Original ArticlePrognosticating Functional Outcome After Intracerebral Hemorrhage: The ICHOP Score
Introduction
Intracerebral hemorrhage (ICH) accounts for 10%–15% of all strokes in the United States.1 The loss of life associated with ICH is high, with approximately 35% mortality at 30 days and mortality at 1 year reaching as high as 59%.2 ICH morbidity is also staggering: 1 study put the aggregate lifetime cost of ICH at $6 billion.3 U.S. hospital admissions for ICH have increased by more than 15% during the 2000s, possibly because of the increasing age of the population.4
Over the past decade, many scoring systems have been created with the intention of prognosticating outcomes for patients with ICH, beginning with the original ICH Score in 2001 and followed by other scales such as the Essen ICH Score, ICH-FOS, FUNC score, and AVICH score.5, 6, 7, 8, 9 Although many of these systems show acceptable performance in prognosticating outcome, each has its own associated problems: limited external validity, not being designed to predict functional outcome, or limited testing in American populations.10, 11, 12, 13 Because of its simplicity and validity throughout multiple cohorts, the ICH Score remains the most commonly used grading scale for ICH mortality and functional outcome.6, 14
Most of these grading systems have similar makeup, including factors such as Glasgow Coma Scale (GCS) or National Institutes of Health Stroke Scale (NIHSS), hematoma volume and location, and age. However, recently, it has been suggested that other physiologic factors may play an important role in predicting prognosis after ICH, particularly those included in the Acute Physiology and Chronic Health Evaluation II (APACHE II) score.15 In addition, most scales use either the GCS or the NIHSS to assess neurologic function, although each is subject to its own problems; for example, the NIHSS score may be identical in patients with different stroke volumes in different hemispheres.16, 17 To our knowledge, no published scores to date account for the previous functional status of the patient before ICH, which has not yet been explored as a predictor of long-term functional outcome. It therefore follows that inclusion of robust physiologic admission data, multiple methods of neurologic assessment, and patient baseline may produce a more accurate and comprehensive prediction algorithm for functional outcome when combined with measures already used by previous scoring systems.
Section snippets
Study Population and Data Collection
Data in the Columbia University Intracerebral Hemorrhage Outcomes Project (ICHOP) database (Columbia University IRB-AAAD4775) was collected prospectively by trained clinicians and researchers from patients presenting to Columbia University Medical Center (CUMC) with a diagnosis of ICH from March 2009 to May 2016 (n = 575). Data collected included patient identification information, admission scores, admission laboratory test results, clotting information, cause, procedures performed, treatments
Results
Of 575 patients who presented to CUMC with ICH between March 2009 and May 2016, 365 had complete information at 3 months and, of these, 321 also had complete information at 12 months. Table 2 shows cohort data for patients with complete information at 3 months. This derivation cohort was similar to the database in terms of important factors such as age, GCS, NIHSS, and hematoma volume.
From our available data, the factors available on admission with the highest Random Forest importance for
Discussion
We present a novel scoring system for predicting functional outcome after ICH at 3 and 12 months follow-up. Both models show robust discrimination as determined by AUC (ICHOP3, 0.89; ICHOP12, 0.87) and are shown to be significantly better than the original ICH Score in predicting 3-month and 12-month functional outcome in our derivation cohort. Several ICH scoring systems have been developed since the induction of the original ICH Score in 2001.5, 6, 7 The ICHOP scores take into account factors
Conclusions
The ideal scoring system for any disease is easy to calculate, highly discriminatory, and universally applicable. When creating the ICHOP score, we chose to sacrifice ease of calculation for increased discrimination, and preliminary external findings hint that it may be applicable in different settings. Although some physicians treating patients with ICH may be hesitant to calculate ICHOP scores, both the high morbidity of the illness and the overwhelming burden that it places on the health
References (22)
- et al.
Intracerebral haemorrhage
Lancet
(2009) - et al.
The AVICH score: a novel grading system to predict clinical outcome in arteriovenous malformation-related intracerebral hemorrhage
World Neurosurg
(2016) - et al.
Functional outcome prediction following intracerebral hemorrhage
J Clin Neurosci
(2012) - et al.
Hematoma volume as the major determinant of outcomes after intracerebral hemorrhage
J Neurol Sci
(2014) - et al.
Long-term mortality after intracerebral hemorrhage
Neurology
(2006) - et al.
Lifetime cost of stroke in the United States
Stroke
(1996) - et al.
Heart disease and stroke statistics–2012 update: a report from the American Heart Association
Circulation
(2012) - et al.
Development and validation of the Essen Intracerebral Haemorrhage Score
J Neurol Neurosurg Psychiatr
(2006) - et al.
The ICH score: a simple, reliable grading scale for intracerebral hemorrhage
Stroke
(2001) - et al.
A novel risk score to predict 1-year functional outcome after intracerebral hemorrhage and comparison with existing scores
Crit Care
(2013)
Prediction of functional outcome in patients with primary intracerebral hemorrhage: the FUNC score
Stroke
Cited by (46)
An artificial intelligence-based prognostic prediction model for hemorrhagic stroke
2023, European Journal of RadiologyDevelopment and validation of machine learning prediction model for post-rehabilitation functional outcome after intracerebral hemorrhage
2022, Interdisciplinary Neurosurgery: Advanced Techniques and Case ManagementCitation Excerpt :In recent years, machine learning has been attempted, which is expected to improve accuracy by incorporating non-linear computations [13,14,15,16,17,18,19,20]. A few studies have focused on the clinical significance of patient selection and explanatory variable settings, which are expected to improve accuracy but have not been fully investigated [12,20]. Detailed outcome prediction has not been focused on so far.
Prediction of Motor Function in Stroke Patients Using Machine Learning Algorithm: Development of Practical Models
2021, Journal of Stroke and Cerebrovascular DiseasesCitation Excerpt :Therefore, we believe that ML algorithms can overcome the limitations of traditional prognostic scoring systems for predicting chronic motor outcomes in individual stroke patients. Although a few studies used ML for predicting motor outcome after stroke,7-9 for the use in clinical practice, further studies are required on this topic. In the present study, by using 14 input variables, we developed a deep neural network (DNN) model and applied 2 well-known ML algorithms, logistic regression and random forest, in predicting motor outcome at 6 months after stroke.
Hemorrhagic stroke
2021, Handbook of Clinical NeurologyMachine Learning for Brain Stroke: A Review
2020, Journal of Stroke and Cerebrovascular DiseasesCitation Excerpt :The performance of the SVM was compared with SEDAN and HAT scores (prognostication tools). Similarly, RF was employed [32] to identify factors or feature selection from 575 intracerebral hemorrhage (ICH) outcomes and developed LR model for prognosticating functional status (mRS score, 0 to 3 is good and mRS score, 4 to 6 is poor) based on these factors. LR model reported 0.89 AUC for 3 months and 0.87 for 12 months.
Conflict of interest statement: This work was supported by the National Institutes of Health (NIA T35AG044303).