Prediction of Length of Stay of First-Ever Ischemic Stroke
Background and Purpose— Accurate information about hospital resource utilization is necessary for management of healthcare service. The purpose of this study was to determine the demographic and clinical predictors of length of hospital stay (LOS) of acute care hospitalization for first-ever ischemic stroke patients.
Methods— A group of 330 patients who suffered from first-ever ischemic stroke and were consecutively admitted to a medical center in southern Taiwan were followed prospectively. Because our intention was to identify the major predictors of LOS from the information available at admission, we evaluated only those factors that could be assessed at the time of admission. Univariate analysis and multiple regression analysis were used to identify the main predictors of LOS.
Results— The median LOS was 7 days (mean, 11 days; range, 1 to 122 days). Among the prespecified demographic and clinical characteristics, National Institutes of Health Stroke Scale (NIHSS) score at admission, the quadratic term of the initial NIHSS score, modified Barthel Index score at admission, small-vessel occlusion stroke, sex, and smoking were the main explanatory factors for LOS. In particular, for each 1-point increase in the total score of NIHSS, LOS increased approximately 1 day for patients with mild or moderate (score 0 to 15 points) neurological impairments, while LOS decreased approximately 1 day for patients with severe (score >15 points) neurological impairments.
Conclusions— The severity of stroke, as rated by the total score on NIHSS, is an important factor that influences LOS after acute stroke hospitalization.
Stroke is the second most common cause of mortality in Taiwan.1 The incidence rate of first-ever stroke for people aged ≥36 years has been reported to be 330 per 100 000, and 71% was due to cerebral infarction.2 The provision of stroke care imposes a major economic burden on the national healthcare system. In facing shrinkage of resources for healthcare, the importance of hospitalization costs cannot be overlooked. In addition, evidence from clinical trials suggests that ischemic stroke can be managed as a medical emergency and outcomes can be improved by using the thrombolytic agent recombinant tissue plasminogen activator (rtPA).3 While it facilitates recovery of some ischemic stroke patients, the cost-effectiveness of this very expensive treatment may vary from country to country because of the different practice patterns of stroke management.4–9⇓⇓⇓⇓⇓ Among variables contributing to the total costs of hospitalization, the use of length of hospital stay (LOS) as a marker of resource utilization is highly predictive of inpatient costs.7,10,11⇓⇓ Hence, accurate prediction of LOS has become increasingly important for the administration of hospitals and healthcare systems.
The objective of this study was to examine the relative importance of demographic, clinical, and functional factors that can be assessed at the time of admission for predicting LOS of acute care hospitalization for patients with first-ever ischemic stroke.
Subjects and Methods
Our study included 368 patients with acute first-ever ischemic stroke consecutively admitted to the Department of Neurology, Chang Gung Memorial Hospital, Kaohsiung, Taiwan (KCGMH) between September 1998 and October 1999. The hospital is a 1900-bed nonprofit proprietary hospital, providing medical-center-level healthcare in an area with a population of approximately 3 million in southern Taiwan. It was the third largest hospital in Taiwan, by total number of medical personnel, at the end of 1999. There are other hospitals, including 2 medical centers and 24 community hospitals, in the same area. All hospitals provide 24-hour CT scanning and laboratory facilities located adjacent to the respective emergency departments. Acute therapy with a thrombolytic agent has not, to date, been approved for clinical use in Taiwan and has been used off-label only in limited cases.
Patients eligible for this prospective follow-up study were enrolled if they met the following criteria: (1) diagnosis with first-ever ischemic stroke and identified as having no history of stroke or transient ischemic attack; and (2) diagnosis of acute stroke and identified as having the qualified ischemic stroke onset within 48 hours before admission. Informed consent was waived by the review board of KCGMH. All the patients in this study had brain CT exams or brain MRIs. All the participating investigators had completed the videotaped training program from the Henry Ford Foundation so that they would perform the evaluations of the National Institutes of Health Stroke Scale (NIHSS) accurately and consistently. The principal investigator assessed the eligible patients once per week.
The discharge date was recorded as the date the patient died or was discharged to home, another hospital, a rehabilitation facility, or any place other than the Department of Neurology in KCGMH. The destination of disposition was made according to the condition of the patients, consideration of the families, and assessment of rehabilitation doctors, according to the practice patterns of stroke management in this area.
