Bravata DM, Wells CK, Lo AC, Nadeau SE, Melillo J, Chodkowski D, Struve F, Williams LS, Peixoto AJ, Gorman M, Goel P, Acompora G, McClain V, Ranjbar N, Tabereaux PB, Boice JL, Jacewicz M, Concato J. Processes of Care Associated With Acute Stroke Outcomes. Arch Intern Med. 2010;170(9):804-810. doi:10.1001/archinternmed.2010.92
Many processes of care have been proposed as metrics to evaluate stroke care. We sought to identify processes of stroke care that are associated with improved patient outcomes after adjustment for both patient characteristics and other process measures.
This retrospective cohort study included patients 18 years or older with an ischemic stroke or transient ischemic attack (TIA) onset no more than 2 days before admission and a neurologic deficit on admission. Patients were excluded if they resided in a skilled nursing facility, were already admitted to the hospital at stroke onset, or were transferred from another acute-care facility. The combined outcome included in-hospital mortality, discharge to hospice, or discharge to a skilled nursing facility. Seven processes of stroke care were evaluated: fever management, hypoxia management, blood pressure management, neurologic evaluation, swallowing evaluation, deep vein thrombosis (DVT) prophylaxis, and early mobilization. Risk adjustment included age, comorbidity (medical history), concomitant medical illness present at admission, preadmission symptom course, prestroke functional status, code status, stroke severity, nonneurologic status, modified APACHE (Acute Physiology and Chronic Health Evaluation) III score, and admission brain imaging findings.
Among 1487 patients, the outcome was observed in 239 (16%). Three processes of care were independently associated with an improvement in the outcome after adjustment: swallowing evaluation (adjusted odds ratio [OR], 0.64; 95% confidence interval [CI], 0.43-0.94); DVT prophylaxis (adjusted OR, 0.60; 95% CI, 0.37-0.96); and treating all episodes of hypoxia with supplemental oxygen (adjusted OR, 0.26; 95% CI, 0.09-0.73).
Outcomes among patients with ischemic stroke or TIA can be improved by attention to swallowing function, DVT prophylaxis, and treatment of hypoxia.
Stroke is the seventh most common reason for hospitalization in the United States: approximately 750 000 strokes are reported annually.1,2 Many processes of stroke care have been proposed as quality indicators.3- 6 For most quality indicators, prospective data support the association between a process of care and a patient outcome, but reports of such associations often lack adjustment for important clinical characteristics (eg, stroke severity) or are not evaluated in the context of overall stroke care quality.
The goal of the present study was to identify processes of acute stroke care that are independently associated with decreased in-hospital mortality, discharge to hospice, or discharge to a skilled nursing facility. The present study differs from much of the prior literature in that our examination of the processes-outcome association includes adjustment for both clinical characteristics of the patients and adjustment for stroke care quality.
This retrospective cohort study evaluated patients admitted to a hospital with a stroke or transient ischemic attack (TIA) at any of 3 Department of Veterans Affairs (VA) hospitals or 2 non-VA hospitals. The hospitals were selected to provide a diversity of stroke care providers, geographic location, and volume of patients with stroke.
Patients were included if they were admitted with an acute ischemic stroke or TIA, had a neurologic symptom onset no more than 2 days before admission, had a neurologic deficit on admission (National Institutes of Health Stroke Scale [NIHSS]7 score, ≥2), and were at least 18 years old. Patients were excluded if they were residing in a skilled nursing facility or were already admitted at the time of the stroke onset or if they were transferred from another acute-care facility. If a patient had 1 or more stroke admissions during the study period (1998-2003), a single hospitalization was randomly sampled for inclusion.
Medical record reviews were conducted by nonneurologist physicians and nurses. Coding questions were addressed during weekly team meetings. Interobserver variability was assessed by a complete reabstraction on 10% of the sample.
We established 3 criteria for determining which processes of care to evaluate: (1) the process was a quality measure proposed by a national organization interested in stroke care and was reviewed by the National Expert Stroke Panel5; (2) prior (published) evidence linked the process with improved mortality or reduced institutionalization; and (3) the process was a component of acute stroke care. Seven processes of stroke care were identified for evaluation in this study: neurological evaluation, swallowing evaluation, deep vein thrombosis (DVT) prophylaxis, early mobilization, blood pressure (BP) management, fever management, and hypoxia management. Neurologic evaluation was assessed as present or absent on the basis of documentation of a neurologic examination by a neurology or neurosurgery attending physician or neurology house officer at any point during the hospital stay. Swallowing evaluation was assessed as present or absent on the basis of a consultation by a speech-language pathologist or documentation of a swallowing evaluation including bedside testing that included liquid and solid components, barium swallow, or fiberoptic examination.
