The prediction of liver decompensation using hepatic collagen deposition assessed by computer-assisted image analysis with Masson’s trichrome stain

Abstract Background & Aim The current pathologic system classifies structural deformation caused by hepatic fibrosis semi-quantitatively, which may lead to a disagreement among pathologists. We measured hepatic fibrosis quantitatively using collagen proportionate area (CPA) in compensated cirrhotic patients and assessed its impact on predicting the development of liver decompensation. Method From January 2010 to June 2018, we assessed 101 patients who went through liver biopsy and received diagnosis as compensated cirrhosis with digital image analysis of CPA. Clinical and laboratory data were collected at the baseline and at the time of the last follow-up or progression to liver decompensation (LD). Result The mean age was 50.8 ± 10.5 years, and the most common etiology of liver disease was chronic hepatitis B (48.5%), followed by alcoholic hepatitis (18.8%). The mean CPA was 16.91 ± 9.60%. The mean CPA values were different in patients with and without LD development (21.8 ± 11.1 vs. 15.2 ± 8.5). During the median follow-up of 60.0 months, 26 out of 101 patients experienced LD. Older age (hazard ratio [HR],1.069; p = 0.015), prolonged international normalized ratio (HR, 6.449; p = 0.019) and higher CPA (HR, 1.049; p = 0.040) were independent predictors of liver decompensation on multivariate cox-regression analysis. When patients were divided according to the optimal CPA threshold (26.8%), higher CPA predicted LD better than lower CPA. (Log-rank test: p < 0.001) Conclusion CPA could be a useful quantitative prognostic value for patients with compensated cirrhosis.


Introduction
Liver cirrhosis progresses over time from compensated to decompensated cirrhosis [1].Five to seven percent of patients with compensated liver cirrhosis progress to decompensated cirrhosis every year, and nearly half of patients with compensated develop decompensated cirrhosis within 10 years after diagnosis [2].Liver decompensation (LD) is a major contributing factor to poor prognosis in patients with end-stage liver disease [3].Accordingly, early identification of compensated cirrhotic patient who has a high risk of progressing LD is important in clinical practice.
Recently, efforts to estimate the degree of fibrosis and clinical outcomes using non-invasive markers have increased [4].Nevertheless, liver biopsy remains the gold standard for evaluating liver fibrosis and disease severity [5].Assessment with liver biopsy classified disease using structural changes and has a risk of inter-observer discrepancy among pathologists.For cirrhotic liver, histologic systems that sub-classify cirrhosis in a semi-quantitative manner, such as Laennec's [6,7] or Kumar's [8] system, have been used.However, such systems are less objective and can induce inter-observer discrepancies.
Several data evaluated the collagen proportionate area (CPA) using digital analysis of liver tissue.Their findings showed its correlation with hepatic fibrosis and proposed CPA as a prognostic marker for predicting long-term outcomes of patients with liver diseases [9][10][11][12].CPA through digital imaging analysis provides objective and quantitative estimation of hepatic fibrosis [13,14].Most previous data measured CPA using Sirius reds-stained liver tissue [9][10][11][12].Although Sirius red stain is highly specific for collagen tissues, it is rarely used in daily practice for assessing liver fibrosis.Meanwhile, Masson's trichrome stain is widely used for the evaluation of liver tissues.
To our knowledge, few data have previously investigated the predictability of CPA using Masson's trichrome stain, but no study has explored the utility of CPA as a prognostic factor in compensated cirrhotic patients [15,16].Thus, the aim of the study is to evaluate the usefulness of the CPA of Masson's trichrome stain using digital image analysis for predicting LD in patients with compensated liver cirrhosis.

Study design and patients
A retrospective review of consecutive patients who underwent liver biopsy for cirrhosis diagnosis from January 2010 to June 2018 at a single tertiary care center was conducted.The study was performed under ethics approval from the Institutional Clinical Research Ethics Committee (IRB no.2022-02-022).The inclusion criteria were patients with biopsy-proven liver cirrhosis and biopsy evaluation using Masson's trichrome stain.The exclusion criteria were (1) failure to measure CPA, (2) decompensated cirrhosis at baseline, (3) cirrhosis due to portal vein anomaly, and (4) follow-up loss.

