Introduction

There are many causes of chronic liver disease in children and young adults, including autoimmune liver disease (AILD). AILD encompasses three separate clinical and histological entities, including primary sclerosing cholangitis (PSC), autoimmune hepatitis (AIH) and autoimmune sclerosing cholangitis (ASC, an overlapping condition with features of both sclerosing cholangitis and autoimmune hepatitis) [1, 2]. As with other chronic liver diseases, liver damage from chronic inflammation can result in myofibroblast activation, deposition of excess collagen and extracellular matrix, and the formation of fibrosis, or scar tissue [3]. Progressive liver fibrosis increases the resistance to blood flow through the liver and increases blood pressure in the portal venous system, ultimately causing portal hypertension, an important source of morbidity and mortality [4]. Portal hypertension can clinically and radiologically manifest as enlargement of the spleen (splenomegaly), development of portosystemic collateral vessels (varices), and ascites [5, 6].

There is increasing evidence that MRI can be used to detect, measure and follow chronic liver disease over time. Quantitative MRI techniques such as MR elastography, T1 mapping, T2 mapping and diffusion-weighted imaging (DWI) have been shown to correlate with the presence and degree of liver fibrosis to various degrees, although mainly in adult populations [7,8,9,10,11]. Although these techniques have been mostly employed to assess liver fibrosis burden, preliminary data suggest that MR elastography of the liver and spleen can be used to predict the risk or presence of portal hypertension and other quantitative MRI techniques might be able to do so, as well [12,13,14]. Quantitative MRI data could also likely be used to determine the rate of liver disease progression, predict the development of varices and ascites, and direct the need for interventions (e.g., endoscopic variceal screening).

Additionally, a small number of laboratory-based clinical biomarkers exist for detecting liver fibrosis and grading its severity. The aspartate aminotransferase (AST)-to-platelet ratio index, or APRI score, has been shown to detect significant liver fibrosis in adults with hepatitis C virus infection as well as numerous other chronic liver diseases, and in children with palliated biliary atresia [15,16,17]. Similarly, the fibrosis-4 (FIB-4) score has been demonstrated, again mostly in adults, to detect significant liver fibrosis in a variety of chronic liver diseases [18, 19]. The diagnostic performance of these clinical biomarkers has not been prospectively assessed for detecting and measuring liver fibrosis or predicting the presence of portal hypertension in a pediatric and young adult AILD cohort.

The primary purpose of our study was to determine whether multiple quantitative MRI biomarkers are associated with the presence of radiologic portal hypertension in children and young adults with AILD. As a secondary outcome, we sought to determine the relationships between laboratory-based APRI and FIB-4 scores and the presence of radiologic portal hypertension in the same patient population. If significant relationships were identified, longitudinal assessment of such biomarkers might be used in the clinic to predict impending portal hypertension, direct personalized patient management, and prognosticate patient outcomes. Finally, we sought to define the relationship between radiologic portal hypertension and important clinical outcomes such as endoscopic varices/gastrointestinal bleeding, clinically relevant ascites and liver transplant listing.

Materials and methods

Our institutional review board approved this cross-sectional study, which complied with the Health Insurance Portability and Accountability Act (HIPAA). Informed consent and informed assent were obtained, as appropriate. Perspectum Diagnostics (Oxford, UK) provided blinded image post-processing support for iron-corrected T1 (cT1) mapping at no cost through a formal research agreement. This study was partly funded by the Cincinnati Children’s Hospital Medical Center’s Center for Autoimmune Liver Disease and Center for Translational Fibrosis Research.

Forty-four children and young adults with AILD were enrolled in a prospective registry at the Center for Autoimmune Liver Disease at Cincinnati Children’s Hospital Medical Center. We approached children and young adults (ages 0 to 25 years) with a diagnosis of AILD (including PSC, AIH and ASC) and seen by a provider in the Division of Gastroenterology, Hepatology, and Nutrition about registry participation (www.clinicaltrials.gov; identifier number NCT03178630); registry enrollment is ongoing. The registry was started in fall 2016 and allows for the collection of pertinent demographic, clinical, laboratory, radiologic and histopathological data.

