Heterogeneous liver on research ultrasound identifies children with cystic fibrosis at high risk of advanced liver disease

Abstract

Background:

This study examines whether heterogeneous (HTG) pattern on liver ultrasound (US) identifies children at risk for advanced cystic fibrosis liver disease (aCFLD).

Methods:

Prospective 6-year multicenter case-controlled cohort study. Children with pancreatic insufficient cystic fibrosis (CF) aged 3–12 years without known cirrhosis underwent screening US. Participants with HTG were matched (by age, Pseudomonas infection status and center) 1:2 with participants with normal (NL) US pattern. Clinical status and laboratory data were obtained annually and US bi-annually for 6 years. Primary endpoint was development of nodular (NOD) US pattern consistent with aCFLD.

Results:

722 participants underwent screening US, with 65 HTG and 592 NL. Final cohort included 55 HTG and 116 NL with ≥ 1 follow-up US. ALT, AST, GGTP, FIB-4, GPR and APRI were higher, and platelets were lower in HTG compared to NL. HTG had a 9.5-fold increased incidence (95% confidence interval [CI]:3.4, 26.7, p<0.0001, 32.7% vs 3.4%) of NOD versus NL. HTG had a sensitivity of 82% and specificity of 75% for subsequent NOD. Negative predictive value of a NL US for subsequent NOD was 96%. Multivariate logistic prediction model that included baseline US, age, and log(GPR) improved the C-index to 0.90 compared to only baseline US (C-index 0.78). Based on survival analysis, 50% of HTG develop NOD after 8 years.

Conclusions:

Research US finding of HTG identifies children with CF with a 30–50% risk for aCFLD. A score based on US pattern, age and GPR may refine the identification of individuals at high risk for aCFLD.

Advanced cystic fibrosis liver disease (aCFLD) manifesting as liver nodularity with or without portal hypertension occurs in about 7% of individuals with CF ^1,2. The cirrhotic form of aCFLD primarily occurs in children and adolescents with a median age at diagnosis of 10 years ³ while the non-cirrhotic form of aCFLD occurs across all age groups and is more common in pediatrics than previously recognized ^4,5. Risk factors for the development of aCFLD in children and adolescents include male gender and Class I-III cystic fibrosis transmembrane regulator (CFTR) mutations ^6,7. Bartlett et al identified that heterozygosity for the Z allele of alpha 1 antitrypsin is associated with a 7-fold increased risk for the development of aCFLD⁸. However, this was only present in 9% of participants with aCFLD⁸. A higher gamma glutamyl transpeptidase (GGTP) level early in life is also associated with increased risk for the development of aCFLD ^9,10.

A prior small single center study in children with CF followed for an average of 10 years showed that a heterogeneous echogenic liver pattern (HTG) on ultrasound (US) identified individuals at risk for aCFLD (defined as a nodular liver pattern on US (NOD)) and could potentially be used as a predictive biomarker for the risk of aCFLD¹¹. In that study, 67% of participants with HTG (10/15) developed NOD of whom 3 developed US features of portal hypertension¹¹. In participants with a normal liver US pattern (NL), 7–13% developed NOD and 5–7.5% developed portal hypertension¹¹. Thus, CF participants with HTG had a 5.2-fold increased incidence of NOD and a 6.1-fold increased incidence of portal hypertension compared to participants with NL ¹¹.

Based on these data, the Cystic Fibrosis Liver Disease Network (CFLD-NET) undertook a multi-center study to determine if US is an effective tool to screen for the risk of the development of aCFLD. We have previously reported the interim results of this study, in which HTG, determined by at least 3 of 4 study radiologists, was associated with a 9.1-fold increased risk of development of NOD ¹². That interim analysis used more stringent criteria for radiologist consensus (3 of 4 agreeing to the grade versus 2 of 3 for this study) and occurred over a shorter period (4 years) than this report. Herein, we report the results of the analysis of the Prospective Study of Ultrasound to Predict Hepatic Cirrhosis in CF (PUSH: NCT 01144507) focusing on whether HTG can predict NOD, the cumulative risk for NOD, timing and identification of additional factors that enhance the prediction of subsequent development of NOD.

Participants and Methods

STUDY DESIGN AND PARTICIPANTS

CFLD-NET is a North American multi-center research consortium which includes 11 clinical sites and a data coordinating center. In 2010, CFLD-NET launched a prospective multicenter case-controlled cohort study to investigate the utility of abdominal US to identify young children with CF at risk for the development of aCFLD (PUSH study). The protocol was reviewed and approved by the Institutional Review Boards at all centers. All authors had access to the study data and reviewed and approved the final manuscript. Study participants were enrolled between January 2010 and February 2014. All guardians provided informed consent and appropriate assent was obtained. Children 3–12 years of age were eligible for the study based on the following inclusion criteria: (1) diagnosis of CF determined by a sweat chloride of >60 mEq/L or 2 disease-causing CFTR genetic mutations with evidence of end organ involvement; (2) enrollment in either the Cystic Fibrosis Foundation (CFF) or Toronto CF registry; and (3) diagnosis of pancreatic insufficiency. Exclusion criteria were known cirrhosis or portal hypertension (i.e., splenomegaly, ascites), prior identification of Burkholderia species on respiratory culture, or short bowel syndrome.

