2024 Volume 31 Issue 11 Pages 1571-1590
Aim: This study investigated the impact of rurality on acute ischemic stroke (AIS) outcomes, emphasizing the hyperacute phase, in which immediate care is crucial.
Methods: This retrospective cohort study analyzed data from a large Japanese hospital network covering AIS patients from 2013-2021, was analyzed. The focus was on patients admitted within 4.5 h of the onset, using the Rurality Index for Japan (RIJ) to categorize patients into rural or urban groups. This study examined treatment methods (intravenous thrombolysis [IVT] and mechanical thrombectomy [MT]) and functional outcomes measured using the modified Rankin Scale (mRS), where scores of 3-6 indicated poor outcomes. Multilevel logistic regression was used to calculate the adjusted odds ratios (ORs) and 95% confidence intervals (CIs) for poor outcomes based on rurality. The study also evaluated the population-attributable fraction (PAF) to estimate potential outcome improvements in urban settings.
Results: Of 27,691 patients, 17,516 were included in the total cohort and 4,954 in the hyperacute cohort. Urban patients constituted 73.7% (12,902), with higher IVT (5.2%) and MT (3.6%) rates than rural patients (4.1% IVT, 2.0% MT). Poor mRS outcomes were more common in rural areas than in urban areas, with adjusted ORs of 1.30 (1.18-1.43) in the total cohort and 1.43 (1.19-1.70) in the hyperacute cohort. The PAF for poor outcomes due to rural residency was 14.8% (0.5%-31.0%).
Conclusion: This study demonstrated a notable association between rurality and poorer AIS outcomes in Japan, particularly in the hyperacute phase.
Acute ischemic stroke (AIS) is a major health challenge affecting approximately 15 million people globally each year, of whom 5 million die and another 5 million are permanently disabled, imposing major burdens on families and society1, 2). The incidence, treatment, and outcomes of AIS can vary based on a broad range of social factors, making it a multifaceted issue that transcends medical and healthcare boundaries. In North America and China, AIS mortality is reportedly higher in remote and rural areas than in urban areas; however, the reasons for this disparity are not well understood3-6). Factors hypothesized to explain this observation include a higher AIS incidence, uneven distribution of specialists, longer transportation times, and a higher prevalence of various risk factors for AIS, such as diabetes and smoking, in rural areas than in urban areas, possibly due to region-specific behavioral and cultural features7-10).
Studies in Japan have shown higher stroke mortality in rural residents than in urban residents11), but this finding was based primarily on a pre-2000 cohort and may not accurately reflect current social conditions that have changed with the development of transportation infrastructure, changes in socioeconomic conditions, and 21st-century standards of medical care. Specifically, a study published using data from to 2010-2015 reported regional disparities in the implementation of intravenous thrombolysis (IVT) and mechanical thrombectomy (MT)12). In particular, MT alone has a Gini coefficient (a number between 0 and 1 used to measure inequality, with 0 representing perfect equality) ranging from 0.25 to 0.49, suggesting inequality in the hyperacute treatment of AIS in Japan.
These recent studies often utilized insurance-based data, frequently lacking important information for the management of AIS, such as disease type and severity, and also did not include information on the hyperacute phase3, 12, 13). Many studies have employed death as the outcome3), and little attention has been paid to the functional prognosis. Furthermore, while regional disparities in IVT have been reported, information on transport times is often underutilized and has not been adequately assessed in patients strictly indicated for the hyperacute treatment of AIS (within 4.5 h of the onset)7, 12, 14). Based on the results of prior studies, it is unclear whether regional disparities exist because of long transport times, lack of treatment access, or barriers to subsequent care despite adequate transport.
Given the above, the present study investigated the association between rurality and subsequent physical outcomes in AIS cases in Japan. An integral aspect of this research was the assessment of the impact of rurality across all AIS patients, while also broadening the analysis to trends among potential candidates for IVT who were transported within 4.5 h of the AIS onset. A thorough understanding of these associations may provide clues to help improve medical services and other high-risk approaches in the rural areas of Japan while also providing valuable insights for other countries facing similar geographic and demographic challenges.
This retrospective cohort study used a nationwide, multicenter registry. Data for this study were extracted from the Saiseikai Stroke Database, which is dedicated to the prospective collection of data on patients with acute stroke within the first week of the event. The Saiseikai Stroke Research Group comprises 27 hospitals under the Social Welfare Organization Saiseikai Imperial Gift Foundation in Japan. The database is active from April 2013 to March 2021. Detailed information and data from the Saiseikai Stroke Database can be found in Fmentary Appendix and have been published previously15). All patients in this database completed their acute care at the same facility and did not include those who were transferred shortly after arriving at the hospital without being admitted.
ParticipantsWe selected patients diagnosed with AIS (excluding transient ischemic attack) between April 2013 and March 2021, as recorded in the database. We excluded patients younger than 16 years old and those with missing outcome data.
