Risk prediction models for intracranial hemorrhage in acute ischemic stroke patients receiving intravenous alteplase treatment: a systematic review

Objectives To identify and compare published models that use related factors to predict the risk of intracranial hemorrhage (ICH) in acute ischemic stroke patients receiving intravenous alteplase treatment. Methods Risk prediction models for ICH in acute ischemic stroke patients receiving intravenous alteplase treatment were collected from PubMed, Embase, Web of Science, and the Cochrane Library up to April 7, 2023. A meta-analysis was performed using Stata 13.0, and the included models were evaluated using the Prediction Model Risk of Bias Assessment Tool (PROBAST). Results A total of 656 references were screened, resulting in 13 studies being included. Among these, one was a prospective cohort study. Ten studies used internal validation; five studies used external validation, with two of them using both. The area under the receiver operating characteristic (ROC) curve for subjects reported in the models ranged from 0.68 to 0.985. Common predictors in the prediction models include National Institutes of Health Stroke Scale (NIHSS) (OR = 1.17, 95% CI 1.09–1.25, p < 0.0001), glucose (OR = 1.54, 95% CI 1.09–2.17, p < 0.05), and advanced age (OR = 1.50, 95% CI 1.15–1.94, p < 0.05), and the meta-analysis shows that these are independent risk factors. After PROBAST evaluation, all studies were assessed as having a high risk of bias but a low risk of applicability concerns. Conclusion This study systematically reviews available evidence on risk prediction models for ICH in acute ischemic stroke patients receiving intravenous alteplase treatment. Few models have been externally validated, while the majority demonstrate significant discriminative power.


Introduction
According to the latest global stroke data, approximately 12.2 million new stroke cases and 101 million existing stroke patients are reported worldwide each year, making it the second leading cause of death globally (1,2).Ischemic strokes account for about 62.4-82.6% of these cases (1,3).Revascularization methods for patients with acute ischemic stroke (AIS) include intravenous thrombolysis, intra-arterial thrombolysis, endovascular thrombectomy, and bridging thrombolytic therapy with endovascular thrombectomy (4,5).The American Stroke Association (ASA) recommends intravenous thrombolysis with alteplase within 4.5 h of AIS onset as the first-line treatment (6).Alteplase is currently the only approved intravenous medication for AIS (7).However, its use can lead to complications such as hemorrhage, vascular edema, and seizures (8,9).Intracranial hemorrhage (ICH) is the most severe complication, potentially resulting in prolonged hospital stays, increased medical costs, and a higher likelihood of disability and death (10).The one-year mortality rate associated of ICH patients is reported to be 52.2%(11).Therefore, the early recognition and management of ICH are crucial for AIS patients receiving intravenous alteplase.Numerous models for predicting ICH risk in AIS patients treated with intravenous alteplase have been developed worldwide.The purpose of this study is to thoroughly review and systematically summarize and compare these risk prediction models for ICH (including asymptomatic and symptomatic intracranial hemorrhage), aiming to enhance their construction and application in the management of AIS patients treated with intravenous alteplase.

Methods and analysis
This research protocol is registered in the International Prospective Register of Systematic Reviews (PROSPERO) with the registration number CRD42023415649.

Search strategy
A literature search was conducted in the following databases: PubMed, Embase, Web of Science, and the Cochrane Library.The search terms included "ischemic stroke," "intracranial hemorrhage," "alteplase," and "factor."Our complete search string can be found in Supplementary material S1.The search deadline was April 7, 2023.

Eligibility criteria
The inclusion criteria for literature are: (1) type of research, either cohort study or case-control study; (2) participants, patients treated with alteplase after an acute ischemic stroke (AIS) who experienced intracranial hemorrhage, including asymptomatic and symptomatic intracranial hemorrhage; (3) studies providing relevant content on model construction and validation.The exclusion criteria are: (1) lack of practicality of predictive factors in clinical practice; (2) inability to obtain the full text; (3) duplicate publications.

Literature screening and data extraction
Two researchers independently screened the literature and conducted cross-checks.Disagreements were resolved by consulting a third researcher.Information from the included literature was extracted regarding basic information, model performance, model composition, etc., as shown in Tables 1, 2.

