Liver ultrasound patterns in children with cystic fibrosis correlate with non-invasive tests of liver disease

Abstract

Objectives

Early identification of children with cystic fibrosis (CF) at risk for severe liver disease (CFLD) would enable targeted study of preventative therapies. There is no gold standard test for CFLD. Ultrasonography (US) is used to identify CFLD, but with concerns for its diagnostic accuracy. We aim to determine if differences in standard blood tests, imaging variables and non-invasive liver fibrosis indices correlate with liver US patterns and thus provide supportive evidence that a heterogeneous US liver pattern reflects clinically relevant liver disease.

Methods

We studied baseline research abdominal US and bloodwork from 244 children with pancreatic insufficient CF, ages 3–12y, enrolled in a prospective study of the ability of US to predict CF cirrhosis (PUSH study). Children with a heterogeneous liver pattern on US (HTG, n=62) were matched 1:2 in design with children with normal US (NL, n=122). Analyses included children with nodular (NOD, n=22) and homogeneous hyperechoic (HMG, n=38) livers.

Results

Univariate analysis showed significant differences between US groups for standard blood tests, spleen size, and non-invasive liver fibrosis indices. Multivariable models discriminated NOD vs. NL with excellent accuracy (AUROC 0.96). Models also distinguish HTG vs. NL (AUROC 0.76), NOD vs. HTG (0.78), and HMG vs NL (0.79).

Conclusion

Liver US patterns in children with CF correlate with platelet count, spleen size and indices of liver fibrosis. Multivariable models of these biomarkers have excellent discriminating ability for NL vs NOD, and good ability to distinguish other US patterns, suggesting that US patterns correlate with clinically relevant liver disease.

Introduction

Cirrhosis occurs in 5–7% of people affected by cystic fibrosis (CF) and is the third leading cause of death in CF. Cirrhosis mostly develops before 20 years of age, although liver involvement may also develop in adults with CF^{1, 2}. There is currently no proven therapy that prevents or reverses CFLD. Early identification of children at risk for cirrhotic CFLD would enable targeted study of preventative therapies.

Ultrasonography (US) is frequently utilized in clinical research and care for children with CF, among whom US detects an abnormal liver parenchyma in approximately 25%, although only a minority of these will progress to severe liver disease^{3, 4}. US patterns have historically shown a poor correlation with evidence of cirrhosis on liver biopsy⁵. However, histological changes in CFLD are known to be patchy and thus, for this disease, liver biopsy does not constitute a gold standard because its diagnostic accuracy is limited by sampling error. It is currently unknown to what extent the poor correlation between US and liver biopsy may reflect the inadequacy of US or the limitations of biopsy to accurately assess severity of CFLD. In the absence of a gold standard test, several definitions of CFLD have been proposed and this further complicates interpretation and comparison of studies.

In a single-center study using data derived from routine clinical care, a heterogeneous liver pattern on abdominal US was previously identified as a potential marker for patients with CF who later develop cirrhosis³. Patients with a heterogeneous echogenicity pattern on US had a 5-fold increased incidence of cirrhosis and a 6-fold increased incidence of portal hypertension compared to children with a normal liver pattern on US. These findings have not been replicated in a prospective multicenter research study and questions remain concerning the utility of US in predicting future development of cirrhosis, the reproducibility of US readings between radiologists and centers, and the extent to which these results can be generalized⁶. The clinical utility of US in defining the progression of CFLD through its early stages therefore remains unclear.

The Cystic Fibrosis Liver Disease Network (CFLD-NET) is a North American multicenter research consortium investigating the development, diagnosis and clinical impact of CFLD by conducting an ongoing, prospective study investigating the utility of abdominal ultrasonography (US) to identify young children at risk for the subsequent development of cirrhosis⁷. The Network’s studies will ultimately determine the ability of US to predict future cirrhosis and will investigate the extent to which additional biomarkers of liver disease may enhance the predictive accuracy of US in achieving an early diagnosis of CFLD.

There is a lack of prospective data on liver US findings in CF and their correlation with other biomarkers of liver involvement. We hypothesize that, among children with CF who have not previously been identified to have cirrhosis, liver US patterns are associated with differences in biomarker indices, including platelet count, liver biochemistries and spleen size. We aim to determine if differences in these biomarkers provide supportive evidence that a heterogeneous liver pattern on US reflects clinically relevant liver disease.

