Characteristics of the training cohort
According to diagram in Figure 1, a total of 426 patients comprising 252 of HCV patients, 27 of HBV patients, 52 of PBC patients, and 95 of AIH patients were enrolled in this study (Table 1). Based on the liver biopsy examinations, 128 patients had pointed out stage 1 fibrosis; 149 had stage 2 fibrosis; 114 had stage 3 fibrosis and 35 had stage 4 fibrosis. Among them, 336 patients were followed up for 1 year or more. The longest follow up period was 32 years. Hepatocellular carcinoma (HCC) was pointed out in 45 patients and 42 ones died.
Generation of Albumin platelet product
To determine items of an equation, linear trend through fibrosis staging was evaluated for age, platelet, total protein, albumin, aspartate aminotransferase (AST), alanine aminotransferase (ALT), total bilirubin (T-Bil), and gamma-glutamyl transpeptidase (γGTP). Among them, age and T-Bil linearly increased, and platelet and albumin decreased (Table 1). R squares were greater than 0.1 in platelet and Alb. Age and T-Bil had lower R squares than 0.1. Combining T-Bil, platelet and albumin mathematically, four equations were generated; Alb × Plt / 1000, Albumin platelet product (APP) ; Alb / T-Bil, Albumin bilirubin quotient; 10 × T-Bil / platelet, Bilirubin platelet quotient; Alb × Plt / (T-Bil × 100), Three math.
Diagnostic ability of novel indices for liver fibrosis staging in the training cohort
Differences in median values between two fibrosis stages were analyzed for four indices using Steel-Dwass test (Figure 2). The results showed that the APP could differentiate any four stages (a). The albumin bilirubin quotient (b), bilirubin platelet quotient (c), and three math Alb × Plt / (T-Bil × 100) (d), could differentiate fibrosis stage 3 from stage 2 and stage 4 from stage 3, whereas they failed to differentiate between stage 1 and 2.
Receiver operating characteristic (ROC) analysis was performed to assess the ability to distinguish advanced fibrosis (F3-4) from nonadvanced fibrosis (F0-2) and cirrhosis (F4) from noncirrhotic stages (F0-3). As shown in Figure 3, area under ROC (AUROC) of four indices to distinguish advanced fibrosis from nonadvanced fibrosis was greater than 0.7 (a-d). A cut-off value of APP to differentiate F3-4 from F0-2 was determined at 6.395 with 0.7383 of sensitivity, 0.7220 of specificity, and 2.656 of positive likelihood ratio (a).
The AUROC of the indices to detect cirrhosis differentially from noncirrhotic stages resulted greater than 0.8 in APP (a), Bilirubin platelet quotient (c) and Three math (d) as shown in Figure 4. The AUROC of Albumin bilirubin quotient stayed smaller than the others (b). A cut-off value of APP to differentiate F4 from F0-1 was determined at 4.349 with 0.7143 of sensitivity, 0.8670 of specificity, and 5.371 of positive likelihood ratio.
The greatest AUROC was presented by APP among four indices. The second was Three math. Based on the analyses above, APP and Three math were extracted as candidates of liver fibrosis staging.
Comparison between APP and conventional indices in the training cohort
Differences in median values between two fibrosis stages were analyzed for FIB-4 (a), AST to platelet ratio index (APRI, b), γGTP to platelet ratio index (GPR, c) and Albumin bilirubin score (ALBI score, d) using the Steel-Dwass test (Figure 5). As a result, FIB-4 and ALBI score differentiated fibrosis stage 3 from stage 2 and stage 4 from stage 3 whereas they failed to differentiate between F0-1 and 2 (p < 0.05). APRI did not distinguish stage 4 from stage 3. GRP was not significantly different between F0-1 and 2 and between F3 and 4.
ROC analysis revealed, as shown in Figure 6, that AUROC of FIB-4 (a) and ALBI score (b) to distinguish advanced fibrosis from nonadvanced fibrosis ranged between 0.7 and 0.8. The AUROCs to detect cirrhosis differentially from noncirrhotic stages were determined between 0.8 and 0.9 (c, d). Compared to FIB-4 and ALBI score, APP revealed competitive in staging of liver fibrosis based on ROC analyses.
Prognosis prediction by APP in the training cohort
In total, 336 patients were followed up for at least one year in the training cohort. Kaplan-Meyer analysis was performed in the training cohort to estimate the contribution of the APP to HCC-free survival and overall survival. As shown in Figure 7, a cut-off value = 6.395 could significantly differentiate HCC-free survival (a) and overall survival (b). Survival rates at 15 year were 91.2 % and 75.9 % for HCC free survival; 95.2 % and 77.7 % for overall survival. Another cut-off value = 4.349 also stratified HCC prevalence (c) and mortality (d) in the training cohort. Survival rates at 15 year were 87.1 % and 71.5 % for HCC free survival; 93.6 % and 55.6 % for overall survival.
Post-hoc power analysis resulted in a power = 0.975 for HCC-free survival and 0.998 for overall survival using a cut-off value = 6.395. When the training cohort was stratified using an alternative cut-off value = 4.349, HCC-free and overall survival was proved with power 0.808 and 1.000.
A multivariate analysis in a training cohort
To investigate predictive ability of APP, Cox proportional hazard model was applied on follow up data of the training cohort. Concerning four variables, age, gender, etiology and APP, hazard ratios were analyzed to determine whether APP independently contribute to HCC prevalence and mortality in the training cohort. As shown in Table 2, both of APP < 6.395 and 4.349 significantly increased HCC prevalence and mortality. The proportional hazard model analysis was validated with 4 variables for 45 patients with HCC or 42 deaths (Table 1) 10.
