Abstract
Background:
Surgical treatment demonstrated a reduction in mortality among patients suffering from severe spontaneous intracerebral hemorrhage (SSICH). However, which SSICH patients could benefit from surgical treatment was unclear. This study aimed to establish and validate a decision tree (DT) model to help determine which SSICH patients could benefit from surgical treatment.
Materials and methods:
SSICH patients from a prospective, multicenter cohort study were analyzed retrospectively. The primary outcome was the incidence of neurological poor outcome (modified Rankin scale as 4–6) on the 180th day posthemorrhage. Then, surgically-treated SSICH patients were set as the derivation cohort (from a referring hospital) and validation cohort (from multiple hospitals). A DT model to evaluate the risk of 180-day poor outcome was developed within the derivation cohort and validated within the validation cohort. The performance of clinicians in identifying patients with poor outcome before and after the help of the DT model was compared using the area under curve (AUC).
Results:
One thousand two hundred sixty SSICH patients were included in this study (middle age as 56, and 984 male patients). Surgically-treated patients had a lower incidence of 180-day poor outcome compared to conservatively-treated patients (147/794 vs. 128/466, P<0.001). Based on 794 surgically-treated patients, multivariate logistic analysis revealed the ischemic cerebro-cardiovascular disease history, renal dysfunction, dual antiplatelet therapy, hematoma volume, and Glasgow coma score at admission as poor outcome factors. The DT model, incorporating these above factors, was highly predictive of 180-day poor outcome within the derivation cohort (AUC, 0.94) and validation cohort (AUC, 0.92). Within 794 surgically-treated patients, the DT improved junior clinicians’ performance to identify patients at risk for poor outcomes (AUC from 0.81 to 0.89, P<0.001).
Conclusions:
This study provided a DT model for predicting the poor outcome of SSICH patients postsurgically, which may serve as a useful tool assisting clinicians in treatment decision-making for SSICH.
Keywords: decision tree, poor outcome, spontaneous intracerebral hemorrhage, surgical treatment, treatment decision-making
Introduction
Highlights
Although surgery could prevent severe spontaneous intracerebral hemorrhage (SSICH) patients from poor outcome, for surgically-treated patients, the current study revealed ischemic cerebro-cardiovascular disease history, renal dysfunction, dual antiplatelet therapy, hematoma volume, and Glasgow coma score at admission as factors of poor outcome.
A decision tree model was developed within 794 surgically-treated SSICH patients and performed well to predict a 180-day poor outcome, which was superior to the existing scale, that is ICH score.
The application of the decision tree could improve junior clinicians’ performance to identify patients at risk for poor outcomes (area under the curve from 0.81 to 0.89, P<0.001).
Severe spontaneous intracerebral hemorrhage (SSICH) is the most critical and lethal variety of hemorrhagic stroke, hallmarked by extensive intracerebral hemorrhage and rapid cerebral herniation progression1,2. Surgery may afford a life-preserving solution for SSICH patients by diminishing intracranial pressure; however, certain patients may have a deterioration in functional status due to factors, for example, postoperative rebleeding, major adverse cerebrovascular, and cardiovascular events (MACCE), among others1–5. These factors and events may culminate in significant disability and a poor surgical outcome, thereby negating the potential benefits of surgery for SSICH patients. Thus, identifying SSICH patients who could benefit from surgical treatment can greatly assist in clinical decision-making processes.
Previous studies reported several factors related to poor outcome of SSICH patients postoperatively, for example, large hematoma volume, deep hematoma location, postoperative rebleeding, and intraventricular hemorrhage (IVH)2,4,6,7. Notably, hemorrhagic stroke frequently afflicts the elderly population8, and neurosurgical treatment can heighten the risk of MACCE5. However, sparse research has investigated the impact of ischemic events on the prognosis of surgically-treated SSICH patients. Furthermore, the absence of a model capable of determining, which SSICH patients might benefit from surgery renders patient stratification for surgery or conservative management challenging. Moreover, previous studies have often been constrained by their small sample size, single-center cross-sectional design, and the paucity of large-scale, multicenter longitudinal data within the Chinese demographic. In addition, a decision tree (DT) model, with its clear cutoff points, presents a user-friendly tool that can significantly aid treatment decision-making in clinical settings.
In the present study, we investigated the clinical and radiological characteristics of sSCIH patients from a prospective, multicenter Chinese cohort (Surgical treatment for AntiPlatelet IntraCerebral Hemorrhage, SAP-ICH cohort)9. Our aim was to establish a DT model capable of forecasting the poor outcome of SSICH patients following surgery. We will also evaluate whether the model can improve the clinicians’ treatment decision.
Methods
Study design and population
The SAP-ICH cohort was a prospective, multicenter cohort study (unique identifier: ChiCTR1900024406), which recruited SSICH patients from seven medical centers from September 2019 through December 2022, with the aim of evaluating the efficacy and safety of surgical interventions for SSICH patients9. This study was complied with STROBE protocol and approved by the institutional review board. All patients (or guardians) provided written informed consents. This study followed the STROCSS guidelines for observational cohort studies10.
The study design is depicted in Supplemental Figure 1.A (Supplemental Digital Content 1, http://links.lww.com/JS9/B301). Initially, we examined whether surgical treatment could preclude death in SSICH patients. Subsequently, we developed a DT model to discern which SSICH patients would benefit from surgical treatment. Finally, the influence of DT model on treatment decisions for SSICH was investigated.