Data collected included demographics, growth parameters, physical findings, CFTR genotype, routine clinically indicated laboratory values and research US findings. The CFF and Toronto CF registries were used for historical and clinical data, including medical insurance, weight and height, symptoms at diagnosis, history of malnutrition, infection, lung function, complications, and medications. All data were integrated in a centralized database at the data coordinating center.

PROCEDURES

Research US details have been previously reported ^12,13. Briefly, US was performed with both gray-scale and Doppler imaging at each site. Liver echogenicity and contours were assessed to classify a participant into one of four US patterns. NL, HTG, Homogeneous (HMG) and NOD ^12,13. NL denoted normal hepatic echogenicity. HTG denoted increased echogenicity that was diffusely patchy or limited to periportal regions. Homogeneous (HMG) denoted diffusely increased hepatic parenchymal echogenicity relative to renal echogenicity, 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 echotexture of the liver parenchyma and obvious nodularity of the liver contour. Participating radiologists from each site completed web-based training for the grading of the US studies and validation as previously reported ^12,13. Sonographer training was previously reported ^12,13.

ASSESSMENTS

At study entry, each participant completed a quality-of-life survey and underwent a standardized research US that included a detailed examination of the liver and spleen and a survey examination of the entire abdomen to assess for ancillary findings.

Measurements of the longest axis of the liver and spleen, greatest diameter of the portal vein, velocities in the main portal and splenic vein and resistive indices in the hepatic and splenic artery were acquired. Each US was independently graded as NL, HTG, HMG, or NOD by 4 study radiologists consisting of the local study radiologist from each center and 3 additional (2 primary and 1 back up) study radiologists randomly assigned from the different participating study sites in regular rotation. All radiologists were blinded to the results of the other interpretations, prior US studies and clinical data. The consensus grade was assigned by majority of the local and 2 primary radiologists. In the absence of consensus among the 3 primary readers, the back-up radiologist grade was used to establish consensus (n=21, 2.9% of 723 screened participants). If 4 different grades were submitted, the participant was excluded from the study (n=1, 0.4%).

HTG participants were matched with 2 NL participants by age (±2 years), center and Pseudomonas infection status. The HTG and matched NL participants were enrolled in longitudinal follow up planned for up to 9 years which included annual evaluations with physical examination, laboratory data collection, biospecimen collection and quality of life surveys. Follow up US was planned for year 2, year 4 and year 6 (±4 months).

During the study there were a total of 1,359 US reads. Among those 58 (4.3%) required a tie break by the backup radiologist and there were 2 instances of no concordance.

ENDPOINTS

The primary endpoint for this study is the development of NOD by year 6 of follow up. Given the potential impact of early evidence of the utility of US to identify children at risk for aCFLD, a planned interim analysis was included in the study design and has been reported¹². The final year 6 US was completed in December 2020 with the last consensus US grade completed in February 2021. Laboratory and physical exam data from the visit closest to the year 6 US is referred to as data at the most recent US.

STATISTICAL ANALYSES

Summary statistics were calculated for baseline demographic, lab, physical examination, medical history features; and lab and physical examination features at the year 6 US. For testing difference between groups, Wilcoxon or t-test were used for continuous variables, and Fisher’s exact test or Chi-square test were used for categorical variables. Data presented are mean ± standard deviation for counts and percentages. For variables with skewed distributions, we also calculated median, quartiles, and range. For comparing the proportion of participants developing NOD by year 6 between groups, relative risk (RR) was calculated with a Chi-square test with type I error 0.025. Additional risk factors present at entry that might improve the prediction of development of NOD were compared to the use of US alone, including age at enrollment, gender, ethnicity, history of meconium ileus, early Pseudomonas infection, newborn screening diagnosis, prior ursodeoxycholic acid use, CFTR gene mutations, GGTP, AST, ALT, albumin, platelet count, age-adjusted spleen size z-score¹⁴, height-adjusted portal vein diameter z-score¹⁵, AST to platelet ratio index (APRI), fibrosis-4 (FIB-4), GGTP to platelet count ratio (GPR)¹⁶, height for age and weight for age z-scores, (HAZ, WAZ, https://www.cdc.gov/nccdphp/dnpao/growthcharts/resources/sas.htm). GPR was calculated using CALIPER normal values as previously described by Sellers¹⁷. Two methods for characterizing CFTR gene mutations were used. The first was a three-category variable grouping all participants based on F508del mutation presence (F508del/F508del, F508del/other and (other/other). The second was based on the classification of mutations (I-VII) ¹⁸

A logistic regression model for development of NOD at year 6 with baseline US as the single covariate was created. The above risk factors were added one at a time into this logistic regression model which we refer to as the univariate analysis. This analysis allows exploration of whether any single additional predictor can improve the prediction versus US alone. Backward selection was used with a selection criteria p<0.05 to select additional predictors for improving prediction of development of NOD. At the beginning for the backward selection process, all candidate predictors were included in the model; at each step, the variable with the largest p-value among those with p>0.05 was excluded from the model; the algorithm stopped when all covariates included in the model had p<0.05. To deal with missing covariates, we generated ten imputed data sets using IVEWARE ¹⁹ and used Rubin’s rule²⁰ to calculate standard errors and p-values.

