Combined Liver Stiffness and Α-fetoprotein Further beyond the Sustained Virologic Response Visit as Predictors of Long-Term Liver-Related Events in Patients with Chronic Hepatitis C

Aims Long-term risk stratification using combined liver stiffness (LS) and clinically relevant blood tests acquired at the baseline further beyond the sustained virologic response (SVR) visit for chronic hepatitis C (CHC) has not been thoroughly investigated. This study retrospectively investigated the prognostics of liver-related events (LREs) further beyond the SVR visit. Methods Cox regression and random forest models identified the key factors, including longitudinal LS and noninvasive test results, that could predict LREs, including hepatocellular carcinoma, during prespecified follow-ups from 2010 to 2021. Kaplan–Meier survival analysis estimated the significance of between-group risk stratification. Results Of the entire eligible cohort (n = 520) of CHC patients with SVR to antiviral therapy, 28 (5.4%) patients developed post-SVR LREs over a median follow-up period of 6.1 years (interquartile range = 3.5–8.7). The multivariate Cox regression analysis identified two significant predictors of LREs after the year 3 post-SVR (Y3PSVR) baseline (LRE, n = 15 of 28, 53.6%, median follow-up = 4.1 [1.6–6.4] years after Y3PSVR): LS at Y3PSVR (adjusted hazard ratio [aHR] = 3.980, 95% confidence interval [CI] = 2.085–7.597, P < 0.001), and α-fetoprotein (AFP) at Y3PSVR (aHR = 1.017, 95% CI = 1.001–1.034, P=0.034). LS ≥1.45 m/s and AFP ≥3.00 ng/mL for Y3PSVR yielded positive likelihood ratios of 4.24 and 2.62, respectively. Kaplan–Meier analysis revealed that among the stratified subgroups, the subgroup with concurrent LS ≥1.45 m/s and AFP ≥3.00 ng/mL at Y3PSVR exhibited the highest risk of LREs after Y3PSVR (log-rank P < 0.001). Conclusion We recommend the combined use of concurrent LS and AFP in future prediction models for LREs in CHC. Patients with concurrently high LS and AFP values further beyond the SVR visit may require a recall policy involving intense surveillance.


Introduction
Global hepatitis C eradication is a major healthcare challenge. Viral eradication based on interferon (IFN) and direct-acting antiviral (DAA) agents benefits patient prognosis by reducing the incidence of extrahepatic morbidity [1], hepatocellular carcinoma (HCC) [2,3], and liver-related events (LREs) [4]. However, HCC and hepatic decompensation can occur after viral eradication, especially in patients with post-treatment cirrhosis or advanced chronic liver disease [3,5]. erefore, posteradication surveillance is essential for patients with chronic hepatitis C (CHC) [6]. e incidence of HCC and LREs after sustained virologic response (SVR) has been widely investigated. Current surveillance programs typically rely on the use of liver elastography [7] and noninvasive tests (NITs) of blood-based markers [8] as baseline predictors. e risks of HCC and LREs can be stratified by values and decline in NITs from antiviral treatment baseline to on-treatment or off-treatment follow-ups [3,5,[8][9][10]. Moreover, the liver stiffness (LS) value indicated by ultrasound measurement is a promising biomarker for risk stratification for antiviral treatment from baseline [11], after SVR [12][13][14], and before HCC treatment [15,16]. Studies have revealed that the declining kinetics of LS values during treatment reflect the early resolution of hepatic necroinflammation and late, sustained regression of post-treatment fibrosis burden [17]. Post-SVR LS values, which may mitigate the confounding effects of hepatic necroinflammation, have been recommended for use as a baseline for post-SVR follow-up [18,19].
e existing liver fibrosis burden may contribute to the common and pivotal upstream pathways that drive both hepatic carcinogenesis and decompensation [20][21][22]. However, the remaining post-treatment fibrosis burden, which can contribute to morbidity and mortality, typically declines over time [23,24]. e optimal threshold values of the predictors acquired further beyond the SVR visit for risk stratification and recall policies [25] remain unclear.
Long-term risk stratification using combined LS and clinically recommended, acceptable, and available blood tests acquired at the baseline further beyond the SVR visit has not been thoroughly investigated. erefore, this study used prespecified protocols to obtain LS values and NIT results further beyond the SVR visit to retrospectively investigate the utility of longitudinal measurements of precancerous LS and NITs in the prognostics of post-SVR LREs from two baselines (SVR and year 3 post-SVR [Y3PSVR]).

