Development and validation of a clinical data-based diagnostic model for neonatal intrahepatic cholestasis caused by citrin deficiency

Abstract

BACKGROUND

Neonatal intrahepatic cholestasis caused by citrin deficiency (NICCD) is a leading cause of infantile chronic liver disease, with the potential to progress to severe complications such as cirrhosis or liver failure. Therefore, early diagnosis and timely intervention are critical. Due to the limitations of newborn screening and genetic testing, novel diagnostic models are needed to improve the early diagnosis of NICCD.

AIM

To develop and validate a diagnostic model using clinical data to facilitate the early diagnosis of NICCD.

METHODS

This retrospective study developed and validated an early diagnostic model for NICCD using data from Hunan Children’s Hospital (2014-2025), including 170 patients with NICCD (NICCD group) and 196 patients with cholestatic hepatitis attributed to other etiologies (control group). Binary logistic regression was employed to construct the combined model. Model performance was evaluated using receiver operating characteristic (ROC) curves and the Hosmer-Lemeshow test, with internal validation performed via bootstrapping. The optimal threshold was established using Youden’s index, with external validation conducted on 39 NICCD cases and 59 controls from First People’s Hospital of Chenzhou.

RESULTS

This study identified 22 differential indicators, with age, albumin (ALB), lactate, and international normalized ratio (INR) selected via univariate and multivariate logistic regression. A joint binary logistic model was developed: Logit (p) = 1.518 + 1.427 × age - 0.423 × ALB + 0.596 × lactate + 6.639 × INR. A cutoff of 0.51 yielded 0.95 sensitivity and 0.97 specificity for NICCD diagnosis. The model achieved an area under the ROC (AUROC) of 0.99 [95% confidence interval (CI): 0.99-1.00], with a Hosmer-Lemeshow goodness-of-fit χ² value of 1.834 (P = 0.986), indicating good discrimination and calibration. Bootstrap internal validation yielded an AUROC of 0.966 (95%CI: 0.948-0.982). The model achieved an AUROC of 0.90 (95%CI: 0.82-0.98) in the validation set.

CONCLUSION

This study developed and validated a high-performing diagnostic model for NICCD using readily available clinical data.

Key Words: Clinical data; Diagnostic model; Neonatal intrahepatic cholestasis caused by citrin deficiency; Binary logistic regression; Early diagnosis

Core Tip: This study developed a novel diagnostic model for citrin deficiency-induced neonatal intrahepatic cholestasis based on clinical variables, representing a new approach in the literature. The model’s excellent diagnostic performance suggests that it could significantly assist in clinical decision-making.

INTRODUCTION

Citrin deficiency (CD) is an autosomal recessive metabolic disorder featuring three age-dependent stages: Neonatal intrahepatic cholestasis caused by CD (NICCD), failure to thrive and dyslipidemia caused by CD (FTTDCD), and adolescent and adult CD (AACD) - formerly known as adult-onset type II citrullinemia[1]. Since its initial documentation in 2006, this disease has been highly prevalent, with China reporting the highest carrier rate of 1:45[2]. NICCD is currently a major cause of cholestatic hepatitis in Asia[3-5]. Patients with NICCD typically present with cholestasis, elevated liver enzymes, hypercitrullinemia, hyperammonemia, and fatty liver disease. Although spontaneous resolution of these clinical conditions is common by age 1 or 2, a subset of patients continue to experience symptoms of FTTDCD. These ongoing clinical issues include dyslipidemia, recurrent hypoglycemia, growth retardation, pancreatitis-related abdominal pain, and fatigue, which persist beyond the typical resolution of NICCD. Delayed diagnosis in some patients can lead to cirrhosis and liver failure, ultimately requiring a liver transplant. Studies have shown that fewer than 20% of patients with CD progress to AACD, a serious and life-threatening metabolic decompensation that typically manifests in adolescence or adulthood[6]. In the majority of NICCD cases, clinical manifestations can be rapidly ameliorated through dietary management with lactose-free and medium-chain triglyceride-fortified formula. Therefore, early diagnosis coupled with prompt treatment are essential to improve patient outcomes, prevent progression to cirrhosis or liver failure, and reduce the risk of developing AACD[7].

