Figures
Abstract
Despite recent advances in care, stroke remains a life-threatening disease. Little is known about current hospital mortality with stroke and how it varies by hospital in a national clinical setting in Japan. Using the Diagnosis Procedure Combination database (a national inpatient database in Japan), we identified patients aged ≥20 years who were admitted to the hospital with a primary diagnosis of stroke within 3 days of stroke onset from April 2012 to March 2013. We constructed a multivariable logistic regression model to predict in-hospital death for each patient with patient-level factors, including age, sex, type of stroke, Japan Coma Scale, and modified Rankin Scale. We defined risk-standardized mortality ratio as the ratio of the actual number of in-hospital deaths to the expected number of such deaths for each hospital. A hospital-level multivariable linear regression was modeled to analyze the association between risk-standardized mortality ratio and hospital-level factors. We performed a patient-level Cox regression analysis to examine the association of in-hospital death with both patient-level and hospital-level factors. Of 176,753 eligible patients from 894 hospitals, overall in-hospital mortality was 10.8%. The risk-standardized mortality ratio for stroke varied widely among the hospitals; the proportions of hospitals with risk-standardized mortality ratio categories of ≤0.50, 0.51–1.00, 1.01–1.50, 1.51–2.00, and >2.00 were 3.9%, 47.9%, 41.4%, 5.2%, and 1.5%, respectively. Academic status, presence of a stroke care unit, higher hospital volume and availability of endovascular therapy had a significantly lower risk-standardized mortality ratio; distance from the patient’s residence to the hospital was not associated with the risk-standardized mortality ratio. Our results suggest that stroke-ready hospitals play an important role in improving stroke mortality in Japan.
Citation: Matsui H, Fushimi K, Yasunaga H (2015) Variation in Risk-Standardized Mortality of Stroke among Hospitals in Japan. PLoS ONE 10(10): e0139216. https://doi.org/10.1371/journal.pone.0139216
Editor: Yoshiaki Taniyama, Osaka University Graduate School of Medicine, JAPAN
Received: June 25, 2015; Accepted: September 10, 2015; Published: October 7, 2015
Copyright: © 2015 Matsui et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited
Data Availability: Data cannot be made publicly available for ethical reasons as the data are patient data. The data are available to interested researchers upon request to the corresponding author, pending ethical approval.
Funding: This work was supported by a grant for Research on Policy Planning and Evaluation from the Ministry of Health, Labour and Welfare, Japan (grant number: H27-Policy-Designated-009). The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.
Competing interests: The authors have declared that no competing interests exist.
Introduction
Stroke remains a potentially life-threatening disease despite recent advances in stroke management, including intravenous thrombolysis with recombinant tissue plasminogen activator [1] and endovascular therapy (thrombectomy and thrombolysis) for ischemic stroke [2]. Poor prognosis of stroke has been found to be related to various patient backgrounds, such as age, consciousness level on admission [3–6], and modified Rankin Scale on admission [7].
Studies have shown access to and mortality after stroke care to be associated with several hospital-level factors, including hospital volume, type of hospital [8], presence of a stroke care unit [9,10], and distance from hospital to home [11,12]. Regarding hospital volume, reports have indicated mixed results about the volume-outcome relationship for stroke [13–17]. However, hospital mortality of stroke in those studies varied widely owing to the small sample sizes from limited geographic areas. Furthermore, the studies lacked sufficient adjustment for the severity of stroke. Little is known about contemporary hospital mortality in stroke in a nationwide clinical setting and how it varies by hospital in Japan.
Assessing the variation in stroke mortality among hospitals is important in promoting optimal quality of care. In the present study, we used hospital-level risk-standardized mortality ratios (RSMRs) after stroke [18,19] with a national database in Japan. With this approach, we compared stroke mortality among hospitals adjusted for patient-level prognostic factors. We also assessed several hospital-level factors that affect stoke mortality.
