A novel clinical nomogram for predicting 3-month unfavorable outcomes of stroke patients treated with mechanical thrombectomy

Xiao-Guang Zhang Tongji University A lliated Yangpu Hospital: Shanghai Yangpu District Central Hospital Jia-Hui Wang Tongji University A lliated Yangpu Hospital: Shanghai Yangpu District Central Hospital Wen-Hao Yang Tongji University A lliated Yangpu Hospital: Shanghai Yangpu District Central Hospital Xiao-Qiong Zhu Tongji University A lliated Yangpu Hospital: Shanghai Yangpu District Central Hospital Jie Xue Tongji University A lliated Yangpu Hospital: Shanghai Yangpu District Central Hospital Zhi-Zhang Li Tongji University A lliated Yangpu Hospital: Shanghai Yangpu District Central Hospital Yu-Ming Kong Tongji University A lliated Yangpu Hospital: Shanghai Yangpu District Central Hospital Liang Hu Tongji University A lliated Yangpu Hospital: Shanghai Yangpu District Central Hospital Shan-Shan Jiang Tongji University A lliated Yangpu Hospital: Shanghai Yangpu District Central Hospital Xu-Shen Xu Tongji University A lliated Yangpu Hospital: Shanghai Yangpu District Central Hospital yun-hua yue (  Yunhua.Yue@tongji.edu.cn ) Tongji University A lliated Yangpu Hospital: Shanghai Yangpu District Central Hospital https://orcid.org/0000-0002-5244-1954


Background
Acute ischemic stroke (AIS) is a common cerebrovascular event that can easily cause disability [1]. Mechanical thrombectomy (MT) is essential for AIS with large-vessel occlusion, of which blood perfusion can rarely be restored by intravenous thrombolysis [2]. With the popularization of MT, more and more patients have successfully achieved recanalization. However, a successful recanalization of the occluded artery does not always lead to a good outcome (90-day modi ed Rankin Scale ≥ 2) [3]. Therefore, studies are still needed to predict the potential factors affecting the prognosis of thrombectomy.
Nomogram is a computational tool that can generate numerical probabilities of clinical events based on each patient's personal pro le by creating graphical representation of statistical prediction models [4], which has demonstrated signi cant advantages over traditional staging systems and has been used to predict many diseases, such as pneumonia, myocardial infarction, and cancer [5][6][7]. In recent years, several prognostic models based on clinical and radiological variables have been used to predict prognosis after thrombectomy for stroke [3,8,9]. The patients included in the studies mainly had symptoms of acute anterior circulation stroke within 6 h of onset, representing the largest population of stroke candidates for thrombectomy according to the 2018 guideline for early management of patients with acute ischemic stroke [10]. With the update of the guideline about endovascular thrombectomy, especially the extension of the time windows and the generation of new devices, more patients with indications can bene t from thrombectomy. Therefore, new nomograms for AIS patients undergoing MT are urgently needed.
The present study aimed to develop and validate a nomogram to predict 3-month unfavorable functional outcomes in Chinese AIS patients undergoing MT by using a limited number of easily available variables.

Study Population
We conducted a retrospective study based on data collected from January 2018 to October 2020 from AIS patients who were treated by MT therapy in the Department of Stroke of the Yangpu Hospital, Tongji University School of Medicine. Participants in this study was reviewed and approved by ethics committee of Yangpu Hospital (ethical approval number LL-2018-SCI-004). Informed consent has been obtained. Patients were included in this study if all the following conditions were met: (1) aged ≥ 18 years; (2) prestroke modi ed Rankin Scale score (mRS) of 0 to 1; (3) ASPECTS of ≥ 6; (4) with occlusion of the internal carotid artery, anterior cerebral artery, middle cerebral artery, or vertebrobasilar artery con rmed by digital subtraction angiography; (5) undergoing MT treatment within 24 hours of the stroke onset. And patients will receive antithrombotic drugs following the MT treatment. Patients with haemorrhage on baseline or within 24 hours after MT, lost to follow-up without 3-month mRS score, baseline or laboratory data and pre-/post-operative imaging examination were unavailable were excluded.
Demographic characteristics, vascular risk factors, clinical data, radiological images, and laboratory data were all collected. These include age, gender, baseline NIHSS score, baseline ASPECTS score, anesthesia, vascular risk factors such as hypertension, diabetes mellitus, coronary heart disease, atrial brillation, smoking and alcoholism, clinical data such as systolic/ diastolic blood pressure, time from onset to puncture, time from puncture to recanalization, bridging therapy, number of retrieval attempts, and strokeassociated pneumonia (SAP), radiological images such as TOAST classi cation, vascular occlusion site, postoperative modi ed Thrombolysis in Cerebral Infraction (mTICI) score and symptomatic intracranial hemorrhage (sICH), preoperative laboratory data such as C-reactive protein(CRP), red blood cell distribution width (RDW) , hemoglobin (HGB), glucose, Na concentration, creatinine and B-natriuretic peptide (BNP). The primary outcome was the unfavorable functional consequences at 3 months after stroke, with mRS between 3 and 6 (i.e., poor prognosis). The 3-month mRS were evaluated by trained physicians with telephone questionnaires or face-to-face interviews.