Data Collection, Variables, and Potential BiasThe following clinical data were collected from the Saiseikai Stroke Database and analyzed: sex, age, medical history (hypertension [HT], diabetes mellitus [DM], dyslipidemia [DL], chronic kidney disease [CKD], congestive heart failure [CHF], atrial fibrillation [AF], and previous stroke), smoking habit, level of consciousness on hospital arrival (Japan Coma Scale [JCS])16-18), clinical classification of AIS according to the Trial of ORG 10172 in Acute Stroke Treatment criteria19), IVT with tissue plasminogen activator, MT procedure, modified Rankin Scale (mRS) score at discharge, and length of hospital stay. The JCS, a consciousness rating scale specific to Japan (Supplementary Table 1), is a reliable tool for assessing intracranial disease and can also be translated into the Glasgow Coma Scale18) (Supplementary Table 2).
| Level of consciousness | ||
|---|---|---|
| 0 | Alert | |
| 1-digit | Awake without any stimuli | |
| 1 | Almost fully conscious but not normal | |
| 2 | Unable to recognize time, place, and person | |
| 3 | Unable to recall name or date of birth | |
| 2-digits | Arousable by some stimuli but reverts to previous state if stimulus stops | |
| 10 | Arousable by being spoken to | |
| 20 | Arousable by loud voice | |
| 30 | Arousable only by repeated mechanical stimuli | |
| 3-digits | Unarousable by any forceful stimuli | |
| 100 | Unarousable but responds to avoid the stimuli | |
| 200 | Unarousable but responds with slight movements, including decerebrate or decorticate postures | |
| 300 | Does not respond at all | |
| Japan Coma Scale | Glasgow Coma Scale |
|---|---|
| 0 | 15 |
| 1 | 15 |
| 2 | 14 |
| 3 | 13 |
| 10 | 12 |
| 20 | 12 |
| 30 | 9 |
| 100 | 7 |
| 200 | 6 |
| 300 | 3 |
As indicated by the above, 1-digit of JCS applies to GCS 13-15 (mild GCS), 2-digits of JCS applies to GCS 9-12 (moderate GCS), 3-digits of JCS applies o GCS 3-8 (severe GCS).
Conversion table with high accuracy reported by Nakajima et al.2)
JCS: Japan Coma Scale, GCS: Glasgow Coma Scale
In addition, hospital-related factors were identified, such as whether or not the hospital was a primary stroke center (PSC), hospital size defined by the number of beds, and hospital volume defined by the number of AIS cases experienced. The Japan Stroke Society accredits PSCs as facilities that meet the necessary conditions for stroke care20, 21). Hospital size was divided into the following three groups based on previous studies: 1-199 (small), 200-499 (medium), and 500 (large)22-24). Hospital volume was defined as “high volume” when the average number of cases was 100 or more and “low volume” when the average number of cases was less than 100, because the criterion for training and education facilities of the Japan Stroke Society is to have more than 100 cases annually.
Rurality was defined by applying the Rurality Index for Japan (RIJ), developed in a previous study25), to the secondary medical care area to which each hospital belonged. Patients come to or are transported to a hospital in the same secondary medical care area as the area in which they reside, but some may later be transported to a different secondary medical care area. A heat map of the secondary medical care areas and RIJ throughout Japan and a plot of the hospitals enrolled in this study are shown in Supplementary Fig.1. The RIJ was calculated based on population density, distance to secondary and tertiary emergency hospitals, isolated islands, and areas with particularly heavy snowfall. The calculated unadjusted values were converted to a scale of 1–100 (1 being the most urban and 100 being the most rural). Moderate negative correlations were reported between the RIJ and the physician disproportionality index, thereby confirming convergent validity and criterion validity in a previous study25).
This figure is reproduced with the permission from the publisher of responsible author of previous study3), and the enrollment hospitals are plotted.
In addition, the classifications of designated cities (designated by government ordinances from among cities with a population of ≥ 500,000), core cities (designated by government ordinances based on applications from cities with a population of ≥ 200,000), and general municipalities (others) were used according to the administrative divisions of Japanese cities26). The above municipal classification was used depending on the location of the hospital in the municipality. Japan has 20 designated cities and 62 core cities. The 23 wards of Tokyo, the central area of Japan’s capital, are not cities and are not classified as designated cities. However, since they have the same administrative scale as designated cities, the hospitals in the 23 wards were categorized as designated cities for the purpose of this study.
OutcomesPoor functional outcomes, as assessed by the mRS at discharge and length of hospital stay, were observed in this study. We defined a good functional outcome as an mRS score of 0-2 and a poor functional outcome as an mRS score of 3-6 15).
Statistical AnalysesIn this study, we first plotted the relationship between outcomes and the RIJ as a continuous variable for visualization. We divided the population into two groups, rural and urban, based on a previously determined cut-off value (unadjusted RIJ: 0.00147=adjusted RIJ: 47/100)25), and descriptively compared the clinical characteristics and treatment contents between these two groups.