Collection of laboratory data and serum non-invasive markers
The laboratory data including albumin, bilirubin, prothrombin time expressed as international normalized ratio (INR), aspartate transaminase (AST), alanine transaminase (ALT), platelet count, and creatinine were collected.Child-Pugh-Turcotte (CTP) score and model for end-stage liver disease (MELD) score were calculated according to the laboratory and clinical data.

Liver biopsy samples and histologic sub-classification
Core liver biopsies were obtained via ultrasound guidance using an 18-gauge automatic side-cutting core biopsy needle (ACECUT; TSK Laboratory, Tochigi, Japan).Biopsy samples were formalin-fixed, paraffin-embedded, and routinely stained with Hematoxylin and Eosin (H & E) stain and Masson's trichrome stain.Histologic assessments ensured the adequacy of the biopsy specimen, with the diagnosis confirmed by a single pathologist (M.J. Jung).The cirrhotic liver was subclassified into three groups according to the Laennec system.The Laennec system subclassifies cirrhotic livers into three groups based on severity: 4 A for mild cirrhosis with thin fibrous septa and small nodules, 4B for moderate cirrhosis with at least two broad septa and larger nodules, and 4 C for severe cirrhosis with a very broad septum or numerous tiny nodules [7].

Collagen proportionate area using digital image analysis
From Masson's trichrome staining specimen, images were captured using a microscope camera (Axiocam, Zeiss, Hertfordshire, UK), and morphometric quantifications of hepatic collagen content were assessed using a computer-assisted system (ZEN 2.3 lite software, Zeiss) by a single investigator (J.W. Park).The investigator was blinded to the biopsy results.For quantification of hepatic collagen, biopsy sections of interest were selected with x10 magnification (Figure 1).Liver capsules and large portal tracts were excluded from areas of interest.Before the digital analysis of CPA, color ranges of interest from the stained samples were adjusted and selected.The result was expressed as percentage values obtained by dividing the total area of fibrosis by the entire biopsy section.

Assessment
The primary outcome of the study was cumulative hazard for progression to LD. Occurrence of LD was defined as the development of ascites, variceal bleeding, newly diagnosed varices in a patient without prior varices, hepatic encephalopathy and more than two points increase in CTP score [18].
The secondary outcomes were as follows: (1) the correlation of CPA with the Laennec staging, CTP score and MELD score; (2) factors associated with progression to LD and adjusted hazard ratio of CPA; and (3) the predictability of CPA.

Statistical analysis
All analyses were performed using R statistical software (version 3.6.2;The R Foundation for Statistical Computing).Continuous variables were presented as mean ± standard deviation, and categorical variables were presented as absolute numbers (percentage).For group comparison, continuous variables were compared using student's t-test or Mann-Whitney test according to normal distribution, and categorical variables were analyzed using the Chi-square test or Fisher's exact test as appropriate.The correlation coefficient was calculated using the Pearson correlation coefficient.The cumulative hazard for hepatic decompensation was plotted according to the Kaplan-Meier analysis, with a comparison of the cumulative rate of events by the Log-Rank test.The observation period was from the date of the liver biopsy to the date of confirmation of decompensation or the latest follow-up for patients in whom decompensation events were not found.A multivariate Cox regression analysis was conducted to assess the effect of CPA by adjusting the potential influence of other covariates.The predictability of progression to LD was evaluated and compared using c-statistics from the Uno method [19].Statistical significance was considered as p < 0.05.

Baseline characteristics
During the study period, 121 patients with cirrhosis confirmed via biopsy specimens were included.After 20 patients were excluded according to exclusion criteria, 101 patients were available for analysis (Figure 2).The baseline characteristics of the included patients are summarized in Table 1.The mean age was 50.8 ± 10.5 years, and 52.5% of the patients were male.Chronic hepatitis B was the main etiology (48.5%), followed by alcohol consumption (10.9%).Among the hepatitis B patients, 12.2% had previously been administered nucleos(t)ide analogues, while the rest started treatment after a biopsy.CTP class A and CTP class B cirrhosis were present in 56.4% and 43.6% of patients, respectively.The mean MELD score was 9.17 ± 2.62.The distribution according to the Laennec classification was 15.8%, 67.3% and 16.8% in classes 4a, 4b and 4c, respectively.The mean total analyzed dimension was 5,785,650 ± 2,411,348 µm 2 , and the mean CPA was 1,025,336 ± 776,786 µm 2 .The mean percentage of CPA was 16.9 ± 9.6.
For patients with CTP Class B, 18 out of 44 developed LD.Neither CPA (HR, 1.00; 95% CI, 0.99-1.01)nor the Laennec stage showed significant association with progression to LD on univariate analysis (Supplementary Table 2).

