Background Avoidable hospitalisations are hospital admissions for medical conditions that could potentially have been prevented by outpatient healthcare. They are used as an indicator of access to and quality of primary healthcare.
Aim To investigate the association between median area income and avoidable hospitalisation and whether potential differences can be explained by contextual or compositional factors.
Method Median area income was calculated for all 43 city districts and municipalities in Stockholm County during 2005–2007 and grouped into quintiles. The association between median area income and avoidable hospitalisation was studied by calculating age-adjusted rates. To disentangle contextual and compositional effects, ORs with 95% CIs were calculated, controlling for individual age, sex, country of birth, marital status and socioeconomic position.
Results Rates of avoidable hospitalisation were higher in areas with lower income, 1535 per 100 000 inhabitants in the lowest area income quintile compared with 1179 in the highest area income quintile after age standardisation. For the age group 18–64 years, comparing the lowest quintile with the highest quintile, adjustment for individual characteristics of residents (compositional factors) reduced the crude OR from 1.52 (95% CI 1.44 to 1.60) to 1.12 (95% CI 1.06 to 1.19). For the age group 65–79 years, the ORs were 1.28 (1.21 to 1.36) and 1.06 (1.00 to 1.13), respectively. For those aged 80+ years, no association was found with area median income.
Conclusions Higher rates of avoidable hospitalisation in low-income areas indicate greater healthcare needs of people living there. This should be addressed by investing in outpatient care for lower socioeconomic groups. The composition of individuals must be considered when studying area characteristics and avoidable hospitalisation.
- Primary care
- Chronic disease management
- Health services research
- Quality measurement
Statistics from Altmetric.com
Health inequalities by socioeconomic status are found in most societies and primary healthcare has a central role in reducing social inequalities in health.1
Healthcare in Sweden is universal and financed through tax revenues with limited user fees and decentralised to the 21 county councils who are responsible for organising healthcare services.2 Healthcare providers may be either public or private.2 Primary care has a central gate-keeping role and is generally delivered through local healthcare centres. The overall aim of healthcare in Sweden is ‘good health, and health care on equal terms for the entire population’.3 Despite legislation addressing the equity dimension and a universal healthcare, inequalities have been observed in Swedish healthcare. For example, individuals in lower socioeconomic positions refrain from seeking healthcare more often than individuals in higher socioeconomic positions.4 In addition, although access and availability to primary care in theory is equitable, the quality of care and continuity of healthcare providers varies between primary care centres, as doctors may find it less attractive to work in disadvantaged areas. For instance, a recent survey of primary care centres in Stockholm County Council found that centres in disadvantaged areas had lower rates of meeting national targets for the treatment of diabetes.5 Hence, in practice, there may be important area differences in how services are provided.
Many indicators of equity aspects in healthcare have limitations and health outcome is suggested to be the most important dimension when measuring equity in healthcare.6 One indicator used internationally to assess primary healthcare access and quality is avoidable hospitalisation, also called hospitalisation for ambulatory care sensitive condition. The indicator is a composition of medical diagnoses for which hospital admission in many cases could have been averted if patients were taken care of properly at a primary healthcare level.7 ,8
Avoidable hospitalisation has been observed to be more common in areas of lower socioeconomic status.9–12 However, in Paris, area socioeconomic status was not associated with avoidable hospitalisation in a multivariate analysis.10 Avoidable hospitalisation has been found to be more common in older people and among men.10 The association with socioeconomic status is less clear in older people, demonstrating an inverse association in some studies13 ,14 and a non-significant association in others.15 ,16 In the USA higher risk has been observed among non-white people,10 while no difference, or even lower rates, have been noted in Australia among people born in refugee source countries.17
The higher rate of avoidable hospitalisation in lower-income areas may be due to lack of appropriate local primary care services, but it may also be due to the composition of the population of the area, with higher disease prevalence. The prevalence of disease is higher in lower socioeconomic groups, but is not always directly translated into higher demand for healthcare services.18 Higher rates may also be related to hospitalisation practices. Distinguishing between explanations related to contextual factors and population composition may be important when analysing health inequalities.19 Avoidable hospitalisation may be expected to be more common in low-income areas because individuals with lower socioeconomic position and greater disease burden tend to reside in those areas, that is, a compositional explanation. In addition, the theory of the inverse care law suggests that high-quality healthcare is less common when the population has greater need for it.20 ,21 If resource allocation and service provision does not compensate for the higher disease burden in lower-income areas it could result in lower access and quality of primary healthcare in those areas, that is, a contextual explanation.
