Vulnerability-related disparities in early childhood healthcare costs in the Northern Territory, Australia: a population-based retrospective cohort study

Abstract

Objective

Early childhood health and economic burdens are disproportionately distributed across population groups. We examined the vulnerability-related (First Nations status and remoteness) disparities of early childhood healthcare costs from birth to age 5 years and quantified the contribution of adverse perinatal outcomes, such as preterm birth (PTB), small-for-gestational-age (SGA) and low birthweight, to these disparities in the Northern Territory (NT), Australia.

Design and participants

We conducted a population-based retrospective cohort study of all singleton live births in the NT, Australia, between 1 July 2000 and 30 June 2016, using the Perinatal Trends dataset. Birth records were linked to hospitalisation, emergency department (ED) and national hospital costing data and followed to age 5.

Outcomes and analysis

Healthcare costs, adjusted to 2024 Australian Dollars (AUD), from a health system perspective, included hospitalisation and ED costs. A generalised linear model with log link and gamma distribution was fitted, and Oaxaca-Blinder decomposition was used to partition cost differences into explained (differences in population characteristics) and unexplained (differential effects/responses) components.

Results

The overall median hospitalisation costs per child were AUD 9608 (IQR: 4873 to 20 681) and ED costs AUD 2156 (IQR: 1075 to 4274) in the first 5 years. Among children with PTB and SGA, median hospitalisation costs were AUD 30 527 (IQR: 16 765 to 54 365) for First Nations and AUD 16 098 (IQR: 7577 to 28 048) for non-First Nations children; within this subgroup, costs were AUD 29 614 (IQR: 15 085 to 51 400) in remote areas compared with AUD 19 776 (IQR: 9620 to 33 263) in urban areas. After adjustment, mean hospitalisation costs were 98% higher for First Nations children (AUD 23 223 vs 11 716; difference AUD 11 507, 95% CI 10 798 to 12 215) and 19% higher for children in remote areas (AUD 19 454 vs 16 322; difference AUD 3132, 2479 to 3786); corresponding differences in mean ED costs were 67% (AUD 1866, 95% CI 1750 to 1982) and 17% (AUD 557, 95% CI 460 to 655). Remoteness-related disparities in hospitalisation costs were largely explained by child characteristics (85.7%; βE=−0.549, 95% CI −0.581 to −0.517), while disparities by First Nations status were primarily driven by differential cost responses (71%; βc=−0.606, 95% CI −0.651 to −0.561). For ED costs, characteristics explained 42% of the remoteness-related gap and 3.4% of the gap by First Nations status.

Conclusions

Our findings highlight that vulnerability-related disparities in healthcare costs were substantial, with adjusted costs 67% to 98% higher for First Nations children and 17% to 19% higher for children in remote areas, and driven by distinct mechanisms. Thus, strategies addressing structural disadvantage and inequitable experiences are needed to reduce avoidable healthcare costs.

STRENGTHS AND LIMITATIONS OF THIS STUDY.

• The study benefits from two decades of population-level linked data, allowing examination of disparities in early-life vulnerability and healthcare costs across a long period.

• By using two dimensions of vulnerability (First Nations status and remoteness) as inequality stratifiers, the study shows how early childhood healthcare cost disparities vary across population groups rather than along a single axis.

• Cost estimates were limited to hospital and emergency department (ED) care; data on primary care, specialist outpatient services and Medicare and Pharmaceutical Benefits Scheme items were unavailable, so total healthcare costs are likely underestimated.

• Infants with no hospital or ED presentations were not included, which may underestimate average birth-admission costs and overstate cost differences between vulnerable and non-vulnerable children at a population level.

• Diagnostic coding may lack clinical specificity, and some cost components, such as patient transport from remote communities and interstate transfers for specialist care, were not captured, despite being substantial in remote settings.

Introduction

Perinatal morbidity, including preterm birth (PTB), small-for-gestational-age (SGA), low birthweight (LBW) and congenital anomalies as well as mortality, such as stillbirth and early neonatal deaths, are unevenly distributed across Australia, with relatively high prevalence in the Northern Territory (NT). In 2022, the NT reported 12.3% PTBs, 9.7% LBW, 13.2% SGA births and 14.7 stillbirths per 1000 total births,¹ compared with the national averages of 8.3% PTBs, 6.5% LBW, 9.5% SGA births and 8 stillbirths per 1000 total births. The burden of these adverse outcomes is also inequitably distributed across population groups, particularly affecting those with low socioeconomic status, residents of remote areas,² Aboriginal and Torres Strait Islander populations (herein respectfully referred to as First Nations)³ and linguistically diverse communities, all of whom face significantly higher burdens. Disparities also persist in child health and developmental outcomes, with inequalities linked to socioeconomic status, gender and jurisdictional differences.^{4 5} Empirical evidence from other settings has also reported clear disparities in perinatal morbidity and mortality across socioeconomic strata, geographic regions and social classes.^6–11 These disparities not only highlight inequities in health outcomes but also translate into increased demands on the healthcare system, particularly among vulnerable populations.

Despite the Australian healthcare system’s aims to provide equitable and accessible health services, including efforts to support the goals of closing the gap for vulnerable population groups,¹² adverse perinatal outcomes are associated with disproportionately higher rates of unplanned healthcare utilisation.¹³ This, in turn, places a substantial burden on the health system, through increasing healthcare resources. The high rates of hospitalisations and emergency department (ED) visits, often accompanied by longer lengths of stay and increased resource use associated with adverse perinatal outcomes, further exacerbate the challenges to the health system. Previous studies also reported higher rates of unscheduled healthcare use among children with disabilities,^14–18 preterm children^19–21 and other morbidities.^{22 23} These challenges in unplanned healthcare utilisation are further heightened by inequalities in child health services, which disproportionately affect vulnerable populations.

