Design and validation of a noninvasive diagnostic criteria for biliary atresia in infants based on the STROBE compliant.

It is difficult for clinicians to distinguish biliary atresia (BA) from other causes of neonatal cholestasis (NC) at an early stage. The aim of this study was to design and validate noninvasive diagnostic criterion for early diagnosis of BA in infants.In this retrospective cohort study, a total of 482 medical records of patients with NC were recruited to design diagnostic criteria. Parameters showing a significant difference between BA (n = 166) and non-BA (n = 316) patients were analyzed by logistic regression to predict the occurrence of BA, and then a nomogram scoring system was designed and validated in another cohort that included 190 cases of NC.A prediction diagnostic criterion with parameters including direct bilirubin, total bilirubin, globulin, albumin, gamma glutamyl transpeptidase, cholesterol, total bile acid, hepatobiliary scintigraphy, birth weight, and stool color was established; the sensitivity and specificity of this diagnostic criterion was 89% and 84%, respectively. The accuracy was 86% and the AUC was 0.91 [95% CI (0.88-0.97)]. The total score ranged from 0 to 402, with a cut-off value of ≥254 discriminating BA from other causes of NC. By applying this score in the validation set with age <60 days, the accuracy was 95.3%, the sensitivity was 93.8% and the specificity was 96.0%, respectively.This prediction diagnostic criterion could facilitate clinicians to distinguish infants with and without BA based on a particular series of parameters, reducing treatment burden and enhancing therapeutic efficiency.

Introduction

Cholestasis is a clinical condition characterized by decreased flow of canalicular bile and direct hyperbilirubinemia.[ Cholestasis in early life arises from a hepatic response to exogenous agents or a specific congenital pathology. In newborns and infants, biliary atresia (BA) is still a challenge for clinicians and researchers.[ Diagnosis of BA at an early stage is of great clinical importance, while misdiagnosis may result in treatment failure and even death.[ In contrast, a significant proportion of neonatal cholestasis (NC) cases do not have BA, and there is unnecessary laparotomy for them. Therefore, distinguishing BA from other causes of NC without intraoperative cholangiography (IOC) would thus be beneficial both for the patients and medical insurance management. Clay-colored stool is considered an indicator and has been used for screening; multiple ultrasonography (US) parameters have also been reported to be examined in a preoperative diagnosis of BA, along with many biochemical and histopathological parameters that may be used to distinguish BA.[ However, there is rarely available noninvasive criterion integrating histopathological observation, biochemical indicators, clinical examination, and imaging parameters for clinicians in the diagnosis of BA.[ The aim of the current study was to design and validate a noninvasive diagnostic prediction criterion combining graphical, biochemical and clinical examination for the early discrimination of BA in infants.

Methods

Subjects

The protocol of this study was reviewed and approved by the National Legislation and Ethical Committee of Xi’an Children's Hospital, China. Guardians of all infants in this study gave informed consent. The records of 2 consecutive cohorts of infants with NC recruited from the Pediatric Hepatology Department, Xian Children's Hospital, China, between January 2011 and June 2016 were examined. Babies with Gilbert syndrome, born prematurely, sepsis, or those receiving total parenteral nutrition were excluded from the study. For the neonates and infants, age, gender, birth weight, birth week, color of stool and urine, serological indicators and imaging examination findings were recorded. The first cohort (design set) included 482 infants, with age<90 days: 166 infants with BA, median age 56.5 days, and 316 infants with cholestasis due to causes other than BA (termed as non-BA), median age 59.5 days. The second cohort (validation set) included 190 infants, with age<60 days (Table 4), and consisted of 64 infants with BA and 126 infants with non-BA. Cholestasis was diagnosis by considering direct bilirubin (DBIL) of more than 20% of the total bilirubin (TBIL), an increased serum bile acid (SBA) concentration, and other biochemical indicators and clinical symptoms.[ The diagnosis of BA was confirmed by laparotomy with or without IOC. Cases without BA at IOC and those with other causes of cholestasis were included in the non-BA group.

