The reporting of reproducibility of cardiac imaging biomarkers

Abstract

Abstract Introduction There is inconsistency in the reporting of reproducibility of imaging biomarkers in published literature. Purpose To assess the reporting of reproducibility of imaging biomarkers described in original articles across the top 3 multi-modality cardiovascular imaging journals. Methods We reviewed all published articles in 2018 in the 3 leading cardiovascular imaging journals: JACC cardiovascular Imaging (A), Circulation. Cardiovascular Imaging (B), and EHJ Cardiovascular Imaging (C). Articles were excluded if they were not original research or reproducibility reporting was not required: case reports, editorials, research supports, meta- analysis, reviews, miscellaneous. The remaining articles were assessed for the following reproducibility reporting parameters: 1) Discuss reproducibility; 2) Assess reproducibility in current study; 3) Reproducibility tools used: Intra-class correlation coefficient (ICC), coefficient of variance (COV), Bland Altman analysis (BA), etc and 5) Proportion of total population used in reproducibility analysis Results Among 951 articles published (Journal A/B/C 521/ 171/ 259), 692 were excluded due to above reasons. 259 articles were assessed for reproducibility data (Figure 1a). 122 of the 259 (47%) articles discussed reproducibility, including 26/122 (21%) referring to previous reproducibility data. Remaining 96/122 (79%) articles demonstrated reproducibility in the current manuscript- there was no difference in the frequency of articles discussing reproducibility across three journals (A 35/85 (41%); B 25/52 (48%); C 62/122 (51%), p=0.39). Modality differences between reporting of reproducibility were as follows: (Echo 66/122 (54%); CT 14/38 (37%); CMR 27/55 (49%); Nuclear 9/26 (35%) and others 6/18 (33%), p=0.04) Measures used to demonstrate reproducibility were as follows: (1.) ICC 64/96 (67%); 2.) BA 28/96 (29%); 3.) COV 16/96 (17%); 4.) Agreement 18/96 (19%); 5.) Correlation (Pearson's r) 3/96 (3%) p<0.001) (Figure 1b). The proportion of the total population used for reproducibility analysis ranged from 0.8% to 100%, median 18.6% IQR (7.8%- 56.3%). Regression revealed proportion of total population used, was inversely related to population size (r=−0.32, p 0.002). Modality differences were as follows: (Echo 12.0% (5.7%- 36.1%); CT 17.1% (7.8%- 100%); CMR 25.0% (16.1%- 97.1%); Nuclear 68.9% (29.6% - 100%) and others 56.1% (32.6%- 62.3%), p=0.03). Conclusions Approximately half original research manuscripts published across 3 major multi-modality imaging journals discussed or demonstrated reproducibility. No differences across journals were identified. There are differences between modalities with respect to reporting reproducibility. The most common tool used to report reproducibility was ICC. There is marked heterogeneity in the proportion of the total population used to demonstrate reproducibility, this is partly determined by the size of population being examined. Figure 1. A Snapshot of reproducibility reporting Funding Acknowledgement Type of funding source: None

Ähnliche Arbeiten