Participants.
Patients with hepatitis C who had achieved an SVR to IFN-or DAA-based therapy and had completed at least two visits for LS measurements at China Medical University Hospital from August 2011 to November 2021 were retrospectively screened. SVR was defined as the absence of a detectable hepatitis C virus (HCV) RNA load on blood test results from 6 months or 3 months after the end of treatment (EOT) with IFN-or DAA-based therapy, respectively. Written informed consent was obtained from all participants. e study protocol, including the use of serial elastography, was approved by the Research Ethics Committee of the medical center (CMUH107-REC3-01). e protocol followed the principles of the Declaration of Helsinki of 1975.
HCV infection was detected using serum HCV RNA (detection limit: 15 IU/mL; COBAS Ampliprep/COBAS TaqMan HCV test, Roche Diagnostics, Branchburg, NJ, USA). e patient exclusion criteria at screening were age <20 years, HCC, coinfection with hepatitis B virus or human immunodeficiency virus, alcohol dependence (score ≥2 on the CAGE questionnaire) [26], and a serum creatinine level of >2.5 mg/dL.

Blood Tests.
e regular on-and off-treatment blood tests, including the serological α-fetoprotein (AFP) test and the tests used for calculating the aspartate aminotransferase (AST)-to-platelet ratio index (APRI) and the Fibrosis-4 index (FIB-4) [17] as concurrent noninvasive liver fibrosis indices, were performed in the central laboratory of the hospital.
After EOT, each patient underwent scheduled blood tests, including the AFP test, every 3 to 6 months.

Clinical Cirrhosis.
Cirrhosis was clinically diagnosed at treatment baseline (TW0) according to the following criteria: a score of ≥8 in parenchymal liver disease on the 2D Bmode ultrasound [27], supplemented by the presence of ascites or gastroesophageal varices, or the stage of META-VIR F4 in liver histology [24], if available.

LS Measurements.
e patients underwent scheduled prespecified LS measurements after 4 h of fasting at TW0, EOT, 12 weeks and 24 weeks after EOT, and every 6 to 12 months thereafter.
Point shear wave elastography is integrated into an ultrasound system (Acuson S2000 with a Siemens 4C1 curvilinear array, 2.67 MHz for push pulses, and 3.08 MHz for detection pulses; Siemens Healthineers, Erlangen, Germany) to measure the patients' LS.
One hepatologist (S. H. C.), who had 10 years of experience in LS measurement using point shear-wave elastography and was blinded to the patient data, performed all LS measurements through the intercostal approach. e measurement results (in meters per second [m/s] for the shear-wave velocities) were deemed reliable when the interquartile range (IQR) was <0.15 of the median of 10 successful LS measurements. e successful LS measurement rate was >60%. All other results were deemed unreliable and were excluded [28].

