Development and internal validation of a nomogram to predict massive blood transfusions in neurosurgical operations
How to cite this article: Sungkaro K, Taweesomboonyat C, Kaewborisutsaku A. Development and internal validation of a nomogram to predict massive blood transfusions in neurosurgical operations. J Neurosci Rural Pract, doi: 10.25259/JNRP-2022-2-31
A massive blood transfusion (MBT) is an unexpected event that may impact mortality. Neurosurgical operations are a major operation involving the vital structures and risk to bleeding. The aims of the present research were (1) to develop a nomogram to predict MBT and (2) to estimate the association between MBT and mortality in neurosurgical operations.
Material and Method:
We conducted a retrospective cohort study including 3660 patients who had undergone neurosurgical operations. Univariate and multivariate logistic regression analyses were used to test the association between clinical factors, pre-operative hematological laboratories, and MBT. A nomogram was developed based on the independent predictors.
The predictive model comprised five predictors as follows: Age group, traumatic brain injury, craniectomy operation, pre-operative hematocrit, and pre-operative international normalized ratio and the good calibration were observed in the predictive model. The concordance statistic index was 0.703. Therefore, the optimism-corrected c-index values of cross-validation and bootstrapping were 0.703 and 0.703, respectively.
MBT is an unexpectedly fatal event that should be considered for appropriate preparation blood components. Further, this nomogram can be implemented for allocation in limited-resource situations in the future.
Massive blood transfusion (MBT) occurs in several situations including for severely injured trauma victims, uncontrolled blood loss,[2,3] and intraoperative incidental injury to major vessels. The incidence of MBT ranges from 1.8% to 5.0%,[1-3] and the common situations that lead to MBT include trauma, cardiac surgery, liver transplantation, ruptured abdominal aortic aneurysm, gastrointestinal hemorrhage, and postpartum hemorrhage.[2,3] However, the association between MBT and mortality remains inconclusive. O’Keeffe et al. studied 8799 patients who had undergone revascularization of the lower extremity and found that MBT was significantly associated with both increased 30-day mortality and complications. However, the study of Rangarajan et al. reported that MBT was not associated with mortality in traumatic patients. In addition, Reppucci et al. reported that MBT did not impact mortality in massively transfused pediatric trauma patients.
Neurosurgical operations usually involve the critical anatomy, meaning there is risk of injury to major vascular structures and unexpected intraoperative bleeding.[8,9] However, there is a lack of evidence mentioned concerning the factors influencing MBT in neurosurgical operation from the literature review.
At present, a nomogram is one of the clinical prediction tools (CPT) widely used to predict outcomes in several fields.[10-12] Several risk factors can be combined to predict and visualize an outcome as a two-dimensional figure or web-based application. Hence, this study aimed to develop a nomogram to predict MBT in neurosurgical operations. In addition, the secondary objective was to estimate the association between MBT and mortality in neurosurgical operations.
MATERIALS AND METHODS
Study design and study population
The present study was a retrospective study design by including patients who had undergone neurosurgical operations at our institute from January 1, 2014, to October 31, 2019. Exclusion criteria were patients who were dead on arrival or who did not have data for cross-match and transfusion. Therefore, various clinical characteristics were reviewed and collected. Therefore, the outcome was an event of massive packed red cell (PRC) transfusion for each patient as the binary classifiers. In addition, massive PRC transfusion was defined as a transfusion using more than 4 units of PRC within 1 h or more than 10 units of PRC within 24 h.[14,15]
Sample size estimation was performed using receiver operating characteristics (ROC) with the area under the ROC curve (AUC) formula. Based on an AUC of 0.850 from the study of Huang et al. with a 0.05 margin of error, a minimum of 302 patients from the testing data would be needed to evaluate the predictability.
The present study was approved by a Human Research Ethics Committee (REC 65-052-10-1). The present study did not require informed consent from patients because it employed a retrospective study design. However, patients’ identification numbers were encoded before analysis.
Clinical characteristics and imaging findings were calculated from descriptive statistics. Continuous variables were presented as mean ± standard deviation for normally distributed data and median with interquartile range for non-normally distributed data. For categorical data, the percentages were used for description.
To construct a predictive model, several factors were assessed using binary logistic regression analysis both univariate analysis and multivariable analysis. The P-values that were reported as being <0.05 were considered to be statistically significant.