On the basis of evidence in the literature and clinical judgment, predictor variables were prospectively determined. Because our intention was to identify the major predictors of LOS from the information available at admission, we evaluated only those factors that could be assessed at the time of admission. Data collected prospectively included the patient’s age and sex, stroke severity at admission, functional independence status at admission, hours after stroke onset (within 24 hours or not), comorbidity (presence or absence of history of hypertension, diabetes mellitus, or hypercholesterolemia), smoking, congestive heart failure, valvular heart disease, atrial fibrillation, history of cardiac disease (history of arrhythmia, angina pectoris, ischemic heart disease, and/or abnormalities identified from the initial ECG), stroke subtypes, and serum total cholesterol and triglyceride levels at admission.
The diagnosis of the subtype of ischemic stroke can be difficult in the beginning of stroke management, and the diagnosis of stroke subtypes often changes as the results of ancillary diagnostic tests become available.12,13⇓ However, we chose to determine the ischemic stroke subtype according to the criteria of the Trial of ORG 10172 in Acute Stroke Treatment.14 Because in our study the number of patients with the cardioembolism subtype of stoke is relatively small, and in view of evidence in the literature that patients with small-vessel occlusion had better outcome,15,16⇓ stroke subtype was dichotomized as small-vessel occlusion or not in our analysis.
All patients had their stroke severity assessed by the NIHSS (range, 0 to 38). Functional independence status at admission was measured with the use of the modified Barthel Index (MBI) based on a scale from 0 to 20 (20=normal). A higher value on the NIHSS indicates more severe neurological impairment, whereas a higher MBI score on admission indicates less severe functional impairment.
In our analysis, age, NIHSS score, MBI score, serum total cholesterol level, and serum triglyceride level were considered continuous variables. Time after stroke onset was classified as onset <24 hours versus ≥24 hours because no detailed information was available. The other category variables were classified as present versus absent.
All predictor variables were first examined by means of univariate analysis to assess the importance of each of them on the LOS. Because the distribution of LOS was positively skewed, the results of the analysis were summarized by median and 25th and 75th percentile values. Additionally, we used the Mann-Whitney U test or the Kruskal-Wallis test to examine the differences between/among different stratified groups. Significant P value was set at 0.05. In the univariate analysis, for the aforementioned continuous variables, we chose certain cutoff points to stratify patients for comparison. The choices, however, were not arbitrary. Age of 65 years was a common cutoff point in the literature. The classifications of initial NIHSS score and initial MBI score were suggested by an often used clinical dichotomy.15,17,18⇓⇓ Following the classification suggested by the American Heart Association, we categorized serum total cholesterol level as desirable (<200 mg/dL), borderline high (200 to 239 mg/dL), and high risk (≥240 mg/dL); serum triglyceride level was classified as normal (<150 mg/dL), borderline high (150 to 199 mg/dL), and high (≥200 mg/dL).
Of the original 368 patients, a total of 38 patients were excluded from the analysis. Data on both lipid levels were missing in 31 patients, and data on serum triglyceride level were missing in another 2 patients. Patients who received acute thrombolytic therapy were also eliminated from the analysis (2 patients received intravenous thrombolysis with rtPA within 3 hours after onset, and 6 patients received intra-arterial thrombolysis with urokinase 6 hours after onset). This resulted in a total of 330 patients included in the analysis (data on both lipid levels were also missing in 3 patients who received intra-arterial thrombolysis with urokinase). In particular, we chose not to exclude outliers if there was no evidence of a mistake.
The logarithmic transformation of LOS produced a more normal distribution (the Kolmogorov-Smirnov Z value changed from 4.77 to 1.71). In this study the natural logarithm of LOS (lnLOS) was used as the dependent variable for our multiple regression analysis. The predicted lnLOS values were then transformed back to obtain the predicted LOS. For simplicity, we did not consider the potentially embedded bias of this transformation.18
Before multiple regression analysis, we plotted lnLOS against each continuous predictor variable to get an idea of model selection. In view of the nonlinear relationship shown on the lnLOS–NIHSS score scatter diagram, we decided to create a regression model that, in addition to including the linear effects of those prespecified predictors, contained a quadratic term of the initial NIHSS score. The cubic term for the NIHSS score would also be examined. Terms representing interaction effects were not considered because little prior knowledge and/or research suggested such implications.