Studies have demonstrated vulnerability-related inequality in child health services across various settings, particularly among children with adverse perinatal outcomes. For example, previous studies in Queensland, Australia, reported significant disparities in health service utilisation among disadvantaged children.^{24 25} Similarly, studies in the USA²⁶ and the UK^{27 28} have highlighted racial and socioeconomic disparities in neonatal intensive care among preterm children. Across these studies, factors such as socioeconomic disadvantage, residence in regional or remote areas, First Nations status and other social determinants of health⁵ drive unplanned and inequitable healthcare use within these groups.

These disparities extend beyond immediate healthcare needs, influencing long-term health outcomes and imposing sustained pressures on individuals, the healthcare system and society.^{4 25 29 30} Furthermore, inequalities in adverse perinatal outcomes also contribute to health and economic divides at multiple levels. The economic cost of adverse perinatal outcomes is particularly pronounced for those at greater risk, including individuals facing socioeconomic challenges, First Nations populations and those living in remote areas. However, evidence on the variation in vulnerability-related health system costs and the contribution of adverse perinatal outcomes to the disparities is scarce, particularly in the NT, where the rates are higher than in other Australian States and Territories.¹

Understanding disparities in healthcare costs across vulnerable population groups and the contribution of adverse perinatal outcomes is crucial for informing policies and interventions aimed at reducing or eliminating these burdens. This includes addressing inequalities explained by socioeconomic status, residence and First Nations status and the unequal distribution of health resources. Identifying such health system cost disparities will enable the development of targeted strategies to ensure equitable resource allocation and access to care, helping to close vulnerability-related gaps. Furthermore, cost estimates from this study can provide a base for future studies on the distributional cost-effectiveness analysis of perinatal health interventions, emphasising both equity and cost-effectiveness for vulnerable populations.^31–33 We, therefore, examined the vulnerability-related disparities, defined by First Nations status and remoteness, in early childhood healthcare costs from birth to age 5 years, from a health system perspective in NT, Australia, and quantified the contribution of adverse perinatal outcomes to these cost disparities.

Methods

Study design and cohort

We conducted a population-based retrospective cohort study using data on births in the NT, Australia, from 1 July 2000 to 30 June 2016. The study included all singleton live births. Adverse perinatal outcomes were defined as the presence of any of the following conditions: PTB (birth before 37 completed weeks of gestational age), SGA (birthweight below the 10th percentile for gestational age), LBW (birthweight less than 2500 g) or congenital anomalies at birth.

Data sources and linkage

We linked all singleton live births recorded in the NT Perinatal Trends dataset between 1 July 2000 and 30 June 2016 to hospitalisation and ED presentation activity and to national hospital cost data, and followed each child to age 5 years. Cost data were drawn from the National Hospital Cost Data Collection (NHCDC), using average Australian Refined Diagnosis Related Groups (AR-DRGs) for inpatient costs and Urgency Related Groups (URG) for ED presentation costs, obtained from the Independent Health and Aged Care Pricing Authority.³⁴ In the NT, care has been delivered across the six public hospitals operated by NT Health. Socioeconomic status was measured using the Socioeconomic Index for Areas (SEIFA), sourced from the Australian Bureau of Statistics.³⁵

Cost valuation approaches

We estimated the healthcare costs per child, covering hospitalisation and ED presentations only, from birth to age 5 years, using an incidence-based approach, in which costs were accumulated for each child in the cohort over the 5-year follow-up period. The cost estimation was performed from a health system perspective (covering both national and NT Health funding), including the direct and overhead costs of admissions and ED presentations. Direct costs are those tied directly to healthcare services delivered, while overhead costs are the indirect costs of care, such as facility and utility services.

Cost data by AR-DRG and URG classifications are publicly available from 2012 onward.³⁴ To estimate AR-DRG and URG costs for years before 2012, we deflated the 2012 values using the national medical services-related consumer price index (CPI). All costs were then expressed in June 2024 Australian dollars (AUD) using the health-related CPI.³⁶

Inequality stratifiers and covariates

In our analysis, we operationalised vulnerability using two variables, remoteness and First Nations status, which were used as inequality stratifiers. We additionally examined socioeconomic status (measured by SEIFA) as a stratifier in a subgroup analysis, as this measure was available only for births from 2013 to 2016.

Residence was classified based on remoteness, with areas categorised as either urban or remote. The second vulnerability-related variable, First Nations status, was classified based on maternal First Nations status. The Perinatal Trends datasets recorded First Nations status for both mother and baby. Because a notable proportion of children had an ‘unknown’ classification (3.8% in the hospitalisation cohort and 7.1% in the ED cohort), we based each child’s First Nations status on the mother’s classification, which categorised all children as First Nations or non-First Nations with no unknown status. Finally, socioeconomic status was assessed using the SEIFA, a composite measure incorporating variables such as income, education, employment, occupation, housing and family structure. For this study, we specifically used the Index of Relative Socio-economic Advantage and Disadvantage (IRSAD), which captures both socioeconomic advantage and disadvantage at the Statistical Area Level 2 (SA2). SA2s ar medium-sized areas in the Australian Statistical Geography Standard, with populations generally ranging from 3000 to 25 000.³⁷ Each SA2 is assigned a SEIFA score and categorised into quintiles from 1 (most disadvantaged) to 5 (most advantaged). However, since SA2 was not available before 2013 in the Perinatal Trends dataset (women were not assigned to specific SA2s before 2013), we restricted the analysis of socioeconomic disparities to birth cohorts from 2013 to 2016.

We included a range of candidate covariates. Child characteristics measured at birth included sex, birthweight, gestational age, diagnosis of congenital malformation at birth, year of birth, mode of delivery, birth hospitalisation outcome (mode of separation) and Apgar scores at 1 and 5 min. Maternal characteristics included parity, maternal age and pregnancy complications, such as antepartum haemorrhage, hypertension and diabetes. Health service-related factors included length of stay for admission and duration of ED presentations and the number of maternal antenatal care visits.

