2019 Volume 1 Issue 3 Pages 86-94
BACKGROUND
The association between physician specialty and stroke mortality remains controversial. The present study evaluated the effectiveness of admission to a hospital with neurologic specialist staffing on 30 day in-hospital mortality after cerebral infarction, controlling for measured and unmeasured hospital and patient characteristics.
METHODS
The study involved 56,866 patients with cerebral infarction who were hospitalized within 1 day after onset between July 1, 2010 and March 31, 2012. Participants were identified using the Japanese Diagnosis Procedure Combination database linked to the Survey of Medical Institution and Hospital Report data and Survey of Physicians data. Well-staffed hospitals were defined as those with ≥3 board-certified neurologic specialists. Poorly-staffed hospitals were those with <3 board-certified neurologic specialists. The association between neurologic specialist staffing and 30-day in-hospital mortality was examined using a generalized estimation equations logistic regression model. Ordinary least square model and two-stage least square model using differential distance to hospitals as an instrumental variable were used for sensitivity analyses.
RESULTS
After adjusting for patient severity and hospital characteristics, 30-day in-hospital mortality at well-staffed hospitals was significantly lower than that at poorly-staffed hospitals (odds ratio, 0.89; 95% confidence interval, 0.79–0.99; P = 0.040). Hausman specification test suggested that admission to well-staffed hospital was exogenous. Ordinary least square model showed 30-day in-hospital mortality at well-staffed hospitals was significantly lower than that at poorly-staffed hospitals (risk difference, −0.6%; 95% confidence interval, −1.2% to −0.0%; P = 0. 044).
CONCLUSIONS
Hospitals with ≥3 neurologic specialists were associated with reduced 30-day in-hospital mortality for cerebral infarction.
The relationship between physician specialty and outcomes for patients with specific medical conditions has been frequently investigated [1–3]. Although board-certified specialty is generally considered to be a quality indicator for physicians and determinant of physician fee, the association between physician specialty and patient outcomes varies widely between specific patient conditions [1].
Previous studies on neurologist care and stroke outcomes have shown conflicting results, with some demonstrating neurologist care to be effective in improving stroke outcomes [4–6], while others did not [7–9]. However, these studies were limited by the impact of unmeasured confounding factors on their observational nature [7].
The present study aimed to determine if neurologic specialist staffing improved short-term mortality in patients with cerebral infarction, using a national inpatient database in Japan. We used instrumental variable method, a statistical method for quasi-experimental studies using observational data [10, 11]. Instrumental variable analyses were used to adjust for hospital- and patient-level unmeasured confounding factors.
The present study used the Diagnosis Procedure Combination (DPC) inpatient database. This is Japan’s national administrative claims and discharge abstract database, details of which have been described elsewhere [12]. The DPC database comprises data from 1119 hospitals, and includes data for 11 million patients discharged between July 1, 2010 and March 31, 2012, representing approximately 50% of all acute-care hospitalization in Japan. The database includes unique identifiers for hospitals; patient age and sex; diagnoses at admission, comorbidities at admission, and complications after admission recorded with text data in Japanese using International Classification of Diseases, Tenth Revision (ICD-10) codes; drugs and devices used; examinations and procedures; the date of stroke onset; modified Rankin Scale (mRS) at admission; Japan Coma Scale (JCS) at admission; length of stay; and discharge status.
Data were also used from the 2011 Survey of Medical Institutions and Hospital Report data and the 2010 Survey of Physicians, Dentists and Pharmacists [13] conducted by the Japanese Ministry of Health, Labour and Welfare. The Survey of Medical Institutions and Hospital Report is a census of hospitals in Japan conducted every 3 years [14]. The census collects structural information from each hospital such as bed numbers, physician and nurse numbers, and use of computed tomography (CT) or magnetic resonance imaging (MRI). The Survey of Physicians data includes physician employment status, workplace and position, and board certification status. Hospital characteristics information from both surveys was merged into the DPC inpatient data using the unique hospital identifiers.
All data were anonymous and informed consent for individual patients was waived. Our study was approved by the Institutional Review Boards and the Ethics Committee of The University of Tokyo.