To avoid overfitting the data to the model,19 all the prespecified predictor variables were entered simultaneously. The model reliability was assessed by the split-sample approach.20,21⇓ Two thirds of the patients were randomly assigned to the training group and one third to the validation group. Additionally, we performed a tougher test by splitting the data in a nonrandom way21: two thirds of the patients who were admitted in time periods earlier than the remaining one third were assigned to the training group. All analyses were done with the use of SPSS version 10.0 for Windows (SPSS Inc).
Table 1 summarizes the demographic and clinical characteristics of the 330 patients. The mean age was 64.3±12.5 years (median, 66 years; range, 18 to 90 years). There were 193 men (58.5%) and 137 women (41.5%) in our study. Mean score of the NIHSS was 8.6±8.4 (median, 6.0). Mean score of the MBI was 11.1±6.5 (median, 12.0). Median LOS was 7 days (mean, 11±14; range, 1 to 122). Stroke subtype was qualified as cardioembolism in 39 (12%), large-artery atherosclerosis in 83 (25%), small-vessel occlusion in 206 (62%), and undetermined etiology in 2 patients (1%).
Approximately four fifths (81%) of patients had an onset-to-admission interval of <24 hours. Sixty-seven percent of patients had at least 1 kind of comorbidity. Twenty-one percent of patients were smokers. Patients of this observational study did not have a high prevalence rate of congestive heart failure (4%), valvular heart disease (2%), or atrial fibrillation (7%). Forty percent of patients had some previous cardiac disease. Of these 330 patients, 17 (5%) died during the acute care hospitalization, 86 (26%) were transferred to the rehabilitation ward located in the same hospital for intensive rehabilitation, and 227 (69%) were discharged to their homes or other care facilities.
As shown in Table 2, among demographic characteristics, patients with atrial fibrillation had a significantly longer LOS. No significant association between LOS and age (≤65 versus >65 years), sex, comorbidity, smoking, congestive heart failure, valvular heart disease, or history of cardiac disease was observed. Among clinical characteristics, mean LOS, not surprisingly, differed significantly by initial neurological severity and by functional severity. Severity of stroke in general prolonged LOS. However, as mentioned, the relationship was expected to be nonlinear. Mean LOS was significantly longer for patients admitted to the hospital sooner (onset <24 hours). There was a strong association between mean LOS and stroke subtype. Small-vessel occlusion stroke was associated with shorter LOS (median, 6 versus 11 days; mean, 7 versus 18). Neither serum total cholesterol level nor serum triglyceride level was significantly associated with LOS.
To further explore whether the noted LOS difference between stratified groups was related to stroke severity, we compared the initial NIHSS score in these categories. Patients with atrial fibrillation had a significantly severe stroke, with mean NIHSS score of 15.5 versus 8.1 (P=0.002). This was also true for patients whose stroke onset was within 24 hours (9.2 versus 6.2; P=0.046) and for patients with other than small-vessel occlusion stroke (13.4 versus 5.7; P<0.001).
In the multiple regression model, the analysis was performed with the use of all the prespecified predictor variables, even though some of them (such as atrial fibrillation, onset <24 hours, or stroke subtype) could be identified as nonconfounders. The cubic term for the NIHSS score did not provide additional implications as opposed to the quadratic term and thus was ignored.
Table 3 summarizes the results of multiple regression analysis. NIHSS score at admission, the quadratic term of initial NIHSS score, MBI score at admission, small-vessel occlusion stroke, sex, and smoking were the main explanatory factors for lnLOS, whereas other variables such as age (continuous), comorbidity, congestive heart failure, valvular heart disease, atrial fibrillation, history of cardiac disease, onset <24 hours, and lipid levels (continuous) had no significant influence on lnLOS. In particular, NIHSS score at admission (along with its quadratic term) was the strongest predictor. The negative sign of the quadratic term indicated that for patients with mild or moderate stroke, lnLOS increased with increasing stroke severity, while for those who had severe strokes, lnLOS decreased with increasing stroke severity. This model explained approximately 37% of the total variance of LOS.
The reliability of the fitted model was evaluated by obtaining a “shrinkage on cross-validation” of 0.1513 for random splitting. The shrinkage statistic for our nonrandom splitting approach was 0.2104. The estimate of the regression coefficients was calculated by pooling all the data (n=330).