Table 4

Performance of the total score in individual infants compared to the final confirmed diagnosis in the validation set (infants with age < 60 days).

Collection of parameters

In addition to general demographic information, routine investigations included complete blood count, total and direct bilirubin (TBIL, DBIL), globulin (GLB), albumin (ALB), alanine transaminase (ALT), aspartate transaminase (AST), alkaline phosphatase (ALP), gamma glutamyl transpeptidase (GGT), cholesterol (CHO), total bile acid (TBA); because of there is a group of BA infants who may be initiated or caused by CMV, these will be cast into the non-BA group, thus, immunoglobulin (Ig)M and DNA concentration for both blood and urine CMV-DNA were detected and calculated.[ Prothrombin time (PT), activated partial prothrombin time (APTT), fibrinogen (FIB) were also detected. The classification of hepatomegaly was based on the size of the liver beyond the edge of the right costal arch (< 3 cm, 3–5 cm, or > 5 cm) when performing a physical examination. Hepatobiliary scintigraphy by emission computed tomography (ECT) assay. The levels of hepatobiliary scintigraphy were: normal, delayed and no. All cases of BA in the validation set were confirmed by laparotomy with or without IOC; non-BA cases were defined as either excluding the possibility of BA by IOC or confirming the etiology.

Statistical analysis

Descriptive results are expressed as mean ± standard deviation (quantitative data with normal distribution), median (Q1, Q3) (quantitative data with non-normal distribution) or number (percentage) (qualitative data) of individuals with a condition. For quantitative data, comparison between 2 groups was made by Student's t-test or Mann–Whitney U test. For qualitative data, significance between groups was tested by chi-square test or Fisher‘s exact test. A P-value of <.05 was regarded as significant.

Relative risk and odds ratio were calculated for variables with a significant difference between groups. Binary logistic regression analysis was performed for each variable and represented in the score by the corresponding raw regression coefficient. The diagnostic performance was expressed as sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), and accuracy. The cut-off values for optimal clinical performance (best sensitivity and best specificity simultaneously) of individual parameters and the overall accuracy of the prediction diagnostic criterion were determined from the receiver-operating characteristic (ROC) curve. A bootstrap procedure was used to compute the standard error for the difference in the area under the curve for ROCs for any 2 criteria. All analyses were performed using the R (http://www.R-project.org) and EmpowerStats software (www.empowerstats.com, X&Y solutions, Inc., Boston, MA).

Results

Participants in the design set

In the first cohort, 166 patients with BA and 316 cases with non-BA were included. There was no difference between the 2 groups in age at diagnosis. Birth weight and gestational age were higher in the BA group. The frequency of clay stool, dark yellow urine, hepatobiliary scintigraphy delay (including contrast agent delay or does not develop) and the median level of TBIL, DBIL, ALT, AST, GGT, TBA; and mean levels of ALB, GLB, CHO, PLT, and PT were significantly higher in the BA than in the non-BA group, while the frequency of hepatomegaly. urine CMV-DNA positive, and mean level of ALP was significantly higher in the non-BA group compared with the BA group. Moreover, female gender was dominant in the BA group than in the non-BA group. There were no significant differences in the other studied clinical and laboratory parameters. The demographic and clinical characteristics of the infants are summarized in Table 1.

Table 1

Baseline characteristics of the whole population (N = 482).

Diagnostic performance of individual parameters and prediction criteria development

