Prognostic scores for IPH

Introduction
Numerous scores have been created to predict outcome in patients with intraparenchymal hemorrhage. Some predict mortality while others predict functional outcome, and at very different time points and different metrics. Some are more well-validated than others. This page summarizes the data that we have on all of these scales, and provides their parameters for easy calculation. The tables below link to more details on each of the scores.

Hemodialysis patients
ICH Outcome Score (ICHOS)

Surgical intervention
Cho's Modified ICH score (basal ganglia hemorrhage only)

Minimally Invasive Surgery (MIS) score (minimally-invasive techniques)

Surgical SwICH score (general surgical evacuation)

Vascular lesions
AVICH score (AVM-related IPH)

AVM-oICH score (AVM-related IPH)

Vascular Structure-Related ICH Score (VSARICH) (vascular-lesion related IPH, mostly AVMs)

Portenoy score (1987)
Derived from 112 patients in the USA. Multivariate regression model using GCS, ICH size, and intraventricular spread used to predict functional outcome. Complicated to calculate, and therefore not very useful.

Tuhrim equation (1988)
Derived from 94 cases of IPH in the USA in 1980-1981. Validated in a cohort of 1013 patients in Helsinki.

Cincinnati score (1993)
Derived from 188 cases of IPH in the USA in 1988. Validated in a cohort of 241 patients, but shown to not be as strong a score as the ICH score. Also validated in a cohort of 1013 patients in Helsinki.

Lisk score (1994)
Derived from 75 patients in the USA. Multivariate regression model using hemorrhage size, IVH, age, and gender to predict functional outcome. Complicated to calculate and therefore not very useful.

Masé score (1995)
Derived from 138 patients in Italy and uses IVH, volume of hemorrhage, and GCS to predict survival. Graphs are used to calculate out survival curves for each permutation. Validated in a cohort of 1013 patients in Helsinki

ICH score (2002)
This has become a standard score utilized for IPH prognosis, although it is derived from a series of only 152 patients at a single center from 1997-1998.

Validation studies include:
 * 142 patients with nontraumatic ICH in Hong Kong in 1999 where a cutoff of 3 or higher was found to be 79% sensitive and 90% specific for mortality, with a Youden's index of 0.69
 * 153 patients from Argentina where a score of 2 or higher had a Youden's index of 0.59.
 * 293 patients in Taiwan
 * 285 patients in Taiwan
 * 274 patients in the USA.
 * 241 patients in the USA, shown to be better than the Cincinnati score
 * 1,013 patients in Helsinki
 * 243 patients in the Phillipines
 * 175 patients in the USA
 * 371 patients in Germany
 * 378 patients in Mexico
 * 37 patients in Brazil
 * 67 patients in the USA
 * 155 patients in the USA
 * 1175 patients in the UK
 * 91 patients in the USA
 * 3,255 patients in China
 * 366 patients in the USA
 * 106 patients in Taiwan
 * 310 patients in China
 * 109 patients in India
 * 399 patients from the Novo Nordisk Recombinant Activated Factor VII Intracerebral Hemorrhage Trial
 * 121 patients in the USA
 * 170 patients in Italy
 * 365 patients in the USA
 * 1338 patients in Singapore
 * 32 patients in the USA
 * 104 patients in India
 * 243 patients in the USA

However, it does not perform as well as the modified ICH score,, the FUNC score, the ICH-GS score, the simplified ICH score, or the ICH-FOS score for prediction of mortality.

It has also used to predict functional outcome (good outcome = mRS 0-2) for which a value less than 3 was 94% sensitive and 61% specific, with a Youden's statistic of 0.54. The same Youden's statistic for determining functional outcome (GOS 4-5) was noted in another study as well but with a cutoff of 1 or less. However, it does not perform as well as the modified ICH score for prediction of  functional outcome.

The score has also been validated for predicting poor functional outcome (mRS ≥ 3) after AVM-associated IPH (AUROC 0.891) but can be improved by modifying some parameters to form the AVM-oICH score (AUROC 0.914).

This score may be more accurate when measured at 24 hours for predicting mortality and functional outcome. For predicting mortality, it may be more sensitive (75% vs 36%) when measured at 72 hours after onset.