Deep vein thrombosis prophylaxis was assessed as present or absent on the basis of whether the patient was treated with subcutaneous, intravenous (IV), or low-molecular-weight heparin or heparinoid; warfarin; pneumatic compression devices; or thrombolytic therapy or there was documentation that the patient was ambulatory within 48 hours of admission. Early mobilization was assessed as present or absent based on documentation that the patient was ambulatory or out of bed to chair within 24 hours of stroke symptom onset. Patients with coma, progressive neurologic symptoms, or acute myocardial infarction were considered to have a possible contraindication to early mobilization.
Guideline-concordant BP management8 within 48 hours of admission was defined as any of the following: (1) systolic BP (SBP) of 220 mm Hg or lower or diastolic BP (DBP) of 120 mm Hg or lower and no new antihypertensive agent prescribed; (2) SBP higher than 220 mm Hg or DBP higher than 120 mm Hg and a new antihypertensive agent prescribed; (3) DBP higher than 140 mm Hg then IV BP treatment within 2 hours; or (4) SBP of 185 mm Hg or lower, DBP of 110 mm Hg or lower prior to thrombolysis. Patients with the following medical conditions were excluded from the BP analysis: serum creatinine level higher than 2.5 mg/dL; aortic dissection; myocardial infarction; congestive heart failure; or hypertensive encephalopathy. (To convert creatinine to micromoles per liter, multiply by 88.4.)
The treatment of fever, during the entire hospital stay, was evaluated by assessing whether a fever event was addressed with the prescription of acetaminophen. Although no accepted standard definition of fever has been used in studies of body temperature in patients with stroke,9- 11 fever was defined as a temperature of 100°F or higher, based on a comprehensive review of studies of body temperature.12 The fever event ended when the temperature fell below 100°F for at least 24 hours. Patients were categorized as having no fever events or as having received treatment for none, some, or all fever events.
Hypoxia was defined as either an oxygen saturation lower than 90% or arterial oxygen pressure (PaO2) lower than 60 mm Hg. Hypoxia management was measured during the first 14 days of admission and was evaluated by determining whether patients with hypoxia received supplemental oxygen. If the patient had multiple episodes of hypoxia on a given day, the episode with the lowest oxygen saturation or PaO2 was examined. Patients were categorized either as having no episodes of hypoxia or as having received treatment for none, some, or all episodes of hypoxia.
The primary outcome was the combined end point of in-hospital mortality, discharge to hospice, or discharge to a skilled nursing facility. We selected the combined end point because it represents clinically important events; it has been widely used in acute stroke studies; it is reliably obtained from medical records; and evidence suggests that processes of stroke care decrease mortality or institutionalization. Discharge disposition was determined from the medical record. Patients who were discharged to a facility for the purpose of receiving rehabilitation services were classified as being discharged to a rehabilitation facility. Patients discharged to an extended care facility were classified as being discharged to a skilled nursing facility. Patients discharged either to home hospice or to inpatient hospice were classified as discharged to hospice.
The variables included in the risk adjustment model were based on a priori clinical judgment and the existing literature and included (1) age; (2) comorbidity based on medical history (Charlson score)13; (3) concomitant medical illness present at the time of admission (acute myocardial infarction, congestive heart failure, aortic dissection, intraoperative or postoperative stroke, trauma, renal failure, hypertensive encephalopathy, or other major medical problem); (4) preadmission symptom course (worsening preadmission course vs other descriptions of the symptom course); (5) prestroke functional status (independent or no record compared with dependent in any activity of daily living); (6) baseline code status (do not resuscitate [DNR], do not intubate [DNI], or “comfort measures only” compared with full code) ; (7) baseline stroke severity (NIHSS score)7; (8) baseline nonneurologic status score (including abnormalities found on physical examination, chest radiography, and electrocardiography); (9) modified APACHE III (Acute Physiology, Age, Chronic Health Evaluation) score14 (including heart rate, mean BP, body temperature, respiratory rate, PaO2, hematocrit level, white blood cell count, creatinine level, urine output, blood urea nitrogen concentration, and levels of sodium, albumin, bilirubin, and glucose); and (10) admission brain imaging findings categorized according to a modified Oxfordshire classification.15,16 Radiology reports were used to classify imaging results. Images were not reread as part of this study, and radiology results documented in clinicians' notes were not used.
