Variations in the intensive use of head Computed Tomography for elderly hemorrhagic stroke patients

Abstract

Purpose

To investigate the variability in head CT scan use for hemorrhagic stroke patients in US hospitals, its association with mortality, and the number of different physicians seen.

Materials and Methods

The study was approved by the Committee for the Protection of Human Subjects at Dartmouth College. We performed a retrospective analysis of the Medicare fee-for-service claims data for elderly patients admitted for hemorrhagic stroke in 2008–2009, with one-year follow-up through 2010. Risk-adjusted primary outcome measures were average number of head CT scans and high-intensity head CT use (six or more head CT scans in the year after admission). We examined the association of high-intensity head CT use with the number of different physicians seen, and mortality.

Results

A total of 53,272 patients (mean age 79.6 years old, 58.9% female) with hemorrhagic stroke were identified in the study period. The average number of head CT scans in the year after admission for stroke was 3.4; 16.4% of the patients had six or more. Among the highest case volume hospitals, with more than 50 hemorrhagic stroke patients, risk-adjusted rates ranged from 8.0% to 48.1%. The correlation coefficient between number of physicians seen and rates of high-intensity head CT use was 0.522 (P<0.01) for all hospitals, and 0.50 (P<0.01) for the highest volume hospitals. We found no improvement in 1-year mortality for patients undergoing 6 or more head CT scans (OR, 0.84; 95% CI, 0.69 to 1.02).

Conclusions

High rates of head CT use for patients with hemorrhagic stroke are frequently observed, without an association with decreased mortality. Higher number of physicians seen was associated with high-intensity head CT use.

INTRODUCTION

There is growing concern over the potential health effects and rising cost attributed to excessive use of Computed Tomography (CT) scans in the United States.(1) In light of these considerations, a significant body of literature is questioning the use of multiple CT scans in various contexts, including the management of blunt trauma(2–5), seizures(6), and chronic headaches(7). It has been estimated(1) that 0.4% of all cancers nationally may be attributable to radiation from CT studies. Excess CT use also contributes to rising costs, with $14 billion spent in 2006 on imaging by Medicare alone.(8) These concerns helped catalyze the “Choosing Wisely” campaign(9) and the CMS Hospital Quality Initiative(10) to measure and report on potentially avoidable CT use. Current efforts by Medicare(10) focus on the inefficient use of abdominal and thoracic CT studies, as well as head CT scans for sinus disease.

Little is known, however, about the magnitude of variations across hospitals in the avoidable use of CT scans or the underlying causes of overuse. Proposed explanations include defensive medicine, clinician uncertainty, consumer demand, financial incentives, and ownership of facilities.(1, 11–15) Some have also hypothesized that many repeat CT scans are performed because information on prior scans may be unavailable to a clinician at the time of care.(1) If this latter explanation were true, advances in technology, such as health information exchanges, or models of payment that encourage integration and coordination, such as ACOs, might offer an approach to reducing overuse.(16)

This phenomenon can particularly be studied in patients with hemorrhagic stroke, who undergo one or several head CT scans at baseline to guide diagnosis and surgical decision-making. Potential overutilization could expose this population to significant radiation risks. In the current analysis, we used a 2008–2009 cohort of Medicare fee-for-service patients with hemorrhagic stroke, with follow-up data running through 2010, to examine hospital-level variation in the use of head CT scans during the index hospital admission and the year after the stroke. We investigated the variability in head CT scan use for hemorrhagic stroke patients in US hospitals (including a sub-analysis of the highest volume hospitals with more than 50 cases of hemorrhagic stroke), its association with mortality, and the number of different physicians seen.

METHODS

Cohort

The study was approved by the Committee for the Protection of Human Subjects at Dartmouth College. We used data for Medicare beneficiaries aged 65 or more, who were enrolled in fee-for-service programs or non-risk bearing health maintenance organizations from 2008–2009. This is a 100% sample of fee-for-service Medicare claims, provided to the analytic core of The Dartmouth Institute for Health Policy and Clinical Practice by the Centers of Medicare and Medicaid Services. The sample included cases of hemorrhagic stroke classified as primary code 430 (subarachnoid hemorrhage) or 431 (intracerebral hemorrhage) of the International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM). Exclusion criteria included: patients with history of hemorrhagic stroke, transfers from hospitals other than the treating facility, ICD-9-CM codes 873.0–873.9 (traumatic hemorrhage), patients with missing race, ZIP, income, or poverty measures, patients not matched to treating hospitals, and subjects older than 100 years. The latter very small group was excluded to prevent any potential compromise in confidentiality, following the Medicare rule of not reporting data involving less than 11 subjects. This cohort is representative of elderly hemorrhagic stroke patients, since the majority of elderly patients in the US receive coverage from Medicare.