As a part of registry participation, participants underwent a multi-parametric research MRI examination at a field strength of 1.5 T (Ingenia; Philips Healthcare, Best, the Netherlands). Imaging parameters are listed in Appendix 1 [11, 20]. Two sets of anatomical images were obtained, including coronal single-shot fast spin-echo and axial T2-weighted fast spin-echo fat-suppressed imaging. Five axial sets of quantitative MRI images also were acquired, including:

  1. 1)

    Two-dimensional (2-D) gradient recalled echo (GRE) MR elastography of the liver (active driver frequency=60 Hz), with the passive driver placed over the right upper quadrant;

  2. 2)

    2-D GRE MR elastography of the spleen (active driver frequency=60 Hz), with the passive driver placed over the left upper quadrant;

  3. 3)

    Iron-corrected T1 mapping of the liver (cT1) using a modified Look-Locker (MOLLI) pulse sequence approach [8];

  4. 4)

    T2 mapping of the liver using a multi-echo fast spin-echo approach; and

  5. 5)

    DWI of the liver, using five b values and quantified as apparent diffusion coefficients (ADC; Fig. 1).

Fig. 1
figure 1

Autoimmune liver disease in a 7-year-old boy. a, b Coronal single-shot fast spin-echo (a) and axial T2-weighted fast spin-echo fat-saturated (b) MR images show areas of geographic liver signal hyperintensity, marked splenomegaly (with an area of splenic infarction) and ascites. c, d MR elastography images of the liver (c) and spleen (d) show abnormally increased stiffness (4.8 kPa and 7.4 kPa, respectively). Note that the splenic region-of-interest was drawn to avoid the area of splenic infarction. e, f Iron-corrected T1 (cT1) (e) and T2 (f) maps show heterogeneous appearance of the liver with areas of focally increased T1 (mean=1,003.4 ms) and T2 (mean=66.0 ms) values, particularly in the right lobe. The T2 map shows mild motion artifacts. g Diffusion-weighted imaging apparent diffusion coefficient (ADC) map shows that the liver is diffusely hypointense, with lower than expected mean ADC value (0.93×10−3 mm2/s)

Full size image

For each quantitative MRI sequence, axial imaging was performed through the mid (widest portion) liver (or spleen) at multiple anatomical levels. Liver regions of interest (ROI) were drawn by an MRI clinical physicist (S.D.S., 10 years clinical MRI experience) on MR elastography elastograms, T2 maps and ADC maps, attempting to stay within the right hepatic lobe and segment IV of the left lobe and maximizing ROI size while avoiding visible blood vessels, bile ducts and areas of artifact. Spleen ROIs were drawn similarly to encompass as much parenchyma as possible. For each quantitative MRI biomarker and patient, we used the mean value for the four images (corresponding to the MR elastography levels) for analysis. For cT1, we used whole-liver ROIs on each of four axial slices, with the mean value provided by Perspectum Diagnostics.

Two radiologists (J.R.D., 9 years of post-fellowship experience, and A.T.T., 5 years of post-fellowship experience) reviewed coronal and axial T2-weighted anatomical images separately to document the presence of splenomegaly, varices and ascites. Reviewing radiologists knew patients had a diagnosis of AILD but were blinded to specific diagnoses as well as all clinical, laboratory and quantitative MRI data. Consensus review was used to manage any discrepancies. We defined splenomegaly as a craniocaudad splenic length that was more than two standard deviations above the mean length for participant age and gender [21]. For participants older than 17 years, we used the 17-year-old mean age and standard deviation. We defined varices as visible portosystemic collateral vessels seen in a variety of locations, such as the gastrohepatic ligament, adjacent to the esophagus and spleen, in the mesentery and omentum, and along the expected course of the umbilical vein. We defined ascites as any volume of free fluid in the peritoneal cavity, including adjacent to the liver or spleen.

Another author (A.E.T., 1 year of post-fellowship experience in pediatric gastroenterology) searched electronic medical records to document gender, age at the time of research MRI, clinical diagnosis (e.g., PSC vs. ASC vs. AIH), pertinent laboratory data, and whether a given patient had experienced any of the following at the time of MRI: endoscopic identification of varices/gastrointestinal bleeding, clinically relevant ascites (i.e., requiring paracentesis or diuretic therapy) and liver transplant listing.

We calculated the APRI and FIB-4 scores for each participant. We calculated the APRI score using the following formula: (AST participant level/AST upper limits of normal)/platelet count, where AST is in U/L and platelet count is 109/L. We calculated the FIB-4 score using the following formula: (age x AST participant level)/(platelet count x √ALT), where age is in years, AST is in U/L, platelet count is 109/L, and ALT is in U/L. Continuous data were summarized as means, standard deviations and ranges, while categorical data were summarized as counts and percentages.

