Introduction

Poverty is expected to be associated with higher rates of child hospital admissions in developed countries [115]. One reason for this is that children from low-income households or households with low socio-economic status tend to experience more frequently clinical problems, and to suffer more severe conditions, than children living in less deprived households [1, 12, 13, 1623]. A second and complementary reason is that developed countries are able to mitigate common access barriers (e.g., availability, affordability) that usually prevent the poor from receiving care.

However, in a previous cross-sectional study, we reported a kind of “inverted U-shaped” relationship between poverty and hospitalization rates among a representative sample of five-and-a-half-month old children in Quebec. Estimated admission rates were 13.3% (95%CI = 9.1–17.5), 21.2% (95%CI = 15.6–26.8), and 11.5% (95%CI = 9.9–13.1) among children from very-low-income, low-income and sufficient-income families, respectively [13, 16]. This observation is incongruent with the income to health problem ratio experienced by children in our study. Compared with sufficient-income groups, mothers in the low-income (Adjusted OR = 1.5; 95%CI: 1.1–2.1) and very-low-income groups (Adjusted OR = 1.8; 95%CI: 1.3–2.6) were more likely to report less positive perception of their baby’s health status, controlling for mothers’ education, immigration, and marital status [12]. Given Canadian universal health care coverage, it is important to understand why the poorest children are the least hospitalized despite their greater needs.

The 3.5-year follow-up of the children in our study provided an opportunity to assess how the poverty–hospitalization relationship evolves over time. We used Andersen and Newman’s model of health service utilization as a conceptual framework [24, 25] to identify appropriate control and explanatory variables. According to this model, health service utilization, indicated here by hospitalization, is influenced by predisposing, enabling, and need-related factors. Predisposing factors are those that exist prior to disease onset. Enabling factors are conditions that facilitate or inhibit individuals’ ability to access the health care system. Need-related variables are linked to illness levels or perceived need for care.

Method

Population

Data originated with the Quebec longitudinal study of child development (QLSCD), a population-based cohort study in which an initial sample of 2,223 infants born in Quebec is observed. All singleton live births recorded in the master registry of Quebec in 1997/1998 were included, except those with the following exclusion characteristics: born to mothers living in the Northern Quebec administrative region or in Cree or Inuit territory; unknown gestational age; born before 24 weeks’ or after 42 weeks’ gestation. Baseline data were collected when the infants were around 5 months old (min–max, 15–36 weeks). From the initial sample, 2,120 children were subsequently resurveyed annually. The current analyses used data from the four first waves, when children were, respectively, about 5, 17, 29, and 41 months old. The database contains information on neonatal health conditions which were retrieved from medical records. It also contains repeated measures of the children’s health, growth, and health care, as well as their parents’ health and socio-economic conditions, and various characteristics of the child rearing environment. Mothers’ reports were the principal data source. Data were collected through face-to-face interviews conducted in respondents’ homes by trained interviewers and through self-administrated questionnaires.

Variables

Hospitalization was defined in our study as a binary variable indicating whether or not the mother reported that the child had ever been admitted to hospital for one night or more, between birth and 41 months of age.

Poverty was the main independent variable pertaining to the enabling group of factors. It was measured via two indicators intended to capture chronic and severe exposure to inadequate income: duration of poverty and severity of poverty. A child’s poverty condition was determined based on the income level of the household in which he/she was being raised. Households were classified against a need specific low-income threshold computed annually by Statistics Canada [26]. A household was considered to have sufficient income if its total annual income was above the low-income threshold. Income was considered insufficient or very insufficient when it was between 60 and 99% or below 60% of the threshold, respectively. In operationalizing duration of poverty, we defined chronic exposure to poverty as living in a household for which the annual income fell below the low-income threshold during three or four follow-up periods (always poor). Children from this group were compared against those who were living in households that reported low-income during one or two periods (sometimes poor) and with children from households that reported no episode of low-income (never poor).