NEUROLOGIC SPECIALIST STAFFINGIn Japan, three academic societies certify neurological specialty: the Japanese Society of Neurology [15], the Japan Neurosurgical Society [16] and the Japanese Society for Neuroendovascular Therapy [17]. It is difficult to differentiate the roles of the three certifications. Some physicians have only one certification, but some have two or more certifications. Thus, we aggregated these three different certifications of the neurological specialty to a single category. We divided the patients into two groups based on the number of neurologic specialists so that the numbers of hospitals in the groups were almost the same. About half of the hospitals had three or more board-certified neurologists. Thus, well-staffed hospitals were defined as those with ≥3 hoard-certified neurologists, while poorly-staffed hospitals were those with were defined as those with <3 hoard-certified neurologists.
PATIENT SELECTIONPatients who were transported to acute hospitals by ambulance between July 1, 2010 and March 31, 2012 for acute cerebral infarction (ICD10 codes: I63$) within 1 day after onset were selected for the study. Patients who were hospitalized for more than 180 days and those who were transferred from another hospital were excluded. In Japan, most stroke patients are reported to be transported to a hospital within 30 km from their residence by ambulance [18]. Therefore, patients who were admitted to a hospital more than 30 km from their residence were excluded because (i) such patients had their onset of cerebral infarction at a place other than their residence, or (ii) they lived more than 30 km away from the nearest hospital (well- or poorly-staffed) and may have been transported to a well-staffed hospital by helicopter. In addition, the Survey of Medical Institutions and Hospital Report in 2011 lacked data on hospitals in Fukushima because of the Great Earthquake, meaning patients from Fukushima prefecture were also excluded.
OUTCOME30-day in-hospital mortality was used as an outcome variable.
COVARIATESData for both patient- and hospital-level characteristics were used. Patient-level variables included age, sex, use of mechanical ventilation at admission, mRS scores at admission (0–2, 3–4, or 5), JCS scores at admission, cerebral infarction subtypes (I633 as atherosclerotic; I634 as cardioembolic; others as other I63x codes), Charlson Comorbidity Index (CCI) scores and use of recombinant tissue-type plasminogen activator (rt-PA). CCI is widely used for scoring patient comorbidity status, and has been validated in Japanese patients [19, 20]. The distance between patient residence and the hospital of admission and duration of onset to admission(days) was also included in the patient-level characteristics.
The hospital-level characteristics included hospital bed numbers, type of hospital (academic or non-academic), number of nurses, and number of physicians. Hospital volume represents the average number of patients with cerebral infarction per year in each hospital. As some studies have shown a relationship between hospital volume and mortality in patients with cerebral infarction [21], hospital volume was a further hospital-level characteristic. We divided hospitals volume into three categories so that the number of hospitals in each group was almost equal (≤18, 19–69, ≥70 patient per year). Population density of patient neighborhood area according to the results of the national census taken on 2010 was used as a proxy for patient residence characteristics. We divided population density into two categories so that the number of cities in each group was almost equal (≤249, 249< person/km2).
STATISTICAL ANALYSESStandardized difference (%) was used to assess the differences between the two groups. An absolute standardized difference of more than 10% indicates a significant difference between the groups [10]. We performed a multivariable logistic regression analysis for 30-day mortality. To account for within-hospital clustering, the regression model was fitted with a generalized estimating equation (GEE) constructed by R 3.0.2 with gee package. We used exchangeable correlation matrix as a working matrix in GEE.
INSTRUMENTAL VARIABLE METHODThe probability of admission to a well-staffed hospital could depend on measured and unmeasured patient characteristics. Therefore, the instrumental variable (IV) method was used for sensitivity analysis [23, 24].