Although newborn screening provides the earliest opportunity to detect CD[3,8], it suffers from low sensitivity[1,3,9]. Research findings have revealed that tandem mass spectrometry, when used for combined evaluation, significantly improves the detection of newborns who subsequently develop NICCD[10-12]. However, prompt diagnosis remains elusive for some cases. Suzuki et al[13] reported that serum blood urea nitrogen (BUN) or the BUN/creatinine ratio may facilitate the early detection of CD in newborn screening. While serum BUN, and particularly the BUN/creatinine ratio (to postnatal day 60 or 100), showed strong positive likelihood ratios at the initial visit, the findings were constrained by low sensitivity (< 80%) and limitations of a single-center design with a small sample size. Diagnostic confirmation requires identifying biallelic carrier family 25 member 13 (SLC25A13) pathogenic variants[3-5]; however, approximately 74% of clinically diagnosed children with CD show this genetic confirmation[1]; some children present with only a single disease mutation or no pathogenic SLC25A13 mutation at all. Moreover, genetic testing is often restricted by high costs and limited accessibility. At the same time, establishing definitive clinical or biochemical criteria for CD is challenging, as its symptoms are heterogenous and overlap with other hepatic and metabolic disorders[14]. This complexity hinders early clinical diagnosis.

Based on the above, the early diagnosis of NICCD can greatly improve prognosis. Given the current limitations in the early diagnosis of NICCD, the aim of this study was to establish a novel, simple, and effective predictive model using multivariate analysis, providing clinicians with a more effective tool for early identification.

MATERIALS AND METHODS

This retrospective study developed and validated a novel diagnostic model for early NICCD detection. It was performed at Hunan Children’s Hospital (Hunan Province, China). Consecutive hospitalized patients with confirmed NICCD were enrolled between January 1, 2014 and December 31, 2025. During the same period, 196 patients with cholestatic hepatitis attributed to other causes were designated the control group. The external validation set included another cohort comprising 39 patients with NICCD and 59 control patients from the First People’s Hospital of Chenzhou (Chenzhou, Hunan Province, China). A retrospective review of medical records was conducted by two independent physicians to ensure the inclusion of all consecutive hospitalized patients with NICCD. Any discrepancies were resolved by consulting a senior physician. The diagnosis was confirmed if at least two of the following criteria were met: (1) Clinical features; (2) Blood tandem mass spectrometry results; and (3) Identification of SLC25A13 mutations by molecular diagnosis. Alternatively, the diagnosis was confirmed if only criterion (3) was met, provided it was a previously reported homozygous or heterozygous mutation. The exclusion criteria were as follows: (1) Co-existing viral infections; (2) Biliary atresia; (3) Hereditary diseases such as familial intrahepatic cholestasis; (4) Severe dysfunction of other organs (e.g., kidney, brain, heart) and other organs that were not fully functional; or (5) Missing data.

Data collection

The ethics committee of Hunan Children’s Hospital approved this study and waived the requirement for written informed consent. Baseline clinical data, including demographics (age, sex) and a full profile of biochemical indicators [routine blood tests, liver and kidney function tests, myocardial enzymes, lipid profile, coagulation, and metabolic levels (glucose, lactic acid, procalcitonin, and ammonia)] were recorded for all patients prior to treatment.

Statistical analysis

All statistical analyses were performed using SPSS 25.0 (IBM Corp., Armonk, NY, United States) and R software 4.0.2 (R Foundation for Statistical Computing, Vienna, Austria). Differences in indicators between groups were assessed using independent samples t-tests and χ² tests. Univariate logistic regression analysis was performed to screen for potential risk factors related to NICCD. Statistically significant factors were entered into a regression model to identify independent factors. Subsequently, statistically significant factors were entered into a binary logistic regression model to establish a combined predictive model. We evaluated the model’s discrimination using receiver operating characteristic (ROC) curves and calibration via the Hosmer-Lemeshow test, followed by internal validation using bootstrapping. The optimal classification threshold was identified using Youden’s index on the training data set, whereas external validation was performed using ROC curves. P < 0.05 was considered statistically significant.

