Use of somatic hospital care services by immigrants in Norway, a retrospective population study

Migration numbers are increasing globally (McAuliffe et al., 2020), and also in Norway immigration is increasing; currently 19.9% of the population is either immigrants or born in Norway to immigrated parents (Facts about Immigration, 2026). Across many countries and over time, immigrants have been shown to underuse specialized care services (Jiménez-Rubio and Hernández-Quevedo, 2011; Sarría-Santamera et al., 2016; Stronks et al., 2001). Underuse is defined as the failure to use medical care, and is documented to be common across all types of health-care systems and health services (Glasziou et al., 2017). Explanations of the underuse may be different in different countries, as it may vary by health system type, funding scheme and with different barriers to access (cost, language, resources etc.).

Norway has a tax-funded, universal health-care system that provides universal coverage, largely free at free at the point of use, likely reducing underuse for immigrants compared to systems with higher out-of-pocket costs. However, underuse for immigrants has also been found in Norway, both for primary health-care services (Diaz et al., 2015; Diaz and Kumar, 2014; Sandvik et al., 2012) and for hospital services (Abebe et al., 2017; Elstad, 2016; Hasting, 2018).

Elstad (2016) studied how geographical origin and reason for migration caused variations in migrants use of somatic hospital services. Immigrants from most regions of the world have lower utilization than non-migrant native Norwegians, with the exception of West and South Asia. There were differences in health-care utilization among different migrant groups: work migrants or family migrants have low hospital utilization, while refugees has quite large health-care use (Elstad, 2016). A recent study of Syrian refugees reported significant changes in health-care utilization following resettlement in Norway, including higher use of general practitioner and emergency services and a decline in specialist care (Haj-Younes et al., 2021). Time since migration does also affect utilization; over time, the utilization pattern more resembles native Norwegians. Abebe (et al. 2017) compared mental health-care utilization between ethnic Norwegians and immigrants from 11 countries. Overall, they found a clear underuse; however, use of services varied with country of origin.

Selection effects among those who migrate may contribute to lower health-care utilization, a phenomenon often referred to as the “healthy migrant effect.” This concept suggests that individuals who migrate tend to be healthier than the general population in their country of origin, either because healthier people are more likely to undertake migration or because migration processes favor those in better physical and mental condition. As a result, newly arrived migrants may initially have fewer health-care needs, which can partly explain observed underuse of health services compared to host populations. This is however very difficult to untangle empirically. An alternative hypothesis has been suggested; there might a “salmon-bias effect” (Namer and Razum, 2018; Pablos-Méndez, 1994) where if the migrants become sick, they return to their country of origin where they possibly have informal care (family) or are better able to negotiate and access the formal health-care system. Nevertheless, the underuse may be critical in the long term, as prevalence of NCDs is high (Tran et al., 2011).

Previous Norwegian studies have often looked at specific country or region of origin, or different time periods of migration (Abebe et al., 2017; Elstad, 2016). However, there is a lack of studies comparing immigrants and second-generation immigrants with non-immigrants. The purpose of this paper is to improve modeling health-care services utilization among immigrants using a complete population approach with retrospective registry data. We extend previous models by including factors beyond immigration status such as age, gender, socioeconomic status and geographic location. Additionally, we examine differences while controlling for diagnoses and type of service. Our analysis distinguishes between first- and second-generation immigrants, non-Western origin, time since migration and proportion of refugees. Unlike earlier studies that only assessed whether somatic hospital services were used, our measure of utilization also capture patient case mix. The underlying hypothesis is that, for comparable demographics, actual need for somatic hospital services may be high, but demand and utilization remain low.

The data used in the analyses are linked data at the individual level from Statistics Norway and hospital use data collected from the Norwegian Patient Registry. Approval for project and data collection was made by the Regional committees for Medical and Health research ethics. The data includes all individuals residing in Norway in the years 2016 and/or 2017, n = 10,834,095 person years. We analyze three different immigration groups in Norway based on immigrant categorization of all inhabitants by Statistics Norway: immigrants, second-generation immigrants (Norwegian-born with immigrant parents) and non-immigrants. The latter group consists of those born in Norway (to two Norwegian-born parents), born in Norway with one parent born abroad and born abroad (with Norwegian parents), but the last two subgroup are too marginal in size to be included as a separate category.

In the present data, it was not possible at the individual level to collect country of origin for immigration population, nor year of arrival. Both factors are assumed to influence the utilization (Elstad, 2016). However, we have information about the immigration categorization of each individual, and about the share of immigrant population from non-western countries in each municipality. We used this to construct a variable that is interpretable as the probability of being non-Western. Non-Western immigration was defined as not migrating from Western-Europe, America or Oceania. To compensate for lack of data at the individual level specifically on country of origin, year of arrival and refuge as reason, we have instead constructed municipal level averages of time since migration (measured in years) and also at municipal level the share of immigrants with refuge as reason for immigration. These two latter variables were generated based on public statistics from the data service microdata.no.

