| Objective:This study aimed to evaluate the predictive ability and clinical utility of COSSH-ACLF Ⅱs in patients with HBV-ACLF,and to explore it’s feasibility in guiding the timing of non-bioartificial liver treatment for HBV-ACLF.Methods:In this study,clinical data from 206 consecutive cirrhotic patients hospitalized with the diagnosis of HBV-ACLF based on the Chinese Group on the Study of Severe Hepatitis B-ACLF(COSSH-ACLF)criteria in Zhujiang Hospital of Southern Medical University and The Third Affiliated Hospital of Sun Yat-sen University from January 2018 to October 2022 were analyzed retrospectively.According to the difference of treatment plan,the enrolled patients were divided into exposed group(n=116)and non-exposed group(n=90).The performance of the new prognostic model in predicting outcomes was compared with that of five other generic prognostic models:COSSH-ACLF s、Child-Turcotte-Pugh、MELD、MELD-Na、and CLIF-C ACLF s.The performance of the prognostic model was evaluated comprehensively through three dimensions,including discrimination,calibration and clinical utility.Concordance statistics was utilized for evaluate the discrimination.The probability density function(PDF)was utilized to define an integral of survival and non-survival density over a certain range,and the overlapping coefficient was calculated to verified the discrimination of models by measuring the similarity of probability distributions between survival and non-survival.Hosmer-Lemeshow goodness-of-fit test was used to evaluate the calibration of the prediction model,and the results of the goodness-of-fit test are visualized through the event risk ten-bit bar chart and the calibration curve.Decision Curve Analysis(DCA)was used to evaluate the clinical utility of the prognostic model.The DCA curve describes the change of net benefit under the condition of intervention according to the predicted value of the model with the change of threshold probability.By comparing the DCA curve of different models,the clinical utility of different models can be evaluated.Patients in the exposed group and non-exposed group were divided into 3 different risk strata based on COSSH-ACLF Ⅱ score:①<7.4;②7.4-8.4;③≥8.4.Kaplan-Meier estimate was used to evaluate the growth rate of 28-/90-days mortality of patients in each risk strata and draw survival curve;The Log-Rank test was used to compare the survival curve and overall survival time among the subgroups.COX proportional hazards model was used to evaluate the effect of artificial liver therapy on the prognosis of each subgroup.Results:1.The C-statistics of 28-day mortality predicted by COSSH-ACLF Ⅱ s(0.926)was higher than that of COSSH-ACLF s(0.914,p=0.4280),and the difference was not statistically significant.But significantly higher than CLIF-C ACLF s(0.868,p=0.0005),MELD-Na(0.822,p=0.0018),MELD(0.851,p=0.0091),CTP(0.821,p=0.0010),The difference was statistically significant.The 90-day mortality prediction of COSSH-ACLF Ⅱ s model was significantly higher than that of COSSH-ACLF s(0.920,p=0.0487)and CLIF-C ACLF s(0.896,p=0.0002),MELD-Na(0.845,p=0.0004),MELD(0.873,p=0.0031),CTP(0.828,p<0.0001),and the difference was statistically significant.2.The proportion of patients with poor prognosis gradually increased with the increase of COSSH-ACLF Ⅱ s,COSSH-ACLF s,CLIF-C ACLF s,MELD-Na,MELD and CTP.Moreover,there were significant differences in the peak values of the probability distribution curves of surviving and dying patients(Figure 1-4).The overlap coefficient of the probability distribution curves of 28/90 days of survival and death in the new prognosis model COSSH-ACLF Ⅱ s was significantly decreased compared with other models(COSSH-ACLF Ⅱ s:30.0%/28.3%;COSSH-ACLF s:31.9%/32.5%;CLIF-C ACLF s:46.3%/41.0%;MELD-Na:46.9%/42.5%;MELD:44.5%/41.3%;CTP:51.7%/51.1%,all p<0.05).The similarity of the probability distribution curves of survival and death patients defined by the new model COSSH-ACLF Ⅱ s was lower than that of the other five prognostic models,suggesting that COSSH-ACLF Ⅱ s had better prognostic performance.3.The 90-day