Off-Hour Admission and In-Hospital Stroke Case Fatality in the Get With The Guidelines-Stroke Program

Whether there are adverse consequences for patients who present to the hospital outside of regular working hours, most notably weekends, has been the focus of several large studies over recent years. Studies of acute hospital admissions in Ontario demonstrated an increased mortality risk for weekend arrival in 23 of 100 conditions¹ as well as delays in the timing of some critical procedures.² However, a large study from California found limited evidence of a “weekend effect” with only 3 of 50 conditions evaluated having a significantly higher mortality with weekend admission.³ Two studies of acute myocardial infarction found an elevated mortality risk for weekend admissions,^4,5 and in both reports, the magnitude of the weekend effect was attenuated after the use and/or delayed timing of specific cardiac procedures were accounted for.

Evidence for a “weekend effect” among acute stroke admissions has also been inconsistent. Studies from Ontario and California found no evidence of a weekend effect on in-hospital mortality for any stroke subtype,^1,3 whereas a large Canadian study did find an elevated mortality risk among ischemic stroke admissions.⁶ Also, a Japanese study based on cases admitted to stroke units found that weekend admission was associated with both a higher risk of in-hospital mortality and poorer functional outcomes at discharge.⁷ However, in both the Canadian and Japanese studies, the absolute increase in mortality associated with weekend admission was small; the 7-day mortality was increased by 1.1% in the Canadian study and approximately 0.5% in the Japanese study. Using data generated from hospitals that participated in the Get With The Guidelines (GWTG)-Stroke program, a voluntary national quality improvement program, our objectives were to first examine the effect of presentation outside of regular working (termed off-hour presentation) on in-hospital mortality among acute stroke admissions and second, to compare the quality of care (QOC) provided to subjects who presented during either off- or on-hours.

Methods

The GWTG-Stroke quality improvement program has been in development since 2000.⁸ An initial pilot quality improvement intervention that involved specific regions within 8 states (Massachusetts, Michigan, Ohio, Florida, Arizona, California, Pennsylvania, Georgia) was first conducted between April 2003 and March 2004. All hospitals within these regions were invited to participate. Starting in April 2004, the program was expanded and made available to any hospital in the United States. Individual hospitals could have joined the program at any time between April 2003 and April 2007 (when this particular data set was closed). The duration of hospital participation, defined as the number of consecutive quarters of participation after a hospital submitted data on a baseline set of 30 cases, therefore varied between hospitals. Each participating institution received human research approval to enroll subjects in GWTG-Stroke without requiring individual patient consent or a waiver of authorization and exemption from Institutional Review Board review based on the use of deidentified records and absence of direct patient contact.

Case Identification and Data Abstraction

Trained hospital personnel were instructed to ascertain consecutive acute ischemic stroke (IS) and hemorrhagic stroke (HS) admissions. Case ascertainment was conducted by prospective clinical identification, retrospective identification using International Classification of Diseases-9 discharge codes, or a combination of the 2 approaches. Exact methods used for prospective identification varied according to the size and organization of the hospital but would have included a combination of regular surveillance of presenting symptoms and chief complaints in the emergency department and review of ward census logs and/or neurological consultations.⁹ Retrospective identification of IS admissions involved use of International Classification of Diseases-9 codes 433, 434, or 436, whereas HS admissions were identified using International Classification of Diseases-9 codes 430 and 431. The eligibility of each acute stroke admission was confirmed at chart review before abstraction.

Data were abstracted by hospital personnel using an Internet-based Patient Management Tool (Outcome Inc, Cambridge, Mass). All users received either online or telephone-based training in the use of the tool. Abstracted data included patient demographics, medical history, initial head CT findings, in-hospital treatment and events, discharge treatment and counseling, and discharge destination. The data collection tool supports both concurrent data collection as well as retrospective data entry and includes predefined logic features, range checks, and user alerts to identify potentially invalid values. Patient confidentiality was maintained on this web-based system by the use of passwords, deidentified data sets, and secure data transmission techniques.

Data on hospital-level characteristics, ie, bed size, academic or nonacademic status (as defined by the American Hospital Association),¹⁰ annual volume of stroke discharges, and geographical region, were collected at the time of initial hospital enrollment.