Severity of poverty was measured by a four category variable resulting from the combination of frequency and level of income sufficiency as follows: Income always sufficient (never poor), no episode of very insufficient income but one to four periods of insufficient income (income never very insufficient), one to two episode(s) of very insufficient income and one to two episode(s) of insufficient or sufficient income (income sometimes very insufficient), and three to four episodes of very insufficient income (income always very insufficient). The following variables were retained as factors predisposing to hospitalization: sex of the child (female, male); age of the child (≤40 months, ≥41 months at follow-up); age of the mother at delivery (<20, 20–34 and 35 years and over); maternal education (no high school diploma, high school or vocational or trade school diploma, partial or completed college or university studies) when the child was 5 months old; maternal immigration status (non-European immigrant vs. born in Canada or Europe); and maternal depression during the 12 months preceding the baseline data collection [46, 12, 13, 2729]. Enabling factors included the following: whether or not the mother had a partner at the beginning of the study [3032]; the total number, from birth to 3.5 years (i.e., 41 months), of medical visits (including general practitioner and pediatricians) [33] and number of visits by a public health nurse. The discrete variables, number of medical visits and number of visits by a public health nurse were categorized for the univariate analysis and subsequently log-transformed for multivariable analysis, as their distribution was highly asymmetric. Health needs considered were: the cumulative number of asthma attacks from birth to 3.5 years; the total number of episodes during which at least one acute health problem (respiratory tract infections, otitis media, gastroenteritis or other infections) was reported; and the index of neonatal cumulative risk. This latter variable is a weighted sum of the number of health problems at birth including preterm birth, small for gestational age, congenital abnormalities, and neonatal complications [16].

Analysis

We used life tables to explore the associations between each independent variable and the incidence of hospital admission. Significant differences in survival were determined using the Wilcoxon statistic. Variables associated with hospitalization at P ≤ 0.25 in univariate analyses were retained for the purpose of fitting them into two proportional hazards (Cox) regression models, for duration of poverty and severity of poverty, respectively. Variables were entered into the equations following the order specified in the Andersen model. The poverty indicator was entered first (Model 1). In Model 2, the predisposing variables were added, followed by the enabling factors (Model 3), and the health needs variables (Model 4). Some of these variables may be confounders while others may be in the causal pathway between poverty and hospitalization. Interactions between poverty and immigration status, presence of a partner, the number of medical visits and the number of visits by a public health nurse were initially tested and later removed from the model as they were not statistically significant. We followed the same model-fitting steps in performing a logistic regression to complement findings from the survival analyses. Logistic regression provides inference on the likelihood of experiencing any hospitalization, while survival analysis provides inference on the speed (earliness) of occurrence of the first hospitalization in the children’s life (birth to 31/2 years). The aim was to verify whether or not modeling the two facets of the poverty–hospitalization relationship indicates similar patterns. All analyses were performed using SPSS software, version 11.5.1.

Results

Complete data on hospital admission were available for 86.5% of the initial sample. Data were incomplete for 13.0% of the sample, and censored at the appropriate step in survival analysis. Eight cases were excluded for missing data on hospitalization at each of the four survey periods. Table 1 illustrates the distribution of children according to their household income level at each follow-up period and the number of times they had been hospitalized per year. At 5 months, 12.3% of children had been hospitalized at least once since birth but information on the number of re-admissions was lacking. The cumulative probability of having been hospitalized at least once by age 3.5 years was 31.0% (not shown). Most of the hospitalized children were admitted only once during the study period. As reported by the mother, admissions were linked to gastro-intestinal problems (5.0%; n = 1,933), respiratory diseases (10.5%; n = 1,941), or other health problems (15.0%; n = 1,940).

Table 1 Distribution of 3.5 years old children in the Quebec longitudinal study of child development (QLSCD), according to household income and hospital admissions by follow-up period

In terms of duration of poverty, children who were exposed to three or four episodes of low-income (chronic poverty) and those with no report of low-income had comparable cumulative probabilities of hospitalization (Table 2). Children from households experiencing intermittent poverty (sometimes poor) were significantly more likely to be hospitalized than those who never experienced poverty. No significant difference was observed between the chronically poor group and the intermittently poor group. Similar results were observed in the analysis of the associations between hospitalizations and severity of poverty. Children of families who experienced poverty but never fell 60% below the low-income threshold (income never very insufficient) and children of families who intermittently went through very insufficient income (income sometimes very insufficient) were the most likely to have been hospitalized by the age of three and a half. The hospitalization rate of those who always reported very insufficient was similar to that of those with sufficient income throughout (Table 2). Table 3 illustrates the single variable associations between hospitalizations and the predisposing, enabling, and needs-related variables, respectively. All of these associations (except mother’s depression) were significant at the 0.05 level.