Statistical analysis
Continue variables were described as median (interquartile intervals 25-75), and univariate comparisons were performed using Mann-Whitney U where appropriate. Categorical variables were expressed as frequencies and percentages, and differences were assessed by the chi-square test or Fisher's exact test.
The glmnet package in R was used for the least absolute shrinkage and selection operator (LASSO) logistic regression, and signi cant factors there in were used to enter into the logistic regression model.
Using forward stepwise selection with Akaike's information criterion (AIC) as the stopping rule, variables included in the nal multivariable logistic regression model were carefully chosen. The AIC value for the nal model was minimized with the fewest number of variables. The Variation In ation Factors (VIF, < 10 indicating no multicollinearity problem) was used to evaluate the collinearity of combination variables entered into the multivariable logistic regression analysis. The odds ratio (OR) and 95% con dence interval (95% CI) were calculated for variables signi cantly associated with the main outcome in the multivariable analysis. Factors remained signi cant in the multivariable analysis were incorporated in the nomogram.
Based on discrimination and calibration, the discriminative power of the model was assessed. The discrimination of the nomogram model was validated using the receiver operating characteristic (ROC) curve and quanti ed by the area under the curve (AUC), while the calibration was visually assessed using the calibration plot and the Spiegelhalter's Z-test. A 45° diagonal line indicates perfect calibration when the predictive value of a well-calibrated model perfectly matches the patient's actual risk. Nomogram construction and calibration was assessed graphically by means of the R package rms.
The statistical analysis was carried out using Stata version 15.0 (Stata Corporation, College Station, TX, USA) statistical software and the statistical software package R version 3.6.0. All tests were two-sided and P < 0.05 was considered statistically signi cant.

Results
Two hundred and thirty-eight AIS patients were nally enrolled in the entire study (mean age, 73.05±12.89 years), 202 patients were treated within 6 hours of symptom onset (Time from onset to groin puncture) and 36 patients treated between 6 and 24 hours after onset of symptom onset. The proportion of patients with poor 3-month pronosis (mRS score 3-6) was 59.7% (142/238) and 21.4% (51/238) of patients died during the follow-up period (mRS score = 6). Among them, 35 died of vascular death, 2 died of recurrent ischemic stroke, and 14 died of Non-vascular death. (Table 1) The clinical, demographic and laboratory characteristics of the patients in the favorable outcome cohorts (n = 96) and unfavorable outcome (n = 142) cohorts are shown in Table 1 The LASSO logistic regression algorithm were used to select the most signi cant predictive factors, which were then used to construct the prediction model. A total of 30 factors shown in Table 1 were used for the LASSO logistic regression, and 11 factors with non-zero coe cients were subsequently selected, with an optimal lambda value of 0.015 (Fig. 1A A nomogram was created by assigning a graphical preliminary score ranging from 0 to 100 to each of the six predictors, then summing them to generate a total score, which was then converted into an individual probability of 3-month unfavorable outcome (from 10% to 90%) (Fig. 2). It was predicted that a higher total score of the nomogram was associated with a higher likelihood of unfavorable outcome, while a lower total score was associated with a lower likelihood of an adverse outcome. In this cohort, the nomogram had an area under the curve (AUC) of 0.848 (95% CI: 0.80-0.90) (Fig. 3). The p-value for Spiegelhalter's Z-test was 0.946, which indicated that there was no signi cant departure from a perfect t.
The Brier was 0.156, which meant that the calibration curve was almost close to the ideal curve. The calibration plot revealed the adequate t of the model, predicting the risk of the poor prognosis at 3 months after MT (Fig. 4).