IV assumes that (i) IV is associated with treatment assignment, (ii) IV is not directly associated with outcomes, and (iii) IV does not share confounders with outcomes. A differential distance (DD) was used for IV, defined as the difference between (i) distance from the patient’s residence to the nearest hospital (D1) and (ii) distance from the patient’s residence to the nearest well-staffed hospital (D2). DD was put into the model as a continuous variable. DD (D1–D2) is a widely used instrumental variable in health outcome research for treating unmeasured confounders [10, 25]. The effect of admission to a well-staffed hospital on mortality was estimated using a two-stage least squares model. The first-stage partial F statistics were evaluated to check for a weak instrument. We divided the patients into those with DD <1 km or DD ≥1 km so that the number of patients in the two groups were almost the same. We also compared patient characteristics between the groups with DD <1 km or DD ≥1 km. Hausman specification test conducted to confirm endogeneity of neurologic specialist staffing. Null hypothesis for the Hausman specification test is that instrumental variable is not endogenous. Hausman specification test examines difference in coefficients between two-stage least-square (2SLS) model and ordinary least square (OLS) model [26–28]. When the difference is significant, we should adopt the results of 2SLS model. Insignificant difference indicates that we can adopt the results of OLS model as well as 2SLS model. The two-stage least squares model was constructed by R 3.0.2 with the AER package.
SENSITIVITY ANALYSESWe constructed several sensitivity analyses to check robustness of our study. First, we further divided the group with neurological specialist staffing ≥3 into those with 3–6 and ≥7, and compared 30-day mortality among the groups with <3, 3–6, and ≥7 (model 1). Then, we performed further sensitivity analyses with cut-off points of neurologic specialist staffing being 2 or 4 (models 2 and 3). We also developed a model which included neurologist staffing divided by the number of hospital beds as the primary exposure (model 4).
Overall, 218,393 patients were admitted to hospitals with a diagnosis of cerebral infarction from July 1, 2010, to March 31, 2012. Of these, 85,393 patients were transported by ambulance within 1 day after onset. We excluded 28,527 patients according to the exclusion criteria and 4,999 patients with missing data. This gave a study sample of 56,866 patients. Table 1 shows the distribution of the number of neurologic specialists in 858 hospitals.
| Neurologic specialist staffing | Hospitals, n (%) | Patients, n (%) |
|---|---|---|
| <2 | 300 (35) | 7669 (13) |
| 2 | 96 (11) | 3852 (7) |
| 3 | 78 (9) | 5170 (9) |
| 4 to 5 | 143 (17) | 13806 (24) |
| 6 to 7 | 52 (6) | 5530 (10) |
| >7 | 189 (22) | 20839 (37) |
| Sum | 858 (100) | 56866 (100) |
Table 2 presents patient characteristics for each group. The patients treated in well-staffed hospitals (with neurologic specialist staffing ≥3) were significantly younger, had less severe functional deficit, and had less severe disturbance of consciousness than patients treated in the poorly-staffed hospitals (with neurologic specialist staffing <3). Patients treated in the well-staffed hospitals were more likely to be admitted to a stroke care unit and to undergo rt-PA treatment. Well-staffed hospitals had larger bed volumes, and more physicians, nurses than poorly-staffed hospitals. The right side of Table 1 shows patient characteristics in the groups with DD <1 km or ≥1 km. Patient-level characteristics were well balanced with standardized differences of less than 10.