Outcome Variables

The risk-adjusted primary outcome variables were (a) average number of head CT scans per patient over a one-year period after admission for hemorrhagic stroke, where the date of admission is the first day of the one-year period, and (b) high-intensity head CT use over the same one-year period. We defined high-intensity head CT utilization as being above one standard deviation from the mean, which corresponded to 6 or more head CT scans in the year post-admission. The number of head CT scans was determined by identifying the appropriate CPT (Current Procedural Terminology) codes in part B Medicare claims. These claims occur whether the patient is in an inpatient, outpatient, or in the post-acute setting. The secondary outcome variable was risk-adjusted mortality in the year post-admission.

Exposure Variables

Sex-age categories (65–69, 70–74, 75–79, 80–84, 85–99) were created, as well as five ethnicity and race categories (Asian, Black, Hispanic, Native American, and other, with white being the reference). The enrollee’s ZIP code was used to match the 2000 Census data on income and poverty, which in turn was used to create quintiles of income (equal number of patients in every quintile). The actual poverty rate was included to reflect the differing distribution of income within the ZIP code.

Comorbidities, for which outcomes were adjusted (eTable 1), included: myocardial infarction, arrhythmia, congestive heart failure, hyperlipidemia, coagulopathy, hypertension, peripheral vascular disease, lung disease, diabetes, and chronic renal failure. These were used as individual variables, instead of already utilized comorbidity indices, since they provide superior disease-specific risk adjustment. We additionally created a dummy variable for SAH, in order to control for hospitals treating a proportionally high number of patients with this condition, since the clinical course of SAH patients is different than that of the general hemorrhagic stroke population.

Hierarchical condition categories (HCCs) during the 6 months prior to admission were created based on the SAS code provided by the CMS. The HCCs reflect the diagnosis of diseases prior to admission, and are highly predictive of mortality at the individual level. Their use in small-area analysis has been questioned in previous studies because regions with greater intensity of diagnostic testing tend to exhibit higher HCC scores even for patients of similar health status(18, 19) This creates a bias towards finding that more intensive health care leads to better health outcomes.(18, 20) For the purpose of the analysis, HCC scores were divided in quintiles, with equal number of patients in every quintile.

We considered hospitals with patients admitted for hemorrhagic stroke during 2008–09 as our basic unit of observation. We further investigated if the disparities persisted among the highest volume hospitals, by selecting institutions with more than 50 hemorrhagic stroke cases. We analyzed this subgroup separately, to investigate whether high volume hospitals, which we expected would treat a more homogeneous patient population, demonstrate the same variability patterns.

We also investigated the association between number of physicians seen and intensity of head CT use. The number of different physicians seen within a year after the admission for hemorrhagic stroke was determined by examining the National Provider Identifying (NPI) numbers assigned to each physician in part B claims. To address concerns that this metric might be a proxy for sicker patients, we calculated the correlation index between one-year mortality and the number of different physicians seen during the same time. This was −0.0086, which is essentially zero.

Statistical Analysis

A linear regression model with random hospital effects was used to estimate hospital-level risk-adjusted measures of average head CT scans and high-intensity head CT scan use. Best linear unbiased prediction (BLUP) estimates for hospital-specific measures were estimated using STATA 11 (XTMIXED with residuals specified with the “reffects” command).(21) Best linear unbiased predictions shrink hospital-level estimates towards the mean depending on sample sizes, and thus resemble empirical Bayes approaches, but use sample-based estimates of error variances. A linear random-effects probability model (as opposed to a non-linear model) was used to avoid computational instability.

We created scatter plots (and calculated Pearson’s rho) to test whether our measure of fragmentation was correlated with risk-adjusted rates of high-intensity head CT use at the hospital level. These are shown for all US hospitals, and separately for the hospitals with more than 50 hemorrhagic stroke cases.

To identify the association of high-intensity head CT use with mortality we performed a separate logistic regression analysis at the level of the patient with hospital-specific random effects, with mortality within one year of admission being the dependent variable. High-intensity head CT use was a binary variable in this regression. The regression also included as independent variables all comorbidity and age adjustment variables discussed previously. This analysis was restricted to hospitals with at least 25 hemorrhagic stroke patients.

