Literature search
We searched for relevant articles in two electronic databases, including PubMed and Web of Science from inception to April 2022. The following search term was used: (NTRK OR NTRK1 OR NTRK3 OR NTRK1/3 OR RET OR RET/PTC OR RET/PTC1 OR RET/PTC3 OR ALK OR BRAF OR MET OR PPARG OR PPAR Gamma OR THADA OR ROS1) AND thyroid AND (rearrangement OR fusion). We also searched for potential studies by reviewing the citations within the included studies and reviews. Our study protocol followed the recommendation of the Preferred Reporting Items for Systematic Review and Meta-analysis (PRISMA) statement [16].
Selection criteria and abstract screening
After deleting the duplicates from two electronic databases, titles and abstracts of included studies were independently screened by four reviewers (HTL, TTBL, TL, and SH). Studies were included if they reported demographic, clinicopathological, and/or prognostic data of any types of gene fusions in PTCs. We excluded studies if they were (i) studies on other thyroid cancer subtypes (ii) case reports, (iii) reviews, and (iv) posters, conference papers, theses, or books. Discordant results between the reviewers were resolved by discussion and consensus.
Full-text screening and data extraction
The full text of all candidate articles was read independently by three reviewers (HTL, TTBL, and TL). Subsequently, data were extracted into a predefined extraction form. The following data were extracted from full-text papers: authors, institution, city, country, publication year, enrollment period, study design, number of patients, fusion detection method, gender, patient age, histologic features, tumor size, focality, vascular invasion, lymph node metastasis, distant metastasis at presentation, extent of resection, administration of radioactive iodine, tumor persistence/recurrence, progression-free survival (PFS) status, PFS time, overall survival (OS) status, and OS time. Any disagreements between two reviewers, if present, were resolved again by discussion and consensus.
Quality assessment and risk of bias analysis
We used the Newcastle – Ottawa Scale (NOS) tool to evaluate the quality of included studies in our meta-analyses [17]. Two reviewers independently awarded stars for cohort or case-control studies (maximum nine stars) based on a developed checklist [17]. Studies awarded at least five stars were considered moderate to high-quality studies and those with a NOS value of less than five were regarded as low-quality studies.
Meta-analysis
We conducted a meta-analysis using summary-level data to compare the differences between fusion-positive and fusion-negative PTCs. For studies with potentially duplicated populations, we selected the studies with the highest number of cases. Review Manager 5.4 (Cochrane Collaborative, Oxford, UK) was utilized for statistical analysis. Pooled estimates of odds ratio (OR), mean difference (MD), and corresponding 95% CIs were calculated using the random-effect model.
To compare different oncogenic fusions in PTCs, individual patient data (IPD) from included papers were used because of their statistical advantage compared to aggregate data. We excluded PTC cases with rare fusions including ROS1 (n = 1), THADA (n = 2), and MET (n = 2), and cases with concomitant fusions (n = 5) because they might bias the analyses. Chi-square and Fisher’s exact tests were used to compare categorical variables while analysis of variance (ANOVA), and Wilcoxon rank-sum tests were used for continuous covariates, if applicable. Kaplan-Meier and Cox regression analyses were conducted to investigate the impact of gene fusions on PFS and OS. IPD analyses were performed by using the R software version 4.1.1 (The R Foundation, Vienna, Austria). A p-value of less than 0.05 indicated a statistically significant value.
Among-study heterogeneity was assessed by the I2 statistic and classified as low (25% < I2 ≤ 50%), moderate (50% < I2 ≤ 75%), and high (I2 > 75%) [18]. We further explored the origin of between-study heterogeneity in cases of moderate and high degrees of heterogeneity using sensitivity and subgroup analyses, if applicable. Egger’s regression test and funnel plot were carried out to further assess the presence of potential publication bias. A p-value of less than 0.05 indicated statistically significant publication bias.