The evaluation of the nomogram’s predictability contained two domains including calibration and discrimination. The Hosmer-Lemeshow goodness-of-fit (GOF) test and a GOF test P-value of 0.05 or higher indicated good calibration of the model. Therefore, the discrimination ability of the nomogram was estimated by the concordance statistic index (c-index) that equaled the AUC in the prediction of binary outcome.[19-21] Thus, internal validation was achieved to detect the overfitting problems of the model. In the present study, both 5-cross-validation and 1000 bootstrapping methods were used for internal validation. The results of internal validation were described as the optimism-corrected c-index for both methods.[20,21] Consequently, a two-dimensional nomogram was used to display the prediction model. Statistical analyses were performed using the R version 4.4.0 software (R Foundation, Vienna, Austria).
Characteristics of patients
A total of 3660 patients underwent neurosurgical operations between January 2014 and October 2019; the baseline characteristics of patients are shown in [Table 1]. The patients comprised 1903 males and 1757 females with a mean age of 46.48 ± 20.55 years. Moreover, the average body mass index was 22.99 ± 4.41 kg/m2. The leading neurosurgical conditions included brain tumor (45.7%), traumatic brain injury (TBI) (15.2%), and cerebral aneurysm (13.8%). The major neurosurgical operations included craniotomy, craniectomy, and burr hole at 36.7%, 11.7%, and 8.7%, respectively. In addition, emergency surgery was observed in 48.0% of the present cohort.
|Characteristics||Non-massive blood transfusion (n=3517)||Massive blood transfusion (n=143)||Total (%)|
|Male||1837 (52.2)||66 (46.2)||1903 (52.0)|
|Female||1680 (47.8)||77 (53.8)||1757 (48.0)|
|<=15||376 (10.7)||7 (4.9)||383 (10.5)|
|>15–60||2245 (63.8)||87 (60.8)||2332 (63.7)|
|>60||896 (25.5)||49 (34.3)||945 (25.8)|
|Mean Age-year (SD)||46.48 (20.55)|
|Hypertension||1059 (30.1)||42 (29.4)||1101 (30.1)|
|Diabetes mellitus||388 (11.0)||16 (11.2)||404 (11.0)|
|Dyslipidemia||521 (14.8)||20 (14.0)||541 (14.8)|
|Liver disease||115 (3.3)||7 (4.9)||122 (3.3)|
|Renal failure||161 (4.6)||9 (6.3)||170 (4.6)|
|Pre-operative current medication|
|Antiplatelet||122 (3.5)||7 (4.9)||129 (3.5)|
|Clexane||11 (0.3)||1 (0.7)||12 (0.3)|
|Warfarin||24 (0.7)||4 (2.8)||28 (0.8)|
|Tumor||1602 (45.6)||69 (48.3)||1671 (45.7)|
|Traumatic brain injury||513 (14.6)||43 (30.1)||556 (15.2)|
|Aneurysm||488 (13.9)||18 (12.6)||506 (13.8)|
|Non-aneurysm cerebrovascular disease||301 (8.6)||9 (6.3)||310 (8.5)|
|Spinal operation-tumor||188 (5.3)||3 (2.1)||191 (5.2)|
|Spinal operation-trauma||137 (3.9)||0||137 (3.7)|
|Spinal operation-degenerative disease||38 (1.1)||1 (0.7)||39 (1.1)|
|Spinal operation-infection||13 (0.4)||0||13 (0.4)|
|Congenital disease-brain||93 (2.6)||0||93 (2.5)|
|Congenital disease-spine||32 (0.9)||0||32 (0.9)|
|Infection (non-surgical site infection)||100 (2.8)||0||100 (2.7)|
|Normal pressure hydrocephalus||12 (0.3)||0||12 (0.3)|
|American Society of Anesthesiologists classification|
|1||2 (0.1)||1 (0.7)||3 (0.1)|
|2||211 (6.0)||3 (2.1)||214 (5.8)|
|3||3275 (93.1)||132 (92.3)||3407 (93.1)|
|4||29 (0.8)||7 (4.9)||36 (1.0)|
|Mean body mass index-kg/m2(SD)||22.9 (4.4)||23.6 (3.6)||22.9 (4.4)|
|Craniotomy||1249 (35.5)||93 (65.0)||1342 (36.7)|
|Craniectomy||385 (10.9)||42 (29.4)||427 (11.7)|
|Suboccipital or rectosigmoid approach||210 (6.0)||3 (2.1)||213 (5.8)|
|Endoscopic approach with tumor removal||173 (4.9)||1 (0.7)||174 (4.8)|
|Cranioplasty||42 (1.2)||0||42 (1.1)|
|Burr hole with biopsy/aspiration/irrigation||320 (9.1)||0||320 (8.7)|
|Spinal operation with instrumentation||204 (5.8)||1 (0.7)||205 (5.6)|
|Spinal operation without instrumentation||168 (4.8)||3 (2.1)||171 (4.7)|
|Spinal operation in congenital condition||29 (0.8)||0||29 (0.8)|