With the use of data from a prospective cohort gathered from a medical center in southern Taiwan, our analysis demonstrated that, among the prespecified predictor variables, initial stroke severity measured by NIHSS was the strongest predictor of LOS for first-ever ischemic stroke patients. Other significant predictors of LOS were initial functional impairment measured by MBI, small-vessel occlusion stroke, sex, and smoking.
In this study age was not significantly associated with LOS. This finding agreed with many previous related studies.7,10⇓ However, the LOS examined in this study was the LOS of acute care hospitalization rather than rehabilitation, which has been studied more often. In addition, only variables for which information is available at the time of admission were considered.
The multiple regression analysis showed that each additional point on the NIHSS would increase the lnLOS by approximately 0.06−0.004N, where N is initial stroke severity measured by NIHSS. More specifically, for patients with mild or moderate neurological impairments (NIHSS score ≤15), a 1-point increase in NIHSS score corresponded to an increase in LOS by approximately 1 day, while for patients with severe neurological impairments (NIHSS score >15), a 1-point increase in NIHSS score corresponded to a decrease in LOS by approximately 1 day.
Our finding of the curvilinear relationship between stroke severity and LOS was compatible with the results of the Copenhagen Stroke Study.7 The decreased LOS in patients with more severe stroke was largely related to the mortality rate. Without consideration of the quadratic term of the NIHSS score, the multiple regression analysis, including only patients with initial NIHSS score ≤15, was repeated (n=274). This analysis showed that in these patients a 1-point increase in NIHSS score corresponded to an increase in LOS by approximately 1 day (regression coefficient=0.02834, SE=0.013, P=0.033). A similar analysis including only patients with initial NIHSS score >15 showed that NIHSS score at admission had no significant influence on LOS; small-vessel occlusion stroke was the only significant predictor variable (regression coefficient=−0.750, SE=0.289, P=0.013). However, caution is warranted when these findings are interpreted because, in this analysis, too many predictor variables were analyzed for the 56 patients.
Male sex corresponded to an increase in LOS by approximately 1.2 days (P=0.004), and smoking decreased LOS by approximately 1.2 days (P=0.043). The 70 patients who smoked were associated with a lower mean NIHSS score (6.2 versus 9.2; P=0.040), as were the male patients (7.7 versus 9.9; P=0.029). It is not clear whether the influence of sex on LOS reflects the impact of culture difference or is due to other factors.
MBI score at admission was, as expected, a significant predictor of LOS. A 1-point decrease of MBI score corresponded to an increase in LOS by approximately 1 day (P=0.042). Stroke subtype was also a strong predictor of LOS, with small-vessel occlusion stroke associated with an approximately 1.5-day shorter LOS than the other subtypes (P<0.001).
This study has some limitations. It is rather difficult for data of an observational study of this type to meet the normality assumption required for using multiple regression models. The LOS variable was thus transformed by a log function to improve the model. Our data were obtained from a hospital-based study. In particular, the fact that the median NIHSS score of the study patients was 6 seemingly indicated that the strokes observed in this study were minor. Consequently, whether the model is transportable to similar patients in different time periods or another location is an important consideration. Even though the shrinkage on cross-validation indicated that the reliability of the model seems fairly acceptable, it is important to have the model externally validated. Nevertheless, the present study sheds some light on the practice patterns of stroke management in Taiwan.
For patients with more severe stroke, the cost of stroke care might not be lower than average because of the extraordinary need of intensive care facilities.22 LOS alone may not correctly measure the costs of acute care hospitalization for first-ever ischemic stroke patients. Future research on this topic is expected to yield potentially fruitful results.
Because initial stroke severity, but not age or comorbidity, was shown to be one of the significant predictors of LOS, we may postulate the hypothesis that initially reducing stroke severity in first-ever ischemic stroke patients with mild or moderately severe stroke might be a wiser way to reduce LOS after acute care hospitalization.
We do not have information specifically related to hours after stroke onset. During the design stage of this study, we thought that the percentage of patients arriving at the hospital within hours after stroke onset might be too low to make any meaningful conclusions. However, with the potential use of intravenous rtPA to treat acute stroke, further studies should address the impact of the therapy on LOS of stroke patients.
Twenty-six percent of patients in this cohort were transferred to the rehabilitation ward within the same hospital. This might reflect the practice pattern of stroke management in this particular area. Additionally, in this area the impact of the caregiver regarding the disposition destination after acute stroke care might be different from that in other countries. Further studies addressing the influence of different practices on LOS are necessary.