Statistical analysis

We first summarised the characteristics of the cohort using descriptive statistics. Categorical variables were reported as frequencies and percentages with accompanying 95% CIs, and continuous variables as the mean with SD or the median with IQR, depending on the distribution.

Our main analysis followed a two-step decomposition framework. First, we estimated mean differences in early childhood healthcare costs between vulnerable and non-vulnerable children. In our analysis, as healthcare costs were positively skewed, we used a generalised linear model (GLM) with a gamma distribution and log link. Models were adjusted for child and maternal characteristics measured at birth (sex, parity, maternal age, maternal smoking during pregnancy and year of birth) and for health service-related characteristics, including length of stay (LoS) following admission and ED presentations. Second, the estimated cost differences were decomposed into explained and unexplained components. For the First Nations (First Nations vs non-First Nations) and remoteness (remote vs urban), we used the Oaxaca-Blinder decomposition method as adapted for non-linear models by Powers et al,^{38 39} to quantify the contributions of each explanatory variable to the cost difference. For socioeconomic-related inequality, examined in a subgroup analysis, we used a Wagstaff modified concentration index⁴⁰ and a corresponding Wagstaff decomposition analysis, which identifies the contributors to the observed cost gaps across the SEIFA quintiles.

Adverse perinatal outcomes, including PTB, SGA and LBW, were included as key explanatory variables to assess their contribution to the disparities. In our covariate selection, we used least absolute shrinkage and selection operator (LASSO) regularised regression technique with an L1 penalty (α=1) to prevent overfitting. The optimal penalty parameter (λ) was chosen by 10-fold cross-validation using the one-standard-error rule (λ.1se), which favours a more parsimonious and interpretable model.

Before running the decompositions, we examined how strongly the two inequality stratifiers, First Nations status and remoteness, were related, using a χ² test and Cramér’s V. We checked whether this affected the stability of the adjusted models by calculating variance inflation factors from the pooled GLM (online supplemental table 1). In each decomposition, the other stratifier was kept in the model as a covariate, so that each disparity was estimated while holding the other dimension constant.

The Oaxaca-Blinder method decomposes cost disparities into two components. The explained component (Endowment) reflects group differences in the distribution of observed characteristics and quantifies how much of the healthcare cost disparity was attributed to these differences. The unexplained component (Coefficient) captures differences in how the same characteristics translate into costs. Additionally, the unexplained component reflects structural factors, including differential effects of the same characteristics, variations in healthcare access, service provision or unmeasured factors contributing to cost disparities.

Our decomposition was applied to two contrasts: remote versus urban and First Nations versus non-First Nations. SEs for the decomposition components were obtained by bootstrapping and statistical significance was assessed at p<0.05. The full mathematical derivation of the decomposition is provided in the Technical Supplement (online supplemental file 1).

Missing data handling and subgroup analysis

Descriptive analyses used all children with complete data on the outcome and the inequality stratifiers, with subtotals where covariates were missing. For the final regression and decomposition models, we used complete-case analysis. In the final hospitalisation model, 879 of 31 169 records (2.8%) were excluded owing to missing or unrecorded covariate values, leaving 30 290; in the ED model, 943 of 42 139 (2.2%) were excluded, leaving 41 196 complete cases for the final modelling and decomposition analysis. Missingness was mostly concentrated in congenital malformation status at birth, mode of separation following birth admission, length of hospital stay and number of antenatal visits. Missingness by covariate is reported in online supplemental table 2. We compared birth cohort characteristics between children linked to a hospitalisation and/or ED presentation and those who were not linked.

Stratified analyses were performed by First Nations status (First Nations vs non-First Nations) and by remoteness (urban vs remote). We conducted a further stratified analysis across four mutually exclusive groups: (1) non-First Nations urban, (2) non-First Nations remote, (3) First Nations urban and (4) First Nations remote. In addition to the median cost per stratum, we estimated the adjusted mean costs for each stratum from a gamma GLM with log link and robust SEs. We also conducted a subgroup analysis of socioeconomic-related disparities in early childhood healthcare costs among the 2013 to 2016 birth cohorts.

Patient and public involvement

Patients and/or the public were not involved in the design, or conduct, or reporting or dissemination plans of this research.

Results

We included 31 169 children linked to hospitalisation records (including birth hospitalisations) and 42 139 children linked to ED presentations (online supplemental figure 1). Of children in the hospitalisation cohort, 47.5% were from remote areas and 52.2% were First Nations. For ED presentations, 38.5% were from remote areas and 40.8% were First Nations children (online supplemental table 3). Cross-classification produced unevenly sized strata in both hospitalisation and ED cohorts. In the hospitalisation cohort, 12 634 children (40.5%) were non-First Nations and urban, 2261 (7.3%) non-First Nations and remote, 3732 (12.0%) First Nations and urban and 12 542 (40.2%) First Nations and remote. In the ED cohort, the corresponding numbers were 21 143 (50.2%), 3819 (9.1%), 4785 (11.4%) and 12 392 (29.4%). A higher proportion of children linked to a hospitalisation and/or ED presentation were born preterm, SGA and LBW than those who were not (online supplemental tables 45).

Hospitalisations and ED presentations cost

The overall median hospitalisation costs per child were AUD 9608 (IQR: AUD 4873 to 20 681) from birth to age 5 years. The median 5-year hospitalisation costs ranged from AUD 6511 (IQR: AUD 4053 to 11 768) in the non-First Nations urban to AUD 16 599 (IQR: AUD 8314 to 31 542) in First Nations remote, with a similar increase in adjusted mean across these strata. The adjusted cost difference associated with remote residence was AUD 1121 (95% CI AUD 441 to 1801) among non-First Nations children compared with AUD 5186 (95% CI AUD 4143 to 6230) among First Nations children, corresponding to relative differences of 10% and 25%, respectively (table 1). All six pairwise differences between strata were statistically significant except for First Nations remote and urban in ED costs (online supplemental file 2).