|Ventriculostomy insertion||181 (5.1)||0||181 (4.9)|
|Shunt insertion||299 (7.3)||0||299 (8.2)|
|Other||257 (7.3)||0||257 (7.0)|
|Emergency operation||1679 (47.7)||79 (55.2)||1758 (48.0)|
|Surgical infection operation||125 (3.6)||1 (0.7)||126 (3.4)|
|Packed red cell transfusion||1361 (38.7)||143 (100.0)||1504 (41.1)|
|Pre-operative hematologic laboratories|
|Hematocrit -%||37.9 (5.8)||35.6 (6.5)||37.8 (5.8)|
|Hemoglobin g/dL||12.6 (2.0)||11.9 (2.2)||12.6 (2.0)|
|White blood cell count -×103/µL||11.2 (5.4)||12.5 (6.0)||11.3 (5.4)|
|Neutrophil l-%||68.3 (16.5)||69.0 (17.3)||68.3 (16.5)|
|Lymphocyte-%||22.8 (13.6)||22.3 (15.1)||22.8 (13.7)|
|Neutrophil-to-lymphocyte ratio||6.1 (9.3)||7.4 (12.0)||6.1 (9.5)|
|Platelet count-×103/µL||298.3 (123.4)||268.9 (111.8)||297.2 (123.1)|
|Prothrombin time ratio||0.9 (0.1)||1.0 (0.3)||0.9 (0.1)|
|International normalized ratio||1.0 (0.1)||1.1(0.5)||1.06 (0.17)|
|Mean intraoperative blood loss-ml||489.3 (564.7)||3342.4 (2439.0)||600.8 (918.1)|
|Hospital-discharge mortality||25 (0.7)||58 (40.6)||83 (2.3)|
For pre-operative hematologic laboratory, the mean hematocrit of the present cohort was 12.60 ± 2.08 g/dL, and average hematocrit was 37.83 ± 5.87 %. Moreover, mean international normalized ratio (INR) was 1.06 ± 0.17. Pre-operative antiplatelet usage was observed in 3.5%, while pre-operative warfarin usage was found in 0.8%. As a result, average intraoperative blood loss was 600.80 ± 918.18 ml, and the blood transfusion rate in the present cohort was 41.1%. For those who were transfused, MBT was observed in 3.9%. In addition, 83 patients (2.3%) in the overall cohort were dead at hospital discharge. For hospital-discharge mortality according to MBT, 69.8% (58/83) of mortality cases in the MBT group. Moreover, MBT was significantly associated with increased mortality with odds ratio (OR) 95.31, 95% confidence interval (CI) 56.89–159.66.
Independent risk factors for MBT
Twelve variables with P < 0.1 in the univariate analysis were included as follows: Age group (<15 years = reference group, >15–60 years = OR 2.08 [95%CI 0.95–4.53], >60 years = OR 2.93 [95%CI 1.31–6.54]), pre-operative warfarin usage (OR 4.18 [95%CI 1.43–12.23]), body mass index (OR 1.03 [95%CI 0.97–1.07]), TBI (OR 2.51 [95%CI 1.74–3.64]), craniectomy operation (OR 3.38, [95%CI 2.32–4.92]), emergency operation (OR 3.38 [95%CI 2.32–4.92]), surgical infection operation (OR 0.19, [95%CI 0.02–1.37]), pre-operative hematocrit level (OR 0.85 [95%CI 0.79–0.92]), pre-operative white blood cell count (OR 1.03, [95%CI 1.01– 1.06]), pre-operative platelet count (OR 0.998 [95%CI 0.996– 0.999]), pre-operative prothrombin time ratio (OR 4.40, [95%CI 2.15–9.01]), and INR (OR 5.53 [95%CI 2.53–12.08]).
Subsequently, multivariable analysis with the backward elimination procedure demonstrated that age group (<15 years = reference group, >15–60 years = OR 2.10 [0.95– 4.68], >60 years = OR 2.68 [95%CI 1.19–6.04]), TBI (OR 1.82 [95%CI 1.23–2.70]), craniectomy operation (OR 2.45 [95%CI 1.63–3.67]), pre-operative hematocrit level (OR 0.95 [95%CI 0.92–0.97]), and pre-operative INR (OR 2.55 [95%CI 1.33–4.86]) were independent predictors of MBT.
Development and internal validation of a nomogram
The predictive model with five predictors was estimated in terms of model performance and internal validation. For calibration, the results of the Hosmer-Lemeshow GOF test gave a P-value of 0.11, which revealed good calibration. Hence, the model discrimination had a c-index value of 0.703, as shown in [Figure 1]. Therefore, the overfitting of the model was evaluated by 5-cross-validation techniques and 1000 bootstrapping. The optimism-corrected c-index values of cross-validation and bootstrapping were 0.703 and 0.703, respectively. Therefore, the nomogram is presented in [Figure 2].
