Introduction

Low socioeconomic status (SES) is associated with higher mortality rates and more years lived with disability and disease [1]. Yet, in most low- and middle-income countries, low SES groups utilize less healthcare and of worse quality [2], a phenomenon referred to as the “inverse care law” [3]. Despite universal healthcare, socioeconomic differences in healthcare utilization (HCU) persist also in many modern welfare states, where low SES groups commonly use more healthcare but not in proportion to their additional needs (sometimes called the “disproportionate care law”) [2,4]. Similar to many other European countries, the Swedish law mandates that the healthcare should be provided in proportion to the need of individuals [5]. However, the degree to which this has been accomplished is not conclusively determined.

In a recent systematic review of 57 studies from high-income countries, Lueckmann and colleagues [4] report that in the majority of studies, higher SES was associated with higher utilization of specialized outpatient care for a given self-reported health status. The findings of primary care utilization were more heterogeneous, and the authors concluded that inequities were more common in specialized outpatient care compared to primary care. Hence, Lueckmann and colleagues [4] pointed out that research on socioeconomic inequalities in HCU should differentiate between primary care and specialized outpatient care, since the results differ greatly according to the type of service.

To evaluate the effects of public policy and healthcare reforms on inequality in HCU and mortality, analysis of time trends are informative [6]. Although Sweden was successful in reducing inequalities, for instance, by providing universal healthcare and free education, relative inequalities in mortality have increased since 1990 [7]. The Swedish healthcare system has, during the same period, undergone major changes, such as the marketization of primary care (spurred by the Primary Health Care Choice Reform in 2010) and a recent increase in virtualization of primary care. Although previous research indicates that marketisation increased inequalities in HCU [8], the evidence is sparse and incomplete [9]. Similarly, studies of trends in inequalities in utilization of specialized care show mixed results [10,11], and updated, large-scale studies are lacking. Furthermore, the extent to which trends in socioeconomic differences in HCU at various levels of care compare to trends in mortality inequalities has not been thoroughly studied.

The Scandinavian health registers provide unique opportunities to investigate this. In Sweden, national population healthcare registers are regularly updated [12], and microdata on HCU and death can be linked to social databases through civic registration numbers. This enables analysis of socioeconomic differences and utilization of primary and specialized outpatient and inpatient care as well as mortality for the general population over time. Unfortunately, Sweden is still lacking a national register of primary care utilization. To our knowledge, this is the first study studying income-related differences at all levels of care at the national level.

The aim of this study is to assess time trends of income-related differences in use of different levels of healthcare in relation to mortality between the years 2004 and 2017. To do this, we compared the lowest (Q1) with the highest income quintile (Q5) with regard to the utilization of primary care, specialized outpatient care, and inpatient care. To facilitate interpretation of income-related group differences in HCU, we included mortality as a comparative outcome measure. We studied each of the 5 disease groups that account for the largest number of deaths in Sweden 2017 (cardiovascular diseases, neoplasms, neurological disorders, chronic respiratory diseases, and diabetes and chronic kidney diseases), as well as all-disease. Specifically, we addressed the following questions: (1) How do low- and high-income groups differ with regard to HCU and mortality in 2017 and over time? (2) To what extent do income-related differences in HCU and mortality vary across disease groups? (3) How do the low- and high-income groups differ in numbers of healthcare encounters ≤5 years prior to death?

Methods

The study is reported according to the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guideline (S1 Checklist). No study protocol was submitted prior to submission.

Study population and data sources

The study population comprised all individuals 16 years and older living in Sweden, any year between 2004 and 2017, that were registered in the Longitudinal Integrated Database for Health Insurance and Labour market Studies [13] by Statistics Sweden (Table 1). Data on specialized out- and inpatient HCU (from the National Patient Register) and mortality (from the Cause of Death Register) were obtained from the Swedish National Board of Health and Welfare (NBHW). Since primary care data are not available at the national level, we contacted the 21 Swedish counties maintaining electronic records of diagnoses in primary care. Seventeen counties provided time series covering one or more years (S1 Table). Whereas the national health registers provided by NBHW have well-documented and high population coverage [12], primary care data lacked estimates of coverage rates, which likely varied by region and over time. Hence, crude rates of primary care utilization could partly be affected by increasing coverage rates and should be interpreted with caution. For disease-specific analyses, we selected the five disease categories causing the highest number of deaths in Sweden in 2017, which together were responsible for 84% of all deaths [14]. The disease groups were (with percentage of the total number of deaths in parentheses): cardiovascular diseases (38%), neoplasms (28%), neurological disorders (8%), chronic respiratory diseases (5%) and diabetes and chronic kidney diseases (pooled) (4%). Diseases were classified using the tenth version of the WHO International Classification of Diseases (ICD-10) codes in accordance with the disease groups defined by the Global Burden of Disease project (S2 Table). Whereas NBHW define specialized care and underlying cause of death by ICD-10, diagnoses in primary care were classified using ICD-10-P which we mapped to ICD-10. Since ICD-10-P is less detailed than the ICD-10, certain disease groups in the primary care contain marginally broader disease categories than the GBD target groups.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Characteristics of the study populations.

