Effects of Institutional Caseload of Subarachnoid Hemorrhage on Mortality
A Secondary Analysis of Administrative Data
Background and Purpose—Procedures requiring specific skill sets often have been shown to depend on institutional volume, that is, centers receiving a higher volume observe better outcomes in those patients. This relationship recently has been shown to exist for subarachnoid hemorrhage(SAH) patients in a large study in the United States. We aim to examine this relationship for SAH patients in England, restricting analysis to specialist neurosurgical units.
Methods—Aggregate counts of patients with SAH in 25 specialist neuroscience centers in England, from 2005 to 2011, were obtained from the Hospital Episode Statistics database maintained by the National Health Service Information Center. These data were linked with national mortality statistics to obtain counts of deaths. Poisson regression was used to investigate the relationship between institutional caseload of SAH and 6-month mortality from any cause. Six-month mortality rates and mortality ratios were computed.
Results—Annual institutional caseload of admissions with SAH was inversely related to 6-month mortality (P=0.009; r2=0.26). Each 100-patient increase in annual patient volume was associated with a 24% reduction in mortality (adjusted mortality ratio, 0.76; confidence interval, 0.67–0.87). This relationship was consistent across the entire range of annual institutional caseloads examined (29–367 cases for the lowest and highest volumes seen in a single center in 1 year).
Conclusions—Our results provide support for management of SAH at high-volume centers and suggest that health care policy in this setting should pursue regionalization while ensuring an adequate geographic spread of access to care.
Primary subarachnoid hemorrhage (SAH) is a devastating illness with an estimated incidence rate of 10 per 100 000 patient-years in the general population.1,2 In the majority of cases (≈85%), SAH is the result of a ruptured arterial aneurysm.2–4 Definitive management of aneurysmal bleeds requires securing of the aneurysm because rerupture is common and increases risk of mortality.5,6 A significant shift in the management of aneurysms, from surgical clipping to angiography-guided coiling, has been observed since the landmark International Subarachnoid Aneurysm Trial (ISAT) in 2005.7
Outcomes (both morbidity and mortality) from procedures requiring specific skill sets have been shown to depend on institutional and operator volume8,9 across numerous surgical disciplines.10–12 Similar findings were reported for the treatment of aneurysmal SAH in an era when surgical management was the prevailing mode of treatment.13 However, volume did not seem to be a factor affecting mortality at centers that offered both forms of treatment during the same era.14,15 More recently, however, a large national study in the United States, based on the National Inpatient Sample (NIS) database (http://www.hcup-us.ahrq.gov/nisoverview.jsp) reported lower mortality at high-volume centers when compared with low-volume centers for both clipping and coiling.16 Interestingly, they also observed that, over time, fewer interventions are being undertaken at low-volume centers.
These data are important but difficult to interpret in the context of health care systems (such as the United Kingdom National Health Service) that have more strategic organization of specialist services. Their threshold for defining a high-volume center was 20 cases per year, and 34% of centers dealt with 10 or fewer cases per year. Given this context, it is difficult to be certain that the differences in observed outcome were truly the consequence of caseload rather than simply reflecting the fact that specialist neurosurgical expertise was not part of the organizational establishment in low-volume centers. Furthermore, it was unclear whether caseload volume would continue to have an impact on outcome at higher levels, which are probably of greater relevance in international comparisons.
Neurosurgical and interventional neuroradiological services in the National Health Service (NHS) are delivered at specialist neuroscience centers across the country, each of which acts as the referral center for a varying number of district hospitals serving a defined regional population. All admissions to each of these centers are captured in a national database, Hospital Episode Statistics (HES; www.hesonline.nhs.uk), which contains information on all NHS hospital admissions, including fee-paying patients and nonresident patients who received health care funded by the NHS. Data contained in the warehouse include demographic data, diagnoses and operative procedures, administrative data, and geographical information for each hospital stay. We have used this large administrative database to examine the impact of center caseload on mortality from SAH in neuroscience centers across England.
Data collection, analysis, and reporting were undertaken in concordance with Strengthening the Reporting of Observational studies in Epidemiology (STROBE) recommendations (STrengthening the Reporting of OBservational studies in Epidemiology, http://www.strobe-statement.org/).