SSICH patients were included from the SAP-ICH cohort (Supplemental Figure 1.B (Supplemental Digital Content 1, http://links.lww.com/JS9/B301). The inclusion criteria entailed: (1) age between 18 and 75 years; (2) diagnosis of severe intracerebral hemorrhage as determined by neurological condition (Glasgow Coma Score [GCS] <13) and radiological findings (supratentorial hematoma volume >30 ml, infratentorial hematoma volume >10 ml, midline shift >1 cm, or severe IVH). Then, we excluded patients with (1) cerebrovascular diseases such as intracranial aneurysm or cerebrovascular malformation; (2) intracranial tumors associated with hemorrhage; (3) hemorrhagic transformation of cerebral infarction; (4) intracerebral hemorrhage precipitated by venous thrombosis; (5) severe coagulation disorders such as hemophilia, or coagulation dysfunction due to malignant tumors and hypohepatia; (6) prior antithrombotic therapy (vitamin K antagonist and others) before hemorrhage; (7) death before hospital arrival or within a short period (6 h) postadmission; (8) COVID-19 positive cases ;(9) absence of sufficient follow-up data.
Follow-up and outcome
All included SSICH patients were monitored through outpatient visits and telephonic consultations every 3 months until either death or 180 days posthemorrhage. During each follow-up, investigators recorded survival status, causes of death (if applicable), and assessed functional state using the modified Rankin Score (mRS)11. Typically, patients underwent radiological examinations, that is, head CT examination, and/or head MRI examination, on the 30th day and 180th day posthemorrhage. The aim of this study was to identify the SSICH patients who could benefit from surgery. The primary outcome was the neurological function outcome of SSICH patients on the 180th day after hemorrhage, evaluated using the modified Rankin score (mRS). A 180-day mRS of 4–6 was defined as the poor outcome, while mRS of 4–6 were deemed as good outcome. The mRS is a globally recognized measure of neurological function, ranging from 0 to 6, with 0–3 indicating no or mild/moderate disability and 4–6 indicating severe disability or death11. To circumvent evaluation bias, in-hospital and follow-up records were collected and independently assessed by two experienced neurosurgeons (with more than 10 years of work experience and blinded to patients’ information). Any disagreements were resolved through consultation with a senior neurosurgeon (with more than 20 years of work experience). The second outcome was the mortality on the 180th day posthemorrhage, and causes of death were duly recorded.
Clinical data collection
We collected baseline information including age, sex, comorbidities (including the history of dyslipidemia, diabetes mellitus, intracerebral hemorrhage, ischemic cerebrovascular and cardiovascular disease and renal dysfunction), antiplatelet therapy before hemorrhage, tobacco and alcohol consumption, and laboratory examination after admission [platelet count, activated partial thromboplastin time (APTT), international normalized ratio (INR) and fibrinogen], GCS at admission, as well as surgical methods.
Ischemic cerebrovascular and cardiovascular diseases included transient ischemic attacks, ischemic stroke, coronary artery disease, and myocardial infarction. Renal dysfunction covered prior acute and chronic renal failure, and renal dysfunction postadmission [glomerular filtration rate <90 ml/(min×1.73m2)]12–14. Antiplatelet therapy prior to hemorrhage was recognized if the patient received antiplatelet therapy for over seven days but discontinued such therapy less than 7 days before hemorrhage15–17. Considering that no patients in this cohort used more than three antiplatelet medications and merely 11 patients took antiplatelet medications other than aspirin and clopidogrel, all included patients were categorized as none, single antiplatelet therapy, and dual antiplatelet therapy (DAPT). Surgical method was classified as craniotomy, craniotomy + craniectomy, endoscopic evacuation, and minimally invasive evacuation.
Alcohol use was classified as regular (one or more drink per week) and nonregular (less than one drink per week, or no drink)18. Smoking status was defined as current if ongoing or smoking cessation less than 1-year19, otherwise as noncurrent (smoking cessation more than 1-year, or no smoking).
Thrombocytopenia was defined as platelet count less than 100×10920. Coagulation disorder was defined as APTT greater than or equal to 60 s, or INR greater than or equal to 1.5, or fibrinogen less than 1.5 mg/dl20.
Radiological images analysis
Radiological assessments were executed based on CT source images. A neurosurgeon and a neuroradiologist, both with work experience exceeding 5 years and blinded to patients’ information, independently evaluated radiological features. These features included hematoma location, hematoma volume, and IVH. Hematoma location was stratified into supratentorial and infratentorial categories. Hematoma volume was calculated employing the 3DSlicer (a free software, available at https://www.slicer.org, see Supplemental Figure 2.A, Supplemental Digital Content 2, http://links.lww.com/JS9/B302). The average of hematoma volumes measured by the two investigators was utilized for further analysis.
Management of SSICH patients
Upon admission, all SSICH patients received standard care as per the guidelines1, inclusive of blood pressure management, airway management, dynamic CT monitoring, among others. Surgical treatment decisions were based on the consensus between physicians and patients’ guardians. In this study, surgical interventions were carried out by senior neurosurgeons (with work experience exceeding 10 years) immediately after admission or within 24 h after the indications for surgery arose. For patients receiving surgical treatment, CT scans were routinely performed at the 4th, 24th, and 48th h postsurgically. Furthermore, should the state of consciousness deteriorate, an immediate CT scan would be conducted.