Table 2 Probabilities of being hospitalized at least once from birth to 3.5 years, in the Quebec longitudinal study of child development (QLSCD): univariate estimates by duration of poverty and severe poverty
Table 3 Estimated probabilities of being hospitalized at least once from birth to 3.5 years according to predisposing, enabling and needs factors, in the Quebec longitudinal study of child development (QLSCD)

Tables 4 and 5 present the results of multivariable analysis of hospital admission as a function of duration and severity of poverty while taking into account predisposing, enabling and needs factors. The fitted hazard ratios of hospital admissions and 95% confidence intervals were obtained from Cox regression. Model 1 in Table 4 shows that, by the age of 3.5 years, the hazard (or risk) of hospitalization among intermittently poor children is 1.42 (95%CI = 1.15–1.76) times that of children who always reported sufficient income, when only the poverty indicator is considered. The estimated risks of hospitalization were comparable among both children experiencing chronic poverty and those with no reported history of poverty. The hazard ratios between the income-groups do not change after adjustment for the effects of socio-demographic indicators including the child’s age and gender as well as the mother’s age, education and immigration status (Model 2, Table 4). By contrast, controlling for the enabling variables reduces the magnitude of poverty-related coefficients (Model 3, Table 4). Furthermore, inclusion of need variables does not change the associations between poverty and hospitalization (Model 4, Table 4). Overall, the risk of hospital admission before the fourth anniversary increases by 1.30 (95%CI = 1.04–1.64) among children exposed to intermittent episodes of insufficient income compared to non-poor children, after controlling for the effects of predisposing, enabling, and needs-related indicators included in the final model. The results for severe poverty parallel those for duration of poverty (Table 5). Children for whom the household income has been always above the poverty threshold and those for whom the household income has been always very insufficient have similar predicted hazard rates. Besides, those who have had any period of insufficient income or intermittent episodes of very insufficient income faced higher risk of hospital admission than the non-poor children. Here again, the adjusting for the effect of enabling variables attenuates the poverty-related coefficients. The analyses based on the multivariable logistic regressions produced equivalent results (results available upon request).

Table 4 Adjusted hazard rates and 95% confidence intervals (CI) for hospital admission duration of poverty among 3.5 years old children in the Quebec longitudinal study of child development (QLSCD)
Table 5 Adjusted hazard rates and 95% confidence intervals (CI) for hospital admission according to severity of poverty among 3.5 years old children in the Quebec longitudinal study of child development (QLSCD)

Discussion

This investigation is prompted by previous unexpected results [13, 16] from a cross-sectional study on the association between poverty and hospital admission among 5-month-old children in a representative sample of Quebec infants. The purpose was to assess how the poverty–hospitalization relationship evolved over 3.5 years among the same population of children studied previously while controlling for predisposing, enabling, and needs-related factors that might explain this relationship. Results show that children living in chronic poverty are not significantly more likely to be hospitalized by age 3 years than children living in households with sufficient income whereas children living in households experiencing intermittent poverty are significantly more likely to be hospitalized at least once during the same ages. These associations remain despite controlling for maternal socio-demographic characteristics, suggesting a unique role of poverty. The association is attenuated once enabling factors are included in the model. Two of the enabling variables stand out as being particularly strongly related to the dependent variable: presence of a male partner and log of the number of consultations to a physician. The first is likely a confounder since most single-headed families are exposed to income insufficiency. The second is probably in the causal pathway. That is, poor households appear to make less use of the primary care system and low utilization is likely to account for the higher rates of hospitalization.

Few studies have specifically examined the effect of different operationalizations of poverty on child hospitalization in the first years of life in developed countries. Existing studies indicate a positive association, wherein the poorest were most likely to be hospitalized [4, 6, 7, 15], or re-admitted [11, 33, 34]. Two European studies detected no substantial impact related to socio-economic conditions on the risk of hospitalization for infectious diseases [5, 32]. Rather, one of them reported that children from lower-status backgrounds experienced the longest hospital stays [5]. Ecological studies also linked high child hospital admission rates to deprived socio-economic areas [14, 35]. A stepwise gradient of positive association is reported for asthma hospitalization [28, 36]. Our study suggests that in a system providing universal health care coverage, the association may not be linear but rather curvilinear, and that the relationship persists over time. This suggestion must be tested further, using adequate samples and analyzing other dimensions of hospitalization such as number of admissions or length of stay.

