Background
In-hospital cardiac arrest (IHCA) is a major adverse event in hospitalized patients. Previous studies have documented the incidence of IHCA between 1 and 6 events per 1000 hospital admissions [
1‐
3], and both short- and long-term survival after IHCA is poor. A meta-analysis yielded a 1-year survival rate of 13.4% but showed substantial heterogeneity between studied cohorts [
4]. A US study also showed heterogeneity in incidence and outcomes after IHCA between centres [
5]. This observed heterogeneity may be attributed in part to differences in case-mix or to differences in improvable facets of care (quality of care) at the provider and hospital level. In other fields, such as stroke, targeted quality improvement measures have led to improved outcomes [
6]. However, it is not known whether outcomes after IHCA can be improved through a similar focus on quality improvement.
Quality of care can be assessed through structures and processes of care, as well as through patient outcomes [
7]. Structure of care indicators pertain to hospital-level factors, which apply to all patients. Notable examples of hospital-level structure of care factors relevant to IHCA are availability of advanced life-support (ALS) trained personnel, cardiopulmonary resuscitation (CPR) training frequency of personnel, assigned roles of specialists in the cardiopulmonary resuscitation team, and the availability of an intensive care physician. These particular structural indicators have been shown to vary substantially between Dutch hospitals [
8]. Secondly, there are process of care indicators, which can vary on the patient-level and can easily be acted upon. A potentially relevant process of care indicator for IHCA is the time until ALS is started, at which point the ALS practitioners can provide additional life-sustaining measures: e.g. endotracheal intubation, administration of epinephrine, and potentially initiate extracorporeal life support [
9]. A shorter time between IHCA and these interventions could improve short- and long-term outcomes. The registration of a rapid response team warning score (RRS) could be an additional relevant process indicator: these scores (the early warning score, EWS; the modified early warning score, the MEWS; the national early warning score, NEWS) may help in identifying patients at-risk for cardiac arrest, in which case extra precautions could be taken [
10]. Finally, outcome metrics such as mortality and cerebral performance category (CPC) score at discharge are relevant patient-level quality indicators [
11].
This study aims to assess variation in outcomes between hospitals and to explain heterogeneity in these outcomes by differences in case-mix or by differences in quality of care stemming from structural and procedural metrics.
Discussion
In this study, we first assessed whether there is substantial variation in outcomes between hospitals in the Netherlands after IHCA. We found small-to-moderate variation in mortality and functional outcomes. Between-centre differences in mortality rates could largely be explained by case-mix, but between-centre differences in CPC scores at discharge persisted after adjustment for case-mix.
To potentially improve functional outcomes, we investigated the reliability and relevance (in terms of association with outcomes) of processes and structure of care indicators. The reliability of the two process indicators was high, but their relevance could not be established with current data. We could not establish this relevance either due to the design of our study, or because our data did not provide evidence against the null hypothesis. In general, quality of care does not often significantly explain variation in outcomes, because treatment effects are generally modest, and not all processes of care apply to all patients [
23,
24]. However, our data did suggest a positive effect of a structure of care indicator in the analysis where we assume differences between hospitals to be random: offering multiple CPR trainings per year to personnel was associated with better functional outcomes of survivors at discharge.
The group of included centres consisted of teaching hospitals with more extensive facilities than the typical Dutch hospital. Within this group of centres, there was little variation in both mortality and CPC score. This finding is in contrast with a US study, which described substantial variation in outcome between centres [
5]. One explanation is that this study included a much broader range of hospital levels, while our sample mainly includes teaching hospitals.
Nevertheless, the finding that the observed variation in mortality is explained by differences in case-mix can be seen as a strong indication for a cohesive hospital system with uniform adherence to guidelines carried out by highly trained personnel. We should consider the possibility that participating hospitals might have performed better, or reported selectively, simply because they were observed within this study (the Hawthorne effect) [
25]. Nevertheless, we hypothesize that the homogeneity in quality of care is an important explanation why survival in our population is higher than described in the literature [
4,
26].
On the contrary, the variation in CPC score could not be entirely explained by differences in case-mix. It can be argued that the explained variance of our models was not high enough. Although the Nagelkerke
R2 is lower than other prognostic studies in cardiac arrest [
27,
28], it is known that
R2 measures for categorical outcomes are much lower than those of continuous outcomes [
29]. Also, because our aim was to explain (and not to predict) outcomes [
30], we think this finding has important implications for cardiac arrest care in the Netherlands: improving care might not improve survival rates, but it might improve functional outcomes. We recommend that other hospital systems identify local processes and structures of care indicators and enact appropriate improvements that could lead to better patient outcomes.