DT model establishment
The DT model was established by using the ʽtidymodelsʼ and ʽrpartʼ package in the R statistical software (version 4.2.1). Seven hundred ninety-four SSICH patients receiving surgical treatments were then divided into the derivation cohort (496 patients recruiting from a referring hospital) and validation cohort (298 patients recruiting from multiple hospitals). Within the derivation cohort, patients were randomly assorted into the training set (396 patients, 80%) and the testing set (100 patients, 20%). Firstly, we used logistic analysis and hyperparameters optimization (with fivefold cross-validation) to select features related to poor outcome of SSICH patients after surgical treatment. To counteract the imbalance between the number of patients with poor outcomes and those with good outcomes, the synthetic minority oversampling technique21 (SMOTE) was employed. For the different effect of supratentorial and infratentorial hematoma on patients’ outcome, the hematoma location was also put into the DT model to investigate the effect of hematoma volume on poor outcome. The parameters of the DT model were set as follows: ʽGiniʼ impurity criterion; ʽclassification modeʼ; the maximum depth of the tree as 30, the minimum number of samples required to split an internal node as 5, and the minimum number of samples required to be at a leaf node as 2; for continuous variables, regression discontinuity analysis was performed per cycle to find the cutoff value for poor outcome. To further assess overfitting, we plotted a DT model learning curve based on the area under the curve (AUC) of the training and testing sets. An average of all repetitions’ training AUCs (based on the training set) was employed, and a 95% CI was calculated based on the fivefold cross-validation results. The model was deemed as not overfitting if the testing AUC (based on the testing set) fell within the 95% CI of the training AUC. The model’s accuracy in classifying patients with poor outcomes versus good outcomes was externally validated using the validation cohort. The accuracy, specificity, sensitivity, positive predictive value, and negative predictive value of the model in differentiating poor and good outcomes were calculated.
Decision-making analysis
To examine whether the DT model could aid in the decision-making process for surgical treatment in SSICH patients, we compared the clinicians’ ability of identifying patients with poor outcomes before and after the assistance of the DT model. Four junior clinicians (with 3 year of independent experience, blinded to SSICH patients’ information and outcome) and two senior clinicians (with over 10 year of independent experience, also blinded to SSICH patients’ information and outcome) were asked to classify all SSICH patients first through their individual methodologies, then with the help of the DT model. To evade recency or order effects, the order of SSICH patient cases with or without the DT model was random and varied. A washout period of 4 weeks was allocated between two assessments. Following this, based on the clinicians’ classifications of patients with poor or good outcomes before and after the DT model’s assistance, the net reclassification improvement (NRI) and integrated discrimination improvement (IDI) were calculated22. If NRI or IDI was greater than 0, the predictive ability of the clinicians was improved.
Statistics analysis
Statistical evaluations were conducted with SPSS software (version 24.0). Normally distributed continuous variables were represented as means and SD, and medians (m) and inter-quartile range for non-normally distributed data. Categorical variables were presented as numbers (no.) and percentages (%). Differences between continuous variables were compared using Student’s t-tests or Wilcoxon rank sum tests, and differences in categorical variables using χ2 tests or Fisher’s exact tests. To identify the factors related to poor outcome of SSICH patients after surgical treatment, univariate and multivariate logistic regression analyses were performed. The result was presented as an odds ratio (OR) and 95% CI. All SSICH patients receiving surgical treatment were stratified into high-risk and low-risk groups by the DT model, and senior and junior clinicians. The performance of models for SSICH patients with poor outcomes postsurgically was evaluated using the AUC and receiver operator characteristic analysis. Model with AUC greater than 0.8 was identified as having clinical utility. The AUC value and accuracy were compared using the Z-test.
Results
Surgical treatment could decrease the risk of poor outcome in SSICH patients
This current study included 1260 patients out of 1416 SSICH patients, during 1351.59 person-years of follow-up (mean follow-up time as 1.07±0.57 year). The median age of all patients was 56 years (range, 46–63), and 984 (78.1%) patients were male. Fifty-nine (4.7%) patients received DAPT and 64 (5.1%) patients had coagulation disorder. The middle GCS score at admission was 10 (6–13). For hematoma, the median hematoma volume was 46.4 (34.0–64.1) ml. One thousand one hundred and fourteen (88.4%) hematoma were supratentorial, and 146 (11.6%) were subtentorial.
Out of the 1260 SSICH patients, 794 (63.0%) patients underwent surgical treatment, and 466 (37.0%) patients received conservative treatment. The differences between both treatment groups are summarized in Table 1. For surgically-treated patients, 362 (45.6%) patients underwent craniotomy and 432 (54.5%) received minimal invasive surgery. Significant differences were found in age (P<0.001), antiplatelet therapy before hemorrhage (P<0.001), hematoma volume (P<0.001), IVH (P=0.030), coagulation disorder (P<0.001), and GCS score at admission (P<0.001), between the surgical and conservative treatment groups.
Table 1.
The information of all SSICH patients.
