Assessment Model to Identify Patients With Stroke With a High Possibility of Discharge to Home
A Retrospective Cohort Study
Background and Purpose—Discharge planning for inpatients with acute stroke can enhance reasonable use of healthcare resources, as well as improve clinical outcomes and decrease financial burden of patients. Especially, prediction for discharge destination is crucial for discharge planning. This study aimed to develop an assessment model to identify patients with a high possibility of discharge to home after an acute stroke.
Methods—We reviewed the electronic medical records of 3200 patients with acute stroke who were admitted to a stroke center in Japan between January 1, 2011, and December 31, 2015. The outcome variable was the discharge destination of postacute stroke patients. The predictive variables were identified through logistic regression analysis. Data were divided into 2 data sets: the learning data set (n=2240) for developing the instrument and the test data set (n=960) for evaluating the predictive capability of the model.
Results—In all, 1548 (48%) patients were discharged to their homes. Multiple logistic regression analysis identified 5 predictive variables for discharge to home: living situation, type of stroke, functional independence measure motor score on admission, functional independence measure cognitive score on admission, and paresis. The assessment model showed a sensitivity of 85.0% and a specificity of 75.3% with an area under the curve equal to 0.88 (95% confidence interval, 0.86–0.89) when the cutoff point was 10. On evaluating the predictive capabilities, the model showed a sensitivity of 88.0% and a specificity of 68.7% with an area under the curve equal to 0.87 (95% confidence interval, 0.85–0.89).
Conclusions—We have developed an assessment model for identifying patients with a high possibility of being discharged to their homes after an acute stroke. This model would be useful for health professionals to adequately plan patients’ discharge soon after their admission.
Stroke is one of the major causes of long-term disability and a leading cause of mortality worldwide.1,2 Discharge planning for inpatients with acute stroke can enhance use of healthcare resources, as well as improve clinical outcomes and decrease financial burden of patients.3 Accurate estimation of expected discharge destinations is a vital issue in discharge planning. Several previous studies discovered factors influencing discharge destinations4 and tried to develop screening instruments for stroke discharge planning5; however, their validity and clinical efficacy are still unclear.
The discharge destination of patients with acute stroke is influenced by various factors, including demographic background, socioeconomic status, and clinical characteristics, such as severity.5–7 Marital status5,7–9 and living situation (with or without someone else)10 are associated with a better chance of being discharged to home. On the contrary, patients living in care facilities before admission are significantly at risk of being institutionalized after discharge.5 In addition, clinical characteristics, such as stroke type, functional independence measure (FIM) score, and comorbidities, have been identified as predictive factors for discharge to home.5,8–11 In particular, better functional status5,9,10,12 and cognitive status5,10,13 were the most significant factors for discharge to home. These studies reported significant associations between individual factors and discharge to home. However, it is essential to take several factors into account simultaneously to assess discharge destination and conduct adequate discharge planning.
A study from the United States has already developed a discharge-to-home planning index for patients with acute stroke,5 but most of the subjects in this study were military veterans. The discharge planning index is, therefore, not appropriate for the general public. Moreover, the factors of functional independence were measured at the final rehabilitation assessment for the index. The validity of the index for poststroke patients in the early period of hospitalization is, thus, unclear.
This study, therefore, aimed to develop a screening instrument with a multifactorial scoring formula to identify patients with a high possibility of discharge to home after an acute stroke.
Study Design and Setting
This study was a retrospective cohort study based on data from the medical records of inpatients who were treated at a stroke center in Japan between January 1, 2011, and December 31, 2015. The stroke center is located in a suburban area in west central Japan, and treats >1000 patients with stroke a year.
We included patients who received treatment for acute stroke (cerebral hemorrhage, cerebral infarction, or subarachnoid hemorrhage) at the stroke center. However, patients who were not living at home before hospital admission were excluded from the analysis because all such patients were not discharged to their homes. Diagnosis and subtype classification of patients’ stroke were made and recorded by neurologists of the research hospital based on the classification of cerebrovascular disease III proposed by the National Institute of Neurological Disorders and Stroke.14 The requirement for informed consent was waived because the data were anonymized. This research was conducted according to the principles of the declaration of Helsinki and ethical guidelines for medical and health research involving human subjects by the Japanese Ministry of Health, Labor, and Welfare.
