Can We Predict Poor Outcome in Stroke Patients Without Imaging Data? A Decision Tree Analysis of Stroke Patients

Abstract

Objective: The aging world population and increased cardiovascular risk factors contribute to stroke and stroke related morbidity. In this study, we aimed to analyze predictors of increased morbidity of ischemic stroke patients in a single stroke unit.

Materials and Methods: Stroke patients recorded in the Istanbul University Stroke Registry between 2014 and 2020 were included and decision tree analyses [chi-squared automatic interaction detection (CHAID) method] were conducted. Gender, diabetes, hypertension, previous stroke, ischemic heart disease, hyperlipidemia, diagnosis of pneumonia during hospitalization in the stroke unit, and atrial fibrillation were determined as possible indicators for poor clinical outcomes.

Results: We included 881 patients with ischemic stroke in the study according to the inclusion and exclusion criteria. The mean age of patients was 66.5±14.4 years and 59% of the patients were male. CHAID analysis revealed that the most important factor for predicting modified Rankin Scale (mRS) score >3 is pneumonia. In patients with mRS score >3 and without pneumonia; hypertension and hyperlipidemia were found to be risk factors for poor functional outcome.

Conclusion: Preventative measures in stroke patients should not be limited to secondary prophylaxis of stroke. Avoiding infections in the acute phase plays an essential role in achieving favorable clinical outcomes.

Introduction

Ischemic stroke (IS) is still one of the leading causes of mortality and morbidity worldwide (1). The aging world population and increased cardiovascular risk factors contribute to the stroke burden. Recent acute treatment options such as thrombolysis and thrombectomy have started to change the course of the disease (2). Nevertheless, the number of patients who are not suitable candidates for acute treatments and those who cannot receive acute treatment due to logistic issues cannot be underestimated even in the 21^st century. The optimal secondary prophylaxis and diligent hospital care were shown to reduce mortality and morbidity, especially in stroke units (3). Not only appropriate antithrombotic treatment but optimal treatment approaches for vascular risk factors and preventative measures for infections and other systemic complications improve quality of life after stroke.

It is important to determine risk factors associated with stroke patients with poor functional outcomes to identify manageable risk factors. To address this question, we aimed to analyze predictors of increased morbidity of IS patients in a single stroke unit.

Material and Methods

We included IS patients admitted to the Istanbul University Stroke Unit between July 2014 and September 2020. Diagnosis of stroke was made with computer tomography or magnetic resonance imaging modalities in patients with acute onset neurological symptoms by a stroke neurologist. Patients with cerebral venous thrombosis, intracerebral hemorrhage (subdural, subarachnoid, or intra-parenchymal hemorrhage) and transient ischemic attacks were not included in the study. In addition, patients younger than 18 years old were excluded. The data of all included patients had been recorded in the Istanbul Medical School Stroke Registry (4).

Patients were grouped according to modified Rankin Scale (mRS) score at discharge. Patients with mRS >3 were determined as poor functional outcome group. Even though the conventional tendency is not to include mRS score of 3 as a good outcome, this rule may not be appropriate for evaluation at discharge, but it is more suitable for evaluating third-month functional status. The National Institutes for Health stroke scale (NIHSS) scores were recorded at admission. Gender, diabetes, hypertension, history of previous stroke, ischemic heart disease, hyperlipidemia, diagnosis of pneumonia during hospitalization in the stroke unit and atrial fibrillation were determined as possible indicators for poor functional outcome. To generalize results and implement preventative practices for the daily care of stroke, patients we did not focus on ischemic-stroke subtypes. Istanbul University Faculty of Medicine Local Ethics Committee approved the study protocol (2020/1543).