Statistical analyses were performed for both the standard logistic regression (LR) model and the multilevel logistic regression (ML-LR) model27, 28). In the LR model (model 1), patient factors such as the sex, age, medical history (HT, DM, DL, CKD, CHF, and prior stroke), smoking habit, JCS classification on arrival, and clinical classification of AIS were used as explanatory variables. In Model 2, in addition to patient factors, the facility factors of PSC, hospital size, and hospital volume were analyzed as explanatory variables. Adjusted odds ratios (ORs) and 95% confidence intervals (CIs) were calculated using mRS failure as the objective variable, and binary rurality as the primary explanatory variable. Based on the LR findings, ML-LR was used to better capture the complexity of the relationships between factors and outcomes. Hospital IDs were entered as a random effect to account for potential clustering and nesting of data, and variables used in the LR analysis served as fixed effects; using the ‘lme4’ package in R for the analysis, adjusted ORs and 95% CIs for provinces as fixed effects were calculated (Supplementary Fig.2)29).
In this study, each Saiseikai hospital is located in a different secondary healthcare service area, and the secondary healthcare service area and hospital are completely linked as shown in the figure. The structure is hierarchical, with individuals existing at the lower levels of those hierarchies.
Additional analyses were conducted to assess the robustness of the model, including the ML-LR model with IVT and MT variables added to fixed effects, time from the onset to arrival added to fixed effects, cases admitted to the PSC, and mortality at discharge as an outcome. From the initial ML-LR model, we extracted the variance of random effects or intercept of the hospital. In addition, the intracluster correlation coefficient (ICC) was calculated to determine the proportion of the total variance attributable to differences in clusters by each hospital28, 30). In addition to the ML-LR models, the population attributable fraction (PAF) and 95% CIs were calculated to quantify the proportion of poor functioning outcomes attributable to rural residence.
We also performed equivalent ML-LR analyses by categorizing the data into three groups based on Japan’s municipal administrative divisions: designated cities, core cities, and general municipalities, with designated cities serving as the reference category26). For the sensitivity analysis, RIJ expressed as 1-100 was categorized into 3 groups: urban (RIJ: 1-33), intermediate (RIJ: 34-66), rural (RIJ: 67-100), and urban (RIJ: 1-33) as the reference for the same ML-LR analysis. All LR and ML-LR were analyzed in exactly the same way for the “hyperacute cohort,” which included only patients transported within 4.5 h of the symptom onset. Because the time from the onset of MT indications has changed over time, analyses were also performed for cohorts that presented within 6 h and cohorts that presented within 8 h of the onset of MT to confirm robustness. Furthermore, restricted cubic splines with three knots were employed to explore the potential nonlinear association between the RIJ as a continuous variable and a poor mRS score at discharge adjusted for the above confounders using the’ rms’ package in R31).
Statistical significance was confirmed when the 95% CI did not overlap with the null effect value or when the two-sided P-value was less than 0.05. All statistical analyses were performed using Python programming language Version 3.8.10 (Python Software Foundation, Wilmington, DE, USA) and R software (version 1.1.456; R Studio Inc., Boston, MA, USA).
Of 27,691 patients enrolled in the Saiseikai Stroke Database, 17,516 were eligible for the primary analysis of the total cohort, and 4,954 were eligible for the secondary analysis of the hyperacute phase cohort. The flowchart of the study is shown in Supplementary Fig.3. Of the total cohort, 4,614 (26.3%) were in the rural group, and 12,902 (73.7%) were in the urban group. In addition, of the 17,516 patients, 4,954 (28.3%) were transported within 4.5 h of the onset or were last known to be well, including 1,279 (25.8%) in the rural group and 3,675 (74.2%) in the urban group. The baseline characteristics of the patients in the urban and rural groups are presented in Table 1 and Supplementary Table 3. In the hyperacute cohort of cases within 4.5 h of the onset, the distribution of the time from the onset to hospital arrival was visualized in histograms in the rural and urban groups, as shown in Supplementary Fig.4.
Flow chart of patient selection
| Parameters |
Total N = 17,516 |
Urban N = 12,902 |
Rural N = 4,624 |
|---|---|---|---|
| Sex (male), n, (%) | 10399 (59.4%) | 7679 (59.5%) | 2720 (59.0%) |
| Age (years), median, (IQR) | 77 (68-84) | 76 (67-83) | 78 (70-85) |
| Medical history, n, (%) | |||