https://doi.org/10.1371/journal.pmed.1004230.t001

Results

For the analyses of mortality and utilization of inpatient and outpatient care, we investigated 105.3 million person-years at risk. For primary care, we had access to data on 77.8 million person-years at risk. Outcomes in the cohort amounted to 1.3 million deaths, 9.9 million year-unique individuals in inpatient care, 38.4 million in outpatient care, and 45.9 million in primary care (Tables 1 and S3).

Fig 1 shows OR for utilization of primary care, outpatient care, inpatient care, and mortality for each of the disease categories in 2017. Considering all-disease, those with lowest income (Q1) utilized slightly more primary care (OR = 1.07, 95% CI [1.07, 1.08]; p < 0.001) and outpatient care (OR = 1.04, 95% CI [1.04, 1.05]; p < 0.001) compared to those with the highest income (Q5). However, Q1 utilized considerably more inpatient care (OR = 1.44, 95% CI [1.43, 1.45]; p < 0.001) and had even higher death rates (OR = 1.78, 95% CI [1.74, 1.82]; p < 0.001) (Fig 1 and S4 Table).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Adjusted ORs of HCU and mortality comparing the lowest with the highest income quintile, in year 2017.

ORs are adjusted for sex, age, age squared, country of birth, and civil status. Unadjusted results are available in S4 Table and yielded largely similar results. All, any ICD-10 code; CVD, cardiovascular diseases; Neuro, neurological disorders; Chro. Resp., chronic respiratory diseases; DM & CKD, diabetes and kidney diseases; HCU, healthcare utilization; OR, odds ratio.

https://doi.org/10.1371/journal.pmed.1004230.g001

Similar patterns of income-related HCU and mortality were observed for each investigated disease. For both cardiovascular diseases and neoplasms, ORs were lowest for utilization of primary and outpatient care and were considerably larger for inpatient care. The largest relative difference was found for mortality. Interestingly, use of primary and outpatient care due to neoplasms were even lower in Q1 than in Q5 (primary care: OR = 0.58, 95% CI [0.57, 0.58]; p < 0.001; outpatient care: OR = 0.78, 95% CI [0.78–0.79]; p < 0.001) (Fig 1 and S4 Table).

Fig 2 shows time series of rates of HCU and mortality for Q1, Q5, and for the total study population. For all the 6 disease categories, the yearly rates of unique individuals in primary and outpatient care are increasing. Rates of inpatient care plateaued around 2012 for most disease categories and decreased by the end of the study period. All-disease mortality rates for Q5 and the total population declined throughout the study period but stayed relatively unchanged for Q1. Cardiovascular diseases contributed most to decreasing mortality rates in the total population (S3 Table). Time series of rates of HCU and mortality for each of the 5 income quantiles are shown in S1 Fig.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Time courses (2004–2017) of rates (cases per 100) and adjusted ORs of HCU and mortality.

HCU (yearly rates of unique individuals) and mortality rates, by income group and for the total study population, are shown in blue colors. Rates are presented in percentages for display purposes, and corresponding numerical rates per 100,000 are found in S3 Table. Adjusted ORs (95% CI) comparing Q1 with Q5 are in red, and shaded areas denote 95% CI. ORs are adjusted for sex, age, age squared, country of birth, and civil status. Unadjusted results are available in S4 Table. Red horizontal grid lines denote OR = 1, and the black grid lines denote the zero levels of rates. All, any ICD-10 code; CVD, cardiovascular diseases; Neuro, neurological disorders; Chro. Resp., chronic respiratory diseases; DM & CKD, diabetes and kidney diseases; HCU, healthcare utilization; OR, odds ratio.

https://doi.org/10.1371/journal.pmed.1004230.g002

Fig 2 also shows time series of adjusted OR comparing Q1 with Q5. We observe an overall pattern of slightly decreasing ORs in primary and outpatient care, but increasing ORs in inpatient care and mortality, for most disease categories. Nearly identical results were obtained from the sensitivity analysis in which we used the same population for specialized care and mortality as was used for primary care (S2 Fig).