We used receiver operating characteristic (ROC) curve analyses to establish the diagnostic performance of each quantitative MRI technique for indicating the presence of (1) splenomegaly (regardless of the presence or absence of other findings of portal hypertension) and (2) radiologic portal hypertension. A diagnosis of radiologic portal hypertension was based on the presence of any of the following: (1) varices alone; (2) splenomegaly and ascites; (3) splenomegaly and varices; or (4) splenomegaly, ascites and varices. We also used ROC curve analyses to assess the diagnostic performance of the laboratory-based APRI and FIB-4 scores for indicating splenomegaly and radiologic portal hypertension. Cut-off values for the various MRI and laboratory-based biomarkers were established based on the Younden J statistic in order to maximize both sensitivity and specificity. Techniques were compared using area-under-the-ROC-curve (AUROC) with 95% confidence intervals (CI).

We used the Fisher exact test to compare the frequency of important clinical outcomes/complications in participants without and with radiologic portal hypertension. A P-value <0.05 was considered statistically significant for inference testing. We performed statistical analyses using GraphPad Prism version 7 for Windows (GraphPad Software, La Jolla, CA) and R 3.4.0 (pROC package). Because our study was exploratory in nature based on an existing prospective, longitudinal institutional registry of pediatric and young adult AILD patients, no formal sample size analysis was performed prior to the data analyses described here.

Results

Our study included 44 registry participants with known AILD. Twenty-three (52%) participants were male. Mean participant age was 15.2±4.0 years (range 7.5–23.5 years). Twenty-three subjects had a clinical diagnosis of AIH, 13 had PSC and eight had ASC. See Table 1 for additional demographics.

Table 1 Baseline demographics and laboratory data (n=44)
Full size table

Based on anatomical imaging, 22 (50%) participants had splenomegaly, while 13 (30%) had radiologic portal hypertension. Study population MRI- and laboratory-based biomarker means, standard deviations and ranges are presented in Table 2.

Table 2 MRI and laboratory-based biomarker means, standard deviations and ranges in our population of children and young adults with autoimmune liver disease
Full size table

Biomarkers and splenomegaly

Of the five quantitative MRI techniques, liver stiffness demonstrated the best diagnostic performance for indicating the presence of splenomegaly (AUROC=0.87; Fig. 2), followed by spleen stiffness (AUROC=0.80). Using a liver stiffness cut-off value of greater than 2.8 kPa, MR elastography had a sensitivity of 86.4% and specificity of 77.3% for indicating the presence of splenomegaly.

Fig. 2
figure 2

Liver stiffness and splenomegaly. a Tukey box plot compares liver stiffness in registry participants with and without splenomegaly. b Receiver operating characteristic curve shows the diagnostic performance of liver stiffness for indicating the presence of splenomegaly (AUROC=0.87)

Full size image

Of the two laboratory-based clinical biomarkers, the APRI score demonstrated the best diagnostic performance for indicating the presence of splenomegaly, with an AUROC and associated confidence interval similar to that of liver MR elastography (AUROC=0.83). Using a cut-off value of >0.42, the APRI score had a sensitivity of 85.7% and specificity of 76.2% for indicating the presence of splenomegaly.

The diagnostic performance for all evaluated MRI and clinical biomarkers for indicating splenomegaly, including AUROCs, sensitivities and specificities, are presented in Table 3.

Table 3 Diagnostic performance of MRI and laboratory-based biomarkers for indicating presence of splenomegaly
Full size table

Biomarkers and radiologic portal hypertension

Of the five quantitative MRI techniques, liver and spleen stiffness demonstrated nearly identical diagnostic performance for indicating the presence of portal hypertension (AUROC=0.98 and 0.96, respectively; Fig. 3). Using a liver stiffness cut-off value of greater than 3.9 kPa, MR elastography had a sensitivity of 92.3% and specificity of 93.6% for indicating the presence of portal hypertension. Using a spleen stiffness cut-off value of greater than 7.3 kPa, MR elastography had a sensitivity of 90.9% and specificity of 93.3% for indicating the presence of portal hypertension.

Fig. 3
figure 3

Liver and spleen stiffness, and portal hypertension. a Tukey box plot compares liver stiffness in registry participants with and without portal hypertension. b Receiver operating characteristic curve shows the diagnostic performance of liver stiffness for indicating the presence of radiologic portal hypertension (AUROC=0.98). c Tukey box plot compares spleen stiffness in registry participants with and without portal hypertension. d Receiver operating characteristic curve shows the diagnostic performance of spleen stiffness for indicating the presence of radiologic portal hypertension (AUROC=0.96). PTL HTN portal hypertension

Full size image

The two laboratory-based clinical biomarkers, the APRI and FIB-4 scores, demonstrated very similar diagnostic performance for indicating the presence of portal hypertension (AUROC=0.87 and 0.88, respectively; Fig. 4). Using a cut-off value of greater than 0.66, the APRI score had a sensitivity of 91.7% and specificity of 80.0% for indicating the presence of portal hypertension. Using a cut-off value of greater than 0.76, the FIB-4 score had a sensitivity of 84.6% and specificity of 93.3% for indicating the presence of portal hypertension.

