Stroke unit demand in Norway – present and future estimates

Timely admission to an SU reduces death and disability and increases the number of patients discharged home [1]. It is therefore imperative that health authorities adequately plan for future SU demand.

The delineation between transient ischemic attack (TIA) and ischemic stroke (IS) may have shifted over time [2], with increasingly sensitive imaging techniques [3]. In Norway, patients with TIA are admitted without delay to an SU for rapid work-up and start-up of secondary prevention [4]. Hence, the hospitalisation of TIA patients, as opposed to out-patient treatment, increase the need of SU beds. Further, patients admitted to the SU for suspected stroke, but without confirmed stroke or TIA at discharge—so-called ‘stroke mimics’ (SM)—might represent up to 50% or more of stroke-like presentations and must also be accounted for in SU capacity planning [5‐7].

Although declining stroke incidence rates have been observed in high-income countries over the last 30 years, the trend in absolute numbers is offset by an aging population [8‐10], and more liberal admission guidelines [3, 11]. Recent estimates of incidence rates in Norway have found a decline in age- and sex-adjusted incidence for IS and non-significant trends for intracerebral haemorrhage (ICH) and TIA during the years 2010–2015 [12, 13]. In addition, there appears to be a general reduction in the average length of stay (LOS) for stroke patients [11]. On the other hand, similar reductions in admission rates for SM are not found [7, 14, 15].

Previously, our research group has published time trend estimates for the incidence of IS, ICH, and TIA, estimated LOS for stroke patients on SUs, and incidence of SU admissions for SMs. The aim of the present study is to integrate our previous findings with demographic projections in order to estimate the demand for SU beds in Norway for the years 2020–2040. We thereby account for trends in incidence rates for stroke and TIA, demographic projections, accounting for the effects of LOS and utilisation of SU beds by patients with SMs [7, 12, 13].

General summary statistics of the input data sets are provided in Table 1. Note that all estimates for stroke (including TIA) were based on a data set covering all of Norway between 2010 and 2015. The estimates regarding stroke mimics were based on single-hospital data as outlined above.

The regression coefficients for the incidence and the LOS models are given in Table 2.

As outlined in Methods, extrapolation of temporal trends for LOS for stroke patients was not used in the combined analysis, and the year predictor was frozen at the last year of the data set (2015). For 2020 this model predicts a national average LOS for the different diagnoses, as TIA: 2.4, IS: 6.2, ICH: 7.3, and SM: 2.5. Since a temporal trend is not used, there are only slight changes to these estimates for the different population scenarios, due to shifts in age and sex distribution.

The calculated demand for 2020 corresponds well with the current capacity, with higher SU bed utilisation in densely populated areas, which is expected due to lower random fluctuations in demand. The baseline scenario shows that the predicted reduction in stroke incidence more than compensates for the population growth and ageing. The net reduction in strokes is replaced by a predicted increase in SMs, resulting in a slightly higher demand for SU beds in 2040 compared to 2020 (316 vs. 302).

This study indicates that the future demand for SU beds is uncertain and extremely sensitive to whether the downward trend in stroke incidence continues. There is also uncertainty in future trends of SM incidence. Our predictions should be considered conservative, since we assume that stroke continues to decline, at the same time as the incidence of mimics does not increase.

The estimated occupancies for 2020 fit the reported capacity reasonably well. Despite random fluctuations, an average demand of 84% of the capacity would imply that most admissions can be handled by a local SU. A level of 92% for the region South-east may be problematic, but this region has larger units, which are less vulnerable to random fluctuation: a low variance in admissions makes it possible with higher average occupancy. In the Oslo area (within region South-east) there are several SUs, which – in theory – adds flexibility in SU utilisation. For the North region, the estimated 2020 average occupancy was substantially lower (57%), which is reasonable by the contrapositive argument: this region has low population density, covers a large geographical area, and smaller SU units, which necessitates a relatively lower average occupancy to accommodate variance in the number of admissions. We note that The Norwegian Stroke Organisation performed their own survey of SU capacity in 2020 that reported 362 beds [22], which differs slightly from our counts.

The analysis of the reference scenario shows a relatively stable demand until 2040, nationally and within regions, due to four interacting trend-assumptions:

The impact of deviations from assumption (1) is illustrated by Fig. 3, and the analysis shows a 50% increase in the demand compared to the reference scenario if the stroke incidence is frozen at the 2020 level [13]. discusses this issue, and we agree that there is a fundamental uncertainty regarding the duration of the favorable cardiovascular trend in the population. From a modelling point of view, however, it seems likely that a possible change in the trend will happen gradually, and not take full effect until late in the period.