| Neurologic specialist staffing <3 (n = 11521) |
Neurologic specialist staffing ≥3 (n = 45345) |
Standardized Difference (%) |
DD <1 km (n = 29395) |
DD ≥1 km (n = 27471) |
Standardized Difference (%) |
|
|---|---|---|---|---|---|---|
| Death in 30 days, n (%) | 924 (8.0) | 3186 (7.0) | 3.8 | 2063 (7.0) | 2047 (7.5) | 1.7 |
| Age (years), mean (sd) | 76.81 (11.82) | 75.45 (12.12) | 11.3 | 75.73 (12.10) | 75.72 (12.03) | 0.1 |
| Sex (female), n (%) | 5166 (44.8) | 19388 (42.8) | 4.2 | 12724 (43.3) | 11830 (43.1) | 0.4 |
| Length of stay (days), mean (sd) | 34.92 (33.83) | 31.08 (29.17) | 12.2 | 31.72 (29.90) | 32.00 (30.54) | 0.9 |
| modified Rankin Scale at admission, n (%) | 5.9 | 2.2 | ||||
| 0–1 | 903 (7.8) | 3781 (8.3) | 2497 (8.5) | 2187 (8.0) | ||
| 2–3 | 2706 (23.5) | 11662 (25.7) | 7339 (25.0) | 7029 (25.6) | ||
| 4–5 | 7912 (68.7) | 29902 (65.9) | 19559 (66.5) | 18255 (66.5) | ||
| Japan Coma Scale, n (%) | 8.8 | 2.8 | ||||
| Alert | 4519 (39.2) | 16571 (36.5) | 10834 (36.9) | 10256 (37.3) | ||
| Drowsiness | 4278 (37.1) | 18522 (40.8) | 11946 (40.6) | 10854 (39.5) | ||
| Somnolence | 1645 (14.3) | 6635 (14.6) | 4272 (14.5) | 4008 (14.6) | ||
| Coma | 1079 (9.4) | 3617 (8.0) | 2343 (8.0) | 2353 (8.6) | ||
| Charlson Comorbidity Index, n (%) | 8.0 | 1.3 | ||||
| 0–1 | 7732 (67.1) | 32104 (70.8) | 20677 (70.3) | 19159 (69.7) | ||
| 2–3 | 3349 (29.1) | 11766 (25.9) | 7730 (26.3) | 7385 (26.9) | ||
| 4– | 440 (3.8) | 1475 (3.3) | 988 (3.4) | 927 (3.4) | ||
| Cerebral infarction subtype, n (%) | 17.5 | 1.3 | ||||
| Atherosclerotic | 3617 (31.4) | 13557 (29.9) | 8889 (30.2) | 8285 (30.2) | ||
| Cardioembolic | 2473 (21.5) | 13089 (28.9) | 8016 (27.3) | 7546 (27.5) | ||
| Others | 5431 (47.1) | 18699 (41.2) | 12490 (442.5) | 11640 (42.4) | ||
| Stroke care unit admission, n (%) | 591 (5.1) | 4298 (9.5) | 16.8 | 2625 (8.9) | 2264 (8.2) | 2.5 |
| Recombinant tissue plasminogen activator, n (%) | 618 (5.4) | 3487 (7.7) | 9.4 | 2183 (7.4) | 1922 (7.0) | 1.7 |
| Magnetic resonance imaging at admission, n (%) | 7084 (61.5) | 30423 (67.1) | 11.7 | 19111 (65.0) | 18396 (67.0) | 4.1 |
| Computed tomography at admission, n (%) | 7279 (63.2) | 31921 (70.4) | 15.4 | 20607 (70.1) | 18593 (67.7) | 5.2 |
| Intubation at admission, n (%) | 130 (1.1) | 576 (1.3) | 1.3 | 364 (1.2) | 342 (1.2) | 0.1 |
| Bed numbers, mean (sd) | 281 (134) | 491 (245) | 106.4 | 458 (233) | 438 (252) | 7.9 |
| Number of physicians, mean (sd) | 48 (33) | 130 (124) | 90.8 | 117 (116) | 110 (117) | 5.8 |
| Number of nurses, mean (sd) | 210 (121) | 433 (242) | 116.7 | 397 (233) | 378 (248) | 8.3 |
| Academic hospital, n (%) | 264 (2.3) | 6342 (14.0) | 43.8 | 3335 (11.3) | 3271 (11.9) | 1.8 |
| Annual hospital volume of cerebral infarction patients | 91.4 | 13.5 | ||||
| ≤18 | 1759 (15.3) | 433 (1.0) | 804 (2.7) | 1388 (5.1) | ||
| 19–69 | 4629 (40.2) | 7277 (16.0) | 5874 (20.0) | 6032 (22.0) | ||
| ≥70 | 5133 (44.6) | 37635 (83.0) | 22717 (77.3) | 20051 (73.0) | ||
| Distance between patient residence and admitted hospital (km), mean (sd) | 5.15 (4.84) | 6.06 (5.45) | 17.7 | 4.28 (4.10) | 7.58 (5.96) | 64.6 |
| Duration between stroke onset and admission (days), mean (sd) | 0.09 (0.29) | 0.11 (0.31) | 4.3 | 0.10 (0.31) | 0.10 (0.30) | 1.3 |
| Population density of patient living area ≤ 249 person/km2, n (%) | 2550 (22.1) | 5687 (12.5) | 25.5 | 3018 (10.3) | 5219 (19.0) | 24.9 |
| Differential Distance (km), mean (sd) | 4.98 (6.39) | 1.85 (3.23) | 61.9 | |||
| Neurologic specialist staffing ≥3 (%) | 25828 (87.9) | 19517 (71.0) | 42.6 |
Crude 30-day mortality for patients treated in well-staffed hospitals was lower than for patients treated in the poorly-staffed hospitals (8.0% vs 7.0%; odds ratio [OR], 0.83; 95% CI, 0.75–0.91).
