Abstract

Objectives: The use of artificial intelligence, including machine learning, has increased in recent years, resulting in subsequent synthesis of results via systematic reviews (SR) or meta-analyses (MA). However, poor reporting quality of methodological information for SRMA studies has been observed; in response, the Transparent Reporting of a Multivariate Prediction Model for Individual Prognosis or Diagnosis (TRIPOD)-SRMA reporting guidelines were developed in 2023. The objective of this review was to assess adherence to the 26-itemTRIPOD-SRMA checklist (itemized broadly under abstract, introduction, methods, results, and discussion).

 

Design: This umbrella review included SRMA publications which applied machine learning methods,  focused on publications pertaining to the most prevalent cancer types (breast, lung, prostate, and colorectal cancer) to maintain a defined scope for the umbrella review. A PUBMED search was conducted on 18 June 2024 for articles published in the 10 years prior.

Results: 12 SR publications and 4 MA publications were identified and abstracted by two independent researchers. For most TRIPOD-SRMA checklist items, between 30 and 75% of publications achieved the maximum possible score. Highest adherence was observed for: rationale, methods: synthesis, methods: heterogeneity, study selection, study and model characteristics, results: synthesis, implications and competing interests. Lowest adherence was observed for abstract and objectives.

 

Conclusion: The SRMA publications in scope of this umbrella review demonstrated varied adherence to the TRIPOD-SRMA checklist. The quality and transparency of machine learning SRMA analysis and publications may be improved with greater adherence to the TRIPOD-SRMA checklist, particularly the components with the lowest average scores. 

Statistics