RESULTS

Clinical features and demographic characteristics at baseline and selection of relevant clinical factors in NICCD

The NICCD group had a median age of 2.47 months and included 105 males and 65 females, whereas the control group had a median age of 1.98 months and included 132 males and 64 females. There was no statistically significant difference in sex distribution between the two groups (P = 0.265). Independent samples t-tests demonstrated significant differences between the NICCD and control groups across multiple clinical parameters (Table 1). These markers included age (months) and a broad range of biochemical indicators: White blood cells, neutrophils, lymphocytes, hemoglobin, platelets, C-reactive protein, bilirubin components (total bilirubin, direct bilirubin, indirect bilirubin), protein levels [albumin (ALB), globulin], liver enzymes [alanine aminotransferase (ALT), aspartate aminotransferase (AST), AST/ALT ratio], total bile acid, lactate, ammonia, and coagulation metrics [prothrombin time (PT), activated partial thromboplastin time (APTT), international normalized ratio (INR), and fibrinogen level]. As shown in Table 2, multivariate regression analysis indicated that age [odds ratio (OR) = 9.743; P < 0.001], ALB level (OR = 0.75; P = 0.005), lactate level (OR = 1.573; P = 0.031), and the INR (OR = 167.614; P = 0.011) were independent factors of NICCD. Next, we conducted thorough collinearity diagnostics on all predictor variables included in the multivariate logistic regression model. All independent variables showed variance inflation factors (commonly referred to as VIFs) below 5, with a mean VIF of 2.12. The results indicated that the variables in our model did not exhibit significant multicollinearity.

Table 1 Comparison of baseline clinical features and demographic characteristics between the neonatal intrahepatic cholestasis caused by citrin deficiency group and the control group.

Parameter	Reference interval	NICCD group, n = 170	Control group, n = 196	T	P value
Age in months		2.47 ± 1.26	1.98 ± 0.70	4.69	0.000
WBC (× 109)	4.3-14.2	13.10 ± 4.14	10.54 ± 3.14	6.72	0.000
N (× 109)	0.6-7.5	4.10 ± 1.96	2.54 ± 1.31	9.02	0.000
L (× 109)	2.4-9.5	7.57 ± 2.99	6.73 ± 2.41	3.00	0.003
HB (g/L)	97-183	100.20 ± 16.91	104.4 ± 17.95	-2.27	0.024
PLT (× 109)	183-614 × 109	501.40 ± 173.50	420.50 ± 135.97	5.00	0.000
CRP (mg/L)	0-5	2.29 ± 2.65	1.31 ± 1.77	4.21	0.000
TBIL (mol/L)	3-20	163.90 ± 65.47	137.20 ± 87.32	3.28	0.001
DBIL (mol/L)	1.7-6.8	78.82 ± 39.05	91.66 ± 55.48	-2.52	0.012
IBIL (mol/L)	1.3-18.3	85.16 ± 45.79	41.62 ± 31.77	10.67	0.000
ALB (g/L)	35-50	30.71 ± 4.67	38.02 ± 3.27	-17.51	0.000
GLO (g/L)	9-27	16.37 ± 4.03	18.62 ± 9.03	-2.988	0.003
ALT (U/L)	8-71	63.18 ± 97.61	187.00 ± 207.65	-7.12	0.000
AST (U/L)	21-80	129.20 ± 115.20	203.20 ± 182.28	-4.56	0.000
ALT/AST		2.60 ± 1.34	1.36 ± 0.77	10.97	0.000
TBAC (mol/L)	0.5-10	180.80 ± 67.36	125.40 ± 59.75	8.35	0.000
GGT (U/L)	9-150	186.10 ± 73.35	198.40 ± 33.66	-0.47	0.642
GLU (mmol/L)	3.89-6.4	4.87 ± 8.94	4.66 ± 1.12	0.33	0.745
TG (mmol/L)	0.33-1.82	1.65 ± 0.76	1.63 ± 0.72	0.27	0.789
TCHO (mmol/L)	0-5.2	4.36 ± 1.32	4.12 ± 2.75	1.06	0.292
LACT (mmol/L)	0.5-2.44	4.58 ± 1.65	3.69 ± 1.44	5.51	0.000
NH4 (mol/L)	10-47	76.53 ± 44.69	51.54 ± 26.04	6.63	0.000
PT (seconds)	10-14	17.29 ± 8.43	13.27 ± 2.87	6.27	0.000
APTT	28-48	49.42 ± 12.26	43.54 ± 5.04	6.14	0.000
INR	0.8-1.5	1.45 ± 1.21	1.00 ± 0.11	5.27	0.000
FIB (mg/dL)	170-450	135.00 ± 54.16	288.80 ± 70.80	-23.06	0.000