Methods

The CFLD-NET is investigating predictors of the development of cirrhosis in CF in a prospective study, “Prediction by Ultrasound of the Risk of Hepatic Cirrhosis in Cystic Fibrosis (PUSH)” (ClinicalTrials.gov: ), whose detailed methodology has been previously reported ⁷. Briefly, children 3–12 years of age attending the CF clinics in any of the eleven collaborating sites were enrolled into the study if they were diagnosed with CF based on a sweat chloride of >60 mEq/L or two disease-causing CFTR genetic mutations with evidence of end organ involvement, were pancreatic insufficient and enrolled in the US or Toronto CF registry. Exclusion criteria included known cirrhosis or portal hypertension (i.e., splenomegaly, ascites), previous Burkholderia species on respiratory cultures (due to the anticipated different clinical course among this small group of patients), or short bowel syndrome requiring parenteral nutrition after 3 months of age.

Baseline liver US studies were performed after participants had fasted for at least 4 hours and were timed not to coincide with an acute respiratory exacerbation of their CF. US findings were classified by liver parenchymal pattern as normal (NL), heterogeneous (HTG), homogeneously hyperechoic (HMG) or nodular (NOD). One study radiologist per site was identified and read all study related ultrasounds. Classification was based on consensus by the local study radiologist and two additional study radiologists from CFLD-NET sites blinded to other reads and clinical and demographic data. In the setting of lack of consensus among the 3 study radiologists, a 4^th study radiologist’s grade was used to establish consensus (n=22 of the 725 baseline studies). All study radiologists participated in web-based training for the grading of US studies with validation and ongoing monitoring of the consistency of reading. HTG denoted increased echogenicity that was patchy or limited to periportal regions, measuring > 2mm in thickness. HMG denoted diffusely increased hepatic parenchymal echogenicity relative to renal cortex echogenicity, which may also show absent or poor definition of portal venous and hepatic structures, and posterior beam attenuation with absent or incomplete diaphragm visualization. NOD pattern denoted a heterogeneous echogenicity and coarse echotexture of the liver parenchyma with obvious nodularity of the liver contour. NOD may represent cirrhosis or other important liver pathology that may contribute to portal hypertension, such as nodular regenerative hyperplasia⁸.

Among the 725 participants who underwent screening ultrasound at the beginning of the PUSH study, all 62 eligible children with HTG liver parenchymal pattern were matched by age (± 2 years), center and Pseudomonas status with two NL children. These HTG and 122 of their 123 originally matched NL participants were followed longitudinally, along with children found to have NOD (n=22) and HMG (n=38) liver patterns. This cohort of 244 children in long-term follow-up forms the cohort for the analyses reported here.

In the current study, we aimed to determine if platelet count, routine biochemical markers of liver disease, spleen size and non-invasive scores of liver fibrosis differ among groups of patients with different baseline US patterns and whether one or a combination of these biomarkers can distinguish US patterns. Spleen size was measured by US as the longest cranio-caudal dimension and expressed as a spleen-size-for-age z-score (SSAZ) using age-specific normal ranges⁹. Portal vein was measured at its widest diameter and expressed as a z-score (PVZ) using height-adjusted normal ranges¹⁰. Baseline blood work variables were recorded including platelet count, aspartate aminotransferase (AST), alanine aminotransferase (ALT), gamma-glutamyl transferase (GGT), alkaline phosphatase (ALP), and albumin. AST-to-platelet ratio index (APRI) was calculated as (AST/upper limit of normal of AST using 40 U/L for this study)/platelet count, and fibrosis index based on 4 factors (FIB4) was calculated as (age × AST)/(platelet count × √ALT)^11-13. Patient weight and height were recorded at baseline.