Statistical Analysis

We used a decision tree method, chi-squared Automatic Interaction Detection (CHAID) to analyze our data set via IBM SPSS (Statistical Package for the Social Sciences) version 25. CHAID is principally a multivariate analysis technique and is used as a prediction model for different fields including neurology (5). CHAID analysis is a very useful tool to analyze big data sets. CHAID is not hypothesis driven and is used to explore associations between different variables. Our stroke data set after applying the inclusion and exclusion criteria consisted of a large amount of patients’ records which were suitable for CHAID analysis. Minimum number of cases was set 100 for the parent node and 50 for the child node. Maximum number of levels was 3 for CHAID. The significance level for splitting nodes and merging categories was determined as 0.05. Significance values were adjusted using the Bonferroni method. CHAID method is widely used to analyze stroke databases. Stroke subtype cannot be identified sufficiently with this model but prediction of prognosis and risk factors associated with stroke are satisfactory (6,7,8).

Results

We included 881 patients with IS in the study according to the inclusion and exclusion criteria mentioned above. Demographical and clinical characteristics of patients are shown in Table 1.

CHAID analysis revealed that the most important factor on mRS score >3 was pneumonia (Figure 1). Overall percentage of the model was found at 74% in the classification table for CHAID growing method.

Figure 1. Decision tree results

Since pneumonia was found as a significant predictor for poor functional outcome, we analyzed possible risk factors associated with pneumonia. As expected, there was a statistically significant relationship between dysphagia and severe walking disability and pneumonia (p<0.00 and p<0.00). The mean age was 73.4 (±11.9) years, and the mean NIHSS score was 12.7 (±6.9) in patients who had pneumonia during hospitalization. The mean age and mean NISSS score were higher compared to patients without pneumonia. These results were also statistically significant (p<0.00 and p<0.00, respectively).

Smoking was also found as a strong risk factor in those patients who had pneumonia. Finally, smokers with pneumonia accumulated at the atrial fibrillation arm. On the other hand, stroke patients who did not have pneumonia had different risk factors for poor functional outcome. Hypertension and hyperlipidemia composed second and third nodes as a predictor, respectively. Hyperlipidemia should be interpreted with caution depending on the presence of hypertension.

Finally, we analyzed relationship between stroke etiologies and mRS scores (p=0.2). The results are given in Table 2.

Discussion

In this study, we analyzed possible risk factors for poor functional outcome in stroke. We found that pneumonia and hypertension were associated with poor outcome. Stroke patients with pneumonia and smoking history also had increased risk for mRS score >3 at discharge.

Patients with mRS score >3 who did not have pneumonia accumulated in hypertension and hyperlipidemia nodes. Surprisingly, diabetes and gender did not have significant results regarding mRS scores in CHAIDS analysis.

Stroke cases may be complicated with pneumonia, especially aspiration pneumonia (9). Derbisz et al. (10,11) showed that prestroke morbidity and elevated C-reactive protein level are associated with infection in stroke patients who received thrombolysis. Association of infection with functional outcome was also shown in different studies (10,11). We found that patients with pneumonia were older and had greater NIHSS scores. Similar to our results, stroke patients with pneumonia were older, had higher stroke severity, and were more likely to have atrial fibrillation and history of ischemic heart disease compared to stroke patients without infection in a different study (9).

We did not include stroke severity and NIHSS score to the decision tree analysis intentionally. Primarily, the main purpose of this decision was to understand which elements affected the clinical outcome other than affected brain regions and stroke severity. Another study based on a stroke database from Netherlands supports our findings regarding the association between infection and poor outcome. They also suggest that stroke associated infections are independent risk factors for worse functional outcome in stroke patients (12). Patients who received appropriate treatment (antibacterial, antifungal drugs, etc.) for stroke associated infections were found to have better outcome (13). Hannawi et al. (14) suggested that aspiration and stroke related immunosuppression might explain the pathophysiology of pneumonia in stroke patients.

As our results suggest, it is crucial to avoid in-hospital pneumonia to improve outcome in stroke patients. Many stroke guidelines such as European stroke guidelines do not support the use of prophylactic antibiotic treatments (15). Preventing hospital infections, especially aspiration pneumonia, should be prioritized for every single patient with stroke. Swallow-tests before feeding, treatment of fever and urinary care are well-known preventative measures. Nevertheless, these simple measures can be ignored very easily and may lead to undesirable consequences.