Sample Population

We obtained data on 365 186 stroke admissions who presented to 857 hospitals that participated in the program during the 4-year period between April 2003 and April 2007. We excluded subjects that were not acute admissions (n=56 641), that did not arrive through the emergency department (n=883), or that had unreliable or missing data for any of the following variables: date of arrival (n=3485), time of arrival (n=52 774), length of stay (n=1086), or vital status at discharge (n=6701). The combined effect of these deletions was a final data set of 300 961 acute stroke admissions (82% of the original total), of which 62% were IS, 12% were HS, 23% were transient ischemic attack, and 3% were stroke of uncertain type. After excluding the last 2 groups, the final data set included 187 699 and 34 845 IS and HS admissions, respectively.

Exposure and Quality of Care Definitions

On-hour presentation was defined as presenting to the hospital emergency department between 7:00 am to 6:00 pm on any weekday. Off-hour presentation was defined as presenting any other time, including evenings, nights, and weekends and during national holidays. The following 8 performance measures⁸ and one safety measure were used to quantify the quality of care (QOC) provided to the IS admissions.

Acute and Subacute Measures

• Intravenous tissue plasminogen activator in patients who arrive <2 hours after symptom onset;

• Antithrombotic medication within 48 hours of admission (early antithrombotics);

• Deep vein thrombosis prophylaxis within 48 hours of admission; and

• Screening for dysphagia before oral intake.

Discharge Measures

• Antithrombotic medication;

• Anticoagulation for atrial fibrillation;

• Treatment for low-density lipoprotein >100 mg/dL; and

• Counseling or medication for smoking cessation.

Safety Measure

• Symptomatic intracranial hemorrhage within 36 hours after intravenous tissue plasminogen activator.

For HS admissions, QOC was quantified using the following 3 performance measures: deep vein thrombosis prophylaxis, screening for dysphagia, and discharge smoking cessation.

Statistical Analysis

All statistical analyses were performed using SAS Version 9.1 software (SAS Institute, Cary, NC). Contingency tables were generated to explore the relationship between important covariates, including demographics, clinical variables, medical history, and hospital-level characteristics, and time of presentation, ie, off-hour versus on-hour presentation. Similarly, contingency tables were generated to explore the relationship between QOC measures and in-hospital complications and off-hour versus on-hour presentation. χ² test for nominal data and Wilcoxon rank sum tests for ordinal and continuous data were used as tests for statistical associations. It should be noted that given the large sample size that even small absolute differences (ie, <1%) reach statistical significance (ie, P<0.05).

The relationship between off-hour presentation and in-hospital mortality was examined using multivariable logistic regression models.¹¹ To account for possible within-hospital clustering, generalized estimating equation methods were used to generate both unadjusted and adjusted models.¹² Given the large size of the data set, traditional statistical approaches to model-building that identify candidate confounders on the basis of statistical significance were not used. Instead, the final models were adjusted for several prespecified patient-level and hospital-level variables that were regarded as potential confounders. These included age, race, gender, body mass index, arrival mode, and medical history and risk factors (including atrial fibrillation, previous stroke/transient ischemic attack, coronary heart disease or prior myocardial infarction, carotid stenosis, diabetes, peripheral vascular disease, hypertension, dyslipidemia, and current smoking), and length of stay. Hospital-level characteristics included bed size, academic hospital, and region.

We undertook 2 analyses to explore interaction effects on the relationship between off-hour presentation and mortality. First, under the hypothesis that the duration of participation in GWTG program might reduce the disparity related to off-hours presentation, we tested the significance of the interaction between off-hours presentation and the duration of hospital participation. Second, under the hypothesis that larger teaching hospitals may have more staffing during off-hours, which could reduce the effect of off-hours presentation on mortality, we tested interaction effects between off-hours presentation and hospital size and teaching status. These analyses was done separately for both IS and HS admissions.

Results

Ischemic Stroke

Of 187 669 IS admissions, half (n=94 008) arrived during off-hours. Table 1 compares the characteristics of IS admissions who presented during off-hours with those who presented during on-hours. The off-hour presentation group was slightly younger than the on-hour group (median age, 74 versus 75 years) and were more likely to arrive by emergency medical services from the scene (56.2% versus 53.6%) or to be transferred by emergency medical services from another hospital (6.2% versus 4.0%). However, despite the presence of several statistically significant (P<0.05) associations, there were no other clinically important differences between the 2 groups in terms of gender, race, medical history, risk factors, or hospital characteristics (Table 1).

