Income-based differences in health care utilization in relation to mortality: Trends in the Swedish population between 2004-2017

Study population and data sources

The study population comprised all individuals 16 years and older living in Sweden any year between 2014 and 2017, that were registered in the Longitudinal Integrated Database for Health Insurance and Labour market Studies (11) by Statistics Sweden (Table 1). Data on specialized out- and inpatient HCU (from the National Patient Register, NPR) and mortality (from the Cause of Death Register, CDR) were obtained from the Swedish National Board of Health and Welfare (NBHW). Since primary care data is 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 (10), primary care data lacked estimates of coverage rates, which likely vary 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 (12). The disease groups were (with percentage of the total number of deaths in parenthesis): cardiovascular disease (38%), neoplasm (28%), neurological disease (8%), chronic respiratory disease (5%) and diabetes and chronic kidney disease (4%). Diseases were classified using 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.

Data on HCU and mortality were individually linked to register based data on family income, sex, birth year, residence county, country of origin, and civil status, as provided by Statistics Sweden.

Income Measures

The study population was grouped by income ranks based on equivalised disposable family income. Family income referred to wages, capital returns, self-employment, pensions, and social benefit, after taxes were deducted (11). For each studied year, an individual’s family income was averaged across the 1-5 preceding years, and the resulting variable was subsequently used for income ranking. Individuals with a family income of zero or below (<2% of the population) were excluded from analysis due to the heterogeneity of this group (13). SES groups were defined as quintiles of time averaged family income, stratified by sex and birth year. Thus, at a given year, each income quintile contained approximately the same number of individuals and were balanced with regard to sex and age compositions.

Statistical analysis

Crude rates of HCU per 100 000 person / year were calculated for each disease group and for each of the three levels of care. Only main diagnoses were considered. Mortality rates were based on underlying cause of death. For main analyses, we used the number of unique individuals, where an individual was counted once per disease group, calendar year and type of data source (i.e., primary care, outpatient, inpatient or death). Given the varying coverage rate of primary care register data, the counties contributing to the population denominator varied across the years, why it ranged from 4.1 million individuals in 2004, to 7.1 million in 2017. The population denominator used for the national health register data ranged from 7.1 million individuals in 2004 to 8.0 million in 2017 (Table 1).

We conducted multiple logistic regressions to calculate adjusted odds ratios (OR) of HCU or death, comparing the lowest (Q1) and highest (Q5) income quintiles. The model adjusted for country of origin, civil status, sex, age and age squared.

To infer time trends in ORs, we regressed log transformed OR on time. To account for uncertainty of the OR, we conducted bootstrapping analysis by simulating new time series (1000 resamples with replacements) drawn from a normal distribution with mean and standard deviation equal to the log transformed ORs. For each simulated time series, we performed linear regressions (GLM), and inferred 95% CI from the sampling distribution of the estimated beta coefficients.

Finally, we investigated whether the income groups differed regarding the number of health care visits during a five-year period prior to death. More specifically, for each year and each disease group, we specified a linear regression model of health care encounters by level of care, using income quintiles as independent variables with Q5 as the reference. We controlled for age, sex, county, and country of origin.

Analyses were carried out in SAS 9.4 and in Python 3.6 using standard data science packages such as statsmodels. All code is available at https://github.com/parflo/TrendsInMortalityHCUInequality.git.

The study was approved by the Regional Ethical Review Board, Stockholm, Sweden (DNR: 2018/1339-31/5, 2018/2292-32, 2019-02185, 2021-00657 and 2022-03111-02).

Trends in HCU inequalities

Our findings of income-based HCU inequalities are in line with previous survey studies. In Italy, Petrelli et al. found that high income earners utilized both more primary and specialized outpatient care, but less inpatient care, compared to low-income earners (14). This pattern is well documented in a wide range of high income countries (2,15,16), although there are a few older studies that only partly confirmed this (17,18).

Analogous to our findings, a longitudinal Norwegian survey study showed that socioeconomic differences in primary care utilization diminished with time between 1986 and 2006 (19). Moreover, high SES groups 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 (19). Similarly, in a Swedish sample from year 1996/1997, Burström found an inverse income gradient related to having the need, but not seeking medical care (8).

It is noteworthy that the association between SES and HCU in part depends on the operationalization of HCU. Lueckmann et al. found that studies that defined HCU in terms of the probability of having visited a physician (“yes or no”), reported larger inequalities in favour of high SES groups, than studies reporting on frequencies (i.e., the number of health care encounters), or more rarely, conditional frequencies (meaning the number of encounters given the individual had visited a physician at least once). Hence Lueckmann et al. suggested that HCU inequalities would best be tackled by removing barriers in initial access to specialized outpatient care for deprived groups (2).

The importance of adjusting for need when analysing socioeconomic differences in HCU is widely acknowledged (20). “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. Thus, the income-related HCU differences detected in the current study are likely due to a combination of differences in prevalence, access to care, and in care seeking behaviour independent of prevalence. The relative importance of each of these mechanisms is however unknown to us. This could hamper comparisons with previous findings of HCU inequalities.