The diagnostic performance of individual parameters was evaluated by logistic regression, and 2 prediction criteria were built. Criterion 1 included all parameters with significant statistical differences between the BA group and non-BA group: gender, hepatomegaly, stool color, urine color, urine CMV-DNA, hepatobiliary scintigraphy, gestational age, birth weight, TBIL, DBIL, ALB, GLB, ALT, AST, ALP, GGT, CHO, TBA, PLT and PT% (see Table S1, Supplemental Content, which illustrates the criteria 1 with all parameters); Criterion 2 was based on selected parameters in order to remove the redundant variables, and thus included ten variables: birth weight, stool color, TBIL, DBIL, ALB, GLB, GGT, CHO, TBA, and hepatobiliary scintigraphy (Table 2). There was no difference between the 2 criteria in the AUC, specificity, sensitivity, accuracy, PPV, or NPV (see Table S2, Supplemental Content, which illustrates the diagnostic performance between criterion 1 and criterion 2). The best performance was stool color, with a sensitivity of 93.9% followed by DBIL (90.9%), birth weight (90.8%), GGT (87.1%), hepatobiliary scintigraphy (84.3%), CHO (83.5%), TBA (83.0%), TBIL (81.2%), ALB (54.0%), and GLB (30.7%). In the design set, for criterion 2 the AUC was [0.91 (95% CI 0.84–0.97)], specificity was 83.6%, sensitivity was 88.6%, accuracy was 85.6%, PPV was 77.5%, and NPV was 92.0% (Fig. 1).

Table 2

Diagnostic performance of selected parameters with significant statistical difference between BA and non-BA groups (criterion 2).

Figure 1

Receiver operating characteristic (ROC) curve for diagnostic prediction criterion of infants with biliary atresia (BA) in the design set.

Derivation of the diagnostic score system by a nomogram

Based on criterion 2, we established a diagnostic scoring system by nomogram according to the regression coefficient of selected parameters (Fig. 2). In this scoring system, in order to facilitate the clinical application, all the continuous variables were transferred into a categorical variable based on the cutoff point. Here, we use the approximate integer of the cutoff values instead of the raw data (cut off values mentioned in table S1). Thus, the corresponding cutoff value of Birth weight, TBIL, DBIL, ALB, GLB, GGT, CHO, and TBA were 2500 g, 150 μmol/L, 90 μmol/L, 40 g/L, 15 g/L, 180 U/L, 4 mmol/L and 110 μmol/L, respectively. According to Figure 2, for example, if one infant Birth weight ≥ 2500 g (52 points (see the “point” line)), normal stool color (0 points), GGT<180 U/L (0 points), ALB <40 g/L (15 points), TBA ≥110 μmol/L (45 points), CHO ≥4 mmol/L (0 points), DBIL ≥90 μmol/L (28 points), GLB ≥15 g/L (0 point), hepatobiliary scintigraphy normal (0 point), TBIL ≥150 μmol/L (5 points), and had a total score of 238 points (see the “Total point” line) corresponding to a risk of about 35% (see the “diagnosis” line) for being diagnosed as BA.

Figure 2

Nomogram for diagnostic prediction criterion of infants with biliary atresia (BA). Nomogram of the prediction criterion for the whole study population. The risk value of BA was calculated by drawing a vertical line to the point on the axis for each of the factors. The points for each factor were summed and located on the total point line. Then, the bottom line corresponding vertically to the above total line illustrated the individual predictive value for BA. For example, if one infant birth weight ≥ 2500 g (52 points (see the “point” line)), normal stool color (0 points), GGT<180 U/L (0 points), ALB <40 g/L (15 points), TBA ≥110 μmol/L (45 points), CHO ≥4 mmol/L (0 points), DBIL ≥90 μmol/L (28 points), GLB ≥15 g/L (0 point), hepatobiliary scintigraphy normal (0 point), TBIL ≥150 μmol/L (5 points), and had a total score of 238 points (see the “Total point” line) corresponding to a risk of about 35% (see the “diagnosis” line) for being diagnosed as BA.

All these parameters in the nomogram were then expressed in a score system. The total score ranged from 0 to 402 and at a cut-off value of 254 it could discriminate patients with BA in the design set with a sensitivity of 89%, and specificity of 84% (Table 3).

Table 3

BA diagnostic score.

Validation of the BA diagnostic prediction criteria

To verify the applicability of this prediction criterion, a second cohort of infants with NC was used. As early diagnosis of BA within 60 days after birth is correlated with good clinical outcomes, this validation group were performed by the records of patients whose age <60 days, including BA (n = 64) and other NC diseases (n = 126). Results showed that in the validation set,

A total of 93.8% of the patients with BA (60/64) and 5 patients with other infantile cholestatic liver diseases had a score ≥254. These patients were found to be wrongly diagnosed as BA by the score, and they were demonstrated to be non-BA by laparotomy; while another 4 patients with BA were wrongly diagnosed as BA by the score<254. The scoring system was able to discriminate BA with a sensitivity of 93.8%, a specificity of 96.0% and an overall diagnostic accuracy of 95.3%.