Data Source and Variables
All data on outcomes and expected predictive factors were extracted from the electronic medical record database. Then, investigators checked the rigor of the data set. If researchers found missing values, outliers, or other aberrant values, one of researchers manually reviewed the individual medical records, including paper-based documents, such as the medical interview sheet of the patients, and then amended or added exact value into the data. As a result, we successfully established a complete data set.
The primary outcome of this study was the discharge destination after hospitalization for acute stroke, namely, home, death, or other locations, such as a rehabilitation facility, hospital, or care facility.
Various sociodemographic factors and clinical characteristics were analyzed as potential predictive factors for discharge to home. The sociodemographic factors included age, sex, living situation (living with someone or living alone), marital status, and insurance use (public assistance, disability insurance, care insurance, or private insurance). The clinical characteristics were type of stroke, paresis, FIM motor score on admission (low, 13–38; intermediate, 39–50; high, 51–91),10 FIM cognitive score on admission (low, 5–20; intermediate, 21–29; high, 30–35),10 modified Rankin Scale at discharge, length of stay, treatment type (surgery, mechanical ventilation, transfusion, and intensive care unit admission), and comorbidity (hypertension, diabetes mellitus, dyslipidemia, electrolyte disorders, liver disease, heart disease, kidney disease, cancer, mental disorder, and anemia).
Development of the discharge-to-home assessment model consisted of three steps: (1) variable selection by logistic regression analysis, (2) scoring method development, and (3) cutoff point setting. To avoid overfitting in a prediction model based on real data, the data were divided into learning data (70%) and test data (30%). Learning data were used for developing the screening instrument, whereas the test data were used for examining the validity of the assessment model.
We performed univariate logistic regression analysis using learning data to investigate the relationship between clinical characteristics/sociodemographic factors and discharge to home. Variables with positive associations with discharge to home (P<0.2) were entered into the initial multivariate logistic regression model. If the Spearman rank correlation for these variables was >0.5, we chose one of the variables for the next step based on clinical relevance. Then we performed multivariate logistic regression using the backward variable selection method. First, the full model consisting of all these variables was established. Then a variable which showed the highest and >0.05 of P value was removed from the model. Then C statistics of the pre- and postselection model were compared for examining the impact of the removal. This variable selection was repeated until all the variables showed statistically significant coefficients or a C statistic of the model showed insufficient goodness of fit. The goodness of fit of the final model was also assessed by the Hosmer–Lemeshow χ2 test.
Scoring Method Development
We developed a weighted scoring method based on the method described by Sullivan et al.15 The resulting score was the screening instrument. Briefly, each coefficient was divided by the smallest coefficient of positive value, after which it was doubled and rounded to the nearest whole number. A patient’s discharge-to-home possibility score was defined as the sum of the individual weighted scores for all variables.
We fitted the model to test data and calculated the discharge-to-home possibility score for each patient.
Cutoff Point Setting
Discrimination of the assessment model was assessed by drawing the receiver-operating characteristic curve and calculating the area under the curve. The cutoff point for identifying patients with a high possibility of discharge to home was determined based on clinical relevance and an acceptable false-positive ratio.
In all, 3340 patients were admitted to the hospital with a primary diagnosis of acute stroke between January 1, 2011, and December 31, 2015. We excluded 140 patients who were not living at home before hospital admission, and, thus, finally included 3200 patients in the study. Subjects were randomly divided into 2 groups with 2240 patients (70%) in the learning data group and 960 patients (30%) in the test data group.
Patients’ clinical and sociodemographic characteristics were similar in both groups (Table 1). Overall, about half of the patients (48%) were discharged to their homes. Of the remaining patients, 32% were discharged to rehabilitation facilities and 18% were discharged to other hospitals. The mean patient age was 72.7 years with an SD of 12.9, and 58% of patients were men. The mean length of stay in hospital was 23.6±12.8 days. In all, 73% of patients had a cerebral infarction, 20% of patients had a cerebral hemorrhage, and 6.3% of patients had experienced a subarachnoid hemorrhage.