Endpoint Definitions.
e LRE of interest was defined as the first de novo occurrence of any LRE. e LREs considered in this study were HCC and liver decompensation after the SVR visit (post-SVR) and after year 3 post-SVR (post-Y3PSVR). e liver decompensation events recorded were portal hypertensive gastrointestinal bleeding (diagnosed by endoscopy), portal hypertension-related ascites (diagnosed through paracentesis or imaging), spontaneous bacterial peritonitis (diagnosed through paracentesis), hepatic encephalopathy (diagnosed clinically), hepatorenal syndrome (diagnosed clinically), acute-on-chronic liver failure necessitating transplantation (diagnosed clinically), and liver-related mortality (diagnosed clinically). HCC was diagnosed according to the criteria of the American Association for the Study of Liver Diseases [29]. HCC was diagnosed following pathologic confirmation or by at least one imaging technique (contrast-enhanced computed tomography, contrast-enhanced magnetic resonance imaging, or contrast-enhanced ultrasound) for liver nodules at least 1 cm in diameter in patients with cirrhosis. e data of the patients who dropped out during the follow-up period were censored at the time they were lost to follow-up. 2 Canadian Journal of Gastroenterology and Hepatology 2.6. Statistical Analysis.
e data are presented as medians (IQRs) or numbers (percentages). Between-group and overall differences were estimated using the Mann-Whitney U and Kruskal-Wallis tests for continuous variables and the chi-square or Fisher's exact test for categorical variables between unpaired or independent groups. Paired parameters from baseline to follow-up visits were compared using the Wilcoxon signed-rank test for continuous variables.
Cox regression models were used to identify the significant predictors of specific LREs after the SVR baseline in the entire cohort and to estimate these factors after the Y3PSVR baseline in the post-Y3PSVR subcohort. e participants with missing data were excluded from the modeling. Case numbers are disclosed for each model. Collinearity between the covariates was evaluated using a variance inflation factor (VIF).
A single decision tree classifier and an ensemble random forest model [30] were preliminarily trained with no validation to assist in predictor identification. Machine learning (ML) used the significant predictors identified using the Cox regression model to predict the binary occurrence (yes or no) of LREs, thereby minimizing the information loss. e performance of the trained model was evaluated using the area under the receiver operating characteristic (ROC) curve To evaluate the clinical utility of the significant predictors in the prediction of LREs, a non-time-dependent ROC analysis was conducted to estimate the discriminative performance of each individual predictor. e AUC, sensitivity, specificity, predictive values, and likelihood ratios based on the cutoff values that maximized the Youden index were recorded. e significance levels of the differences in AUCs were evaluated using the DeLong test. Kaplan-Meier survival analysis was conducted to estimate the significance of between-group risk stratification using predictors classified according to the cutoffs.
e key predictors were grouped into combination classes through the Kaplan-Meier analysis to enhance the predictive performance and utility of the model. e significance levels were compared using a log-rank test. e data were analyzed using SPSS version 22.0 for Windows (IBM, Armonk, NY, USA). A two-sided P value <0.05 indicated statistical significance.

Participant Characteristics.
Each of the 520 eligible participants with an SVR to HCV monoinfection underwent at least one session of LS measurements (Figure 1). e median age was 56 years (IQR, 48-63), and 260 (50%) of the participants were male.

LRE Prediction Post-SVR.
After adjustment for baseline patient parameters acquired at the SVR visit, the multivariate Cox regression analysis identified the following three Canadian Journal of Gastroenterology and Hepatology  Table 2). e median post-SVR follow-up period was 6.1 years. e results of Kaplan-Meier survival analyses of clinical cirrhosis (TW0), age (TW0), LS (SVR), and AFP (SVR) are presented in Figures 2(a)-2(d), respectively. After SVR, 5 LREs (7.9%) developed in patients with cirrhosis (TW0) (n � 63) and 23 (5.0%) in those without cirrhosis (TW0) (n � 457). Clinical cirrhosis at baseline did not significantly affect the stratification of the risk prediction for post-SVR LREs (log-rank P � 0.221) (Figure 2(a)).
e single decision tree classifier model used LS (SVR), age, AFP (TW0), and AFP (SVR) to preliminarily classify the occurrence (yes or no) of LREs for each patient ( Figure S1). e average performance indices of the trained decision tree model in the prediction of LREs were as follows: AUC � 0.790, accuracy � 0.967, F1 � 0.960, precision � 0.968, and recall � 0.967. e performance indices of the trained random forest model in the prediction of LREs were as follows: AUC � 0.911, accuracy � 0.952, F1 � 0.933, precision � 0.954, and recall � 0.952 (Table 3, Figure S2).
e Kaplan-Meier analyses ( Figure 3) identified two classes: high-risk and low-risk (5/32 and 7/260 LREs, respectively) based on Y3PSVR LS and AFP values (log-rank P < 0.001) to optimize the stratification of the post-Y3PSVR risk of LREs. e high-risk subgroup (LS ≥1.45 m/s and AFP ≥3.00 ng/mL at Y3PSVR) exhibited a higher risk of LREs after Y3PSVR. e proportion of patients with LREs after Y3PSVR was 15.6% (5/32), with an incidence rate of 3.4 per 100 person-years and a 5-year cumulative incidence rate of 21.3%. By contrast, the low-risk subgroup (AFP <3.00 ng/ mL or LS <1.45 m/s at Y3PSVR) exhibited a lower risk of LREs after Y3PSVR. e proportion of patients with LREs post-Y3PSVR was 2.7% (7/260), with an incidence rate of 0.5 per 100 person-years and a 5-year cumulative incidence rate of 2.1%.
While validating the Y3PSVR thresholds of LS and AFP, the results of Kaplan-Meier analyses were both significant in the group with cirrhosis at baseline (n � 63) (P � 0.042, Figure 3(f )) and in that without cirrhosis at the baseline (n � 457) (P � 0.006, Figure 3(g)), respectively.