Fig. 4
figure 4

Fibrosis-4 score performance and portal hypertension. a Tukey box plot compares fibrosis-4 (FIB-4) score in registry participants with and without radiologic portal hypertension. b Receiver operating characteristic curve shows the diagnostic performance of the FIB-4 score for indicating the presence of radiologic portal hypertension (AUROC=0.88). PTL HTN portal hypertension

Full size image

The diagnostic performance for all evaluated MRI and laboratory-based biomarkers for indicating the presence of portal hypertension, including AUROCs, sensitivities and specificities, is presented in Table 4.

Table 4 Diagnostic performance of MRI and laboratory-based biomarkers for indicating presence of radiologic portal hypertension
Full size table

Association between radiologic portal hypertension and important clinical outcomes

Thirteen of 44 (30%) registry participants had radiologic findings of portal hypertension, including 8 with AIH, 3 with ASC and 2 with PSC. Ten of these 13 (77%) participants experienced one or more clinically important outcomes, compared to 0/31 (0%) participants without radiologic portal hypertension (P<0.0001; odds ratio >100 [>10.5–1,133]). Important clinical outcomes in these 10 individuals included endoscopic identification of varices/gastrointestinal tract bleeding (n=6), clinically relevant ascites (n=2) and liver transplant listing (n=7).

Discussion

A review of Table 2 shows that all of the quantitative MRI biomarkers assessed in our study are outside the normal range, on average, when compared to reported values in the literature for patients with no or mild liver disease. For example, for MR elastography of the liver, our population’s mean stiffness was 3.4 kPa. In a paper by Xanthakos et al. [22], a stiffness of 2.71 kPa showed excellent diagnostic performance (AUROC=0.92) for detecting significant liver fibrosis in a pediatric population. A more recent study by Trout et al. [7] showed that a stiffness cut-off value of 2.49 kPa in a pediatric population without fatty liver disease was able to detect significant liver fibrosis with an AUROC of 0.82 (sensitivity=86%, specificity=71%). In a study of adults evaluating the diagnostic performance of cT1 in 79 unselected patients with paired MR imaging and biopsy data, and in 7 healthy volunteers, the volunteer values were on average 717±48 ms and the biopsied patients with no fibrosis were 750±42 ms [8]. This is in contradistinction to our patient population, which had a mean cT1 measurement of 987.8±73.4 ms.

Our data suggest that a single quantitative biomarker has satisfactory diagnostic performance for indicating the presence of splenomegaly, perhaps the earliest imaging and physical examination manifestation of portal hypertension. Liver stiffness measurements using MR elastography had an AUROC of 0.87 and was 86% sensitive and 77% specific, when maximizing diagnostic performance. When comparing other MRI biomarkers, AUROCs would not be clinically acceptable, with cT1, T2 and DWI ADC AUROCs being no better or only slightly better than that what would be observed by chance when considering 95% confidence intervals. Surprisingly, the APRI score also showed respectable diagnostic performance for indicating the presence of splenomegaly with an AUROC of 0.83, having substantial overlap of the 95% confidence intervals for the AUROC values with liver stiffness. Further research comparing liver MRE and the APRI score for predicting impending splenomegaly is needed. This is in part because the APRI score is based on routine bloodwork and is likely to be a much cheaper biomarker, and the biomarker of choice, if shown to be equivalent or non-inferior to MR elastography liver stiffness.

Our data suggest that both liver and splenic stiffness acquired using MR elastography can detect radiologic portal hypertension with a high degree of diagnostic accuracy. AUROCs (0.98 and 0.96, respectively) were essentially identical for liver and spleen MR elastography with relatively similar 95% confidence intervals. These findings are similar to those in a prior study, by Ronot et al. [14], which found both liver and spleen stiffness to correlate with hepatic venous pressure gradients in adults with cirrhosis, some of whom had portal hypertension. Of note from their study, on multivariable analysis, spleen loss modulus (a quantitative measure that can be extracted from 3-D MR elastography that relates to a tissue’s viscosity) had the highest predictive accuracy for both severe portal hypertension (AUROC=0.81) and high-risk varices (AUROC=0.93).

