To quantify the effect of socioeconomic status (SES) on health outcomes during the first year after newborn discharge among infants with complex chronic conditions (CCCs) insured through a universal health plan.
Longitudinal, population-based cohort study.
Infants born in hospitals from April 1, 1996, through March 31, 2000. Infants with CCCs were identified from their newborn discharge records. Neighborhood income quintiles were obtained by linking participants' postal codes to census data.
Main Outcome Measures
Mortality and hospital admissions in the first year after newborn discharge. Logistic and Poisson regression analyses were used to examine the relationship between neighborhood income quintiles and outcomes, adjusting for important covariates such as low birth weight and rural residence.
A total of 512 768 infants were included, of whom 2.3% had CCCs at newborn discharge. Infants with CCCs accounted for 37.8% of deaths and 11.0% of hospitalizations during the first year after the newborn discharge. Infants with CCCs living in the lowest-income neighborhoods had a 1.26-fold higher mortality risk (95% confidence interval, 0.83-1.90; P = .28) and a 1.24-fold higher hospitalization rate (1.09-1.40; P < .001) compared with those living in the highest-income neighborhoods. Although the income gradients associated with mortality and hospitalization were less pronounced among infants with CCCs compared with infants without CCCs, the absolute interquintile risk differences attributable to SES were higher among infants with CCCs.
Despite universal health insurance, SES-related inequality affects hospitalization and, possibly, mortality rates among medically vulnerable infants.
Socioeconomic status (SES) has been widely recognized as an important social determinant of health and a powerful predictor of health status,1 mortality,2- 6 and health services use in the general infant population.1,7,8 Previous Canadian studies1- 8 showed that infants from less affluent neighborhoods had a significantly higher risk of mortality and hospitalization compared with those from more affluent neighborhoods, even after accounting for infant and maternal characteristics, such as infant sex, parity, plurality, maternal age, maternal ethnicity, marital status, mode of delivery, and maternal illness. Although markedly diminished during the past 2 decades in Canada, socioeconomic disparities in infant mortality remain obvious over time and across geographical areas, despite a universal health insurance system.2,5,6
Although the adverse effects of low SES on mortality and morbidity in the general infant population are well documented, few studies have included medically fragile infants, who have greater health care needs and a much higher risk of poor outcomes. Much of this risk may be mediated by health services, whereas the effect of health care in turn may be modified by socioeconomic factors.9 Studies of Canadian adults suggest that access to primary health care services in general is not related to SES, although a gradient in specialty care use may exist.10 A more recent Canadian study found that low SES was associated with hospitalizations for ambulatory care–sensitive conditions among urban children11; however, it was unclear whether this outcome was driven purely by differential access to health care.
This study aimed to give a closer look at the association between low SES and health outcomes among medically fragile infants whose complex chronic conditions (CCCs) are identified at birth. This creates a specific opportunity to examine the effect of SES in the context of an identified condition and a universal health insurance scheme in which access to care should not be impaired by the identification of the primary condition or the ability to pay for services. There are challenges in classifying these infants, who are nonhomogeneous and have a variety of severe and rare chronic conditions. One approach to identifying these infants is to use the definition of CCCs. The definition proposed by Feudtner et al12(p206) refers to CCCs as “any medical condition that can be reasonably expected to last at least 12 months (unless death intervenes) and to involve either several different organ systems or 1 organ system severely enough to require specialty pediatric care and probably some period of hospitalization in a tertiary care center.” By aggregating a broad range of rare but severe childhood chronic conditions, this definition approximates a “noncategorical” approach to classifying CCCs that is based on the shared common attributes of these conditions and the care they require, including special expertise, management of voluminous information, skillful care planning and coordination, and often the need for extensive child and family support services.12- 15
The goals of this study were to describe mortality and hospitalization rates among infants with CCCs using health administrative data and to test whether there was a similar SES gradient compared with infants without CCCs. We hypothesized that the association between SES and health outcomes among CCC infants would be weaker compared with non-CCC infants because CCC infants are expected to frequently access specialized health care regardless of SES and thus may modify their risks of poor health outcomes.