Clinically important differences in the quality of care provided to patients who presented during off- or on-hours were small to nonexistent (Table 2) The proportion of patients who arrived within 2 hours who were treated with intravenous tissue plasminogen activator was slightly lower during off-hours (56.4% versus 58.8%), but deep vein thrombosis prophylaxis rates were slightly higher in the off-hour group (67.4% versus 65.6%). In terms of in-hospital complications, the proportion of patients treated for pneumonia was slightly higher in the off-hour group (6.1% versus 5.5%).

A total of 10 326 IS admissions died in-hospital resulting in an overall case fatality rate of 5.5%. The in-hospital case fatality rate was higher for admissions that arrived in the off-hours (5494 of 94 008 [5.8%]) compared with those that arrived during regular work hours (4832 of 93 661 [5.2%]). The absolute difference in mortality (0.6%) translates into number needed to harm of 166 for off-hour presentation. As expected, subjects who died during hospitalization were older than those that survived (median age, 80 versus 74 years) and were more likely to be female (57.6% versus 53.0%), white (78.1% versus 74.1%), to have a medical history of atrial fibrillation (34.1% versus 16.9%) or heart disease (34.8% versus 26.6%), to have been transported by emergency medical services (88.5% versus 58.3%), and to require treatment for pneumonia during hospitalization (18.7% versus 5.1%). Subjects who died in-hospital were less likely to be current smokers (10.9% versus 17.5%) or to have a history of dyslipidemia (26.0% versus 34.1%). The in-hospital mortality rate was higher in larger, academic hospitals (5.7% academic versus 5.1% nonacademic).

The median length of stay was 4 days (interquartile range, 3 to 7) for both off-hour and on-hour presentations. Among the 177 343 IS subjects discharged alive from the hospital, 48.3% were discharged home, 22.9% to a nursing home, 21.5% to rehabilitation, 3.4% were transferred to another acute care facility, and 3.4% were discharge to hospice. Slightly fewer admissions who arrived during off-hours were discharged home (47.0%) compared with those who presented during on-hours (49.5%; P<0.001).

Hemorrhagic Stroke

Of 34 845 HS admissions, 79.5% (n=27 710) had intracerebral hemorrhage and the remainder subarachnoid hemorrhage. Just over half of the HS admissions (56.7% [n=19 767]) arrived during off-hours. Table 3 compares the demographic and clinical characteristics of HS admissions who presented during off- and on-hours. Subjects who presented during off-hours were a little younger than the on-hour group (median age, 69 versus 71 years) and were more likely to arrive by emergency medical services hospital transfer (22.1% versus 16.2%). Whites were slightly more likely to present during on-hours and blacks were slightly more likely to present during off-hours, but no other clinically important differences were noted between off- and on-hour presentations among the HS admissions (Table 3).

Similar to IS admissions, clinically important differences in the QOC provided to patients with HS who presented during off- or on-hours were small to nonexistent (Table 4). Among the quality indicators relevant to HS care, the proportion of patients who received dysphagia screening was slightly lower in the off-hour group (40.2% versus 41.2%). The proportion of patients treated for pneumonia was slightly higher in the off-hour group (10.0% versus 8.9%).

The case fatality rate for all HS cases was 25.9%, whereas the rates for intracerebral hemorrhage and subarachnoid hemorrhage were 26.6% and 22.9%, respectively. The in-hospital case fatality rate was higher for admissions that arrived in the off-hours (5368 of 19 767 [27.2%]) compared with those that arrived during regular work hours (3640 of 15 078 [24.1%]). The 3.1% absolute difference in mortality translates into a number needed to harm of 32 for off-hour presentation. As expected, subjects who died during hospitalization were older than those that survived (median age, 74 versus 69 years), and they were also more likely to arrive by emergency medical services (72.9% versus 51.6%) and to have a medical history of atrial fibrillation (16.9% versus 12.2%) or heart disease (22.0% versus 18.1%). Similar to IS cases, they were less likely to smoke or report a history of dyslipidemia.

The median length of stay was 5 days (interquartile range, 2 to 11) for both off-hour and on-hour presentations. Among the 25 837 HS subjects discharged alive from the hospital, 35.8% were discharged home, 24.2% to a nursing home, 24.3% to rehabilitation, 8.6% were transferred to another acute care facility, and 6.6% were discharge to hospice. Similar to the IS cases, slightly fewer admissions who arrived during off-hours were discharged home (34.9%) compared with on-hour admissions (36.9%; P<0.001).