Aggregate data were provided by the Information Center of the NHS, which maintains the HES database. Ethical approval was not required to obtain these data because they are aggregated and not patient-identifiable. We obtained data of a number of SAH cases by center, from April 1, 2005 to March 31, 2011, across all neurosurgical centers in England. These centers were identified from the records of the web site of the Society of British Neurological Surgeons, as documented on their web site (http://www.sbns.org.uk/files/8113/3545/4150/Neurosurgical_Units.pdf). We included all hospitals on this list except for the following exclusions: hospitals in Scotland, Northern Ireland, Wales, and the Republic of Ireland; dedicated pediatric hospitals; and 1 neurosurgical center in England, where the true counts could not be ascertained reliably from the hospital records.
For the 25 included centers, counts were obtained, within each annual period, based on the number of unique patients who had an episode of care with primary diagnosis recorded as I60 (SAH) according to International Classification of Diseases, 10th revision (http://apps.who.int/classifications/icd10/browse/2010/en). To obtain all-cause mortality data within 6 months from the end of their hospital episode, SAH patients from each year were linked using patient identifiers unique to the mortality database for England and Wales, which is maintained by the UK Office for National Statistics. The use of these data for anonymized, linked analyses was compliant with national regulatory and ethical requirements.
Crude 6-month mortality rates were computed across all centers, within each year, using counts of SAH patients and the number of those patients who died within 6 months. A Poisson regression model was fitted, adjusting for SAH patient caseload, year of hospitalization, and center of treatment. Year and center were fitted as categorical variables, whereas fractional polynomials were used for patient volume to allow for a possibly nonlinear relationship. Using this model, fitted mortality rates were estimated for each center and plotted against the average number of SAH cases per year in the corresponding center. Mortality ratios were derived for patient caseload (based on single and 100-patient increases in caseload) and year of hospitalization.
The NHS Information Center provided data from >300 NHS centers where patients had been admitted with a diagnosis of SAH. From these data, we extracted the relevant data for the 25 neurosurgical centers we were interested in. From April 1, 2005 to March 31, 2011, >23 000 patients with SAH were managed in these 25 centers. Over this 6-year period, the mean number of SAH patients seen per year, across all centers, was 3861; the mean number of deaths within 6 months among those patients was 525 per year (Table 1). This corresponds to an average 6-month mortality of 13.6%. The center treating the largest number of SAH patients saw an average of 260 SAH patients per year between 2005 and 2011, whereas the center with the smallest number saw an average of 54 (Figure 1).
Fitted mortality rates for each center, adjusted by year of hospitalization, are presented in Figure 2. The highest 6-month mortality of SAH patients at a single center was 22% and the lowest was 10% (Table I in the online Data Supplement). There was an inverse relationship between mortality and average caseload of SAH patients per year. Lower mortality rates were observed in centers with larger caseloads of SAH patients. A linear relationship was fitted between these mortality rates, and average annual caseload and was found to be statistically significant (P=0.009).
The 6-month mortality ratio for SAH patient caseload confirmed a linear relationship between caseload and mortality. The ratio was 0.997 per 1-patient increase in caseload (95% confidence interval [CI], 0.996–0.999; Table 2), corresponding to a reduction in mortality of 24% for every 100-patient increase in SAH caseload per year (adjusted rate ratio, 0.76; 95% CI, 0.67–0.87). This effect of caseload on 6-month mortality is shown graphically in Figure 3, in which mortality for a range of caseloads is compared with a caseload of 100 SAH patients in 1 year. For example, a center treating 350 SAH patients in 1 year has an almost 50% lower 6-month mortality rate for those patients compared with a center treating 100 SAH patients (Table II in the online Data Supplement). Although there was an annual increase in the number of SAH patients over the 6-year period, and although the overall mortality rate declined over time (Table 1), this trend with time did not reach significance in the Poisson regression model (Table 2).
We show that 6-month mortality after SAH across 25 English neuroscience centers is associated with the annual caseload of patients admitted with this diagnosis. Centers with higher SAH patient caseloads had lower 6-month mortality rates, and this benefit appeared to persist for annual caseloads of up to >350 cases (the highest volume in our data was 367 cases in 1 year). Six-month mortality after SAH did not significantly change over the 6-year study period across all centers as a whole.