A major problem with this score is that it includes a self-fulfilling prophecy. It overestimates mortality and underestimates the number of patients with favorable outcome when patients' care is limited (DNR order or care withdrawn within 24 hours). One study suggested that clinician judgment may actually be superior to it. The ICH score is associated with 12 month functional outcome as well.

Hallevy ICH score (2003)
Derived from 184 IPH patients in Israel, with poor outcome defined as mRS 4-6 at discharge.

Modified ICH score (2003)
This score is similar to the Hemphill et al. ICH score, but uses NIHSS instead of GCS. It is derived from a sample of 142 patients with nontraumatic ICH in Hong Kong in 1999. For mortality a cutoff of ≥ 3 is 92% sensitive and 79% specific, with a Youden's index of 0.711, which gives it better performance than the original ICH score and the new ICH score. For good outcome (mRS 0-2), a cutoff of < 3 is 86% sensitive and 78% specific with a Youden's index of 0.640, which also gives it better performance than the original ICH score and the new ICH score. Thus, of the original ICH score, the modified ICH score, and the new ICH score, the modified ICH score performs the best for both prediction of mortality and functional outcome at 30 days. However, it does not perform as well as the ICH-FOS score for prediction of mortality or functional outcome. Validated in a cohort of 1013 patients in Helsinki, 67 patients in the USA, and 371 patients in Germany

New ICH score (2003)
Derived from a sample of 142 patients with nontraumatic IPH in Hong Kong in 1999. For mortality a cutoff of ≥ 3 is 91% sensitive and 73% specific, with a Youden's index of 0.640. For good outcome (mRS 0-2), a cutoff of < 3 was 70% sensitive and 92% specific, for a Youden's index of 0.617. For neither of these parameters does it perform as well as the modified ICH score. Validated in a cohort of 1013 patients in Helsinki and 67 patients in the USA.

Essen ICH score (2006)
This score was developed in 340 German patients and externally validated in a separate cohort of 371 patients. A score of > 7 was 44% sensitive but 96% specific for mortality (AUROC 0.851), and a score of < 3 best predicted complete recovery (Barthel Index ≥ 95) with sensitivity 85%, specificity 86%, and AUROC of 0.913. For mortality, this score did not perform as well as Cheung's Modified ICH score, and performed equally to the original ICH score. However, for complete recovery (BI ≥ 95), it performed better than both. It does not perform as well as the ICH-FOS score for prediction of mortality or functional outcome. Validated in a cohort of 1013 patients in Helsinki and 67 patients in the USA.

GP on Stage Score (2006)
Derived from 995 IPH patients in seven Asian countries from 2000-2002. It utilizes fever, GCS, IPH volume, and IVH to derive a logistic regression equation to calculate out the likelihood of grave outcome (mRS 5-6). It is too complex to calculate at the bedside and therefore is less clinically useful. Validated in a cohort of 1013 patients in Helsinki.

Modified ICH-A score (2006)
Derived from 153 Argentinian IPH patients. For a score of ≥ 4, the score was 79% sensitive and 80% specific for mortality, with a Youden's statistic of 0.59. For a score of ≤ 3, the score was 85% sensitive and 75% specific for good outcome (GOS 4-5), with a Youden's statistic of 0.60. For neither of these does the score perform as well as Cheung's Modified ICH score, but that score predicts 30 day outcome and this one predicts 6 month outcome. Validated in a cohort of 1013 patients in Helsinki and 67 patients in the USA.

Modified ICH-B score (2006)
Derived from 153 Argentinian IPH patients. A score of ≥ 3 is 85% sensitive and 73% specific for mortality, with a Youden's statistic of 0.58. For good outcome (GOS 4-5), a score of ≤ 3 is 70% sensitive and 90% specific, with a Youden's statistic of 0.60. For neither of these does the score perform as well as Cheung's Modified ICH score, but that score predicts 30 day outcome and this one predicts 6 month outcome. Validated in a cohort of 1013 patients in Helsinki and 67 patients in the USA.