Multivariable Analysis of the Effect of Off-Hour Presentation on In-Hospital Mortality

The crude and adjusted OR estimates with 95% CIs for off-hour presentation compared with on-hour presentation are shown in the Figure for both IS and HS admissions. In unadjusted analyses, presentation during off-hours increased the odds of in-hospital mortality by 13% (OR, 1.13; 95% CI, 1.09 to 1.18) and 17% (OR, 1.17; 95% CI, 1.11 to 1.23) for IS and HS, respectively. These results changed little after adjustment for a wide range of patient-level and hospital-level characteristics; off-hours presentation was associated with a 9% elevated odds of in-hospital mortality among IS admissions (OR, 1.09; 95% CI, 1.03 to 1.14) and an 19% elevated odds for HS admissions (OR, 1.19; 95% CI, 1.12 to 1.27). Full multivariable model results for both IS and HS admissions can be found in Supplemental Tables I and II, available online at http://stroke.ahajournals.org.

• Download figure
• Download PowerPoint

Interaction effects between off-hour presentation and hospital size or academic status were not significant. However, among IS admissions, a statistically significant interaction was observed between the duration of participation in the GWTG-Stroke program and the effect of off-hour presentation on in-hospital mortality (P=0.002). The ORs for in-hospital mortality in the off-hour group declined with increased duration of participation; the OR for off-hour presentation (versus on-hour presentation) was 1.18 (95% CI, 1.09 to 1.27) for the first quarter of participation and declined to the null by the end of the second year (OR, 1.02; 95% CI, 0.95 to 1.09). The interaction remained statistically significant after controlling for calendar time, which was not associated with either in-hospital mortality or off-hour presentation. There was no statistically significant interaction between off-hours presentation and duration of program participation among HS admissions (P=0.93).

Discussion

In this study of over 220 000 acute stroke admissions, in-hospital mortality was higher for those that presented outside of regular working hours. Although the absolute effect of off-hours presentation on in-hospital mortality among IS admissions was small (ie, 0.6%), on a relative basis, the odds of mortality was almost 10% higher, even after adjusting for differences in patient and hospital characteristics. Among HS admissions, the off-hours effect was even stronger with both a higher absolute difference (ie, 3.1%) and relative difference (ie, adjusted OR, 1.18). The number needed to harm estimates indicate that for every 166 IS admissions and 32 HS admissions that present during off-hours, one extra death would be expected to occur. Although these absolute effects appear modest, the overall impact of presenting during off-hours is more considerable when one takes into account that more than half of all patients with stroke present during these hours. A metric for expressing the clinical impact of the off-hours effect is the population-attributable risk fraction, which expresses the proportion of in-hospital mortality that is attributable to presenting during off-hours. For IS admissions, the population-attributable risk fraction was estimated to be 4.2%, whereas for HS, it was 6.9%; thus, approximately one in 20 in-hospital stroke deaths could be avoided if the higher mortality associated with off-hour presentation was eliminated.

The most directly comparable studies that have examined the effect of presentation during nonregular work hours on acute stroke outcomes include 2 studies from Canada^1,6 and one from California.³ The first Canadian study examined data from almost 4 million acute emergency department admissions in Ontario between 1988 and 1997.¹ The study found no evidence of a “weekend effect” on in-hospital mortality for intracerebral hemorrhage (OR, 1.01) or IS (OR, 1.00), although it did identify a statistically significant elevated risk among cases of unspecified intracranial hemorrhage (OR, 1.23) and a trend toward an increased risk among subarachnoid hemorrhage (OR, 1.10). The second Canadian study used national-level data from over 26 000 IS cases discharged from 606 hospitals during a 1-year period (2003 to 2004).⁶ Weekend presentation was associated with a 8% higher adjusted odds of in-hospital mortality, an estimate similar to that seen in our study. The California study used data from over 24 000 IS cases admitted during a 1-year period (1998) to all acute care hospitals in the state. In contrast to our findings, they found no effect of weekend admission among cerebral infarction cases admitted through the emergency room (OR, 0.99).