Discussion

The development of BA involving inflammatory and progressive fibrotic obstruction leads to end stage liver disease. Therefore, early diagnosis of BA in infants with NC is particularly important. However, no single diagnostic procedure is clearly superior in BA diagnosis. In the current study, we designed and validated a prediction criterion composed of demographic, clinical, laboratory, and radiological variables to discriminate between BA and other NC disorders. This BA diagnostic prediction criterion was able to accurately discriminate BA with a specificity of 96.0%, a sensitivity of 93.8% and an accuracy of 95.3% in the validation set.

No single biochemical test can be successfully used to differentiate BA and non-BA cases, and diagnosis can only be established by using all available methods.[ Our hospital is a reference center for infants with BA. Large numbers of cases of BA are referred to our hospital from the north-west regions of China. We have therefore seen large numbers of cases of biliary atresia over the past 5 years. In this retrospective cohort study, we used a total of 482 medical records collected from February 2013 to June 2016 as the first cohort to design the diagnostic prediction criterion, and used another 190 medical records collected from January 2011 to January 2013 as a second cohort for criterion validation. We did not exclude patients older than 60 days for that older patients with BA do not always have a bad outcome.[ Because it is well known that the performance of Kasai portoenterostomy before the age of 60 days has a better outcome,[ we further validated the score system by a patients group with age ≤60 days.

Reports in the literature suggest that BA can be diagnosed at early stages using stool color cards.[ Stool color cards were used in Taiwan for mass screening of all live births for the presence of BA.[ Some studies have reported a specificity and sensitivity of acholic stool in diagnosing BA to be 55% to 86% and 72% to 100%, respectively.[ In our study, of all parameters included, the presence of acholic stool was one key feature suggestive of BA. Since the main purpose of this study was to help patients with BA received necessary intraoperative cholangiography and Kasai portoenterostomy in time, improving sensitivity was the highest priority. Thus, unlike previous studies, the current study divided the stool color into normal, light yellow and clay like, rather than separating it into clay-like and non-clay like, in order to improve the sensitivity of the diagnosis. Here, stool color discriminated BA with a specificity of 59.8%, a sensitivity of 93.9%, a PPV of 54.8% and an NPV of 95.0%.

Liver biochemical tests are important clinical tests to determine the presence of liver damage, treatment efficacies and prognoses of liver disease.[ Since the predominant effect is extrahepatic obstruction, TBIL may have an inordinately weighted contribution as a biochemical manifestation of cholestasis.[ Here, TBIL showed a specificity of 73.0%, a sensitivity of 81.2%, a PPV of 61.2%, an NPV of 88.1%, and an accuracy level of 75.8% in discriminating BA; and DBIL showed a specificity of 65.6%, a sensitivity of 90.9%, a PPV of 58.1%, an NPV of 93.2%, and an accuracy level of 74.3% in discriminating BA. Besides TBIL and DBIL, GGT has been reported as the most reliable parameter in discriminating BA and non-BA patients. One study reported an accuracy level of 85% for GGT levels above 300 U/L when BA presented before 10 weeks of age; another study reported a sensitivity of 83.1%, a specificity of 98.1%, and an accuracy level of 65.6% for γ-GT levels above 300 U/L.[ In this study, there was a statistically significant difference in the γ-GT levels between the BA and non-BA groups. We found that GGT at a cut-off value above 180 U/L exhibited a sensitivity of 87.1%, a specificity of 67.0%, a PPV of 57.7%, an NPV of 91.0%, and an accuracy level of 73.9% for differentiating BA and non-BA cases. Interestingly, we also found a specificity of 80.6% for GLB levels above 15 g/L, with a sensitivity of 30.7% and an accuracy level of 55.6%, as well as a specificity of 56.4% for ALB levels above 40 g/L, with a sensitivity of 54.0% and an accuracy level of 55.6%, in differentiating BA and non-BA cases. CHO and TBA were also contributed to the criterion with the sensitivity of 83.5% and of 83.0%, respectively.