This study used 2 Ontario health care administrative databases housed at the Institute for Clinical Evaluative Sciences: the hospital discharge abstract database, which includes hospitalization records from all hospitals, the data quality of which is monitored by the Canadian Institute for Health Information; and the Registered Persons Database, which contains demographic and vital statistic data for all Ontario residents eligible for the Ontario Health Insurance Plan. Data from the Canadian Institute for Health Information database included each resident's unique identifier, postal code, admission and discharge dates, and as many as 16 diagnosis codes. Variables from the Registered Persons Database included each resident's unique identifier, sex, birth date, and, if applicable, date of death. The unique identifier included in each database is a reproducible scrambled identifier, which permits the linkage of an individual's records across databases and time but also preserves patient confidentiality. We linked postal codes recorded at birth to the Canadian Census 2001 using Postal Code Conversion Files Plus16 to obtain neighborhood income quintiles and community size (rural vs urban residence). Ethics approval for this study was received from the research ethics boards of The Hospital for Sick Children, University of Toronto, and Sunnybrook Health Sciences Center, Toronto.
This study used a population-based retrospective cohort study design and included all infants born in Ontario hospitals and discharged alive from April 1, 1996, through March 31, 2000, who were residents of Ontario and, therefore, eligible for the Ontario Health Insurance Plan. All infants who met these criteria were observed for 12 months after discharge from newborn hospitalization. We excluded infants who developed CCCs within 1 year after newborn discharge.
We defined CCCs using the list of International Classification of Diseases, Ninth revision (ICD-9), codes developed by Feudtner et al,12 which mapped specific ICD-9 codes to 9 diagnostic categories: neuromuscular, cardiovascular, respiratory, renal, gastrointestinal, hematologic or immunologic, metabolic, malignancy, and genetic or other congenital defect conditions. Eligible infants were assigned to the CCC or non-CCC group based on diagnoses documented at newborn discharge. Data on birth weight were obtained from birth hospitalization records.
We used the neighborhood income quintile as our measure of SES. “Neighborhood” was defined at the level of dissemination area, the smallest geographical unit in the 2001 Canadian Census with 400 to 700 residents.17 Neighborhood income quintiles were constructed based on the household-size–adjusted mean neighborhood income per single person equivalent,18 calculated using the person-weights implicit in the Statistics Canada low-income cutoff points.19 Within the same census metropolitan area or census agglomerations, dissemination areas were divided into 5 income quintiles based on the mean neighborhood income. Each neighborhood income quintile contained approximately one-fifth of the total population of each census metropolitan area or agglomeration, but not necessarily one-fifth of all births. We assigned those with missing neighborhood income quintile values because of unstable populations, absent census income data, or missing patient postal codes to a separate category (unassigned neighborhood income).
To account for potential differentials in access to health care at the community level,4- 8 we measured the percentage of residents living in rural areas using the rural and small town definition recommended by Statistics Canada.20 All areas with a community size of fewer than 10 000 residents were considered rural, whereas the rest were considered urban.21
We followed up all infants for 1 year after discharge from newborn hospitalization and determined all-cause deaths from the hospitalization and demographic data sets and all-cause hospital admissions from the hospitalization data set.
We used a χ2 test in 2-group comparison for nominal data and the t test for continuous variables. Rates (per 1000) of all-cause mortality and all-cause hospitalization by neighborhood income quintile were calculated. We analyzed trends in mortality rates across neighborhood income quintiles (ordinal data) with the Cochran-Armitage trend test and the trend in hospitalization rates with Poisson regression analysis. Logistic regressions were used to examine the effect of independent risk factors on all-cause mortality and Poisson regressions for all-cause hospitalization. To determine whether the effect of low SES on mortality and hospitalization rates differed significantly between infants with CCCs and those without, both analyses were stratified by CCC status. Model significance was determined by the Hosmer-Lemeshow test for logistic regression analysis and the log likelihood ratio test for Poisson regression analysis. Multilevel analyses were also conducted to assess whether there was clustering of outcomes by neighborhood. We used SAS statistical software, version 8.0 (SAS Inc, Cary, North Carolina) for all analyses.
We identified 512 768 infants at discharge from newborn hospitalization. Of these infants, 2.3% had CCCs identified during newborn hospitalization, 51.2% were boys, and 11.6% lived in rural areas. Infants with CCCs had a significantly lower mean birth weight (2775 g vs 3420 g; P < .001) compared with non-CCC infants. Cardiovascular and respiratory conditions and other congenital or genetic defects accounted for the vast majority of CCC diagnoses (Table 1).