Previous studies have relied on crude measures of activity. Hasting (2018) used a dummy for contact with different services based on self-reported data from surveys. Diaz et al. (2015) uses number of consultations, while Elstad (2016) and Abebe et al. (2017) measure use of health care as a dummy for at least one yearly contact. Elstad also measures number of admissions for some specific diagnostic groups. We aim to improve these approaches by using a case mix adjusted measure that will capture both the amount/intensity of treatments/services provided, and the severity of each treatment/patient. Output is thus measured at individual level as yearly use of hospital services calculated as the weighted sum of diagnosis-related groups (DRG) points. DRG weights are used in Norway for financing purposes; thus, all inpatient, day surgery and outpatient treatments were already grouped and assigned a DRG weight ^[1]. DRG weights are derived from standardized cost data and clinical complexity; they provide a robust proxy for case mix, enabling fair comparisons across hospitals and patient populations. This makes DRG weights not only a financing tool but also an analytically sound measure of treatment intensity and patient severity, widely recognized in health services research. One DRG point in 2017 had the value of 42,753 NOK or 5,174 EUR. Activities include all somatic hospital care use: inpatient, day care, outpatient, private rehabilitation and contract specialists. All hospital activity data was collected on the individual level from the Norwegian Patient Registry and linked to explanatory variables.

Previous studies have primarily looked at country/region of origin, time since migration and only to a limited extent used demographic and socioeconomic controls. One of the aims of this study is to assess the impact of individual level variables as well as municipal properties regarding immigration and supply side factors. By controlling for several individual level variables, we can rule out differences between immigrants and non-immigrants that is caused by differences in demographics or socioeconomic differences.

We test for differences in age/gender, supply-side effects (hospital trust dummies generated from hospital catchment area aggregated from each individuals home municipality, travelling time to nearest hospital generated by Statistics Norway), socioeconomic variables (children with single parent, children in child welfare services, education (measured linearly as a variable from 0 to 5 with each step representing the completion of a level of education), dummy for low income, out of work, sick leave, disability pension or work assessment allowance all collected from Statistics Norway) and dummy for mortality within same year or the next (source: Statistics Norway). Mortality is split in four distinct age groups (0–19, 20–39, 40–79, 80+), and education (25–66) and out of work and welfare benefits (18–66) are only included for relevant age groups.

All explanatory variables on the individual level were collected from databases at Statistics Norway and linked to hospital utilization data. Furthermore, there may also be supply-side effects that influence the use of services, and we thus include hospital trust dummies and travelling time to nearest hospital to capture any association of settlement pattern and utilization.

Probability of being non-Western immigrant was estimated based on population statistics for each municipality based on data from Statistics Norway. Mean number of years since arrival and probability of being refugee was estimated at municipal level with data from public Norwegian data service microdata.no.

We first present descriptive statistics for the use of specialized hospital services between immigrants and non-immigrants. Second, we use ordinary least squares regression to estimate association between immigration status and use of services. This is done by comparing six different models ^[2]; see Table 1 below. Models 1A, 1B, 1C and 1D are tested to compare operationalizations of immigration, while comparison of models 2 and 3 are presented to test differences between immigrants and non-immigrants. All models are tested both with linear probability models (OLS) and in a two-part model. Model 3 is also tested for probability of AMI.

The purpose of testing different models is to examine how the effect of immigration does play out. Comparing models with and without immigration information, and with and without controls will allow us to measure the importance of immigration and controls respectively. Models 1A, 1B and 1C is equal to a bivariate test of immigrant status, in other word it only shows the difference between immigrants, second-generation immigrants and non-immigrants, before any controls. The difference of models 1A and 1B will inform us of whether the non-western immigrants are different than immigrants overall. Comparing models 1A, 1B and 1C will assist in deciding which operationalization of immigration status that is most important. There might further be differences between different immigration groups, so we have also included a test of Model 1D that introduces also share of refuge immigrants and years since immigration.

If immigration does not have an impact, we expect the model without immigration status (Model 2) to be as statistically good (measured as explained variation, R²) as Model 3. Model 3 will thus be an improvement (interpreted by significant increase of adjusted R-squared) to Model 2 only if there are large differences between immigrants and non-immigrants.