Discussion
In the present study, CPA measurement predicted clinical outcomes in patients with compensated liver cirrhosis and enabled subclassification of cirrhosis according to the risk of LD progression.Meanwhile, the Laennec staging system could not predict decompensated events in patients with compensated cirrhosis.Quantification of fibrosis was better at predicting hepatic decompensation than evaluation of cirrhosis based on structural changes (c-statistics, 0.665 vs. 0.607).After the adjustment of other confounding clinical parameters, CPA was independently associated with the risk of LD progression.In clinical practice, it is important to stratify patients into high and low-risk liver complications.Utilizing CPA in clinical practice would allow the identification of patients currently with compensation, who have a high risk of developing decompensated cirrhosis.
Although it is well-validated to subclassify patients with chronic hepatitis according to the stage of fibrosis, the evidence on stratifying compensated cirrhosis according to the risk of developing decompensation has not been fully established.Previous semiquantitative staging systems, such as the Laennec scoring system, provide a non-linear, categorical assessment for cirrhosis, especially in cases with conflicting results from two staging systems.CPA alternatively provides continuous quantitative scales, which enable more sensitive and accurate assessments.CPA using Sirius red stain showed its usefulness as a prognostic indicator in various etiologies of liver diseases, such as chronic hepatitis C [10], alcoholic hepatitis [20,21] and non-alcoholic steatohepatitis [22].
In this study, unlike other previous studies, we applied Masson's trichrome stain to measure CPA.Although Masson's trichrome stain is less specific for collagen staining and is used to detect all connective tissue components, it is widely used as a routine assessment of liver fibrosis [10,16].Thus, we believe CPA using Masson's trichrome could be easily accessible in daily clinical practice despite its weakness compared to Sirius's red stain.To our knowledge, this study is the first to assess the predictability of CPA using trichrome stain for progression to decompensating events.There were few previous literatures that used Masson's trichrome stain to measure CPA.According to an article from China [15], the study measured CPA using Masson's trichrome stain in liver transplant subjects with hepatitis-B-induced decompensated cirrhosis.The measured CPA was 35.93 ± 14.42%, which was higher than the value from our study on compensated cirrhosis.The other study regarding CPA using Masson's trichrome stain from Australia [16], compared it with CPA using Sirius red stain.Another study from Australia compared Masson's trichrome stain and Sirius red stain in the evaluation of CPA.Sirius red stain showed less variability with an interclass correlation coefficient (ICC) of 0.99; in contrast, trichrome stain exhibited a moderate ICC of 0.51 to 0.66 in variability and an excellent ICC as 0.99 in reproducibility.Both stains showed good correlations with the METAvIR staging.The reported area under the receiving operating curve of CPA using Masson's trichrome stain was 0.86 for detecting cirrhosis.The mean CPA values from Masson's trichrome stain were found to be smaller than the values from Sirius red stain.
In this study, the suggested cut-off for stratifying patients was 26.8%.According to an Italian study on evaluating CPA using Sirius red stain in patients with hepatitis C cirrhosis [10], CPA showed independent prediction of LD (cut-off: 18%) and esophageal varices (cut-off: 14%).Another study from the United Kingdom that investigated alcoholic hepatitis concluded that the estimated cut-off value was 8.4%.The cut-off value suggested by our study was higher than those from the previous literature.The discrepancy in the cut-off values may be due to the differences in cohort characteristics or staining methods used in the involved studies.Identifying the optimal cut-off to stratify cirrhotic patients should be verified in future studies.
From the results of our study, a subgroup analysis that categorized patients according to the CTP classification showed no significant association with progression to LD after multivariate Cox regression analysis.This result could be due to the small sample size of each subgroup.In patients with CTP class A, the association was only verified at univariate analysis on CPA (HR, 1.10; 95% CI, 1.03-1.17).Further investigation with larger numbers is necessary.