Improved outcomes at high-volume centers have been previously demonstrated in other clinical contexts and in previous studies in aneurysmal SAH.16–22 However, most previous work originates from the United States and is predominantly based on data from specific states. Even the most recent United States national study using NIS data is based on only a 20% sampling of community hospital admissions.16 Furthermore, Leake et al16 distinguished between low-volume and high-volume centers based on a threshold of 20 cases per year and included all varieties of hospital. This makes it difficult to distinguish between the impact of specialist neurosurgical care, which may have been less reliably available (or absent) at the smaller hospitals and may have been a true volume effect. We present robust data from a large-scale database representing almost all neuroscience centers in England (1 excluded for unreliability of data). The fact that we have restricted our analysis to clinical neuroscience centers provides strong evidence supporting an effect of varying caseload rather than difference in access to basic neurosurgical care. Furthermore, we show that the impact of caseload on outcom e persists well beyond the 20-case threshold demonstrated previously, with persisting improvements in outcome for institutional caseloads of up to approximately 350 cases in 1 year.
The strength of the study lies in the large amount of data collected over 6 years, obtained from the Information Center of the NHS, which maintains comprehensive hospital records of all admissions in the HES database. However, it is important to acknowledge that our analyses suffer from significant limitations.
We could not obtain reliable information regarding specific cause of SAH (aneurysmal vs nonaneurysmal) or treatment modality (surgical clipping vs interventional coiling) because these data were not reliably available in the HES database. Our inability to explore specific diagnoses and treatment differences may have resulted in significant confounding if there were major differences in case mix between centers, or if the higher-volume centers were providing more patients with a more effective treatment option and the observed effect was related to this, rather than directly to caseload. However, overall mortality was stable over the 6 years, and the majority of the study period postdates any presumed swing in treatment modality after the publication of the ISAT trial,7 suggesting that changes in overall management trends are unlikely to have affected the results.
Given that this is not a randomized controlled trial, an additional limitation is the use of all-cause mortality outcome rather than a more specific end point of SAH-related mortality. We also were unable to control confounding of differences in overall mortality rates between centers. Nonetheless, given that the majority of the centers are large hospitals in large cities, we do not expect great variation in mortality rates. Therefore, even though any bias in numbers of SAH-related deaths could affect the magnitude of our computed mortality rates, it is unlikely to change the shape of our relationship of interest.
Furthermore, because we obtained aggregate data, we were unable to adjust for possible confounding by patient characteristics, such as age and sex, or by characteristics of the diagnosis, such as severity at presentation. If there was large variability in such characteristics across centers, then this may have confounded the relationship between SAH volume and mortality. However, the possibility of more complex high-risk patients remaining at lower-volume centers is balanced by the more likely situation of being transferred to higher-volume centers, affecting mortality rates adversely in these latter hospitals.
It is important to note that, unlike many previous studies, we used mortality rates at 6 months rather than at hospital discharge as our primary outcome. This approach accounted for clinical practice in the NHS, and patients who are experiencing slow neurological recovery are often discharged from specialist neuroscience centers to referring district general hospitals or to a rehabilitation facility in another institution. The use of 6-month outcomes meant that the uncertainties associated with such transitions of care were less of a concern. In addition, this also allowed comparison with a previous United Kingdom audit,15 which provided Glasgow outcome data (and hence mortality) at 6 months and showed similar levels of mortality in the 2 studies (14.3% vs 13.6%). It also accounts for our plans to undertake a subsequent study in a smaller number of centers with full risk adjustment (SAH grade, age, sex, and comorbidities) and full assessment of Glasgow outcome score collected (as is conventional) at 6 months postictus. Given this context, comparison of 6-month mortality in our current study with that observed in the planned study will allow cross-validation.