Zooming in on the time trends of OR, Fig 3 displays the beta estimates (i.e., the slope) of log-transformed ORs regressed on time, which denotes the annual rate of change in the relative differences between Q1 and Q5. Time trends of ORs for all-disease followed a similar pattern as the one observed for 2017. Thus, for each year, Q1 (relative to Q5) displayed increasing rates of inpatient care, and even larger increase in mortality rates, but decreasing rates of primary and, especially, outpatient care. For most of the specific disease categories, including cardiovascular diseases and neoplasms, we observed a similar pattern of HCU and mortality trends (Fig 3 and S4 Table).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. Yearly change in log-transformed adjusted ORs.

Bars show the beta-coefficients (i.e., rates of change or “slopes”) estimated by linear regression of the log-transformed time-series of OR regressed on years (2004–2017). Error bars denotes 95% CI. Analyses are adjusted for sex, age, age squared, country of birth, and civil status. Numerical values for both adjusted and unadjusted are found in S4 Table. All, any ICD-10 code; CVD, cardiovascular diseases; Neuro, neurological disorders; Chro. Resp., Chronic respiratory diseases; DM & CKD, Diabetes and kidney diseases; OR, odds ratio.

https://doi.org/10.1371/journal.pmed.1004230.g003

Fig 4 shows income-related difference in number of disease-specific healthcare encounters within 5 years prior to death. In 2017, Q5 had significantly higher rates of both primary and outpatient care utilization across all 5 disease categories. Notably, for neoplasms, Q1 had only half the number of primary care encounters compared to Q5 (RR 0.50, 95% CI [0.40, 0.61]; p < 0.001) and a slightly less reduced number in outpatient care (RR 0.83, 95% CI [0.79, 0.88]; p < 0.001). Group differences in outpatient care were largest for neurological disorders (RR 0.50, 95% CI [0.45, 0.56]; p <0.001). For inpatient care, RRs did not significantly differ from 1 in any disease, except from all-disease, where Q1 displayed higher rates (RR 1.17, 95% CI [1.11, 1.23]; p < 0.001).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 4. Average number of healthcare encounters and adjusted rate ratios (RRs) ≤5 years prior to death.

The number of healthcare encounters by income group (Q1 and Q5) and for the total study population are displayed in blue. Adjusted RRs comparing Q1 with Q5 are in red, and shaded red areas denote 95% CI. RRs are adjusted for sex, age, age squared, country of birth, and civil status. Both adjusted and unadjusted results are available in S5 Table. Horizontal grid lines for RR = 1 (red) and encounters = 0 (black) are shown in the figures.

https://doi.org/10.1371/journal.pmed.1004230.g004

The post hoc sensitivity analysis of unadjusted ORs and unadjusted RRs did not alter the original results in any way that significantly impacted the conclusions, as detailed in S4 and S5 Tables. However, in the sensitivity analyses where county was controlled for (i.e., Model 3), the OR for primary care was generally higher than in the original analysis. The strongest modifications were the marked reductions in income-related differences in number of primary healthcare encounters ≤5 years prior to death: In 2017, all group differences were erased in each of the 6 disease categories, apart from neurological disorders (S5 Table).

Discussion

In this study, we investigated time trends of income-stratified HCU, along with mortality as a comparative outcome. Whereas income-related differences were small for primary and outpatient care, they were larger for inpatient care. This was true for all the investigated disease groups. However, except for neurological disorders and diabetes and chronic kidney diseases, the largest inequalities were observed for mortality. Considering all-disease at the end of the study period (2017), the lowest income quintile had 7% higher odds of utilizing primary care, 4% higher odds of utilizing specialized outpatient care, 44% increased odds of hospitalization, and 78% higher odds of dying. The largest discrepancy between group differences in utilization of primary and outpatient care as compared to mortality was observed for neoplasms. Overall, the large inequality in mortality rates were not reflected in correspondingly large group differences in HCU, and the discrepancy between large differences in mortality and lower in HCU was further widened each year.