| Characteristics | Overall n=1260 | Surgical n=794 | Conservative n=466 | P |
|---|---|---|---|---|
| Age, m (IQR), y | 56 (46–63) | 54 (44–62) | 59 (50–64) | <0.001┼ |
| Male, no. (%) | 984 (78.1%) | 608 (76.6%) | 376 (80.7%) | 0.089 |
| Comorbidities, no. (%) | ||||
| Diabetes mellitus | 150 (11.9%) | 102 (12.8%) | 48 (10.3%) | 0.178 |
| Dyslipidemia | 99 (7.9%) | 56 (7.1%) | 43 (9.2%) | 0.166 |
| History of ICCD | 620 (49.2%) | 396 (49.9%) | 224 (48.1%) | 0.536 |
| History of ICH | 34 (2.7%) | 22 (2.8%) | 12 (2.6%) | 0.836 |
| Renal dysfunction | 194 (15.4%) | 123 (15.5%) | 71 (15.2%) | 0.904 |
| Current-or-ever smoker, no. (%) | 364 (28.9%) | 244 (28.2%) | 140 (30%) | 0.489 |
| Regular drinkers, no. (%) | 206 (16.3%) | 122 (15.4%) | 84 (18%) | 0.311 |
| Antiplatelet therapy, no. (%) | <0.001┼ | |||
| No antiplatelet therapy | 902 (71.6%) | 610 (76.8%) | 292 (62.7%) | |
| Single antiplatelet therapy | 299 (23.7%) | 153 (19.3%) | 146 (31.3%) | |
| DAPT | 59 (4.7%) | 31 (3.9%) | 28 (6%) | |
| Hematoma location, no. (%) | 0.145 | |||
| Supratentorial | 1114 (88.4%) | 710 (89.4%) | 404 (86.7%) | |
| Subtentorial | 146 (11.6%) | 84 (10.6%) | 62 (13.3%) | |
| Hematoma volume, m (IQR), ml | 46.4 (34.0–64.1) | 55.1 (36.0–72.9) | 39.7 (32.9–48.2) | <0.001┼ |
| IVH, no. (%) | 656 (52.1%) | 432 (54.4%) | 224 (48.1%) | 0.030┼ |
| Platelet count, m (IQR), ×109 | 214 (178–256) | 212 (175–254) | 216 (186–259) | 0.211 |
| Thrombocytopenia, no. (%) | 30 (2.4%) | 16 (2%) | 14 (3%) | 0.266 |
| APTT, m (IQR), s | 26.6 (24.4–29.9) | 26.5 (24.0–29.8) | 27.2 (24.7–30.1) | 0.137 |
| INR, m (IQR) | 1.00 (0.95–1.06) | 1.01 (0.95–1.06) | 1.00 (0.94–1.07) | 0.137 |
| Fibrinogen, m (IQR), mg/dl | 3.22 (2.59–3.91) | 3.26 (2.59–2.74) | 3.13 (2.55–4.24) | 0.429 |
| Coagulation disorder, no. (%) | 64 (5.1%) | 54 (6.8%) | 10 | <0.001┼ |
| GCS score at admission, m (IQR) | 10 (6–13) | 8 (5–12) | 12 (9–15) | <0.001┼ |
| ICH score, m (IQR) | 2 (2–3) | 2 (2–3) | 1 (1–2) | <0.001┼ |
| Surgical methods, no. (%) | NE | |||
| Craniotomy/craniotomy + craniectomy | 362 (45.6%) | 362 (45.6%) | N/A | |
| Minimal invasive surgery | 432 (54.4%) | 432 (54.4%) | N/A | |
the difference is significant.
APTT, activated partial thromboplastin time; DAPT, dual antiplatelet therapy; GCS, Glasgow coma score; ICCD, ischemic cerebrovascular and cardiovascular disease; ICH, intracerebral hemorrhage; INR, international normalized ratio; IVH, intraventricular hemorrhage; N/A, not applicate; NE, not evaluation; SSICH, severe spontaneous intracerebral hemorrhage.
Of 1260 SSICH patients, 275 (21.8%) SSICH patients were identified with 180-day poor outcomes, including 184 mortalities. The proportion of patient mortality due to MACCE and pulmonary embolization increased on the 180th day compared to the 30-day condition (Supplemental Figure 2.B, Supplemental Digital Content 2, http://links.lww.com/JS9/B302). Based on the 180-day outcome, patients receiving surgical treatment had a lower incidence of poor outcome (18.5 vs. 27.5%, P<0.011) and mortality (12.4 vs. 18.2%, P<0.011), compared with patients receiving conservative treatment (Fig. 1A). For patients taking DAPT prior to hemorrhage, no significant difference in poor outcome risk was found between the two treatment groups (Fig. 1B, P=0.604).
Figure 1.
Surgical treatment could decrease the risk of death in SSICH patients. A. Histograms depicting the 180-day outcome for SSICH patients receiving surgical versus conservative treatment. B. Subgroup analysis focusing on antiplatelet therapy preceding the hemorrhage, comparing outcomes between SSICH patients undergoing surgical treatment versus those receiving conservative treatment. C. Relative risk evaluation for 180-day outcomes among SSICH patients receiving surgical versus conservative treatment. The orange arrows highlight subgroups that did not benefit from surgical treatment. DAPT, dual antiplatelet therapy; GCS, Glasgow coma score; RR, relative risk; SSICH, severe spontaneous intracerebral hemorrhage.
Subgroup analysis, after controlling for modified factors associated with posthemorrhage mortality, indicated that surgically-treated patients continued to display a lower incidence of poor outcomes, compared to those receiving conservative treatment, except for patients administered DAPT, and those with infratentorial hematomas, coagulation disorders, and renal dysfunction (Fig. 1C).
Factors related to poor outcome of SSICH patients after surgical treatment
Further investigation was conducted to identify the factors contributing to the poor outcome of SSICH patients receiving surgical treatment (Fig. 2A). Detailed information regarding SSICH patients with poor and good outcomes postsurgery is provided in Supplemental Table 1 (Supplemental Digital Content 7 http://links.lww.com/JS9/B307). Of the 794 surgically-treated patients, 147 (18.5%) patients had poor 180-day outcome. More patient with poor outcome had a history of ICCD (Fig. 2B, P=0.007) and renal dysfunction (Fig. 2C, P=0.038). Patients on DAPT therapy exhibited a higher risk of poor outcomes compared to those on single or without any antiplatelet therapy (Fig. 2D, P<0.001).
Figure 2.
Factors related to poor outcome of SSICH patients after surgical treatment. A. Depiction of the identification process distinguishing SSICH patients with poor (180-day mRS as 4–6) and good (180-day mRS as 1–3) outcomes. Patients with poor outcomes were deemed nonbeneficiaries of surgical treatment. B. Comparison of outcomes between patients with and without a history of ICCD. C. Comparison of outcomes between patients with and without renal dysfunction. D. Comparison of outcomes between patients receiving and not receiving DAPT before hemorrhage. E. The forest plot presents the result of univariate logistic analysis for poor outcome. DAPT, dual antiplatelet therapy; GCS, Glasgow coma score; ICCD, ischemic cerebrovascular and cardiovascular disease; mRS, modified Rankin score; SSICH, severe spontaneous intracerebral hemorrhage; SAPT, single antiplatelet therapy.