Table 1. Cost per child in the first 5 years, by First Nations status and remoteness.

Stratum	n (%)	Median cost (IQR, AUD)	Adjusted mean cost (95% CI)	Difference vs ref (95% CI)	Cost ratio (95% CI)
Hospitalisation costs
Non-First Nations, urban (ref)	12 634 (40.5)	6511 (4053 to 11 768)	10 853 (10 553 to 11 154)	Ref.	1.00
Non-First Nations, remote	2261 (7.3)	7245 (4059 to 13 325)	11 975 (11 350 to 12 600)	1121 (441 to 1801)	1.10 (1.04 to 1.17)
First Nations, urban	3732 (12.0)	11 440 (5485 to 23 902)	20 436 (19 545 to 21 327)	9582 (8622 to 10 543)	1.88 (1.79 to 1.99)
First Nations, remote	12 542 (40.2)	16 599 (8314 to 31 542)	25 622 (25 017 to 26 228)	14 769 (14 051 to 15 486)	2.36 (2.27 to 2.46)
ED costs
Non-First Nations, urban (ref)	21 143 (50.2)	1647 (851 to 3106)	2529 (2489 to 2570)	Ref.	1.00
Non-First Nations, remote	3819 (9.1)	2234 (1084 to 4338)	3505 (3387 to 3622)	975 (853 to 1 097)	1.39 (1.34 to 1.44)
First Nations, urban	4785 (11.4)	3084 (1449 to 6244)	4893 (4730 to 5057)	2364 (2 193 to 2 536)	1.93 (1.86 to 2.01)
First Nations, remote	12 392 (29.4)	3011 (1392 to 6181)	4842 (4733 to 4952)	2313 (2 191 to 2 435)	1.91 (1.86 to 1.97)

Median and IQRs describe the observed distribution; adjusted means, differences and cost ratios are from the GLM model.

AUD, Australian Dollars in June 2024 value; CI, Confidence Interval; ED, emergency department; GLM, generalised linear model; IQR, interquartile range; ref, Reference group.

The median hospitalisation costs per child with PTB and SGA were AUD 30 527 (IQR: 16 765 to 54 365) among First Nations children and AUD 16 098 (IQR: 7577 to 28 048) among non-First Nations children. Among preterm SGA children, hospitalisation costs were higher for those from remote areas (median AUD 29 614, IQR: 15 085 to 51 400) than for those from urban areas (AUD 19 776, IQR: 9620 to 33 263) (online supplemental figure 2, online supplemental table 6).

We found that the overall median cost of ED presentations per child was AUD 2156 (IQR: 1075 to 4274) in the first 5 years. Median ED costs were lowest among non-First Nations urban children (AUD 1647, IQR: 851 to 3106) and highest among First Nations urban children (AUD 3084, IQR: 1449 to 6244), with First Nations remote children at AUD 3011 (IQR: 1392 to 6181) (table 1). The median ED costs per child born preterm and SGA was AUD 3861 (IQR: 2067 to 8079) among First Nations children, and AUD 2172 (IQR: 1024 to 4670) among non-First Nations children. Among preterm children with SGA, the median ED presentation cost was AUD 3845 (IQR: 1993 to 7740) for those from remote areas and AUD 2590 (IQR: 1244 to 5474) for those from urban areas (online supplemental figure 3, online supplemental table 7).

Decomposition of vulnerability-related disparities in hospitalisation costs

We found differences in remoteness and First Nations-related early childhood hospitalisation costs. Accordingly, 85.7% (${\beta }_E$= −0.549, 95% CI −0.581 to −0.517) of the remoteness-related disparities in hospitalisation costs were attributed to differences in children’s observed characteristics (endowment effect), while 71% (βс= −0.606, 95% CI −0.651 to −0.561) of the First Nations-related hospitalisation cost gap was due to differences in how these characteristics influenced hospitalisation costs (coefficient effect) (table 2).

Table 2. Summary of decomposition components of hospitalisation cost disparities by remoteness and First Nations status, NT, Australia (2000 to 2020).

Vulnerability-related factors	Decomposition components	β (95% CI)	Percentage contribution to variation in cost
Remoteness	Difference due to composition of characteristics (endowments, E)	−0.549 (−0.581 to −0.517)	85.7
	Differences due to coefficients (effect of characteristics, C)	−0.092 (−0.131 to −0.052)	14.3
	Total difference	−0.641	100.0
First Nations	Differences due to composition of characteristics (endowments, E)	−0.248 (−0.285 to −0.211)	29.0
	Differences due to coefficients (effect of characteristics, C)	−0.606 (−0.651 to −0.561)	71.0
	Total difference	−0.854	100.0

In the remoteness-related disparities, First Nations status was the largest contributor to the explained difference in hospitalisation costs. Equalising the proportion of First Nations children in remote areas to urban levels would reduce the remoteness-related cost gap by 66.6% ($exp\left({\beta }_E\right)$=0.652, 95% CI: 0.629 to 0.676). Newborns from remote areas have longer birth hospital stays; aligning their LoS with that of urban newborns would reduce the gap by 1.7% ($exp\left({\beta }_E\right)$ = 0.989, 95% CI: 0.985 to 0.993), and aligning LoS during readmission would reduce it by 6% ($exp\left({\beta }_E\right)$ = 0.962, 95% CI: 0.954 to 0.970). Similarly, reducing disparities in postnatal transfers to other healthcare facilities would narrow the cost gap by 2.7% ($exp\left({\beta }_E\right)$ =0.983, 95% CI: 0.979 to 0.987). Equalising parity levels between remote and urban areas would further reduce the gap by 1.3% (${\beta }_E$= 0.992, 95% CI: 0.986 to 0.998).

