INTRODUCTION

Stroke can lead to severe disability and death and is the second leading cause of death worldwide. In Korea, stroke is the fourth leading cause of death, accounting for 42.6 deaths per 100,000 population in 2020. Stroke mortality also varies substantially by region; for example, the age-standardized mortality rate per 100,000 people was highest in Ulsan Metropolitan City at 30.0 and lowest in Jeju at 15.9 [1]. Because timely medical care can substantially reduce premature death and disability after an incident event [2,3], regional inequalities in stroke mortality in Korea may be partly attributable to unequal access to appropriate care.

Previous studies have identified individual- and area-level characteristics associated with regional inequalities in stroke mortality in Korea [4,5]. These characteristics include socioeconomic status, underlying health conditions, rural residence, and limited access to healthcare facilities. However, few studies have examined hospital-level factors associated with outcomes among patients with stroke [6,7].

In addition to timely access, the ability of hospitals to provide appropriate care is critical to patient outcomes. This study aimed to identify hospital-level factors associated with acute stroke mortality and, in particular, to determine the appropriate level of hospital resources needed for high-quality care while accounting for individual-level characteristics.

MATERIALS AND METHODS

Data source

This nationwide retrospective cohort study included all patients with acute stroke in Korea from January 1, 2018, to December 31, 2020. Data were extracted from the National Health Insurance Service (NHIS) claims database and the Medical Resource Report of the Health Insurance Review & Assessment Service. These data included patients’ socio-demographic information, such as sex, age, health insurance type, residential area, comorbidities, and outcomes, including death, as well as hospital characteristics.

Study population

We used inpatient claims data from the NHIS for patients with hemorrhagic or ischemic stroke (International Classification of Diseases, 10th revision [ICD-10] codes I60–I64) who were admitted through emergency departments. A total of 315,762 patients were included in the final sample, with 541,556 claims. The unit of analysis was the inpatient claim; each death was assigned to a single claim per patient to avoid double counting. Of the 315,762 patients, 75,706 died within 30 days of admission.

Variables

The main outcome variable was all-cause mortality, defined as death within 30 days of admission regardless of the location of death. Restricting the outcome to in-hospital deaths in individual hospitals could overestimate mortality in final-destination hospitals because of their higher clinical capacity and underestimate mortality in hospitals with high referral rates because of their lower capacity. Because NHIS claims data do not include cause of death, and because total mortality has been widely used in disease-burden studies, we defined total mortality, rather than stroke-specific mortality, as the outcome variable [8].

Individual-level variables included socio-demographic characteristics, such as sex; age categorized into three groups (<15, 15–64, and ≥65 years); income, measured by insurance premium quintile; and residence, measured by health service areas [9]. The Charlson comorbidity index (CCI) was used to measure underlying health conditions that may influence patient outcomes. The CCI was based on the revised version used in the United Kingdom Summary Hospital-level Mortality Indicator calculation [10,11]. To avoid underestimating comorbid conditions, this study used up to five diagnosis codes for major and minor conditions in the NHIS claims data to calculate the score.

Acute stroke severity was measured using the stroke severity index (SSI), which was developed by Sung et al. [12]. Using NHIS claims data, Sung et al. [12] obtained regression coefficients for seven stroke severity items: airway suction, bacterial sensitivity testing, general ward stay, intensive care unit stay, nasogastric intubation, osmotherapy, and urinary catheterization. They then estimated SSI scores for each case by applying these coefficients to a multiple regression model. The SSI has been validated against the National Institutes of Health Stroke Scale (NIHSS) in both ischemic and hemorrhagic stroke. We estimated the SSI for each stroke case using the same regression model and coefficient values [13-15].

