Summary
Given the rapid increase in health care spending in Switzerland in the last few years, it is pivotal to better understand the factors associated with this increase. This study was the first to decompose inpatient care costs by disease in Switzerland, using detailed cost data for the canton of Zurich, between 2013 and 2017. It assessed the impact of six fundamental factors on the change of disease costs over time. The most important factor was the increase in the costs per day, which was partly offset by a reduction in the average length of stay. Population growth and age and sex structure were related to about half of the increase. A minor part was associated with an increase in the number of people treated relative to the population. This study contributed to the existing literature in three important ways. First, it included one additional factor in the decomposition specification, namely length of stay. Second, it further decomposed the costs per day factor into five cost components. Third, it reported results not only for the average, but also for the full cost distribution.
Discussion
The data we used does not reflect spending at prices charged to payers but production costs incurred by the hospitals. It is therefore a more objective measure of the resources needed to provide treatment for each disease than spending. However, the cost changes observed in this study do not necessarily correspond to the actual health care spending changes by disease, since hospitals might either realize profits or the costs by disease might not perfectly correspond to the spending for the diseases.
Although we used data at the individual level, an aggregation of the cost data at the disease and age/sex group level was appropriate for the question we aimed to answer. Decomposition methods usually used with individual level data, i.e. Oaxaca and Blinder type decompositions [
33,
34], are of limited use here since they aim at explaining the mean outcome (in our case mean costs per day or case) and thus ignore the factors leading to the aggregate change such as population growth or utilization. One strength of our study lies in its ability to separate the propensity to be treated from the utilization as well as the intensity and price of treatment (captured by length of stay and costs per day).
The observed positive association between population structure and costs is due to the fact that patients in higher age groups exhibit more stays and higher costs per case. It has been shown for Switzerland that costs per patient and hospital days per person in the inpatient sector increase significantly beyond the age of 60 [
36]. Since we controlled for other factors, the cost increase that is linked to a change in population structure is small.
Several reasons may explain the increase in the share of people being treated, which was associated with a cost increase of 1.9%. The first one might be supplier-induced demand. A recent study for Germany has shown that the steady increase in hospital admissions can be explained only partially by demand-side factors such as time-to-death and morbidity; about 80% is accounted for by supply-side effects resulting from the DRG reimbursement system [
37]. The second reason is linked to technological progress: new diagnostic and treatment options allow more patients to be successfully treated, which possibly leads to better health outcomes [
38]. The third reason is a more technical one. We measured the treated prevalence as the share of unique patients within each age/sex group. Since some patients live outside of the canton of Zurich, but this information was not accessible to us, the factor is also influenced by inter-cantonal patient migration and might thus be slightly biased. The share of the net patient migration has increased from 9.5 to 10.7% of the stays between 2013 and 2017 [
39]. We can therefore not rule out that for some diseases, part of the effect of the treated factors is driven by patients coming from other cantons. A fourth explanation is the underlying disease prevalence: if more people suffer from a disease, the number of treated patients is likely to increase too. However, our study was not able to capture the health-related needs of the patients and thus ignored the fact that some diseases become more or less prevalent. The studies by Dieleman et al. [
5] and Zhai et al. [
23] included age/sex-specific prevalence estimates for each disease from the GBD project in their decompositions and reported utilization per prevalent case instead of per treated patient. Even though national prevalence estimates are also available for Switzerland from the GBD project, we did not include these in our analysis of sub-national data. Since we only focus on one health care sector, our two factors treated and utilization provide policy makers more easily interpretable results than utilization per (clinically) prevalent case.
6 For some communicable diseases and injuries, we observed a reduction in the share of treated patients. This might be due to a shift from inpatient to outpatient treatments. Despite this development, the number of treated people has further increased for most diseases. This suggests that surgeries in the inpatient setting do not decrease to the same extent as the outpatient treatments increase [
42].
The observed decrease in the length of stay and the increase in the utilization factor might both be linked to the implementation of the DRG reimbursement system. Hospitals have been incentivized to discharge patients early and to produce more stays (e.g. [
43] for the United States). Previous research for Switzerland provided evidence that the system change led to an increase in hospital readmissions [
44]. Readmissions within 18 days of treatment are, however, already included in our records and do not constitute an additional stay, i.e. the utilization factor as defined here captures the increase in stays per patient on top of that. The differentiation between the number of stays per patient and the length of stay thus proves to be relevant, as the effects go in opposite directions. Utilization defined as number of days per patient like in [
5] captures both effects at the same time. Furthermore, our study confirms previous evidence for Switzerland, as it found a strong association between the reduced length of stay and costs [
45]. The heterogeneity of the length of stay factor association across diseases might indicate different levels of efficiency margins for different diseases.
