Development Of A Dynamic Predictive Model For Postoperative Transfusion Risk In Spinal Tuberculosis

Objective: Lesion debridement combined with vertebral fusion is the primary surgical procedure for the treatment of spinal tuberculosis.The purpose of this study was to report the incidence and risk factors of postoperative allograft transfusion in patients with spinal tuberculosis and to develop a dynamic predictive model to help guide clinical practice.Methods: This retrospective study recorded 115 patients with spinal tuberculosis referred to our institution between May 2010 and April 2020.Based on a restrictive transfusion strategy,allogeneic transfusions were performed in patients who met the following criteria: 1.hemoglobin less than 70g/L;2.hemoglobin of 70-100g/L in cases of advanced age and poor cardiopulmonary function.Data of all subjects were obtained from electronic medical records,including demographic characteristics,laboratory test results and surgical details.Stepwise logistic regression analysis and six machine learning algorithms(support vector machine,decision tree,multilayer perceptron,Naive Bayesian,k-nearest neighbors and random forest)were used to analyze the data and develop predictive models.A 10-fold cross-validation was used to compare the prediction performance of the seven developed models and an open network calculator was developed relying on the model with the best prediction performance.Results: There were 41 cases in the transfusion group and 74 cases in the nontransfusion group,with an incidence of allogeneic transfusion of 35.65%.Compared with the intergroup differences in each subgroup,the following parameters confirmed to be statistically significant: age(P=0.001),anticoagulant history(P=0.019),length of hospital admission(P=0.046),surgical duration(P=0.001),intraoperative blood loss(P=0.017),number of fused vertebrae(P=0.048),hemoglobin level(P=0.016),mean hemoglobin concentration(P=0.001),albumin level(P= 0.009)and hospital costs(P<0.001).Age,proportion of anticoagulant use,length of hospital admission,surgical duration,intraoperative blood loss,number of fused vertebrae,and hospital costs were significantly higher in the transfusion group than in the nontransfusion group,whereas hemoglobin level,mean hemoglobin concentration,and albumin level were significantly lower than in the non-transfusion group.Stepwise logistic regression analysis finally determined the surgical duration(OR = 1.009,95% CI 1.001-1.016,P = 0.025),intraoperative blood loss(OR =1.002,95% CI 1.001-1.003,P = 0.004),number of fused vertebrae(OR = 2.641,95% CI 1.402-4.977,P = 0.003),history of anticoagulant drugs(OR = 3.994,95% CI 1.182-13.487,P = 0.026),hemoglobin level(OR = 0.951,95% CI0.906-1.000,P = 0.048),albumin level(OR = 0.856,95% CI 0.734-0.998,P=0.047)and mean hemoglobin concentration(OR=0.941,95% CI 0.886-1.000,P=0.049)were independent risk factors significantly associated with the incidence of postoperative transfusion in spinal TB.10-fold cross-validation ROC curves evaluated the prediction performance of nomogram and machine learning algorithms,and the nomogram had the best prediction performance with mean AUC 0.76 and maximum AUC 0.93.Therefore,the clinical value of the nomogram was further validated using calibration plots,decision curve analysis and clinical impact curves.The results suggested good agreement between actual and nomogram predictions,and patients receiving decision guidance benefited as a result.The nomogram with the best predictive power(results of the logistic analysis)were converted into a dynamic web-based transfusion risk calculatorConclusion: In summary,because of the combined effects of anemia secondary to tuberculosis and surgical procedures,patients with spinal tuberculosis are moderately likely to receive allogeneic blood transfusions in the perioperative period.We identified seven independent risk factors influencing transfusion and illustrated them with a nomogram and a web calculator.This will help to enhance perioperative management and rational allocation of health care resources for patients with spinal tuberculosis as clinical prediction models are promoted.