Most unexplained disparities of hospitalisation costs linked to remoteness were attributed to First Nations status, gestational age, birthweight, LoS following birth hospitalisation, parity, and delivery by emergency caesarean section (CS). Equalising the proportion of parity between remote and urban areas would reduce the unexplained component by 3.9% (exp(${\beta }_C$)= 1.026, 95% CI: 1.001 to 1.051). Furthermore, First Nations status and delivery by emergency CS contributed 4.9% and 2.5%, respectively, to the unexplained differences in remoteness-related hospitalisation costs (table 3, online supplemental table 8).

Table 3. Remoteness-related disparities of early childhood hospitalisation cost in the first 5 years, NT, Australia (2000 to 2020).

Variable	Exp (${\beta }_E$) (95% CI)	Proportional cost change (%)	Percentage contribution to variation in cost (absolute)	Exp (${\beta }_C$) (95% CI)	Proportional cost change (%)	Percentage contribution to variation in cost (absolute)
Gestational age, weeks	0.993 (0.988 to 0.998)*	−0.72	1.12	0.912 (0.432 to 1.924)	−8.85	14.45
Birthweight, grams	0.994 (0.987 to 1.001)	−0.63	0.99	0.880 (0.736 to 1.052)	−11.99	19.93
LoS during birth hospitalisation, days	0.989 (0.985 to 0.993)*	−1.08	1.70	1.008 (0.997 to 1.019)	0.81	1.26
Age of mother, years	0.973 (0.959 to 0.987)*	−2.70	4.27	0.889 (0.770 to 1.027)	−11.09	18.34
Female sex at birth	1.001 (1.00 to 1.002)	0.08	0.12	0.992 (0.970 to 1.015)	−0.80	1.25
First Nations	0.652 (0.629 to 0.676)*	−34.77	66.64	0.969 (0.952 to 0.985)*	−3.14	4.97
Appropriate birthweight for gestational age	0.998 (0.997 to 0.999)*	−0.19	0.30	0.993 (0.949 to 1.039)	−0.72	1.12
LoS during readmission, days	0.962 (0.954 to 0.970)*	−3.83	6.10	1.009 (0.975 to 1.045)	0.94	1.45
Congenital malformation status at birth
Diagnosed	0.999 (0.998 to 1.00)	−0.07	0.10	1.003 (1.001 to 1.005)*	0.29	0.45
Under investigation	1.000 (0.999 to 1.001)	−0.04	0.06	1.003 (0.999 to 1.007)	0.31	0.49
Birth hospitalisation outcome
Transferred	0.983 (0.979 to 0.987)*	−1.72	2.72	1.004 (1.001 to 1.007)*	0.42	0.65
Refused care	0.998 (0.996 to 0.999)*	−0.23	0.36	1.000 (0.999 to 1.001)	−0.02	0.03
Other	0.998 (0.997 to 0.999)*	−0.19	0.29	1.017 (1.006 to 1.028)*	1.71	2.64
Parity	0.992 (0.986 to 0.998)*	−0.80	1.24	1.026 (1.001 to 1.051)*	2.56	3.94
Apgar score at 1 min	1.000 (0.999 to 1.001)	0.04	0.06	0.982 (0.937 to 1.028)	−1.85	2.91
Method of delivery
Emergency CS	0.999 (0.998 to 1.000)	−0.08	0.12	1.016 (1.002 to 1.030)*	1.61	2.49
Year of birth (2001)	1.001 (1.00 to 1.002)	0.07	0.11	0.996 (0.989 to 1.002)	−0.42	0.66
Constant				1.260 (0.588 to 2.697)	26.00	35.99

Coefficient (unexplained components).

^*Statistically significant at p value<0.05.

^†Apgar score measured at 1-minute post-birth; values range from 0 to 10, with lower scores indicating poorer neonatal condition.

CS, caesarean section; Endowments, (explained components); LoS, length of hospital stay; βC, β value for coefficient; βE, β value for endowments.

The largest contributor to the explained component in First Nations-related disparities in hospitalisation costs was remoteness (11.8%, exp (${\beta }_E)$) = 0.902, 95% CI: 0.878 to 0.926), followed by maternal age and LoS following birth hospitalisation. Equalising the LoS following birth hospitalisation among First Nations and non-First Nations would reduce the hospitalisation gap by 2.5% ($exp\left({\beta }_E\right)$ = 0.983, 95% CI: 0.978 to 0.988), and equalising the LoS during readmission would reduce the gap by 3.7% ($exp\left({\beta }_E\right)$ = 0.963, 95% CI: 0.954 to 0.972). Most disparities linked to First Nations status were attributable to unexplained components. Gestational age, LoS during readmission, remoteness and transfer to other health facilities following birth hospitalisation contributed to the unexplained component (table 4, online supplemental table 9). Gestational age accounts for 70% of the unexplained First Nations hospitalisation cost gap (exp(βc) = 0.448, 95% CI: 0.212 to 0.947). Furthermore, in subgroup analyses restricted to births from 2013 to 2016, we found higher costs disproportionately incurred for children from socioeconomically disadvantaged families (online supplemental file 3).

Table 4. First Nations-related disparities of early childhood hospitalisation cost in the first 5 years, NT, Australia (2000 to 2020).