| Hypertension | 11573 (66.1%) | 8321 (64.5%) | 3252 (70.5%) |
| Dyslipidemia | 6431 (36.7%) | 4626 (35.9%) | 1805 (39.1%) |
| Diabetes mellitus | 4735 (27.0%) | 3483 (27.0%) | 1252 (27.1%) |
| Chronic kidney disease | 2503 (14.3%) | 1747 (13.5%) | 756 (16.4%) |
| Smoking | |||
| Current smoker | 3292 (18.8%) | 2567 (19.9%) | 725 (15.7%) |
| Past smoker | 2950 (16.8%) | 2121 (16.4%) | 829 (18.0%) |
| Congestive heart failure | 1294 (7.4%) | 1042 (8.1%) | 252 (5.5%) |
| Atrial fibrillation | |||
| Previously diagnosed | 3103 (17.7%) | 2333 (18.1%) | 770 (16.7%) |
| Newly diagnosed | 703 (4.0%) | 403 (3.1%) | 300 (6.5%) |
| Previous stroke | 4530 (25.9%) | 3260 (25.3%) | 1270 (27.5%) |
| Consciousness (JCS), n, (%) | |||
| 1-digit | 15763 (90.1%) | 11535 (89.5%) | 4228 (91.6%) |
| 2-digits | 1167 (6.7%) | 905 (7.0%) | 262 (5.7%) |
| 3-digits | 569 (3.3%) | 445 (3.4%) | 124 (2.7%) |
| Clinical classification of AIS, n, (%) | |||
| Small vessel occlusion | 4421 (25.2%) | 3260 (25.3%) | 1161 (25.2%) |
| Large artery atherosclerosis | 5272 (30.1%) | 3817 (29.6%) | 1455 (31.5%) |
| Cardioembolism | 3914 (22.4%) | 2873 (22.3%) | 1041 (22.6%) |
| Other determined etiology | 2701 (15.4%) | 1936 (15.0%) | 765 (16.5%) |
| Undetermined etiology | 1208 (6.9%) | 1016 (7.9%) | 192 (4.2%) |
| PSC, n, (%) | 15020 (85.8%) | 12141 (94.1%) | 2879 (62.3%) |
| Hospital size, n, (%) | |||
| Large (500- beds) | 3299 (18.83%) | 3299 (25.57%) | 0 (0%) |
| Medium (200-499 beds) | 12386 (70.7%) | 9507 (73.7%) | 2879 (62.4%) |
| Small (1-199 beds) | 1831 (10.5%) | 96 (0.7%) | 1735 (37.6%) |
| Hospital volume, n, (%) | |||
| High volume | 12369 (70.6%) | 8580 (66.5%) | 3789 (81.9%) |
| Low volume | 5147 (29.4%) | 4322 (23.5%) | 825 (18.1%) |
IQR: interquartile range, JCS: Japan Coma Scale, AIS: acute ischemic stroke, ESUS: embolic stroke of undetermined sources, PSC: primary stroke center
| Parameters | Total N= 4,954 | Urban N= 3,675 | Rural N= 1,279 |
|---|---|---|---|
| Sex (male), n, (%) | 2923 (59.0%) | 2192 (59.7%) | 731 (57.2%) |
| Age(years), median, (IQR) | 77 (69-85) | 76 (68-84) | 80 (71-86) |
| Past medical history, n, (%) | |||
| Hypertension | 3123 (63.0%) | 2270 (61.8%) | 853 (66.7%) |
| Dyslipidemia | 1705 (34.4%) | 1238 (33.7%) | 467 (36.5%) |
| Diabetes mellitus | 1164 (23.5%) | 862 (23.5%) | 302 (23.6%) |
| Chronic kidney disease | 722 (14.6%) | 487 (13.3%) | 235 (18.4%) |
| Smoking | |||
| Current smoker | 801 (16.2%) | 639 (17.4%) | 162 (12.7%) |
| Past smoker | 861 (17.4%) | 620 (16.9%) | 241 (18.8%) |
| Congestive heart failure | 474 (9.6%) | 380 (10.3%) | 94 (7.4%) |
| Atrial fibrillation | |||
| Previously diagnosed | 1269 (25.6%) | 946 (25.8%) | 323 (25.3%) |
| Newly diagnosed | 294 (5.9%) | 172 (4.7%) | 122 (9.6%) |
| Previous stroke | 1298 (26.2%) | 957 (26.0%) | 341 (26.7%) |
| Consciousness (JCS), n, (%) | |||
| 1-digit | 4141 (83.6%) | 3045 (82.9%) | 1096 (85.7%) |
| 2-digits | 545 (11.0%) | 422 (11.5%) | 123 (9.6%) |
| 3-digits | 267 (5.4%) | 207 (5.6%) | 60 (4.7%) |
| Clinical classification of AIS, n, (%) | |||
| Small vessel occlusion | 889 (18.0%) | 660 (18.0%) | 229 (17.9%) |
| Large artery atherosclerosis | 1418 (28.6%) | 1049 (28.5%) | 369 (28.9%) |
| Cardioembolism | 1652 (33.4%) | 1200 (32.7%) | 452 (35.3%) |
| Other determined etiology | 534 (10.8%) | 448 (12.2%) | 86 (6.7%) |
| Undetermined etiology | 341 (6.9%) | 287 (7.8%) | 54 (4.2%) |
| PSC, n, (%) | 4272 (86.2%) | 3438 (93.6%) | 834 (65.2%) |
| Hospital size, n, (%) | |||
| Large (500- beds) | 1144 (23.1%) | 1144 (31.1%) | 0 (0%) |
| Medium (200-499 beds) | 3350 (67.6%) | 2516 (68.5%) | 834 (65.2%) |
| Small (1-199 beds) | 460 (9.3%) | 15 (0.4%) | 445 (34.8%) |
| Hospital volume, n, (%) | |||
| High volume | 1433 (28.9%) | 1243 (33.8%) | 190 (14.9%) |
| Low volume | 3521 (71.1%) | 2432 (66.2%) | 1089 (85.1%) |
IQR: Interquartile range, JCS: Japan Coma Scale, AIS: Acute ischemic stroke, ESUS: Embolic stroke of undetermined sources, PSC: Primary stroke center
Histogram of time from onset to hospital arrival (hours) in the hyperacute cohort for rural and urban
A histogram of the distribution of outcomes using continuous RIJ values is shown in Fig.1. The proportion of patients receiving IVT and MT was lower in the rural group in both the total and hyperacute cohorts (Fig.2). In the total cohort, IVT was performed in 672 (5.2%) urban group patients and 190 (4.1%) rural group patients, and the MT numbers were 465 (3.6%) and 94 (2.0%), respectively. In addition, the proportion of patients with a poor mRS score was higher, and the length of hospital stay tended to be longer in the rural group than in the urban group (Fig.2). Detailed treatment and outcome data are presented in Supplementary Tables 4 and 5, respectively. In addition, a chart showing the proportion of each mRS score (0-6) for the rural and urban groups is presented in Supplementary Fig.5.