The backward selection was based on p-values generated using Rubin’s rule. To account for censoring and determine the timeline of NOD development we calculated the NOD free survival probability over time, treating time of first NOD grade as interval-censored event time using nonparametric maximum likelihood method via EM algorithm²¹. In the first analysis we denoted time 0 as the time of the screening US and determined the NOD free survival for HTG and NL. For the second analysis, we used the age in years of the participant to develop the survival analysis to represent a lifespan risk adjusting for the various ages of entry. It is important to note that HTG and NL participants were matched for age. When using age as the time variable, left truncation was accounted for (only NOD free survival children were included in the NL and HTG groups at the age of screening US) in the observed data. Due to lack of software dealing with data in presence of both left truncation and interval censoring, the event time of converting to NOD was imputed as the midpoint between two US visits and Kaplan-Meier was used for estimating the NOD free survival probabilities.

In addition, to determine the timeframe of development of the NOD pattern in the overall CF population, a survival analysis using age as the time variable was conducted, including all HTG, NL, and HMG participants at screening US visit who were included in the longitudinal follow-up study. Because only NL participants who matched the HTG participants were enrolled into the longitudinal analysis, propensity score weighting was used to correct the selection bias. A logistic regression model including age, site, pancreatic insufficiency, and pseudomonas status as covariates was developed for modeling the probabilities of being selected into the longitudinal study among NL participants. For NL participants in the longitudinal analysis, the inversed probability of being included in the longitudinal cohort was used as their weights in this survival analysis. Similar strategies were used to deal with left truncation and interval censoring as described above. R package “survey” with robust estimator was used to provide 95% confidence interval for the estimated survival probabilities.

To determine if longitudinal changes in laboratory parameters differ between HTG participants who progressed to NOD vs. those who did not progress to NOD, the rate of change in biomarkers previously suggested to be associated with a risk of aCFLD was investigated using linear mixed effect models: APRI, platelet count, and spleen length z score. APRI was log transformed prior to this analysis.

Results

A total of 774 participants were enrolled. 722 participants had a consensus grade at screening, of whom 65 were HTG and 592 NL. Participants with baseline HMG or NOD are not included in this analysis. The consort diagram is presented in Figure 1.

Figure 1:

Consort Diagram

Legend: US: Research ultrasound, NL: Normal liver US pattern, HTG: Heterogeneous liver US pattern, HMG: Homogeneous liver US pattern, NOD: nodular liver US Pattern

Baseline data from all participants were previously reported²². All participants with at least one follow-up US were included. The final cohort for this analysis included 55 HTG and 116 NL who were also included in the interim analysis. Participants missing their year 6 US had their year 4 US utilized (n=7 (11.3%) for HTG and 11 (8.9%) for NL) and for those missing year 6 and 4, year 2 US utilized (n=1 (1.6%) for HTG and 4 (3.3%) for NL). Participant disposition is shown in Table 1. There was no significant difference in proportion of participants who completed either year 2, 4, or 6 US visit (p=0.17).

Table 1:

Participant Disposition

Status	HTG	NL
Enrolled in longitudinal follow-up study	62	123
Discontinued study before Year 2 US visit	7 (11.3%)	7 (5.7%)
Discontinued study between Year 2 and Year 4 US visit	1 (1.6%)	4 (3.3%)
Discontinued study between Year 4 and Year 6 US visit	6 (9.7%)	11 (8.9%)
Completed Year 6 US visit with consensus grade missing	1 (1.6%)	0 (0.0%)
Completed Year 6 US visit with consensus grade available	47 (75.8%)	101 (82.1%)
Completed either Year 2, 4 or 6 US visit (analysis population)	55 (88.7%)	116 (94.3%)

US = Abdominal ultrasound

HTG = Heterogeneous US pattern

NL = Normal US pattern

The baseline demographic and laboratory information are shown in Table 2. Participants were well matched for age and Pseudomonas infection status. GGTP, AST, ALT, GPR, APRI and FIB-4 were significantly higher, and platelets lower at baseline in HTG compared to NL.

Table 2:

Baseline characteristics of participants by US consensus grade at screening

Variables (median (min, max) for skewed variables)	HTG N=55	NL N=116	P value
Demographics/Matching Features
Age (years), mean (SD)	8.7 (3.2) n=55	8.4 (3.1); n=116	0.51a
Pseudomonas positive at enrollment, n (%)	14 (25.5%); n=55	28 (24.1%); n=116	0.85b
Male Gender, n (%)	35 (63.6%); n=55	58 (50.0%); n=116	0.09b
Hispanic, n (%)	0 (0.0%); n=55	7 (6.0%); n=116	0.10c
Number with year 6 US available n (%)	47 (85.5%); n=55	101 (87.1%); n=116	0.77b
Number with only up to year 4 US available n (%)	7 (12.7%); n=55	11 (9.5%); n=116	0.52b
Number with only year 2 US available n (%)	1 (1.8%); n=55	4 (3.4%); n=116	1.00c
Lab and physical exam
GGTP (U/L), mean (SD); median (min, max)	36 (35); 23 (7, 204); n=51	15 (8); 13 (4, 46); n=107	<.0001 d
AST (U/L), mean (SD); median (min, max)	45 (30); 39 (22, 236); n=54	34 (13); 31 (16, 108); n=114	<.0001 d
ALT (U/L), mean (SD); median (min, max)	42 (22); 41 (10, 97); n=51	32 (19); 27 (6, 154); n=110	0.003 d
Albumin (g/dL), mean (SD)	4.3 (0.4); n=50	4.2 (0.4); n=106	0.63d
Platelet (103/mm3), mean (SD)	301 (78); n=55	331 (69); n=114	0.01 e
Spleen size Z-score, mean (SD)	0.29 (1.67) n=55	0.01 (1.21) n=116	0.27e
Portal vein diameter Z-score, mean (SD)	0.15 (1.33) n=55	0.20 (1.19) n=114	0.79e
APRI, mean (SD); median (min, max)	0.65 (0.49); 0.56 (0.20, 3.73); n=54	0.43 (0.19); 0.39 (0.16, 1.57); n=113	<.0001 d
FIB-4, mean (SD); median (min, max)	0.22 (0.16); 0.20 (0.06, 0.99); n=51	0.16 (0.08); 0.14 (0.05, 0.38); n=108	0.007 d
GPR, mean (SD); median (min, max)	0.74 (0.69); 0.47 (0.10, 3.16); n=51	0.27 (0.14); 0.23 (0.07, 0.85); n=106	<.0001 d
Height-age Z-score, mean (SD)	−0.3 (0.8); n=55	−0.3 (0.8); n=115	0.66e
Weight-age Z-score, mean (SD)	−0.4 (0.5); n=55	−0.4 (0.4); n=115	0.23e
Medical History
Meconium ileus present, n (%)	11 (20.0%); n=55	24 (20.7%); n=116	0.92b
Newborn screen diagnosis, n (%)	14 (25.5%); n=55	28 (24.1%); n=116	0.85b
Early Pseudomonas (<= 2 years), n (%)	22 (46.8%); n=47	52 (54.2%); n=96	0.41b
F508del mutation, n (%)
Heterozygous	19 (34.5%); n=55	42 (36.2%); n=116	0.96c
Homozygous	34 (61.8%); n=55	69 (59.5%); n=116
Other	2 (3.6%); n=55	5 (4.3%); n=116
Mutation classification by PI, n (%)
Class1/Class1	2 (3.8%); n=53	3 (2.7%); n=112	0.61c
Class1/Class2	9 (17.0%); n=53	31 (27.7%); n=112
Class1/Other	0 (0.0%); n=53	1 (0.9%); n=112
Class2/Class2	38 (71.7%); n=53	70 (62.5%); n=112
Class2/Other	4 (7.5%); n=53	6 (5.4%); n=112
Other/Other	0 (0.0%); n=53	1 (0.9%); n=112

^aWilcoxon test

^bChi-square test

^cFisher’s exact test

^dTwo sample t-test based log-transformed scale

^eTwo sample t-test based on original scale

ENDPOINTS

Over the six years of follow up, there was a significant difference between HTG and NL in the development of NOD. Thirty three percent (18 of 55) of participants with baseline HTG developed NOD within 6 years compared to only 3.4% (4 of 116) with baseline NL (Table 3). This difference translates into a relative risk of 9.5 (95% confidence interval (CI): 3.4, 26.7) for the development of NOD for participants with baseline HTG.

Table 3:

Development of NOD by year 6

		Development of NOD N (%)
Consensus Grade at Screening US	N	NOD	Non-NOD	RR NOD (CI)	P value
HTG	55	18(32.7%)	37 (67.3%)	9.5(3.4, 26.7)	<0.0001
NL	116	4 (3.4%)	112 (96.6%

A finding of HTG has a sensitivity of 82% (95% CI: 66%, 98%) and specificity of 75% (95% CI: 68%, 82%) for the development of NOD with a positive predictive value of 33% (95% CI: 21%, 47%) for developing NOD. A finding of NL at baseline has a negative predictive value of 96% for developing NOD. The survival curves for remaining NOD free for HTG and NL, and the overall CF population are shown in Figure 2. The significant difference in development of NOD in HTG versus NL is evident in both 2A and B with a 50% development of NOD by 8 years after the finding of HTG (Figure 2A), with the majority identified by 15 years of age (Fig 2B). The overall population risk is shown in Figure 2C demonstrating an estimate of 10% of participants developing NOD by 16 years of age.

Figure 2:

Survival analysis for remaining NOD free comparing NL and HTG (2A) based on time from initial US or (2B) on age of development of NOD and remaining NOD free considering the entire cohort (2C)

There were important differences in laboratory and clinical data at baseline between participants with HTG who developed NOD and those that did not (Table 4). At baseline, participants with HTG who developed NOD were younger at entry, had a better WAZ score and a higher GPR compared to those with HTG who did not develop NOD. There were no significant differences in sex, pseudomonas positivity, laboratory parameters, APRI, FIB-4 or HAZ score. We did not perform a formal comparison of participants with NL who developed NOD to those who did not develop NOD as only 4 participants with NL developed NOD.