Intracerebral Hemorrhage Grading Scale (ICH-GS) (2007)
Derived from 378 IPH patients in Mexico. This score had higher sensitivity for in-hospital (78% vs 64%, AUROC 0.880 vs 0.831) and 30-day mortality (79% vs 65%, AUROC 0.882 vs 0.832) than the original ICH score. For good functional outcome (GOS 4-5 at 30 days), ICH-GS also outperformed the original ICH score (AUROC 0.862 vs 0.821). The score was validated in a separate cohort of 366 patients and showed an AUC of 0.888 for predicting 90-day mortality. . However, it does not perform as well as the ICH-FOS score for predicting mortality or functional outcome. Validated in a cohort of 1013 patients in Helsinki, 67 patients in the USA, 366 patients in the USA, 1175 patients in the UK, 3,255 patients in China, 170 patients in Italy, and 1338 patients in Singapore.

FUNC score (2008)
Derived in 418 patients with primary IPH in the USA and validated on an additional 211 patients with an AUROC of 0.88. This score was also validated in a separate 366 person cohort, with AUC 0.87 for predicting mortality, but the ICH-GS score performed better in this regard. An online calculator is available here. However, it does not perform as well as the ICH-FOS score for functional outcome. Validated in a cohort of 1013 patients in Helsinki, 366 patients in the USA, 121 patients in the USA, 92 patients in the USA, and 170 patients in Italy.

Simplified ICH score (2009)
Derived from 293 patients with IPH in Taiwan to predict mortality with 83% sensitivity, 80% specificity, and AUROC 0.89, which was better than the original ICH score (AUROC 0.74) and the ICH-GS (AUROC 0.74). However, it does not perform as well as the ICH-FOS score for prediction of mortality. Validated in a cohort of 1013 patients in Helsinki, 67 patients in the USA, and 1338 patients in Singapore.

Creutzfeldt ICH score (2011)
This score has great accuracy but is derived from logistic regression which makes it cumbersome to calculate for regular use. However, this study was crucial for highlighting the self-fulfilling prophecy of other scores, by showing that the original ICH score (for example) underestimates functional outcome because it includes patients for whom care is withdrawn.

FOUR + FUNC score (F2 score) (2011)
This score utilizes a combination of the FOUR score and the aforementioned FUNC score. If the FOUR score is ≤ 10 and the FUNC score is ≤ 5, then the F2 score is considered positive, and a positive F2 score has 95% PPV for mortality and 100% PPV for poor outcome (GOS 1-3) in one study of 92 patients in the USA.

Landseed ICH score (2011)
Derived from 285 IPH patients in Taiwan from 2003-2006. However, it did not perform better than the original ICH score for prediction of in-hospital mortality (AUROC 0.867 for both) or unfavorable outcome (mortality or BI <40 at discharge) (AUROC 0.817 for Landseed ICH score and 0.834 for original ICH score). It also does not perform as well as the ICH-FOS score for prediction of mortality or functional outcome. Validated in a cohort of 1013 patients in Helsinki.

ICH Index (2012)
This score was derived from 227 patients in China and is calculated with the following equation:"ICHI = [age (years) / 10] + [serum glucose (mmol/L)] + [WBC (109/L)] - [ GCS score]."A value of 18 or higher was 65% sensitive and 95% specific for in-hospital mortality (Youden's index 0.60). Validated in a cohort of 1013 patients in Helsinki.

GWTG Stroke score (2013)
This score as derived from an extremely large sample of over 300,000 stroke patients (all types) from the Get With the Guidelines - Stroke Database. It included 37,509 patients with IPH, for which the AUROC for mortality was 0.82. Validated in a cohort of 1013 patients in Helsinki.

ICH-Functional Outcome Score (ICH-FOS) (2013)
This score was derived from 1,953 IPH patients and validated in 1,302 IPH patients in China for prediction of 1 year mortality. In this cohort the score outperformed (with higher AUROC) the original ICH score, Modified ICH score, Essen ICH score, ICH-GS score, FUNC score, MICH score, Simplified ICH score, and Landseed ICH score. . In another cohort of 1013 patients in Helsinki, this score outperformed nearly all of the other scores published by 2017 in prediction of 3-month mortality.

EDICH (2014) and Modified EDICH (2015)
Derived from 191 patients with IPH in Greece. A score of ≥ 3 is 84% sensitive and 87% specific for 30-day mortality, while a score of ≥ 6 is 100% specific for 30-day mortality. This score was validated and modified in 2015 in a cohort of 170 Italian patients, which gave it better performance for 48 hours hospital mortality (AUROC 0.971 vs 0.929) and poor functional outcome defined as mRS ≥ 4 (AUROC 0.850 vs 0.807).