Characteristics of HCV-specific subpopulation in the training cohort
To probe diagnostic accuracy of liver fibrosis and contribution to prognosis prediction by APP further, a cohort with HCV infection was extracted from the training cohort. As shown in Table 3, 252 patients presented a distribution of liver fibrosis stage as follows; 58 of stage 1, 93 of stage 2, 72 of stage 3 and 29 of stage 4. Similar to results in the training cohort, Total bilirubin increased along with fibrosis progression while platelet and albumin decreased.
Among the sub-cohort with HCV infection, 191 patients were followed up for the median 10 years. The longest follow up period recorded 29 years. Sustained viral response was confirmed 86 patients after antiviral therapy subsequent to baseline liver biopsy examination. Thirty-seven patients were complicated with HCC and 25 ones were described died.
Diagnostic ability of liver fibrosis staging in HCV-specific subpopulation in the training cohort
Differences in median values between two fibrosis stages were evaluated for APP (a) and FIB-4 (b), as shown in Figure 8. Both of the indices differentiated fibrosis stage 3 from stage 2 and stage 4 from stage 3 (p<0.05).
ROC analysis revealed that AUROC of APP to distinguish advanced fibrosis from nonadvanced fibrosis was greater than 0.8 (a), as shown in Figure 9. Youden index determined a cut-off value of 6.395, which is identical to the cut-off value in the training cohort. Cirrhosis was also divided from noncirrhotic status with AUROC > 0.8 by APP (b). A cut-off value = 4.349 was also identical to the value in the training cohort.
The AUROCs of APP were greater than that of FIB-4 in both case of F0-2 versus F3-4 (c) and F0-3 versus F4 (d). Diagnostic accuracy of APP was also clarified in HCV-specific population.
Influence of Hepatitis Activity on Fibrosis indices
Indices of liver fibrosis have been reported to fluctuate according to hepatitis activity grading 9. In 93 patients with stage 2, APP in grade 0-1 patients did not differ from that in grade 2 patients (Figure 8c). In case of 72 patients with stage 3, APP was not significantly different between grade 1-2 and 3 patients (d). However, FIB-4 was significantly fluctuated in stage 2 and 3 patients (e, f).
Prognosis prediction by APP in HCV-specific subpopulation in the training cohort
To determine prognostic value of APP in HCV-specific subpopulation, Kaplan Meier curve was calculated using follow up data of 191 of HCV patients. As a result, APP < 6.395 at baseline meant poorer HCC-free survival compared to those with APP ≥ 6.395, while overall survival was not differentiated by the cut-off value = 6.395 (Figure 10a, b). Survival rate at 15 year was 96.0 % and 67.2 % for HCC free survival. APP < 4.349 predicted significantly poorer HCC-free survivals and overall survival (c, d). Survival rate at 15 year was 83.3 % and 62.0 % for HCC free survival and 92.1 % and 59.6 % for overall survival.
Post-hoc power analysis resulted in a power = 0.998 for HCC-free survival using a cut-off value = 6.395. When the training cohort was stratified using an alternative cut-off value = 4.349, HCC-free and overall survival was proved with power 0.819 and 0.998.
A multivariate analysis in HCV-specific subpopulation in the training cohort
Cox proportional hazard model was applied on follow up data of HCV-specific subpopulation in the training cohort. Considering four variables, age, gender, viral response to antiviral therapy and APP, hazard ratios for HCC prevalence and mortality were calculated. As shown in Table 4, APP < 6.395 yielded significant increase of HCC prevalence. APP < 4.349 also increased of both HCC prevalence and mortality.
However, based on the criteria, number of variables exceeded 10 times of events; 37 patients with HCC and 25 deaths, as shown in Table 3 10. Thus, a validation study was prepared to clarify clinical significance of APP for HCV-specific cohort.
Diagnostic ability of liver fibrosis staging in the validation cohort
To evaluate the diagnostic ability of liver fibrosis staging, the APP was calculated for each fibrosis stage in the validation cohort. As shown in Figure 11, the APP was able to differentiate the four stages of fibrosis (p < 0.05).
ROC analysis revealed that the AUROCs of the APP for distinguishing advanced fibrosis from nonadvanced fibrosis (a) and cirrhosis from noncirrhotic status (b) were greater than 0.8 (c), as shown in Figure 12. The AUROCs of the APP were greater than that of FIB-4 or APRI (c). The diagnostic abilities of the APP with two cut-off values are summarized in Table 5. Both cut-off values, APP = 6.395 and = 4.349, were characterized by negative predictive values relatively greater than 80%.
Prognosis prediction by the APP in a validation cohort
The clinical impact of the APP on HCC-free survival and overall survival was confirmed using Kaplan-Myer analysis in the validation cohort through 15 years observation (Figure 13). Patient number of HCC complication and death at 15 year was 143 and 73. Each cut-off value, APP = 6.395 and 4.349, could differentiate HCC-free survival in 707 patients with HCV infection (a, b). Overall survival was also stratified by two cut-off values (c, d).
Post-hoc power analysis resulted in a power = 1.000 for four comparison above between greater and smaller APP groups.
The Cox proportional hazard model was also applied on the validation cohort using a stepwise method, as shown in Table 6. The performances of interferon therapy, sex, age, serum AFP and WFA+-M2BP levels were included in the multivariate analyses. The results showed that APP < 6.395 contributed to a greater risk of HCC complication. APP < 4.349 also indicated increased prevalence of HCC and death. Number of variables did not exceed 10 times of HCC patients or death 10,11.