Table 4:

Baseline characteristics by baseline US grade and NOD progression

Variables (median (min, max) for skewed variables)	HTG developed NOD N=18	HTG did not develop NOD N=37	P value HTG NOD vs HTG no NOD	NL developed NOD N=4	NL did not develop NOD N=112
Demographics/Matching Features
Age (years), mean (SD)	7.3 (3.2); n=18	9.3 (3.0); n=37	0.049 a	5.9 (1.6); 6.2 (3.8, 7.2); n=4	8.5 (3.1); 9.2 (3.0, 13.4); n=112
Pseudomonas positive at enrollment, n (%)	6 (33.3%); n=18	8 (21.6%); n=37	0.51b	1 (25.0%); n=4	27 (24.1%); n=112
Male Sex, n (%)	11 (61.1%); n=18	24 (64.9%); n=37	0.77c	2 (50.0%); n=4	56 (50.0%); n=112
Hispanic, n (%)	0 (0.0%); n=18	0 (0.0%); n=37	--	0 (0.0%); n=4	7 (6.3%); n=112
Lab and physical exam
GGTP (U/L), mean (SD); median (min, max)	49 (48); 28 (10, 204); n=18	29 (22); 18 (7, 82); n=33	0.06d	19 (5); 17 (16, 26); n=4	15 (8); 13 (4, 46); n=103
AST (U/L), mean (SD); median (min, max)	47 (15); 40 (28, 74); n=18	44 (35); 36 (22, 236); n=36	0.21d	41 (19); 35 (26, 69); n=4	34 (13); 31 (16, 108); n=110
ALT (U/L), mean (SD); median (min, max)	46 (25); 46 (11, 97); n=17	40 (21); 39 (10, 95); n=34	0.44d	25 (3); 25 (22, 28); n=4	32 (19); 27 (6, 154); n=106
Albumin (g/dL), mean (SD)	4.4 (0.4); n=15	4.2 (0.4); n=35	0.16d	4.4 (0.3); n=4	4.2 (0.4); n=102
Platelet (103/mm3), mean (SD)	293 (85); n=18	304 (75); n=37	0.62e	332 (68); n=4	331 (69); n=110
Spleen size Z-score, mean (SD)	0.45 (1.20) n=18	0.21 (1.86) n=37	0.63e	−0.46 (0.76) n=4	0.03 (1.23) n=112
Portal vein diameter Z-score, mean (SD)	0.03 (1.19) n=18	0.20 (1.41) n=37	0.66e	0.55 (1.35) n=4	0.19 (1.18) n=110
APRI, mean (SD); median (min, max)	0.70 (0.29); 0.66 (0.30, 1.24); n=18	0.63 (0.57); 0.54 (0.20, 3.73); n=36	0.19d	0.50 (0.20); 0.49 (0.32, 0.71); n=4	0.43 (0.19); 0.39 (0.16, 1.57); n=109
FIB-4, mean (SD); median (min, max)	0.20 (0.11); 0.17 (0.07, 0.44); n=17	0.23 (0.18); 0.20 (0.06, 0.99); n=34	0.50d	0.14 (0.03); 0.13 (0.11, 0.18); n=4	0.16 (0.08); 0.15 (0.05, 0.38); n=104
GPR, mean (SD); median (min, max)	1.03 (0.89); 0.71 (0.13, 3.16); n=18	0.57 (0.48); 0.38 (0.10, 1.76); n=33	0.032d	0.37 (0.06); 0.37 (0.31, 0.42); n=4	0.27 (0.14); 0.23 (0.07, 0.85); n=102
Height-age Z-score, mean (SD)	0.0 (0.9); n=18	−0.4 (0.7); n=37	0.10e	−0.8 (0.3); n=4	−0.3 (0.8); n=111
Weight-age Z-score, mean (SD)	−0.1 (0.6); n=18	−0.5 (0.4); n=37	0.02e	−0.5 (0.3); n=4	−0.4 (0.4); n=111
Medical History
Meconium ileus present, n (%)	2 (11.1%); n=18	9 (24.3%); n=37	0.31b	1 (25.0%); n=4	23 (20.5%); n=112
Newborn screen diagnosis, n (%)	4 (22.2%); n=18	10 (27.0%); n=37	1.00b	3 (75.0%); n=4	25 (22.3%); n=112
Early Pseudomonas (<= 2 years), n (%)	7 (46.7%); n=15	15 (46.9%); n=32	0.99c	1 (25.0%); n=4	51 (55.4%); n=92
F508del mutation, n (%)
Heterozygous	5 (27.8%); n=18	14 (37.8%); n=37	0.60b	1 (25.0%); n=4	41 (36.6%); n=112
Homozygous	12 (66.7%); n=18	22 (59.5%); n=37		3 (75.0%); n=4	66 (58.9%); n=112
Other	1 (5.6%); n=18	1 (2.7%); n=37		0 (0.0%); n=4	5 (4.5%); n=112
Mutation classification by PI, n (%)
Class1/Class1	1 (5.6%); n=18	1 (2.9%); n=35	0.20b	0 (0.0%); n=4	3 (2.8%); n=108
Class1/Class2	5 (27.8%); n=18	4 (11.4%); n=35		1 (25.0%); n=4	30 (27.8%); n=108
Class1/Other	0 (0.0%); n=18	0 (0.0%); n=35		0 (0.0%); n=4	1 (0.9%); n=108
Class2/Class2	12 (66.7%); n=18	26 (74.3%); n=35		3 (75.0%); n=4	67 (62.0%); n=108
Class2/Other	0 (0.0%); n=18	4 (11.4%); n=35		0 (0.0%); n=4	6 (5.6%); n=108
Other/Other	0 (0.0%); n=18	0 (0.0%); n=35		0 (0.0%); n=4	1 (0.9%); n=108

^aWilcoxon test

^bFisher’s exact test

^cChi-square test

^dTwo sample t-test based on log-transformed scale

^eTwo sample t-test based on original scale

The laboratory and clinical data closest to the year 6 US for each group are shown in Table 5. HTG participants who developed NOD had higher APRI, GPR and spleen size z-scores, lower platelet count and were younger than HTG participants who did not develop NOD.