A logistic regression model was developed for the combined outcome and included all risk adjustment variables. A risk adjustment score17 was constructed for each patient using the equation obtained from the logistic regression model. As in examples from other clinical contexts,18 the risk adjustment score was then used for risk adjustment (1 variable) in a multivariable model that assessed the relationship between stroke care processes and outcome.
Descriptive statistics (eg, means [SDs], medians, ranges, and proportions) were used to describe the independent variables, the prevalence of each process of care, and the outcome rate. Unadjusted analyses were used to evaluate the association between each process measure and the combined outcome.
Our goal for the multivariable analysis was to identify the processes of care that were independently associated with the combined outcome, adjusting for patients' clinical characteristics (all studied domains) and stroke care quality (all the other processes of care). We began by modeling the combined outcome and included each of the process measures separately. We subsequently adjusted for all of the process measures and the risk adjustment score. The fever management and hypoxia management variables were entered into the multivariable models using dummy variables coded to retain their ordinal nature (eg, no episodes treated, some episodes treated, all episodes treated).19 This approach was used to adjust both for the presence of the condition as well as for the treatment of the condition because it was expected that patients with either fever or hypoxia would be more likely to experience an adverse event than patients without such conditions. By definition, patients included in the “some episodes treated” category must have had more than 1 episode; therefore, the “some episodes treated” patients likely had greater disease severity than patients in the “all episodes treated” category. No additional adjustments were made for the severity of the fever or hypoxia.
The sample size for this study was designed to provide sufficient patients to conduct the multivariable analysis while maintaining an event per variable ratio higher than 10 to 1 in all multivariable models.20,21 Specifically, given 7 processes of care plus the 1 risk adjustment variable, the sample size was designed to provide at least 80 outcome events.
In separate multivariable analyses, we adjusted for hospital site, restricted the analysis to patients with stroke only, examined the adjusted odds ratios (ORs) modeling the outcome of death and discharge to hospice (without considering discharge to a skilled nursing facility), and excluded patients with DNR, DNI, or “comfort measures only” code status.
Among 1487 patients, 1363 had a definitive stroke (92%); 34 had a definitive TIA (2%); 28 had a stroke vs TIA (2%); and 62 had a probable stroke or TIA (4%). The combined outcome was observed in 239 patients (16%), and 91 patients died in the hospital (6%). Neurologic worsening or death was observed in 208 patients (14%). Table 1 summarizes baseline patient characteristics and outcomes.
A total of 968 data elements were collected for each medical record, with an overall agreement rate of 96%.
The prevalences of the individual processes of care varied from a low of 29% receiving treatment of all episodes of fever to a high of 92% receiving a neurology evaluation (Table 2). Fever was observed in 510 of 1487 patients (34%), with 333 patients having 1 episode, 112 patients having 2 episodes, 37 patients having 3 episodes, and 28 patients having 4 or more episodes of fever.
Oxygen saturation or PaO2 was measured daily for the first 14 days in 1149 of 1487 patients (77%). Hypoxia was observed in 192 of 1487 patients (13%), with 122 patients having 1 episode, 40 patients having 2 episodes, 15 patients having 3 episodes, and 15 patients having 4 or more episodes of hypoxia. Hypoxia was first noted on the day of admission in 29% of patients; on the second hospital day in 20%; on the third hospital day in 12%; and on or after the fourth hospital day in 39% of patients. Among the 192 patients with at least 1 hypoxia episode, 36% had no treatment with supplemental oxygen.
The unadjusted associations between the individual items in the risk adjustment score and the primary outcome are provided in Table 3. Overall, the unadjusted OR for risk score (as a single dimensional variable) and the primary outcome was 2.72 (95%CI 2.41-3.07) per unit increase in risk score.
Three processes of care were independently associated with a reduction in the combined outcome after adjusting for the other processes of care and the risk adjustment score: swallowing evaluation (adjusted OR, 0.64; 95% confidence interval [CI], 0.43-0.94); DVT prophylaxis (adjusted OR, 0.60 95% CI, 0.37-0.96); and treating all episodes of hypoxia with supplemental oxygen (adjusted OR, 0.26; 95% CI, 0.09-0.73) (Table 4).
The results were essentially unchanged when adjusted for hospital site or when the analysis was restricted to patients with stroke (data not shown). The results were also similar when we modeled death or discharge to hospice, although the CIs around the adjusted ORs widened (data not shown). When the analysis was restricted to patients with full code status, the results were very similar (adjusted OR, 0.68; 95% CI, 0.45-1.04), except that the CI around the dysphagia measure included 1.0.