SUSPEKT (2014)
This score was devised from a sample of 156 supratentorial IPH patients in Hungary. They developed a formula using serum glucose, IPH size, admission systolic BP, presence or absence of IVH, serum potassium, and age to predict mortality. It is based on a logistic regression equation and complex to calculate which limits its clinical usefulness.

APACHE II for IPH (2016)
In a series of 546 acute IPH patients in China, APACHE II scores were calculated and correlated with functional outcome and mortality. It was validated in 104 patients, but the ICH score performed better.

Intracerebral Hemorrhage Outcomes Project (ICHOP12) score (2017)
This score was derived from a database of 321 patients with IPH at Columbia University Medical Center from 2009-2016 to predict poor functional outcome, which it did with AUROC of 0.87, which was better than the original ICH score. However, as this includes the APACHE II physiological score it is more complicated to calculate.

Max-ICH score (2017)
This score was derived from a cohort of 583 patients in Germany. 471 of these patients were maximally treated without care limitations in the first 24 hours, and the Max-ICH score was developed from this cohort, for which it had higher AUROC compared with the original ICH score (0.81 vs 0.72) for long-term outcome and for mortality (0.77 vs 0.69). . This has a major advantage in that it is not limited by self-fulfilling prophecy.

ICH scoreFS (2018)
This score was evaluated in 274 patients in the US and incorporates aspects of the FOUR score into the original ICH score instead of GCS. It did not clearly perform better than the original ICH score for predicting mortality.

AVM-oICH score (2012)
This score is a modification of the original ICH score to make it more predictive of poor 3-month functional outcome (mRS ≥ 3). The age cutoff is changed from 80 to 41 years and the ICH volume cutoff is changed from 30 to 37 mL. These changes increase the AUROC from 0.891 to 0.914.

AVICH score (2016)
This score was derived from 67 patients and incorporated elements of the Spetzler-Martin grading system, the supplemented Spetzler-Martin scale, and the original ICH score. In its study it was more accurate than the original ICH score and the supplemented Spetzler-Martin grade for predicting functional outcome at 3 months (AUROC 0.837 vs 0.729 for ICH score and 0.792 for supplemented Spetzler-Martin) or 1 year (AUROC 0.811 vs 0.644 for ICH score and 0.800 for supplemented Spetzler-Martin). A cutoff of 9 or higher was highly specific for unfavorable outcome (mRS 3-6).

Vascular Structure-Related ICH Score (VSARICH) (2015)
This score was derived in 307 vascular-lesion IPH patients and validated in another 99 patients, all in China, to evaluate prognosis in vascular structural abnormality-related IPH, most of whom (~50%) had AVMs, but also aneurysms (27-31%), moyamoya disease (12-13%) and cavernous hemangiomas (7%). The AUROCs were 0.93-0.94 for mortality and 0.81-0.84 for good clinical outcome in the derivation and validation cohorts.

Cho's Modified ICH score (2008)
Derived from 226 patients with basal ganglia hemorrhage to help guide which patients are most likely to benefit from surgery. Best cutoff for mortality is ≥ 3, with Youden's index of 0.66. When used for all patients, however, it does not perform as well as the ICH-FOS score for prediction of mortality or functional outcome. Validated in a cohort of 1013 patients in Helsinki, 285 patients from Taiwan, 32 patients from China, and in 1175 patients in the UK.

Minimally Invasive Surgery (MIS) score (2017)
Derived from 101 patients with spontaneous IPH in China who underwent minimally-invasive surgery (drainage tubes placed into the hematomas, urokinase injected twice daily). They showed that a score of 0-1 was associated with a 70% chance of good functional outcome (mRS 0-3), while a score of ≥2 was only associated with a 20% chance of good functional outcome.

Surgical SwICH score (2019)
As most patients in other series were treated conservatively, this score was developed in 401 Swedish IPH patients from 2011-2015. Patients treated with surgery had considerably lower mortality than expected based on the original ICH score.

ICH Outcome Score (ICHOS) (2009)
This score was developed from 107 hemodialysis patients with IPH from 1994 to 2004 in Taiwan and seemed to outperform the original ICH score in this population. It was subsequently validated in a cohort of 38 additional patients with similar results.