Patient demographic and clinical characteristics were compared between groups using Kruskal Wallis, Chi square, and Fisher’s exact test as appropriate. Univariate logistic regression was first used to study association between biomarkers and US patterns individually. We then selected the best prediction models using multivariate logistic regression and step-wise selection with entering and removing criteria with p-value <0.05. Two alternative sets of candidate predictors were initially entered into the model for distinguishing between NOD vs. NL, HTG vs. NL and NOD vs. HTG. The first set included APRI (including AST, platelets), age, GGT, SSAZ, PVZ, ALT and ALP. The second set included FIB-4 (including age, AST, ALT, platelets), GGT, SSAZ, PVZ, albumin and ALP. Variables entered for modeling the associations between HMG and NL were age, gender, ethnicity, ALT, GGT, body mass index z-core (BMIZ), APRI, ALP and albumin. The predictive ability was classified as excellent if the area under the receiver operating characteristic curve (AUROC) was >0.90, good if 0.80–0.89, and fair if 0.70–0.79. Prior to model selection, we used IVEWare to impute missing independent variables¹⁴ to avoid losing a significant portion of our sample size due to missing values. One HTG subject had a large AST value, and consequently a large FIB-4 value. Model diagnostics showed that it was an influential point and was excluded in both univariate and multivariate analyses.

The study was approved by the Institutional Review Board or Research Ethics Board of each participating Institution. Written informed consent was obtained from the parents or guardians of the children who served as subjects of the investigation and, when appropriate, assent from the subjects themselves.

Results

The entire longitudinal PUSH study cohort of 244 children was included in the current study of their baseline data, representing 62 children with HTG, 122 NL, 22 NOD and 38HMG. Baseline characteristics show that children with HTG and NOD patterns were slightly older than NL and HMG (Table 1). The trend to fewer females in HTG and NOD was not statistically significant. Children with HMG were more commonly Hispanic and had a higher mean BMI z-score.

Table 1.

Baseline characteristics of the study cohort

		Normal (N = 122)	HTG (N = 62)	NOD (N= 22)	HMG (N = 38)	p-value
Age at Ultrasound (years)		8.4±3.1	8.6±3.2	9.7±2.7	9.6±3.0	0.074£
Ethnicity	Hispanic	7 (5.7%)	0	1 (4.6%)	8 (21.1%)	0.0009$
	Non-Hispanic	115 (94.3%)	62 (100%)	21 (95.4%)	30 (79.0%)
Height z-score		−0.2±1.0 N=119	−0.1±0.9 N=61	−0.3±0.9 N=21	−0.4±0.9 N=38	0.63£
Weight z-score		−0.2±0.9 N=119	−0.1±0.9 N=61	−0.1±1.0 N=21	0.1±1.0 N=37	0.55£
BMI z-score		0.0±0.8 N=119	0.0±1.0 N=61	0.1±0.9 N=21	0.5±0.9 N=37	0.042£
Female		61 (50.0%)	22 (35.5%)	9 (40.9%)	21 (55.3%)	0.18*
CFRD	No	112 (94.1%)	55 (88.5%)	18 (85.7%)	36 (94.7%)	0.23$
	IGT	2 (1.7%)	4 (6.6%)	2 (9.5%)	2 (5.3%)
	CFRD	5 (4.2%)	3 (4.9%)	1 (4.8%)	0 (0%)
F508del	Heterozygous	43 (35.3%)	21 (33.9%)	6 (27.3%)	10 (26.3%)	0.52$
	Homozygous	74 (60.7%)	39 (62.9%)	13 (59.1%)	25 (65.8%)
	Other	5 (4.1%)	2 (3.2%)	3 (13.6%)	3 (7.9%)
Mutation	Minimal	34 (28.8%)	15 (25.0%)	5 (23.8%)	7 (18.4%)	0.64*
	Unclassified	84 (71.2%)	45 (75.0%)	16 (76.2%)	31 (81.6%)
Platelet count (x109/L)		333.7±71.9 N=120	302.8±84.4 N = 62	244.5±89.0 N=22	334.3±63.6 N=38	< 0.0001£
AST (IU/L)		34.2±12.7 N=119	44.4±28.7 N=61	54.1±23.7 N=22	58.8±60.8 N=36	< 0.0001£
ALT (IU/L)		31.4±18.5 N=115	41.3±22.0 N=57	56.4±24.1 N=22	62.0±82.9 N=36	< 0.0001£
GGTP (IU/L)		15.3±7.9 N=113	33.6±32.8 N=58	60.9±46.6 N=20	24.2±20.5 N=36	< 0.0001£
ALP (IU/L)		253.0±81.2 N=110	256.1±71.7 N=58	318.9±87.5 N=22	267.9±65.5 N=36	0.0093£
Albumin (g/dL)		4.2±0.4 N=111	4.3±0.4 N=57	4.1±0.4 N=20	4.3±0.4 N=36	0.43£
APRI		0.27±0.12 N=118	0.40±0.30 N=61	0.65±0.43 N=22	0.43±0.38 N=36	< 0.0001£
FIB-4		0.16±0.08 N=109	0.21±0.15 N=57	0.29±0.13 N=21	0.20±0.10 N=36	<0.0001£
Height adjusted Portal vein diameter z-score		0.35±1.12 N=118	0.31±1.22 N=60	0.75±1.41 N=21	0.35±1.53 N=37	0.56£
Spleen size by ultrasonography (cm)		8.9±1.5 N=119	9.2±1.8 N=61	10.5±2.3 N=21	9.2±1.4 N=38	0.0066£
Spleen size-for-age z-score		−0.0±1.19 N=119	0.30±1.62 N=61	1.43±1.96 N=21	0.09±1.19 N=38	0.0068£