To examine patients’ access to hospitals, we measured the mean distance in kilometers from patients’ homes to the hospitals where they were treated and the percentage of patients treated within their own service area. Hospital-level variables included the total numbers of beds, specialists, neurologists and neurosurgeons, and nurses, as well as hospital accreditation level. Acute intervention capacity was defined according to stroke center designation by the Korean Stroke Society. This designation classifies certified hospitals into two tiers: Stroke Centers (SCs), which provide intravenous thrombolysis and are analogous to Primary Stroke Centers in the international literature, and Thrombectomy-Capable Stroke Centers (TSCs), which additionally provide endovascular thrombectomy. In the present study, this variable was operationalized as a binary indicator (designated vs. non-designated), with SCs and TSCs combined into a single category.

Statistical analysis

Descriptive statistics were calculated to examine bivariate associations between individual-level and hospital-level characteristics and mortality among patients with stroke. The chi-squared test was used for categorical variables, and the t-test was used for continuous variables. The association between hospital-level characteristics and 30-day mortality among patients with stroke was then examined using hierarchical logistic regression analysis with generalized linear mixed models through the GLIMMIX procedure, which accounted for patient characteristics. All statistical analyses were performed using SAS version 9.3 (SAS Institute Inc., Cary, NC, USA). All p-values were two-sided, and p-value <0.05 was considered statistically significant.

The optimal numbers of hospital resources and staff were defined as the points at which risk-adjusted mortality rates significantly decreased. Change-point analysis (CPA) using Taylor’s tool was performed to determine the optimal number of hospital resources and staff. CPA identifies the point at which a change has occurred by using a combination of cumulative sum charts and bootstrapping, and it characterizes detected changes with confidence levels [16]. This method has been used to identify optimal resource levels in emergency medical care in Korea [17,18]. CPA was performed for the numbers of beds, all specialists, neurologists and neurosurgeons, and nurses (Figure 1).

Ethics statement

This study was approved by the Institutional Review Board of the National Medical Center (IRB No. NMC-2022-02-025).

RESULTS

Descriptive statistics

Among 541,556 inpatient claims for acute stroke, 75,706 (14.0%) resulted in death within 30 days of admission. The proportions of females, older adults, and patients in lower income quintiles were higher in the non-survivor group. The deceased group had higher medical expenditures and higher CCI and SSI values. They were also more likely to travel longer distances from home to the hospital and less likely to receive care in their own health service area (HSA) (p<0.001; Table 1).

When hospitals where patients survived were compared with hospitals where patients died, the former had larger numbers of beds, specialists, neurologists and neurosurgeons, and nurses. Patients treated in hospitals with lower accreditation grades and limited intervention capacity were also more likely to die (Table 1).

CPA identified hospital-factor thresholds associated with declining mortality trends: 600 beds, 25 specialists, 8 neurologists and neurosurgeons, and 540 nurses. These thresholds were used to categorize hospital attributes (Figure 1).

Multilevel analysis

To examine the effects of hospital-level characteristics on 30-day mortality among patients with acute stroke while accounting for individual-level characteristics, multilevel models were constructed from a null model to progressively more complex models. Model 1, a random-intercept model, did not include explanatory variables. The interhospital variance in deaths among patients with acute stroke was 0.156, and the intraclass correlation coefficient (ICC) was 4.52%. The model fit was statistically significant (p<0.001). In model 2, which included individual-level variables, the between-hospital variance increased to 0.160, and the ICC increased from 4.52% to 4.65%. The model’s goodness-of-fit was statistically significant (p<0.001). In model 3, which included only hospital-level variables, the between-hospital variance was 0.149, which was smaller than that in model 1. The ICC decreased from 4.52% to 4.35%.

In model 4, the effects of both individual-level and hospital-level variables on deaths among patients with acute stroke were examined. The between-group variance was 0.140, which was smaller than those in models 2 and 3, and the ICC was 4.07%. The explanatory power of the model, expressed as the percentage of variance explained, was 10.46%, which was higher than that of the previous models. After individual-level characteristics were accounted for, patients with acute stroke were more likely to die when treated in hospitals with fewer resources and lower clinical capacity, as indicated by fewer beds (<600), fewer neurologists and neurosurgeons (<8), lower accreditation grades, and no intervention capacity (Table 2).