Discussion
Major ndings For precise clinical and therapeutic management of AIS patients, early prediction of unfavorable outcome is undoubtedly a valuable perspective. In this study, we developed a nomogram model to predict the unfavorable outcome after MT in AIS with large-vessel occlusion. Our nomogram provides a quantitative prediction of a 10% to 90% probability of 3-month unfavorable outcome in Chinese AIS patients undergoing MT, and it exhibits an excellent predictive power with a C-index of 0.848. Gender, bridging therapy, postoperative mTICI, SAP, preoperative creatinine and Na are independently associated with poor prognosis at 3-months.
Although 88.24% of the patients (210/238) bene ted from the progressive development of MT intervention materials and prolonged treatment time window to achieve postoperative mTICI (≥2b), only 43.81% of the patients (92/210) bene ted from the reperfusion. At present, only limited data are available for clinicians to determine the risk of poor prognosis after MT in AIS with large-vessel occlusion. Due to the paucity of data, this nomogram was developed speci cally for patients undergoing MT within 24 h from onset to treatment since 2018. This nomogram is based on six basic clinical data, three preoperative data (gender, creatinine and Na), one intraoperative data (bridging therapy), and two postoperative data (mTICI, SAP), which demonstrated an excellent discrimination in terms of internal validation.

Nomogram variables interpretation
In our study, by using a combination of six easily available predictors before and after endovascular procedure, the area under the curve (AUC) of our nomogram could reach 0.848, which means it may be a novel and reliable graphical computational tool that can provide guidance for AIS patient management after revascularization treatments. Gender differences is also one of the factors affecting the prognosis of stroke. Previous studies reported that females consistently fare worse than males following MT for large vessel ischemic stroke [11,12]. One of the possible reasons is that males have larger cerebral arterial diameters than females, which may improve the odds for favorable clinical outcomes [13]. Serum creatinine is widely used as an indicator of renal function, and impaired renal function is also associated with long-term mortality and poor prognosis after stroke [14]. The present study is consistent with previous ndings that the worse the renal function, the worse the prognosis. Sodium ions play an important role in maintaining intracellular and extracellular electrolyte gradients, nerve conduction velocity, and muscle excitability [15]. Mild hyponatremia is known to be an independent prognostic factor for mortality in AIS patients [16]. Although the average preoperative Na concentration in the poor prognosis group has not reached hyponatremia, but was lower than that in the good prognosis group, indicating that it could affect the prognosis to some extent. The lower the preoperative Na concentration, the greater the risk of unfavorable outcomes. For patients with acute ischemic stroke due to anterior circulation large-vessel occlusion, they are eligible for both intravenous alteplase and MT according to most guidelines. Although current DIRECT-MT trial have shown that direct intervention with MT was noninferior to the combination of intravenous thrombolysis followed by MT [17], the trial was still limited to the anterior circulation and time window was 4.5 hours. However, after extending the treatment time window, we found that the bene t of bridging therapy was still greater than that of direct thrombectomy. Successful recanalization after MT is the strongest adjustable predictor of patient prognosis, and higher recanalization rates are associated with functional improvement at 3 months [18]. Successful recanalization of large-vessel occlusions is traditionally de ned by mTICI grades of 2b or 3. Our study also con rmed that the patient prognosis of mTICI greater than 2b was better. Moreover, SAP is another strong predictor of poor outcome in AIS patients undergoing MT. Aspiration caused by dysphagia, reduced consciousness, impaired bulbar re exes, and hypokinesia are more likely to cause poststroke cumulus pneumonia in AIS patients with MT [19]. For patients with acute stroke, the occurrence of pneumonia has been shown to be associated with poor functional-reelated outcomes and increased risk of death [20]. Therefore, our nomogram covers preoperative, intraoperative and postoperative variables and may be more practical in some respects.
Comparison with prior studies Previously, several models have been developed to predict the probability of 3-month mortality in AIS patients receiving MT [3,8,9]. Among them, an Italian cohort study developed a nomogram that included NIHSS score, age, pre-stroke mRS score, bridging therapy or direct thrombectomy, grade of recanalization according to the mTICI grading system, and onset-to-end procedure time [8]. While two Chinese studies identi ed many demographic and clinical characteristics, including baseline NIHSS score, age, creatinine, blood glucose level, collateral status, and sICH [3,9]. However, due to the inconsistent inclusion and exclusion criteria, the potential predictors nally included in the nomogram in their studies are still quite different. The patients in Cappellari's study had anterior and posterior large-vessel occlusion who underwent thrombectomy within 6 h of stroke onset [8]. With the update of the guidelines on thrombectomy, the time window for thrombectomy was expanded, and the number of patients requiring endovascular treatment has increased accordingly. The 2018 Guidelines for the Early Management of Acute Ischemic Stroke extended the time window from 6 hours to 24 hours for patients with large-vessel occlusion in the anterior circulation and meeting DAWN or DEFUSE 3 eligibility criteria [10]. So it is currently necessary to update the nomogram to predict the probability of unfavorable outcome in AIS patients who are candidates for thrombectomy. Zhang et al. developed a nomogram that predicted the likelihood of 3-month mortality in patients with anterior circulation stroke who had a successful EVT within 6 hours of symptom onset [3]. Patients with unsuccessful recanalization, non-anterior circulation stroke, exceeding 6 hours of last known normal values were excluded from their study. Based on the fact that unsuccessful recanalization is often accompanied with poor functional outcomes, many patients with poor prognosis may be excluded. While our study assessed AIS patients undergoing MT treatment within 24 hours of the stroke onset with both anterior and posterior circulation infarction.