Statistical analysis
A meta-analysis was performed using Stata (version 13.0).The relationship between risk factors and ICH in patients with AIS receiving intravenous alteplase treatment was explored through the odds ratio (OR) and corresponding 95% confidence interval (CI).Heterogeneity between studies was detected using the Q-test, and appropriate effect models were selected.A sensitivity analysis was conducted by sequentially excluding studies.If Begg's test and/or Egger's test (p < 0.05) indicated publication bias, the trim-and-fill method was employed for reassessment.

Bias risk assessment
Two researchers used the Prediction Model Risk of Bias Assessment Tool (PROBAST) (24) to evaluate the included study's bias risk and consulted a third researcher in cases of disagreement.PROBAST evaluates the risk of bias and the applicability of the model across four domains: participants, predictors, outcomes, and analysis.

Predictive performance
The predictive performance of the model was evaluated through discrimination and calibration.The discrimination index is the Area Under the Curve (AUC), with an AUC ≥ 0.7 indicating good discrimination of the model, and externally tested AUC being more convincing.Calibration indicators include the Hosmer-Lemeshow test and calibration plots.A model is considered well-fitted if the Hosmer-Lemeshow test yields a p-value >0.05, or if the calibration plot's slope is close to 1.

Meta-analysis for risk factors
A meta-analysis was conducted for NIHSS, glucose, and advanced age.Some studies did not provide effective OR values and 95% CIs, hence they were excluded from the meta-analysis.The results indicated that NIHSS, glucose, and advanced age are independent risk factors for ICH in AIS patients treated with intravenous alteplase (Table 3).Nine studies (10, 13, 14, 16, 18-22) examined the impact of NIHSS on ICH in patients treated with intravenous alteplase post-AIS (heterogeneity: p = 0.001, I 2 = 84.2%).There was a significant difference between the groups (95% CI: 1.09-1.25,p < 0.0001; Figure 2).A subgroup meta-analysis based on sample size was conducted due to significant differences in sample sizes included.The meta-analysis revealed NIHSS as an independent risk factor for ICH in all patient subgroups (Table 4).Sensitivity analysis indicated that excluding any study did not significantly alter the meta-analysis results (Figure 3).Begg's test (p < 0.05) and Egger's test (p < 0.05) suggested publication bias (Figures 4A,B).The trim-and-fill method was used to correct for publication bias.After three iterations using the linear method, the adjusted OR remained significant (OR = 1.096, 95% CI 1.013-1.185,p < 0.05), indicating the reliability of the results (Table 4).

Methodological quality evaluation
Among the thirteen pieces of literature included, the bias risk assessment results indicated a high risk for all, primarily due to  Sensitivity analysis for the association between NIHSS and ICH.
Meta-analysis of NIHSS on ICH in patients receiving intravenous alteplase therapy after AIS.Meta-analysis of glucose on ICH in patients receiving intravenous alteplase therapy after AIS.
Sensitivity analysis for the association between glucose and ICH.Plots for publication bias test in meta-analysis for the association between glucose and ICH.(A) Begg's funnel plot; (B) Egger's publication bias plot.
Meta-analysis of advanced age on ICH in patients receiving intravenous alteplase therapy after AIS.
participant selection and statistical analysis aspects.However, the applicability evaluation results for all the studies were considered low-risk (Table 5).

Discussion
Overall, the development of risk prediction models for intracranial hemorrhage (ICH) in patients with acute ischemic stroke (AIS) receiving intravenous alteplase treatment is still in its early stages.The research spans a considerable time frame, with a primary focus on America, Thailand, Italy, and China.Most of the models lack external validation and have not yet been implemented clinically.

Discussion on overall bias risk
The risk of bias in prediction models is closely associated with participants, predictors, outcomes, and analysis.All 13 articles included in our study exhibited a high risk of bias (10, [12][13][14][15][16][17][18][19][20][21][22][23].According to PROBAST, data from prospective cohort studies are considered more reliable than those from retrospective cohort studies (24).However, our study included only one prospective cohort study (10).PROBAST also stipulates that to avoid overfitting, the modeling sample size should include more than 20 events per variable (EPV), and the validation sample size should  Sensitivity analysis for the association between advanced age and ICH.
Plots for publication bias test in meta-analysis for the association between advanced age and ICH.