All models are for all-purpose somatic hospital use and include all possible treatments and diagnoses. We controlled for age/gender, supply-side effects (hospital trust dummies, travelling time), socioeconomic variables and mortality. Results of individual levels associations are shown in the paper, but for brevity and presumed little interest outside of the trust themselves, supply-side associations were not reported. Age-effects are also omitted from results, but relevant age differences can be seen in Figures 1 and 2. Regressions control for clustering on the hospital trust level.

The analysis described in these models are group comparisons of the average hospital utilization at the individual level. However, hospital utilization can be decomposed as the product of the number of patients per inhabitant and the utilization per patient. In plain words: the number of patients per inhabitant is a measure of the probability of being treated, and the intensity of treatment among those that are treated. This is estimated as a two-part model (Belotti et al., 2015).

For some diseases, we expect that all patients with the relevant diagnosis must use a hospital, for instance acute myocardial infarction, so the utilization rates should be comparable between different immigration groups. However, it is suggested that the incidence is higher for some immigration cohorts or regions of birth, due to underlying causes such as general cardiovascular risk and diabetes (Abdelnoor et al., 2012; Hedlund et al., 2007; Rabanal et al., 2017), and as such we would expect higher utilization when looking at AMI in hospital. We therefore test the probability of being diagnosed with AMI (acute myocardial infarction (ICD-10 diagnosis I21) or subsequent myocardial infarction (ICD-10 diagnosis I22)) within a year, as a logistic regression.

Table 2 shows that 43.6% of all inhabitants in the population visited the hospital in the years 2016–2017. However, among the immigrants, only 27.4% visited the hospital. If we split the activity according to acute/elective status, we see that the differences are largest when it comes to elective treatments (of which most is outpatient treatments), but surprisingly the difference is also very much present for acute care. These differences may possibly be explained by differences in age, as the table does not control for differences in age between the categories. Table 3 below presents the average annual amount of hospital services related to each individual, measured as DRG-points. On average, each inhabitant receives hospital services equal to 0.300 DRG points each year. For immigrants, the number is half of that at 0.151.

Figure 1 below shows the average somatic hospital care utilization by age as measured by average annual volume of DRG-points per capita. It shows for both immigrants and non-immigrants low utilization from early age until around age 50, after which there is an increase until a peak in utilization around ages between 80 and 90. The pattern is very similar for both immigrants and non-immigrants, aside from newborns, who by definition cannot be immigrants. Figure 2 below plots rates related to average per age. The figure shows that non-immigrants are above average for all age groups, especially between ages 25 and 55. Immigrants use below average for all age groups, but with higher age, the groups seem to converge slightly. However, the group of second-generation immigrants is much smaller, especially aged 40 and above; thus there is much more variation for that category. 75% of the second-generation population is aged 0–17.

We have used OLS regressions to test the specified models, and the results with regression coefficients and standard error are shown in Table 4. The immigrant dummy and probability of being non-Western immigrant correlate extremely high on the individual level, resulting in very similar results in models 1A and 1B. Testing differences of these models with controls gives marginal preference to operationalizing immigrants only with a dummy-variable instead of probability; thus, Model 3 is only presented in one version. Testing both variables simultaneously also favor the dummy over the interaction. Second-generation dummy was only significantly different from 0 in Model 1C, and therefore not included further in Model 3. Including age groups as controls in regression did explain all the differences in the utilization of the second-generation immigrants.

Model 1A show that there is a difference in somatic hospital utilization between immigrants and non-immigrants; however, this is before any controls. Including controls and age-group significantly improves the model, as seen from R-squared. However, even in Model 3 there is a significant difference of somatic hospital utilization between immigrants and non-immigrants, even after controlling for age/sex and socioeconomic variables. The difference after control is 0.0763 DRG points per year, which is smaller than the absolute difference from Table 2 above. Note that the change in adjusted R-squared from Model 2 to Model 3 is very small, indicating that not a lot of the unexplained variation in the overall health-care use is due to immigration-status. The coefficients that change the most from Model 2 to Model 3 are education and income, indicating that immigration captures some of the education and income association related to use of health-care services.

Table 5 below shows the results from a two-part model that first estimates the probability of using somatic hospital services while the second model is a regression on amount of utilization, conditional on being a patient. Shown here are only the Z-statistics from each step, while complete results of the full Model 3 is available as online supplementary file. In all models, the immigrant variables are far more important explaining probability of being a patient than explaining the total volume of services used at the individual level for those that are patients. The combined marginal effect of the two-part model is −0.0934. Indicating that the average marginal effect of being an immigrant, after considering both the probability of being a patient, and then in the second stage the health-care use per patient, is 31% lower than for the average population, holding every other variable constant.