The distribution for CCC diagnostic categories was similar between neighborhood income quintiles (data not shown). Infants with CCCs from the lowest neighborhood income quintile were more likely to have a birth weight less than 1500 g (21.1% vs 17.9%; P = .01) and less likely to live in rural areas (7.8% vs 13.2%; P = .03) compared with those from the highest neighborhood income quintile.
Despite accounting for only 2.3% of the population, CCC infants accounted for 241 of the 637 deaths (37.8%) in the first year after newborn discharge (Table 2). They also had a significantly higher overall mortality rate compared with non-CCC infants (20.1/1000 vs 0.8/1000; P < .001). This difference remained highly significant when further stratified by other risk factors such as sex, birth weight, SES, and rural residence. The test for trend showed that mortality rates decreased significantly with successive increases in neighborhood income among both infants with CCCs (P = .03) and without CCCs (P < .001). The difference in absolute mortality rate between the lowest (quintile 1) and the highest (quintile 5) SES quintiles was 5.5 per 1000 among CCC infants vs 0.6 per 1000 among non-CCC infants, suggesting that the SES-attributable mortality rate was higher among CCC infants.
Infants with CCCs accounted for 11.0% (n = 7983) of all hospitalizations (n = 72 538) during the first year after discharge from newborn hospitalization and had a significantly higher overall hospitalization rate compared with other infants (665.9 per 1000 vs 128.9 per 1000; P < .001) (Table 2). This difference remained highly significant when further stratified by other risk factors. The test for trend showed that hospitalization rates decreased significantly with successive increases in neighborhood income for both groups (P < .001). The difference in absolute hospitalization rate between the lowest and the highest SES quintiles was 150.1 per 1000 for CCC infants and 38.6 per 1000 for non-CCC infants, suggesting that the SES-attributable hospitalization rate was higher among CCC infants.
There was tremendous heterogeneity in causes of hospital admission among CCC infants (Table 3). No single cause was responsible for a substantial proportion of hospitalizations. Although CCC infants were often admitted for diagnoses specifically related to their primary conditions, such as congenital defects of the heart and face, they were also frequently admitted for complicating diagnoses, such as bronchiolitis, respiratory tract infection, urinary tract infection, feeding problems, and failure to thrive. The rates of these admissions were higher than the corresponding rates of admissions among non-CCC infants.
Having a CCC was a significant predictor for mortality (adjusted odds ratio, 15.41; 95% confidence interval, 12.68-18.72; P < .001) within 1 year after discharge from newborn hospitalization when adjusted for SES, birth weight, sex, and rural residence (Table 4). Among CCC infants, living in the lowest SES neighborhoods was associated with a 1.26-fold higher mortality risk (95% confidence interval, 0.83-1.90; P = .28) compared with infants living in the highest SES neighborhoods after adjusting for other risk factors. Adjusting for the various CCC categories in the model did not change the risk estimates (data not shown). The mortality risk ratio of the lowest neighborhood income quintile vs the highest was smaller among CCC infants compared with non-CCC infants, although the trend was not statistically significant.
Having a CCC was associated with significantly higher hospital use (adjusted rate ratio, 4.18; 95% confidence interval, 4.07-4.31; P < .001) within 1 year after discharge from newborn hospitalization (Table 5). Among CCC infants, living in the lowest SES neighborhoods was associated with a 1.24-fold higher hospitalization rate (95% confidence interval, 1.09-1.40; P < .001) compared with living in the highest SES neighborhoods, after adjusting for other risk factors. Adjusting for CCC categories in the model did not change the estimated rate ratios (data not shown). The hospitalization rate ratio for the lowest neighborhood income quintile vs the highest was smaller among CCC infants compared with non-CCC infants, although the trend was not statistically significant.
Multilevel regression analyses found 0% and 1.7% neighborhood variance for mortality and hospitalization rates, respectively, among CCC infants and did not change the risk estimates or the confidence intervals for the effect of SES (data not shown).
This study is the first to use population-based data to identify infants with CCCs at the time of newborn discharge and to examine the association between neighborhood SES and subsequent infant health outcomes in a universal health insurance system. We showed that about 2.3% of Ontario infants had at least 1 CCC at newborn discharge and that these infants accounted for a major proportion of adverse outcomes in this age group. The infants with CCCs and lower SES were significantly more likely to be hospitalized than those of higher SES. We found a very similar SES gradient for mortality risk, although this did not reach statistical significance. Collectively, these results demonstrate that low SES is related to poor outcomes even when infants have been identified at birth as having a CCC and when they have the same universal insurance coverage for medical services, regardless of SES.