Although the reliability for the investigated processes and structure of care indicators was high, only the relevance for one of the structure of care indicators could be confirmed with the current study. We will here discuss the investigated processes and structure of care indicators, and the implication of our evaluation.
First, we found an indication that CPR training frequency of twice per year might improve functional outcomes. However, patients in centres who trained twice per year were younger and had slightly better pre-existing neurological status, coincidentally. Either training twice per year somehow results in a healthier population being resuscitated, or the difference is based on chance. We hypothesize that hospitals that train twice have more awareness of in-hospital cardiac arrest than hospitals that train less. If there is more awareness, we believe cardiac arrests might be noticed earlier, and possible some unnecessary arrests would be prevented. However, we think that the prevented arrests are more likely those in patients with more physiological reserve, since there is more time to prevent cardiac arrest. Therefore, intervening earlier in the process due to higher awareness should result in the remaining patients being older and with worse pre-arrest functional status. Because of this, we think that it is more likely that this difference in case-mix is based on chance, especially because it entails a post-hoc analysis. As only 45% of the Dutch hospitals are described to offer CPR training twice per year [
8], increasing adherence to this structure of care indicator could result in improvements in outcome: decreasing intervals between CPR training increases CPR quality in terms of compression depth and rate, and complete chest recoils [
31,
32].
Second, our results did not suggest that 24/7 availability of intensivists improves outcomes, in spite of evidence to the contrary [
33‐
35]. We believe that the 24/7 availability of intensivists could indeed improve neurological outcomes, but that our study lacks sufficient power to detect an effect due to the small number of included centres with an intensivist 24/7. With 24/7 intensivist coverage, similar mortality between weekdays and weekends has been reported [
36,
37]. It might be hypothesized that we would have found a significant effect if we would have included more hospitals without 24/7 availability of intensivists.
Third, the absolute variation in time to ALS was limited, but consistent and reliable: the rankability was more than the 70% threshold that is suggested as reasonable for quality indicator to be valid [
19]. The effect on outcome, however, could not be established: the assumed mechanism through which a lower time to ALS improves functional outcome is by enabling early treatment of reversible causes [
38]. We recommend that future studies register whether a reversible cause was present, and whether this was effectively resolved, to better establish the relevance of this process indicator.
Fourth, the reporting of an RRS varied substantially between hospitals and was again a reliable process indicator. The presumed effect of RRS on outcomes, however, primarily impacts outcomes through preventing cardiac arrest [
10]. Therefore, a study which only includes patients with cardiac arrest cannot evaluate the relevance of this indicator. Nevertheless, as other studies have showed evidence for effective prevention of cardiac arrest [
10,
39], our results mainly indicate that the implementation of these scores in clinical practice could be more stringent.
This study is limited because we study a selected group of centres due to logistical reasons. The observed variation in outcome could partly be explained by case-mix in these centres, but perhaps this cannot be generalized to all centres. Fortunately, we collected data about characteristics of these centres and were able to compare our sample’s characteristics to those of the universe of hospitals in the Netherlands. Because we are transparent about these differences, the data can be interpreted with more context. Another limitation of our study is the presence of missing data. We dealt with missing data by using multiple imputations. Using this method, we have assumed that the data were missing at random. Unfortunately, there is no empirical way to check this assumption. The fact that a complete case analysis showed same direction and uncertainty of effects is reassuring. Finally, we only were able to assess the process and structure of care indicators which we collected in this study. Other potential process indicators are the time to defibrillation in patients with IHCA by shockable rhythm, or time to BLS. Both indicators were not (accurately) collected and therefore could be of interest in future studies. That is, if unexplained differences in outcome are found between centres.
This study introduces metrics for the evaluation and improvement of resuscitation care. Notable strengths of our study include the large sample size and the comprehensive adjustment for both random variation and case-mix. Based on our findings, the following two recommendations for clinical management and research for IHCA can be proposed: we should improve care for IHCA mainly to improve neurological outcomes, i.e. through more frequent CPR training of staff; existing outcome measures of IHCA cannot be reliably used to compare hospitals on quality of care, as opposed to processes and structure of care indicators.
Acknowledgements
We are grateful for Virginia E. Tangel (Department of Anesthesiology, Erasmus University Medical Center, Rotterdam, the Netherlands, and Department of Anesthesiology, Weill Cornell Medicine, New York, NY, United States) for her revisions of the final manuscript. ROUTiNE study consortium: Evert-Jan Wils, Cees Kuijs, Michiel Blans, Bas van den Bogaard, Ankie Koopman – van Gemert, Chris Hukshorn, Nardo van der Meer, Marco Knook, Trudy van Melsen, René Peters, Patrick Perik, Jan Assink, Gerben Spijkers and Wytze Vermeijden.
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.