Trial sequential analysis in meta-analyses: A clinically oriented approach with real-world example

Central MessageTrial sequential analysis can be used to assess the conclusiveness of meta-analytical findings. Clinicians should learn understand and interpret such analyses, aware its limitations.See Commentaries on pages 174, 175, 176. Trial limitations. See Ranked high hierarchy evidence, meta-analyses are important summaries existing literature, often influencing clinical practice guidelines. However, as evidence accumulates, random errors may at times lead spuriously significant results, an increasing frequency statistical testing by increases likelihood being reported; that is, type 1 error.1Borm G.F. Donders A.R.T. Updating leads larger than publication bias.J Clin Epidemiol. 2009; 62: 825-830.e10Abstract Full Text PDF PubMed Scopus (68) Google Scholar Therefore, trial (TSA) has been developed adjust for this increase in error. TSA is conceptually similar interim analyses randomized controlled trials, where results tested regular intervals determine whether a certain difference due intervention, or lack thereof, conclusively demonstrated data, allowing terminated.2Brok J. Thorlund K. Gluud C. Wetterslev reveals insufficient information size potentially false positive many meta-analyses.J 2008; 61: 763-769Abstract (710) Scholar,3Brok Apparently conclusive inconclusive—trial adjustment error risk repetitive accumulating data apparently neonatal meta-analyses.Int J 38: 287-298Crossref (636) In meta-analyses, from each study analogous added during analysis. Statistical techniques threshold significance (z-score threshold; conventionally corresponds P value .05) based effect observed, incidence outcome control arm, (a measure amount analyzed), heterogeneity. Simply put, large (eg, 90% relative reduction [RRR]) observed with minimal heterogeneity requires smaller sizes observation small 10% RRR) As such, effects (ie, differences between 2 arms, 25% considered meta-analysis. Results found not require much further investigation, whereas premature spurious conclusions unadjusted also avoided. Meta-analyses TSA4Chan J.S.K. Kot T.K.M. Ng M. Harky A. Continuous infusion versus intermittent boluses furosemide acute heart failure: systematic review meta-analysis.J Card Fail. November 12, 2019; ([Epub ahead print])Abstract (3) Scholar, 5Zhang X. Wu Q. Wei Ding Y. Gu Liu S. et al.Low-residual diet clear liquid bowel preparation before colonoscopy: meta-analysis trials.Gastrointest Endosc. May 3, 2020; (9) 6Yokoyama Mihara T. Kashiwagi Koga Goto Effects intravenous dextrose preventing postoperative nausea vomiting: analysis.PLoS One. 15: e0231958Crossref 7Ge Wang W. Hou L. Yang Fa Inspiratory muscle training associated decreased pulmonary complications: trials.J Thorac Cardiovasc Surg. 2018; 156: 1290-1300Abstract (25) 8Hristovska A.-M. Duch P. Allingstrup Afshari Efficacy safety sugammadex neostigmine reversing neuromuscular blockade adults.Cochrane Database Syst Rev. 2017; 8: CD012763PubMed post hoc TSA9Chan Lau D.H.H. Does lysing make life better? A analysis.J Neurol. 267: 1842-1845Crossref (1) 10Chan Colchicine prevents stroke patients coronary artery disease—a analysis.Eur 27: e28Crossref 11Chan The in: cardiovascular benefits glucagon-like peptide-1 receptor agonists—a Prev Cardiol. April 10, print])Crossref 12Manogaran S.S. Data beta-blockade ACLS—a analysis.Resuscitation. 150: 191-192Abstract have become more common recent years, having incorporated some Cochrane reviews well.8Hristovska It therefore clinicians principles, interpretation, limitations TSA. article, we aim illustrate these using Woldendorp colleagues.13Woldendorp Doyle M.P. Black D. Keech Grieve S.M. al.Subclinical valve thrombosis transcatheter aortic implantation: February 20, (12) shows association subclinical (SCVT) implantation stroke, superiority oral anticoagulants (OAC) single (SAPT) dual (DAPT) antiplatelet therapy SCVT. To perform TSA, study-level summary all meta-analyzed comparisons were obtained published Forrest plots, odds ratios estimates accordance choice Single-arm studies excluded. Two-sided z-score thresholds adjusted O'Brien-Fleming α-spending function 2-sided 5% 80% power. DerSimonian-Laird model was used. Control arm incidences calculated event rates studies. Information estimated available (Fischer information). If impossible, sample instead. cases, estimations repeated (event sizes) ensure robustness calculations. RRR values studies.14Thorlund Engstrøm Brok Imberger G. User Manual Sequential Analysis (TSA). Copenhagen Unit, Copenhagen, Denmark2017Google demonstrate interpretation utility performed prespecified values, deemed appropriate comparison