Nomogram clinical consideration
We could easily obtain the most signi cant variables in the nomogram for each patient, indicating that this nomogram is highly practical as a predictive tool. Unlike previous prognostic models [21,22], our nomogram assigns a probability (from 10 to 90%) of unfavorable outcome. For example, our nomogram assigns > 90% probability of adverse consequence when a male patient (0 point), with no bridging therapy (18.75 points), postoperative mTICI 2b is 22.5 points, SAP is 27.5 points, preoperative creatinine concentration is 160mmol/l (53.75 points) and Na concentration is 130mmol/l (57.5 points), with a total score of 180 points. On the other hand, a male patient (0 point), with bridging therapy (0 points), postoperative mTICI ≥2b (0 point), no SAP (0 points), preoperative creatinine concentration is 60mmol/l (15.5 points) and Na concentration is 144mmol/l (12.5 points), with a total score of 28 points, produce a probability of adverse outcome of < 10%.
By converting the total score into a continuum of individual probabilities, our nomogram can more precisely reclassify the risk of 3-month adverse outcome. Thus, the nomogram may better identify different risk predictors in AIS patients undergoing MT than previous prognostic models based on inadequate risk-grouping categorization [23].

Limitation
There are several limitations in this study. First, the data we collected were from a single center, so the sample size was not large enough, which might limit the generalizability of our ndings. Further expansion of the sample size is needed to improve the nomogram. Second, biomarkers, such as brain edema and infarct size, were not included in the model because these parameters were not available in the current imaging system. Our future prospective study will include more parameters and determine whether the integration of other biomarkers may help to improve the accuracy of the nomogram. Third, this study is based on a retrospective analysis of prospectively collected data. Despite the good discriminative performance of this model, external validation in completely different patient cohorts is still required.    Figure 1 Factor selection using the least absolute shrinkage and selection operator (LASSO) logistic regression.

Conclusions
(A) The LASSO coe cient pro les of the 30 candidate variables. The binomial deviance is plotted versus log (λ). (B) Tuning parameter (λ) selection in the LASSO logistic regression performed using 10-fold cross-validation via the minimum criteria. A coe cient pro le plot is produced versus the log (λ). The left vertical line represents the minimum error, and the right vertical line represents the cross-validated error within 1 standard error of the minimum.

Figure 2
Nomogram for predicting the probability of 3-month unfavorable outcome in acute ischemic stroke patients undergoing mechanical thrombectomy based on gender, bridging therapy, postoperative mTICI, stroke-associated pneumonia, preoperative creatinine and Na.

Figure 3
Receiver operating characteristic (ROC) curve of the nomogram for predicting 3-month unfavorable outcomes of stroke patients treated with mechanical thrombectomy.

Figure 4
The calibration plot of the nomogram for predicting 3-month unfavorable outcomes of stroke patients treated with mechanical thrombectomy. The x-axis represents the predicted probability of unfavorable outcome calculated using the nomogram. The y-axis represents the actual rate of unfavorable outcome. The dashed line is the reference line where an ideal nomogram would lie. The dotted line is the performance of the nomogram, while the solid line corrects for any deviation of the nomogram.