Table 5:

Last visit characteristics by baseline US grade and NOD progression

Variables (median (min, max) for skewed variables)	HTG developed NOD N=18	HTG did not develop NOD N=37	P value HTG NOD vs HTG no NOD	NL developed NOD N=4	NL did not develop NOD N=112
Age (years), mean (SD)	13.3 (3.5); n=18	15.4 (3.0); n=37	0.03 a	12.2 (1.8); n=4	14.5 (3.6); n=112
GGTP (U/L), mean (SD); median (min, max)	69 (121); 23 (11, 491); n=16	37 (38); 18.5 (7, 156); n=32	0.34a	24 (24); 13 (10, 59); n=4	16 (7); 14 (4, 45); n=103
AST (U/L), mean (SD); median (min, max)	48.5 (50); 35 (15, 226); n=16	36 (28); 25 (13, 137); n=36	0.24a	30 (11); 25 (23, 47); n=4	28.3 (13.5); 24.0 (8.0, 73.0); n=108
ALT (U/L), mean (SD); median (min, max)	45.6 (27); 34 (20, 103); n=16	43 (35); 33 (8, 161); n=36	0.44b	30 (10); 31 (17, 40); n=4	31 (19); 26 (7, 117); n=107
Albumin (g/dL), mean (SD)	4.1 (0.3); n=15	4.2 (0.5); n=35	0.56b	4.1 (0.5); n=4	4.2 (0.5); n=104
Platelet (103/mm3), mean (SD)	196 (100); n=16	271 (87); n=34	0.01 c	252 (68); n=4	311 (76); n=110
Spleen size Z-score, mean (SD)	3.85 (4.44); n=18	1.54 (2.06); n=37	0.048 c	0.78 (1.43); n=4	0.67 (1.54); n=108
Portal vein diameter Z-score, mean (SD)	1.76 (1.78); n=16	1.38 (2.22); n=34	0.55c	1.03 (1.19); n=4	0.92 (1.68); n=106
APRI, mean (SD); median (min, max)	1.37 (1.43); 0.79 (0.24, 4.86); n=16	0.65 (0.65); 0.39 (0.17, 3.20); n=34	0.02 a	0.49 (0.19); 0.47 (0.29, 0.74); n=4	0.38 (0.18); 0.35 (0.06, 0.96); n=107
FIB-4, mean (SD); median (min, max)	0.61 (0.53); 0.39 (0.17, 1.91); n=16	0.37 (0.27); 0.29 (0.13, 1.43); n=34	0.07a	0.27 (0.08); 0.27 (0.19, 0.36); n=4	0.25 (0.09); 0.25 (0.06, 0.55); n=105
GPR, mean (SD); median (min, max)	2.38 (4.07); 0.93 (0.20, 16.50); n=16	0.97 (1.61); 0.38 (0.13, 8.44); n=30	0.05 b	0.51 (0.48); 0.34 (0.14, 1.22); n=4	0.27 (0.13); 0.24 (0.06, 0.71); n=102
Height-age Z-score, mean (SD)	0.1 (0.9); n=17	−0.1 (0.9); n=36	0..50c	−0.7 (0.9); n=4	0.5 (6.8); n=110
Weight-age Z-score, mean (SD)	0.3 (0.7); n=17	−0.2 (1.0); n=34	0.08c	−0.2 (1.1); n=4	−0.1 (1.0); n=110

^aWilcoxon test

^bTwo sample t-test based on log-transformed scale

^cTwo sample t-test based on original scale

UNIVARIATE AND MULTIVARIATE ANALYSIS AND RECEIVER OPERATING CURVE

Univariate logistic regression (Table 6) showed that baseline US (HTG vs NL), age, log GGTP, log GPR, APRI, and WAZ are associated with development of NOD. FIB-4 and APRI are not independent as both use AST and platelet count in their calculation.

Table 6:

Univariate and multivariate logistic regression analysis for the selected model.

	Univariate		Selected parsimonious model
Variables	Odds Ratio Estimates	P-value	Odds Ratio Estimates	P-value
US at Screening (HTG vs. NL)	13.6 (4.3, 42.8)	<.0001	8.1 (2.1, 31.0)	0.002
Age (years)	0.8 (0.7, 0.9)	0.008	0.69 (0.55, 0.86)	0.001
Gender (female vs. male)	1.1 (0.4, 3.1)	0.81	-	-
Log(GGTP) (U/L)	2.3 (1.1, 5.0)	0.03
AST (U/L)	1.0 (1.0, 1.0)	0.56
ALT (U/L)	1.0 (1.0, 1.0)	0.55
Albumin (g/dL)	2.9 (0.7, 11.5)	0.13	-
Platelet (103/mm3)	1.0 (1.0, 1.0)	0.66	-
APRI (per 0.1 unit)	1.2 (1.0, 1.5)	0.02
Log(APRI)	4.0 (1.1, 14.4)	0.03	-	-
APRI cut point (>=0.425 vs. <0.425)	1.7 (0.6, 5.0)	0.35	-	-
Log(FIB-4)	0.7 (0.3, 1.7)	0.47	-	-
Log-GPR	2.4 (1.2, 4.9)	0.014	4.2 (1.8, 10.3)	0.001
Spleen size Z-score	1.0 (0.7, 1.4)	0.87	-	-
Portal vein diameter Z-score	1.0 (0.7, 1.4)	0.90	-	-
Height-age Z-score	1.3 (0.7, 2.5)	0.36	-	-
Weight-age Z-score	3.9 (1.2, 12.7)	0.02	-	-
Mutation classification
Class1/Class1 vs. Class2/Class2	1.5 (0.1, 19.3)	0.76	-
Class1/Class2 vs. Class2/Class2	1.8 (0.5, 6.0)	0.33	-
Other vs. Class2/Class2	<0.001 (<0.001, >999.999)	0.98	-
F508del mutation
Heterozygous vs. Homozygous	0.6 (0.2, 1.9)	0.40	-
Other vs. Homozygous	1.1 (0.1, 13.0)	0.92	-
C-index	-	-	0.90