NICCD: Neonatal intrahepatic cholestasis caused by citrin deficiency; ALB: Albumin; ALT: Alanine aminotransferase; APTT: Activated partial thromboplastin time; AST: Aspartate aminotransferase; CRP: C-reactive protein; DBIL: Direct bilirubin; FIB: Fibrinogen; GGT: Gamma-glutamyl transferase; GLO: Globulin; GLU: Glucose; HB: Hemoglobin; IBIL: Indirect bilirubin; L: Lymphocytes; INR: International normalized ratio; LACT: Lactate; N: Neutrophils; NH4: Blood ammonia; PLT: Platelets; PT: Prothrombin time; TBAC: Total bile acids; TBIL: Total bilirubin; TCHO: Total cholesterol; TG: Triglycerides; WBC: White blood cell.

Table 2 Logistic regression analyses of predictors of neonatal intrahepatic cholestasis caused by citrin deficiency.

Characteristic	Univariate analysis		Multivariate analysis
	OR (95%CI)	P value	OR (95%CI)	P value
Age	1.73 (1.34-2.23)	0.000	9.74 (3.56-26.69)	0.000
WBC	1.22 (1.15-1.31)	0.000	2.23 (0.56-8.95)	0.258
N	1.96 (1.64-2.35)	0.000	0.42 (0.08-2.16)	0.299
L	1.12 (1.04-1.22)	0.004	0.49 (0.11-2.19)	0.347
HB	0.99 (0.97-1.00)	0.026	1.01 (0.97-1.05)	0.727
PLT	1.00 (1.00-1.01)	0.000	1.00 (1.00-1.01)	0.352
CRP	1.31 (1.15-1.50)	0.000	1.09 (0.77-1.53)	0.631
TBIL	1.01 (1.00-1.01)	0.002	0.92 (0.61-1.39)	0.696
DBIL	0.99 (0.99-1.00)	0.014	1.06 (0.70-1.61)	0.782
IBIL	1.03 (1.02-1.04)	0.000	1.11 (0.73-1.69)	0.615
ALB	0.65 (0.59-0.70)	0.000	0.75 (0.61-0.92)	0.005
GLO	0.90 (0.85-0.96)	0.001	0.99 (0.90-1.10)	0.894
ALT	0.99 (0.98-0.99)	0.000	1.00 (0.99-1.01)	0.494
AST	1.00 (0.99-1.00)	0.000	1.00 (1.00-1.01)	0.641
AST/ALT	4.15 (2.96-5.80)	0.000	1.85 (0.76-4.47)	0.173
TBAC	1.01 (1.01-1.02)	0.000	0.99 (0.98-1.00)	0.097
LCAT	1.47 (1.27-1.70)	0.000	1.57 (1.04-2.37)	0.031
NH4	1.02 (1.02-1.03)	0.000	1.02 (01.00-1.04)	0.07
PT	2.22 (1.86-2.65)	0.000	1.08 (0.94-1.23)	0.291
APTT	1.13 (1.08-1.17)	0.000	0.98 (0.87-1.11)	0.773
INR	33437.21 (3651.70-306172.16)	0.000	167.61 (3.28-8557.58)	0.011
FIB	0.97 (0.96-0.97)	0.000	0.98 (0.97-0.99)	0.001

ALB: Albumin; ALT: Alanine aminotransferase; APTT: Activated partial thromboplastin time; AST: Aspartate aminotransferase; CI: Confidence interval; CRP: C-reactive protein; DBIL: Direct bilirubin; FIB: Fibrinogen; GGT: Gamma-glutamyl transferase; GLO: Globulin; GLU: Glucose; HB: Hemoglobin; IBIL: Indirect bilirubin; INR: International normalized ratio; L: Lymphocytes; LACT: Lactate; N: Neutrophils; NH₄: Blood ammonia; OR: Odds ratio; PLT: Platelet; PT: Prothrombin time; TBAC: Total bile acids; TBIL: Total bilirubin; TCHO: Total cholesterol; TG: Triglycerides; WBC: White blood cell.

