Demographics and clinical data
A total of 66 patients from the derivation cohort, 75 patients from the validation cohort, and 37 age-matched controls were enrolled. The mean age was 50.2 ± 14.6 (median: 52) days and 55.3 ± 16.9 (median: 58) days for the derivation cohort and validation cohort, respectively. There were no significant differences in birth weight (3134.4 ± 618.9 g vs 2984.0 ± 502.2 g, p>0.05) and weight at admission (4254.8± 872.2 g vs 4466.6 ± 1112.3 g, p>0.05) between the two cohorts. Overall, 63 female and 78 male cholestatic infants were enrolled. A total of 105 infants underwent intraoperative cholangiography and liver biopsy, of which 74 BA cases and 31 non-BA cases were identified. Another 36 infants were assumed to have no BA due to the recovery of cholestasis during the clinical follow-up.
Baseline patient characteristics of the derivation cohort are shown in Table 1. Demographic and clinical parameters including birth weight, age and weight at admission, sex, parity, recurrent jaundice, and splenomegaly showed no significant differences between the BA and non-BA groups (p>0.05, all). The frequency of clay stool and hepatomegaly was higher in the BA group than in the non-BA group. There was no apparent significant difference in liver function tests except for total bilirubin and γ-GT between the BA and non-BA group. γ-GT levels were much higher in the BA group than in the non-BA group (p<0.001). The frequency of abnormal gallbladder size and positive findings on hepatobiliary scintigraphy were also significantly higher in the BA group than in the non-BA group.
Serum bile acid concentration in BA, non-BA, and normal controls
Among the 15 IBAs, seven bile acids could be quantitatively detected in all infants (Table 2). Compared to controls, levels of CA and CDCA were significantly lower, while levels of GCA, GCDCA, TCA, and TCDCA were significantly higher in BA and non-BA infants. Differences in IBAs were also found between BA and non-BA infants. CDCA levels were significantly lower in the BA group, while GCA and GCDCA levels were significantly higher in the BA group than in the non-BA group.
GCDCA is generated by glycine conjugation of CDCA in the liver. Because there were higher GCDCA levels and lower CDCA levels in BA, we used the ratio of GCDCA/CDCA to compare BA infants with non-BA infants. The ratio of GCDCA/CDCA was significantly higher in BA infants than in non-BA infants (p< 0.05). The median ratio of GCDCA/CDCA was 685 (range, 394–1288) in BA infants and was 266 (range, 100–596) in non-BA infants (Table 2).
The variables that were statistically significant with p<0.01 in the univariate analysis were included into the multivariate logistic regression analysis by stepwise selection. The final multivariable model included (1) γ-GT, (2) GCDCA/CDCA ratio, and (3) clay stool.
In the derivation cohort, using a ROC curve, the diagnostic performance of the three selected variables based on the occurrence of BA was evaluated individually and compositely. A combination of these three parameters was proven to be significantly related to the identification of BA compared with each parameter (p<0.05) (Figure 1).
Accordingly, a formula with the three aforementioned variables was developed by stepwise algorithms for discriminating patients with BA from those with infantile cholestasis. The probability of BA = exp (-2.4672 + 0.1377 × γ-GT + 0.0319 × GCDCA/CDCA + 1.5779 × clay stool). It was obvious that the function is complicated and difficult for clinical application; therefore, a composite score system was then established for easy prediction (Table 3). Of note, because the IBA value varied depending on the instruments or reagents used,[12, 21] we standardized the serum bile acid value using the multiple of the median (MoM) value, which is defined as the ratio of the actual measured value over the normal median value of IBA (Supplemental Table S1). Thus, the GCDCA/CDCA ratio could be practically used in any institution and hospital where serum bile acid profiles are measured. Similarly, the scoring system also contained MoM values of γ-GT to eliminate the different values in various laboratories. MoM values of GCDCA/CDDCA and γ-GT could replace the original value in the formula above, by which we could figure out the probability of BA.
The BA score system derived from the multivariable model (score range, 0 to 41) linearly corresponded to the risk estimate. A ROC curve analysis was applied to evaluate the diagnostic efficacy of the score system. A cutoff point was selected to stratify BA risk (low risk, ≤15 points; high risk, >15 points; Supplemental Table S2). The AUC of the scoring system was 0.87 (95% CI, 0.77–0.94). A scatter plot showed the diagnostic sensitivity of 85.3%, and specificity of 81.3% with a cutoff point of 15 (Figure 2).
To verify the applicability of the proposed scoring system, a validation cohort of infants with cholestatic liver diseases was tested, including BA (n = 40) and non-BA (n = 35). The diagnostic sensitivity and specificity were 90.0% and 80.0%, respectively (Figure 2).
The validation test characteristics for all point values are shown in Supplemental Table S3, in which the performance of the three-variable scoring system in individual infants was compared to the final confirmed diagnosis in the validation cohort with a cutoff point of 15. In the BA group, 4/40 (10%) infants were misdiagnosed, while 11 of all infants were misdiagnosed based on this scoring system, for an accuracy of 85.3%.