Taken together, studies on HCU inequalities in high income countries mainly report higher HCU of specialized outpatient care among high SES groups, and higher rates of hospital admission among low SES groups. Results of primary care utilization have been more ambiguous, but our findings show that high income earners utilize primary care almost to the same extent as Q5, despite their substantially lower utilization of inpatient care and odds of dying. Although more updated analyses of trends are rare, the available evidence suggests that our findings of increasing use of outpatient care among high SES is not unique for Sweden.

Trends in mortality inequalities

Overall, previous studies of income-based inequalities in mortality rates have reported trends comparable to our findings. Kondo et al. measured the income (Q1 vs Q5) ratio in mortality rates of the Swedish working age population between 1994 and 2004 and reported a yearly widening of mortality rates of 11% (5), and a more recent study report similarly (21). In Norway, Kinge et al. (22) observed an increasing income-related gap in life expectancy from 2005 to 2015 among individuals aged 60 to 69 years. Comparing four Scandinavian countries, Brønnum-Hansenfi concluded that the “Nordic countries displays the same overall pattern of increasing gap between income quartiles in mortality [between 1997 and 2017]” (23).

In this study, we observed that the largest inequalities in mortality were for chronic respiratory disease, followed by diabetes and cardiovascular disease. No income-related differences in neurological mortality rates were observed, in line with previous findings of larger health inequalities for more ‘‘preventable” diseases (1). If using the population attributable fractions (i.e., the percentage of disease burden that are explained by avoidable risk factors) provided by GBD (12) as a proxy for “preventability”, we observe a clear (albeit not perfect) relationship to inequality magnitudes: in Sweden in 2017, the population attributable fraction in mortality were 99.9% for diabetes, 81% for cardiovascular disease, 66% for chronic respiratory disease, 40% for neoplasm and 23% for neurological disorders.

Relative inequalities in mortality tend to increase as the overall mortality rates in the population declines (1). This is also what we found. Between 2004 -2017, all-disease mortality decreased from 1285 per 100 000 to 1143 per 100 000 in the general population. Although both Q1 and Q5 displayed decrease in all-disease mortality rates, the relative inequality increased from OR=1.50 to OR=1.74. However, increased relative inequalities are probably due to a combination of many factors, including social drift, socioeconomic differences in lifestyle, exposures to risk factors, and HCU.

HCU and possible explanations for increased inequality in mortality

From a public health perspective, understanding the causal factors driving health inequalities is important since these provide potential targets for interventions. However, explaining the origin of the observed increase in mortality inequalities is challenging, and differences in HCU likely only play a partial role.

Income-related differences in utilization of primary- and outpatient care could be due to differences in both prevalence and care seeking behaviour, and here we cannot confidently disentangle the two. In contrast, income-related differences in inpatient care are likely driven almost exclusively by differences in prevalence of the conditions requiring hospital admission, a claim that 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 possibly be due to many reasons. For instance, if low SES individuals would be more prone to more severe conditions (e.g., 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 proper control for group differences in medical need.

Health care availability

Although financial barriers to access to care have been largely removed in modern welfare states, the so-called inverse care law suggests that “availability of good medical care tends to vary inversely with the need of the population served” ([1] p 459, Hart 1971). An increasing number of private health providers and the free right to establish care centres 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). In addition, availability is further skewed in favour of high SES HCU due to the increase in private insurances and out of pocket health care spending. However, since our study is based on records of publicly financed health care 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.

The fundamental cause theory

The fundamental cause theory states that “SES embodies an array of resources, such as money, knowledge, prestige, power, and beneficial social connections that protect health no matter what mechanisms are relevant at any given time” (25 p28). HCU is one important mechanism by which people can protect and promote their health. It is well documented that high SES is associated with higher adherence to a range of preventive health programs, e.g. cervical cancer screenings (26) and vaccinations (27). Taken together, the current study adds to the growing body of evidence of an inverse care law (28), where the healthy and wealthy utilize primary and special care more than the low-income earners, adjusting for differences in need (either inferred from self-reports, or as in this study, -indicated by mortality).

Strengths and limitations

To our knowledge, this is the first study to present time trajectories of inequalities both in mortality rates and health care utilization, covering the Swedish adult population for more than a decade. This was possible due to a unique, comprehensive database of linked microdata on socio demographics, cause of death, and health care utilization in specialized care, as well as on primary care in a majority of the 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 enables unbiased definitions of income groups. Finally, the continuously recorded HCU register data enabled detection of time trends with high precision.

This study presents a bird’s-eye view of inequalities in HCU and morality over time at the expense of a more fine-grained description of patterns of group differences in HCU, which would be valuable when attempting to intervene the mechanisms causing avoidable health inequalities. 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 (i.e., yearly health care prevalence) tells little about inequalities in the treatments provided.

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 could 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 data coverage has been consistently high (29).

Future studies

Further studies should investigate socioeconomic differences in HCU patterns with higher resolution in terms of disease categories, treatments, and patient flows (30). 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 disease, and chronic respiratory disease 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.