We then constructed 2 models using multivariate logistic regression, one with APRI excluding FIB-4 and the other with FIB-4 excluding APRI. The best fit model excluded FIB-4 (predictors include US pattern, age, and log GPR) yielding an ROC curve for prediction of NOD with an area under the curve (AUC) of 0.90 in participants with baseline HTG (versus NL) (Table 6, Figure 3) compared to using baseline US only (AUC 0.78). Figure 4 shows the nomogram related to this model, which can be used to evaluate the risk of progression to NOD pattern among patients with NL and HTG US patterns. Alternatively, one can use the following equations for calculating the probability of progression to NOD pattern in six years:

$$\text{a}=1.15+2.1\times \text{\hspace{0.17em}I}\left(\text{US=HTG}\right)-0.37\times \text{\hspace{0.17em}age\hspace{0.17em}}+1.44\times \log \left(\text{e-base}\right)\text{\hspace{0.17em}GPR}$$

$$\text{Probability\hspace{0.17em}of\hspace{0.17em}progression\hspace{0.17em}to\hspace{0.17em}NOD}=\exp \left(\text{a}\right)/\left[\exp \left(\text{a}\right)+1\right]$$

Figure 3:

Receiver operating characteristics (ROC) curves

Legend: ROC curve for prediction of NOD pattern (HTG vs NL) using baseline US grade alone and age and log GPR (solid line, area under the curve=0.90)

Figure 4.

Nomogram predicting probability of developing NOD pattern in CF children in 6 years. To obtain the nomogram-predicted probability of develoing NOD pattern, locate patient values on each axis. Draw a vertical line to the “Point” axis to determine how many points should be attributed for each variable. Sum the points for all variables to obtain the total point. Locate the total point on the “Total Points” line, so that the individual probability of developing NOD pattern can be assessed on the “Risk of Progression to NOD” line. For example, a 4-year-old CF patient (~55 points) with a HTG US pattern (~32 points) and GPR 0.6 (~55 points), would have total score of ~142. Using this model, this would confer ~90% chance of progression to NOD pattern in 6 years.

I (US=HTG) = 1 for subjects with US HTG pattern and 0 for subjects with US NL pattern.

As the development of NOD is dynamic, we investigated the rate of change in key parameters over time. Figure 5 demonstrates that there are significant differences in the rate of decline in platelet count (4 A & B, p=0.0001) and increase in spleen z score (4 C & D, p<0.0001) and APRI (4 E & F, p=0.0012) over time in those participants with HTG who developed NOD compared to HTG who did not develop NOD.

Figure 5:

Longitudinal changes in selected parameters in HTG participants who progressed to NOD and those who did not progress.

Legend: Changes by individual participants in Platelet count (4A & B), Spleen size z score (4C & D) and APRI (4E & F) over time (years) for participants with HTG who developed NOD (4A, C & E) and did not develop NOD (4B, D & F). Solid line is mean population trajectory and gray area is 95% confidence band for the mean trajectory. The differences in the rates of change between groups was significant for all selected parameters

To assess the impact of a single radiologist reading (like the clinical setting) instead of consensus reading, we used the initial US grade by the site radiologist alone to determine the prediction of risk for NOD compared to consensus reading. Using only the local single radiologist reads, 40/55 (73%) of HTG and 105/116 (91%) of NL would have been identified. This would have resulted in missing 6/18 (33%) of the participants who developed NOD.

Discussion

We have demonstrated that research-based US screening of children with CF can identify children with a high risk for the development of advanced CF liver disease. We chose the development of NOD as a marker for advanced CF liver disease as the development of clinically significant portal hypertension, variceal hemorrhage and liver transplant are late events in CF. We have previously demonstrated that NOD correlates well with declining platelets, increasing spleen size and subsequent risk for varices, variceal bleeding, and liver transplantation^23,24. When this readily available tool is combined with clinical data routinely collected in the care of individuals with CF there is very good performance for the prediction of the development of a nodular liver consistent with aCFLD. The isolated finding of HTG is associated with a 9.5-fold increased risk for the development of NOD compared with NL over just a 6-year period. Indeed, survival analysis demonstrates that most of the HTG participants who develop NOD do so by 5–6 years after the identification of HTG. When using age as the time frame and considering the overall CF population, most participants who developed NOD did so between 5 and 13 years of age. In constructing a model of NOD development from these data for all participants, we found that approximately 10% would be predicted to develop NOD by 18 years of age, most of whom would develop NOD by 15 years of age. This is very consistent with large observational studies ^3,25,26 It is also notable that the finding of a NL pattern had a very low risk for the development of NOD. Thus, US can also identify children at much lower risk for the development of NOD.