Regarding assumption (2), the physiological mechanisms of SM are largely independent of cardiovascular risk factors, and there is no evidence of a falling incidence trend. On the contrary, Barra et al. [7] show an increasing trend in incidence of hospitalisations for SMs between 2008 and 2017, but conclude that the trend appears to flatten out near the end of that period; an assumption we have incorporated in our models. If SM incidence continues to rise, however, a substantial increase in demand would result. The findings from this study are somewhat higher than reported in another Norwegian study, which reported a share of SMs of 38% in 2012 [5, 7], and is substantially higher than found in many studies from other countries [14, 15, 23, 24]. None of these report on time-series, however, and we are not aware of any recent studies that contradict the strong increasing trend of SU admittance for SMs between 2008 and 2017. We remark here that the incidence of admittance to an SU with SMs in our forecasts is based on a study from only one (large) Norwegian hospital and may be higher there than for other populations. There is, however, no evidence that Haukeland University Hospital’s admittance policies differ in a systematic way from other Norwegian hospitals. (Indeed, at Akershus University Hospital, we were able to check during the review process to confirm that in 2020 the share of SMs surpassed 50%).

The large share of hospitalisations due to SMs – even if overestimated here – highlights the importance of pre-hospital or early screening and rapid work up. We hope advances are made in the diagnostic pathway for suspected strokes, resulting in better specificity, so that fewer SMs are admitted to the SU in the future. Given such advances, our estimates will overestimate the demand for SU beds.

Assumption (3) is in fact contrary to empirical observations: the NPR stroke dataset shows a downward trend in SU LOS. However, if we extend this trend through to 2040, the average LOS for ICH patients would fall to 2.5 days, which is very unreasonable and medically indefensible [1]. There are several reasons why we choose to keep the LOS stable in the analysis: Importantly, the current average LOS is consistent with the recommended shortest SU LOS for stroke patients from Norwegian national guidelines for standards of care, and further reductions are not advisable [25]. Second, the period 2010–2015 coincided with the implementation of the Norwegian Coordination reform, which included measures designed to reduce (excess) hospital LOS. Specifically, the reform introduced a substantial remuneration from the municipalities to the hospitals if they were unable to receive patients declared ready for discharge. Third, we consider the changes in LOS largely as an endogenous variable, which captures the hospitals’ responses to changes in the demand. We find it most informative to make predictions under the assumption that the level of care is kept constant. Still, there are uncertainties in this area also, and it may be possible to reduce the LOS slightly without compromising the level of care. In addition to a trend of fewer strokes, there is also a trend toward milder strokes [11], which may lead to shorter SU LOS in the future. Also, for SM patients, the SU stays are mostly about diagnostics, and technical progress in medical imaging may reduce their LOS.

Assumption (4) is least problematic, although the demographic trends explain nearly 50% of the increase in demand when stroke incidence is frozen. The SN scenarios of high and low ageing project changes in 2040 occupancy of approximately 5% up or down from the reference scenario, which is substantial, but still moderate compared to the other uncertainties discussed above. The SN scenarios represent internal migration, but the results in Table 3 show no important between-region effects.

There is a specific demographic uncertainty for the catchment area of Akershus University Hospital, however, which has a substantial number of immigrants that are suspected to have an elevated stroke risk [26]. At present, this group is relatively young, but during the forecasted period these individuals will shift into high-risk-for-stroke age-groups. We also note that although evidence from Norway suggest elevated stroke-risk in some immigrant groups, the opposite might be true. In neighboring Denmark, a contemporary study reported lower stroke incidence for immigrants after adjusting for sociodemographic covariates [27]. We have not estimated the magnitude of such effects.

Sensitivity analyses of the predictors included in the component models show that only age and year have an impact on the predictions that is large enough to be important for policy makers. It may be surprising that a predictor like sex, which has a strong effect on incidence, has little impact on the overall demand. The explanation is that an incidence model without sex as a predictor will make unbiased average estimates over the whole baseline population, and if the sex distribution in the population remains relatively stable, the aggregate estimates remain sound. The two main drivers in the predictions are a) the population aging combined with a strong age gradient in incidence and b) the time trend in incidence. It thus stands to reason that age and year affect estimates the most.

The main strength of the present study is the integration of separate state-of-the-art models for incidence and LOS for SU patients in Norway, combined with approved demographic scenarios. The models for stroke and TIA diagnoses are based on nationwide data. For other model inputs we include data from several hospitals. Within the model suite, the SM models are the least general since these are estimated from smaller datasets that do not cover the whole country. Still, and as always, the main limitations of the study are given by fundamental uncertainties in the extrapolation of the historic trends.

A slightly higher average SU bed demand is predicted in 2040 compared to today under the reference scenario. A reduction in numbers of strokes, particularly ischemic stroke, is predicted despite expected population growth and ageing. An increase in stroke mimic admissions replace the net reduction in strokes and TIAs. The main uncertainties to our predictions relate to future incidence trends for both stroke and stroke mimics.

Since the most important uncertainty factors point toward a potential increase, which may be as large as 50%, we would recommend that the health authorities plan for a potential increase in the demand for SU bed capacity. At the very least, we would strongly advise against a reduction in this important capacity before such uncertainties are substantially reduced.