RIJ: rurality index for Japan, mRS: modified Rankin Scale
IVT: intravenous thrombolysis, MT: mechanical thrombectomy, mRS: modified Rankin Scale
| Parameters |
Total N= 17,516 |
Urban N= 12,902 | Rural N= 4,614 |
|---|---|---|---|
| Treatment in hyperacute phase, n, (%) | |||
| Intravenous thrombolysis | 862 (4.9%) | 672 (5.2%) | 190 (4.1%) |
| Mechanical thrombectomy | 559 (3.2%) | 465 (3.6%) | 94 (2.0%) |
| Outcome | |||
| Length of hospital stay, median, (IQR) | 16 (10-28) | 15 (10-26) | 21 (13-35) |
| Poor functional outcome, n, (%) | 7905 (45.1%) | 5648 (43.8%) | 2257 (48.9%) |
IQR: Interquartile range
| Parameters | Total N= 4,954 | Urban N= 3,675 | Rural N= 1,279 |
|---|---|---|---|
| Treatment in hyperacute phase, n, (%) | |||
| Intravenous thrombolysis | 811 (16.4%) | 634 (17.3%) | 177 (13.8%) |
| Mechanical thrombectomy | 413 (8.3%) | 341 (9.3%) | 72 (5.6%) |
| Outcome | |||
| Length of hospital stay, median, (IQR) | 17 (10-19) | 16 (10-27) | 21 (13-35) |
| Poor functional outcome, n, (%) | 2477 (50.0%) | 1771 (48.2%) | 706 (55.2%) |
IQR: Interquartile range
mRS: modified Rankin Scale
The results of each analysis, analyzed using the LR and ML-LR models, are summarized in Table 2. In the main analysis of the total cohort, the adjusted OR (95% CI) was 1.22 (1.12-1.32) for the LR model 1, 1.48 (1.34-1.64) for the LR model 2, and 1.66 (1.02-2.71) for the ML-LR model with urban serving as the reference. The ML-LR model included the hospital ID as a random intercept. The variance in the random intercept for the hospital was estimated to be 0.326. The ICC was 0.096, indicating that 9.6% of the total variability in the outcome was due to differences in the hospital, and the median OR was 1.40. The ML-LR model in the hyperacute cohort yielded an OR of 1.71 (1.19-2.44) for rural as compared to urban. In addition, the PAF for poor outcomes due to rural residence was calculated as 14.8% (95% CI: 0.5-31.0%) in the ML-LR model. The results of the analyses described above for the ML-LR model with IVT and MT variables added to fixed effects, time from the onset to arrival added to fixed effects, cases admitted to the PSC (N=15,020), and mortality at discharge as an outcome are presented in Supplementary Table 6 and showed similar trends.
| Variables | Total cohort Adjusted OR [95% CI] | Hyperacute cohort Adjusted OR [95% CI] |
|---|---|---|
| Main analysis by LR in model 1 | ||
| Urban | Ref | Ref |
| Rural | 1.22 [1.12-1.32] | 1.29 [1.11-1.50] |
| Main analysis by LR in model 2 | ||
| Urban | Ref | Ref |
| Rural | 1.47 [1.34-1.64] | 1.69 [1.39-2.05] |
| Main analysis by ML-LR | ||
| Urban | Ref | Ref |
| Rural | 1.66 [1.02-2.71] | 1.71 [1.19-2.44] |
| Analysis of municipal category by ML-LR | ||
| Designated | Ref | Ref |
| Core | 1.55 [1.02-2.35] | 1.43 [1.01-2.02] |
| General | 1.17 [0.76-1.80] | 1.22 [0.85-1.76] |
| Analysis of RIJ divided into three by ML-LR | ||
| Urban (RIJ: 1-33) | Ref | Ref |
| Suburban (RIJ: 34-66) | 1.08 [0.70-1.66] | 1.00 [0.70-1.43] |
| Rural (RIJ: 67-100) | 1.62 [0.96-1.73] | 1.66 [1.19-2.32] |
OR: odds ratio, LR: logistic regression, ML-LR: multilevel logistic regression, RIJ: rurality index for Japan
| Variables | Total cohort Adjusted OR [95% CI] | Hyperacute cohort Adjusted OR [95% CI] |
|---|---|---|
| Main analysis by ML-LR adjusted by treatment | ||
| Urban | Ref | Ref |
| Rural | 1.69 [1.04-2.75] | 1.83 [1.32-2.53] |
| Main analysis by ML-LR adjusted by time from onset to arrival | ||
| Urban | N/A | Ref |
| Rural | N/A | 1.70 [1.19-2.45] |
| Main analysis by ML-LR in PSC cohort | ||
| Urban | Ref | Ref |
| Rural | 1.45 [0.80-2.63] | 1.64 [1.07-2.53] |
| Main analysis by ML-LR (mortality at discharge as outcome) | ||
| Urban | Ref | Ref |
| Rural | 1.34 [0.63-2.88] | 1.49 (0.67-3.31) |
ML-LR: Multilevel logistic regression, OR: Odds ratio, CI: Confidence interval
The results of an additional analysis using the Japanese municipal divisions and a sensitivity analysis using the three RIJ delimited categories are summarized in Table 2. Regarding municipal divisions, 6,918 (39.5%) patients were admitted to hospitals in designated cities, 4,601 (26.3%) to hospitals in core cities, and 5,997 (34.2%) to hospitals in general municipalities. There were 3 rough RIJ groups: 10,903 (62.2%) in the urban group (RIJ: 1-33), 2,824 (16.1%) residing in suburban areas (RIJ: 34-66), and 3,789 (21.6%) in the rural group (RIJ: 67-100). Restricted cubic splines are shown in Fig.3, and a near-linear relationship was found between RIJ as a continuous variable and poor functional outcomes.