Establishment of clinical diagnostic models of NICCD

Utilizing the independent risk factors previously identified, a combined model was established through further binary logistic regression analysis. The NICCD disease prediction model was formulated using the constant and the regression coefficients for each independent variable: Logit (p) = 1.518 + 1.427 × age - 0.423 × ALB + 0.596 × lactate + 6.639 × INR. For the training data set, the model had an area under the ROC curve (AUROC) of 0.99 [95% confidence interval (CI): 0.99-1.00]. At a threshold greater than 0.51, the sensitivity was 0.95, and the specificity was 0.97 (Figure 1A).

Figure 1 Area under the curve for predicting neonatal intrahepatic cholestasis caused by citrin deficiency. A: Training set; B: Validation set. The area under the curve of the new model in predicting neonatal intrahepatic cholestasis caused by citrin deficiency (NICCD) patients was 0.99 in the modeling set. The area under the curve of the new model in predicting NICCD patients was 0.9 in the validation set. The receiver operating characteristic curve demonstrates that the model had good diagnostic value for the training set and the validation set. CI: Confidence interval; AUC: Area under the curve.

Model evaluation and verification

The Hosmer-Lemeshow test yielded a χ² value of 1.834, with a significance level of 0.986; this was further confirmed by the calibration curve’s close alignment with the ideal line. Overall, the model showed strong discrimination and calibration, indicating excellent diagnostic performance for NICCD (Figure 2).

Figure 2 Calibration curve graph for calibration degree assessment. The calibration curve demonstrates that the model has good calibration.

The bootstrap ROC analysis yielded an area under the curve of 0.966, standard error of 0.009, and 95%CI: 0.948-0.982. These results suggest that the model is robust against individual case variations and holds strong diagnostic value (Figure 3). The model achieved an AUROC of 0.90 (95%CI: 0.82-0.98) on the validation set (Figure 1B).

Figure 3 Receiver operating characteristic curve for evaluating diagnostic performance. The bootstrap receiver operating characteristic curve demonstrates that the model had good diagnostic value. ROC: Receiver operating characteristic.

DISCUSSION

This study identified 22 distinct clinical indicators in patients with NICCD, of which age, ALB, lactate, and INR were identified as independent risk factors. The binary logistic regression models showed good discrimination and calibration. Internal and external bootstrap validation confirmed high diagnostic accuracy.

Consistent with the literature, primary manifestations included metabolic abnormalities such as progressive cholestasis, elevated transaminases, hyperammonemia, hyperlactatemia, hyperlipidemia, hypoproteinemia, and prolonged PT[1]. Feeding infants with NICCD a diet of medium-chain fatty acids and a lactose‐free formula can normalize abnormal biochemical indicators. When introduced to solid foods, patients show a clear preference for high-fat options and consume very few carbohydrates. Despite outward health, patients with NICCD remain at risk for FTTDCD and AACD, making timely clinical management the cornerstone of a favorable prognosis. Therefore, annual follow-ups for children with NICCD are essential to assess growth and development, neurocognitive function, and biochemistry. Regular imaging and timely adjustments to nutritional therapy are necessary to reduce the risk of AACD and improve patient outcomes.

Children with NICDD exhibited higher age, ALB levels, lactic acid levels, and INR compared to children with cholestatic hepatitis of other causes. In this study, the median age at diagnosis for children with NICCD was 2.47 months. ALB levels were significantly lower in children with NICCD children compared with controls (30.71 ± 4.67 vs 38.02 ± 3.27). As the earliest abnormal biochemical marker, ALB represents a potential indicator for early diagnosis[15]. Both liver damage and delayed maturation of liver function can lead to hypoproteinemia in children with NICCD[16]. Given that the median ALB concentration is approximately 30 g/L[15,16], an ALB level < 30 g/L acts as a key indicator for potential NICCD[17].