Avoidable hospitalisation has not been extensively studied in Sweden but public reports indicate higher rates of avoidable hospitalisation among men, foreign born and individuals with lower educational attainment.22 Psychiatric patients are another group with increased rates of avoidable hospitalisation compared with non-psychiatric patients.23
Previous international studies of socioeconomic status and avoidable hospitalisation commonly use neighbourhood data as a proxy for individual socioeconomic position,9 or use ecological data with areas as the unit of analysis.10–16 In these studies it was not possible to link a person's individual income with the risk of avoidable hospitalisation; hence it is uncertain whether the higher rates in low-income neighbourhoods originate from contextual or compositional factors.
In this study we have the unique opportunity to investigate the association between avoidable hospitalisation and median area income in Stockholm, Sweden, and whether potential differences could be explained by individual income and other socioeconomic factors.
A population-based cohort study was performed with residents 18 years and older living in Stockholm County between 1 January 2005 and 31 December 2007. The size of the study population varied from 1 524 554 in 2005 to 1 575 303 in 2007, in total 4 648 974 observations.
Source of data
The patient administrative register of Stockholm County Council was used to identify cases of avoidable hospital admissions. The patient administrative register of Stockholm County includes diagnoses for admission using codes from the International Classification of Diseases 10 and a personal identity number which makes it possible to link these data to other registers. Admission data cover public and private hospitals.
Individual income from the compounded dataset LISA (integrated database for labour market research), Statistics Sweden was used to categorise median area income level. Other sociodemographic variables were also retrieved from LISA and avoidable hospitalisations were subsequently linked at an individual level to data from LISA. Statistics Sweden manages several administrative registers, including the Register of the Total Population, the National Education Register and Swedish tax agency with individual-level information on all residents of Sweden, which have been linked into the LISA database.24
The areas under study were the 25 municipalities in Stockholm County and, at that time, the 18 city districts of Stockholm municipality, in total 43 areas. Municipalities collect taxes and are responsible for schools, social and caring services, and collaborate closely with the County Council, for instance in providing care for older people. Stockholm municipality with some 900 000 inhabitants is divided into city districts, which operate services locally under the umbrella of the municipality. These administrative units are larger than neighbourhoods. Municipalities and city districts vary in size from about 10 000 to 900 000 residents, the median size being about 40 000. Stockholm municipality was represented by its constituent 18 city districts. Based on their median income, municipalities and city districts in Stockholm County were grouped into quintiles. Median income was calculated from individualised disposable household income (the family's compounded disposable income divided by the family members’ total consumption weight).24 Individuals with a disposable income of zero or below (from capital losses or deficits) were excluded in the calculation of area median income because of the heterogeneity of this group.
Avoidable hospitalisation was defined by a hospitalisation for any ambulatory care sensitive condition as categorised by the National Board of Health and Welfare and Swedish Association of Local Authorities and Regions.25 This categorisation has been used in Sweden for several years and is a modified version of the one used in Australia, for example, vaccine preventable diseases have been omitted as Sweden has a close to complete national coverage of vaccination in children.17 ,26 Avoidable hospitalisation included the following chronic conditions: anaemia, asthma, diabetes complications, heart failure, hypertension, chronic obstructive pulmonary disease and angina, and the acute conditions: bleeding stomach ulcer, diarrhoea, pelvic inflammatory disease, pyelonephritis, and ear, nose and throat infections (see online appendix 1).25 The outcome was binary and an individual either had or had not been admitted to hospital for any of those conditions.