predicted mortality rate of the COSSH-ACLF Ⅱ s model is generally similar to the observed mortality rate(-2Log likelihood=118.379;Hosmer-Lemeshowx2=8.008,p=0.433),the best fit of the model,the predictive power of short-term prognosis was superior to that of COSSH-ACLFs(-2Log likelihood=144.926;Hosmer-Lemeshow χ2=13.271,p=0.103),CLIF-C ACLF score(-2Log likelihood=158.150;Hosmer-Lemeshow x2=11.084,p=0.197),the MELD-Na(-2Log likelihood=195.998;Hosmer-Lemeshow χ2=14.202,p=0.077),the MELD(-2 Log likelihood=181.682;Hosmer-Lemeshow x2=16.126,p=0.041).4.The results of decision curve analysis showed that COSSH-ACLF Ⅱ s model could achieve the highest net benefit rate in guiding clinical decision making within a large threshold probability range,and its clinical application value was better than that of COSSH-ACLF s、CLIF-C ACLF s、MELD、MELD-Na and Child-Turcotte-Pugh.5.The results of survival analysis showed that after artificial liver therapy and medical treatment,there were statistically significant differences in the 28/90 day survival curve between the exposed group and the non-exposed group.The average survival time of the exposed group was significantly longer than that of the non-exposed group(Log-rank x2=10.425/5.446,p=0.0012/p=0.020);Univariate COX proportional hazards model analysis showed that artificial liver therapy was a prognostic factor for patients(HR:0.582;95%CI:0.366-0.926;p=0.022),and further multivariate COX regression analysis also showed that artificial liver therapy was still the main prognose-related factor after adjusting for age,sex,HE grade,severity of disease,and total bilirubin(HR:0.393;95%CI:0.226-0.684;p=0.001).6.Patients in the exposed group and the non-exposed group were divided into 3 different risk strata based on COSSH-ACLF Ⅱ s score:low risk(COSSH-ACLFⅡ s<7.4),intermediate risk(7.4≤COSSH-ACLFⅡ s<8.4)and high risk(COSSH-ACLFⅡs≥8.4).In the low risk strata,there was no significant difference in the 28-/90-day mortality rate and survival curve between the exposed group and the non-exposed group(5.1%/6.8%vs 2.1%/2.1%,Pearson Chi-Square test χ2=0.630/1.260,p=0.427/0.262),in the meanwhile,there was no significant difference in the survival curve between this two subgroups(Log-rank test χ2=0.598/1.205,p=0.439/0.272).Multivariate COX proportional hazards model analyses showed that artificial liver therapy was not a prognostic factor in the low-risk range(HR:0.606;95%CI:0.238-1.358;p=0.723).In the intermediate risk strata,the 28-/90-day mortality rate in the exposed group was significantly lower than that in the non-exposed group(16.2%/32.4%vs 79.2%/79.2%,Pearson Chi-Square test x2=23.851/12.721,all p<0.001).At the same time,in the medium risk interval,there was also a significant difference in the survival curve between the two groups,and the average survival time of the exposed group was significantly longer than that of the non-exposed group(Log-rank test χ2=27.303/20.135,all p<0.001).Multivariate COX proportional hazards model analysis showed that artificial liver therapy was an independent prognostic factor in the medium risk interval,and the risk of death in patients receiving artificial liver therapy was only 0.211 times higher than that in patients receiving medical treatment(HR:0.211;95%CI:0.099-0.448;p<0.001).In the high risk strata,there was no significant difference in the 28-/90-day mortality rate between the exposed group and the non-exposed group(75.0%/90.0%vs 84.2%/94.7%,Pearson Chi-Square test χ2=0.507/0.308,p=0.476/0.579),and there was no significant difference in survival curve between the two groups(Log-rank test χ2=1.274/1.652,p=0.259/0.199).COX proportional hazards model analysis showed that artificial liver therapy was not a prognostic factor in the high risk strata(HR:0.437;95%CI:0.177-1.079;p=0.073).Conclusion:1.The prediction performance and clinical application value of COSSH-ACLFⅡs is significantly better than that of COSSH-ACLF s、CLIF-C ACLF s、MELD、MELD-Na and Child-Turcotte-Pugh.2.COSSH-ACLF Ⅱ s can effectively guide the timing of non-bioartificial liver for HBV-ACLF. |
PDF Full Text Request