A machine learning model to predict liver-related outcomes after the functional cure of chronic hepatitis B.

BACKGROUND & AIMS: The risk of hepatocellular carcinoma (HCC) and hepatic decompensation persists after hepatitis B surface antigen (HBsAg) seroclearance. This study aimed to develop and validate a machine learning model to predict the risk of liver-related outcomes (LROs) following HBsAg seroclearance. METHODS: A total of 4,787 consecutive patients who achieved HBsAg seroclearance between 2000 and 2022 were enrolled from 6 centers in South Korea and a territory-wide database in Hong Kong, comprising the training (n=944), internal validation (n=1,102), and external validation (n=2,741) cohorts. Three machine learning-based models were developed and compared in each cohort. The primary outcome was the development of any LRO, including HCC, decompensation, and liver-related death. RESULTS: During a median follow-up of 55.2 (interquartile range=30.1-92.3) months, 123 LROs were confirmed (1.1%/person-year) in the Korean cohort. A model with the best predictive performance in the training cohort was selected as the final model (designated as PLAN-B-CURE), which was constructed using a gradient boosting algorithm and 7 variables (age, sex, diabetes, alcohol consumption, cirrhosis, albumin, and platelet count). Compared to previous HCC prediction models, PLAN-B-CURE showed significantly superior accuracy in the training cohort (c-index: 0.82 vs. 0.63-0.70, all P<0.001; area under the receiver operating characteristic curve: 0.86 vs. 0.62-0.72, all P<0.01; area under the precision-recall curve: 0.53 vs. 0.13-0.29, all P<0.01). PLAN-B-CURE showed a reliable calibration function (Hosmer-Lemeshow test P>0.05) and these results were reproduced in the internal and external validation cohorts. CONCLUSION: This novel machine learning model consisting of 7 variables provides reliable risk prediction of LRO after HBsAg seroclearance that can be used for personalized surveillance.

Dynamic Changes in ELF Score Predict Hepatocellular Carcinoma in Chronic Hepatitis B Patients Receiving Antiviral Treatment.

Enhanced liver fibrosis (ELF) score is a noninvasive assessment for liver fibrosis. We aimed to evaluate the performance of changes in ELF score 3 years apart in combination with liver stiffness measurement (LSM)-hepatocellular carcinoma (HCC) score to predict HCC in chronic hepatitis B (CHB) patients. This is a prospective cohort study. Patients who underwent transient elastography (TE) examinations and at intermediate or high risk of HCC defined by LSM-HCC score were invited to repeat the examination about 3 years later. Their serum samples at these two time points were retrieved to assess the ELF score changes. The primary endpoint was HCC. There were 445 CHB patients (males: 73.9%; mean age: 51.6 ± 10.3 years) who received two TE examinations and ELF scores. Among them, 252 (56.6%) and 193 (43.4%) patients were at intermediate and high HCC risk at first assessment defined by LSM-HCC score, respectively. Kaplan-Meier analysis showed that the changes in ELF score could stratify the HCC risk in both intermediate- and high-risk patients defined by LSM-HCC score (p < 0.001 for intermediate-risk group; p = 0.011 for high-risk group). Patients remained having mild or moderate fibrosis at both assessments had the lowest risk of HCC (4.0%), followed by patients with fibrosis regressed (11.3%; p = 0.014) during a mean follow-up of 163 months. Patients remained having or progressed to severe fibrosis were at highest risk of HCC (>20%). Consistent findings were demonstrated in patients at both intermediate and high risk of HCC defined by LSM-HCC score. Dynamic changes in ELF score provided additional value to LSM-HCC score for stratifying HCC risk in CHB patients.

Long-term liver-related outcomes and liver stiffness progression of statin usage in steatotic liver disease.

BACKGROUND: Statins have multiple benefits in patients with metabolic-associated steatotic liver disease (MASLD). AIM: To explore the effects of statins on the long-term risk of all-cause mortality, liver-related clinical events (LREs) and liver stiffness progression in patients with MASLD. METHODS: This cohort study collected data on patients with MASLD undergoing at least two vibration-controlled transient elastography examinations at 16 tertiary referral centres. Cox regression analysis was performed to examine the association between statin usage and long-term risk of all-cause mortality and LREs stratified by compensated advanced chronic liver disease (cACLD): baseline liver stiffness measurement (LSM) of ≥10 kPa. Liver stiffness progression was defined as an LSM increase of ≥20% for cACLD and from <10 kPa to ≥10 or LSM for non-cACLD. Liver stiffness regression was defined as LSM reduction from ≥10 kPa to <10 or LSM decrease of ≥20% for cACLD. RESULTS: We followed up 7988 patients with baseline LSM 5.9 kPa (IQR 4.6-8.2) for a median of 4.6 years. At baseline, 40.5% of patients used statins, and cACLD was present in 17%. Statin usage was significantly associated with a lower risk of all-cause mortality (adjusted HR=0.233; 95% CI 0.127 to 0.426) and LREs (adjusted HR=0.380; 95% CI 0.268 to 0.539). Statin usage was also associated with lower liver stiffness progression rates in cACLD (HR=0.542; 95% CI 0.389 to 0.755) and non-cACLD (adjusted HR=0.450; 95% CI 0.342 to 0.592), but not with liver stiffness regression (adjusted HR=0.914; 95% CI 0.778 to 1.074). CONCLUSIONS: Statin usage was associated with a relatively lower long-term risk of all-cause mortality, LREs and liver stiffness progression in patients with MASLD.

Baseline Viral Load and On-treatment Hepatocellular Carcinoma Risk in Chronic Hepatitis B: A Multinational Cohort Study.

BACKGROUND & AIMS: Hepatocellular carcinoma (HCC) risk persists in patients with chronic hepatitis B (CHB) despite antiviral therapy. The relationship between pre-treatment baseline hepatitis B virus (HBV) viral load and HCC risk during antiviral treatment remains uncertain. METHODS: This multinational cohort study aimed to investigate the association between baseline HBV viral load and on-treatment HCC risk in 20,826 noncirrhotic, hepatitis B e antigen (HBeAg)-positive and HBeAg-negative patients with baseline HBV DNA levels ≥2000 IU/mL (3.30 log10 IU/mL) who initiated entecavir or tenofovir treatment. The primary outcome was on-treatment HCC incidence, stratified by baseline HBV viral load as a categorical variable. RESULTS: In total, 663 patients developed HCC over a median follow-up of 4.1 years, with an incidence rate of 0.81 per 100 person-years (95% confidence interval [CI], 0.75-0.87). Baseline HBV viral load was significantly associated with HCC risk in a non-linear parabolic pattern, independent of other factors. Patients with baseline viral load between 6.00 and 7.00 log10 IU/mL had the highest on-treatment HCC risk (adjusted hazard ratio, 4.28; 95% CI, 2.15-8.52; P < .0001) compared with those with baseline viral load ≥8.00 log10 IU/mL, who exhibited the lowest HCC risk. CONCLUSION: Baseline viral load showed a significant, non-linear, parabolic association with HCC risk during antiviral treatment in noncirrhotic patients with CHB. Early initiation of antiviral treatment based on HBV viral load may help prevent irreversible HCC risk accumulation in patients with CHB.

The effectiveness of magnetic resonance imaging (MRI) iron corrected T1 in monitoring metabolic dysfunction-associated steatohepatitis in obesity following bariatric surgery and lifestyle modification: a prospective cohort study.

Background: Bariatric surgery and lifestyle modification are important treatments for obesity, a risk factor for metabolic dysfunction-associated steatohepatitis (MASH). Studies have related weight reduction with changes in MASH, however, few have used imaging to investigate effects on liver health. We evaluated differences in liver response to obesity treatment using disease activity iron corrected T1 (cT1) and proton density fat fraction (PDFF) in patients with both obesity and metabolic dysfunction-associated steatotic liver disease (MASLD). Methods: Thirty-four patients with obesity and MASLD were recruited between March 2019 to February 2022 from a tertiary hospital in this longitudinal study; 13 underwent laparoscopic sleeve gastrectomy (LSG) alongside intraoperative liver biopsy, and 21 underwent a 4-month lifestyle modification program (LMP). All patients had multi-parametric magnetic resonance imaging (MRI) at baseline and 4-months. Diagnostic accuracy to identify MASH was assessed using the area under receiver operating characteristic (AUROC) curve. Results: Four (31%) of patients in the LSG group had MASH [non-alcoholic steatohepatitis (NAS) activity score ≥4] on liver biopsy. PDFF and cT1 correlated with the NAS activity score [r=0.81, 95% confidence interval (CI): 0.453 to 0.943, P<0.001] and (r=0.70, 95% CI: 0.228 to 0.907, P=0.008, respectively). There was good AUROC curve for cT1 (0.89, 95% CI: 0.67 to 1.00, P=0.031) and PDFF (0.83, 95% CI: 0.57 to 1.00, P=0.064) to identify MASH. At follow-up, weight reduction -22.8% (P=0.013) vs. -1.3% (P=0.262) resulted in cT1 reduction of -8.04% (864 ms, P=0.025) vs. -3.87% (907 ms, P=0.083) in the LSG vs. LMP group, respectively. Significant differences between interventions were observed for percentage PDFF decrease (-64.52% vs. -29.16%, P=0.001). Both biomarkers were significantly reduced in the LSG group (cT1 by -8.04%, P=0.025, PDFF by -64.52%, P=0.012), while only PDFF (-29.16%, P=0.012) was significantly reduced in the LMP group. Conclusions: MRI biomarkers may have some utility to monitor MASH following intervention in patients with obesity allowing objective comparison between intervention strategies. Compared to LMP, LSG was more effective in improving liver health.

Role of noninvasive tests in the prognostication of metabolic dysfunction-associated steatotic liver disease.

In managing metabolic dysfunction-associated steatotic liver disease, which affects over 30% of the general population, effective noninvasive biomarkers for assessing disease severity, monitoring disease progression, predicting the development of liver-related complications, and assessing treatment response are crucial. The advantage of simple fibrosis scores lies in their widespread accessibility through routinely performed blood tests and extensive validation in different clinical settings. They have shown reasonable accuracy in diagnosing advanced fibrosis and good performance in excluding the majority of patients with a low risk of liver-related complications. Among patients with elevated serum fibrosis scores, a more specific fibrosis and imaging biomarker has proved useful to accurately identify patients at risk of liver-related complications. Among specific fibrosis blood biomarkers, enhanced liver fibrosis is the most widely utilized and has been approved in the United States as a prognostic biomarker. For imaging biomarkers, the availability of vibration-controlled transient elastography has been largely improved over the past years, enabling the use of liver stiffness measurement (LSM) for accurate assessment of significant and advanced fibrosis, and cirrhosis. Combining LSM with other routinely available blood tests enhances the ability to diagnose at-risk metabolic dysfunction-associated steatohepatitis; and predict liver-related complications, some reaching an accuracy comparable to that of liver biopsy. Magnetic resonance imaging-based modalities provide the most accurate quantification of liver fibrosis, though the current utilization is limited to research settings. Expanding their future use in clinical practice depends on factors such as cost and facility availability.

Serum PRO-C3 is useful for risk prediction and fibrosis assessment in MAFLD with chronic kidney disease in an Asian cohort.

BACKGROUND: Metabolic dysfunction-associated fatty liver disease (MAFLD) is an emerging risk factor for chronic kidney disease (CKD). N-terminal propeptide of collagen type 3 (PRO-C3) is a biomarker of advanced fibrosis in MAFLD and PRO-C3 may be involved in renal fibrosis. We aimed to use PRO-C3 measurements to generate a new algorithmic score to test the prediction of MAFLD with chronic kidney disease (MAFLD-CKD). METHODS: A derivation and independent validation cohort of 750 and 129 Asian patients with biopsy-confirmed MAFLD were included. Serum PRO-C3 concentration was measured and regression analyses were performed to examine associations with MAFLD-CKD. A derivative algorithm for MAFLD-CKD risk prediction was evaluated with receiver operator characteristic (ROC) curve analysis. RESULTS: The study included two Asian cohorts (n = 180 with MAFLD-CKD; mean-eGFR: 94.93 mL/min/1.73 m2; median-urinary albumin-to-creatinine ratio: 6.58 mg/mmol). PRO-C3 was associated with the severity of MAFLD-CKD and independently associated with MAFLD-CKD (adjusted odds ratio = 1.16, 95% confidence interval [CI]: 1.08-1.23, p < .001). A new non-invasive score (termed PERIOD) including PRO-C3 efficiently predicted MAFLD-CKD (AUROC = .842, 95% CI: .805-.875). Accuracy, specificity and negative predictive values were 80.2%, 85.1% and 88.4%, respectively. In the validation cohort, the PERIOD score had good diagnostic performance (AUROC = .807, 95% CI: .691-.893) with similar results in all patient subgroups. In the MAFLD-CKD subgroup, the accuracy for identifying advanced fibrosis was further improved by combining the PRO-C3-based ADAPT with the Agile 3+ scores (AUROC = .90, 95% CI: .836-.964). CONCLUSIONS: The PERIOD score is helpful for accurately predicting the risk of MAFLD-CKD. PRO-C3 can also be used to assess liver fibrosis in people with MAFLD-CKD.

Vibration-Controlled Transient Elastography Scores to Predict Liver-Related Events in Steatotic Liver Disease.

Importance: Metabolic dysfunction-associated steatotic liver disease (MASLD) is currently the most common chronic liver disease worldwide. It is important to develop noninvasive tests to assess the disease severity and prognosis. Objective: To study the prognostic implications of baseline levels and dynamic changes of the vibration-controlled transient elastography (VCTE)-based scores developed for the diagnosis of advanced fibrosis (Agile 3+) and cirrhosis (Agile 4) in patients with MASLD. Design, Setting, and Participants: This cohort study included data from a natural history cohort of patients with MASLD who underwent VCTE examination at 16 tertiary referral centers in the US, Europe, and Asia from February 2004 to January 2023, of which the data were collected prospectively at 14 centers. Eligible patients were adults aged at least 18 years with hepatic steatosis diagnosed by histologic methods (steatosis in ≥5% of hepatocytes) or imaging studies (ultrasonography, computed tomography or magnetic resonance imaging, or controlled attenuation parameter ≥248 dB/m by VCTE). Main Outcomes and Measures: The primary outcome was liver-related events (LREs), defined as hepatocellular carcinoma or hepatic decompensation (ascites, variceal hemorrhage, hepatic encephalopathy, or hepatorenal syndrome), liver transplant, and liver-related deaths. The Agile scores were compared with histologic and 8 other noninvasive tests. Results: A total of 16 603 patients underwent VCTE examination at baseline (mean [SD] age, 52.5 [13.7] years; 9600 [57.8%] were male). At a median follow-up of 51.7 (IQR, 25.2-85.2) months, 316 patients (1.9%) developed LREs. Both Agile 3+ and Agile 4 scores classified fewer patients between the low and high cutoffs than most fibrosis scores and achieved the highest discriminatory power in predicting LREs (integrated area under the time-dependent receiver-operating characteristic curve, 0.89). A total of 10 920 patients (65.8%) had repeated VCTE examination at a median interval of 15 (IQR, 11.3-27.7) months and were included in the serial analysis. A total of 81.9% of patients (7208 of 8810) had stable Agile 3+ scores and 92.6% of patients (8163 of 8810) had stable Agile 4 scores (same risk categories at both assessments). The incidence of LREs was 0.6 per 1000 person-years in patients with persistently low Agile 3+ scores and 30.1 per 1000 person-years in patients with persistently high Agile 3+ scores. In patients with high Agile 3+ score at baseline, a decrease in the score by more than 20% was associated with substantial reduction in the risk of LREs. A similar trend was observed for the Agile 4 score, although it missed more LREs in the low-risk group. Conclusions and Relevance: Findings of this study suggest that single or serial Agile scores are highly accurate in predicting LREs in patients with MASLD, making them suitable alternatives to liver biopsy in routine clinical practice and in phase 2b and 3 clinical trials for steatohepatitis.

Chronic hepatitis B baseline viral load and on-treatment liver cancer risk: A multinational cohort study of HBeAg-positive patients.

BACKGROUND AND AIMS: A single-nation study reported that pretreatment HBV viral load is associated with on-treatment risk of HCC in patients who are HBeAg-positive without cirrhosis and with chronic hepatitis B initiating antiviral treatment. We aimed to validate the association between baseline HBV viral load and on-treatment HCC risk in a larger, multinational cohort. APPROACH AND RESULTS: Using a multinational cohort from Korea, Hong Kong, and Taiwan involving 7545 adult patients with HBeAg-positive, without cirrhosis and with chronic hepatitis B who started entecavir or tenofovir treatment with baseline HBV viral load ≥5.00 log 10 IU/mL, HCC risk was estimated by baseline viral load. HBV viral load was analyzed as a categorical variable. During continuous antiviral treatment (median, 4.28 y), HCC developed in 200 patients (incidence rate, 0.61 per 100 person-years). Baseline HBV DNA level was independently associated with on-treatment HCC risk in a nonlinear pattern. HCC risk was lowest with the highest baseline viral load (≥8.00 log 10 IU/mL; incidence rate, 0.10 per 100 person-years), but increased sharply as baseline viral load decreased. The adjusted HCC risk was 8.05 times higher (95% CI, 3.34-19.35) with baseline viral load ≥6.00 and <7.00 log 10 IU/mL (incidence rate, 1.38 per 100 person-years) compared with high (≥8.00 log 10 IU/mL) baseline viral load ( p <0.001). CONCLUSIONS: In a multinational cohort of adult patients with HBeAg-positive without cirrhosis and with chronic hepatitis B, baseline HBV viral load was significantly associated with HCC risk despite antiviral treatment. Patients with the highest viral load who initiated treatment had the lowest long-term risk of HCC development.

CAMP-B score predicts the risk of hepatocellular carcinoma in patients with chronic hepatitis B after HBsAg seroclearance.

BACKGROUND: Risk of hepatocellular carcinoma (HCC) persists after hepatitis B surface antigen (HBsAg) seroclearance in patients with chronic hepatitis B (CHB). AIMS: To identify risk factors and construct a predictive model for HCC development. METHODS: We retrospectively analysed patients with CHB with HBsAg seroclearance. Primary outcome was HCC development. Factors identified from a multivariate Cox model in the training cohort, consisting of 3476 patients from two Korean hospitals, were used to construct the prediction model. External validation was performed using data from 5255 patients in Hong Kong. RESULTS: In the training cohort, HCC occurred in 102 patients during 24,019 person-years of observation (0.43%/year). Risk scores were assigned to cirrhosis (C:3), age ≥50 years (A:2), male sex (M:3) and platelet count <150,000/mm3 (P:1); all were independently associated with an increased risk of HCC in multivariate analysis The time-dependent area under receiver operating characteristic curves for 5, 10 and 15 years in the training and validation cohorts were 0.782, 0.817 and 0.825 and 0.785, 0.771 and 0.796, respectively. In the validation cohort, 85 patients developed HCC (0.24%/year). The corresponding incidence of HCC in the low-, intermediate- and high-risk groups were 0.07%, 0.37% and 0.90%, respectively. CONCLUSIONS: The CAMP-B score (cirrhosis, age ≥50 years, male sex and platelet count <150,000/mm3/L) was significantly associated with HCC development after HBsAg seroclearance. CAMP-B score can be easily implemented in real-world clinical practice and helps stratify HCC risk in patients with CHB following HBsAg seroclearance.

Reply to Allaire, M.; Thabut, D. Comment on "Wu et al. Baveno VII Criteria Is an Accurate Risk Stratification Tool to Predict High-Risk Varices Requiring Intervention and Hepatic Events in Patients with Advanced Hepatocellular Carcinoma. Cancers 2023, 15, 2480".

We thank Allaire et al [...].

A Liver Stiffness-Based Etiology-Independent Machine Learning Algorithm to Predict Hepatocellular Carcinoma.

BACKGROUND & AIMS: The existing hepatocellular carcinoma (HCC) risk scores have modest accuracy, and most are specific to chronic hepatitis B infection. In this study, we developed and validated a liver stiffness-based machine learning algorithm (ML) for prediction and risk stratification of HCC in various chronic liver diseases (CLDs). METHODS: MLs were trained for prediction of HCC in 5155 adult patients with various CLDs in Korea and further tested in 2 prospective cohorts from Hong Kong (HK) (N = 2732) and Europe (N = 2384). Model performance was assessed according to Harrell's C-index and time-dependent receiver operating characteristic (ROC) curve. RESULTS: We developed the SMART-HCC score, a liver stiffness-based ML HCC risk score, with liver stiffness measurement ranked as the most important among 9 clinical features. The Harrell's C-index of the SMART-HCC score in HK and Europe validation cohorts were 0.89 (95% confidence interval, 0.85-0.92) and 0.91 (95% confidence interval, 0.87-0.95), respectively. The area under ROC curves of the SMART-HCC score for HCC in 5 years was ≥0.89 in both validation cohorts. The performance of SMART-HCC score was significantly better than existing HCC risk scores including aMAP score, Toronto HCC risk index, and 7 hepatitis B-related risk scores. Using dual cutoffs of 0.043 and 0.080, the annual HCC incidence was 0.09%-0.11% for low-risk group and 2.54%-4.64% for high-risk group in the HK and Europe validation cohorts. CONCLUSIONS: The SMART-HCC score is a useful machine learning-based tool for clinicians to stratify HCC risk in patients with CLDs.

Machine-learning model comprising five clinical indices and liver stiffness measurement can accurately identify MASLD-related liver fibrosis.

BACKGROUND & AIMS: aMAP score, as a hepatocellular carcinoma risk score, is proven to be associated with the degree of chronic hepatitis B-related liver fibrosis. We aimed to evaluate the ability of aMAP score for metabolic dysfunction-associated steatotic liver disease (MASLD; formerly NAFLD)-related fibrosis diagnosis and establish a machine-learning (ML) model to improve the diagnostic performance. METHODS: A total of 946 biopsy-proved MASLD patients from China and the United States were included in the analysis. The aMAP score, demographic/clinical indices and liver stiffness measurement (LSM) were included in seven ML algorithms to build fibrosis diagnostic models in the training set (N = 703). The performance of ML models was evaluated in the external validation set (N = 125). RESULTS: The AUROCs of aMAP versus fibrosis-4 index (FIB-4) and aspartate aminotransferase-platelet ratio (APRI) in cirrhosis and advanced fibrosis were (0.850 vs. 0.857 [P = 0.734], 0.735 [P = 0.001]) and (0.759 vs. 0.795 [P = 0.027], 0.709 [P = 0.049]). When using dual cut-off values, aMAP had a smaller uncertainty area and higher accuracy (26.9%, 86.6%) than FIB-4 (37.3%, 85.0%) and APRI (59.0%, 77.3%) in cirrhosis diagnosis. The seven ML models performed satisfactorily in most cases. In the validation set, the ML model comprising LSM and 5 indices (including age, sex, platelets, albumin and total bilirubin used in aMAP calculator), built by logistic regression algorithm (called LSM-plus model), exhibited excellent performance. In cirrhosis and advanced fibrosis detection, the LSM-plus model had higher accuracy (96.8%, 91.2%) than LSM alone (86.4%, 67.2%) and Agile score (76.0%, 83.2%), respectively. Additionally, the LSM-plus model also displayed high specificity (cirrhosis: 98.3%; advanced fibrosis: 92.6%) with satisfactory AUROC (0.932, 0.875, respectively) and sensitivity (88.9%, 82.4%, respectively). CONCLUSIONS: The aMAP score is capable of diagnosing MASLD-related fibrosis. The LSM-plus model could accurately identify MASLD-related cirrhosis and advanced fibrosis.

Long-term use of tenofovir disoproxil fumarate increases fracture risk in elderly patients with chronic hepatitis B.

BACKGROUND & AIMS: The use of tenofovir disoproxil fumarate (TDF) is associated with a reduction in bone mineral density and an increase in bone metabolism biomarkers. However, data on clinical bone fractures remain limited. We evaluated the impact of TDF compared to entecavir on the risk of fracture in elderly patients with chronic hepatitis B (CHB). METHODS: Patients with CHB aged ≥60 years receiving entecavir or TDF between January 2008 and December 2022 were identified using a territory-wide database in Hong Kong. The risk of incident fracture in entecavir- and TDF-treated patients before and after month 24 were compared after propensity score matching. RESULTS: A total of 41,531 patients with CHB (mean age 69.8±7.8 years, 61.6% male) receiving entecavir (n = 39,897 [96.1%]) and TDF (n = 1,634 [3.9%]) were analysed. At a median follow-up of 25.3 (9.1-58.5) months, 1,733 (4.2%) patients developed incident fracture. Patients with incident fracture were more likely to have diabetes, hypertension, congestive heart failure, rheumatoid arthritis, osteoporosis, and a history of fracture. Compared with propensity score-matched entecavir-treated patients, the risk of incident fracture in TDF-treated patients was comparable in the first 24 months (weighted subdistribution hazard ratio [sHR] 0.99, 95% CI 0.56-1.73, p = 0.960) but increased after month 24 (weighted sHR 1.80, 95% CI 1.11-2.93, p = 0.019). The 24-, 60-, and 96-month cumulative incidences (95% CI) of fracture in TDF-treated and entecavir-treated patients were 2.3% (1.6%-3.4%) vs. 2.6% (1.9%-3.5%), 6.4% (5.0%-8.2%) vs. 4.7% (3.8%-6.0%), and 10.2% (8.3%-12.6%) vs. 6.8% (5.4%-8.5%), respectively. CONCLUSIONS: The risk of fracture increased with TDF treatment for ≥24 months in elderly patients with CHB. Selection of nucleos(t)ide analogues should be individualised based on age and comorbidities. IMPACT AND IMPLICATIONS: Previous literature suggested that the use of tenofovir disoproxil fumarate (TDF) is associated with a decrease in bone mineral density. However, data on the impact of TDF on long-term incident clinical fracture remains scarce. In this real-world territory-wide study of 41,531 treated patients with chronic hepatitis B in Hong Kong, patients who received TDF were at a higher risk of fracture after 2 years of treatment than those who received entecavir. Given the ageing population of patients with chronic hepatitis B and the rising prevalence of comorbidities, our findings support the current treatment guidelines that recommend selecting antiviral treatment based on age and comorbidities.