Methodological quality and risk of bias in non-randomised comparative effectiveness studies of invasive mould...

Many treatment recommendations for invasive mould infections, particularly those caused by rare moulds, rely on non-randomised comparative effectiveness research (CER), which is vulnerable to design and analytical limitations that can bias effect estimates. We systematically reviewed 117 non-randomised studies evaluating interventions for aspergillosis, mucormycosis, fusariosis, scedosporiosis, lomentosporiosis, or phaeohyphomycosis published from 2003 to 2024. We assessed structural susceptibility to bias from misalignment of eligibility, treatment assignment, and start of follow-up, and approaches to confounding adjustment. 112 (96%) of 117 studies had at least one timepoint misalignment. Of 140 evaluable CER questions, 116 (83%) were susceptible to immortal time bias, 47 (34%) to selection bias from post-baseline eligibility, and 13 (9%) to prevalent-user bias. 76 (65%) of 117 studies adjusted for confounding but variable selection often relied on statistical methods and 26 (34%) of those 76 studies included post-baseline variables. This structural susceptibility to bias was present across all pathogen groups, not only rare moulds. Only one study avoided timepoint misalignment and accounted for confounding without incorporating post-baseline variables. These findings could provide a framework to improve the design of future non-randomised CER studies.