The univariate logistic analysis identified the risk factors associated with poor outcomes on the 180th day (Fig. 2D), revealing a history of ICCD (P=0.008), renal dysfunction (P=0.039), DAPT before hemorrhage (P<0.001), hematoma volume (P<0.001), IVH (P<0.001), coagulation disorder (P=0.013), and GCS at admission (P<0.001) as contributing factors. The subsequent multivariate logistic regression analysis, performed on the parameters significant in the univariate analysis (Table 2), confirmed that a history of ICCD (OR, 1.86; 95% CI: 1.15–3.01; P=0.013), renal dysfunction (OR, 1.66; 95% CI: 1.11–2.47; P=0.011), DAPT (OR, 4.34; 95% CI: 1.98–9.50; P<0.001), hematoma volume (OR, 1.11; 95% CI: 1.05–1.18; P<0.001), and GCS score at admission (OR, 0.86; 95% CI: 0.81–0.91; P<0.001) were risk factors related to poor outcome independently.
Table 2.
Multivariate logistic analysis for poor outcome after surgical treatmenta
| Crude model | Adjusted modelb | |||
|---|---|---|---|---|
| Parameters | ORs | P | ORs | P |
| GCS score at admission | 0.86 (0.81–0.91) | <0.001 | 0.86 (0.82–0.91) | <0.001 |
| Hematoma volume, ×10 | 1.11 (1.05–1.18) | <0.001 | 1.13 (1.06–1.21) | <0.001 |
| DAPT before hemorrhage | 4.34 (1.98–9.50) | <0.001 | 4.69 (2.12–10.40) | <0.001 |
| History of ICCD | 1.86 (1.15–3.01) | 0.011 | 1.82 (1.12–2.94) | 0.015 |
| Renal dysfunction | 1.66 (1.11–2.47) | 0.013 | 1.57 (1.05–2.36) | 0.030 |
| IVH | Omitted | |||
| Coagulation disorder | Omitted | |||
Multivariate logistic analysis was performed using the backward method.
The result was adjusted by age, sex, hematoma location, and surgical methods.
DAPT, dual antiplatelet therapy; GCS, Glasgow coma score; ICCD, ischemic cerebrovascular and cardiovascular disease; IVH, intraventricular hemorrhage.
A DT model to help treatment decision-making for SSICH patients
Seven hundred ninety-four SSICH patients receiving surgical treatment were set as the derivation cohort (496 patients enrolling from a referring hospital) and the validation cohort (298 patients enrolling from multiple hospitals) (Fig. 3A).
Figure 3.
A decision tree model to predict the outcome of SSICH patients after surgical treatment. A. Illustration of the establishment and validation of the DT model for predicting outcomes of SSICH patients post-surgical treatment. The DT model was built based on a derivation cohort (SSICH patients from a referral hospital), and its performance was validated using a validation cohort (SSICH patients from multiple hospitals). B. The DT model for the outcome of SSICH patients after surgical treatment. The derivation cohort was split into the training set (80%) and testing set (20%). C. Evaluation of the DT model’s performance in classifying SSICH patients with poor and good outcomes within the training and testing set. D. Evaluation of the DT model’s performance in classifying SSICH patients with poor and good outcomes within the derivation and validation cohorts. E. The AUC values of model in predicting outcomes of SSICH patients after surgical treatment. SSICH, severe spontaneous intracerebral hemorrhage; DT, decision tree; AUC, area under curve; mRS, modified Rankin score; ICCD, ischemic cerebrovascular and cardiovascular disease; DAPT, dual antiplatelet therapy; GCS, Glasgow coma score.
Characteristics of SSICH patients in the derivation cohort is given in Supplemental Table 2 (Supplemental Digital Content 8, http://links.lww.com/JS9/B308). This cohort was then randomly separated as the training set (80%) and testing set (20%). After being adjusted by the SMOTE21, the model was acceptable without overfitting issue (Supplemental Figure 3.A, Supplemental Digital Content 3, http://links.lww.com/JS9/B303). The process of feature selection, based on the training set, identified several key indicators of poor outcomes: the GCS score at admission, hematoma volume, DAPT before hemorrhage, history of ICCD, and renal dysfunction (Supplemental Figure 3.B, Supplemental Digital Content 3, http://links.lww.com/JS9/B303). Incorporating above features of poor outcome, a DT model was trained (Fig. 3B). The DT model performed well to classify patients with poor outcome and good outcome both the training set (AUC, 0.95) and testing set (AUC, 0.90) (Fig. 3C). After controlling the effect of age, hematoma location, IVH, coagulation disorder and surgical methods, the model remained robust in poor outcome evaluation (Supplemental Figure 4, Supplemental Digital Content 4, http://links.lww.com/JS9/B304).
Patient characteristics in the validation cohort are detailed in Supplemental Table 3 (Supplemental Digital Content 9, http://links.lww.com/JS9/B309). Within the validation cohort, the DT model categorized 56 patients as the high-risk and 242 as the low-risk group (Fig. 3D). The model also showed high efficacy in predicting patients with poor outcomes within the validation cohort (AUC, 0.92). Subsequent analysis indicated that the DT model surpassed the ICH score in accurately assessing the risk of poor outcomes in every dataset (all P<0.001, Fig. 3E and Supplemental Figure 3.C–E, Supplemental Digital Content 3, http://links.lww.com/JS9/B303). After controlling the effect of age, hematoma location, IVH, coagulation disorder and surgical methods, subgroup analysis demonstrated that the DT model continued to reliably predict poor outcome (Supplemental Figure 5, Supplemental Digital Content 5, http://links.lww.com/JS9/B305). A summary of the DT model’s performance in distinguishing between poor and good outcomes after surgical treatment is presented in Supplemental Table 4 (Supplemental Digital Content 10, http://links.lww.com/JS9/B310).