Some Canadian [6] and American [27, 37] studies of hospital admissions during the first 12–36 months of life reported, in addition to poverty, an independent effect generated by mothers’ depression. We did not reproduce this relationship in this sample. Rather, consistent with other reports [10, 32], we found that, in addition to poverty, low maternal education and single parenthood are positively associated with hospital admission. Having one’s mother born outside of Canada is associated with a lower risk of hospitalization, suggesting a different pattern of hospital use among those born outside of Canada. This is consistent with results in other studies on child hospitalization in Ontario and Sweden, but the reasons for the lower hospitalization rates among the children of immigrants remain unknown [31, 38].

In our study, a high number of medical visits tends to decrease the likelihood of hospitalization, suggesting that the availability of primary care and pediatric services may decrease the likelihood of hospitalizations among these young children. As far as we know, this negative relationship between availability of primary care and pediatric services and hospitalization has not been previously examined in Canada. As expected, all health needs indicators are associated with increased probability of hospitalization. Asthma deserves particular attention. Asthma is generally more frequent among poor children and children with asthma are also more likely to be hospitalized. This should lead to higher hospitalization rates among the poorest and controlling for asthma should attenuate the association between income and hospitalization. This is not the case in our dataset. The asthma indicator (total number of asthma attacks from birth to the age of 3.5 years) is correlated with hospital admission but is not significantly correlated with our poverty indicators (number of episodes of insufficient income or very insufficient income).

Limitations of the Study

Hospitalization information was based on mothers’ reports and its validity may be subject to question. However, the consistency of the associations between known risk factors for child hospitalization (low maternal education, single parenthood, non-immigrant status, being a boy, and health needs) and hospitalization risk supports the validity of the hospitalization information given by the mothers. Poverty status and the two dynamic measures of poverty used in this study have indicated a significant association with poor health and growth retardation in previous analyses of QLSCD data [810, 12, 13, 16], providing further evidence for their construct validity.

The results may be subject to selection and attrition bias. Selection bias is present in the baseline sample, since 48% of mothers who had not completed secondary school and 45% of mothers who did not speak English or French at home agreed to participate in the original QLSCD sample. However, the truncation of the sample to the more educated mothers and more integrated immigrants could only weaken the associations between hospitalization and these two risk factors. Differential attrition by poverty level is likely to be present, as evidenced in Table 1, since the proportion of the poorest sector of the population decreases in the consecutive waves. It could be argued that the economic situation improved from 1998 to 2003, but it is hard to believe that this improvement resulted in halving the percentage of the very poor from 12 to 6%. Attrition results in lack of precision in multivariable model estimates. In addition, it may bias the results towards the null hypothesis since it results in the shifting of the study sample towards the less poor sector.

Various alternative hypotheses may be advanced for the comparatively lower hospitalization risks among the most disadvantaged children. Our results may reflect poverty-related imbalances in access to family doctors and referrals by primary care physicians, or differences in utilization of emergency departments. This could be the case if providers caring for the poorest children have qualitatively different levels of professional expertise, practice in ways differing systematically from providers who care for more affluent children, or perceive fewer options in terms of appropriate hospital facilities in their practice location. Alternatively, all children may be equally likely to encounter a health care provider who makes the decision to hospitalize (at a common level of illness severity); however parents of the poorest children may be less likely to take their children to health care facilities for various reasons, e.g., lack of health facilities in the underserved areas where they live; lack of transportation; economic consequences of missed work; prior discriminatory or otherwise negative interactions with health care providers, etc.

Results of the stratification of medical visits and hospitalization by poverty level support the existence of barriers to care. Chronically poor children are less likely to visit physicians in ambulatory care as compared to wealthier children (Fig. 1a). When observing hospitalizations by duration of poverty, Fig. 1b shows that those children who seldom visit doctors in ambulatory care are more often hospitalized regardless of their income. However, Fig. 1b also indicates that among chronically poor children whose parents take them to ambulatory care facilities frequently, hospitalization is very low—approximately half the rate of the hospitalization among comparable children from wealthy families. We were, however, unable to provide strong evidence to support our stated hypothesis, since the QLSCD database did not contain sufficient information. This particularly applied to the way in which hospital admissions took place, reasons for non-utilization, or physician practices.

Fig. 1
figure 1

(a) Frequencies of medical visits from birth to 31/2 years, (b) hospitalizations according to frequency of medical visits and poverty status

In conclusion, we provide evidence for lower-than-expected rates of hospitalization among very poor children. Results suggest that socio-economic inequalities in utilization of health care services in infancy and early childhood persist, even in a universal system of care. Qualitative research on health care itineraries may increase our understanding of barriers encountered by parents of these very poor children.