放射線医学研究論文における統計的サポートレベルの妥当性に関する方法論的レビュー

We assessed if there was disparity between qualified statisticians and other researchers regarding the level of statistical assistance deemed necessary to support radiological research.

We categorised 50 consecutive, eligible original research articles published in an indexed imaging journal (European Radiology) 2024, according to authors' statements regarding statistical support, declared in the "statistics and biometry" section. Two reviewers extracted data related to study design, statistical methods, and analysis. Two medical statisticians categorised each study as presenting "complex" statistical methods or not and then compared this with authors' own assessment of statistical complexity, stated in the published article. We performed descriptive analyses.

Most studies were observational (49, 98%) and retrospective (38, 76%). 35 (70%) studies were diagnostic, 7 (14%) prognostic, and 6 (12%) mixed. Malignancy was the most frequent topic (29 studies, 58%), and MRI the most frequent modality (35 studies, 70%). We deemed most studies (33, 66%) presented complex statistical methods. Of these, 13 studies (26% overall) declared that "no complex statistical methods were necessary for this paper". However, 10 of these employed hypothesis testing, frequently using multiple methods; 9 employed agreement and/or reliability analyses; all presented accuracy measures; 11 (85%) presented a regression model.

We found that approximately one quarter of original research articles published in our sample stated that "no complex statistical methods were necessary", but then presented complex analyses.

Some radiological researchers may underestimate the complexities of statistical analysis and requirement for specialist statistical support, which risks inappropriate analyses and misleading results.