In sensitivity analysis, to avoid spurious associations between the exposure variable and the outcome variable – for example, patients who die within 3 days of admission are unlikely to have as many head CT scans as those surviving the entire year – we create measures of both average number of head CT scans and high-intensity head CT use at the hospital level among those who survived the entire year, again restricted to hospitals with at least 25 stroke patients. We also used a measure of high-intensity head CT use (6 or more head CT scans) limited to those scans performed at the admitting hospital for the index admission. The results of these analyses were nearly identical to our primary analyses, and are therefore not reported here. All probability values are the results of two-sided tests, and the level of significance was set at P < 0.05. SAS 9.4 (SAS Institute, Cary, North Carolina), and Stata 11 (StataCorp, College Station, TX) were used for the analysis.

RESULTS

Hospital variation in one-year head CT utilization

A total of 53,272 patients (mean age 79.6 years old, 58.9% female) with hemorrhagic stroke were identified in the Medicare Claims Data from 2008–2009, after applying the exclusion criteria (eFigure 1, Table 1). The average number of head CT scans in the year after admission for hemorrhagic stroke was 3.4 (95% CI, 3.3 to 3.5), the median was 2 (IQR, 1–4), while the average one-year mortality rate was 56.9% (95% CI, 56.0 to 56.8).

Table 1.

Patient characteristics

Variable	Overall		50-best hospitals	Non 50-best hospitals
	Mean	Standard Deviation	Mean	Mean
Age	79.6	7.7	77.7	79.1*
	%	Standard Deviation	%	%
Race/Ethnicity
Black	8.9	28.5	12.4	8.1*
Hispanic	1.9	13.8	1.5	2.0*
Gender
Female	58.9	49.1	59.7	58.9*
Comorbidities
Arrhythmia	25.2	43.4	22.7	33.4*
Diabetes	19.4	39.6	16.0	25.0
Coagulopathy	4.1	19.8	5.6	1.8*
Myocardial Infarction	21.3	41.0	18.8	28.7
Chronic Renal Failure	7.4	26.2	6.0	10.5*
Hypertension	69.7	45.9	62.0	72.4*
Hyperlipidemia	24.1	42.8	20.4	35.1*
Lung Disease	1.2	10.9	1.5	2.1
Peripheral Vascular Disease	3.9	19.3	3.8	6.0
Congestive Heart Failure	10.5	30.6	10.1	14.8*
	Mean	Standard Deviation	Mean	Mean
Health System Measures
Number of different physicians seen (1 year post-admission)	19.5	16.4	26.7	18.8*
Number of Head CT Scans	3.4	3.1	4.3	3.3*

Notes: SD=Standard Deviation, N = 53,631 for total sample, N = 4,082 for 50-best hospitals, and N = 49,549 for non 50-best hospitals.

^*Denotes significance at p < .05 level in t-test comparison of 50-best with non 50-best.

Figure 1 demonstrates the hospital-level variation of risk-adjusted high-intensity head CT use (six or more head CT scans) in the first year post-admission. The rate of high-intensity head CT use ranged from 12.5% for the 10th percentile hospital to 25.3% for the 90th percentile hospital, with the highest rate equal to 48.1% for Temple University Hospital.

Figure 1.

Histogram demonstrating the ratio of risk-adjusted high-intensity head CT use (six or more head CT scans) among US hospitals

Considerable variation was shown among the highest hemorrhagic stroke volume hospitals. As shown in Table 2, Hartford Hospital, University Hospitals Case Medical Center, Emory University Hospital, and the Mayo Clinic demonstrated risk-adjusted rates of average head CT scans (2.8, 3.3, 2.9, and 3.2 respectively), as well as risk-adjusted rates of high-intensity head CT use (9.6%, 10.1%, 12.9%, and 14.5% respectively) that are below the national average for patients with hemorrhagic stroke. Other high hemorrhagic stroke volume hospitals, such as Northwestern Memorial Hospital, University of Illinois Medical Center at Chicago, and Temple University Hospital exhibited significantly higher risk-adjusted average rates of head CT scans (7.4, 7.5, and 6.9 respectively) and risk-adjusted rates of high-intensity use (39.5%, 41.4% and 48.1%, respectively). Variability for high-intensity use was observed even within the same city: for example between Beth Israel Deaconess Medical Center and the Brigham and Women’s Hospital (18.7% versus 27.8%).

Table 2.