Discussion
Surveillance using prediction factors over time may assist in identifying growing post-treatment patient populations at high or minimal risk of LREs. Nevertheless, risk stratification according to the baseline values (measurement of noninvasive markers) further beyond the SVR visit for CHC has received insufficient attention.
e optimal threshold values of the predictors acquired further beyond the SVR visit for risk stratification remain unclear. e present study identified LS and AFP as key predictors of post-SVR LREs ( Table 2). e predictor values acquired at Y3PSVR were subsequently applied to evaluate the discriminative performance in the prediction of LREs further beyond the SVR visit (post-Y3PSVR) in the subcohort with available predictor values at the Y3PSVR baseline (Table 4). e incidences of LREs after the SVR and Y3PSVR baselines were not zero (Table 1). e residual post-treatment liver fibrosis burden may contribute to the common and pivotal upstream mechanisms that drive the development of LREs related to hepatocarcinogenesis and portal hypertension. ese LREs remain emerging problems in the growing population of patients achieving viral eradication. After viral eradication, HCC risk gene signatures [20] may not be fully reversed in patients who have developed HCC. Previous studies have also revealed that epigenetic alterations persist and predispose patients to hepatocarcinogenesis after SVR. ese remaining epigenetic signatures, including HCV protein-associated post-translational modifications of histones, may constitute the treatment targets for post-SVR carcinogenic risk reduction [21].
Post-treatment LS values may reflect the burden of remnant liver fibrosis. Post-treatment LS values may change over time by mitigating the confounding effects of hepatic necroinflammation on LS measurements [31]. A study revealed that post-treatment AFP >5.4 ng/mL is a predictor of a lack of histological fibrosis regression [9]. However, these predictors, which serve as surrogate markers of liver fibrosis, exhibit changing and declining kinetics across the treatment timeline [9,17,24]. A study reported that, compared with pretreatment liver fibrosis, post-treatment liver fibrosis is a more accurate prognostic marker of LREs, including HCC [23]. In the present study, the clinical diagnosis of cirrhosis at treatment baseline was not sufficiently discriminatory for post-SVR or post-Y3PSVR risk stratification to predict LREs after treatment completion (Tables 2  and 4; Figures 2(a) and 3(a)). In other words, the absence of cirrhosis at baseline did not preclude the need for LRE surveillance. Overall, 23 (5.0%) and 12 (3.3%) patients without baseline cirrhosis developed LREs after SVR and Y3PSVR, respectively (Figures 2(a) and 3(a)). e significant declines in LS and AFP values from the SVR visit to Y3PSVR imply the importance of validating the utility of predictor values acquired at Y3PSVR as baseline values in post-Y3PSVR prognostics (Table 4; Figure 3). Furthermore, the AUC of LS (Y3PSVR) being numerically higher than that of LS (SVR) is consistent with the results of a previous study [23]. LS and AFP values obtained at Y3PSVR may be incorporated into models for predicting post-Y3PSVR LREs to improve the efficacy and cost-effectiveness of recall policies. For patients who did not complete early on-or off-treatment surveillance sessions through blood tests, histology, ultrasound, or LS measurements, the measurements of Y3PSVR may compensate for the missed surveillance, thereby enabling the continuation of long-term surveillance. Regardless of the availability of data from early on-and off-treatment surveillance sessions, concurrently high LS and AFP values indicating leftover liver fibrosis and regeneration burden at Y3PSVR alert both physicians and patients that continuing or starting careful long-term surveillance is necessary. e thresholds of precancerous AFP in the present study were much lower than those for HCC detection in naïve or posteradication settings [32,33]. AFP, an oncofetal glycoprotein discovered decades ago, has also been used as a serum and gene expression marker to evaluate hepatocyte regeneration activity closely related to concurrent hepatic necroinflammation or damage predisposing the liver to progression to advanced fibrosis, cirrhosis, or HCC [34]. Growing evidence has revealed that mildly elevated AFP levels (<150 ng/mL) might reflect the status of liver regeneration and damage [35,36]. AFP may be used as an additional predictor of post-SVR LREs to enhance the performance of predictive models for liver-related pathogenesis, which may not be explained by liver