Materials and Methods
Data Source
The Japanese Diagnosis Procedure Combination database is a national database of in-hospital patients in Japan. Details of the database have been described elsewhere [20]. In brief, the database contains administrative claims and discharge abstract data including the following: unique hospital identifiers; 7-digit postal codes of the patients’ residential areas; patients’ age and sex; diagnoses and comorbidities on admission coded according to the International Classification of Diseases and Related Health Problems, 10th Revision; procedures; Japan Coma Scale (JCS) on admission; modified Rankin Scale (mRS) on admission for stroke patients; days from stroke onset to hospital admission (≤3 days, 4–7 days, or ≥8 days); length of stay; and discharge status. As of 2013, the database included data on approximately 7 million inpatients from more than 1,000 hospitals, representing approximately 50% of all acute-care inpatient hospitalizations in Japan. All the academic hospitals were obliged to participate in the database, but participation of non-academic community hospitals was voluntary.
We also used data from the Survey of Medical Institutions, which is conducted every year by the Ministry of Health, Labour and Welfare in Japan.[21] The Survey of Medical Institutions is a census of hospitals in Japan, including hospital structural information such as the presence of specific physicians in each hospital.
This study was approved by the Institutional Review Board of The University of Tokyo. Because of the anonymous nature of the data, the requirement for informed consent was waived.
Patient Selection and Data
We included patients aged ≥20 years who were admitted to hospital with a primary diagnosis of stroke within 3 days of stroke onset during the 12 months from April 1, 2012 to March 31, 2013. The International Classification of Diseases and Related Health Problems, 10th Revision, codes I60, I61, and I63 were used to determine the diagnoses of subarachnoid hemorrhage, cerebral hemorrhage, and cerebral infarction, respectively. For a subset analysis, we excluded patients with subarachnoid hemorrhage and focused on those with cerebral hemorrhage or cerebral infarction.
Patient-level data included age, sex, type of stroke (subarachnoid hemorrhage, cerebral hemorrhage, or cerebral infarction), JCS on admission, mRS on admission, length of stay, and in-hospital death. The JCS is widely used in Japan to measure impaired consciousness [22]. A JCS score of 0 indicates alert consciousness. Single-digit scores (1, 2, 3) signify patients who are awake without any stimuli. Double-digit scores (10, 20, 30) denote patients who can be aroused by some stimuli. Triple-digit scores (100, 200, 300) indicate coma. The JCS and Glasgow Coma Scale assessments are well correlated [23]. The mRS on admission consisted of the following: 0 (no symptoms at all); 1 (no significant disability despite symptoms—able to carry out all usual duties and activities); 2 (slight disability—unable to carry out all previous activities, but able to look after own affairs without assistance); 3 (moderate disability—require some help, but able to walk without assistance); 4 (moderately severe disability—unable to walk without assistance and unable to attend to own bodily needs without assistance); and 5 (severe disability—bedridden, incontinent, and requiring constant nursing care and attention) [7].
The hospital-level characteristics included hospital volume, hospital type (academic or non-academic hospital), presence of a stroke care unit, availability of endovascular therapy (thrombectomy and thrombolysis) and presence of neurologists. Hospital volume was defined as the number of patients with stroke treated at an individual facility during the study period. In the Diagnosis Procedure Combination database, we checked the availability of endovascular therapy for each hospital. Data on the presence of neurologists in each hospital was obtained from the Survey of Medical Institutions data.
Distance from Patient’s Residence to Hospital
Full addresses of the hospitals and postal codes of the patients’ residential areas were converted to latitudes and longitudes with address-match geocoding using ArcGIS software, version 10.3 (Esri Inc., Redlands, CA, USA). The area of the Japanese archipelago is approximately 378 000 km2. About two-thirds of which is uninhabited mountainous terrain, and the inhabited area is roughly 121 000 km2. There are roughly 140 000 postal codes in Japan (unpublished data). A postal code covers approximately 1 km2 on average. The measurements were taken from the center of the postal code to the hospital.