Association between the RIJ and poor mRS at discharge allowing for nonlinear effects and 95% CIs. The model was fitted with a three-knot restricted cubic spline for RIJ and adjusted for the age, sex, medical history, consciousness level, hospital size, and clinical classification of ischemic stroke. RIJ: rurality index for Japan
In a cohort that arrived at the hospital within 6 h (N=5,666), the OR (95% CI) for poor functional outcome of rural was 1.72 (95% CI: 1.15-2.57). In a cohort that arrived within 8 h (N=6,374), the OR (95% CI) was 1.75 (1.16-2.63).
In this multilevel analysis of a nationwide, multicenter registry study, rural areas were independently associated with an OR (95% CI) of 1.66 (1.02-2.71) for poor functional outcomes in AIS patients, with lower rates of IVT and MT and longer hospital stays than urban areas. The proportion of patients in the hyperacute group was similar among the groups, and the relationship with the outcome was either similar or slightly worse in the hyperacute cohort than X [please define X]. A nearly linear dose-response relationship between RIJ and poor functional outcomes was also observed.
Strengths of the StudyThis study addressed several important gaps in the literature on AIS and its outcomes in different geographic settings. Unlike many previous studies that used insurance databases and were limited to evaluating mortality, our study used a comprehensive hospital network registry to examine functional outcomes3, 5, 12). This key outcome of AIS has implications for caregiving needs and societal reintegration. The dataset allowed us to adjust for relevant confounders, such as the clinical type of AIS, thus providing a particularly nuanced perspective.
Our study also focused on cases admitted within 4.5 h of the symptom onset, which is a requirement for the implementation of IVT. While earlier studies touched upon regional differences in IVT and MT rates, they often did not specify whether these differences were due to factors before or after hospital admission7, 12, 32-35). Our analysis included a similar proportion of hyperacute cases from both rural and urban areas, suggesting that poorer outcomes in rural settings might be related more to in-hospital factors than to prehospital delays.
In summary, this study adds to the existing body of work by offering findings related to functional outcomes and the significance of the hyperacute phase of AIS, especially in rural areas.
InterpretationThe independent association between the degree of rurality and the AIS prognosis shown in this study and a slightly stronger trend seen in the hyperacute phase than X [please define X] might be explained by the disadvantages pertaining to in-hospital factors, including medical personnel and hospital infrastructure, as well as unmeasured patient factors inherent in rural areas.
First, in-hospital factors in rural areas may be related to geographical maldistribution of medical personnel and hospital infrastructure36-38). Indeed, a correlation was reported between RIJ and the geographic maldistribution of physicians25), as has been reported worldwide38). The distribution of stroke specialists is reportedly uneven, showing no correlation with the number of full-time physicians or hospital beds39), which may have affected the inability to perform IVT and MT, even when patients arrived at the hospital within the appropriate timeframe. In addition, the maldistribution of physiotherapists may also have affected the availability of rehabilitation in the acute phase37), and the lack of medical social workers may have hindered the efficient coordination of hospital transfers. The rurality difference in the length of stay may have been influenced by the uneven distribution of medical personnel40). Indeed, a questionnaire survey conducted in the United States also revealed a shortage of medical personnel in remote areas and differences in stroke care capacity and services among hospitals, suggesting that a similar situation exists in Japan41). The above suggests that the maldistribution of medical personnel and the limited infrastructure of facilities in rural areas may have been factors impacting outcomes.