The reduction in length of stay was offset by an even higher increase in costs per day.
7 This result is in line with the results by Dieleman et al. for the US, who found an increase in costs per hospital day and a reduction in bed-days per prevalent case [
5]. The reason might be a compression of (more intensive) treatment into a shorter period of time. The relative size of the two competing effects of length of stay (negative) and costs per day (positive) in our study (7.3%/12.1% = 60%) is the same as the ratio of bed-days per patient and costs per day in [
5]. This accordance is surprising, even though they defined the utilization term in a slightly different way than we did. The costs per day factor likely captures to some extent higher input prices, e.g. due to technological progress in drugs and higher wages, and an increase in treatment intensity, e.g. due to technological progress or supplier-induced demand. This is, however, only likely in the case that the additional treatment leads to higher reimbursement for the hospital. A significant part of the increase in costs per day is probably due to an increase in input prices for medical staff. Our costs per day decomposition showed that 8.1 of the 12.1%-points cost increase association was due to medical and physician costs, two categories mainly consisting of labor expenses. The Swiss statistics on hospitals (Krankenhausstatistik) by the Federal Statistical Office reports that the total wage bill in general hospitals in the canton of Zurich increased by 13% between 2013 and 2017 [
2]. The highest increase was observed for physicians’ salaries. Their share of the total wages increased from 24.0 to 25.7%, which is equal to a rise of the wage sum for physicians by 21% in the 4 years period. A similar focus on the sources of increasing costs per inpatient case was taken previously by a study in the United States which showed that supplies and devices, room and board, intensive care and operating rooms as well as drugs contributed to more than 75% of the increase in costs per case between 2001 and 2006 [
46].
Accounting for length of stay as well as costs per day is important. However, policy makers are more likely to be interested in their joint effect, i.e. costs per case. Despite a modest overall increase in average costs per case, average costs decreased for some diseases. This is surprising, as input costs, and particularly wages of hospital personnel, increased substantially over the study period.
Our analysis of distributional aspects revealed that for most diseases, the 10% most expensive cases showed the highest contribution to the factor associations. However, it also showed that the direction of the effect might differ across the distributional sections and by disease. Only recently, the literature aiming at identifying cost drivers has started to look at other measures than just the average [
47,
48]. DeMeijer et al. [
48] used Dutch data and showed that the impacts of various explaining variables measured at the patient level differed across the spending distribution. For hospital spending, growth was found to be largest in the middle of the distribution. In contrast to their study, we did not use an individual-level decomposition, but our results provide insights at the disease level.
Policy implications and future research
A better understanding of the epidemiological, technological and demographic trends on health care costs may be particularly useful for a sound definition of global spending budgets currently discussed in Switzerland. Effective cost containment policies require reliable estimates of the key factors influencing costs at a very granular level such as specific diseases. The majority of the cost increase was due to only three groups that caused an increasing health and cost burden and should be monitored carefully. Furthermore, it is important to know the consequences of a change in the hospital payment system for health care utilization. Our study suggests that the number of cases per treated (in)patient within each disease group is increasing, which would somehow conflict with the goal to increase efficiency in hospitals.
Future research should include other health care sectors to investigate whether there has been a shift to outpatient treatments and if yes to what extent and for which diseases. It is further important to research the reasons for the factor associations we found, e.g. the increased number of treated inpatients relative to the population and the increased number of stays per (in)patient over time. In addition, it would be interesting to assess whether the additional health care use is due to increased care needs or due to hospital-induced demand. Accounting for the cost impact of comorbidities is a very important and demanding task for future research. Some regression-based approaches have recently been proposed by the literature to adjust for concurrent diseases at the patient or the encounter level [
49,
50]. Our framework could also be used to assess the impact of lower length of stay on treatment quality at the level of specific diseases. The reduction in length of stay is only desirable if there are no “bloody exits” associated with worse health outcomes. Finally, expanding the analysis to more than just average effect has shown how important it is to have a closer look at which cases contributed most to the cost increase. It is therefore vital to include distributional aspects into future cost-of-illness studies and other research looking at drivers of health care spending.