made. Studies reporting no handled adding constant both counts. All software version 0.9.5.10 Beta, unit.14Thorlund files online (Online Supplement 1-3). generates graphical outputs (Figure 1, Figure 2, 4). y-axis represents cumulative included studies, which strength x-axis size. blue line curve, plot plotted chronologically. These points represent size, combination number trials (also called Fischer information), depending specification Generally, preferred superior other options. brown, dotted lines conventional ±1.96 (these optional always present plots). red boundaries 1)—a pair outer, oblique form boundaries, inner futility boundaries. Crossing either imply existing, statistically conclusive; similarly, crossing insignificant conclusive. final vertical required (RIS)—upon reaching observations said stable because unlikely fluctuate significantly even upon addition new conclusion likely reached. narrower approaches increases, reflecting penalized heavily when smaller.Figure 2Result comparing valvular taking (DAPT). Here, curve (blue line) crossed (vertical boundary corresponding .05 (dotted yellow lines). Such indicates rate OAC could evidence. numbers indicate chronologically arranged set analysis: 1: Leetmaa T, Hansson NC, Leipsic J, Jensen K, Poulsen SH, Andersen HR, al. Early thrombosis. Circ Interv. 2015;8:e001596, 2: Makkar RR, Fontana G, Jilaihawi H, Chakravarty Kofoed KF, de Backer O, Possible leaflet bioprosthetic valves. N Engl Med. 2015;373:2015-24, 3: Grove EL, Mathiassen ON, JM, Transcatheter Am Coll 2016;68:2059-69, 4: Søndergaard L, Friedman Berman D, Subclinical surgical valves: observational study. Lancet. 2017;389:2383-92, 5: Marwan M, Mekkhala N, Göller Röther Bittner Schuhbaeck A, al Leaflet following implantation. Comput Tomogr. 2018;12:8-13, 6: Vollema EM, Kong WKF, Katsanos S, Kamperidis V, van Rosendael PJ, der Kley F, thrombosis: relation hypo-attenuated thickening, abnormal haemodynamics, stroke. Eur Heart 2017;38:1207-17, 7: Yanagisawa R, Hayashida Yamada Y, Tanaka Yashima Inohara Incidence, predictors, mid-term outcomes possible after TAVR. Cardiol Imaging 2017;10:1-11, Basra SS, Gopal Hebeler KR, Baumgarten Anderson Potluri SP, Clinical valves four-dimensional computed tomography. Ann 2018;106:1716-25, 9: Ruile P, Minners Breitbart Schoechlin Gick Pache Medium-term follow-up early replacement. 2018;11:1164-71, 10: Jimenez C, Ohana Marchandot B, Kibler Carmona Peillex Impact antithrombotic regimen platelet inhibition extent detected cardiac MDCT 2019;8:506-17.View Large Image ViewerDownload Hi-res image Download (PPT)Figure 3Result did cross any (red arms Med Blanke Dorfs Jander Arepalli CD, thickening balloon-expandable 2016;37:2263-71, Lancet Imaging. 2018;106:1716-25.View 4Result (DAPT), –33.3% reduction. sufficient conclude SAPT usage SCVT 33.3% compared DAPT. 2016;68:2059-206, et. (PPT) normal motions (NLM) (stroke latter, 2.75%). Sample estimate too little use (−141.9%). RIS, indicating effect, occurring frequently SCVT, With −188%, −100%. This revealed boundary, reach there least 100% those NLM. So, given current it sizes. Similar made instead, suggesting calculation robust. DAPT (SAPT incidence, 18.71%). Again, 68.6%, 20%. Similarly, 20% OAC, 3 (DAPT 16.06%). (−63.12%). boundary. conclusive, −16.4%, −20%, confirm rule out −33.3%, shown 4, show Although attractive, one must bear mind Most importantly, virtually applicable well, makes meta-analyses. Any bias, including affect significantly. ideally bias. Nonetheless, most inevitably include risks ameliorate issue, only low On note, affects validity results—in above example, interpreted caution (I2 = 68%). More detailed analysis, subgroup meta-regression, delineate underlying cause refine conclusion. An alternative would without heterogeneity, need addressed definitive anyway. useful conclusions, criticized conservative.15Wetterslev Jakobsen J.C. meta-analysis.BMC Res Methodol. 17: 39Crossref (472) relates selected analysis—choosing conservative easily showing chosen level RRR. Indeed, mostly arbitrary cases relies expertise experience author. example almost meaningless repeat –20%, TSA: does used, resultant unhelpful strong, risk.16Roshanov P.S. Dennis B.B. Pasic N. Garg A.X. Walsh When conclusive? guide remote ischemic preconditioning renoprotection undergoing surgery.Nephrol Dial Transplant. 32: ii23-ii30Crossref (19) seem counterintuitive, but simply result previously mentioned principle directly influences relevance appropriateness interpreting colleagues13Woldendorp NLM, well use. who take SAPT, DAPT; despite more. Our key evolving tool expected play increasingly roles familiarize themselves note results. appropriately, powerful capable assessing