A similar pattern was found when investigating the number of healthcare encounters ≤5 years prior to death: The largest group differences were observed in primary care, followed by outpatient care. For inpatient care, there were no significant differences. Again, group differences in primary care were largest for neoplasms, whereas group differences in utilization of outpatient care were largest for neurological disorders.

Although trend analyses of income-based HCU inequalities are sparse, our results are in line with previous cross-sectional survey studies. In Italy, Petrelli and colleagues found that high-income earners utilized both more primary and outpatient care, but less inpatient care, compared to low-income earners [16]. This pattern is well documented in a wide range of high-income countries [4,17,18], although there are a few older studies that only partly confirmed this [19,20].

Analogous to our findings, a longitudinal Norwegian survey study showed that socioeconomic differences in primary care utilization diminished over time between 1986 and 2006 [21]. Moreover, groups with high SES were more prone to utilize outpatient care, after adjustment of self-reported need. Finally, the income-based ratio of inpatient care was stable through the period, with considerably higher rates among low-income groups [21]. Similarly, in a Swedish sample from the years 1996/1997, Burström found an inverse income gradient related to having the need, but not seeking medical care [10].

This study adds to the previous literature by providing time trends in income-based differences in care utilization, stratified by disease groups. We observe a consistent and notable difference in rates of all-disease inpatient care and mortality between Q1 and the other income quintiles. Our results indicate that low-income earners tend to underutilize primary and outpatient care, especially for neoplasms and chronic respiratory diseases. Lower rates of healthcare for prevention and early treatments might contribute to higher rates of inpatient care. Meanwhile, low-income groups have higher HCU for neurological disorders that are less responsive to treatment. Further research exploring the benefits of allocating resources toward prevention and early treatments particularly for neoplasms and chronic respiratory diseases among low-income groups would be beneficial.

The importance of improving preventive and early care for low SES groups is also highlighted by a recent report on disparities in Swedish cancer care. The study revealed that individuals with lower education experienced delayed access to care, longer waiting times, and suboptimal treatments for various types of cancers, compared to individuals with higher education [22].

It is noteworthy that the association between SES and HCU in part depends on the operationalization of HCU. Lueckmann and colleagues [4] found that studies that defined HCU in terms of the probability of having visited a physician (“yes or no”) reported larger inequalities in favor of high SES groups than studies reporting on frequencies (i.e., the number of healthcare encounters) or, more rarely, conditional frequencies (meaning the number of encounters given the individual had visited a physician at least once). Hence Lueckmann and colleagues suggested that HCU inequalities would best be tackled by removing barriers in initial access to specialized outpatient care for deprived groups [4]. In the current study, we found that low income was associated with both reduced probability and reduced frequency (in the last 5 years of life) of primary and outpatient care utilization.

The importance of adjusting for need when analyzing socioeconomic differences in HCU is widely acknowledged [23]. “Need” may be conceptualized as the capacity to benefit from care. Need is commonly measured either as self-reported general health status or as self-reported presence of disease. Among the studies listed by Lueckmann, the majority employed the former. However, for population-wide register data, measures of subjective health are typically absent (as in [6]), which is also the case in our study. Instead, we compared income-related differences in HCU with group differences in mortality. Since we did not adjust income-related HCU differences for differences in need, they were likely due to a combination of differences in prevalence, access to care, and care-seeking behavior independent of prevalence. The relative importance of each of these factors is, however, unknown to us and likely play out differently in primary care, outpatient care, and inpatient care, respectively. Whereas income-related differences in utilization of primary and outpatient care likely are driven by differences in both prevalence and care-seeking behavior, income-related differences in inpatient care are probably driven almost exclusively by differences in prevalence of the conditions requiring hospital admission. This is further substantiated by the observed corresponding magnitudes of mortality differences. It is tempting to speculate that given perfect horizontal equity (i.e., equal treatment for equal need), any relative group differences in utilization rates of primary and outpatient care should be of similar magnitude as differences in inpatient use and mortality. However, the fact that we found substantially larger OR for mortality and inpatient care could also be due to reasons other than HCU inequalities. For instance, if low SES individuals would be more prone to more severe conditions (for instance, comorbidities) for which primary and outpatient care are not a primary point of entry, one would expect larger relative group differences in inpatient care and death. To better determine this, future studies should use appropriate control for group differences in medical need.

Despite not adjusting for HCU need, we can still draw important conclusions about the trends in HCU inequalities. If we assume a stable relationship between need and mortality throughout our study, the increasing discrepancy between income-related differences in HCU and mortality suggests an increase in HCU inequalities.