Study Limitations

There are some limitations of this study. Firstly, even though our stroke database was registered by a neurologist and reviewed by experienced stroke neurologists, it is based on a single center experience. Secondly, this study is a retrospective observational study and therefore selection bias could not be entirely excluded.

Conclusion

In conclusion, to achieve better outcomes in patients with stroke, preventative measures should not be limited to only secondary prophylaxis. Avoiding infections plays an essential role. New data-driven approaches might be beneficial in stroke registry data research.

Istanbul University Faculty of Medicine Local Ethics Committee approved the study protocol (2020/1543).

Obtained from patients and/or their relatives.

Externally peer-reviewed.

Concept: M.S., M.G.G., N.Y., Design: M.S., M.G.G., Data Collection or Processing: M.S., N.Y., Analysis or Interpretation: M.S., M.G.G., Literature Search: M.S., Writing: M.S., N.Y.

No conflict of interest was declared by the authors.

The authors declared that this study received no financial support.

References

1. Furie K. Epidemiology and primary prevention of stroke. Continuum (Minneap Minn) 2020;26:260-267.
2. McCarthy DJ, Diaz A, Sheinberg DL, et al. Long-term outcomes of mechanical thrombectomy for stroke: a meta-analysis. ScientificWorldJournal 2019;2019:7403104.
3. Stroke Unit Trialists' Collaboration. Organised inpatient (stroke unit) care for stroke. Cochrane Database Syst Rev 2013;2013:CD000197.
4. Yesilot NF, Koyuncu BA, Coban O, Tuncay R, Bahar SZ. Gender differences in acute stroke: Istanbul medical school stroke registry. Neurol India 2011;59:174-179.
5. Doğan N, Özdamar K. Chaid analysis and an application related with family planning. T Klin J Med Sci 2003;23:392-397.
6. Naftulin IS, Rebrova OY. WizWhy algorithms for stroke type diagnosis. Artificial Intelligence and Soft Computing 2010:651-656.
7. Imura T, Iwamoto Y, Inagawa T, et al. Decision tree algorithm identifies stroke patients likely discharge home after rehabilitation using functional and environmental predictors. J Stroke Cerebrovasc Dis 2021;30:105636.
8. Shao Z, Xiang Y, Zhu Y, Fan A, Zhang P. Influences of daily life habits on risk factors of stroke based on decision tree and correlation matrix. Comput Math Methods Med 2020;2020:3217356.
9. Suda S, Aoki J, Shimoyama T, et al. Stroke-associated infection independently predicts 3-month poor functional outcome and mortality. J Neurol 2018;265:370-375.
10. Kwan J, Hand P. Infection after acute stroke is associated with poor shortterm outcome. Acta Neurol Scand 2007;115:331-338.
11. Derbisz J, Nowak K, Wnuk M, et al. Prognostic significance of strokeassociated infection and other readily available parameters in acute ischemic stroke treated by intravenous thrombolysis. J Stroke Cerebrovasc Dis 2021;30:105525.
12. Vermeij FH, Scholte op Reimer WJ, de Man P, et al. Stroke-associated infection is an independent risk factor for poor outcome after acute ischemic stroke: data from the Netherlands Stroke Survey. Cerebrovasc Dis 2009;27:465-471.
13. Grabska K, Gromadzka G, Cz onkowska A. Infections and ischemic stroke outcome. Neurol Res Int 2011;2011:691348.
14. Hannawi Y, Hannawi B, Rao CP, Suarez JI, Bershad EM. Stroke-associated pneumonia: major advances and obstacles. Cerebrovasc Dis 2013;35:430- 443.
15. The European Stroke Organisation (ESO) Executive Committee and the ESO Writing Committee. Guidelines for management of ischaemic stroke and transient ischaemic attack 2008. Cerebrovasc Dis 2008;25:457-507.