To our knowledge, our study represents a novel evaluation of the association between specific processes of acute stroke care and the combined outcome of in-hospital mortality, discharge to hospice, or discharge to a skilled nursing facility, adjusted for both patient characteristics and stroke care quality. We found that 3 processes of care were independently associated with a reduction in the combined outcome: swallowing evaluation; DVT prophylaxis; and treating all episodes of hypoxia with supplemental oxygen. Because sophisticated structures of care are not needed for the successful implementation of these 3 processes of care, they could be applied across the full spectrum of medical facilities.
Two of the 3 processes (swallowing evaluation and DVT prophylaxis) are similar to existing stroke quality metrics,22 but the treatment of hypoxia is not a current performance measure. Clinicians caring for patients with stroke and organizations interested in evaluating and improving stroke care should consider including a focus on hypoxia treatment in addition to swallowing evaluation and DVT prophylaxis services.
A growing body of evidence suggests that hypoxia is relatively common and is associated with adverse events after stroke.23,24 Our finding of 13% hypoxia prevalence is similar to the 15% hypoxia rate reported by Rocco et al,23 yet our finding that only 77% of patients received daily oxygenation measurement suggests that hypoxia may be underdiagnosed in routine practice. Our finding that only 47% of patients with hypoxia had every episode treated with oxygen indicates that this component of poststroke care can be improved. Additional studies are needed to examine the relationship between the degree and timing of hypoxia and poststroke outcomes, the causes of poststroke hypoxia, and whether stroke subtype influences the effect of hypoxia on outcomes.
Prior research has found that 20% to 50% of patients with stroke have a swallowing abnormality, depending on the detection method used, timing of screening, and stroke severity.25,26 Abnormal swallowing is an independent predictor of chest infection26 and has been associated with a higher risk of death.25 The detection of dysphagia is necessary if patients are to be offered possible therapies, but amidst a controversy about the best way to screen for dysphagia and the difficulty in evaluating dysphagia screening from chart review, several national organizations have recently proposed that dysphagia screening be eliminated from the evaluation of stroke care quality. Given both the robust literature reports and the present results, we recommend that these organizations reconsider including dysphagia screening as a stroke care quality measure.22
Our results confirm the importance of adequate risk adjustment in studies of poststroke outcomes. Although adjustments for age, baseline stroke severity, and brain imaging findings are often performed in stroke outcomes studies, our results highlight the importance of also adjusting for multimorbidity, preadmission symptom course, prestroke functional status, baseline nonneurologic factors, baseline code status, and baseline APACHE III scores. Although some stroke outcome studies adjust for medical comorbidity, we found that concomitant medical problems were more strongly associated with poststroke outcomes than medical history comorbidity. This finding is clinically expected—a patient having a stroke and an acute myocardial infarction is more likely to have an adverse outcome than a patient with a stroke and a medical history of myocardial infarction.
Clinically relevant differences were observed in the associations between processes and outcome in the unadjusted vs adjusted analyses (Table 4). This phenomenon is clinically expected, given that, for example, dysphagia risk increases with worsening stroke severity, receiving a swallowing assessment may be a marker for increasing stroke severity, and increasing stroke severity is associated with poor poststroke outcomes. Accordingly, only by adequate risk adjustment can the beneficial role of swallowing assessment become evident.
Data for this study were obtained from patients who were admitted to a hospital during the period 1998 through 2003, and although the association between the process of care and outcomes may not change over several years, the prevalences of the processes may differ today. In addition, this study focused on processes of stroke care and not structures of care. Structures of care are important to the delivery of quality stroke care, but few structural components of stroke care have been recommended for use as quality measures,5 and the specific structural elements associated with patient outcomes are not known.27 Also, although structural components of care might be in place in a hospital, they might not be available for an individual patient. For example, if a hospital has neurologic expertise in general, but a given patient is admitted on a day when the neurologist is unavailable, then that patient cannot benefit from that structure. In contrast, the actual application of a structure for a patient can be examined (such as whether a neurologist examined a patient).
The determination of whether a process of care was delivered to an individual patient was based on documentation in the medical record, and such data may be incomplete, although the components of care that were the focus of this study should be routinely documented.