At the baseline US, GPR was significantly higher in the HTG who developed NOD compared to HTG who did not develop NOD (Table 4) and remained higher using the values closest to their year 6 US (Table 5). There was a significant difference in the platelet count in the HTG who developed NOD using the values closest to their year 6 US which certainly contributes to the differences in GPR. However, at baseline, the platelet count was not different between the HTG who did or did not develop NOD. This is consistent with other data suggesting GGT and GPR may be an early marker for the risk of development of aCFLD¹⁷. Although there was a statistically significantly higher WAZ in the HTG that eventually progressed to NOD, we do not feel that it is a clinically relevant difference given that the values never ranged below -1 or above 1.5.

Univariate and multivariate analysis of our data showed that in addition to the US pattern, there are other laboratory findings that can improve the identification of individuals with CF at increased risk for developing aCFLD. With multivariate modeling, we found that the addition of age and log GPR at the time of the identification of HTG US pattern into the model significantly improve the AUC for predicting development of aCFLD. These data are also consistent with reports of higher GGTP and GPR in individuals at risk for or with aCFLD ^9,10,17.

In addition, we found significant differences in the rate of change of platelet count, APRI, and spleen size between HTG that developed NOD compared to HTG that did not. This suggests that the development of NOD is likely a dynamic process and that longitudinal changes in these parameters should be part of the evaluation of children with CF identified as at risk for the development of advanced liver disease.

A potential drawback to this work is that the research data may not immediately apply to the clinical interpretation of routine ultrasounds or predict the risk of progression to aCFLD in the clinical realm. Using individual site radiologists instead of consensus readings missed one third of the individuals with HTG who eventually developed NOD. Thus, individual clinical US readings did not perform nearly as well as consensus reads in this study performed by radiologists who underwent specific training and who were blinded to clinical data. There should be caution in translating these results, from specially trained radiologists and ultrasound technicians to the clinical setting where there are ranges in experience and more individuals are involved. In this study, we did not see differences based on the manufacturer of the local US machines, but that may also introduce variability.

While this study has identified a subset of children with CF who are at high risk for aCFLD, 50% did not develop aCFLD within 6–8 years. Adding baseline age, and log GPR to the model improved the ability to identify those at risk for aCFLD. Our data (figure 5) suggest that attention to of longitudinal changes in platelets, APRI, and spleen size could be additional parameters to identify evolving aCFLD in conjunction with US.

Our study has several limitations. No histological correlation was available for the imaging patterns that are used in the grading system as liver biopsy was not part of this study. However, NOD has long been interpreted as cirrhosis and more recently as nodular regenerative hyperplasia, both findings of advanced liver disease given their association with portal hypertension. We previously demonstrated that individuals with NOD have lower platelet counts and larger spleens, consistent with aCFLD¹². In the current study, we found that HTG participants who developed NOD had more rapid decline in platelet count and faster increases in spleen size z score and APRI, all consistent with aCFLD. Our 6 year follow up may still miss some individuals who will eventually develop aCFLD, but we believe the likelihood is small given that the mean age of the cohort at the end follow up is 14.5 years and Stonebraker et al showed that very few adolescents with aCFLD are identified after 15 years of age³. Modeling based on our data (figure 2C) is also consistent with large observational studies. Ongoing clinical follow-up of this study cohort for up to 3 additional years is planned to further define the utility of screening abdominal US and the significance of the spectrum of US findings in young children with CF into young adulthood. We did not incorporate elastography, which measures liver stiffness and has been shown to correlate with advanced hepatic fibrosis in pediatric liver diseases including CFLD^27,28 as it was not readily available at the initiation of this study. We recognize that, in the interim, elastography based imaging has been shown to be a reliable marker of advanced liver disease including CF. While liver stiffness was not included as an endpoint in the current study, we have previously reported the use of elastography in a subset of this cohort²⁹. Thus, our focus was on the use of conventional US and other readily available laboratory and clinical data, to optimize the identification of children at risk for aCFLD. In summary, we present the findings from the first large multicenter study that attempts to identify children at risk for the development of aCFLD. We have shown, in this study, that the presence of HTG on research liver US identifies a cohort of children at high risk for the development of aCFLD. A combination of baseline age, log GPR and HTG ultrasound pattern can identify children with CF at risk for aCFLD with an AUC of 0.90.

While our results will require further study in the clinical realm, they already have potential to identify children with a high risk for aCFLD in future clinical trials.

APRI=AST to platelet ratio index FIB-4=Fibrosis-4 score GPR=GGT to platelet ratio

Highlights.

• CF Children with heterogeneous ultrasound pattern of the liver have a 30–50% risk for advanced liver disease over 6 years

• Normal ultrasound pattern of the liver has good negative predictive value for advanced liver disease in CF

• A score based on US pattern, age and GGT to platelet ratio may refine the identification of individuals at high risk for aCFLD.

List of abbreviations:

CF

cystic fibrosis

CFLD

cystic fibrosis liver disease

aCFLD

advanced cystic fibrosis liver disease

CFTR

cystic fibrosis transmembrane regulator

GGTP

gamma glutamyl transpeptidase

HTG

heterogeneous liver ultrasound pattern

US

research ultrasound

NOD

nodular liver ultrasound pattern

NL

normal liver ultrasound pattern

CFLD NET

Cystic Fibrosis Liver Disease Network

CFF

Cystic Fibrosis Foundation

HMG

homogeneous liver ultrasound pattern