Table 3 presents the adjusted ORs of patient and hospital-level characteristics for 30-day mortality in the GEE logistic regression model. Compared with patients in poorly-staffed hospitals, those in well-staffed hospitals had lower adjusted 30 days mortality (OR, 0.91; 95% CI, 0.82–1.00; P = 0.045) with adjustment for patient- and hospital-level characteristics.
| Odds ratio (95% confidence interval) | P value | |
|---|---|---|
| Neurologic specialist staffing | ||
| <3 | Reference | |
| ≥3 | 0.91 (0.82–1.00) | 0.045 |
| Age (years) | 1.02 (1.02–1.02) | 0.000 |
| Sex | ||
| Male | Reference | |
| Female | 0.95 (0.88–1.02) | 0.175 |
| Charlson Comorbidity Index | ||
| 0–1 | Reference | |
| 2–3 | 0.84 (0.77–0.91) | 0.000 |
| 4– | 1.19 (1.00–1.42) | 0.045 |
| modified Rankin Scale | ||
| 0–1 | Reference | |
| 2–3 | 0.94 (0.71–1.24) | 0.663 |
| 4–5 | 2.86 (2.23–3.67) | 0.000 |
| Japan Coma Scale | ||
| Alert | Reference | |
| Drowsiness | 2.22 (1.95–2.54) | 0.000 |
| Somnolence | 5.52 (4.80–6.36) | 0.000 |
| Coma | 15.17(13.14–17.50) | 0.000 |
| Cerebral infarction subtype | ||
| Atherosclerotic | Reference | |
| Cardioembolic | 1.90 (1.72–2.10) | 0.000 |
| Others | 1.62 (1.46–1.78) | 0.000 |
| Intubation at admission | 3.68 (3.05–4.44) | 0.000 |
| Using rtPA | 0.97 (0.86–1.10) | 0.636 |
| Distance between patient residence and hospital | 1.00 (0.99–1.00) | 0.275 |
| Bed numbers (divided by 100) | 1.07 (1.02–1.11) | 0.004 |
| Number of physicians (divided by 100) | 0.98 (0.91–1.06) | 0.627 |
| Number of nurses (divided by 100) | 0.96 (0.91–1.00) | 0.070 |
| Academic hospital | 0.89 (0.74–1.06) | 0.196 |
| Hospital volume of cerebral infarction patients | ||
| ≤18 | Reference | |
| 19–69 | 1.03 (0.85–1.26) | 0.729 |
| ≥70 | 1.01 (0.84–1.23) | 0.890 |
| Population density of patient living area ≤249 person/km2 | 1.13 (1.03–1.25) | 0.013 |
| Duration between stroke onset and admission | 0.85 (0.75–0.97) | 0.017 |
rtPA, recombinant tissue-type plasminogen activator
Table 4 shows the adjusted coefficient estimator of 30-day mortality for well-staffed hospitals in the OLS and 2SLS models. Admission to a well-staffed hospital was significantly associated with lower 30-day mortality (coefficient = −0.006; 95% CI, −0.012 to −0.001; P = 0.031) in OLS model, but not significantly associated in 2SLS model (coefficient = −0.018; 95% CI, −0.043 to 0.006; P = 0.142) with adjustment for patient- and hospital-level characteristics. The partial F statistics from the first stage regression indicated that the differential distance was an instrumental variable with sufficient strength for predicting admission to well-staffed hospitals. The Hausman specification test did not reject the null hypothesis that admission to a well-staffed hospital was exogenous.