Values shown are mean±SD or count (%)

^£Kruskal-Wallis for continuous variables and Chi-square test or Fisher’s exact test for categorical variables

^*χ² test

^$Fisher’s exact test

Among all screening US for the PUSH study, the concordance levels were 3 out of 3 for 49.6%, 2 out of 3 for 44.8%, and 2 out of 4 for 5.6%.

When comparing biomarker values among NL, HTG and NOD, the mean values for platelets (p<0.0001) were higher in NL. ALT, AST, GGT (all with p<0.0001), SSAZ (p=0.0068), APRI (p<0.0001), and FIB4 (p<0.0001) were lower in NL compared to NOD. For each variable, the value for HTG was intermediate between values for NOD and NL (Table 1 and Figure 1).

Figure 1.

Levels of biomarkers in children with different US patterns.

The width of each figure represents the frequency of observations at that level. The upper and lower limits of the figure represent the maximum and minimum observed values, respectively. The dot represents the average. Compared to NL, *p<0.05, **p<0.01, ***p<0.001, # p<0.0001

No significant differences between US pattern groups were identified for height z-score, weight z-score, albumin, portal vein diameter z-score, or Pseudomonas aeruginosa status.

When comparing HMG with the other groups, values for AST, ALT, and GGT were elevated compared to NL, and intermediate between values for HTG and NOD for AST and ALT but not GGT. Values for SSAZ, APRI, and FIB4 were similar in HMG and HTG.

We compared US patterns by univariate logistic regression analysis using four models, NOD vs. NL, HTG vs. NL, NOD vs. HTG and HMG vs. NL. This revealed several significant associations (Table 2). APRI, GGTP, ALT, and AST were statistically significant in all four models comparing US patterns. Platelet count was significant in two models (NOD vs. NL, HTG vs. NL). FIB4, ALP, and SSAZ were significant in models comparing NOD vs. NL and NOD vs. HTG. Height-adjusted portal vein diameter was significant only in the model comparing NOD vs. NL. Albumin was not significant in any of the univariate models.

Table 2.

Univariate logistic regression models examining the association between biomarkers and US patterns individually

Biomarkers	Odds Ratio (95% CI)
	NOD vs. NL	HTG vs. NL	NOD vs. HTG	HMG vs .NL
APRI (per 0.1 unit)	2.23 (1.53, 3.25)*	1.69 (1.30, 2.20)*	1.55 (1.19, 2.04)*	1.48 (1.16, 1.89)*
FIB-4 (per 0.1 unit)	3.25 (1.88, 5.61)*	1.50 (1.05, 2.13)*	1.86 (1.19, 2.92)*	1.68 (1.09, 2.58)*
Spleen size-for-age z-score	1.90 (1.35, 2.67)*	1.18 (0.94, 1.49)	1.45 (1.07, 1.97)*	1.07 (0.79, 1.45)
Height adjusted portal vein diameter	1.33 (0.90, 1.95)	0.96 (0.73, 1.26)	1.33 (0.89, 1.98)	1.00 (0.74, 1.36)
ALT (IU/L)	1.05 (1.03, 1.07)*	1.02 (1.01, 1.04)*	1.03 (1.01, 1.06)*	1.03 (1.01, 1.06)*
AST (IU/L)	1.07 (1.03, 1.10)*	1.04 (1.01, 1.06)*	1.04 (1.01, 1.07)*	1.04 (1.02, 1.06)*
GGT (IU/L)	1.15 (1.08, 1.23)*	1.08 (1.04, 1.12)*	1.020 (1.003, 1.032)*	1.06 (1.02, 1.10)*
Platelet count (x109/L)	0.98 (0.97, 0.99)*	0.995 (0.990, 0.999)*	0.991 (0.985, 0.998)*	1.000 (0.995, 1.005)
ALP (IU/L)	1.009 (1.003, 1.015)*	1.000 (0.996, 1.004)	1.012 (1.004, 1.019)*	1.002 (0.998,1.007)
Albumin (g/dL)	0.52 (0.17, 1.65)	1.30 (0.56, 2.99)	0.46 (0.14, 1.48)	1.84 (0.68, 4.97)