fibrosis or its surrogate markers alone [3]. e present study has several limitations. Firstly, NIT results, including APRI and FIB-4, were not included in the proposed predictive models for LREs because they were already known to exhibit promising predictive performance [8], and they exhibited significant collinearity with concurrent LS values in the present study. e declines in LS from the SVR visit to Y3PSVR were also excluded. e utility of absolute values of biomarkers, such as LS and AFP,       acquired at post-treatment visits instead of the delta (decline) values over time may facilitate the generalization of simple noninvasive surveillance programs, with no need for calculations in clinical settings, for the risk stratification of post-treatment LREs. In the Asia-Pacific region, guidelines [33] recommend the assessment of serum AFP levels on and off antiviral treatment every 6 months for the HCC surveillance of patients with chronic liver diseases. By contrast, regular and longitudinal measurements of post-SVR patient parameters, such as platelet count, have not been strongly recommended. erefore, NIT results, such as platelet-based indices, had a high rate of missing data in the present cohort, in both the post-SVR cohort and post-Y3PSVR subcohort. Secondly, ML models may exhibit higher accuracy than do regression models [3]. Preliminary ML-based algorithms were, therefore, applied to assist in identifying the key predictors of LREs in the present study. However, the limited number of post-Y3PSVR LREs hampered the model training and optimization. irdly, the values for the same markers acquired at different visits (e.g., LS at SVR and LS at Y3PSVR) were not simultaneously included in the same Cox regression modeling because of the dependence and collinearity between the same markers. In addition, the values collected at the time of SVR were not used as the baseline for post-Y3PSVR surveillance. Fourth, baseline age, which was one of the significant covariates identified through Cox regression analysis for post-SVR LRE prediction (Table 2), did not retain significance for post-Y3PSVR LRE prediction (Table 4). erefore, age was not included in the subsequent Kaplan-Meier analysis (Figure 3). Studies involving larger cohorts are warranted to reappraise the significance of age in prognostics further beyond the SVR visit. Fifth, longitudinal spleen stiffness values closely reflect changes in portal hypertension in patients with advanced chronic liver disease treated with DAA [37]. Clinically significant portal hypertension and related LREs may not be radically abolished, despite SVR or a decrease in LS [4,37]. Future LRE-predicting algorithms should employ a combination [38] of LS and spleen stiffness measurements, including delta values [39,40], to maximize the diagnostic performances, as well as by considering remnant portal hypertension after antiviral treatment completion. Sixth, concerns and controversies regarding differences in HCC risk after viral eradication by IFN-or DAA-based therapies have been clarified. Recent meta-analyses have revealed no significant differences [2]. erefore, the present study combined both IFN-or DAA-based cohorts despite their distinct characteristics (Table 1). Seventh, censoring may lead to biased estimates during the follow-up period and potential overestimation of LREs after Y3PSVR. Future studies may also need to evaluate LRE recurrence to optimize risk estimation for longer terms after SVR. Eighth, the LREs may have been subclinical before the first clinical identification or diagnosis [41]. Ninth, clinical cirrhosis was only evaluated at treatment baseline rather than at the two points after treatment completion: SVR and Y3PSVR. Lastly, the utility of the identified predictors can benefit surveillance by facilitating the identification of patients at high risk for LREs further beyond the SVR visit, for whom recall policies should be implemented. Additional studies are warranted to establish a prespecified actionable recall policy involving additional and more intensive return visits and diagnostic tests and to measure its cost-effectiveness.

Conclusion
We recommend the combined utility of concurrent LS and AFP acquired further beyond the SVR visit in future models for predicting the risk of LREs further beyond the SVR visit in patients with CHC. Patients with concurrently high LS and AFP values should undergo prespecified lifelong surveillance after viral eradication. Future studies should employ a larger cohort to validate the results of the present study, develop a prespecified recall policy, and evaluate the cost-effectiveness of the policy.