dates or through maternal or child HDMC data. Child-level HDMC data were not obtained for the second-generation children, however, as stillbirths were captured for the second-generation this assisted in identifying multiple gestation pregnancies.

The separate and combined effects of key pregnancy complications identified from maternal hospital records were evaluated. Pregnancy complications included the effects of infection-related hospitalisations, anaemia, diabetes, hypertension, preeclampsia, eclampsia, abruptio placentae, placenta previa, other placental disorders, premature rupture of membranes and renal disorders during pregnancy on infant mortality (Additional file 1) [19]. Complications were excluded due to low incidence or non-significance (p > 0.05), including hospitalisations for anaemia, diabetes, hypertension, preeclampsia, and eclampsia (Additional file 1).

Maternal substance use (including alcohol) and poisoning-related service contacts during pregnancy were identified from HMDC and MHIS data. Maternal hospital admissions for any injuries from external causes, excluding substance use, self-harm and poisoning-related contacts were identified during pregnancy. Maternal hospital admissions for mental and behavioural disorders, self-harm, and mental health service presentations, both excluding substance use and poisoning-related contacts, were identified during pregnancy from HMDC and MHIS data. Having an older sibling in contact with child protection services during the infant’s pregnancy was also identified using sibling’s child protection data.

All analyses were conducted using Stata Version 14.0. All analyses were stratified by Indigenous status of the infants.

Infant mortality rates (per 1000 population) were calculated for singleton infants by maternal corrections history. Prevalence of demographic and pregnancy-related risk factors were calculated for Indigenous and non-Indigenous populations. Log-binomial regression was used to calculate the Relative Risk of infant death for all univariate and multivariate analyses.

The strength of correlation between all variables of interest was assessed using Chi-square tests with Cramer’s V statistic. For variable pairs with a medium effect size (> 0.3) [28], one variable was excluded from further multivariate analysis on the basis of the univariate Relative Risk and level of statistical significance of each variable with infant mortality. Multivariate regression was then conducted for each grouping of variables (demographic factors, baseline pregnancy risk factors, pregnancy complications, and other maternal factors and exposures) separately with infant mortality, and variables that were not statistically significant (p < 0.05) were excluded from further analysis (Additional file 2). All remaining variables were entered with maternal corrections history into a full regression model (Model 1). Variables were removed by key groups; other maternal factors and exposures (Model 2), pregnancy complications (Model 3), baseline pregnancy risk factors (Model 4), leaving the combined effects of maternal corrections history and demographic factors with infant mortality. Goodness of model fit was assessed from the Akaike Information Criterion (AIC) value. There were insufficient numbers of non-Indigenous infant deaths whose mothers were imprisoned before or during pregnancy to undertake multivariate regression for non-Indigenous children.

Prevalence of key demographic and pregnancy-related risk factors, as determined from the univariate and multivariate analyses, were calculated for each maternal corrections history grouping and by Indigenous status.

For singleton births, Indigenous infants whose mothers were imprisoned at any time had a significantly higher risk of death than infants of mothers with community corrections orders alone, or with no corrections history (Table 1). Indigenous infants whose mothers were imprisoned during pregnancy had the highest rates of infant mortality (32.1 per 1000), compared to infants whose mothers were imprisoned either only before pregnancy (22.1 per 1000) or after birth (23.6 per 1000) (Table 1).

In non-Indigenous infants, those whose mothers were imprisoned before (9.4 per 1000) or during (8.2 per 1000) pregnancy had apparently lower rates of infant mortality than those whose mothers were imprisoned for the first time after their birth (14.2 per 1000) or who had community orders alone (12.6 per 1000). This difference was not significant and confidence intervals wide, possibly due to the low numbers of infant deaths within the sample of non-Indigenous children with mothers imprisoned before or during pregnancy.

As shown in Table 2, abruptio placentae and other placental disorders (excluding placenta previa) were associated with the highest risk of death for both Indigenous and non-Indigenous infants. Other important risk factors for Indigenous and non-Indigenous infants were low socioeconomic status, multiple gestation pregnancies, birth spacing < 18-months, having an older sibling in contact with the child protection system during an infant’s gestation, and maternal substance use or poisoning-related service contacts during pregnancy.

Maternal hospitalisations for premature rupture of membranes, or external causes of injury were important risk factors for infant mortality only for Indigenous infants. Male sex was also only a risk factor for infants in the Indigenous subgroup. Young maternal age and geographical location for those living in outer regional areas, were only significant risk factors in the non-Indigenous sample.

Factors not strongly associated with infant death included parity, infection related hospitalisations, or mental health related service contact (not related to substance use) in pregnancy, and hospital admissions for placenta previa.

Multivariate regression was only performed for Indigenous children because of the low number of deaths in the sample of non-Indigenous infants with incarcerated mothers.