| Coefficient (95% confidence interval) | P value | |
|---|---|---|
| Neurologic specialist staffing ≥3 in OLS model | −0.006 (−0.012, −0.001) | 0.031 |
| Neurologic specialist staffing ≥3 in 2SLS model | -0.019 (−0.044, 0.006) | 0.142 |
| F statistics | 5761 | <0.001 |
| Hausman specification test | 0.260 | 0.610 |
OLS, Ordinary least square; 2SLS, Two-stage least square
Table 5 shows the adjusted ORs for each model in the sensitivity analyses. In model 1, the group with neurological specialist staffing 3–6 was significantly associated with lower 30-day mortality compared with the group with <3. Although not significant, the group with ≥7 had a lower 30-day mortality than the group with <3. The ORs for the groups with 3–6 and ≥7 were similar, being 0.89 and 0.91, respectively. When we compared between neurologic specialist staffing <2 and ≥2, no significant difference in 30-day mortality was shown (model 2). Neurologic specialist staffing ≥4 was significantly associated with lower 30-day mortality than neurologic specialist staffing <4 (model 3). The neurologist staffing per 100 beds was significantly associated with 30-day mortality (model 4).
| Model name | Variable | Odds ratio (95% confidence interval) | P value |
|---|---|---|---|
| Model 1 | Neurological Specialist Staffing <3 | Reference | |
| Neurological Specialist Staffing = 3 to 6 | 0.89 (0.79–1.00) | 0.038 | |
| Neurological Specialist Staffing ≥7 | 0.91 (0.83–1.01) | 0.073 | |
| Model 2 | Neurological Specialist Staffing <2 | Reference | |
| Neurological Specialist Staffing ≥2 | 1.00 (0.89–1.12) | 0.947 | |
| Model 3 | Neurological Specialist Staffing <4 | Reference | |
| Neurological Specialist Staffing ≥4 | 0.88 (0.80–0.96) | 0.003 | |
| Model 4 | Neurological Specialist Staffing per 100 beds | 0.98(0.96–1.00) | 0.027 |
Our results showed that patients who were admitted to well-staffed hospitals exhibited significantly lower 30-day mortality than patients treated in poorly-staffed hospitals with adjustment for patient-level and hospital-level factors.
Well-staffed hospitals exhibited significantly lower 30-day mortality than patients treated in poorly-staffed hospitals in the OLS model, but not significant in the 2SLS model. In the present study, null hypothesis for the Hausman test was that admission to well-staffed hospitals was not endogenous. The results of the test indicates that the null hypothesis was not rejected; that is, if the measured confounders were equal, there was no evidence that admission to well-staffed hospitals was endogenous. Consequently, we can adopt the results of the OLS model as well as the 2SLS model. Generally, variance of instrumental variable estimator is larger than that of OLS estimator [23]. Based on the results of the OLS model, patients who were admitted to well-staffed hospitals exhibited significantly lower 30-day mortality than those who were admitted to poorly-staffed hospitals.
A small-size randomized control study (n = 232) in 1987–1989 showed that patients treated in the neurology department had a shorter length of stay and a lower likelihood of being discharged to home compared with those treated in the department of medicine, but showed no significant difference in mortality [5]. In our study, better neurologic specialist staffing was associated with a reduction in 30-day mortality. Previous observational studies showed mixed results for the association between neurologist care and improved outcomes in stroke patients [4, 6, 7, 9]. Smith et al found that neurologist care reduced 30-day mortality compared with generalist care in an analysis of 44,099 stroke patients [6]. They included patients’ socioeconomic backgrounds and comorbidities in their analysis, but did not include severity of disability or score of impaired consciousness. Gillum and Johnson found no significant association between in-hospital mortality and neurologist care using instrumental variable analysis [7]. However, although they accounted for patient-level unmeasured confounders, they controlled only a few hospital-level confounders, meaning their results may be biased by hospital-level confounding between instrumental variable outcomes [29].