^*p-value < 0.05

Multi-variable models confirmed the ability of the variables analyzed to predict the US pattern (Figure 2 and Supplementary Table 1). AUROC confirmed excellent discriminating ability for NOD vs. NL (AUROC=0.96, 96%CI: 0.92, 1.00), in a final model that included GGT and APRI. A fair discriminating ability was found for HTG vs. NL (AUROC=0.76, 95%CI: 0.67, 0.83), with also GGT and APRI included in the model. For NOD vs. HTG, the predictive ability was good (AUROC=0.78, 95%CI: 0.66, 0.90) with GGT and SSAZ as the predictive variables. Finally, the ability to predict HMG vs. NL was also fair (AUROC=0.79, 95%CI: 0.71, 0.88), with ALT, BMIZ, age and ethnicity in the model.

Figure 2.

Receiver Operating Characteristic (ROC) curves for the final prediction models.

Discussion

There is much debate about the clinical significance of a heterogeneous appearance of the liver parenchyma on US for children with CF. CFLD-NET is addressing this question with a rigorous, prospective research approach that aims to carefully characterize patients and minimize the variability often associated with reading ultrasound studies. Using this approach, our analysis of baseline data from the PUSH study cohort shows a strong association between US patterns and non-invasive biomarkers that correlate with more advanced liver disease. In the absence of a gold standard diagnostic test for CFLD and with concerns about the limited accuracy of liver biopsy for this condition, we believe that our findings validate the likely clinical significance for HTG pattern on US by showing its correlation with other known biomarkers of chronic liver disease and portal hypertension, including platelet count and spleen size.

Patients with CF frequently have abnormalities of AST, ALT and GGT^{15, 16}. Interestingly, our finding that GGT is an important discriminator of NOD and HTG from NL is consistent with studies from the United States and Netherlands that showed strong associations between persistently high-normal or mildly elevated GGT and the development of cirrhotic CFLD within 2 years^16,17.

Biomarker indices such as APRI and FIB4 have proven useful to discriminate severity of liver fibrosis in many adult and pediatric liver disorders. In a study of 51 children with CFLD who had dual-pass liver biopsies, compared to 107 children with CF and no liver disease, the AUROC for diagnosis of CFLD was 0.75 for APRI and 0.60 for FIB4, and for diagnosing severe CFLD (greater than stage 3 fibrosis) was 0.81 for APRI and 0.70 for FIB4¹³. In a separate multi-center study of 497 children and adults with CF cirrhosis with portal hypertension, APRI and FIB4 scores were above the diagnostic thresholds for CFLD described by Leung et al in 96% and 90% of patients, respectively¹⁸. Our data further support these findings as APRI and FIB4 were both significantly increased among the NOD group with approximately two-fold higher scores compared to patients with normal US pattern.

Mean portal vein diameter is known to be increased in patients with CF, compared to those without CF ^{19, 20}. However, there are limited data exploring the correlation of PV diameter with severity of CFLD. Measurement of the portal vein has not entered common usage in the evaluation of cirrhosis and portal hypertension. Our data shows a trend towards greater PV diameter in HTG and NOD compared to NL, but this was not statistically significant.