Second, unmeasured patient factors, such as inadequate screening, have been cited in previous studies. Rural patients also reported higher rates of comorbidities that were not prescreened or detected as comorbidities than urban patients8, 10). Notably, in the present study, the proportion of cases with previously diagnosed AF was similar between the rural and urban groups, but the proportion of cases with newly diagnosed AF, i.e. AF detected during hospitalization, was higher in the rural group than in the urban group (rural: 6.5%, urban: 3.1%). Other possible influences include selection bias, such as rural patients being less likely than urban patients to be seen for minor illnesses, and unmeasured confounding, such as rural patients being more vulnerable than urban patients. In summary, rural patients may include individuals with an undiagnosed risk, and the true slope of the risk factor prevalence may therefore be greater than indicated42).
Finally, regarding the prehospital transport component, the rural and urban groups had almost the same percentage of patients that could be transferred to the hospital during the hyperacute phase. This may not have had a significant impact on transport time because of the Japanese system, in which patients can request an ambulance without co-payment, and the well-developed transport network in Japan43). Although previous studies have shown that delays in recognition and transport time in rural areas are associated with poor outcomes in these areas, the causes of our study results may be different44, 45). Therefore, in-hospital and unmeasured patient factors in rural areas may, at least in part, explain the urban-rural gap in AIS outcomes.
Clinical ImplicationsThe results of this study highlight the actionable targets for public health interventions aimed at improving AIS outcomes in rural areas. The PAF, estimated under the assumption that the estimates in this study are all causal and that the study sample represents the healthcare system of typical general hospitals in Japan, indicated that 14.8% of poor functional outcomes in AIS could be averted by addressing the disparities between rural and urban healthcare. This corresponded to 692 of the 4,614 rural patients among the 17,516 enrolled in our study. This finding provides a measurable target for healthcare policy improvements, particularly for reducing the risk of AIS in rural areas. Regional differences in the implementation of reperfusion therapies, such as IVT and MT, are likely to involve a combination of factors, ranging from facility infrastructure to maldistribution of specialists, along with unmeasured patient factors. Issues such as inequities in the distribution of specialists and medical staff will also change over time and will be difficult to address through centralized interventions and policies. Data-driven improvements in stroke care systems may effectively reduce regional disparities, such as those observed in this study, and minimize the burden on the healthcare system.
In addition, with the accumulation of evidence in recent years46, 47), hyperacute treatment of AIS is evolving rapidly, and ever-greater demands are being placed on specialists as the number of indicated cases increases. In Japan, there has been a move to meet the demand for MT by establishing a qualification with relaxed requirements for certification as a “practitioner of mechanical thrombectomy” in addition to “specialist of neuroendovascular therapy” starting in 2020. Similar efforts have been made in other countries to increase the number of physicians who can perform MT and improve the speed at which MT can be performed48). There may be an urgent need to create a system that allows for these important efforts to reach high-risk rural areas more effectively.
Several limitations associated with the present study warrant mention. First, the baseline Rankin Scale, National Institutes of Health Stroke Scale, and long-term functional outcomes were not included in the database. The lack of a baseline Rankin Scale made it difficult to speculate whether remote vulnerability was a factor before or after onset; however, baseline comorbidity information may have covered that vulnerability to some extent. In addition, stroke severity is generally assessed using the National Institutes of Health Stroke Scale; however, in the present study, as in previous studies, the level of consciousness was used as a proxy. This was considered a limitation because it may not accurately reflect the stroke severity49). In this study, outcomes were assessed at the time of discharge from the hospital, which was considered a limitation because the timing of discharge differed among cases, and the timing of outcome assessment should have been standardized, such as after three or six months. However, considering the actual practice in Japan, it is possible that the timing of the evaluation was constant for post-acute care status since patients were discharged home or moved to rehabilitation or convalescent hospitals when treatment at acute care hospitals was stable. Second, individual socioeconomic status could not be obtained. However, while we believe that the direct pathway from socioeconomic status to the prognosis could not be adjusted for, the pathway through risk factors regarded as highly specific was adjustable, as previous studies suggested that stroke risk factors (such as diabetes and smoking) might be intermediate factors8). Third, because this database is based on the hospital network of Saiseikai, it cannot be transferred to extremely rural areas (e.g. remote islands) where these comprehensive facilities do not exist. In addition, generalizability may be limited because facilities in the Tohoku region of Japan, which has a particularly high RIJ and is also a medically underpopulated area, were not included in this study50). Fourth, facility factors that varied over time may not have been adjusted for. The number of specialists at the facility and opening of the Stroke Care Unit varied over time. We believe that the effect of facility factors is small because the analysis was carefully adjusted for hospital size, hospital volume, and PSC and also took random effects into account, but it is still possible that they were not fully adjusted for.
Despite these limitations, the use of a high-quality population database allowed us to capture functional outcomes and adjust for multiple risk factors, thereby potentially providing reasonable estimates of the association between rural residence and stroke incidence and outcomes. Furthermore, the direction and interpretation of the present findings are likely to be generalizable to other high-income countries and regions.
This study demonstrated an independent association between rurality and poor functional outcomes in AIS patients in Japan, with a slightly stronger impact observed in hyperacute cases than X [please define X]. These poor outcomes suggest substantial population impacts and highlight the need for targeted health care policies. Addressing the maldistribution of specialists and medical staff may serve as an effective strategy for reducing these regional disparities, particularly for high-risk populations in rural areas. Future research should focus on generating additional evidence to inform and refine healthcare policies aimed at minimizing these disparities.