Variable	Exp (${\beta }_E$) (95% CI)	Proportional cost change (%)	Percentage contribution to variation in cost (absolute)	Exp (${\beta }_C$) (95% CI)	Proportional cost change (%)	Percentage contribution to variation in cost (absolute)
Gestational age, weeks	0.995 (0.989 to 1.001)	−0.52	2.12	0.448 (0.212 to 0.947)*	−55.2	70.04
Birthweight, grams	0.990 (0.982 to 0.999)*	−0.98	0.98	0.941 (0.786 to 1.127)	−5.9	6.00
LoS during birth hospitalisation, days	0.983 (0.978 to 0.988)*	−1.74	2.47	0.984 (0.956 to 1.013)	−1.6	0.09
LoS during readmission, days	0.963 (0.954 to 0.972)*	−3.67	3.67	1.041 (1.008 to 1.076)*	4.1	4.30
Age of mother, years	0.963 (0.945 to 0.981)*	−3.73	4.86	0.920 (0.799 to 1.060)	-8	5.45
Female sex at birth	1.002 (1.000 to 1.003)*	0.16	0.23	0.994 (0.971 to 1.017)	−0.6	0.33
Remoteness
Remote	0.902 (0.878 to 0.926)*	−9.84	11.75	0.983 (0.973 to 0.993)*	−1.7	1.63
Appropriate birthweight for gestational age	0.996 (0.994 to 0.999)*	−0.38	0.53	1.009 (0.962 to 1.059)	0.9	2.50
Congenital malformation status at birth
Diagnosed	0.999 (0.998 to 1.000)	−0.09	0.11	1.002 (1.00 to 1.004)*	0.2	0.23
Under investigation	0.999 (0.998 to 1.000)	−0.09	0.08	1.005 (1.001 to 1.009)*	0.5	0.50
Birth hospitalisation outcome
Transferred	0.989 (0.987 to 0.993)*	−1.01	1.24	1.003 (0.999 to 1.006)	0.3	0.43
Refused care	0.997 (0.995 to 0.999)*	−0.30	0.31	1.000 (0.999 to 1.001)	0	0.01
Others	0.996 (0.995 to 0.998)*	−0.34	0.45	1.010 (1.000 to 1.021)*	1	1.18
Parity	0.985 (0.976 to 0.994)*	−1.53	1.86	1.021 (0.998 to 1.045)	2.1	2.83
Apgar score at 1 min†	1.001 (0.998 to 1.004)	0.04	0.04	0.970 (0.928 to 1.014)	-3	5.35
Method of delivery
Emergency CS	0.999 (0.998 to 1.000)	−0.08	0.13	1.014 (1.001 to 1.027)*	1.4	1.42
Year of birth (2001)	0.999 (0.998 to 1.001)	−0.01	0.01	0.999 (0.993 to 1.007)	−0.1	0.46
Constant				1.358 (0.657 to 2.89)	35.8	3.81

^*Statistically significant at p value<0.05

^†Apgar score measured at 1-minute post-birth; values range from 0 to 10, with lower scores indicating poorer neonatal condition.

CS, caesarean section; LoS, length of hospital stay; βC, β value for coefficient; βE, β value for endowments.

Decomposition of vulnerability-related disparities in ED presentation costs

We found that 42.3% (${\beta }_E$ = -0.204, 95% CI: −0.233 to −0.175) of the remoteness-related disparities in ED presentation costs were attributable to differences in observed characteristics (endowment effect). In contrast, 96.6% (${\beta }_c$= -0.635, 95% CI: −0.665 to −0.604) of the First Nations-related ED presentation cost disparities resulted from differences in how these characteristics influenced costs (coefficient effect).

In the remoteness-related ED presentation cost disparities, First Nations status was the largest contributor to the explained difference. Equalising the proportion of First Nations children in remote areas to that of urban areas would reduce the remoteness-related ED presentation cost gap by 8.7% ($exp\left({\beta }_E\right)$ = 0.827, 95% CI: 0.805 to 0.849). Additionally, differences in maternal age, postnatal transfer following birth hospitalisation, congenital malformations, and births occurring after 2013 contributed to this gap. The majority of the unexplained component of the remoteness-related disparity in ED presentation costs was attributable to gestational age, LoS during ED presentation, maternal age and First Nations status (table 5, online supplemental table 10).

Table 5. Remoteness-related disparities in ED presentation cost of children in the first 5 years, NT, Australia (2000 to 2020).

Variable	Exp (${\beta }_E$) (95% CI)	Proportional cost change (%)	Percentage contribution to variation in cost (absolute)	Exp (${\beta }_C$) (95% CI)	Proportional cost change (%)	Percentage contribution to variation in cost (absolute)
Gestational age, weeks	0.991 (0.988 to 0.995)*	−0.86	0.39	0.538 (0.340 to 0.852)*	−46.19	28.30
Age of mother, years	0.982 (0.973 to 0.991)*	−1.82	0.84	0.719 (0.650 to 0.795)*	−28.11	15.07
Female sex at birth	1.000 (0.999 to 1.001)	−0.03	0.01	0.985 (0.966 to 1.003)	−1.54	0.71
First Nations	0.827 (0.805 to 0.849)*	−17.31	8.68	1.057 (1.045 to 1.069)*	5.71	2.53
LoS during ED presentation, minutes	1.005 (0.995 to 1.015)	0.51	0.23	0.885 (0.830 to 0.943)*	−11.52	5.59
Congenital malformation status at birth	0.999 (0.998 to 1.000)	−0.07	0.03	1.000 (0.998 to 1.002)	0.01	0.00
Birth hospitalisation outcome
Transferred	0.993 (0.990 to 0.996)*	−0.68	0.31	1.001 (0.999 to 1.003)	0.10	0.04
Refused care	0.998 (0.996 to 0.999)*	−0.25	0.11	0.999 (0.998 to 1.000)	−0.09	0.04
Year of birth
2003	1.001 (1.000 to 1.001)*	0.07	0.03	1.000 (0.996 to 1.004)	0.03	0.01
2012	1.000 (0.999 to 1.001)	−0.02	0.01	1.003 (0.997 to 1.009)	0.31	0.14
2013	1.004 (1.002 to 1.005)*	0.35	0.16	1.000 (0.992 to 1.008)	−0.05	0.02
2014	1.002 (1.001 to 1.004)*	0.23	0.10	0.997 (0.990 to 1.004)	−0.31	0.14
2015	1.008 (1.006 to 1.010)*	0.80	0.36	0.986 (0.979 to 0.994)*	−1.35	0.62
2016	1.004 (1.002 to 1.006)*	0.41	0.19	0.997 (0.992 to 1.002)	−0.30	0.14
Constant				2.160 (1.325 to 3.520)	115.99	35.16

^*Statistically significant at p value<0.05.