Increased spleen size in patients with chronic liver disease is almost always caused by increased portal pressure but correlates poorly with the severity of portal hypertension²¹. Splenomegaly measured by ultrasound has good sensitivity but poor specificity for the non-invasive diagnosis of esophageal varices^22-24. Splenomegaly is commonly used in clinical practice as a marker for the development of portal hypertension, including in patients with cystic fibrosis⁶, although we are not aware of published data examining changes in spleen size as a predictor of progressive CFLD. Our study shows that spleen size measured by US is greater in HTG compared to NL, and is greater still in NOD.

A single-center study has suggested that the the combination of an abnormal liver US pattern and elevated serum aminotransferases indicates a risk for the development of cirrhosis in CFLD²⁵. Published data concerning the utility of combining bloodwork and imaging results to predict CFLD are otherwise limited. Our data highlight the potential for combinations of bloodwork and imaging variables to more powerfully identify CFLD and to distinguish NL, HTG and NOD patterns.

The importance of HMG pattern is currently unknown and its long-term outcomes have been poorly studied. Due to the common finding of HMG on US of obese children and adults with non-alcoholic fatty liver disease, it is often assumed that the HMG pattern in children with CF is due to hepatic steatosis. Our multivariable model comparing HMG to NL significantly included BMIZ and ethnicity, unlike the modelling comparing HTG and NOD to NL. This therefore raises the intriguing possibility that an HMG pattern is caused by a different mechanism or abnormality than that causing HTG and NOD. Further longitudinal follow-up of this cohort of patients will help to clarify this question.

Our study did not include liver biopsy, due to research ethics constraints and the limitations introduced by sampling error due to the focal variability of the liver pathology in CF. We therefore cannot determine the proportion of patients with a NOD parenchymal pattern by US who have cirrhosis or whose nodularity arises from a different pathological process, such as nodular regenerative hyperplasia secondary to portal venopathy⁸. Transient elastography and magnetic resonance elastography were not available at the time of baseline data collection, although have since been incorporated into the study as we believe these techniques have potential to add important information to our understanding of the nature and progression of liver abnormalities in CF. The small number of patients in each of our subgroups limited the number of variables that could be incorporated in the predictive models. However, we have carefully and prospectively characterized these children with CF and utilized radiologists who were trained for the purposes of the study and have explicitly demonstrated reliably reproducible reading of the US patterns.

We have entered a new era in the management of cystic fibrosis and cholestatic liver diseases, with CFTR-modifying, anti-cholestatic and antifibrotic therapies now available or in late stages of development for clinical use. It is therefore imperative that we identify biomarkers that accurately reflect changes in the severity of CFLD over time, and that identify patients who will progress to cirrhosis whilst they are still at an early stage of disease that is likely to be more responsive to therapeutic interventions. In summary, our study data show that liver US patterns correlate with biomarkers of liver disease severity, including blood work and spleen size variables. This observation provides supportive evidence that a heterogeneous liver pattern on US reflects clinically relevant, progressive liver disease. Important follow-up of this patient cohort will enable analysis of these biomarkers among children who transition between US patterns over time, such as from NL to HTG and HTG to NOD.

What is known

• There is no single gold standard diagnostic test for cystic fibrosis liver disease (CFLD)

• The clinical relevance of liver parenchymal abnormalities identified by liver ultrasonography (US) is unclear

What is new

• Liver US parenchymal abnormalities in children with cystic fibrosis correlate with biomarkers of severity of liver disease, including platelet count, spleen size and non-invasive indices of liver fibrosis

• Biomarkers distinguish groups of children with different liver parenchymal appearances

• These findings suggest that liver US abnormalities reflect the presence of clinically relevant liver disease

Abbreviations

ALT

alanine aminotransferase

ALP

alkaline phosphatase

APRI

aspartate aminotransferase to platelet ratio index

AUROC

area under the receiver operating characteristic curve

BMIZ

body mass index z-score

CF

cystic fibrosis

CFLD

cystic fibrosis liver disease

FIB4

fibrosis index based on 4 factors

GGT

gamma-glutamyl transferase

HMG

homogeneous

HTG

heterogeneous

NL

normal

NOD

nodular

PUSH

prediction by ultrasound of the risk of hepatic cirrhosis in cystic fibrosis

PVZ

portal vein diameter z-score

SSAZ

spleen-size-for-age z-score

US

ultrasonography