Components of care may be associated with improved patient outcomes that differ from the outcomes evaluated in this study. For example, we did not evaluate whether patient education during the acute-care hospital stay would result in improved hospital outcomes. It is also possible that the processes that were the focus of this study are surrogates for other components of care that are associated with the outcome. Although this possibility cannot be eliminated, the processes that were selected to be the focus of this study are likely independent predictors of the primary outcome for several reasons. First, the literature supports an association between each of the processes and the outcome. Second, adjustment for hospital site did not alter the results. Third, the expected temporal relationship between earlier intervention and improved outcome was observed for some processes (eg, the earlier the DVT prophylaxis, the better the protective effect [data not shown]). Fourth, the expected intermediate outcome relationship existed for some processes (eg, patients receiving swallowing evaluation were less likely to have pneumonia [data not shown]).
We sought to include a broad range of hospital types to increase the generalizability of our findings. However, these results may not be generalizable to patients admitted to the hospital from a nursing home or to patients who have a stroke after hospitalization. Given that almost 90% of patients with stroke are admitted from home,11 our results should be generalizable to most patients.
The present study used a combined primary outcome. The relatively low short-term mortality rates after stroke makes risk adjustment difficult, and many studies of stroke therapies therefore include institutionalization as an outcome. Institutionalization is an undesired stroke outcome: patients often rate dependency and institutional care as being a less desirable state than death.28 Death or institutionalization are often combined as a single end point that is both common and the least desirable stroke outcome.29- 32 Functional status is another important patient outcome, but such information is not collected in routine clinical practice, nor is it documented in most medical records. For these reasons, the primary outcome for this study was the combined end point of in-hospital mortality, discharge to hospice, or discharge to a skilled nursing facility. Our research findings remained essentially unchanged, however, when we restricted the analysis to death or discharge to hospice.
When evaluating stroke care quality, most researchers exclude patients who are not ideal candidates for a particular performance measure from the denominator of that measure. In the current analysis, we did not exclude patients from the analysis of the processes of care because our objective was to evaluate the association between processes of care and patient outcomes across the full spectrum of patients with clinical stroke. For example, we did not exclude patients who were restricted to “nothing by mouth” throughout the hospitalization from the evaluation of the dysphagia screening process. Similarly, patients who died within 48 hours of admission were not dropped from the analysis of DVT prophylaxis.
We found that patients with stroke and TIA who received a swallowing evaluation, DVT prophylaxis, and oxygen treatment for all episodes of hypoxia were less likely to have a poor outcome. Accordingly, we recommend that national organizations that establish national performance measures add treatment of hypoxia to their assessment of stroke care quality and continue to measure DVT prophylaxis and swallowing assessment among patients with stroke. In addition, given the heterogeneous and complex characteristics of both patients with stroke and their health care, evaluations of associations between processes of stroke care and patient outcome require detailed risk adjustment.
Correspondence: Dawn M. Bravata, MD, Center of Excellence on Implementing Evidence-Based Practice, Richard L. Roudebush VA Medical Center, HSR&D Mail Code 11H, 1481 W 10th St, Indianapolis, IN 46202 (firstname.lastname@example.org).
Accepted for Publication: November 2, 2009.
Author Contributions:Study concept and design: Bravata, Wells, Lo, Nadeau, Ranjbar, and Concato. Acquisition of data: Bravata, Wells, Melillo, Chodkowski, Struve, McClain, Ranjbar, Tabereaux, and Boice. Analysis and interpretation of data: Bravata, Wells, Lo, Nadeau, Williams, Peixoto, Gorman, Goel, Acompora, Tabereaux, Jacewicz, and Concato. Drafting of the manuscript: Bravata, Wells, and Concato. Critical revision of the manuscript for important intellectual content: Bravata, Wells, Lo, Nadeau, Melillo, Chodkowski, Struve, Williams, Peixoto, Gorman, Goel, Acompora, McClain, Ranjbar, Tabereaux, Boice, Jacewicz, and Concato. Statistical analysis: Bravata and Wells. Obtained funding: Bravata, and Concato. Administrative, technical, and material support: Bravata, Wells, Melillo, Chodkowski, Struve, Williams, Peixoto, Acompora, McClain, Ranjbar, Tabereaux, Boice, and Concato. Study supervision: Bravata, Wells, Gorman, and Concato. Data Quality Assurance: Goel.
Funding/Support: This study was supported by grant Merit IIR-01-104-3 (Dr Bravata) and an Advanced Career Development Award (Dr Bravata) from the Department of Veterans Affairs Health Services Research & Development Service; the Max Patterson Stroke Research Fund at Yale University (Dr Bravata); and the Robert Wood Johnson Generalist Physician Faculty Scholar Program (Dr Bravata).