Prediction factor analysis
The models included in this study encompass a variety of predictive factors, including general data, disease-related information, biochemical indicators, and imaging results.While each model comprises different predictive factors, common elements are present.Notably, NIHSS, glucose, and advanced age were found to be strongly associated with intracranial hemorrhage (ICH) in patients with acute ischemic stroke (AIS) receiving intravenous alteplase treatment.Meta-analysis confirms that these factors are independent risk factors, aligning with the findings of numerous related studies (28)(29)(30).The National Institutes of Health Stroke Scale (NIHSS) is widely used for assessing stroke severity and is endorsed by the American Stroke Association (ASA) guidelines as an effective tool for emergency departments to evaluate stroke severity (31,32).All models included in our study featured NIHSS, and meta-analysis validated that a higher NIHSS score is an independent risk factor for ICH.Whiteley WN et al. conducted a meta-analysis of 55 articles on ICH risk factors in AIS patients treated with intravenous alteplase, concluding that a higher NIHSS score is a risk factor for ICH (33).Teekaput et al.'s retrospective study on 725 AIS patients who received alteplase treatment demonstrated that the incidence of ICH in patients with a higher baseline NIHSS was 1.9 times that of the control group (11).Patients with AIS and a high baseline NIHSS typically have larger infarct areas and more extensive vascular damage, increasing their  (37).This is particularly relevant for patients over 80 years old receiving alteplase treatment (38).As patients age, their overall physical condition deteriorates, the prevalence of cardiovascular diseases rises, blood vessel elasticity decreases, and brain parenchyma undergoes degenerative changes, all of which contribute to a higher risk of ICH after thrombolysis (38,39).

Advantages and limitations
Advantages (1) This study systematically integrates recent risk prediction models for intracranial hemorrhage (ICH) in patients with acute ischemic stroke (AIS) receiving intravenous alteplase treatment, emphasizing the modeling method, model performance, predictive factors, and factor assignment. (2) It employs PROBAST to evaluate the quality of these models, analyze risk sources, and provide a reference for future research.
(3) The study enhances the reliability of its conclusions by supplementing quantitative analysis through meta-analysis.

Limitations
(1) The study is limited to literature published in English.
(2) There are variations in the study populations and the standards for defining ICH among the included studies.(3) Some models lack validation, necessitating further research to verify their applicability.

Conclusion
In summary, there has been a steady increase in the number of risk prediction models for ICH in AIS patients treated with intravenous alteplase.However, the performance of these models varies.The 13 models included in this study present a high risk of bias due to statistical analysis issues, but they generally exhibit a low risk regarding overall applicability, which is beneficial for the early screening of highrisk patients.Due to the high overall bias risk in the included studies, it is not advisable to directly apply these predictive models in clinical practice.Medical professionals should aim to facilitate the application and generalization of existing models through external validation involving multiple centers and large samples or conduct large-sample prospective research to develop new predictive models in accordance with TRIPOD and PROBAST guidelines.

FIGURE 1 PRISMA
FIGURE 1 PRISMA flow diagram of study selection process.

FIGURE 4
FIGURE 4 Plots for publication bias test in meta-analysis for the association between NIHSS and ICH.(A) Begg's funnel plot; (B) Egger's publication bias plot; (C) Filled funnel plot.

TABLE 1
Basic characteristics and effectiveness evaluation of the included models.

TABLE 2
Predictors and stratification methods included in the study.

TABLE 3
The meta-analysis for risk factors.

TABLE 4 Subgroup
meta-analysis of NIHSS based on sample size.REM, random-effects model; FEM, fixed-effects model.Frontiers in Neurology 07 frontiersin.org

TABLE 5
(35,36) bias assessment results included in the model (PROBAST).Therefore, careful consideration of treatment methods, enhanced monitoring, and proactive intervention are crucial for patients with high baseline NIHSS scores to effectively prevent ICH.Glucose is another independent predictor of ICH, consistent with previous research findings.Hyperglycemia is common in the acute phase of AIS, often resulting from stress-induced increases in cortisol and catecholamine levels following ischemic injury.Elevated glucose levels are associated with stroke severity and adverse outcomes(35,36).Advanced age is also a significant predictor.Dong S. et al.'s meta-analysis of 25 cohort studies established that advanced age is an independent predictor of ICH in AIS patients