The prospective collection and subsequent analysis of the database to be used in this study was approved by the Institutional Ethics Review Board of Saiseikai Shiga Hospital (notice number: 226) and Tokyo Saiseikai Central Hospital, which is the core hospital associated with this registry (approval number: 28-15).
The ethics committee waived the requirement for informed consent because of the anonymity of data.
Not applicable.
The data that support the findings of this study are available from the Saiseikai Research group, and there are restrictions that apply to the availability of these data, which were used under license for the current study and are thus not publicly available.
Koji Kawakami has received research funds from Eisai Co., Ltd., Kyowa Kirin Co., Ltd., Mitsubishi Corporation, OMRON Corporation, Real World Data Co., Ltd., Sumitomo Pharma Co., Ltd., and Toppan Inc.; consulting fees from Advanced Medical Care Inc., JMDC Inc., LEBER Inc., and Shin Nippon Biomedical Laboratories Ltd.; executive compensation from Cancer Intelligence Care Systems, Inc.; and held stock in Real World Data Co., Ltd.
None.
GF conceived and designed the study, analyzed and interpreted the data, and drafted the manuscript. NK designed the study, interpreted the data, critically revised the manuscript for important intellectual content, and approved the final manuscript. HO and AF collected and assembled the data, critically revised the manuscript for important intellectual content, and approved the final manuscript. KK critically revised the article for important intellectual content and approved the final version.
We thank the members of the Saiseikai Stroke Research Group for their assistance with clinical data collection.
This registry collected patient data and stroke profiles, including age, sex, and stroke type (ischemic stroke, hemorrhagic stroke, or subarachnoid hemorrhage). In the current study, the collection of this database began in April 2013 and ended in March 2021 (27,691 cases). This study was approved by the medical ethics board in each institute of the Saiseikai Stroke Research Group. Since we obtained general consent to use clinical data from each subject on admission, individual written informed consent was not obtained at the time of the study based on the Ethical Guideline for Medical and Health Research Involving Human Subjects outlined by the Ministry of Education, Culture, Sports, Science and Technology and the Ministry of Health, Labor and Welfare in Japan.
The datasheet contained more than 50 items such as patient characteristics, date of admission, past medical history, medication, time of transportation, consciousness at the time of admission, imaging findings, acute-phase treatment, surgical or endovascular intervention, mortality, and modified Rankin Scales at discharge, and in-hospital recurrence of stroke. Stroke diagnoses were classified according to The Trial of Org 10172 in Acute Stroke Treatment (TOAST)-criteria1).The diagnosis of vascular risk factors (hypertension, diabetes mellitus, and dyslipidemia) was made according to the Japanese diagnostic criteria. Hypertension was defined as a history of hypertension or use of medication for hypertension, as were diabetes and dyslipidemia. Chronic kidney disease was defined as an estimated glomerular filtration rate ＜60 mL/min; proteinuria was not assessed. Congestive heart failure was defined as a history of heart failure and its current treatment. Smoking was defined as current smoking status or past smoker with smoking within one year. These parameters are collected by representative administrators of the hospital based on electric in-hospital charts.
Affiliations in the Saiseikai Stroke Database:Mito Saiseikai General Hospital, Ryugasaki Saiseikai Hospital, Saiseikai Kawaguchi General Hospital, Saiseikai Kurihashi Hospital, Tokyo Saiseikai Central Hospital, Saiseikai Yokohamashi Tobu Hospital, Saisekai Yokohamashi Nanbu Hospital, Saiseikai Wakakusa Hospital, Saiseikai Toyama Hospital, Fukui-ken Saiseikai Hospital, Saiseikai Matsuzaka General Hospital, Saiseikai Shiga Hospital, Saiseikai Kyoto Hospital, Saiseikai Noe Hospital, Saiseikai Chuwa Hospital, Saiseikai Wakayama Hospital, Saiseikai Sakaiminato General Hospital, Saiseikai Hiroshima Hospital, Saiseikai Yamaguchi General Hospital, Saiseikai Matsuyama Hospital, Saiseikai Saijo Hospital, Saiseikai Fukuoka General Hospital, Saiseikai Yahata General Hospital, Saiseikai Kumamoto Hospital, Saiseikai Misumi Hospital, Saiseikai Futsukaichi Hospital, Saiseikai Nagasaki Hospital.
Supplementary References1)Adams HP, Bendixen BH, Kappelle LJ, Biller J, Love BB, Gordon DL, Marsh EE: Classification of subtype of acute ischemic stroke. Stroke, 1993; 23: 35-41
2)Nakajima M, Okada Y, Sonoo T, Goto T: Development and validation of a novel method for converting the Japan Coma Scale to Glasgow Coma Scale. J Epidemiol, 2022; 22: 1-8
3)Kaneko M, Ikeda T, Inoue M, Sugiyama K, Saito M, Ohta R, Cooray U, Vingilis E, Freeman TR, Mathews M: Development and validation of a rurality index for healthcare research in Japan: a modified Delphi study. BMJ Open, 2023; 13: 68-800