Information on covariates on an individual level was derived from LISA. Demographic variables were age, sex, marital status and country of birth. Age was categorised into groups of 10 years, and country of birth was categorised into born in Sweden/born outside Sweden. Socioeconomic variables were individual income and education, the latter categorised as compulsory school/upper secondary school/higher education. Individual income was the same individualised disposable household income used to calculate median area income, but grouped into quintiles based on individuals. Additional socioeconomic variables studied were gainful employment, sickness benefit and social assistance within the family, all categorised as yes or no.
Crude and age standardised rates were calculated for area income quintiles of the proportion with avoidable hospitalisation per 100 000 inhabitants per year. Individuals contributed to one, two or three observations depending on how many of the years 2005–2007 they were residing in Stockholm County.
In logistic regression analyses, observations are assumed to be independent,27 hence an individual could only contribute to one observation. To fulfil this, a new summarised dataset was created with each individual contributing to only one observation, including only individuals living in Stockholm during all 3 years. From the original dataset, 354 960 observations were excluded for this reason and a resulting 1 431 338 observations were left in the summarised dataset. For the summarised dataset, background characteristics from 2005 were used and a new variable with avoidable hospitalisation was created with admitted/not admitted during the 3-year period 2005–2007. ORs of avoidable hospitalisation were first calculated for each of the individual covariates, controlling for age and sex. ORs were then calculated for area income quintiles. Except for a crude model adjusted for age and sex, a multivariate logistic regression model was performed with each of the covariates. A final logistic regression model was performed with all covariates included. The study population used in the multivariate logistic regression models was restricted to only individuals with valid values on all covariates, which equalled 1 404 937 observations. Analyses were stratified by sex and age, and the final model for age groups over 64 years did not include sickness benefit and gainful employment since these variables are not relevant in these ages. The final model was checked for multicollinearity using the procedure regression and the option variance inflation factors VIF in SAS.28 Multicollinearity appeared not to be a substantial problem since VIF values were under 2.
The average composition of individual characteristics during 2005–2007 is presented for area income quintile one, three and five, and for Stockholm County in total in table 1. Lower-income areas had slightly higher proportions of young people, men, married and divorced people. Being born outside Sweden, having only compulsory school, not being gainfully employed and receiving sickness benefit and social assistance were also more common in lower-income areas.
Rates of avoidable hospitalisation
The mean number of people with avoidable hospitalisation per year was 18 956. However, since an individual could have been admitted more than once during a year the actual number of avoidable hospitalisations was higher, on average 29 841. The crude rate, that is, people with avoidable hospitalisation per 100 000 inhabitants, increased over the years from 1215 in 2005 to 1235 in 2007, on average 1223 persons per 100 000 inhabitants. The crude and age-standardised rates of avoidable hospitalisations per area income quintile are presented in Otable 2. The crude rates of avoidable hospitalisation were higher in lower-income areas with the lowest area income quintile having 18% higher rates than the highest area income quintile (1330 and 1130, respectively). After age standardisation, the rates were 30% higher in the lowest income quintile compared with the highest income quintile (1535 and 1179, respectively). There was a similar inverse relationship between area income and avoidable hospitalisations among men and women when analysed separately.
Rates of hospitalisation for chronic and acute conditions per area income quintiles are presented in table 2. There was an inverse association by area income quintile for chronic and acute conditions in the age group 18–64 years: the higher the area income quintile, the lower the rates of chronic and acute avoidable hospitalisation. Also for the age group 65–79 years, lower-income areas had higher rates of chronic and acute conditions, although the gradient was less pronounced in this age group. Rates of hospitalisations for the population 80 years and above showed no clear relationship with area income, either for chronic or acute conditions.