The results from models 1A, 1B, 1C, 1D and 3, both with OLS-regression and two-step model, indicate that there is a distinct underuse of hospital services. All these tested models have been for all-purpose use of hospital, including all possible diagnoses and treatments. One particular condition worth looking into is AMI. We test Model 3 again, with being treated with ICD-10 diagnosis I21 or I22 (as main or secondary diagnosis) as the dependent variable in a logistic regression. Results are provided in Table 6 below. Again, the results show that the immigrant group clearly stand out with an odds-ratio of 0.734 (95% confidence interval 0.665–0.809). This analysis thus shows, that even after controlling for age and other relevant variables, the immigrants clearly show an underuse of health-care utilization.

Differences between Model 1C and Model 1D reveal that there is some variation in the immigrant population that is correlated to the other immigrant variables. However, none of the other variables are highly significant, but we see that both share of refugees and time since immigration tend to increase the probability of being a patient as we see in the first step of the model. However, in the second step we see that average number of years (by municipality) since immigration does capture a large part of the effect of the immigration status dummy.

Earlier studies have shown that immigrants underuse health-care services. The present study confirms past results. Table 2 above showed that immigrants’ use of somatic hospital services amounted to 0.175 DRG points less use per year than non-immigrants. After controlling for differences in supply side factors and individual level variables such as age/sex and socioeconomic factors, the difference between the groups was only 0.076 DRG points per year. This means that even though immigrants on average only use half the level of average Norwegians, a very large part of this is due to age composition and socioeconomic factors. Nevertheless, there is still a gap in the utilization between immigrants and non-immigrants. Our analysis reveal that differences between immigrants and non-immigrants are larger for probability of being a patient, than for amount of hospital utilization for those that are patients.

The underuse has potentially severe public health effects, as it may increase the risk of comorbidities and mortality in the long term. Our analysis show that even for AMI, which should be provided with equal access in a free and universal health system, we find evidence of underuse. This is despite immigrants are often having increased risk for cardiovascular diseases and chronic conditions such as diabetes.

The general, and specific underuse, reduces actual equality and may increase health inequities. In times of higher immigration, the availability and use of specialized care is of high importance, for instance in times such as the immigration following the 2022 Russian invasion of Ukraine. Health policies should be directed to increasing short term utilization of care for immigrants to improve health and reduce costs in the long term. Also, in times of medical emergencies, such as COVID-19, it is important that all have equal access to health care.

The strength and the primary contribution of this study is to extend previous research with an improved measure of hospital use, and by including important controls both at municipal and individual level. Further, the study looks both at average effect on inhabitant level as well as on patient level. This indicates that the use of health care is not very different when looking at those that are patients, so the underuse is a matter of actually using hospital services. The results in this paper of underuse are comparable to earlier studies, but it must be noted that only somatic hospital services are included in this paper; and different mechanisms may cause different utilizations among other health-care services.

It was not possible in this study to randomize immigration status or hospital treatment for patients. So, we have relied on control variables to capture any group difference that we are interested in, and while immigration status is set from birth one cannot draw causal conclusions that it is immigration status itself that is causing underuse. This study has not been able to use country of origin in the analysis. This has previously been shown to be an important factor determining individual level variations. We have used information about country of origin on municipal level to construct the probability of being non-western on the individual level. This however did not yield vastly different results than immigration status itself. Furthermore, including municipal averages of refuge immigration and time since immigration (in Model 1D) did not yield significant results. However, in the two-part model, we see that conditioning the patient status clearly caused differences between immigrants as we saw that time since immigration reduced the effect size of the immigration dummy in models 1A and 1C.

As a registry study on the whole population, we have demonstrated that immigrants use less hospital services. A large part of this underuse is related to measurable issues, but not all. We can tentatively conclude that the remaining causes of underuse are likely to be complex and related to issues outside the scope of health registry data. We can only speculate these issues may be cultural issues, language barriers, perceived cost, accessibility and trust. These issues are qualitatively important for the use of health care, but registry studies are not able to collect data on this nor perfectly operationalize the complexity of the matter into a simple statistical model.

Data collection was funded by the Norwegian Ministry of Health and Care for the purpose of using in Official Norwegian Report 2019:24. The funding body had no influence of the analysis, interpretation of results and in writing the manuscript. All other activities regarding analysis and writing the manuscript had no external funding.

The authors are grateful for valuable feedback from Kirsti Wahlberg and other participants in the Norwegian Health Services Research conference in 2022. Data for the study was provided by Statistics Norway and the Norwegian Patient Register. The interpretation and reporting of these data are the sole responsibility of the authors, and no endorsement by the data providers is intended nor should be inferred.

The supplementary material for this article can be found online.