DISCUSSION

In Korea, most advanced medical institutions are concentrated in metropolitan areas, particularly Seoul. This uneven distribution of healthcare resources may contribute to inequalities in healthcare access and, consequently, inequalities in health outcomes. In emergency conditions in which outcomes can change within minutes, healthcare facilities within each HSA need to provide appropriate care promptly. This study aimed to identify hospital-level factors associated with mortality among patients with acute stroke while accounting for patient characteristics. We also estimated hospital-resource thresholds associated with better outcomes among patients with acute stroke [19,20].

The thresholds at which statistically significant reductions in mortality were observed were 600 hospital beds, 25 specialists, 8 neurologists and neurosurgeons combined, and 540 nurses. Multilevel analysis showed that patients treated in hospitals with fewer resources than these thresholds had a higher risk of mortality. These findings are consistent with previous studies showing that stroke mortality is higher in smaller hospitals. In addition to quantitative indicators, such as the number of beds and personnel, this study identified qualitative indicators, including accreditation grade and procedural capability, that were associated with mortality.

However, the multilevel models showed an unexpected finding: mortality risk was lower among patients treated in hospitals with fewer than 540 nurses (OR, 0.83; p=0.012). Previous studies have reported a significant positive association between the number of patients per nurse and in-hospital mortality. This discrepancy is most plausibly explained by confounding by indication arising from referral patterns in the Korean healthcare system. Tertiary hospitals, which typically employ larger nursing staff, serve as referral centers and disproportionately manage patients with severe underlying conditions, such as those undergoing active cancer treatment, whose baseline mortality risk is inherently higher and may not be fully captured by administrative comorbidity measures. When such patients present with acute stroke, their elevated baseline risk may inflate the apparent mortality rate in well-resourced hospitals, biasing the observed nurse–mortality association toward the null or even reversing its direction. Similarly, the total number of specialists was negatively associated with mortality, although this association was not statistically significant. Different findings might have been observed if the number of nurses working in stroke units could have been disaggregated, as was done for neurologists and neurosurgeons. Future studies incorporating hospital-tier classification, such as tertiary versus general hospitals, or stratified analyses by hospital size are warranted to distinguish the structural effects of nursing staff levels from referral-driven casemix differences.

As expected, patients who received care at institutions with higher accreditation grades or emergency stroke intervention recognition had a lower risk of mortality. Such hospitals are more likely to have more highly specialized medical personnel, adequate facilities, and necessary equipment. They may also maintain higher quality by performing a sufficient volume of procedures.

