Risk Factors And Predictive Model For Early Death In De Novo Acute Promyelocytic Leukemia Patients

ObjectiveEarly death(ED)is the major reason for treatment failure in patients with acute promyelocytic leukemia(APL).The Sanz risk stratification was originally used to assess the risk of relapse rather than early death.There is no widely accepted predictive model for early death of APL patients.The purpose of our study was to evaluate the early mortality of newly diagnosed APL patients in the First Afiliated Hospital of Soochow University.Understanding the risk of early death and establishing a new risk score system for predicting it were also the purpose of the research.Methods1.Patients:Collected the clinical and laboratory data of de novo APL patients that admitted to the First Affiliated Hospital of Soochow University from May 2005 to November 2019.570 well-documented patients were included in our research.2.Establishment of the risk model:570 patients were randomly divided into training cohort and a validation cohort.Univariate logistic regression analysis was performed in the training cohort to determine the risk factors that may affect early death.Variables with significant statistical differences in univariate analysis(P<0.05)were included in multivariate regression analysis.Multivariate logistic regression analysis determined the risk factor included in the risk model.3.Evaluation of the risk model:(1)Verification of discrimination(C index;Integrated Discrimination Improvement,IDI),consistency(calibration plot),and clinical effectiveness(Decision Curve Analysis,DCA)in the training cohort.In addition to training cohort(internal)cohort,the model also verified with the validation cohort(external).(2)Compared the predictive ability between the traditional Sanz risk stratification and the newly established risk model.Results1.Early mortalityThere are 570 de nove APL patients in the First Affiliated Hospital of Soochow University from May 2015 to November 2019.The early mortality of newly diagnosed APL patients was 7.45%(43/570).We assessed the early mortality of APL patients in recent years.It was found that the early mortality rates of 2005.5-2013.12 and 2014.1-2019.11 were 8.72%(19/218)and 6.82%(24/352),respectively.Early mortality was not significantly decreased(P=0.440).Fully understanding early death and related risk factors is of great significance for reducing early mortality.2.Causes of early deathFatal bleeding(79.07%,34/43)was the leading cause of early death in APL patients.The bleeding-related mortality in our study was 5.96%(34/570).Cerebral hemorrhage(70.59%,24/34)was the most common,followed by pulmonary hemorrhage(26.47%,9/34).44.12%(15/34)bleeding-related mortality was concurrent with differentiation syndrome,suggesting that differentiation syndrome was closely related to bleeding.WBC and LDH were independent risk factors for simple fatal hemorrhage,while age,WBC,PLT,and LDH were associated with fatal hemorrhage accompanying differentiation syndrome.Differentiation syndrome was the second cause of early death(18.60%,8/43).The incidence of early death related to differentiation syndrome in this study was 1.4%(8/570).Age,PLT,WBC,and BMI were independent risk factors for early death associated with differentiation syndrome.Fatal bleeding was the leading cause of death before day 7(65.22%,15/23),differentiation syndrome with or without fatal bleeding(N=15/20,75%)or severe infection(1/20,5%)may be responsible for the death beyond day 7.The risk factors were not exactly the same at different stages of induction.WBC,LDH,age,and creatinine were associated with the former;age,WBC,LDH,PLT,albumin,Cr,and BMI were associated with the later.3.Establishment of the risk modelMultivariate logistic regression analysis in the training cohort showed that age>52 years(OR=5.170,P=0.002),WBC count>10×109/L(OR=9.062,P<0.001),PLT count≤10×109/L(OR=4.254,P<0.001)and LDH level>500U/L(OR=3.002,P=0.046)were independent risk factors of ED in the patients.We established the risk model based on the results above:risk score=9.062×(WBC>10×109/L)+5.170×(age>52 years old)+3.002×(LDH>500 U/L)+4.254×(PLT≤10×109/L).Taking clinical practical into consideration.Modify the model as following:2 x(WBC>10×109/L)+1.5 x(age>52 years old)+1 x(LDH>500 U/L)+1 x(PLT≤10 x 109/L).Newly diagnosed APL patients can get a risk score based on the above observations and determine the risk of early death:low risk:0 points(early mortality:0%);moderate risk:1-2 points(early mortality:5.52%);high risk:2.5-4 points(early mortality:22.73%);ultra-high risk:>4.5 points(early mortality:61.54%).Besides,we built a nomogram according to regression analysis,which could visualize regression results and provides individualized predictions of the risk of early death.4.the validation of the modelDiscrimination:We compared the AUC of model with traditional Sanz risk stratification.In the training cohort,the risk model(AUC:0.878)shows a better predictive compared with traditional Sanz risk stratification(AUC:0.799,P＝0.001118).In the validation cohort,the model(AUC:0.863)also shows a better ED prediction than the Sanz risk stratification(AUC:0.9792,P=0.01063).Compared with traditional Sanz risk stratification,the new risk model shows comprehensive discriminative improvement(IDI)index greater than 0(IDI training cohorn=0.0965,95%CI:0.015-0.178,P-value=0.02032;IDI validation cohort=0.0965,95%CI:0.0192-0.1739,P-value:0.01441).Consistency:the calibration plot indicates that the new risk model shows a good consistency in predicting early death(slope=1.0009;slope verified=0.9879).The Hosmer-Lemeshow test of the training and validation cohort show that there was no difference between the calibration curve and the ideal curve(both P value>0.05).Clinical effectiveness:the decision curve analysis(DCA)curve shows that patients may benefits from Sanz risk stratification and the new risk model.Compared with Sanz risk stratification,more patients could benefit from the new risk model.Predictive ability of risk model in different induction stage:We analyzed the risk factors of dijfferent induction stages.The risk factors were not exactly the same in different stages of period.The new risk score model also shows higher sensitivity and specificity in diferent period of induction in spite of different risk factorsConclusionIn the era of all-trans retinoic acid,early death,occurring from the period before therapy initiation until the end of induction,has become the major obstacle of APL.The risk model based on age,WBC count,PLT count and LDH shows higher discrimination and consistency in predicting early death compared with traditional Sanz risk stratification.