In the present study, the incidence of MBT was 3.9%, concordant with prior studies that have been reported in a range from 1.8% to 5.0%.[1-3] MBT can occur unexpectedly as an event, which is important to consider because such an event significantly increased mortality in the present study. This finding is similar to what was previously shown in a study by O’Keeffe et al., which demonstrated that MBT was significantly associated with 30-day mortality. However, the association between MBT and mortality continues to be debated. Other previous studies did not find that MBT influenced mortality.[6,7]
The factors associated with MBT comprised age group, TBI, craniectomy operation, pre-operative hematocrit, and pre-operative INR following multivariable analysis. Older patients had a higher risk for MBT than younger patients. Similarly, our results are in concordant with prior studies.[22,23] Akaraborworn et al. studied 867 patients with trauma and found that those aged >60 years were more associated with MBT. In the present study, patients suffering from TBI and craniectomy operation showed predictors influencing MBT, which was a novel finding in terms of the authors’ knowledge. Moreover, pre-operative anemia and coagulopathy have been reported as predictors of MBT. These associations among predictors may be explained by traumatic-induced coagulopathy caused by tissue factor release from severe tissue damage and tissue hypoperfusion following trauma.[24,25] Therefore, the vicious cycle between coagulopathy and vigorous bleeding is promoted and needs appropriate resuscitation and treatment strategies. Hence, severe brain damage can develop brain swelling intraoperatively, thus leading to performance of decompressive craniectomy.
The nomogram in the present study was developed based on various predictors from multivariable analysis. Hence, the predictability of the nomogram for MBT had a c-index of 0.703, which was acceptable in the range of 0.7–0.8. After optimism correction, the c-index values of internal validation did not drop, meaning no overfitting performance.[20,21] From prior studies, Kang et al. developed a scoring model for predicting MBT in placenta previa. They found that the c-index of the model was 0.922. Moreover, Chico-Fernández studied the scoring systems for MBT in trauma patients and reported that the Assessment of Blood Consumption score had the highest c-index at 0.779.
The nomogram and other prediction scores should be estimated with the unseen dataset to ensure generalizability in the future. At present, machine learning is proposed to predict various clinical outcomes as the CPT. For intraoperative transfusion, Chang et al. used the support vector machine algorithm to forecast intraoperative transfusions in orthopedic operations with an AUC of 0.707, while Mitterecker et al. reported AUC of transfusion prediction using the gradient boosting, neural network, and random forest algorithms at 0.966, 0.966, and 0.963, respectively. Therefore, comparison of predictability between nomogram and various machine learning algorithms is challenged to perform. Tunthanathip et al. compared the predictability of intracranial injury in pediatric TBI between nomogram and machine learning-based algorithms. Therefore, the best performance of the CPT can be implicated in real-world practice.
To the best of the authors’ knowledge, this study is the first to reveal the acceptable performance of a nomogram for predicting MBT in neurosurgery. Besides, the limitations of the present study are considered. First, MBT is an uncommon event and incidence reported in <10% of cases; multicenter trials should be conducted in the future using a large number of MBT. Second, the study design was a retrospective approach that could have led to selection and information bias. However, the authors attempted to use multivariable analysis to mitigate this limitation.[32-34] Finally, this prediction tool needs to be validated externally with unseen data before implementation in real-world practice.
MBT is an unexpectedly fatal event that should be considered for appropriate preparation blood component. Further, this nomogram can be implemented for allocation in limited-resource situations in the future.
This research was a part of a retrospective and cohort study that will be published elsewhere, whereas this study focused on nomogram predicting MBT.
Declaration of patient consent
Patient’s consent not required as there are no patients in this study.
Financial support and sponsorship
Conflicts of interest
There are no conflicts of interest.
- Field J Korean Neurosurg Soc. 1976;5:185-94.Problems of massive transfusion in neurosurgical.
- [Google Scholar]
- Translational Medicine in Neurosurgery. Bangkok: Sahamit Pattana Printing; 2022. p. 111-34.Development of Nomograms for Neurosurgery In:
- [Google Scholar]
- Oearsakul T, ed. Traumatic Brain Injury and Cervical Spine Injury. Bangkok: Sahamit Pattana Printing; 2022. p. 373-84.Traumatic cerebrovascular injury In:
- [Google Scholar]
- Gastroenterol Hepatol Bed Bench. 2012;5:79-83.How to control confounding effects by statistical analysis.
- [Google Scholar]