Imaging techniques occupy an important place in the differential diagnosis of BA. Numerous US features have been described as useful pointers to the diagnosis of BA. Features such as GB contractility, GB length, and TC-sign have been described as helpful in the diagnosis of BA. In the current study, which was limited by the incomplete medical records used, only some of the US indicators were included in the diagnosis of BA. Among these, hepatomegaly was selected in the first prediction criterion (criterion 1) with a specificity of 51.6%, a sensitivity of 68.9%, a PPV of 42.5%, an NPV of 76.2%, and an accuracy level of 57.5% in discriminating BA. However, for its poor specificity and sensitivity, it failed to enter the final criterion. A further prospective study including the overall performance of US parameters (including the TC-sign, abnormal GB wall, shape, an absent common bile duct) should be performed.

Liver biopsy is an invasive procedure used to evaluate NC, but inadequate biopsy material may be obtained, especially via the percutaneous route.[ We do not consider it to be useful in the diagnosis of BA in NC, and excluded the parameters of liver biopsy from this study. Another reason for us did not brought the results from liver biopsy in the criterion is that in our country, it is difficult for the parents to allow their baby receive such an invasive test. Thus, in the current study, we use another noninvasive examination, hepatobiliary scintigraphy, a delayed hepatobiliary scintigraphy exhibited a specificity of 54.6%, a sensitivity of 84.3% and an accuracy level of 72.0%, in differentiating BA and non-BA cases.

Moreover, in the current study, it is interesting that among premature infants, the proportion of BA is smaller. This results is consistent with some reports that speculated BA may have a similar pathogenesis of infant hepatitis, it is acquired, but not congenital disease.[

One previous study used some parameters including ultrasonographic characters to built a scoring system which accurately separated infants with BA and those without, rendering intraoperative cholangiography for confirming or excluding BA in a substantial proportion of patients.[ Though the sensitivity and specificity is higher in that study, the patients enrolled in that study was only 135, and all the patients came from Egypt (an African country). This made that criterion may not be suitable for the Asians. Thus, it is necessary for us to explore a criterion that is more suitable for the Asian population.

The most important limitation of the criterion in predicting infants with BA lies that our study was a retrospective cohort study based on medical records, and therefore some indicators, especially US parameters, were missing and could not be analyzed in detail. Secondly, the sample size was still small, especially in the validation set. Based on these limitations, our results need to be validated in independent patient populations from other hospitals. And further studies are needed to determine the potential utility of our scoring system, which may need fine tuning.

In conclusion, by using the BA scoring system, we were able to clearly separate patients with BA from others with a relatively high accuracy (95.3%), sensitivity (93.8%) and specificity (96.0%) in the validation group whose age <60 days. Thus, patients with BA could receiving necessary intraoperative cholangiography and Kasai portoenterostomy in time.

Author contributions

Conceptualization: Xiaoguai Liu, Shuangsuo Dang.

Data curation: Xiaoguai Liu, Xiaokang Peng, Pan Liu, Shuangsuo Dang.

Formal analysis: Xiaoguai Liu, Yanxia Huang.

Investigation: Xiaoguai Liu, Xiaokang Peng, Yanxia Huang, Chang Shu, Pan Liu, Weike Xie, Shuangsuo Dang.

Methodology: Xiaoguai Liu, Xiaokang Peng, Yanxia Huang, Chang Shu, Pan Liu, Weike Xie, Shuangsuo Dang.

Supervision: Xiaoguai Liu, Yanxia Huang, Weike Xie, Shuangsuo Dang.

Writing – original draft: Xiaoguai Liu, Xiaokang Peng, Shuangsuo Dang.

Writing – review & editing: Xiaoguai Liu, Yanxia Huang, Chang Shu, Pan Liu, Weike Xie, Shuangsuo Dang.