Several explanations for increasing HCU inequalities could be hypothesized. Although financial barriers to access to care have been largely removed in modern welfare states, the so-called disproportionate care law suggests that socially disadvantaged people receive more healthcare but of worse quality and insufficient quantity to meet their additional needs [2]. An increasing number of private health providers and the free right to establish care centers in any location has, in recent years, reduced the number of child and maternal care units in poorer areas where the need is greater [24]. There is also evidence of a preference for establishing healthcare facilities in high-income areas among private providers [25]. Also, the virtualization of primary care likely contribute to HCU inequalities, considering higher rates of digital HCU among individuals with higher education and income, even after adjusting for age and prevalence of chronic diseases [26]. In addition, availability is further skewed in favor of high SES HCU due to the increase in private insurances and out-of-pocket healthcare spending [27]. However, since our study is based on records of publicly financed healthcare usage, privately paid care is omitted from our analyses. If anything, including it would likely further accentuate the observed discrepancy between income-based differences in HCU and mortality.

Interestingly, in the sensitivity analyses in which we also adjusted for county effects, the ORs for primary care were higher, and income-related disparities in the number of primary healthcare encounters during the last 5 years of life were greatly diminished. The most significant reduction in group differences was observed for neoplasms. These findings align with a recent report that highlighted regional and socioeconomic disparities in Swedish cancer care, concluding that regional factors play a pivotal role in driving healthcare inequalities in cancer treatments [22].

To the best of our knowledge, this is the first study to present time series of income-related group differences in the utilization of primary, outpatient, and inpatient care alongside differences in mortality rates, covering the Swedish adult population for more than a decade. This was made possible through a unique, comprehensive database of linked microdata, which consisted of sociodemographic information, cause of death, and HCU both in specialized care and in primary care across most Swedish counties.

Most previous studies on SES differences in HCU rely on survey data. Although these provide valuable information on self-reported need of care, survey studies are typically based on just a small fraction of the general population, interviewed at only a few time points. National population registers, on the other hand, provide (near) full population coverage and enable unbiased definitions of income groups [13]. Importantly, the continuously recorded HCU register data enabled detection of time trends with high precision.

This study has several limitations. While it provides a comprehensive overview of time trends in income-related differences in HCU and mortality, it does not delve into a detailed analysis of specific patterns of group differences in HCU. A more nuanced understanding of these patterns would be valuable for identifying the mechanisms underlying avoidable health inequalities and to guide targeted interventions. For instance, pooling relatively heterogeneous diseases into broad disease categories masks the more extreme and possibly preventable income-related differences in HCU. Further, our metrics of HCU tell little about inequalities in the quality and content of the provided treatments.

Another limitation is the unknown and time-dependent coverage of primary care data. Coverage is likely higher in the more population-dense counties, which typically also coincide with higher income gradients and higher concentrations of high-income individuals. If true, this might partly explain the relatively high primary care utilization among high-income earners. However, similar patterns of high HCU among high-income earners are also observed in the outpatient care, for which the data coverage has been consistently high [28]. A national register of primary care data akin to what is available in, for example, Norway and Finland, would prove very valuable for future research.

Further studies should investigate socioeconomic differences in HCU patterns with higher resolution in terms of disease categories, treatments, and patient flows [29], for instance, by including income, HCU, and mortality in a common statistical model. Our study indicates that a promising target for reducing health (care) inequalities is to facilitate access to primary and outpatient care for low-income groups. Increased knowledge on how to promote care-seeking related to neoplasms, cardiovascular diseases, and chronic respiratory diseases in low SES groups would be particularly valuable, considering their high prevalence and the large discrepancies between income-related differences in HCU and income-related differences in mortality for these disease categories.

To conclude, in this study, we have identified major discrepancies between income-related differences in HCU and income-related differences in mortality, which widened with time. While high-income earners utilized an increasing share of primary and outpatient care, the inequality in mortality increased each year. This was especially noticeable for neoplasms. Facilitating motivated use of primary and outpatient care among low-income groups could help mitigate the growing health inequalities.

Supporting information

Acknowledgments

We are grateful to all providers of register data and, in particular, to the providers of regional primary care data. We also thank Hugo Sjöqvist, Lode Van Der Velde, and Dorien Beeres for valuable discussions on the data analyses, and Kasra Zarei for language proofreading.