This study has several limitations. First, although it would have been ideal to consider the number of nurses assigned specifically to stroke unit beds, hospital-level data on nurse allocation at the ward-bed level were not available. Second, as discussed above, the present analysis could not fully address potential referral bias and confounding by indication. Because tertiary hospitals in Korea disproportionately care for medically complex patients with severe baseline comorbidities, the observed associations between hospital resources and mortality may partly reflect differences in patient case mix rather than the structural effects of hospital-level care capacity. This unmeasured confounding may have biased the estimated effect sizes. Third, owing to the inherent limitations of claims data, 30-day all-cause mortality was used as the outcome rather than stroke-specific mortality. A proportion of observed deaths may therefore have been attributable to underlying conditions rather than the cerebrovascular event itself, particularly among patients with severe comorbidities concentrated in tertiary hospitals. As a result, the estimated associations between hospital characteristics and stroke-related mortality may have been biased toward the null because of non-differential outcome misclassification. To partially reduce this limitation, the analysis was restricted to patients with cerebrovascular disease admitted through the emergency department. When reliable cause-of-death information becomes available, future studies should focus on stroke-specific mortality. Fourth, the claims data did not capture critical prehospital determinants of acute stroke prognosis, including symptom onset-to-door time, mode of arrival, and use of emergency medical services. The absence of these temporal and prehospital variables is an important source of residual confounding that may have distorted the estimated effect sizes for hospital-level characteristics. Linking claims data with the National Emergency Medical Information Network and Fire Department rescue and emergency activity datasets would enable more valid and comprehensive analyses in future research. Fifth, severity adjustment in this study relied on administrative proxies—the SSI and CCI—rather than validated clinical scales such as the NIHSS. Although the SSI has been validated against the NIHSS in both ischemic and hemorrhagic stroke populations [12,14], administrative diagnosis codes and procedural proxies cannot fully capture baseline neurological deficits. Consequently, residual confounding by unmeasured stroke severity remains possible, and the findings should be interpreted with this caveat in mind. Finally, the acute stroke intervention capacity variable was based on the Korean Stroke Society’s SC designation, which classifies certified hospitals into two tiers. In the present analysis, however, as described in the Methods section, this variable was operationalized as a binary indicator (designated vs. non-designated), with SCs and TSCs aggregated into a single category. This approach may obscure the differential contribution of advanced interventional capabilities, particularly endovascular thrombectomy, to stroke survival. Future studies should disaggregate these designations to clarify the impact of intervention complexity on mortality outcomes.

CONCLUSION

This study suggests that each HSA should have a hospital with sufficient resources and acute stroke care capacity that can provide appropriate care for acute stroke to improve survival among patients with acute stroke and reduce regional inequalities. The staffing and resource-allocation standards derived from this study may also serve as benchmarks for capacity building in healthcare institutions.

NOTES

Conflict of interest

The authors have no conflicts of interest to declare for this study.

Funding

Open access funding was enabled and organized by the National Medical Center.

Acknowledgements

None.

Author contributions

Both authors contributed equally to conceiving the study, analyzing the data, and writing this paper.

Figure 1.

The optimal number of hospital resources and staffing by change-point analysis. To reflect the quality of medical care, we used (A) number of hospital beds, (B) total number of specialists, (C) number of neurologists and neurosurgeons, and (D) number of nurses as variables, and performed change-point analysis to derive the optimal resource and staffing thresholds at which stroke mortality showed a significant decrease.

