Systematic Review With Meta-analysis

The Effects of Non-steroidal Anti-inflammatory Drugs and Anti-platelet Therapy on the Incidence and Recurrence of Hepatocellular Carcinoma

Regina Zi Hwei Tan; Ian Lockart; Christina Abdel Shaheed; Mark Danta


Aliment Pharmacol Ther. 2021;54(4):356-367. 

In This Article


This meta-analysis was performed according to PRISMA and registered with International Prospective Register of Systematic Review [PROSPERO:CRD42020183495 (5/7/2020)].[17]

A systematic literature search was carried out using Pubmed, Scopus, Medline, Embase and Cochrane Library (from inception to 31 May 2021) for published studies fulfilling the inclusion criteria. Presentations at the major viral hepatitis conferences were also searched, including International Liver Congress™, The Liver Meeting®. was searched for registered on-going clinical trials. The following search terms were used, without date restrictions: 'hepatocellular carcinoma', 'liver cancer', 'liver neoplasm', 'non-steroidal anti-inflammatory drug', 'cyclooxygenase inhibitor', 'platelet aggregation inhibitors', 'aspirin', 'clopidogrel', 'ticagrelor', 'celecoxib', 'diclofenac', 'indomethacin', 'naproxen', 'rofecoxib', 'lumiracoxib', 'valdecoxib', 'etoricoxib' and 'mefenamic acid'. Electronic searches were supplemented with extensive manual searches of the reference lists of all relevant retrieved review articles and primary studies to identify overlooked studies. Search strategies were developed based on the PICO (Population, Intervention, Comparison and Outcome) framework, whereas no search terms were included for comparison and outcome to increase the sensitivity of the search strategies.

We included observational studies (case-control and cohort), and randomised controlled trials in the English language evaluating incident and recurrent HCC. To be eligible for inclusion, studies had to include adult participants with a confirmed diagnosis of HBV, HCV, ALD or NAFLD administered with either aspirin, non-aspirin NSAIDs or APT (eg, clopidogrel or dual therapy, including aspirin) for a defined period of time and were followed for at least 6 months. Studies must have reported on at least one of the outcome measures of interest: HCC incidence, HCC recurrence or liver-related mortality data. Studies with sample size <10 were excluded.

Two researchers independently screened abstracts of studies for eligibility and then independently read the full texts of included studies. Discrepancies were resolved by consensus and by discussion with the study team. The extracted data included study characteristics, participant characteristics and outcome variables (Table 1, Table 2 and Table 3). Authors were contacted if the key data were not reported.

Comparative studies evaluating NSAIDs, or APT, vs control group were used to determine the overall HCC incidence, HCC recurrence after treatment, liver-related mortality and bleeding complications. HCC diagnoses were confirmed by specialists using established pathological or radiographic criteria. HCC recurrence was defined as confirmation of HCC relapse by radiographic criteria. Liver mortality was either determined by Cancer and Cause of Death registries and defined by their primary International Classification of Diseases (ICD) codes or documented using Common Terminology Criteria for Adverse events (CTCAE). Liver mortality outcomes assessed include: HCC progression, liver failure, hepatorenal syndrome, spontaneous bacterial peritonitis and variceal haemorrhage. Adverse bleeding complications were defined as gastrointestinal bleeding events including peptic ulcer bleeding and variceal bleeding.

The treatment cohorts were considered matched if pre-specified criteria were applied to select patients for matching. Patients matched by propensity score were considered matched. Single-arm studies that did not meet the matching criteria were considered 'unmatched'. Adjusted hazard ratio (aHR) was derived from matched comparative studies or from multivariable-adjusted regression modelling. Common covariates that were adjusted for included: age, gender, cirrhosis, comorbidities (eg, diabetes, coronary artery disease, hypertension, hyperlipidaemia, alcohol misuse) and concurrent drug usage (eg, insulin, metformin, statins, interferon, nucleoside analogue).

All statistical analysis was performed using the Comprehensive Meta-Analysis (CMA) software v3, Englewood, USA. The effect measures of interest were the proportions, HR, OR and the corresponding 95% CI. The effect of drug therapy on HCC incidence and recurrence was evaluated by the pooled aHR at 95% CI using random effects modelling with external and internal heterogeneity assessed by I2 statistics and Cochran Q test respectively.[18] The effect of drug therapy on liver-related mortality and bleeding outcomes was evaluated using proportion data obtained from the index date to the end date of the study.

We performed subgroup analysis for aspirin and non-aspirin drugs separately for the HCC incidence. Data from the included studies were divided into matched or unmatched studies (if ≥2 studies), with the pooled adjusted HR presented for each group separately. In addition, we performed subgroup analysis for studies with data stratified by cirrhosis where possible. In analyses with few trials (<5 studies), we did not perform matched comparisons. Instead, we pooled data based on drug subtype. To avoid unit of analysis error by duplicating the number in the control arm, we did not combine multiple comparisons from the same study.[19]

The Grading of Recommendations, Assessment, Development and Evaluation (GRADE) criteria were used to evaluate the overall quality of evidence.[20] The quality of evidence began at high quality for RCTs and at low quality for observational studies and was downgraded by one level for each of the following factors: study limitations (risk of bias); inconsistency of results (I2 >50%); imprecision (wide variance of point estimates across studies); and publication bias (assessed using funnel plot analysis or Egger's regression test for ≥10 comparisons).[21–25] Overall quality of evidence was defined as high, moderate, low or very low quality[21] (Table 4). The sample sizes and effect estimates representing the magnitude of the treatment effect were included in the analysis. Risk of bias of individual studies was assessed using the Newcastle-Ottawa Scale (NOS) for observational studiesl[26] or the seven-item Cochrane Risk of Bias tool for RCTs.[27] Risk of bias assessment was conducted by two authors independently, with discrepancies discussed with a third author to reach consensus.