Logistic regression analyses of avoidable hospitalisation
All covariates studied were associated with avoidable hospitalisation, with older individuals, men, divorced people and widows/widowers, and people with lower educational attainment having higher age-adjusted and sex-adjusted odds of avoidable hospitalisations (data not shown). Being born outside Sweden was associated with higher odds of avoidable hospitalisation. However, closer investigation revealed that individuals born in some parts of the world had lower odds than Swedish-born people, while others had higher odds than Swedish-born people (data not shown). Individuals with lower income and those not gainfully employed had increased likelihood of avoidable hospitalisation, as did recipients of sickness benefit and social assistance.
ORs per area income quintile are presented in table 3. In the crude model, controlling for sex and age, lower-income areas had higher odds of avoidable hospitalisation compared with higher-income areas (quintile 1 (Q1) OR=1.29, 95% CI 1.25 to 1.33). Marital status was significant in the model but did not explain the association. Controlling for country of birth partially explained the higher odds in lower-income areas. Income on an individual level explained a substantial part of the differences between high-income and low-income areas, and so did educational level. Gainful employment, sickness benefit and social assistance all partly explained the higher odds in lower-income areas when analysed separately. In the final model, all covariates simultaneously included, the three lowest area income quintiles still had significantly higher odds of avoidable hospitalisation than the highest area income quintile, although the ORs were reduced (Q1 OR=1.05, 95% CI 1.01 to 1.08).
Stratified analyses were made for men and women with similar results (data not shown). Logistic regression analyses were also stratified by age group and showed higher ORs for lower-income areas in the crude model in age groups 18–64 and 65–79, while no significant difference was seen between areas among 80+ years old. The final model, with all covariates included, showed that people in the youngest age group living in lower-income areas had higher odds of avoidable hospitalisation than those living in the highest income areas, even after controlling for individual characteristics (Q1 OR=1.12, 95% CI 1.06 to 1.19). All variables included were statistically significant except country of birth. For the age group 65–79 years, the odds were significantly higher in all area quintiles, except for the lowest income quintile, compared with the highest income areas. For the oldest age group no differences were observed between area income quintiles after controlling for individual characteristics. For the two older age groups, all control variables in the final model were statistically significant (including individual-level income).
In addition, analyses stratified by age group, individual income quintile and area income quintile showed that individuals on a low income residing in lower-income areas tended to have increased odds of avoidable hospitalisation compared with individuals on a low income residing in more affluent areas (table 4). This tendency was more pronounced in age groups 18–64 years and 65–79 years, and was not seen among those aged 80+ years.
This population-based cohort study showed higher rates of avoidable hospitalisations in low-income areas compared with high-income areas. Among those 18–64 years old avoidable hospitalisation for chronic and acute conditions was more common in lower-income areas. An inverse association was also found among those aged 65–79 years, although less pronounced, especially for acute conditions. In the age group 80+ no relationship with area income was observed. Adjustment for individual characteristics explained a large part of the differences seen between high-income and low-income areas.
Strengths and limitations
The strengths of this study include the large sample size of the whole population aged 18 years and older in Stockholm County, and the complete coverage of avoidable hospitalisations. A further major advantage was the use of individual income to calculate area income level. Also, the possibility to adjust for individual-level characteristics, including several measures of socioeconomic position, was a major strength. The main weakness was the use of municipalities and city districts to define geographical areas, because these geographical boundaries did not directly correspond to administrative healthcare areas. The variation in size of the geographical areas limits the degree to which an area represents a neighbourhood; a lower level of aggregation would better approximate a neighbourhood. However, using municipalities and city districts to define areas may facilitate the interpretation of the results and policy implications. Although the results indicate that the major problem may not be contextual factors, data on access to and quality of primary care of the areas, for example, number of visits per physician or physicians per inhabitants, would have provided better evidence, but was not available. In addition, the risk for avoidable hospitalisation may be influenced not only by primary care but also by the vicinity to other forms of outpatient care, for example, hospital-based specialist care.
Nations with universal healthcare have demonstrated lower social disparities in avoidable hospitalisation than nations with other healthcare systems.29 ,30 Equity in access to healthcare has long been an overarching goal in Sweden,3 although not entirely realised.4 Generalising our results to countries with other healthcare systems should be made with caution. Swedish healthcare has become more market oriented in the last few years, and studies with more recent data are needed to investigate whether primary care reforms have affected inequalities in avoidable hospitalisation. However, the data in the present study refer to a previous time period which would not be affected by the reforms.