The distance from the patient’s residence to the hospital was calculated using Hubeny’s distance calculation formula [24]. For patient-level analyses, the distance from the patient’s residence to the hospital (in kilometers) was categorized into tertiles. For hospital-level analyses, the median of the distance from the patients’ residences to each hospital was calculated.
Risk-Standardized Mortality Ratio
First, we constructed a multivariable logistic regression model to predict in-hospital death for each patient with the following patient-level factors: age (as a continuous variable); sex; type of stroke (subarachnoid hemorrhage, cerebral hemorrhage, or cerebral infarction); JCS on admission; and mRS on admission. The predicted probability of death for each patient (ranging from 0 to 1) was calculated using a logistic regression model. We determined the prediction accuracy of the logistic model using the c-index (area under receiver operating curve) [19,25]. A c-index value of 0.5 suggests that the model is no better than random chance in predicting death; a value of 1.0 indicates perfect discrimination.
In this study, RSMR was defined as the ratio of the actual number of in-hospital deaths to the expected number of such deaths for each hospital [18,19]. The expected number of in-hospital deaths was equal to the sum of the predicted probabilities of the patients in each hospital. RSMR was calculated only for facilities with a hospital volume of ≥50.
Statistical Analyses
We employed chi-square tests to compare crude in-hospital mortality between the categories of each patient-level factor. The averages and standard deviations of RSMR were compared between the categories of each hospital-level factor using t tests or analyses of variance. A hospital-level multivariable linear regression was modeled to analyze the association between RSMRs and hospital-level factors (type of hospital, presence of stroke care unit, presence of neurologists, availability of endovascular therapy, hospital volume, and median distance from the patient’s residence to the hospital). Because all hospitals with stroke care unit had neurologists, the variables “stroke care unit” and “presence of neurologists” were merged to a single variable with the following three categories: “absence of neurologists”, “presence of neurologists, without stroke care unit” and “presence of neurologists, with stroke care unit”. Availability of endovascular therapy and hospital volume were merged to a single variable including the following four categories: “hospital volume ≤399, endovascular therapy unavailable”, “hospital volume ≥400, endovascular therapy unavailable”, “hospital volume ≤399, endovascular therapy available” and “hospital volume ≥400, endovascular therapy available”. We also performed a patient-level Cox regression analysis to examine the association of in-hospital death with both patient-level and hospital-level factors. We considered a two-sided p value of less than 0.05 significant. For all the statistical analyses, we employed Statistical Package for Social Sciences version 22.0 (IBM SPSS Corp., Armonk, NY, USA).
Results
During the study period, we identified 176,753 eligible patients from 894 hospitals. The average (standard deviation) age was 72.8 (13.1) years. Table 1 shows the patient-level characteristics and in-hospital mortality for each category. Overall crude in-hospital mortality was 10.8%. Higher crude in-hospital mortality was significantly associated with the following: higher age; female gender; hemorrhagic stroke; higher JCS; and higher mRS. Non-academic hospitals, hospitals with stroke care units, and higher-volume hospitals (≥600) were more likely to have lower crude in-hospital mortality. Patients with a greater distance from their residence to the hospital were significantly likely to have higher crude in-hospital mortality.
Table 2 shows the RSMR for each category of hospital-level factors in facilities with a hospital volume of ≥50. The RSMR was significantly lower in academic hospitals, hospitals with a stroke care unit, higher-volume hospitals and hospitals in which endovascular therapy was available. The cut-off value for a median distance from the patient’s residence to the facility was set as 4.3 km so that the number of hospitals were equal between the categories (349 hospitals with median distance ≤4.3 km and 349 hospitals with median distance >4.3 km, excluding 26 hospitals with missing data on median distance). No significant difference in RSMR was evident between hospitals with a median distance from the patient’s residence to the facility of less than 4.3 km and those with a distance greater than that.
Fig 1 is a scatter diagram of hospital volume and RSMR for the 724 facilities with a hospital volume of ≥50. The numbers of hospitals in the RSMR categories of ≤0.50, 0.51–1.00, 1.01–1.50, 1.51–2.00, and >2.00 were 28 (3.9%), 347 (47.9%), 300 (41.4%), 38 (5.2%), and 11 (1.5%), respectively.