Preferential access to health care for higher-income individuals is considered by some to be less of an issue in a universal insurance health care system. In Canada, general practitioner care is distributed fairly equally across income groups according to needs, although people with higher income are more likely to seek specialist care than those with lower income.22 We hypothesized that the SES gradient would be attenuated among CCC infants, who are expected to be connected to the health care system on newborn discharge. The barriers to receiving ongoing health care after discharge may exist with these diagnoses, but they should be minimized in the Canadian universal health insurance scheme. Our study showed that SES-related disparities in outcomes may be attenuated in the CCC group in terms of relative risk/rate ratio differences across quintiles. However, this does not mean that health outcomes attributable to SES are less important for CCC infants. In fact, the SES-attributable fraction of the absolute hospitalization rate was substantially larger among CCC infants (150.1 per 1000 vs 38.6 per 1000 in the non-CCC group) because of the much higher rates of hospital admission in this group. A similar trend was also observed in mortality rates (5.5 per 1000 in the CCC group vs 0.6 per 1000 in the non-CCC group). This is to say that, for an individual child with low SES and a CCC, SES can pose an important added dimension of risk. Although these results may not be directly generalizable to jurisdictions without universal health insurance systems, they should be quite relevant for those jurisdictions striving to provide universal coverage for children.23
The mechanisms through which SES is linked to health outcomes are still being debated.24- 26 Besides access to care, other factors, such as the safety levels of home care and environmental risks, could all have contributed to the SES gradients in health outcomes for CCC infants, which remained significant even after adjustment for birth weight and primary medical conditions. Although it is beyond the scope of this study to delineate the complex causal factors underlying these relationships, we hypothesize that the seemingly attenuated relative risk effect of SES on mortality and hospitalization rates in CCC infants may be owing to the different causes of death and hospitalization in this group compared with non-CCC infants. Although cause of death is not available in this study, it is expected that non-CCC infants would be more likely to die of sudden infant death syndrome (SIDS) and injuries12,27 or rapidly emerging unexpected acute conditions, which are highly sensitive to social risk factors. In contrast, CCC infants would be more likely to die from causes arising from their underlying medical conditions, such as complications of congenital anomalies or metabolic disorders, which may be more easily anticipated. This may partly explain the attenuated SES gradient in relative mortality risk ratio for CCC infants. Similarly, although the most common readmission diagnoses overlapped between the 2 groups, CCC infants were more likely to be admitted for complications caused by congenital defects. The need for hospitalization for such severe chronic medical conditions is less likely to be influenced by socioeconomic and environmental factors compared with ambulatory care–sensitive conditions (eg, urinary tract infections, gastroenteritis, and dehydration), which are more commonly seen among non-CCC infants.28
Several limitations of this study should be noted. First, although useful in identifying infants with many health care needs, the definition of CCCs is arbitrary and includes a broad range of conditions that are relatively less lethal to highly lethal. Because the association between SES and health outcomes may vary depending on the specific disease, the findings of this study may not be completely applicable to subgroups of infants with specific CCCs. Nevertheless, because of the low prevalence of each condition, such condition-specific differences are expected to have a small effect overall and are not likely to affect health care planning at the population level. Other methods have also been proposed to aggregate childhood chronic conditions for the purposes of research, service delivery, and health policy development. One example is the US initiative to identify children with special health care needs using functional questionnaire data.29,30 Because of the lack of clinical data and functional status assessment in health administrative databases, such an approach was not feasible for the present study, but it remains a future avenue for analysis.