Head CT utilization for hemorrhagic stroke in the 50-best hospitals in Neurology and Neurosurgery based on the 2012 report by the US News and World Report (Hospitals are ranked alphabetically)

50-best hospitals in Neurology and Neurosurgery	Average number of head CT	Lower 95% CI	Higher 95% CI	Percentage of high-intensity head CT use	Lower 95% CI	Higher 95% CI
Abbott Northwestern Hospital	3.8	3.2	4.4	16.4	9.1	23.6
Aurora St. Luke’s Medical Center	5.1	4.7	5.6	24.9	19.1	30.7
Barnes-Jewish Hospital/Washington University	4.7	4.3	5.2	23.3	18.2	28.5
Beaumont Hospital - Royal Oak	4.7	4.1	5.2	24.7	18.4	30.9
Brigham and Women’s Hospital	5.5	5.0	5.9	27.8	22.1	33.4
Cedars-Sinai Medical Center	4.3	3.7	4.9	20.1	13.4	26.9
Clarian Health Partners	4.4	4.0	4.8	18.7	14.2	23.2
Cleveland Clinic Foundation	4.9	4.5	5.4	23.1	17.5	28.8
Duke University Hospital	3.5	3.0	4.1	13.4	7.0	19.9
Emory University Hospital	3.9	3.4	4.3	12.8	7.5	18.2
Hackensack University Med Center	3.7	3.0	4.3	19.1	12.0	26.2
Hospital of the University of Penn.	5.7	5.0	6.3	24.8	17.0	32.6
Johns Hopkins Hospital	5.8	5.2	6.5	26.5	19.3	33.6
Massachusetts General Hospital	4.6	4.2	4.9	23.1	18.6	27.7
Mayo Clinic/St Mary’s Hospital	3.7	3.2	4.2	14.5	8.6	20.4
Mount Sinai Medical Center	3.7	2.9	4.5	13.3	4.2	22.5
Nebraska Medical Center	4.6	3.9	5.2	22.5	15.1	29.9
New York-Presbyterian Hospital	5.0	4.6	5.5	25.6	19.8	31.4
North Carolina Baptist Hospital	4.4	3.8	5.0	18.4	11.4	25.4
North Shore University Hospital	5.3	4.9	5.7	26.3	21.2	31.3
Northwestern Memorial Hospital	7.9	7.3	8.5	39.4	32.6	46.2
NYU Medical Center	5.4	4.5	6.2	34.5	25.2	43.9
Ochsner Clinic Foundation (LA)	4.5	3.7	5.3	21.4	12.4	30.3
Rush University Medical Center	5.1	4.6	5.5	25.4	20.3	30.5
Shands at the University of Florida	3.7	3.2	4.2	15.3	9.6	21.1
St. Joseph’s Hospital & Medical Center	5.3	4.9	5.7	29.3	24.3	34.3
St. Vincent Indianapolis Hospital	3.8	3.2	4.3	14.8	8.1	21.5
Stanford Hospital and Clinics	4.1	3.5	4.7	18.0	10.7	25.2
Strong Memorial Hospital of University of Rochester	4.0	3.4	4.7	17.0	9.5	24.6
Tampa General Hospital	5.1	4.6	5.5	25.7	20.5	31.0
The Methodist Hospital	4.7	4.2	5.2	23.3	17.3	29.2
UCLA Medical Center	4.7	4.0	5.3	24.4	15.0	29.9
UCSF Medical Center	4.0	3.5	4.5	13.9	8.0	19.8
University Hospitals Case Medical Center	3.8	3.3	4.3	10.1	4.2	16.0
University of Chicago Hospital	5.9	5.2	6.6	25.1	17.2	33.1
University of Iowa Hospitals & Clinics	4.8	4.4	5.1	24.1	19.6	28.6
University of Kansas Hospital	5.7	5.0	6.4	22.4	14.3	30.5
University of Maryland Med Center	5.3	4.8	5.7	25.5	19.8	31.3
University of Michigan Hospitals	5.6	4.9	6.2	29.3	21.9	36.8
University of Wisconsin Hospital & Clinics	4.4	3.9	5.0	23.2	16.9	29.4
UPMC Mercy	4.2	3.5	4.9	19.0	10.8	27.1
UT Southwestern Zale Lipshy Hospital	6.5	5.8	7.2	30.6	22.5	38.7
Vanderbilt University Medical Center	3.6	3.0	4.1	15.7	9.2	22.1

