Stroke units, certification, and outcomes in German hospitals: a longitudinal study of patient-based 30-day mortality for 2006–2014

We linked hospital data from different sources based on standardized institutional codes, which are unique mandatory identifiers for each hospital in Germany. First, we obtained structural hospital data (e.g. case volume, hospital teaching status, type of ownership) for the available years 2006, 2008, 2010, 2012, 2013, and 2014 from the German mandatory quality monitoring system, operated by the executive authority of the German health care system, the Federal Joint Committee (Gemeinsamer Bundesausschuss, G-BA). The G-BA provides publicly available hospital report cards for research purposes upon request via XML files on hospital and annual level.

Second, we integrated risk-adjusted, patient-based stroke outcome data (for the stroke diagnoses I. intracerebral hemorrhage, ICD Code I61; II. ischemic stroke, I63, and III. stroke not specified as hemorrhage or ischemic, I64) from the Quality Assurance with Routine Data (Qualitätssicherung mit Routinedaten, QSR) program. The QSR is operated by the AOK, the largest German sickness fund, and employs routine in- and outpatient data of AOK insured patients. It provides a risk-adjusted 30-day SMR, comparing observed vs. expected events. For risk-adjustment purposes, the QSR calculates 30-day expected mortality by means of logit regressions which includes patient-specific risk-factors like age, gender, and a set of comorbidities [17, 18]. To ensure comparability across years, we applied the 2014 logit risk-adjustment model to the AOK patient data for all data years.

Third, we included information on SU certification from the German Stroke Society (Deutsche Schlaganfall Gesellschaft, DSG), the premier German SU certification scheme [19]. The data provides information on which hospitals have DSG-certified SUs and the period of certification. A DSG certificate, granted for three years, requires minimum patient volume, minimum volume of certain interventions, staff level resources, and training obligations. Hospitals with non-certified SUs were identified by two specific procedure codes (OPS 8-891and 8-89b), which capture provision of complex stroke care [20].We assumed the existence of a SU when a hospital reported at least ten such procedures per year [6]. Structural standards are generally higher for DSG certification than for documenting complex stroke procedures.

Fourth, we integrated data from the THQ certificate Cooperation for Transparency and Quality in Health Care (Kooperation für Transparenz und Qualität im Gesundheitswesen, KTQ), comparable to JCAHO accreditation. Central components include continuous quality improvement in: patient orientation, employee orientation, patient safety, quality management, communication, transparency, and leadership [21]. Like the DSG SU certificate, the certificate is granted for 3 years. Hospital specific information on both certification schemes were provided from the mentioned organizations and integrated via standardized institutional codes and address information.

Based on Donabedian’s quality framework [16], we hypothesize better stroke outcome quality for hospitals that organize care through: (ii) a dedicated SU facility, (iii) SU certification, and (iv) total hospital quality (THQ) certification, relative to the (i) conventional, non-SU care model. We employ a fixed effects model with a within-regression estimator at hospital level. To quantify the influence of (certified) SU care on stroke outcomes, we regress the log of stroke 30-day SMR (SMR_it) on separate dummy variables, specifying the existence of a SU (SU_it), a DSG-certified SU (acc _ SU_it) and a THQ certification (acc _ THQ_it). We add the log of stroke case volume (stroke _ CV_it) to model stroke treatment experience, and a flattening learning curve. We include the share of stroke patients relative to all patients treated to account for relative importance of and organizational focus on stroke care. Hospital beds (beds_it), dummy variables for hospital teaching status and ownership type, and a category medical specialization (CMS) [22] index reflect important time-variant characteristics. For time-variant trends that affect each hospital equally such as technological advances, regulatory changes, and judicial decisions, we specify time effects (τ_t), excluding 2006 as the reference year. To adjust for the optimal level of stroke quality of care with a 0 SMR value (0 observed mortality), we adapt Battese’s (1997) approach to include an dummy explanatory variable (\( {D}_{it}^{SMR} \)), which takes on the value of 1 when the SMR is 0, and add \( {D}_{it}^{SMR} \) to SMR_it before taking the log [23]. We further adjust for the fact that hospitals treat variable amounts of stroke patients using AOK patient stroke case volume as analytical weights. The main model is specified in Eq. 1:

In addition to the variables specified above, β₀ is the intercept, α_i is individual time invariant hospital-fixed effects, and ε_it is the error term. To assess result robustness, we further estimate the model using the log of the number of SU complex procedures instead of the dummy indicator variable for stroke units. The data comprise repeated measurements at the hospital level which may involve autocorrelation in the error term ε_it. A Hausman test indicates that a random effects specification would likely yield inconsistent estimates. We therefore use hospital fixed effects α_i to control for unobserved hospital characteristics and avoid inconsistencies. Testing the time-fixed effects τ_t for joint significance indicates systematic differences in mortality across years. All statistical inferences are based on heteroscedasticity- and autocorrelation-consistent estimates for the standard errors.

Our analysis confirms the positive trend over time of SMR reduction after stroke in Germany, although to a much lower degree than prior studies have shown [6]. This can be attributed to the use of patient-based 30-day mortality data, including time after patient discharge. This data enables a cross-sectoral perspective on stroke care and demonstrates the shortcomings of admission-based data.