Factors Associated With Discharge to Home After Acute Stroke: Univariate Logistic Regression
Univariate logistic regression analysis showed that most of the sociodemographic and clinical factors were associated with discharge to home (Table 2). In particular, patients with high FIM scores had a significantly higher possibility of discharge to home than those with low FIM scores. For instance, compared with patients with low FIM motor scores on admission, the odds ratios (95% confidence intervals) of those with intermediate and high scores for discharge to home were 6.05 (4.63–7.93) and 29.52 (22.69–38.83), respectively (Table 2). Similarly, compared with patients with low FIM cognitive scores on admission, the odds ratios (95% confidence intervals) of those with intermediate and high scores for discharge to home were 5.30 (4.24–6.64) and 17.83 (13.89–23.05), respectively (Table 2).
Development of Assessment Model: Multivariable Regression Model and Scoring
Variables with P<0.2 in univariate analysis were selected for the initial multivariable logistic regression model. After backward variable selection, 5 variables were kept in the final logistic regression model, namely, living situation, stroke type, FIM motor score at admission, FIM cognitive score at admission, and paresis (Table 3). The Hosmer–Lemeshow statistic was 10.1 (degree of freedom, 8; P=0.26) for the model. C statistics of the initial full model and final model were 0.885 and 0.882, respectively. Therefore, a significant lack in goodness of fit were not observed.
Each coefficient was divided by the smallest coefficient of positive value, that is, 0.55 for cerebral infarction, doubled and rounded to the nearest whole number. The total score ranged from 0 to 25 in the model, with a higher score indicating a higher possibility of discharge to home. The mean score was 11.5±6.97 for the learning data, and 12.8±5.95 for the test data.
Figure 1 shows the receiver-operating characteristic curve of the total score for the learning data (area under the curve, 0.88; 95% confidence interval, 0.86–0.89). When the cutoff point was 10, the model showed a sensitivity of 85.0% and specificity of 75.3%. For the test data, the receiver-operating characteristic curve of the total score is shown in Figure 2 (area under the curve, 0.87; 95% confidence interval, 0.85–0.89). The sensitivity and specificity were 88.0% and 68.7%, respectively. Figure 3 shows the final assessment model.
In this study, we developed a screening instrument for predicting the possibility of discharge to home after an acute stroke. We developed a multidimensional model based on factors associated with discharge to home, such as degree of disability in daily life, disease characteristics like stroke type or paresis, physical disability, and social factors like living status. The model showed a high sensitivity with both learning and test data.
Our study showed that factors related to physical and cognitive function were especially useful predictors for discharge to home, which was consistent with previous studies.5,8–11 FIM motor and cognitive scores at admission showed remarkably higher score weights, with the former showing the highest predictive ability in the model. Currently, the measures of independence in activity of daily life, such as the FIM, are not routinely collected or recorded at admission. However, because of the importance of the FIM to predict a discharge to home shown in this research, routine measurement of independence in activity of daily life at admission should be recommended. Interestingly, the weighted score for the FIM motor score at admission was ≈2× higher than that for the FIM cognitive score. All variables related to physical function (eg, FIM motor score at admission and paresis) were retained in the model, whereas those related to cognition, such as mental disorders, were not. These results suggest that strategic physical independency is more important than cognitive independence for discharge to home directly from acute-care hospital.
Our model also showed that patients without paresis are more likely to be discharged to their homes. However, we found no evidence that patients with lateral paresis are more likely to be discharged to their homes than those with bilateral paresis. Bilateral paresis usually causes more severe physical dysfunction in patients with stroke than unilateral paresis. Our result shows that paresis, regardless of its severity, independently influences the possibility of discharge to home.
The type of stroke was also a significant factor. Previous studies have already stated that cerebral hemorrhages are more severe than ischemic strokes.16,17 Analogously, patients with higher ischemic stroke have a higher possibility of discharge to home. Surprisingly, our model shows that patients with subarachnoid hemorrhages have a higher possibility for discharge to home than those with other types of stroke. The World Federation of Neurosurgical Societies Cerebrovascular Diseases and Therapy Committee recently reported a bimodal distribution of severity among patients with subarachnoid hemorrhage. The report showed that 47.4% of patients with subarachnoid hemorrhage have a poor outcome like death (World Federation of Neurosurgical Societies scale, grade V, 28.7%) or persistent vegetative state (grade IV, 18.6%). In contrast, about half of the patients have a favorable outcome, such as low disability (grade I, 29.1%) or moderate disability (grade II, 21.0%).18 Similarly, half of the patients with subarachnoid hemorrhages in this study were discharged to home. We suspect the bimodality in the severity of subarachnoid hemorrhage caused the model to emphasize the occurrence of discharge to home. However, the number of patients with subarachnoid hemorrhages was relatively small compared with those with cerebral infarction or hemorrhage. Therefore, the predictive ability of the model for patients with subarachnoid hemorrhage should be checked in a larger population.