Influence of DT model on treatment decisions for SSICH
We further investigated whether the DT model could help treatment decision-making for SSICH patients (Fig. 4A). Utilizing the DT model, the accuracy (0.92 vs. 0.88, P<0.001) and AUC value (0.81 vs. 0.89, P<0.001) of clinicians to identify patients with poor outcome postsurgery was improved (Supplemental Figure 6.A, Supplemental Digital Content 6, http://links.lww.com/JS9/B306). Specifically, junior clinicians’ performance in identifying SSICH patients with poor outcomes was improved, with increased accuracy (0.84 vs. 0.91, P<0.001) and AUC value (0.75 vs. 0.88, P<0.001) when using the DT model; moreover, with the help of DT model, the performance of junior clinicians to classify patients with poor outcome and good outcome was not inferior to senior clinicians (AUC, 0.88 vs. 0.92, P=0.170 (Fig. 4B-C, also seen in Supplemental Figure 6.B, Supplemental Digital Content 6, http://links.lww.com/JS9/B306). However, for senior clinicians, the accuracy (0.94 vs. 0.94, P=0.902) and AUC (0.92 vs. 0.92, P=0.781) to identify patients with poor outcome were not improved significantly with the usage of the DT model (Fig. 4B-C and Supplemental Figure 6.C, Supplemental Digital Content 6, http://links.lww.com/JS9/B306).
Figure 4.
The DT model could help treatment decision-making for SSICH patients. A. Depiction of the assessment to determine if the DT model could assist in treatment decisions for SSICH patients. Four junior and two senior clinicians were tasked with identifying whether SSICH patients could benefit from surgical treatment (with poor outcomes defined as a 180-day mRS score of 4-6), with and without the assistance of the DT model. B. The accuracy of clinicians in classifying SSICH patients with good outcome and with poor outcome, with and without the help of DT model. C. The AUC values of clinicians in classifying SSICH patients with poor outcome and good outcome, with and without the help of DT model. AUC, area under curve; DT, decision tree; SSICH, severe spontaneous intracerebral hemorrhage.
Subsequent analysis using NRI (0.26, P<0.001) and IDI (0.20, P<0.001) confirmed that the DT model substantially improved clinicians’ ability to identify SSCIH patients likely to have poor outcomes postsurgery (Supplemental Table 5, Supplemental Digital Content 11, http://links.lww.com/JS9/B311).
Discussion
Surgical treatment is an efficacious method to mitigate the risk of mortality or unfavorable outcomes in SSICH patients by rapidly alleviating intracranial pressure. Our study highlights GCS score at admission, hematoma volume, DAPT before hemorrhage, history of ICCD and renal dysfunction were factors related to poor outcome of SSICH patients receiving surgical treatment. The DT model incorporating above factors could identify SSICH patients at high-risk of poor outcome after surgical treatment, which could evaluate whether SSICH patients can benefit from surgical treatment. Further analysis showed that the use of the DT significantly improved the performance of junior clinicians to identify patients with poor outcome. To the best of our knowledge, this study constitutes the largest cohort analysis of SSICH patients within the Chinese population, providing a practical tool to aid clinicians in assessing the potential benefit of surgical treatment for SSICH patients.
Our findings underscore that surgical treatment can safeguard SSICH patients against unfavorable outcomes. However, for patients with prehemorrhage DAPT use, renal dysfunction, coagulation disorders, and infratentorial hematoma, surgical treatment did not appear to significantly improve outcomes. DAPT use and coagulation disorders can impede patient hemostasis, potentially leading to bleeding complications both during and after surgery. An infratentorial hematoma can precipitate a rapid surge in intracranial pressure, causing severe brainstem compression and cerebellar tonsillar herniation within a brief span, particularly within the confined space of the posterior fossa23–25. Previous research has also established that renal dysfunction augments the risk of acute cerebrovascular events due to various underlying pathologies, including large artery stroke and rebleeding26–28. Therefore, SSICH patients with these factors may not benefit from surgical treatment. Additional analyses of surgically-managed patients disclosed that apart from the above factors, the GCS score at admission, hematoma volume, and history of ICCD were also related to poot outcome of SSICH patients postsurgically27,29. The GCS score at admission could signify the progression and severity of SSICH, while hematoma volume is related to SSICH severity. These two features were previously demonstrated as predictors of poor outcomes4,6,7. Patients with a history of ICCD are known to be at heightened risk of MACCE2. Our data indicates that MACCE represented an increasing cause of mortality on the 180th day posthemorrhage, thereby establishing a previous ICCD history as a predictor of poor outcomes. Notably, our study discovered that infratentorial hematoma, IVH and coagulation disorder did not emerge as risk factors associated with poor outcomes in further analysis. This could potentially be attributed to the limited number of patients with infratentorial hematoma or coagulation disorders, and the fact that surgery may also help alleviate IVH.
It is acknowledged that not all SSICH patients can benefit from surgical treatment. Acknowledging this reality, we developed a DT incorporating the above prognostic factors. This model had well performance in discerning which SSICH patients could potentially benefit from surgical treatment. Our results indicated that this DT model can improve the ability of junior clinicians to identify patients likely to experience poor postoperative outcomes. However, its efficacy in assisting senior clinicians was less profound. It could deepen junior clinicians’ understanding of the clinical outcome of SSICH and help them determine which patients are likely to benefit from surgical intervention, even in the face of limited SSICH cases and clinical training. For patients identified as low-risk for poor outcome postsurgery, surgical treatment should be recommended to mitigate the risk of unfavorable outcomes. Conversely, for patients at high-risk of poor outcome, the economic and physical burden of the surgical treatment may outweigh its advantages, therefore a conservative management approach may be more appropriate. Conclusively, this DT model could aid junior clinicians in identifying which SSICH patients might benefit from surgical treatment, thereby assisting in treatment decision-making for SSICH.