In conclusion, given the increasing demand on health services in an aging population, it is crucial to identify the factors that hamper discharge, particularly before clinicians, patients (consumers), and policy makers can evaluate the most effective, efficient, and acceptable methods of managing patients with acute stroke. However, most studies of LOS were conducted by rehabilitation specialists, and the study patients were largely undergoing postacute care. With the potential positive impacts from drugs available to treat ischemic stroke, the present study of LOS after acute care hospitalization is valuable for further analysis of cost-effectiveness.
We are indebted to Dr Harold P. Adams, Jr, for advice during the preparation of the manuscript.
- Received January 2, 2002.
- Revision received May 30, 2002.
- Accepted June 11, 2002.
- ↵Department of Health, Executive Yuan of the Republic of China. Health Statistic Annual, Republic of China, 1971–1998. Taipei, Taiwan, ROC: National Health Administration; 1998.
- ↵Hu HH, Sheng WY, Chu FL, Lan CF, Chiang BN. Incidence of stroke in Taiwan. Stroke. 1992; 23: 1237–1241.
- ↵Mushinski M. Variations in average charges for strokes and TIAs: United States, 1995. Stat Bull Metrop Insur Co. 1997; 78: 9–18.
- ↵Samsa GP, Bian J, Lipscomb J, Matchar DB. Epidemiology of recurrent cerebral infarction: a Medicare claims–based comparison of first and recurrent strokes on 2-year survival and cost. Stroke. 1999; 30: 338–349.
- ↵Smurawska LT, Alexandrov AV, Bladin CF, Norris JW. Cost of acute stroke care in Toronto, Canada. Stroke. 1994; 25: 1628–1631.
- ↵Jørgensen HS, Nakayama H, Raaschou HO, Olsen TS. Acute stroke care and rehabilitation: an analysis of the direct cost and its clinical and social determinants: the Copenhagen Stroke Study. Stroke. 1997; 28: 1138–1141.
- ↵Mamoli A, Censori B, Casto L, Sileo C, Cesana B, Camerlingo M. An analysis of the costs of ischemic stroke in an Italian stroke unit. Neurology. 1999; 53: 112–116.
- ↵Albanese MA, Clarke WR, Adams HP Jr, Woolson RF. Ensuring reliability of outcome measures in multicenter clinical trials of treatments for acute ischemic stroke. Stroke. 1994; 25: 1746–1751.
- ↵Gordon DL, Bendixen BH, Adams HP Jr, Kappelle LJ, Woolson RF, for the TOAST Investigators. Interphysician agreement in the diagnosis of subtypes of acute ischemic stroke: implications for clinical trials. Neurology. 1993; 43: 1021–1027.
- ↵Madden KP, Karanjia PN, Adams HP Jr, Clarke WR. Accuracy of initial stroke subtype diagnosis in the TOAST study: Trial of ORG 10172 in acute stroke treatment. Neurology. 1995; 45: 1975–1979.
- ↵DeGraba TJ, Hallenbeck JM, Pettigrew KD, Dutka AJ, Kelly BJ. Progression in acute stroke: value of the initial NIH Stroke Scale score on patient stratification in future trials. Stroke. 1999; 30: 1208–1212.
- ↵Adams HP Jr, Davis PH, Leira EC, Chang KC, Bendixen BH, Clarke WR, Woolson RF, Hansen MD. Baseline NIH Stroke Scale score strongly predicts outcome after stroke: a report of the Trial of ORG 10172 in Acute Stroke Treatment (TOAST). Neurology. 1999; 53: 126–131.
- ↵Johnston KC, Connors AF Jr, Wagner DP, Knaus WA, Wang X, Haley EC Jr. A predictive risk model for outcomes of ischemic stroke. Stroke. 2000; 31: 448–455.
- ↵Caro JJ, Huybrechts KF, Kelley HE. Predicting treatment costs after acute ischemic stroke on the basis of patient characteristics at presentation and early dysfunction. Stroke. 2001; 32: 100–106.
- ↵Kleinbaum DG, Kupper LL, Muller KE, Nizam A. Applied Regression Analysis and Other Multivariable Methods. Pacific Grove, Calif: Duxbury Press; 1998.
- ↵Reed SD, Blough DK, Meyer K, Jarvik JG. Inpatient costs, length of stay, and mortality for cerebrovascular events in community hospitals. Neurology. 2001; 57: 305–314.