In our study, the AUROC for cT1 (0.86) for indicating radiologic portal hypertension was somewhat lower than those for liver and spleen MR elastography, although the 95% confidence intervals overlap. Liver T2 and DWI ADC values were poor indicators of portal hypertension. Interestingly, both the APRI and FIB-4 scores demonstrated essentially identical good diagnostic performance for indicating the presence of radiologic portal hypertension (AUROCs of 0.87 and 0.88, respectively). Again, more research is needed to determine whether there is a true difference in diagnostic performance between the imaging biomarkers (liver and spleen MR elastography), which utilize more resources, and the clinical laboratory-based biomarkers for indicating the presence of portal hypertension as well as predicting its impending onset.

Regarding the relationship between splenomegaly and radiologic portal hypertension, the onset of splenomegaly appears to occur at a lower liver stiffness, on average, than radiologic portal hypertension (2.8 kPa vs. 3.9 kPa, based on ROC curve analyses). Similarly, splenomegaly appears to be present, on average, at lower APRI and FIB-4 scores than radiologic portal hypertension (again, based on ROC curve analysis). This suggests that splenic enlargement might be an earlier manifestation of increasing portal pressure.

We also have demonstrated that the presence of radiologic portal hypertension is a risk factor for important clinical outcomes and morbidity in this specific patient population. In particular, radiologic portal hypertension, as defined in our study, seems to predict endoscopic identification of varices/gastrointestinal tract bleeding, clinically relevant ascites and liver transplant listing. Equally important, the absence of radiologic portal hypertension seems to indicate very low risk for these outcomes (0 such outcomes in 31 participants).

So, what do our results mean to children with chronic liver diseases? Given that there are significant positive associations between both MRI and laboratory biomarkers and splenomegaly (which may be an early feature of portal hypertension) and radiologic portal hypertension, we believe that these biomarkers and, in particular, their serial change over time might allow for the prediction of both portal hypertension and important associated clinical outcomes. Such outcomes might include rate of liver disease progression, both the short- and long-term risks of developing portal hypertension and its complications (e.g., variceal bleeding, intractable ascites), and need for intervention or changes in management such as endoscopic variceal screening. The cut-off values for liver and spleen stiffness — which are highly indicative of likely or impending portal hypertension — might help inform the selection of surrogate endpoints when designing clinical trials in the setting of pediatric-onset AILD. It is also conceivable that these biomarkers and their change over time might predict even more important outcomes such as time to liver transplant listing and time to actual transplant. It has not been established whether either the MRI- or laboratory-based biomarkers will ultimately be the better predictor, or whether there are additional predictive value and improved outcomes by combining multiple biomarkers (a so-called multiparametric approach).

Our study has limitations. First, it includes only 44 participants, 13 of whom had radiologic findings of portal hypertension. However our study is prospective and investigates a rare group of diseases, yields several significant results, and provides promising pilot data that should serve as the basis for future validation investigations, including multi-institutional and longitudinal studies. Second, our primary outcome is radiologic portal hypertension. The most sensitive methods for detecting early portal hypertension are invasive catheter-based techniques that can provide measurements of pressure in the portal venous system. That said, our ultimate outcome is generalizable, easy to assess (is there splenomegaly, ascites or varices on cross-sectional imaging?), and clinically relevant. In fact, we have demonstrated that its presence is highly associated with important clinical outcomes, such as varices/gastrointestinal tract bleeding and liver transplant listing. Finally, while we have shown that numerous biomarkers are clearly associated with and can indicate the presence of splenomegaly and radiologic portal hypertension with satisfactory diagnostic performance, additional prospective studies employing longitudinal assessments are necessary to determine their predictive abilities and value in clinical practice.

Conclusion

Both quantitative MRI- and clinical laboratory-based biomarkers are associated the presence of splenomegaly and radiologic portal hypertension in pediatric and young adult patients with AILD. It is possible that MRI- or laboratory-based biomarkers, particularly when assessed longitudinally over time, can be used to predict impending portal hypertension and its associated risks, and thus can be used to more personally guide patient care and predict key outcomes. Additionally, we have demonstrated that the presence of radiologic portal hypertension is highly associated with important clinical outcomes in this patient population, while its absence is associated with likely very low risk for such outcomes, at least in the short term. Further research, including prospective longitudinal investigations, is needed to confirm our findings and to establish whether MRI-based biomarkers, laboratory-based biomarkers, or some combination of both provides the greatest predictive abilities and correlates best with the presence of (or impending) portal hypertension and other key clinical outcomes.