Despite the insightful findings, several limitations inherent to the current study must be acknowledged. First, the study cohort was exclusively Chinese, which might limit the generalizability of the results to other population demographics. It is recognized that the Chinese population has a higher incidence of MACCE30, thus introducing a potential population bias. Secondly, the study was susceptible to a multicenter bias due to the involvement of multiple research centers, potentially introducing variations in study protocols and patient characteristics. Thirdly, the relatively small cohort size implies that certain factors may not have revealed significant differences due to insufficient statistical power. Fourthly, our primary focus was on identifying patients who might benefit from surgery; therefore, we did not evaluate the specific impact of different surgical procedures on patient outcomes. This is a subject of ongoing research, and we plan to address it in forthcoming publications. Finally, the potential exists for unaccounted factors not incorporated in the analysis that could influence study outcomes.
Conclusion
The current study demonstrated GCS score at admission, hematoma volume, DAPT before hemorrhage, history of ICCD and renal dysfunction as factors related to poor outcome following surgical treatment for SSICH. This study provided a DT model for evaluating the poor outcome of SSICH patients after surgery, which is an applicable tool to help clinicians in treatment decision-making for SSICH. Larger-scale validation of this scale is needed to demonstrate further clinical utility.
Ethical approval
The study was approved by the institutional review board of Beijing Tiantan hospital (KY2019-096-02). Written informed consents were obtained and the privacy of patients was effectively protected.
Consent
Written informed consent was obtained from the patient for publication and any accompanying images. A copy of the written consent is available for review by the Editor-in-Chief of this journal on request.
Sources of funding
This study was supported by the ʽWuxi Taihu Lake Talent Plan, Team in Medical and Health Profession (Grant No. TH202109)ʼ, ʽNational Key Research and Development Program of the 14th Five-Year Plan (Grant No. 2021YFC2501100)ʼ, ʽNational Natural Science Foundation of China (Grant No. 82071296, 81801158, 81671129 and 81471210)ʼ, and ʽTop Talent Support Program for young and middle-aged people of Wuxi Health Committee (Grant No.202014)ʼ.
Author contribution
K.W., Q.L., and S.M.: conceptualization, methodology, project administration, writing original draft and review and editing; K.Z., X.L., and J.L.: data collection and curation; S.C., X.T., and Y.C.: data analysis and interpretation; Z.L.: study concept and design, critically revising the paper; J.W.: concept and design, critically revising the paper; S.W.: concept and design, critically revising the paper, study supervision. All authors confirm that they contributed to manuscript reviews and critical revision for important intellectual content, and read and approved the final draft for submission. All authors agree to be accountable for the content of this study.
Conflicts of interest disclosure
The authors stated that they had no conflict of interest.
Research registration unique identifying number (UIN)
Name of the registry: Effect and safety of surgical intervention for severe spontaneous intracerebral hemorrhage patients on long-term oral antiplatelet treatment.
Unique identifying number or registration ID: Chinese Clinical Trial Registry, ChiCTR1900024406.
Hyperlink to your specific registration (must be publicly accessible and will be checked): https://www.chictr.org.cn/showproj.html?proj=40640.
Guarantor
Shuo Wang, MD, Department of Neurosurgery, Beijing Tiantan Hospital, Capital Medical University, Beijing, People’s Republic of China. E-mail: captain9858@126.com.
Data availability statement
The data supporting the findings of this study are available from the corresponding authors upon reasonable request.
Provenance and peer review
This paper was not invited.
Supplementary Material
Acknowledgements
The authors thank the participants of the study, and all clinical and research staff who contributed to this work.
Footnotes
Kaiwen Wang, Qingyuan Liu, and Shaohua Mo contributed equally to this work.
Sponsorships or competing interests that may be relevant to content are disclosed at the end of this article.
Published online 7 November 2023
Supplemental Digital Content is available for this article. Direct URL citations are provided in the HTML and PDF versions of this article on the journal’s website, www.lww.com/international-journal-of-surgery.
Contributor Information
Kaiwen Wang, Email: ccmukevin@163.com.
Qingyuan Liu, Email: 13260457220@163.com.
Shaohua Mo, Email: shmo2018@outlook.com.
Kaige Zheng, Email: zhengkg109@126.com.
Xiong Li, Email: lili9579@sina.com.
Jiangan Li, Email: lijiangan2014@163.com.
Shanwen Chen, Email: csw_1981_1999@126.com.
Xianzeng Tong, Email: noahtong@163.com.
Yong Cao, Email: Caoyong6@hotmail.com.
Zhi Li, Email: lzsjwk@163.com.
Jun Wu, Email: wujunslf@126.com.
Shuo Wang, Email: captain9858@126.com.