Table 3 shows the results of the hospital-level multivariable linear regression analysis for RSMR. With adjustment for other variables, academic hospitals and hospitals with a stroke care unit were significantly likely to have lower RSMRs. Higher hospital volume and availability of endovascular therapy were significantly associated with lower RSMRs. The median distance from the patient’s residence to hospital was not significantly associated with the RSMR.
Table 4 shows the results of the Cox regression analysis for in-hospital death (n = 171,194). Patients with higher age, male gender, hemorrhagic stroke, higher JCS on admission, or higher mRS were significantly more likely to die during hospitalization. No significant association was found between in-hospital mortality and distance from the patient’s residence to hospital. After adjustment for these variables, the hazard ratio for the group with a hospital volume of ≥600 was 0.91 (95% confidence interval, 0.86–0.96; p = 0.001) compared with the reference group (hospital volume ≤199).
The results of the subset analysis that focused on those with cerebral hemorrhage or infarction were shown in the supporting tables (S1 Table, S2 Table and S3 Table). These results were almost similar to those in the all-patient analysis.
Discussion
Using a national inpatient database, this study found that RSMRs varied significantly among hospitals in Japan. Approximately 4% of hospitals had an RSMR of ≤0.50, and approximately 8% of hospitals had an RSMR of >1.50. Academic hospitals, the presence of a stroke care unit, higher hospital volume and the availability of endovascular therapy were associated with lower RSMRs, whereas the distance from the patient’s residence to hospital was not associated with the RSMR.
The present study has several strengths. First, the national inpatient database enabled us to conduct a larger study than previous investigations that assessed the hospital-level factors affecting stroke mortality. Second, this study determined the variation in stroke mortality among hospitals and related factors in a nationwide setting with standardization of patient-level risks.
Previous studies have found that various patient-level prognostic factors affect hospital mortality of stroke. The mRS is a well-known predictor of early mortality for stroke patients [7]. One US study determined that poor prognostic factors for mortality included advanced age, low Glasgow Coma Score, and infratentorial location [4]. In another study, the independent predictors of mortality included male gender, advanced age, hemorrhage location, and decreased consciousness level [5]. The results of the present study are mostly consistent with those of previous studies.
The most notable finding in the present study was a significant disparity in the RSMR of stroke among hospitals. This result can be used as a benchmark for hospitals in seeking better stroke management for improved outcome.
Another important finding of this study was the association between RSMR for stroke and several hospital-level factors. A number of studies have identified a relationship between higher hospital volume and lower stroke mortality [13–17]. However, most reports have lacked sufficient adjustment for patient backgrounds and were based on small sample sizes from limited geographic areas. A recent study using the Diagnostic Procedure Combination database found an association between higher hospital volume and better patient outcomes for ischemic stroke after adjustment for stroke severity [26]. The present investigation determined that variation in RSMR for stroke was partly explained by hospital volume.
In a previous study, academic hospitals were found to have slightly more favorable hospital-level performance in risk-adjusted outcomes [8]. Several reports have shown the presence of a stroke care unit to be associated with a decrease in stroke mortality [9,10]. Our findings confirm superior outcomes in academic hospitals and stroke care units in a nationwide clinical setting.
Our results showed that the availability of endovascular therapy was significantly associated with reduced RSMR. A previous study showed that the majority of patients with acute ischemic stroke did not receive endovascular therapy owing to lack of availability and limited indications for this therapy [26]. Our results suggest that the availability of endovascular therapy is an important part for stroke readiness and for reducing stroke mortality.
Several limitations of the present study should be noted. First, the database only included 50% of acute-care inpatients in Japan. Participating hospitals were skewed to large hospitals because participation of non-academic smaller hospitals was voluntary. Second, although we included a large population, this was a retrospective observational database study in which we could not control the collection or reporting of the variables we were measuring. There could be bias in the form of unmeasured confounders. We were unable to identify several important factors, including National Institute Health Stroke Scale and imaging results (e.g., size of infarction), owing to lack of data availability. Third, the database does not have strict data on discharge to independent living or into care.