Until recent years, the clinical use of CPA had been somewhat limited.However, CPA may soon serve as a useful indicator, as artificial intelligence (AI) has been explored and applied in various fields of clinical practice during recent years.Computer-assisted digital analysis of CPA using AI could increase reliability and efficiency, especially with Masson's trichrome stain.Future study using AI assisted digital image analysis is warranted.
Recently, transient elastography and newly developed shear wave elastography have partly replaced the role of liver biopsies [23,24] Such methods have the advantages of longitudinal and serial assessment as well as high reliability [25].Some literatures showed that the measurements of liver stiffness from transient elastography and other shear wave elastography were well-correlated with CPA [26,27].However, recent data from Taiwan [28] reported that CPA exhibited superior diagnostic performance than the performance of acoustic radiation force impulse elastography in detecting significant fibrosis.There has been a lack of data comparing CPA to elastography techniques in determining the prognosis of cirrhotic patients.
There are several limitations in this study.First, the lack of a direct comparison with other markers, such as Sirius Red staining, is a major limitation of this study.Due to the retrospective nature of our research, we only had access to scanned slides with Masson's Trichrome, making it impossible to apply alternative staining techniques.Direct comparisons would be essential and should be addressed in another large-scale study to better understand the potential differences and advantages of these staining methods.Secondly, this study was a retrospective analysis with a small number of patients.Nevertheless, we included consecutive patients with a relatively long-term follow-up.Third, the lack of evaluation on the inter-and intra-observer agreement could be another limitation.However, the CPA measurement of our study exhibited a meaningful correlation with other prognostic values, and the presented values were valid based on previous reports [15].Furthermore, reported ICC using Masson's trichrome stain from past literature was within the affordable range [16].Fourth, we did not evaluate the influence of magnification.However, according to previous data comparing x4 and x10 magnification, CPA calculated by both magnifications showed an independent prognostic role for liver-related clinical outcomes [22].Lastly, our study cohort included a wide range of cirrhosis etiologies.As the fibrosis pattern may differ between etiologies, it is important to consider these factors in the evaluation process.However, due to the limited number of patients in our study, it was not possible to perform a subgroup analysis based on etiology.This limitation further emphasizes the need for large-scale, well-designed comparative studies to thoroughly investigate the impact of different etiologies of cirrhosis in more detail.
In conclusion, CPA using Masson's trichrome stain is useful for predicting clinical outcomes in patients with compensated cirrhosis.It is expected to be a useful quantitative determination of prognosis in patients with cirrhosis.Future exploration of well-established risk-stratified classification of CPA would be required.
During a median follow-up of 60.0 (interquartile range, 31.0-92.0)months, 26 patients (25.7%) with compensated cirrhosis experienced LD.Differences in baseline characteristics are summarized in Table2.Patients who experienced LD during the study period showed higher bilirubin levels, prolongation of INR, lower platelet count, lower albumin level and higher MELD score.The mean CPA values with and without LD development were 21.8 ± 11.1 and 15.2 ± 8.5, respectively.In patients with LD progression, the Laennec staging at baseline did not show significant differences in predicting LD progression (p = 0.297).From a univariate Cox regression analysis, the risk of LD increased in patients with relatively high baseline CPA (hazard ratio [HR], 1.055; 95% confidence interval [CI], 1.017-1.094;p = 0.005).For the prediction of LD risk, the optimal threshold for CPA at the baseline was 26.8% (sensitivity, 38.5%; specificity 89.3%).The HR for developing LD with CPA larger than 26.8% was 4.16 (95% CI, 1.857-9.302).The cumulative hazard to LD development according to the CPA threshold and the baseline Laennec stage is plotted in Figures3(a) and 3(b).Thus, the baseline CPA predicted LD progression more accurately (c-statistics, 0.665) than the Laennec classification (c-statistics, 0.607).

Figure 3 .
Figure 3. Cumulative hazard for progression to decompensated cirrhosis according to the baseline collagen proportionate area and the Laennec staging.

Table 2 .
comparsion of characteristics for patients with and without progression of liver decompensation.

Table 3 .
factors associated with progression to decompensated cirrhosis.