Our model also indicates that support from a partner or family member who lives with the patient is important. Such support can help patients with stroke perform activities of daily living more efficiently. Our study confirms the results of earlier studies. Some previous studies stated that marital status had a significant influence on discharge destinations.5,7–9 However, our study clarifies that the living situation is a better predictor than marital status. Consequently, medical professionals should consider support available from partners or family members during discharge planning.
The assessment model was based on electronic health records from a single large-scale stroke center. Hence, the data were homogeneous, which is a strength of this study. Additionally, all the data were collected within a week of hospital admission. This model is, therefore, useful for clinicians in making more valid and immediate informed decision about discharge planning for patients with acute stroke. Receiver-operating characteristic curves and area under the curves indicated that the model was accurate, with high sensitivity (85%–88%) and specificity (69%–75%), suggesting that the instrument will be effective in identifying poststroke patients with moderate-to-high possibility of discharge to home.
Several limitations of this study should be acknowledged. First, participants were limited to patients from a stroke center located in a suburban area in Japan. Although it is not clear whether this would cause a sampling bias, it would be better to check the model’s performance before applying it to other populations. Second, the clinical treatment for acute stroke is different in different countries. For example, the average length of stay in acute-care hospitals in Japan was 17.2 days compared with 5.4 days in the United States in 2013.19 Therefore, further research investigating whether this model is applicable to other settings is needed, especially in a nation with a shorter length of stay than Japan. Third, the cutoff point for identifying discharge-to-home patients in individual medical institutions is expected to be highly influenced by available medical resources. For example, a higher cutoff point is recommended for institutions with fewer resources to suppress overestimating the discharge-to-home possibility. Fourth, we cannot evaluate patient outcomes, such as the quality of life, after discharge using this model. Therefore, there were other important variables which we didn’t investigate in this study. For example, socioeconomic status would influence patients and their family’s choice for discharge destination. Additionally, it is unclear that how many patients were inappropriately discharged. Further evaluation of inappropriate discharge events, such as early readmission or transfer to care facility after home discharge, is needed.
In summary, we have developed an assessment model for identifying patients with a higher possibility of discharge to home after an acute stroke. In turn, this screening instrument may provide useful information for clinicians to plan the discharge of patients with acute stroke.
Sources of Funding
This study was conducted with financial support from Brain Attack Center Ota Memorial Hospital to Dr Nishigaki.
- Received May 17, 2017.
- Revision received July 20, 2017.
- Accepted August 4, 2017.
- © 2017 American Heart Association, Inc.
- Lozano R,
- Naghavi M,
- Foreman K,
- Lim S,
- Shibuya K,
- Aboyans V,
- et al
- Murray CJ,
- Vos T,
- Lozano R,
- Naghavi M,
- Flaxman AD,
- Michaud C,
- et al
- Summers D,
- Leonard A,
- Wentworth D,
- Saver JL,
- Simpson J,
- Spilker JA,
- et al
- Mees M,
- Klein J,
- Yperzeele L,
- Vanacker P,
- Cras P
- Stineman MG,
- Kwong PL,
- Bates BE,
- Kurichi JE,
- Ripley DC,
- Xie D
- Tanwir S,
- Montgomery K,
- Chari V,
- Nesathurai S
- Badriah F,
- Abe T,
- Miyamoto H,
- Moriya M,
- Babazono A,
- Hagihara A
- 14.↵Special report from the National Institute of Neurological Disorders and Stroke. Classification of cerebrovascular diseases III. Stroke. 1990;21:637–676.
- Sheedy R,
- Bernhardt J,
- Levi CR,
- Longworth M,
- Churilov L,
- Kilkenny MF,
- et al
- Sano H,
- Satoh A,
- Murayama Y,
- Kato Y,
- Origasa H,
- Inamasu J,
- et al
- 19.↵Organisation for Economic Co-operation and Development. Length of hospital stay. https://data.oecd.org/healthcare/length-of-hospital-stay.htm. Accessed December 6, 2016.