Notes: Hospitals in Bold have an average number of head CTs above the mean for the 50-best, and those in Italics have an average number of head CTs below the mean. The Table includes 43 hospitals. Harper University Hospital, St. Francis Hospital, St. Luke’s Hospital, USC University Hospital, U of Minnesota Hospital, U of Washington Hospital had less than 25 observations and where therefore excluded from the analysis. The Hospital for Special Surgery, although included in the 50-best Neurology and Neurosurgery programs by the US News and World Report, does not treat any stroke patients. High-intensity CT use=6 or more head CTs in the year after admission for hemorrhagic stroke; CT=Computed Tomography; CI=Confidence Interval; UCSF=University of California in San Francisco; USC=University of Southern California; UCLA=University of California in Los Angeles; UPMC=University of Pittsburg Medical Center

We found similar patterns of variability in high-intensity head CT use limited to the cohort who survived the entire year, with a correlation coefficient of 0.89 (P < 0.01) with the entire cohort. We also considered the sensitivity of our results to using only the number of head CT scans performed during the initial index admission. While the fraction of patients with 6 or more head CT scans was obviously lower, the rankings of the hospitals were similar.

Number of different physicians seen

Medicare enrollees visited on average 19 different physicians in the year after a hemorrhagic stroke, with variation across hospitals ranging from 12.8 to 28.1. The association between the average number of physicians seen and high-intensity head CT use was positive and significant in the sample of 819 U.S. hospitals (Figure 2) with at least 25 hemorrhagic stroke patients (overall rho=0.52, P<0.01). There was also a strong positive association between the level of number of physicians seen and the risk-adjusted rates of high-intensity head CT use among the highest volume hospitals (rho=0.50, P<0.01).

Figure 2.

Scatter plot demonstrating the association between risk-adjusted high-intensity CT use (six or more head CT scans) and risk-adjusted number of different physicians seen among hospitals with the highest case volume of hemorrhagic stroke (Pearson’s rho=0.5, P < 0.01)

One-year risk-adjusted mortality

The association of one-year mortality with hospital-level high-intensity head CT use was evaluated in a regression analysis controlling for socioeconomic factors, comorbidities, age, quintiles of HCC score, and race (eTable 2). As expected, increasing age (OR, 2.56; 95% CI, 2.36 to 2.78, for age over 85, relative to 65–69) was associated with higher mortality. The opposite was true for higher income levels (OR, 0.89; 95% CI, 0.83 to 0.96, for the highest quintile of income in comparison to the lowest). Finally, there was no significant difference in mortality with respect to high-intensity head CT use (OR, 0.84; 95% CI, 0.69 to 1.02). Because of the well-known biases associated with the use of HCC risk adjustment,(18) we also estimated the model without HCC measures (eTable 3), again failing the reject the null hypothesis (OR, 1.06; 95% CI, 0.89 to 1.27).

DISCUSSION

In this retrospective study of Medicare enrollees, we demonstrated wide hospital-level variation in the rate of high-intensity head CT use for patients with hemorrhagic stroke. High-intensity was associated with greater numbers of different physicians treating each patient, but was not associated with improved survival. These results underscore the importance of initiatives to monitor the overuse of expensive and potentially harmful imaging(9, 10). This is of particular significance for our cohort of patients with hemorrhagic stroke, who undergo multiple CT scans.

Hospitals throughout the United States were characterized by different intensity of head CT scan use for similar patients. A 6-fold variation in the use of intensive head CT scanning across hospitals was present, even after adjusting for differences across hospitals in patient health status, and after shrinkage methods were used to reduce the degree of “overfitting” in regression analysis. These disparities persisted among the highest volume centers. In particular, patients with hemorrhagic stroke admitted at some hospitals had on average twice the number of head CT studies in comparison to other highly regarded institutions such as UCSF and Mayo Clinic, highlighting the lack of universal practices in regards to CT utilization in this patient population.

Higher average number of different physicians seen was associated with increased high-intensity head CT use, both nationwide and among the highest volume hospitals. There can be various explanations for this association. Although, our results were risk-adjusted for several comorbidities, including HCC score, there is a possibility that the care of sicker patients necessitated more consultations, and eventually more head CT scans. Another plausible explanation would be the lack of coordinated care in the hospital systems with more physicians seen, resulting in more head CT scans.(22) This is a challenge that has been identified as a driver of health care costs.(23) Whether the adoption of value-based payment models for services would improve coordination and minimize unnecessary diagnostic testing, however, is not yet known.(23)

Improving the continuity of care, and minimizing the number of different physicians overseeing patient care, is a recognized goal of several current health care reforms, such as Accountable Care Organizations (ACOs). In the ACO setting, the health care providers within the ACO would be held accountable for the entire year of post-discharge services such as head CT scans, regardless of which institution actually performs the services.(24) Hospitals (or ACOs) may be able to reduce the unnecessary use of post-acute head CT scans by introducing case managers, discharge planning, electronic medical records, and evidence-based neuro-rehabilitation guidelines.(25, 26) O’Toole et al(25) have demonstrated that system fragmentation, gaps in human and financial resources, and complexity at the inter-organizational and operational levels are common barriers for effective resource utilization. A holistic approach to the patient with frequent post-discharge phone calls and home-visits by the same group of physicians could minimize such fragmentation.