ED, emergency department; LoS, length of hospital stay; βC, β value for coefficient; βE, β value for endowments.

Maternal age difference was the largest contributor to the explained component of First Nations-related disparities in ED presentation costs, accounting for 1.2% (${exp(\beta }_E)$= 0.979, 95% CI 0.966 to 0.991). Other contributors to the explained component included gestational age at birth, birth hospitalisation outcomes, under investigation for congenital malformations and LoS during ED presentation. However, most disparities associated with First Nations status were attributable to unexplained components (table 6, online supplemental table 11).

Table 6. First Nations-related disparities in ED presentation cost of children in the first 5 years, NT, Australia (2000 to 2020).

Variable	Exp (${\beta }_E$) (95% CI)	Proportional cost change (%)	Percentage contribution to variation in cost (absolute)	Exp (${\beta }_C$) (95% CI)	Proportional cost change (%)	Percentage contribution to variation in cost (absolute)
Gestational age, weeks	0.987 (0.982 to 0.992)*	−1.3	0.75	0.498 (0.315 to 0.789)*	−50.1	39.63
Age of mother, years	0.979 (0.966 to 0.991)*	−2.1	1.22	0.671 (0.606 to 0.741)*	−32.9	22.76
Female sex at birth	0.999 (0.998 to 1.001)	0.0	0.03	0.985 (0.965 to 1.005)	−1.5	0.85
Remote areas	1.012 (0.991 to 1.034)	1.2	0.70	1.051 (1.043 to 1.059)*	5.1	2.84
LoS during ED presentation, minutes	0.997 (0.994 to 1.001)	−0.2	0.13	0.963 (0.915 to 1.015)	−3.7	2.13
Under investigations for congenital malformation at birth	0.998 (0.997 to 1.00)	−0.1	0.06	1.002 (0.999 to 1.004)	0.2	0.08
Birth hospitalisation outcome
Transferred	0.994 (0.992 to 0.997)*	−0.5	0.31	0.999 (0.998 to 1.002)	−0.1	0.01
Refused care	0.996 (0.994 to 0.998)*	−0.4	0.21	0.999 (0.999 to 1.000)	−0.1	0.04
Year of birth
2003	1.001 (0.999 to 1.001)	0.1	0.03	1.001 (0.996 to 1.007)	0.1	0.08
2012	1.001 (0.999 to 1.001)	0.0	0.01	1.002 (0.996 to 1.009)	0.2	0.13
2013	1.003 (1.001 to 1.004)*	0.3	0.16	1.001 (0.993 to 1.007)	0.0	0.01
2014	1.002 (1.001 to 1.004)*	0.2	0.11	0.999 (0.993 to 1.006)	−0.1	0.04
2015	1.005 (1.003 to 1.007)*	0.5	0.28	0.993 (0.985 to 1.000)	−0.7	0.43
2016	1.003 (1.001 to 1.004)*	0.3	0.15	0.998 (0.993 to 1.003)	−0.2	0.12
Constant				1.598 (0.986 to 2.591)	59.8	26.71

^*Statistically significant at p value<0.05.

CI, Confidence Interval; ED, emergency department; LoS, Length of hospital stay; βC, β value for coefficient; βE, β value for endowments.

Discussion

Our study highlights that most of the remoteness-related gap in hospitalisation costs (86%) was explained by differences in observed characteristics (eg, First Nations status, birthweight, gestational age, diagnosed for congenital malformations) and only 29% of the First Nations-related gap was explained this way. The remaining (14% for remoteness and 71% for First Nations) was attributable to differences in how these characteristics translated into healthcare costs, which may reflect variations in access to healthcare, treatment decisions or the way costs are structured across the system. Similar patterns were observed in ED presentation costs, where 42% of the remoteness-related and 3.4% of the First Nations-related disparities were explained by differences in observed characteristics, whereas 58% and 96.6% were due to differences in how these characteristics translate into costs. These findings suggest that children in remote areas and First Nations children may experience distinct healthcare trajectories and costs, influenced by unobserved differences in healthcare access, service delivery models, and clinical decision-making.²⁴

These contrasting patterns may suggest that the cost gap reflects not only clinical need but also how the health system responds to it. For children from remote areas, most of the gap came from differences in the children’s clinical characteristics, more adverse perinatal outcomes, longer stays and greater clinical need. In that sense, the higher costs are consistent with a system responding proportionately to need, allocating more resources where clinical need and geographic distance are greatest, and covering retrieval, transfer and the additional care that local services cannot provide. The pattern for First Nations children differed. Most of the gap was explained not by the children’s clinical characteristics but by how those characteristics translated into cost, a pattern that suggesting maybe due to differences in system response rather than in underlying need. These two patterns are not, however, separable. More than three-quarters of First Nations children lived remotely, and a comparable share of children from remote areas were First Nations, indicating that neither dimension can be treated as an independent axis of disadvantage. Our stratified analysis highlighted that the cost difference associated with remote residence was roughly five times higher among First Nations children than among non-First Nations children. The higher health-system costs for First Nations children are better read as a marker of need that was not prevented upstream, of long-standing gaps in accessible, culturally safe care, rather than as a cost the population imposes.^{12 41} The issue is not that one group is more expensive, but that the system may not yet have invested early enough to stop avoidable need from arising.⁴² Addressing these disparities may, therefore, require targeted perinatal health interventions such as birthing on country services,⁴³ midwifery continuity care and Aboriginal Community Controlled Health Organisations based holistic and maternity models of care,^{44 45} to ensure equitable healthcare access and reduce excess cost across different population groups and geographic areas.⁴⁶

How these differences played out in practice can be seen in a few recurring patterns. We found that hospitalisations due to respiratory and other infections were more common in all groups. However, the costs of those hospitalisations and ED presentations associated with adverse perinatal outcomes were substantially higher among children from remote areas or First Nations populations. Several factors may contribute to these higher costs.

