Presentation (CTI)

OBJECTIVES: OBJECTIVES: The use of artificial intelligence (including supervised machine learning) has increased substantially in recent years. However, poor reporting quality of methodological information for these studies has been observed; in response, the Transparent Reporting of a multivariate prediction model for Individual Prognosis or Diagnosis for Systematic Reviews and Meta Analyses (TRIPOD-SRMA) statement was developed. The objective of this umbrella review is to assess TRIPOD-SRMA adherence in recent supervised machine learning systematic reviews (SR) and meta-analyses (MA) for select cancer outcomes.
METHODS: METHODS: A scoring guideline based on the 26-item TRIPOD-SRMA checklist was developed and pilot-tested prior to use. A PUBMED search was conducted on 18 June 2024, including search terms pertaining to supervised machine learning and breast, lung, prostate, or colorectal cancer, and restricted to SR and MA published in the 10 years prior. Abstracts were screened, reviewed, and extracted by two independent reviewers.
RESULTS: RESULTS: From 108 articles identified by the search terms, 12 SR and 4 MA publications were included in this umbrella review. The median scores were 33 for SR (maximum possible score: 50), and 48.5 for MA (maximum possible score: 59). The domains with the lowest scores (i.e. fewer than 75% of studies met the lowest possible score) were study objectives, data performance measures, risk of bias and applicability (both methods and results), sensitivity methods, certainty assessment (method), registration, and protocol availability.
CONCLUSIONS: CONCLUSIONS: This review identified several domains of the TRIPOD-SRMA checklist which had low levels of adherence in SR/MA studies for a selection of cancer types. Better adherence to the TRIPOD-SRMA statement across future studies that use machine learning may improve the reporting quality of individual SR and MA publications and reduce the heterogeneity across umbrella reviews.