OBJECTIVES: The number of first-line treatments for unresectable hepatocellular carcinoma (uHCC) under investigation in randomized clinical trials (RCTs) has risen dramatically in recent years, complicating regular updates of evidence sythesis efforts. The aim was to implement a multilevel network meta-regression (ML-NMR) using a function-oriented pipeline tool in R to efficiently incorporate new trial data while accounting for differences in trial populations.

METHODS: A targeted literature search was conducted in PubMed up to April 2024 to identify RCTs of systemic treatments and selective internal radiation therapy for uHCC. Study meta-data, patient characteristics, and counts of adverse events (AEs), including diarrhea, fatigue, hypertension, and rash (all ≥grade 3), were extracted. All data were recorded in a tabular format and used as input to a ‘targets’ pipeline. ‘targets’ is an open-source pipeline tool for R, linking inputs to intermediate and final outputs. ‘targets’ pipelines enable computationally-efficient execution and re-execution of entire analyses from a single function call as only pipeline elements changed since the previous run (and their dependencies) are re-run.

RESULTS: Fourteen RCTs and >13,500 data points were incorporated into the analysis. Data were processed with custom functions specified as entities in the ‘targets’ pipeline, ensuring that only changes in the data or processing functions would trigger re-execution. Another set of functions was included to create and analyze the four networks, using the ‘multinma’ package. Initial pipeline execution took <20 minutes (16 GB of RAM, 12 cores, R v4.3.3), and included aggregate and individual-level data informing fixed-effect and random-effects models, without covariates, using four Markov chains with 2,000 iterations each per model. Subsequent modifications to a single model typically required <60 seconds to re-run the entire pipeline.

CONCLUSIONS: Implementing an ML-NMR in a ‘targets’ pipeline proved to be efficient and practical and could represent an option for “living syntheses” in dynamic research fields such as uHCC.