Citrin is primarily expressed on the inner mitochondrial membrane of hepatocytes and acts as an aspartate/glutamate carrier. It facilitates a key transport step for the malate-aspartate shuttle by exporting aspartate into the cytoplasm while simultaneously importing glutamate and protons into the mitochondrial matrix. Under physiological conditions, glycolysis in the cytoplasm produces nicotinamide adenine dinucleotide (NADH). Because this NADH cannot directly enter the mitochondria, its reducing equivalents are transported via a malate shuttle, subsequently generating adenosine triphosphate through the tricarboxylic acid cycle and oxidative phosphorylation. This biochemical process pairs with citrin-mediated aspartate transport to establish the malate/aspartate shuttle. Thus, citrin plays an indispensable and critical role in hepatic energy production[18]. Typically, mitochondria oxidize NADH back to NAD⁺ to sustain continuous glycolysis. In cases of CD, accumulated NADH in the cytoplasm promotes the conversion of pyruvate into lactate to regenerate NAD⁺, ensuring the continuation of glycolysis. Furthermore, because aspartate cannot be transported into the cytoplasm, there is less oxaloacetate for gluconeogenesis. This prevents lactate from being converted into glucose, resulting in lactate accumulation. Consistent with the literature, the affected child in our case also had elevated lactate levels (4.58 ± 1.65 vs 3.69 ± 1.44). Hyperlactatemia may exacerbate metabolic disturbances in children, aggravate liver function impairment, and hinder growth and development. Restricting galactose intake and supplementing with a special formula rich in medium-chain triglycerides can rapidly correct metabolic disorders in infants with NICCD, improving their clinical condition.

These patients frequently exhibit a prolonged PT that typically improves with vitamin K supplementation[3,19-21]. Chen et al[4] reported that the PT with INR [1.43 (1.01-2.34) vs 1.03 (0.88-1.30)] increased in children with NICCD. Accordingly, we also found that children with NICCD had abnormal coagulation function. In particular, the INR (1.45 ± 1.21 vs 1.00 ± 0.11) significantly increased. Prompt dietary management facilitates the rapid recovery of coagulation profiles (thrombin time, PT, and APTT)[15], eliminating the need for plasma transfusion or cryoprecipitation. Delayed diagnosis, however, correlates with higher transfusion requirements and adverse clinical outcomes[1].

Based on a cohort of 20 Taiwanese patients diagnosed with NICCD, Chen et al[4] developed a 6-point scoring system to distinguish NICCD from biliary atresia and other infantile cholestatic liver diseases. The scoring parameters included a direct bilirubin/total bilirubin ratio < 0.67, AST/ALT > 2, gamma gamma-glutamyl transferase concentration < 300 U/L, PT with INR > 1.3, and alkaline phosphatase/ALT > 10. The majority of patients with NICCD had a score ≥ 4 points, whereas those with biliary atresia or other causes of infantile cholestasis scored < 4 points. However, due to limitations and a small sample size, its use in clinical practice remains limited. An early diagnosis model for NICCD was developed using binary logistic regression analysis: Logit (p) = 1.518 + 1.427 × age - 0.423 × ALB + 0.596 × lactate + 6.639 × INR. Using a 0.51 threshold, the training set results showed 0.95 sensitivity and 0.97 specificity. The model demonstrated good diagnostic performance, confirmed through robust calibration, discrimination, and validation procedures. Because it relies on low-cost, routinely collected data, this study’s diagnostic model is simple to implement and holds significant potential for widespread clinical application without added expense.

Three main limitations merit consideration. Primarily, the small sample size may have resulted in an overestimation of model performance. Therefore, prospective, multicenter studies with larger cohorts are required to validate our findings. Second, because this study was conducted in a provincial tertiary hospital - which often handles more severe cases - the current sample may not be representative of patients managed in primary healthcare settings, introducing potential selection bias. Third, the model’s generalizability is constrained by its single-center validation. Future research should focus on validating the model across diverse geographical regions and varied healthcare settings.

CONCLUSION

In summary, we developed and validated a simple, novel tool for the diagnosis of NICCD. Utilizing four routine variables - age, ALB, lactic acid, and the INR - this model performs robustly in children with NICCD. Its simplicity allows for straightforward clinical application, enabling early identification and timely intervention for patients with NICCD.

ACKNOWLEDGEMENTS

We thank all of the participants in this study.