References
- 1. Greenberg SM, Ziai WC, Cordonnier C, et al. 2022 guideline for the management of patients with spontaneous intracerebral hemorrhage: a guideline from the American Heart Association/American stroke association. Stroke 2022;53:e282–ee361. [DOI] [PubMed] [Google Scholar]
- 2. Sheth KN. Spontaneous intracerebral hemorrhage. N Engl J Med 2022;387:1589–1596. [DOI] [PubMed] [Google Scholar]
- 3. Shen J, Shao X, Ge R, et al. Risk factors for postoperative rebleeding and short-term prognosis of spontaneous cerebellar hemorrhage. RMHP 2021;14:2045–2053. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4. Casolla B, Moulin S, Kyheng M, et al. Five-year risk of major ischemic and hemorrhagic events after intracerebral hemorrhage. Stroke 2019;50:1100–1107. [DOI] [PubMed] [Google Scholar]
- 5. Smilowitz NR, Gupta N, Ramakrishna H, et al. Perioperative major adverse cardiovascular and cerebrovascular events associated with noncardiac surgery. JAMA Cardiol 2017;2:181. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6. Murthy SB, Diaz I, Wu X, et al. Risk of arterial ischemic events after intracerebral hemorrhage. Stroke 2020;51:137–142. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7. Murthy SB, Zhang C, Diaz I, et al. Association between intracerebral hemorrhage and subsequent arterial ischemic events in participants from 4 population-based cohort studies. JAMA Neurol 2021;78:809. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8. Qureshi AI, Mendelow AD, Hanley DF. Intracerebral haemorrhage. Lancet 2009;373:1632–1644. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9. Wu J, Sun X, Liu Q, et al. Surgical treatment for antiplatelet intracerebral hemorrhage (SAP-ICH): protocol for a prospective cohort study of emergency surgery for severe spontaneous intracerebral hemorrhage patients on long-term oral antiplatelet treatment. Chin Neurosurg J. 2021;7:5. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10. Mathew G, Agha R, Albrecht J, et al. STROCSS 2021: strengthening the reporting of cohort, cross-sectional and case-control studies in surgery. Int J Surg 2021;96:106165. [DOI] [PubMed] [Google Scholar]
- 11. Rankin J. Cerebral vascular accidents in patients over the age of 60: ii. prognosis. Scott Med J 1957;2:200–215. [DOI] [PubMed] [Google Scholar]
- 12. Wang A, Xie X, Tian X, et al. Ticagrelor–aspirin versus clopidogrel–aspirin among CYP2C19 loss-of-function carriers with minor stroke or transient ischemic attack in relation to renal function: a post hoc analysis of the CHANCE-2 trial. Ann Intern Med 2022;175:1534–1542. [DOI] [PubMed] [Google Scholar]
- 13. Levey AS, Coresh J. Chronic kidney disease. Lancet 2012;379:165–180. [DOI] [PubMed] [Google Scholar]
- 14. Tonelli M, Wanner C. Lipid management in chronic kidney disease: synopsis of the kidney disease: improving global outcomes 2013 clinical practice guideline. Ann Intern Med 2014;160:182–189. [DOI] [PubMed] [Google Scholar]
- 15. Awtry EH, Loscalzo J. Aspirin. Circulation 2000;101:1206–1218. [DOI] [PubMed] [Google Scholar]
- 16. Wu J, Liu Q, Wang K, et al. Emergency surgery is an effective way to improve the outcome of severe spontaneous intracerebral hemorrhage patients on long-term oral antiplatelet therapy. Neurosurg Rev 2020;44:1205–1216. [DOI] [PubMed] [Google Scholar]
- 17. Liu Q, Li X, Wang N, et al. A novel coagulation classification and postoperative bleeding in severe spontaneous intracerebral hemorrhage patients on antiplatelet therapy. Front Aging Neurosci 2022;14:793129. doi: 10.3389/fnagi.2022.793129 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18. Cochrane J. Alcohol use in CHINA. Alcohol Alcohol 2003;38:537–542. [DOI] [PubMed] [Google Scholar]
- 19. Can A, Castro VM, Ozdemir YH, et al. Association of intracranial aneurysm rupture with smoking duration, intensity, and cessation. Neurology 2017;89:1408–1415. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20. Maegele M, Schöchl H, Menovsky T, et al. Coagulopathy and haemorrhagic progression in traumatic brain injury: advances in mechanisms, diagnosis, and management. Lancet Neurol 2017;16:630–647. [DOI] [PubMed] [Google Scholar]
- 21. Nakamura M, Kajiwara Y, Otsuka A, et al. LVQ-SMOTE – learning vector quantization based synthetic minority over–sampling technique for biomedical data. BioData Min 2013;6:16. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 22. Alba AC, Agoritsas T, Walsh M, et al. Discrimination and calibration of clinical prediction models. JAMA 2017;318:1377. [DOI] [PubMed] [Google Scholar]
- 23. Mayer SA, Sacco RL, Shi T, et al. Neurologic deterioration in noncomatose patients with supratentorial intracerebral hemorrhage. Neurology 1994;44:1379. [DOI] [PubMed] [Google Scholar]
- 24. Kalita J, Misra UK, Vajpeyee A, et al. Brain herniations in patients with intracerebral hemorrhage. Acta Neurol Scand 2009;119:254–260. [DOI] [PubMed] [Google Scholar]
- 25. Wilkinson DA, Pandey AS, Thompson BG, et al. Injury mechanisms in acute intracerebral hemorrhage. Neuropharmacology 2018;134:240–248. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 26. Currie CJ, Berni ER, Berni TR, et al. Major adverse cardiovascular events in people with chronic kidney disease in relation to disease severity and diabetes status. PLoS ONE 2019;14:e0221044. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 27. Woo D, Comeau ME, Venema SU, et al. Risk factors associated with mortality and neurologic disability after intracerebral hemorrhage in a racially and ethnically diverse cohort. JAMA Netw Open 2022;5:e221103. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 28. Kelly DM, Ademi Z, Doehner W, et al. Chronic kidney disease and cerebrovascular disease. Stroke 2021;52:e328–ee346. [DOI] [PubMed] [Google Scholar]
- 29. Shah VA, Thompson RE, Yenokyan G, et al. One-year outcome trajectories and factors associated with functional recovery among survivors of intracerebral and intraventricular hemorrhage with initial severe disability. JAMA Neurol 2022;79:856. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 30. Zhou M, Wang H, Zeng X, et al. Mortality, morbidity, and risk factors in China and its provinces, 1990–2017: a systematic analysis for the Global Burden of Disease Study 2017. Lancet 2019;394:1145–1158. [DOI] [PMC free article] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Data Availability Statement
The data supporting the findings of this study are available from the corresponding authors upon reasonable request.