PRO-C3 and ADAPT Algorithm Accurately Identify Patients With Advanced Fibrosis due to Alcohol-related Liver Disease

Bjørn S. Madsen; Maja Thiele; Sönke Detlefsen; Maria Kjærgaard; Linda S. Møller; Jonel Trebicka; Mette J. Nielsen; Natasja Stæhr Gudmann; Diana J. Leeming; Morten A. Karsdal; Aleksander Krag;


Aliment Pharmacol Ther. 2021;54(5):699-708. 

In This Article


We included 426 participants as described in the flow chart in Figure 1. Patients were randomly assigned to a test or validation cohort. The characteristics of the enrolled participants in the total, test- and validation cohorts are shown in Table 1. In general, participants in the test- and validation cohort did not differ statistically in core parameters.

Figure 1.

Study flow chart

PRO-C3 is Highly Associated With the Severity of ALF

The median concentration of PRO-C3 was 12.9 (±10.3) ng/ml in the total cohort, 13.2 (±10.5) ng/ml in the test cohort and 12.7 (±9.6) ng/ml in the validation cohort (P = 0.713 Mann-Whitney test). PRO-C3 strongly correlated with Kleiner fibrosis stage in the total cohort (rho = 0.61, P = 0.000). Kruskal-Wallis confirmed that the concentration of PRO-C3 differed significantly between Kleiner fibrosis stages (P = 0.000). Post hoc Dunns test confirmed statistically significant differences in PRO-C3 concentrations between all consecutive stages apart from stage 2 fibrosis vs stage 3 fibrosis (P = 0.481). Dotplot of PRO-C3 serum concentration related to Kleiner fibrosis stage is depicted in Figure 2. We subsequently performed a logistic regression model to validate the association between advanced fibrosis and PRO-C3 when adjusted for various clinical variables. As seen in Table 2, PRO-C3 remained independently associated with advanced fibrosis (OR = 1.07, 95% CI 1.04–1.10, P = 0.000).

Figure 2.

PRO-C3 concentration in serum according to Kleiner fibrosis stage. Dotplot of the serum concentration of PRO-C3 in the healthy population group and its relation to the Kleiner fibrosis stage in the total cohort of patients with current or prior alcohol overuse. The brown line indicates the median value

Accuracy of PRO-C3 to Detect Advanced ALF

The PRO-C3 had good diagnostic accuracy to detect advanced ALF with an AUROC of 0.85 (95% CI 0.79–0.90) in the total cohort, as seen in Table 3. When analysing the test and validation cohorts, the AUROC was 0.86 (95% CI 0.79–0.92) in the former and 0.83 (95% CI 0.75–0.92 P = 0.7056) in the latter. ROC curves are seen in Figure 3. When applying the suggested cut-off value of 15.6 ng/ml to detect advanced fibrosis, the sensitivity of PRO-C3 in the total cohort was 81%, specificity 73% and PPV 38% and NPV 95%. The corresponding results from the test and validation cohorts are seen in Table 3.

Figure 3.

Receiver operating characteristic for PRO-C3 and ADAPT score to diagnose advanced fibrosis and cirrhosis. Receiver operating characteristics curves for PRO-C3 and ADAPT algorithms to detect advanced liver fibrosis in the total cohort (A), test cohort (B) and validation cohort (C)

PRO-C3 was not well calibrated (Hosmer-Lemeshow chi square = 21.86, P = 0.005). Risk prediction and calibration plots are seen in Figure 4. Misclassifications were mainly driven by false-positive results. We subsequently performed a logistic regression model to identify risk factors for being falsely classified as having advanced ALF with the variables diabetes, alcohol consumption, gender, age, steatosis, ballooning, lobar inflammation, GGT, plates and biopsy length. Only the degree of steatosis, ballooning, lobular inflammation and concentration of GGT were associated with an increased risk to be falsely classified as having advanced ALF (data not shown). In a subsequent subgroup analysis, the AUROC was significantly lower among patients classified as with GGT >260 U/L compared to the group with GGT below the threshold (AUROC = 0.70 vs AUROC = 0.86, P = 0.034). In contrast, neither drinking status (abstinent vs alcohol consuming), diabetes (diabetic vs non-diabetic) nor a high ALT level (comparing ALT levels above or below 49 U/L) reduced the AUROC significantly as depicted in Figure 5. Likewise, obesity did not impact the diagnostic accuracy (Supporting Information). Further subgroup analysis was performed to mimic the diagnostic accuracy of PRO-C3 in a primary care setting. Patients were divided into a low-risk group if recruited from municipal alcohol rehabilitation clinics or advertisements and a high-risk group if they were recruited from hospital liver clinics. The negative predictive value was 98% (95–100) when PRO-C3 was used to exclude advanced fibrosis in the low-risk group with a disease prevalence of 7%. The positive predictive value of PRO-C3 dropped to 22% (12–36) in this setting. The full analysis based on referral is available in the Supporting Information.

Figure 4.

Risk prediction and calibration curves according to serum PRO-C3. (A and B) Risk-prediction curves to evaluate the probability of advanced fibrosis according to the serum concentration of PRO-C3 and the ADAPT score. (C and D) Calibration slopes for PRO-C3 and the ADAPT score in the total cohort. The marron line graphs the agreement between predicted probability of advanced fibrosis on the x-axis and observed proportion with advanced fibrosis on the y-axis. The perfect calibration with 100% agreement is marked with a black dashed line

Figure 5.

Receiver operating characteristic for PRO-C3 in subgroups. ROC curves for the detection of advanced fibrosis by PRO-C3 in four subgroups. (A) Abstinent vs. alcohol using participants (B) Diabetic vs. non-diabetics participants (C) Participants with an ALT level above or below 49 U/L (D) Participants with an GGT level above or below 260 U/Ls

Diagnostic Accuracy of the ADAPT Algorithm to Detect Advanced ALF

The median ADAPT score was 5.5098 (±4.7929) in the total cohort. ADAPT had higher diagnostic accuracy compared to PRO-C3 alone in the total cohort with AUROC = 0.88 (95% CI 0.83–0.93, P = 0.010). The corresponding AUROC of ADAPT was 0.91 (95% CI 0.86–0.96) in the test cohort and 0.85 (95% CI 0.75–0.94, P = 0.230) in the validation cohort (Table 3). The reported optimal cut-off value of 6.3287 for ADAPT to detect advanced fibrosis was used.[14] By applying this cut-off value, the sensitivity was 86%, specificity 78%, PPV 44% and NPV 97% in the total cohort. The results from the test and validation cohorts are reported in Table 3. Results when using optimized cut-offs and rule-in and rule-out criteria are available in the Supporting Information.

Head-to-head Comparison of Diagnostic Accuracy With Other Serological Fibrosis Markers

Results of the head-to-head comparison with non-patented serological fibrosis markers are seen in Table 4. Forns index was the best-performing non-patented biomarker with an AUROC of 0.83 (95% CI 0.78–0.89). The ADAPT score, but not PRO-C3, performed significantly better than the Forn index in the total cohort. When using the recommended cut-off value to detect advanced fibrosis, the sensitivity of the Forns index was 67%, specificity 89%, PPV 55% and NPV 93% (Table 3).