In conclusion, this study found that the RSMR for stroke varied widely among hospitals in Japan. Academic status, presence of a stroke care unit, hospital volume and availability of endovascular therapy were significant predictors of lower RSMRs. These results suggest that the availability of stroke-ready hospitals plays an important role in improving mortality.
Supporting Information
S1 Table. Number of patients and in-hospital mortality in each category for those with cerebral infarction or hemorrhage.
https://doi.org/10.1371/journal.pone.0139216.s001
(DOCX)
S2 Table. Hospital-level characteristics and RSMR for cerebral infarction or hemorrhage.
https://doi.org/10.1371/journal.pone.0139216.s002
(DOCX)
S3 Table. Hospital-level multivariable linear regression for risk standardized mortality ratio for cerebral infarction or hemorrhage (n = 724).
https://doi.org/10.1371/journal.pone.0139216.s003
(DOCX)
Acknowledgments
This work was supported by a grant for Research on Policy Planning and Evaluation from the Ministry of Health, Labour and Welfare, Japan (grant number: H27-Policy-Designated–009).
Author Contributions
Conceived and designed the experiments: HM KF HY. Performed the experiments: HM KF HY. Analyzed the data: HM KF HY. Contributed reagents/materials/analysis tools: HM KF HY. Wrote the paper: HM KF HY.
References
- 1. Hacke W, Kaste M, Bluhmki E, Brozman M, Dávalos A, Guidetti D, et al. Thrombolysis with alteplase 3 to 4.5 hours after acute ischemic stroke. N Engl J Med. 2008;359: 1317–1329. pmid:18815396
- 2. Goyal M, Demchuk AM, Menon BK, Eesa M, Rempel JL, Thornton J, et al. Randomized assessment of rapid endovascular treatment of ischemic stroke. N Engl J Med. 2015;372: 1019–1030. pmid:25671798
- 3. Ruiz-Sandoval JL, Chiquete E, Romero-Vargas S, Padilla-Martínez JJ, González-Cornejo S. Grading scale for prediction of outcome in primary intracerebral hemorrhages. Stroke. 2007;38: 1641–164. pmid:17379820
- 4. Hemphill JC 3rd, Bonovich DC, Besmertis L, Manley GT, Johnston SC. The intracerebral hemorrhage score: a simple, reliable grading scale for intracerebral hemorrhage. Stroke 2001;32: 891–897. pmid:11283388
- 5. Zia E, Engstrom G, Svensson PJ, Norrving B, Pessah-Rasmussen H. Three-year survival and stroke recurrence rates in patients with primary intracerebral hemorrhage. Stroke 2009;40: 3567–3573. pmid:19729603
- 6. Heuschmann PU, Kolominsky-Rabas PL, Misselwitz B, Hermanek P, Leffmann C, Janzen RW, et al. Predictors of in-hospital mortality and attributable risks of death after ischemic stroke: the German Stroke Registers Study Group. Arch Intern Med. 2004;164: 1761–178. pmid:15364669
- 7. Hankey GJ. Long-term outcome after ischaemic stroke/transient ischaemic attack. Cerebrovasc Dis. 2003;16 (suppl 1): 14–19. pmid:12698014
- 8. Fonarow GC, Smith EE, Reeves MJ, Pan W, Olson D, Hernandez AF, et al. Hospital-level variation in mortality and rehospitalization for Medicare beneficiaries with acute ischemic stroke. Stroke. 2011;42: 159–166. pmid:21164109
- 9. Tamm A, Siddiqui M, Shuaib A, Butcher K, Jassal R, Muratoglu M, et al. Impact of stroke care unit on patient outcomes in a community hospital. Stroke 2014;45: 211–216. pmid:24253547
- 10. Inoue T, Fushimi K. Stroke care units versus general medical wards for acute management of stroke in Japan. Stroke. 2013;44: 3142–3147. pmid:23988645