Finally, there are inherent limitations in dealing with administrative databases. Minor biases will exist in the counts of SAH patients and deaths because some patients could be missed if they did not have a completed episode of care (because primary diagnosis information is only added once the episode is complete). Some deaths may have been missed if they occurred outside England and Wales. These numbers should be low, and the effect is likely to impact all centers in the same way. Diagnostic coding (such as International Classification of Diseases, 10th revision) is also not without its limitations.23 However, SAH is one of the stronger performing subcodes of strokes, with a reported positive predicted value of 65% to 100%.24 One additional potential confounder is that the International Classification of Diseases, 10th revision, coding for our data identified patients with primary SAH from any cause, rather than specifying the cause of SAH. It is possible that the differences in outcome at different sites may have been partly accounted for by differences in pathogenesis at different centers. However, this seems unlikely because the dominant contribution of aneurysmal SAH in population studies of SAH across many countries, and over time, has remained consistent.25–30
Given these limitations, it is important to acknowledge that our results are discordant from an earlier United Kingdom audit of SAH outcomes15 undertaken during the transition from an era of surgical clipping to one of mixed modalities of treatment, which showed no effect of volume on mortality, outcome, or case-mix adjusted outcome. However, the United Kingdom SAH audit15 varied from our study in several features. It involved a much smaller sample size (3174 patients), reported outcomes only for patients with confirmed aneurysms, did not provide full case ascertainment, and varied in whether centers reported all-comers or chose to only recruit patients who had been treated with clipping or coiling (the proportion of patients with aneurysmal SAH in this study who were not treated varied from 0%–28.6%).
Although our data show that the outcome of aneurysmal SAH is significantly associated with caseload, it remains unclear from our analyses whether this is truly a center effect or is primarily driven by operator caseload, with a smaller number of highly experienced operators delivering the improved outcomes in high-volume centers. Even if this was a center effect, it is unclear whether the expertise in care relates to the definitive treatment options available (neurosurgical clipping or interventional coiling) or is the consequence of more distributed expertise across neurosurgery, neurocritical care, neuroimaging, interventional radiology, and neurorehabilitation. Furthermore, the number of cases that distinguishes an institution or operator from being low-volume or high-volume in terms of impact on outcome also is not clear because a threshold effect was not apparent in our analyses. These are important issues and deserve further study to allow rational planning of services, but our results provide a strong impetus for a future study that takes account of robust risk adjustment and uses the full Glasgow outcome score as an outcome variable. If this subsequent study confirms our findings, then it would provide further support for management of SAH at high-volume centers and would suggest that health care policy in this setting should pursue regionalization while ensuring an adequate geographic spread of access to care.
Sources of Funding
This study was supported by the Neuroscience Theme of the Cambridge Biomedical Research Center, funded by the National Institute for Health Research (NIHR), UK. D.K. Menon is supported by a Senior Investigator award of the NIHR.
The online-only Data Supplement is available with this article at http://stroke.ahajournals.org/lookup/suppl/doi:10.1161/STROKEAHA.112.681254/-/DC1.
- Received October 29, 2012.
- Revision received December 3, 2012.
- Accepted December 19, 2012.
- © 2013 American Heart Association, Inc.
- Ostbye T,
- Levy AR,
- Mayo NE
- Bonita R,
- Beaglehole R,
- North JD
- Rosenørn J,
- Eskesen V,
- Schmidt K,
- Rønde F
- Molyneux AJ,
- Kerr RS,
- Yu LM,
- Clarke M,
- Sneade M,
- Yarnold JA,
- et al
- Katz JN,
- Barrett J,
- Mahomed NN,
- Baron JA,
- Wright RJ,
- Losina E
- Jain N,
- Pietrobon R,
- Hocker S,
- Guller U,
- Shankar A,
- Higgins LD
- Johnston SC
- 15.↵The Royal College of Surgeons of England. National Study of Subarachnoid Haemorrhage: Final report of an audit carried out in 34 Neurosurgical Units in the UK and Ireland between 14 September 2001 to 13 September 2002. London, UK: The Royal College of Surgeons of England; 2006.
- Solomon RA,
- Mayer SA,
- Tarmey JJ
- Berman MF,
- Solomon RA,
- Mayer SA,
- Johnston SC,
- Yung PP
- Bardach NS,
- Zhao S,
- Gress DR,
- Lawton MT,
- Johnston SC
- Manuel DG,
- Rosella LC,
- Stukel TA
- Williams GR,
- Jiang JG,
- Matchar DB,
- Samsa GP
- Rinkel GJ,
- van Gijn J,
- Wijdicks EF
- van Gijn J,
- Rinkel GJ