Variables for maternal Indigenous status, geographical remoteness, maternal age, parity, placenta previa, infection-related hospitalisations, external injury-related hospitalisations, and mental health related service contacts in pregnancy (other than for substance use) were excluded based on their lack of association with infant mortality in our data (Additional file 2). The remaining variables were included and are shown in Model 1 (Table 3). In Models 2–4, variables were removed by group in order of; other maternal factors and exposures (Model 2), pregnancy complications (Model 3), baseline pregnancy risk factors (Model 4), leaving the combined effects of maternal corrections history and demographic factors on infant mortality in Model 4.

The best model fit was achieved with the inclusion of all variables (i.e., Model 1). Abruptio placentae and other placental disorders contributed the highest risk of infant death, followed by maternal imprisonment during pregnancy, and multiple gestation pregnancy. The effect of birth year remained stable across Models 1–4 and approximated to a 4% reduction in risk of infant mortality each year.

Table 4 shows the prevalence of pregnancy risk factors selected from the previous univariate and multivariate analyses, for Indigenous and non-Indigenous infants reported by their mothers’ contact with corrective services.

Infants of mothers with any corrections history had, in most instances, higher prevalence of maternal service contact in pregnancy for substance use (including poisoning), external causes of injury, mental health related service contacts, and sibling contact with the child protection system compared to infants whose mothers had no record with corrective services. Prevalence of these service contacts in pregnancy was highest for infants whose mothers were imprisoned during pregnancy, and higher for infants whose mothers were imprisoned before pregnancy compared to those imprisoned for the first time after birth or who had community corrections orders for these service contacts.

The proportion of infants born in areas of very low socioeconomic status was high for all groups of Indigenous children (> 20%) and did not differ by maternal corrections history. There was a higher proportion of low socioeconomic status for non-Indigenous children with any maternal corrections record (12–13%) compared to those with no maternal corrections history (4%).

There was a higher prevalence of a birth spacing of less than 18-months for infants with any maternal corrections record compared to no maternal corrections history, except the difference was not significant for those infants whose mothers were imprisoned during pregnancy. The prevalence of infants born to mothers aged less than 20 years was highest among infants whose mothers were first imprisoned after birth or had community-corrections orders.

For Indigenous children, maternal hospitalisation for abruptio placentae and other placental disorders (excluding placenta previa) was not different between infants by maternal corrections history, whereas prevalence of maternal hospitalisation for premature rupture of membranes was higher where there had been any record of maternal contact with the corrections system.

For the non-Indigenous children, hospitalisation of the mother for abruptio placentae and other placental disorders (excluding placenta previa) was higher where infants’ mothers were first imprisoned after birth or had community-correctional orders, and prevalence of hospitalisation for premature rupture of membranes was higher where infants’ mothers were imprisoned before or after pregnancy.

These results need to be considered within the context of the study’s limitations, for example, the power to detect relationships between infant mortality and maternal imprisonment before or during pregnancy in the non-Indigenous sample was limited by the small numbers of non-Indigenous infant deaths reported within those groups.

As gestational age was not available, pregnancy was taken to begin nine-months prior to month of birth for all infants. Accordingly, some records of exposure to maternal imprisonment, and service contacts related to substance use, mental health or child protection, that were attributed to having occurred during pregnancy may have occurred prior to pregnancy if gestation was shorter than nine-months. There is evidence, however, that adverse events within the preconception period (6–0 months before pregnancy) can increase infant mortality risk [49, 50]. Therefore any records of exposure misclassified as having occurred during pregnancy, may still have been expected to have an impact on infant mortality risk.

Additionally, without gestational age we have not been able to measure preterm birth (delivery before 37 weeks of gestation), which is associated with infant mortality [51]. However, preterm birth shares many of the same risk factors for infant mortality identified within the study including low socioeconomic status, maternal age, maternal stress, infections, and multiple gestation pregnancies [51]. The study has provided the first evidence of an association between maternal incarceration in pregnancy and infant mortality, further research is needed investigate the impact of factors not able to be measured in the current study, notably preterm birth, antenatal care and caesarean section rates.

Administrative data alone cannot fully capture occurrences of heavy drinking or substance use, mental illness, or injuries resulting from domestic violence, within pregnancy [52]. In addition, the study only obtained the primary diagnosis code (not co-diagnoses) for hospital and mental health service records. Consequently, the associations observed in our study with drinking or other substance use in pregnancy, mental illness or injuries are likely under ascertained.

Hospital data were not obtained for second-generation children. While explanatory variables based on hospital data were primarily taken from maternal hospital records (available for all mothers), there may have been some missed cases of chromosomal abnormalities, and substance use related service contacts in pregnancy, for the second-generation children.