- 11. Acharya AB, Nyirenda JC, Higgs GB, Bloomfield MS, Cruz-Flores S, Connor LT, et al. Distance from home to hospital and thrombolytic utilization for acute ischemic stroke. J Stroke Cerebrovasc Dis. 2011;20: 295–301. pmid:20634096
- 12. El Khoury R, Jung R, Nanda A, Sila C, Abraham MG, Castonguay AC, et al. Overview of key factors in improving access to acute stroke care. Neurology. 2012;79 (13 Suppl 1): S26–34. pmid:23008407
- 13. Reed SD, Cramer SC, Blough DK, Meyer K, Jarvik JG. Treatment with tissue plasminogen activator and inpatient mortality rates for patients with ischemic stroke treated in community hospitals. Stroke 2001;32: 1832–1840. pmid:11486113
- 14. Heuschmann PU, Kolominsky-Rabas PL, Misselwitz B, Hermanek P, Leffmann C, Janzen RW, et al. Predictors of in-hospital mortality and attributable risks of death after ischemic stroke: the German Stroke Registers Study Group. Arch Intern Med. 2004;164: 1761–1768. pmid:15364669
- 15. Votruba ME, Cebul RD. Redirecting patients to improve stroke outcomes: implications of a volume-based approach in one urban market. Med Care. 2006;44: 1129–1136 pmid:17122718
- 16. Saposnik G, Baibergenova A, O'Donnell M, Hill MD, Kapral MK, Hachinski V, et al. Hospital volume and stroke outcome: does it matter? Neurology. 2007;69: 1142–1151. pmid:17634420
- 17. Tung YC, Chang GM, Chen YH. Associations of physician volume and weekend admissions with ischemic stroke outcome in Taiwan: a nationwide population-based study. Med Care. 2009;47: 1018–1025. pmid:19648828
- 18. Lichtman JH, Leifheit-Limson EC, Jones SB, Wang Y, Goldstein LB. 30-day risk-standardized mortality and readmission rates after ischemic stroke in critical access hospitals. Stroke. 2012;43: 2741–2747. pmid:22935397
- 19. Miyata H, Hashimoto H, Horiguchi H, Matsuda S, Motomura N, Takamoto S. Performance of in-hospital mortality prediction models for acute hospitalization: hospital standardized mortality ratio in Japan. BMC Health Serv Res. 2008;8: 229. pmid:18990251
- 20. Matsuda S, Fujimori K, Kuwabara K, Ishikawa KB, Fushimi K. Diagnosis procedure combination as an infrastructure for the clinical study. Asian Pac J Dis Manag 2011;5: 81–87.
- 21.
Ministry of Health Labour and Welfare. Survey of Medical Institutions. Available: http://www.mhlw.go.jp/english/database/db-hss/mi.html.
- 22. Shigematsu K, Nakano H, Watanabe Y. The eye response test alone is sufficient to predict stroke outcome—reintroduction of Japan Coma Scale: a cohort study. BMJ Open 2013;3: 4.
- 23. Ono K, Wada K, Takahara T, Shirotani T. Indications for computed tomography in patients with mild head injury. Neurol Med Chir (Tokyo) 2007;47: 291–297.
- 24. Hubeny I. Vertical structure of accretion disks—A simplified analytical model. Astrophys J, Part 1. 1990;351: 632–641.
- 25.
Ash A, Schwartz M. Evaluating the performance of risk-adjustment methods: dichotomous variables. In: Iezzoni L, editor. Risk adjustment for measuring health care outcomes. Ann Arbor: Health Administration Press; 1994. pp. 313–346.
- 26. Tsugawa Y, Kumamaru H, Yasunaga H, Hashimoto H, Horiguchi H, Ayanian JZ. The association of hospital volume with mortality and costs of care for stroke in Japan. Med Care. 2013;51: 782–788. pmid:23942219