Systematic and comprehensive identification of confounders for three non-randomized studies of interventions (NRSI) emulating 'target' randomized controlled trials (RCT)

Session: 

Oral session: Inclusion of non-randomized designs

Date: 

Wednesday 23 October 2019 - 14:00 to 15:30

Location: 

All authors in correct order:

Pufulete M1, Mahadevan K2, Johnson TW2, Reeves BC1
1 University of Bristol, UK
2 Bristol Heart Institute, UK
Presenting author and contact person

Presenting author:

Barnaby Reeves

Contact person:

Abstract text
Background: the ROBINS-I tool is recommended for assessing risk of bias in reviews that include non-randomized studies of interventions (NRSIs). ROBINS-I requires important confounding domains to be specified but there is no established method to do this. We designed three NRSIs to emulate hypothetical 'target' randomized controlled trials (RCTs) to quantify 'real world’ bleeding in populations: 1) undergoing percutaneous coronary intervention (PCI); 2) coronary artery bypass grafting (CABG); or 3) treated conservatively (medication only) and exposed to different regimens of dual antiplatelet therapy (DAPT).

Objectives: as part of our NRSIs, we identified potential confounders and co-interventions systematically using literature review, semi-structured interviews with clinicians and short surveys with additional clinicians.

Methods: we searched the literature (Ovid MEDLINE, Ovid Embase and the Cochrane Library from inception to August 2016) to identify RCTs and cohort studies of DAPT interventions. Two authors extracted data independently. We randomly ordered relevant citations and discontinued data extraction after reviewing 10 consecutive studies without identifying an additional confounder/co-intervention. We also conducted semi-structured interviews with six cardiologists and six cardiac surgeons. We combined all factors identified from the literature review and clinician interviews in two short surveys, administered online to members of professional cardiology and cardiac surgery organizations.

Results: we screened 2544 abstracts, obtained full-text papers on 386 eligible studies and extracted data from 62. We identified 58 potential confounders from the literature and 20 from interviews with clinicians. Figure 1 shows their classification and the overlap between literature review and clinician interviews. We identified most potential confounders (58/66, 88%) from the literature only; just under a fifth (12/66, 18%) by both methods, while we identified relatively few (8/66, 12%) from clinician interviews only. The literature review also identified 10 co-interventions.

Conclusions: although the review identified most factors, clinicians identified 'hard-to-measure' factors (e.g. perceived patient adherence, local prescribing guidelines). The challenges to conducting the review included the large number of studies identified and uncertainty about the data to extract from different designs. Work is ongoing to identify 'true' confounders that predict the choice of DAPT intervention (awaiting survey results) and independently influence bleeding risk. The factors identified from the literature and interviews may vary by health field but we anticipate that the methods will apply widely.

Patient involvement: patients reviewed the findings of the clinician interviews and provided confirmation that decisions we made about patient-related factors that may influence prescribing were sensible.

Attachments: