Chronic hepatitis B (CHB)

Hepatic steatosis (HS)

Both CHB and HS are major contributors to chronic liver injury, increasing the risk of cirrhosis and hepatocellular carcinoma

As the coexistence of CHB and HS becomes increasingly common, simple, noninvasive diagnostic methods are needed to accurately assess the severity of liver inflammation

Establishing a machine learning (ML)-based model to detect liver inflammation in patients with CHB combined with HS

We conducted a multicenter, retrospective cohort study in China

Treatment-naïve CHB patients with HS confirmed by liver biopsy between April 2004 and September 2022 were included.

Moderate to severe liver inflammation (Scheuer’s system ≥ G3)

The best features for model development were explained by SHapley Additive and the ML algorithm for determining the best accuracy in diagnosing moderate to severe liver inflammation was evaluated by DCA and calibration curves.

From 1,787 treatment-naive CHB and HS patients from 11 hospitals, 689 patients from 9 hospitals were selected to develop the diagnostic model

The remaining two hospitals participated in two independent external validation cohorts, including 509 patients in validation cohort 1 and 589 patients in validation cohort 2

Subgroup analysis was used to determine the performance of the GBC model in subgroups stratified by age, sex, body mass index, HBeAg status, HBV DNA levels, and presence of diabetes

To ensure data relevance and consistency, only laboratory tests performed within 14 days before liver biopsy were considered

Pathological examinations were performed by two independent board-certified pathologists from their respective centers without blinding to the clinical data.

Disagreements were resolved through discussion and consensus was reached. Assessing interobserver variability to assess interpathologist agreement

11 features identified regarding inflammation, liver and metabolic function

Gradient boosting classifier (GBC) model showed best performance in predicting moderate to severe liver inflammation

GBC model uses simple parameters to predict liver inflammation in CHB patients complicated by HS

It has the potential to guide clinical management and improve patient outcomes

A publicly accessible web tool was generated for the model

Baseline characteristics of the cohort

Diagnostic performance of machine learning models for moderate to severe liver inflammation

Flowchart of the study population

SHapley additional explanation diagram

Diagnostic performance of final machine learning model for moderate to severe liver inflammation in training cohort and external validation cohort

GBC Model Net Benefit in Diagnosing Moderate to Severe Liver Inflammation via Decision Curve Analysis

interobserver variability

To ensure the robustness of the diagnosis of inflammation, we selected cases from the two largest cohorts, Nanjing Drum Tower Hospital and Hangzhou Normal University Hospital, and evaluated the two pathologists who performed pathological examinations on these cases

Both pathologists are board certified and have at least 5 years of general pathology experience at the time of examination

Using kappa statistics to assess interobserver variability between two pathologists

We use the following classification to evaluate the quality of kappa statistics: 0-20% is poor consistency, 21-40% is fair consistency, 41-60% is fair consistency, 61-80% is good consistency, and 81-100% For excellent consistency

The Kappa statistic is a measure of agreement between observers and is calculated as (Po - Pe) / (1 - Pe), where Po represents the observed agreement between raters and Pe represents chance alone. expected consistency

The result kappa values ​​were 0.76 (Nanjing Drum Tower Hospital) and 0.63 (Hangzhou Normal University Affiliated Hospital) respectively.

Population Characteristics of Moderate to Severe Hepatic Inflammation and Non-Moderate to Severe Hepatic Inflammation in the Training and Validation Cohorts

Mixed baseline of positive events for 3 datasets

Diagnostic value of different subgroups of GBC models for moderate to severe liver inflammation

6 different subgroups

Diagnostic performance of pre-machine learning model for severe liver inflammation in training cohort

Evaluation of the efficacy of the GBC model in diagnosing moderate to severe liver inflammation using calibration curves