The reasons why an effect of weekend presentation on in-hospital mortality is observed in some stroke studies but not others is unclear. All of these studies have used large representative samples and have adjusted for similar factors such as demographic factors and the presence of comorbidities. One explanation for the presence of a “weekend effect” is that acute stroke admissions are systematically different from those that present during the week. It is known that there is circadian variation in the timing of stroke onset with more stroke cases occurring in the morning hours between 6 am and noon¹³ and on Mondays.^14,15 It has also been hypothesized that weekend admissions may differ in terms of stroke subtype or severity¹⁶ as has been observed in acute myocardial infarction.¹⁷ However, solid evidence for differences in stroke subtype or severity by day of admission is lacking. Only one study that evaluated weekend effects on stroke outcomes was able to adjust for stroke severity, and an effect of weekend admission was still observed.⁷

Although the potential for the “weekend effect” to be accounted for by confounding or other biases inherent to observational data cannot be completely discounted, many observers believe these effects to be real^6,18 and that they reflect differences in the QOC provided to patients outside of regular work hours. The belief that differences in the QOC affect in-hospital mortality is bolstered by the observation that mortality is higher in for-profit hospitals compared with not-for-profit hospitals¹⁹ and in for-profit hemodialysis centers compared with not-for-profit centers.²⁰ These findings have been shown to be related to the reduced availability of highly skilled personnel per risk-adjusted bed, a factor that is strongly associated with hospital mortality.^21,22 Given the reduction in staffing that occurs during off-hours, there is clearly the potential for substandard care. Reduced access to specialists on weekends results in a concomitant reduction in access to urgent procedures.² The fact that the negative effect of weekend admission on in-hospital mortality of patients with acute myocardial infarction was attenuated after accounting for the lower use of cardiac procedures^4,5 is seen as strong evidence that the weekend effect is primarily driven by QOC.¹⁸ A recent study from the United Kingdom has shown that the quality of acute stroke care is influenced by when stroke cases are admitted.²³ In this study of over 8000 stroke cases treated at 246 hospitals, patients admitted on the weekend had longer delays in obtaining a brain scan or being admitted into a stroke unit.

The GWTG-Stroke program monitors the quality of in-hospital stroke care through the generation of core stroke performance measures that address acute, subacute, and discharge care.⁸ We did not find any clinically important differences in compliance with these performance measures between subjects who presented during off- or on-hours. Such findings could be interpreted as showing that the mortality differences found between off- and on-hour presentations are not due to underlying differences in the QOC. However, it could be argued that the performance measures used by the GWTG program do not reflect the acute care processes that likely influence in-hospital mortality such as the control of fluid and electrolyte imbalances, blood pressure, glycemia, and pyrexia. Moreover, determining the exact influence of QOC on in-hospital mortality is difficult, because it requires knowledge of the timing of the in-hospital death and the expected time of delivery of each care process. Details on the decision to withdraw care due to a terminal condition or the presence of do-not-resuscitate or comfort care orders would also need to be carefully documented. Unfortunately, such details are not available in the GWTG-Stroke program.

Given that the mortality differences for cases that present out of hours are thought to be driven primarily by differences in the quality and intensity of care, the “weekend effect” has been identified as a target for policy changes and quality improvement efforts.¹⁸ Increased reimbursement of hospitals to provide greater staffing of critical services on the weekends has been discussed in both Canada and the United States^2,24,25 Whether interventions to increase the availability of stroke nursing and/or specialty stroke services on weekends is both efficacious and cost-effective requires further study. We found that the mortality effect of off-hours presentation on IS admissions declined with longer program participation, suggesting that the weekend effect could be ameliorated by hospitals participating in stroke quality improvement initiatives over the long-term. It is possible that longer participation is associated with more consistent application of quality stroke care regardless of time of presentation.

This study has several limitations. First, the GWTG program is voluntary and the hospitals that participate are more likely to be larger teaching hospitals with a strong interest in stroke and quality improvement. Thus, the generalizability of these findings remains to be determined. Second, it was not possible to account for stroke severity because National Institutes of Health Stroke Scale data are poorly documented in this registry. Third, we cannot confirm that the subjects entered into the GWTG program represent a consecutive or unbiased patient sample. Hospitals are instructed to include all consecutive admissions or to take a systematic sample after selecting a random starting point. However, because these processes are not audited, the potential exists for selection bias.⁸ Finally, only in-hospital mortality was assessed and so deaths that occurred soon after discharge were not accounted for. However, to assess whether this could have introduced a bias, we repeated the analysis using a combined end point of in-hospital death or discharge to hospice, and the results were essentially unchanged (data not shown).

Footnotes