The descriptive stroke SMR trends for the different hospital sub-groups suggest progressively better stroke outcomes in hospitals with SU infrastructure, a SU that is also DSG-certified, and a certified SU within a THQ certified hospital. Results of the fixed effects regression models also show that having a SU alone significantly enhances outcome quality of care. The results align with previous research and confirm the benefits of treating patients in a dedicated SU facility [7, 8, 14].

Conversely, both certifications do not show significant effects. The structural and process differences between non-certified and certified SUs might be too small to show significant impact, and the overall hospital quality management improvements associated with the THQ certification might not be meaningful enough to influence outcomes in emergency medical conditions such as stroke.

On a health system level, our results question why a large share of German stroke patients is still treated in non-specialized facilities, and, related, why the shift towards a centralized stroke treatment model is sluggish [6]. Our findings suggest that treating all stroke patients at hospitals with a SU may result in a decrease in the absolute 30-day stroke mortality by 5.6%, from 16.2 to 15.3% even after adjusting for case volume and share of stroke cases. For those roughly 50,300 stroke patients currently treated at hospitals without SUs, this would correspond to 460 fewer annual stroke-related deaths. Considerable reductions in stroke-related disabilities and in medical and economic costs are additional expected benefits [7].

Experience in other European countries demonstrates the positive outcome impact of stroke care centralization in SUs [25, 26]. Underpinning the centralization argument is the positive volume-outcome relationship, which has also been shown to hold for stroke [27]. In the mid-term, national and regional policy makers should ensure that all stroke patients are treated in SUs by requiring SU infrastructure for stroke care and centralizing stroke care with hospitals that already operate a well-performing SU.

The German certification of SUs sets high procedural, personnel, and infrastructural standards; however, as above, in contrast to expectations, the SU service line certification shows no additional significant improvement with 30-day stroke SMR when non-certified SU existence is controlled for. Several explanations are possible. First, DSG certification confirms the SU set-up externally, with some additional staffing and process requirements. These enhancements might not have a large enough additional effect on the 30-day mortality compared to the standard SU characteristics.

Second, mortality is a valid and well-accepted outcome parameter [28], but it is only one of the outcomes that matters in stroke care [29]. Others, such as readmissions, degree of disabilities, and quality of life are also important [7, 29]. Standardized and risk-adjusted data for these outcome parameters are not currently available in Germany. Certified SUs, however, might have better outcomes for these indicators because the DSG certification takes a holistic approach, focusing on reducing disabilities after stroke [19]. Third, certified SU might have improved outcomes over a longer timeframe than the 30 days after hospital admission examined here.

Likewise, certified SUs might provide care for more severe patients, as they have on average substantially higher case volumes (Fig. 1). While the standardized 30-day stroke mortality is adjusted for co-morbidities, stroke severity (e.g. National Institutes of Health Stroke Scale from 0 to 42) is not fully reflected by administrative data [30]. However, the impact of severity adjustment on risk-adjusted indicators that already are adjusted for co-morbidities, age and other patient characteristics has been shown to be limited [31]. Lastly, the suspension of the DSG SU certification process in 2008 and first months of 2009, which resulted in delays for about 100 re- or new stroke unit certifications [32], might have also reduced the effectiveness of the DSG certification for the time span 2008–2012 and the amount of 30-day stroke SMR improvement attributable to the DSG certification.

THQ certification showed no additional significant effect on 30-day stroke mortality, in line with previous studies in other countries [11, 13]. The primary purpose of this certification is the general improvement of hospital quality management; its achievement might not be appropriately reflected by 30-day mortality in one specific emergency condition. Other measures such as patient safety, patient and employee responsiveness and satisfaction, and operational efficiency at the hospital level might be more affected by THQ certification. For example, Lindlbauer et al. (2016) show improved technical efficiency for THQ-certified hospitals. A downward bias of the THQ effect could be possible due to the fact that no consolidated and standardized data on ISO 9001 certification, which is a universal quality certificate also applied in hospitals, is available. Hospitals without a KTQ certification might alternatively have an ISO 9001 THQ certification even though they appear without THQ certification in our dataset. However, the number of ISO 9001 certifications is likely substantially smaller compared to the KTQ-certified hospitals [22].

Lastly, there are benefits from certification schemes that are not captured by outcome data. Both the SU and the THQ certification provide quality signals for patients, emergency teams, and admitting physicians, which can facilitate hospital choice decisions.

Besides the limitations mentioned above, the results of this study should be viewed considering some data and methodological limitations. The validity of self-reported hospital data might be compromised, due to reputational concerns by hospitals and different coding practices. Annual, random validity checks and cross-checks with administrative patient data, demonstrated for 5% of hospital reports some validity issues affecting 15–60% of the examined reporting data(26, 57).

The analyzed post-discharge timeframe of 30 days for stroke mortality provides substantial information on outcome quality, but an extended period like365 days might provide additional insights. While the AOK QSR indicators have some advantages, they only rely on data for patients insured by the AOK sickness fund. This might lead to biased outcome indicators, but the high share of AOK insured patients in all German hospitals (35% average market share) and results from previous studies (58) demonstrate the representativeness of the AOK QSR data.

Even though the outcome data is risk-adjusted for a large set of comorbidities and age, some bias might be affecting the results as the outcome data is not fully adjusted for severity. This might especially affect certified stroke unit hospitals as they could receive more severe cases, also via transfer from non-certified stroke units, leading to higher mortality that is not accounted for in the patient-based risk adjustment. Therefore, the effect of a SU certification or a full hospital certification is possibly underestimated in our data.