Comments on findings
The findings of this study suggest that the higher rates in low-income areas are mainly explained by the higher proportion of individuals in lower socioeconomic positions and other vulnerable groups living there, whose outpatient healthcare needs do not seem to be adequately met. However, individuals on a low income residing in low-income areas had higher odds of avoidable hospitalisation compared with individuals on a low income residing in more affluent areas. The inverse association of individuals’ own socioeconomic position and avoidable hospitalisation has also been observed in the USA.31
The higher rates among individuals in lower socioeconomic positions are probably due to multiple causes. The inequalities in avoidable hospitalisation are likely to be shaped not only by primary care access and quality but also by broader social circumstances. Disease strikes earlier in lower socioeconomic groups, as shown in the age standardised rates and also observed elsewhere.9–14 ,29 The higher rates may also be partly attributed to individual characteristics not studied, for example, smoking and severe alcohol misuse which have been associated with avoidable hospitalisation.32 A higher disease prevalence is likely to be one major contributing factor, but the risk of being hospitalised among those with a disease may also be higher for lower socioeconomic groups. Health literacy, material resources, healthcare-seeking behaviour, physician communication skills and practices could possibly play a role in this.
The remaining differences in avoidable hospitalisation by area income for the younger age groups could be due to area contextual factors, for example, lower access and quality of primary care. Healthcare services could potentially compensate for higher disease prevalence and adverse social circumstances in lower socioeconomic groups by providing more appropriate and comprehensive primary healthcare to this group. The fact that individuals on a low income residing in low-income areas have higher rates of avoidable hospitalisation than individuals on a low income residing in high-income areas indicates that there may be a difference in the provision and quality of primary care services between high-income and low-income areas, as also found in a previous study,5 suggesting that there is potential for improvement of services in low-income areas.
The validity of avoidable hospitalisation as an indicator of primary care access and quality has been discussed, and a recent systematic review found that lower access to primary care was associated with avoidable hospitalisation in general, but that adjustment for socioeconomic factors is important.8 A study of primary healthcare visits prior to an event of avoidable hospitalisation revealed that people living in lower-income areas make more visits to physicians before being hospitalised for avoidable conditions than individuals in higher-income areas,12 suggesting that it is more than accessibility of primary healthcare of an area that matters. Studies of communication between healthcare provider and patients in different settings may be warranted.
Future studies may focus on evaluating interventions at the primary healthcare level with the potential to reduce inequalities in avoidable hospitalisation, and at a structural level evaluating which resource allocation and reimbursement systems are most efficient in decreasing the higher rates among individuals in lower socioeconomic positions. However, increasing rates of avoidable hospitalisations and persistent inequalities in the UK have shown that it is difficult to pinpoint appropriate policy.33 ,34
This study shows that population characteristics play a major role in observed area healthcare inequalities. However, higher rates of avoidable hospitalisation in lower-income areas indicate that healthcare needs are not adequately met. This suggests a need for interventions and investments in outpatient care for lower-income groups, preferably directed to low-income areas.
We would like to thank colleagues in the Equity and Health Policy research group, Karolinska Institutet, for giving valuable feedback on this study.
This web only file has been produced by the BMJ Publishing Group from an electronic file supplied by the author(s) and has not been edited for content.
Files in this Data Supplement:
- Data supplement 1 - Online appendix
Contributors TL, BB and RL conceptualised the idea and designed the study. TL and AW analysed data and all authors contributed to the interpretation of the results. TL led the drafting of the article, but all authors participated in critical revisions and have granted final approval of the version to be published.
Funding The study was partly funded by Stockholm County Council.
Competing interests None.
Ethics approval The study was approved by the Regional Ethical Review Board in Stockholm (Dnr 2008/1004-32).
Provenance and peer review Not commissioned; externally peer reviewed.
If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.