Portal Vein Thrombosis in Cirrhosis: Incidence, Outcomes and Anticoagulation

Background and rationale Portal vein thrombosis (PVT) complicates ≈10-25 % of cirrhosis cases and is linked to variceal bleeding, ascites, hepatic encephalopathy and inferior transplant-free survival. Existing data are fragmented, largely single-centre, and often exclude patients managed without anticoagulation. Consequently, clinicians still lack robust, generalisable tools for (a) predicting incident PVT, (b) quantifying its true prognostic impact, and (c) balancing the benefits and risks of real-world anticoagulant therapy.

Study design This is a seven-centre, bidirectional cohort comprising a retrospective arm (chart review of 1 682 cirrhotic adults hospitalised) and a prospective arm that will enrol ≈ 840 additional patients, yielding a total target size of ≈ 2 500 subjects. Eligible participants are aged ≥18 years, have imaging-confirmed cirrhosis (any aetiology), and can be stratified into three baseline states: (1) no PVT, (2) focal or partial PVT, or (3) occlusive/complete PVT. Key exclusions are hepatocellular carcinoma invading the portal vein, Budd-Chiari syndrome, current pregnancy, or life expectancy < 3 months.

Data collection and follow-up At index admission (or first study visit) the team records demographics, liver disease aetiology, Child-Pugh/MELD scores, thrombophilia panel, abdominal Doppler/CT, and treatment decisions (including anticoagulant class, dose and duration). Follow-up contacts are scheduled at 6 ± 1 months, 12 ± 2 months and 24 ± 3 months via clinic visit, electronic health record extraction or structured telephone interview . End-points captured at each contact include incident PVT, progression/regression of an existing thrombosis, major bleeding (ISTH criteria), liver-related decompensation, transplantation and all-cause mortality.

Statistical plan Predictors of incident PVT will be screened with univariable Fine-Gray competing-risk models (death/transplant as competing events); independent factors will enter a multivariable model to derive a PVT-Risk Score. Model performance will be quantified by Harrell's C-index, time-dependent AUROC and calibration plots, with bootstrap internal validation and external temporal validation using the prospective sub-cohort. Propensity-score overlap weighting will compare anticoagulation versus no-anticoagulation for key outcomes, while multi-state Markov models will characterise transitions between "no PVT", "partial PVT", "complete PVT" and "death/transplant". Pre-specified subgroup analyses include Child-Pugh class, non-selective β-blocker use and thrombophilia status.