TY - JOUR
T1 - Major mistakes or errors in the use of trial sequential analysis in systematic reviews or meta-analyses - the METSA systematic review
AU - Riberholt, Christian Gunge
AU - Olsen, Markus Harboe
AU - Milan, Joachim Birch
AU - Hafliðadóttir, Sigurlaug Hanna
AU - Svanholm, Jeppe Houmann
AU - Pedersen, Elisabeth Buck
AU - Lew, Charles Chin Han
AU - Asante, Mark Aninakwah
AU - Pereira Ribeiro, Johanne
AU - Wagner, Vibeke
AU - Kumburegama, Buddheera W M B
AU - Lee, Zheng-Yii
AU - Schaug, Julie Perrine
AU - Madsen, Christina
AU - Gluud, Christian
N1 - © 2024. The Author(s).
PY - 2024/9/9
Y1 - 2024/9/9
N2 - BACKGROUND: Systematic reviews and data synthesis of randomised clinical trials play a crucial role in clinical practice, research, and health policy. Trial sequential analysis can be used in systematic reviews to control type I and type II errors, but methodological errors including lack of protocols and transparency are cause for concern. We assessed the reporting of trial sequential analysis.METHODS: We searched Medline and the Cochrane Database of Systematic Reviews from 1 January 2018 to 31 December 2021 for systematic reviews and meta-analysis reports that include a trial sequential analysis. Only studies with at least two randomised clinical trials analysed in a forest plot and a trial sequential analysis were included. Two independent investigators assessed the studies. We evaluated protocolisation, reporting, and interpretation of the analyses, including their effect on any GRADE evaluation of imprecision.RESULTS: We included 270 systematic reviews and 274 meta-analysis reports and extracted data from 624 trial sequential analyses. Only 134/270 (50%) systematic reviews planned the trial sequential analysis in the protocol. For analyses on dichotomous outcomes, the proportion of events in the control group was missing in 181/439 (41%), relative risk reduction in 105/439 (24%), alpha in 30/439 (7%), beta in 128/439 (29%), and heterogeneity in 232/439 (53%). For analyses on continuous outcomes, the minimally relevant difference was missing in 125/185 (68%), variance (or standard deviation) in 144/185 (78%), alpha in 23/185 (12%), beta in 63/185 (34%), and heterogeneity in 105/185 (57%). Graphical illustration of the trial sequential analysis was present in 93% of the analyses, however, the Z-curve was wrongly displayed in 135/624 (22%) and 227/624 (36%) did not include futility boundaries. The overall transparency of all 624 analyses was very poor in 236 (38%) and poor in 173 (28%).CONCLUSIONS: The majority of trial sequential analyses are not transparent when preparing or presenting the required parameters, partly due to missing or poorly conducted protocols. This hampers interpretation, reproducibility, and validity.STUDY REGISTRATION: PROSPERO CRD42021273811.
AB - BACKGROUND: Systematic reviews and data synthesis of randomised clinical trials play a crucial role in clinical practice, research, and health policy. Trial sequential analysis can be used in systematic reviews to control type I and type II errors, but methodological errors including lack of protocols and transparency are cause for concern. We assessed the reporting of trial sequential analysis.METHODS: We searched Medline and the Cochrane Database of Systematic Reviews from 1 January 2018 to 31 December 2021 for systematic reviews and meta-analysis reports that include a trial sequential analysis. Only studies with at least two randomised clinical trials analysed in a forest plot and a trial sequential analysis were included. Two independent investigators assessed the studies. We evaluated protocolisation, reporting, and interpretation of the analyses, including their effect on any GRADE evaluation of imprecision.RESULTS: We included 270 systematic reviews and 274 meta-analysis reports and extracted data from 624 trial sequential analyses. Only 134/270 (50%) systematic reviews planned the trial sequential analysis in the protocol. For analyses on dichotomous outcomes, the proportion of events in the control group was missing in 181/439 (41%), relative risk reduction in 105/439 (24%), alpha in 30/439 (7%), beta in 128/439 (29%), and heterogeneity in 232/439 (53%). For analyses on continuous outcomes, the minimally relevant difference was missing in 125/185 (68%), variance (or standard deviation) in 144/185 (78%), alpha in 23/185 (12%), beta in 63/185 (34%), and heterogeneity in 105/185 (57%). Graphical illustration of the trial sequential analysis was present in 93% of the analyses, however, the Z-curve was wrongly displayed in 135/624 (22%) and 227/624 (36%) did not include futility boundaries. The overall transparency of all 624 analyses was very poor in 236 (38%) and poor in 173 (28%).CONCLUSIONS: The majority of trial sequential analyses are not transparent when preparing or presenting the required parameters, partly due to missing or poorly conducted protocols. This hampers interpretation, reproducibility, and validity.STUDY REGISTRATION: PROSPERO CRD42021273811.
KW - Humans
KW - Systematic Reviews as Topic/methods
KW - Meta-Analysis as Topic
KW - Randomized Controlled Trials as Topic/methods
KW - Research Design/standards
UR - http://www.scopus.com/inward/record.url?scp=85203393011&partnerID=8YFLogxK
U2 - 10.1186/s12874-024-02318-y
DO - 10.1186/s12874-024-02318-y
M3 - Journal article
C2 - 39251912
SN - 1471-2288
VL - 24
SP - 196
JO - BMC Medical Research Methodology
JF - BMC Medical Research Methodology
IS - 1
M1 - 196
ER -