The results of model 1 suggest that neurological specialist staffing ≥3 were associated with better outcomes, but incremental effect of the number of neurological specialist staffing was not detected. These results for models 1, 2, and 3 suggest that neurologic specialist staffing = 3 was an appropriate cut-off value for well/poorly staffed hospitals.
The present study has several strengths. First, as national inpatient data were used, the findings are generalizable to all patients with cerebral infarction in Japan. Second, using instrumental variable analysis, unmeasured confounders which may have had an impact on mortality were adjusted for.
Our findings indicated that admission to hospitals well-staffed by neurologic specialists was associated with improved outcomes for acute ischemic stroke patients. There are several possible explanations for the better outcomes of cerebral infarction in hospitals with better neurologic specialist staffing [4, 30]. First, neurologists may be well trained and ready for stroke care. Stroke care consists of several components including diagnosis, treatment, avoiding complications and rehabilitation. Previous studies showed that neurologic specialists and generalists were different in terms of several practice patterns. For example, neurologists were more likely to use CT, MRT, and rt-PA [7, 9]. Use of CT and MRI can provide important information for determining diagnosis of stroke and differentiating cerebral infarction from hemorrhage. Thus, prompt use of such modalities may affect treatment selection and then patients’ prognosis. The present study did show that hospitals with better neurologic specialist staffing were more likely to use rt-PA. This suggests that neurologists may have been more aware of the effect of rt-PA at the very early phase from onset. Second, we speculate that neurologist as a specialist may have roles of the leader in a stroke care team. They are expected to organize prompt diagnosis and treatment, nursing care, and rehabilitation. Based on our results, these neurologic specialist staffing effects may not be sufficient if there are only 1 or 2 neurologists in a hospital. A mortality-reducing effect may only be expected if a hospital has ≥3 neurologists. Although the reason for this remains unclear, we speculated that a hospital with only 1 or 2 neurologists cannot maintain 365-day 24-hour readiness for acute stroke care admissions. Further studies are needed to verify the mechanism of better outcomes associated with better neurologic specialist staffing.
Our analysis did not show a significant effect of rtPA, possibly because the numbers of rtPA users were small. Further studies are needed to verify the effectiveness of rtPA.
Our findings highlight the relevance of establishing a certification system for stroke centers for any country. To date, there are guidelines for the certification of stroke centers, and variation in defining neurologic specialist staffing. The guideline from the European Stroke Organization recommends involving a neurologist or stroke physician in acute care for stroke patients [31]. This recommendation was based on previous observational studies or small sized randomized controlled trials (RCT) [4, 32]. On the other hand, the Primary Stroke Center guideline from the Brain Attack Coalition in the United States did not mandate that a neurologist be the sole provider of acute care for stroke patients [32]. Our study adds to the evidence on the association between neurologic specialist staffing and better outcomes, which justifies development of health policies to support better neurologic specialist staffing of stroke centers.
The present study has some limitations. First, our instrumental variable analysis may not completely exclude instrument-outcome confounder bias arising from hospital characteristics or patient residence characteristics [29]. For example, owing to the limited census data available on hospital characteristics, we could not separate the neurologist care effect from hospital processes for stroke management. Second, our data did not contain post-discharge long-term outcomes. Third, the present study used physical DD. However, time DD might be more significant than physical DD in a small country like Japan because the public transportation have relatively developed compared to a large country like the United States. In order to calculate DD, we selected patients who were admitted by ambulance. It could lead to a lack of generalizability.
In summary, hospitals well-staffed with neurologic specialists were associated with reduced 30-day mortality of ischemic stroke. This suggests that neurologist care should be included in the criteria for certifying acute stroke centers.
This study was founded by Grants-in-Aid for Research on Policy Planning and Evaluation (grant numbers: H27-Policy-Designated-009, H27-Policy-Strategy-011, and H26-Statistics-001) from the Japanese Ministry of Health, Labour and Welfare, and by a Grant-in-Aid for Young Scientists B (grant number: 26860455) from the Japanese Ministry of Culture, Sports, Science and Technology. There were no other conflict of interest disclosure.