Increased high-intensity head CT use was not associated with lower mortality in the year after admission for hemorrhagic stroke, a result that is consistent with other studies showing weak or insignificant associations between discretionary medical treatment intensity and health outcomes.(27–29) To address concerns that HCCs tend to over-adjust for risk in higher intensity hospitals,(18) we repeated the analysis without including the HCCs as covariates. There was still no association of high-intensity head CT with mortality.

From a clinical perspective, it is important to recognize that this analysis does not address the question of how the amount of care for an individual patient in a specific case would affect the patient’s clinical outcome. Nor does our study indicate whether it is possible to reduce overutilization and spending. However, if the United States as a whole could safely achieve intensity levels comparable to those of the lowest-utilizing regions, significant cost saving could potentially be realized. Further research is needed to investigate whether such an impact is possible.

The present study has several limitations. First, indication bias (a potential confusion between cause and effect when exposure is dependent on indication) and residual confounding (variables introducing bias that the current study could not control for) could account for some of the observed associations. In addition, coding inaccuracies can affect our estimates, although several reports have demonstrated that coding for hemorrhagic stroke has shown good association with medical record review.(30, 31) Second, our data is based on the Medicare population, with potentially different results for the commercially insured. Third, we were not able to test the hypothesis that local factors, such as the malpractice environment affected the propensity to perform additional head CT scans. We found variation in utilization within cities or states, despite common state malpractice statutes.

Fourth, although using the number of different physicians seen as a proxy for continuity of care is imperfect, it has been adapted from other prior studies(22, 36, 37) and appears to reflect the fundamental challenges of coordinating care across multiple physicians. Fifth, we cannot control perfectly for the severity of the hemorrhagic stroke, since measures such as the Glasgow Coma Scale and the NIH stroke scale are not available. However, we have risk-adjusted for socioeconomic factors that are associated with disease severity, and for HCCs that account for patient health status and are highly predictive of mortality. Despite this, we were still not able to examine the correlation of high-intensity head CT use with improved neurologic function in 1 year, and other functional outcomes. Sixth, we could not account for the proximity of the patient to a CT scanner, while they were hospitalized, or the overall use of that scanner. Finally, causality cannot easily be established based on ecologic data. Although the variation described here is strongly suggestive of the inefficient use of resources, it is challenging to predict the effect of different policy initiatives on utilization patterns and continuity of care.

Conclusions

Patients with hemorrhagic stroke are frequently undergoing several head CT scans as part of their diagnostic workup and follow-up. Inefficient utilization can have detrimental effects on population-health and cost.(38) We demonstrate significant variability in high-intensity head CT use among all US hospitals, and among the highest volume hospitals in hemorrhagic stroke cases. Higher average number of different physicians seen was positively associated with high-intensity head CT use. Increased utilization however did not correlate with improved mortality in our patient population. That there is so much variation in high-intensity head CT use suggests a need for coordinated efforts to minimize disparities.

Advances in Knowledge.

1. Hospitals throughout the United States are characterized by different intensity of head CT scan use (up to 6-fold variation) for similar hemorrhagic stroke patients.

2. These disparities persisted among the highest hemorrhagic stroke volume centers, with risk-adjusted rates of utilizing more than 6 head CT scans in 1 year ranging from 8.0% to 48.1%.

3. Higher average number of different physicians seen was associated with increased high-intensity head CT use, both nationwide and among the highest volume hospitals.

4. There was no improvement in 1-year mortality for patients undergoing 6 or more head CT scans (OR, 0.84; 95% CI, 0.69 to 1.02).

Implications in Patient Care.

1. Coordinated efforts are needed to minimize disparities in the use of head CT scans in patients with hemorrhagic stroke.

Summary Statement.

High rates of head CT use for patients with hemorrhagic stroke are frequently observed, and while they demonstrate an association with higher number of physicians seen, they are not associated with mortality.