First, a greater proportion of vulnerable infants required interfacility transfers within the NT following birth hospitalisation, compared with those from urban and non-First Nations populations. This additional need for transfer and access to specialist care likely reflects limited capacity at smaller regional facilities to manage complex cases, contributing to higher overall costs. Prior studies have similarly reported higher hospitalisation and ED presentation rates among First Nations and socioeconomically disadvantaged infants,^{13 47–50} and greater clinical severity may further explain the elevated costs observed.⁵¹

Second, prolonged hospital stays following both admission and ED presentations were more frequent among children from remote areas and First Nations children. Length of stay is a major cost driver and contributes to the disparities observed. Consistent with our findings, previous studies from other settings showed that infants with adverse perinatal outcomes experience disproportionately longer hospital stays, particularly among racially and socioeconomically disadvantaged populations.^{26 52–54} In the setting of patients from remote communities, prolonged admission is often a proactive strategy to conduct necessary investigations, initiate specialist or allied healthcare and ensure neonatal stability before discharge, particularly where access to follow-up services is limited.

Third, greater socioeconomic vulnerability in these populations may exacerbate barriers to health education and timely care, increasing the risk of adverse perinatal outcomes and severe health conditions that necessitate costly interventions. We found that rates of adverse perinatal outcomes were higher among First Nations groups and in remote areas compared with the national average, consistent with prior studies.^{51 55–58} Socioeconomic disadvantage has also been associated with increased risk of readmission and prolonged hospital stay, both of which elevate costs.^{59 60}

Finally, a range of changes to health service delivery in the study area over the last decades may have influenced hospitalisation and ED presentation patterns and associated costs. Some of the initiatives that could have influenced healthcare costs include expanded health service delivery models and child health check initiatives,⁶¹ introduced as part of the Closing the Gap agenda¹² to narrow long-standing primary healthcare funding gaps in remote and First Nations areas. Infrastructure such as the Patient Assistance Travel Scheme,⁶² which subsidises travel and accommodation for those requiring specialist care unavailable locally, also forms part of the broader system of remote healthcare access that our cost estimates may partly reflect.

Our study has several strengths. It draws on two decades of linked administrative data which traced cost disparities over a long period and revealed patterns that shorter or single-timepoint studies would miss. Because the data cover all singleton births across the Northern Territory, our findings are less prone to selection bias. We looked at two dimensions of vulnerability, First Nations status and remoteness, showing how disparities vary across population groups rather than along a single axis. A further strength was our decomposition approach, which went beyond measuring the size of the disparities to identify what drives them, separating differences due to measurable characteristics from those reflecting structural or unmeasured factors. These strengths must be interpreted in light of some limitations. Cost estimations were limited to hospitalisations and ED presentations, as data on general practitioner and specialist outpatient visits, as well as Medicare Benefits Schedule and Pharmaceutical Benefits Scheme items, were unavailable. Thus, our cost valuations might differ from the actual costs incurred per child at a population level, as children with no hospital or ED presentation contribute no recorded cost, although many will have incurred costs in these other settings. Our estimates accordingly describe disparities in hospital-based costs rather than in total healthcare costs, and the direction of any resulting bias depends on how substitution between care settings varies across groups. Related to this, healthy children without inpatient admissions or ED presentations were excluded, potentially underestimating average birth hospitalisation costs and overestimating cost differences between children with and without adverse perinatal outcomes. Beyond how cost was defined, observed costs reflect service use rather than need. Because the likelihood of having any presentation may differ by First Nations status and remoteness, children in vulnerable groups who faced barriers to accessing care^{24 63} may not appear in the records at all. Where such barriers occur, our estimates could understate the true disparities in need, and the differences reported here should be read as a conservative lower bound. Survival to 5 years was assumed for preterm children who survived the birth hospitalisation, which may underestimate long-term costs by excluding children who died earlier.^{64 65} Reliance on administrative data presents inherent challenges. Diagnostic coding may be non-specific or imprecise and may not accurately reflect the severity of conditions, which is an important determinant of healthcare utilisation and associated costs.⁶⁶ Additionally, patient transport costs, including transfers from remote communities or interstate for specialist procedures (eg, cardiac surgery), were not included. These costs, often absorbed within the NT Health hospital budget, accounts for a substantial proportion of healthcare expenditure, particularly for patients living in remote areas. Our socioeconomic measure, the IRSAD, has recognised limitations. It does not differentiate between First Nations and non-First Nations populations as comprehensively as newer indices such as the Pooled Indigenous and Non-Indigenous Relative Socioeconomic Outcomes measure. The generalisability of our findings may also be limited, as the burden^{3 67} and drivers of adverse perinatal outcomes in the NT differ from those in other Australian jurisdictions due to variation in population demographics, healthcare access and social determinants of health.⁶⁸ Analyses based on First Nations status relied on maternal identification due to limited data on child identification, which may have influenced observed disparities. Finally, our data captured care provided within the NT only; children who moved interstate during follow-up would not be observed in NT hospital and ED records, so their costs may be incomplete. Given the mobility of the Territory’s population,⁶⁹ this may have affected cost estimation, most likely by underestimating total costs for children who out-migrated.

Conclusions

Our study highlights marked disparities in early childhood hospitalisation and emergency care costs largely attributed to adverse perinatal outcomes, varying by remoteness and First Nations status. Cost differences by remoteness were largely explained by children’s characteristics, whereas disparities for First Nations families reflected systemic barriers and structural inequities. These findings emphasise the importance of policies that target both individual and structural determinants to ensure fair access and efficient use of resources. Tailored strategies to improve perinatal care, especially for remote and First Nations communities, are essential to reduce preventable hospitalisations, associated costs and long-term health inequalities.

Data availability statement

All data relevant to the study are included in the article or uploaded as supplementary information.