Second, we defined neighborhood at the level of dissemination area, which is the smallest geographical unit in the Canadian Census 2001 with a relatively small and homogeneous population. Studies have shown that neighborhood income calculated from small geographical units is moderately correlated with individual family income31 and that risk estimates derived from ecologic income measures approximate estimates obtained from individual household income.32,33 Several US studies also demonstrated that neighborhood socioeconomic characteristics (eg, income inequality and access to health and social services), or “contextual factors,” are important independent determinants of health outcomes,34- 36 although similar associations were not found in Canada.26,36 In the present study, SES was measured on the neighborhood level only. It is, therefore, difficult to tell whether the effects of SES observed in this study were contextual, merely reflect the aggregate effect of individual-level SES, or represent a combination of both.37,38 Most researchers agree that neighborhood income quintile is a surrogate measure of neighborhood socioeconomic characteristics and reflects neighborhood-related difference in individual socioeconomic positions.24- 26 Therefore, the risk/rate ratio of adverse health outcomes associated with neighborhood income may serve as a good indicator of socioeconomic disparities in health at the community level. It should also be noted that geocoding determination of neighborhoods is less precise in rural areas, thus resulting in random misclassification of neighborhood income and a bias toward the null hypothesis. Fortunately, our population-based administrative databases have large enough sample sizes to minimize the effect of random misclassification error.
Third, although the severity of CCCs could be unequally distributed among income quintiles (eg, greater in lower-income quintiles) leading to disproportionately higher morbidity and mortality according to SES, it was not included as a confounder in the analyses because of the lack of clinical data in health administrative databases. Nevertheless, had such data been available, measurement of disease severity would have still posed a significant challenge owing to the broad range of conditions included in the CCC list. We tried to use CCC diagnosis categories as proxy measures of overall severity and spectrum of illness. However, adjusting for these factors did not alter the results of regression analyses.
Finally, the number of CCC infants included in this study may not be adequate for testing the effect of low SES on mortality using logistic regression analysis because the SES-associated mortality risk ratio in this group was smaller than that of the general population. Findings of this study will guide sample size calculations for further studies, and continuous inclusion of incident cases of newborns with CCCs will be desirable to achieve a higher power in future analysis.
In summary, the present study showed that SES gradients in risks of mortality and hospitalization existed in medically vulnerable infants who participated in a universal health insurance system. Infants with CCCs identified at birth who have higher risks of poor outcomes, some of which may be mediated by health services use, should have reduced SES gradients within a universal health care system. This is consistent with our finding that the SES relative risk effects on mortality and hospitalization rates were attenuated among CCC infants identified at birth. However, the SES gradients still exist for these infants, and the absolute risk of poor outcomes attributable to SES is still increased. This represents the ironic paradigm of attenuated relative risk but increased attributable absolute risk as a result of the substantially high baseline risk of poor outcomes for these infants. Finally, SES may be a useful factor to stratify these infants for further investigations of practical strategies to reduce the SES gradients in health outcomes in this population.
Correspondence: Paul T. Dick, MDCM, MSc, FRCPC, Grey Bruce Health Services, Owen Sound, Ontario, N4K 6M9 Canada (PDick@gbhs.on.ca).
Accepted for Publication: October 11, 2008.
Author Contributions: Drs Wang and Dick had full access to all the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis. Study concept and design: Wang, Guttmann, To, and Dick. Acquisition of data: Wang and Dick. Analysis and interpretation of data: Wang, Guttmann, To, and Dick. Drafting of the manuscript: Wang and Dick. Critical revision of the manuscript for important intellectual content: Wang, Guttmann, To, and Dick. Statistical analysis: Wang, To, and Dick. Obtained funding: Dick. Administrative, technical, and material support: Wang and Guttmann. Study supervision: Guttmann, To, and Dick.
Financial Disclosure: None reported.
Funding/Support: This study was supported by a Salary Award from the Canadian Institutes of Health Research (Dr Guttmann), a 2007 Dales Award in Medical Research from the University of Toronto Life Sciences Committee (Dr To), and the Institute for Clinical Evaluative Sciences, an independent nonprofit organization funded by the Ontario Ministry of Health and Long-Term Care, with provision of population-based data.
Role of the Sponsors: The opinions, results, and conclusions are those of the authors, and no endorsement by the Ontario Ministry of Health and Long-Term Care or by the Institute for Clinical Evaluative Sciences is intended or should be inferred.
Additional Contributions: Kelvin Lam, MSc, and Jun Guan, MSc, statisticians at the Institute for Clinical Evaluative Sciences, provided programming and statistical support for this project.
Wang C, Guttmann A, To T, T. Dick P. Neighborhood Income and Health Outcomes in InfantsHow Do Those With Complex Chronic Conditions Fare?. Arch Pediatr Adolesc Med. 2009;163(7):608-615. doi:10.1001/archpediatrics.2009.36