REFERENCES

• 1. Statistics Korea. 2020 cause of death statistics; 2021 [cited 2025 Sep 1]. Available from: https://kostat.go.kr/board.es?bid=218&mid=a10301060200 (Korean)
• 2. Feigin VL, Norrving B, Mensah GA. Global burden of stroke. Circ Res 2017;120:439-448. https://doi.org/10.1161/CIRCRESAHA.116.308413ArticlePubMed
• 3. World Health Organization. Global status report on noncommunicable diseases 2010; 2011 [cited 2025 Sep 25]. Available from: https://iris.who.int/handle/10665/44579
• 4. Park YY, Park JH, Park YH, Lee KS. A study on the regional factors affecting the death rates of cardio-cerebrovascular disease using the spatial analysis. Health Policy Manag 2020;30:26-36. (Korean). https://doi.org/10.4332/KJHPA.2020.30.1.26Article
• 5. Kang M, Son K. Risk of death and occurrence of secondary disease of cancer and cardiovascular disease patient by income level in Korea. J Korea Contents Assoc 2018;18:145-157. (Korean). https://doi.org/10.5392/JKCA.2018.18.10.145Article
• 6. Kim SH, Lee HJ. Factors affecting the outcome indicators in patients with stroke. Health Policy Manag 2015;25:31-39. (Korean). https://doi.org/10.4332/KJHPA.2015.25.1.31Article
• 7. Kim JE, Bae SH. Association between nurse staffing and mortality in patients with ischemic and hemorrhagic stroke in the intensive care units. J Korean Acad Nurs Adm 2021;27:311-322. (Korean). https://doi.org/10.11111/jkana.2021.27.5.311Article
• 8. Pak H, Pak YS. Construction of medical episode data using National Health Insurance Service data. J Converg Inf Technol 2019;9:195-200. (Korean). https://doi.org/10.22156/CS4SMB.2019.9.9.195Article
• 9. Kim SB, Lee BM, Park JW, Kwak MY, Jang WM. Weekend effect on 30-day mortality for ischemic and hemorrhagic stroke analyzed using severity index and staffing level. PLoS One 2023;18:e0283491. https://doi.org/10.1371/journal.pone.0283491ArticlePubMedPMC
• 10. Charlson ME, Pompei P, Ales KL, MacKenzie CR. A new method of classifying prognostic comorbidity in longitudinal studies: development and validation. J Chronic Dis 1987;40:373-383.ArticlePubMedPMC
• 11. Glasheen WP, Cordier T, Gumpina R, Haugh G, Davis J, Renda A. Charlson comorbidity index: ICD-9 update and ICD-10 translation. Am Health Drug Benefits 2019;12:188-197.PubMedPMC
• 12. Sung SF, Hsieh CY, Kao Yang YH, Lin HJ, Chen CH, Chen YW, et al. Developing a stroke severity index based on administrative data was feasible using data mining techniques. J Clin Epidemiol 2015;68:1292-1300. https://doi.org/10.1016/j.jclinepi.2015.01.009ArticlePubMed
• 13. Shim DH, Kim Y, Roh J, Kang J, Park KP, Cha JK, et al. Hospital volume threshold associated with higher survival after endovascular recanalization therapy for acute ischemic stroke. J Stroke 2020;22:141-149. https://doi.org/10.5853/jos.2019.00955ArticlePubMedPMC
• 14. Hung LC, Sung SF, Hsieh CY, Hu YH, Lin HJ, Chen YW, et al. Validation of a novel claims-based stroke severity index in patients with intracerebral hemorrhage. J Epidemiol 2017;27:24-29. https://doi.org/10.1016/j.je.2016.08.003ArticlePubMedPMC
• 15. Shim DH. Volume-outcome relationships for endovascular recanalization therapy of acute ischemic stroke in Korea and identifying volume thresholds [dissertation]. Seoul: Seoul National University; 2018. (Korean).
• 16. Taylor WA. Change-point analysis: a powerful new tool for detecting changes; 2000 [cited 2025 Sep 12]. Available from: https://variation.com/wp-content/uploads/change-point-analyzer/changepoint-analysis-a-powerful-new-tool-for-detecting-changes.pdf
• 17. Kwak MY, Lee SM, Kim HJ, Eun SJ, Jang WM, Jung H, et al. How far is too far? A nationwide cross-sectional study for establishing optimal hospital access time for Korean pregnant women. BMJ Open 2019;9:e031882. https://doi.org/10.1136/bmjopen-2019-031882ArticlePubMedPMC
• 18. Jang WM, Lee J, Eun SJ, Yim J, Kim Y, Kwak MY. Travel time to emergency care not by geographic time, but by optimal time: a nationwide cross-sectional study for establishing optimal hospital access time to emergency medical care in South Korea. PLoS One 2021;16:e0251116. https://doi.org/10.1371/journal.pone.0251116ArticlePubMedPMC
• 19. Farley DE, Ozminkowski RJ. Volume-outcome relationships and in-hospital mortality: the effect of changes in volume over time. Med Care 1992;30:77-94. https://doi.org/10.1097/00005650-199201000-00009ArticlePubMed
• 20. Ross JS, Normand SL, Wang Y, Ko DT, Chen J, Drye EE, et al. Hospital volume and 30-day mortality for three common medical conditions. N Engl J Med 2010;362:1110-1118. https://doi.org/10.1056/NEJMsa0907130ArticlePubMedPMC

Figure & Data

References

Citations

Citations to this article as recorded by