An overview of Cochrane reviews of predictive, prognostic or prognostic modelling studies
Background: Prognostic or prognostic modelling studies are developed to give accurate and discriminating predictions about the likelihood of a future outcome in individuals with a given disease or health condition; and are increasingly being used in evidence-based medicine of the XXI century. Predictors or prognostic factors typically include laboratory variables, biomarkers, patient/disease/clinical characteristics, comorbidities, treatment-related factors, or concomitant medications.
Objectives: This overview aims to a.) synthesise evidence from Cochrane systematic reviews (CSRs) of predictive, prognostic or prognostic modelling studies; b.) map conditions/outcomes evaluated in CSRs against non-CSRs.
Methods: Cochrane Central (2023, Issue 2) and Medline (from inception until 03.02.2023) were searched (without language restrictions) for all relevant studies. CSRs and non-CSRs of predictive, prognostic or prognostic modelling studies in any healthcare area were eligible. Other types of reviews i.e., effectiveness, diagnostic, or safety were excluded.
Results: Fourteen CSRs of predictive, prognostic or prognostic modelling studies were included. The conditions involved unprovoked seizure (n=1), cognitive impairment or dementia (n=2), autism spectrum disorder (n=1), major depressive disorder (n=1), noncardiac surgery (n=1), cancer (n=6), non-specific low back pain (n=1) and venous leg ulcers (n=1). Those condtions were mapped against non-CSRs; and included: seizures (n=19), cognitive impairment or dementia (n=24), autism spectrum disorder (n=7), major depressive disorder (n=80), noncardiac surgery (n=8), cancer (n=1,190), non-specific low back pain (n=33), venous leg ulcers (n=1). The ratio of non-CSRs versus CSRs of prognostic modelling studies in oncology is 99.5% versus 0.5%; and 98.75% versus 1.25% in psychiatry.
Conclusions: There is a paucity of CSRs of predictive, prognostic or prognostic modelling studies; and a plethora of conditions or outcomes evaluated in non-CSRs. The biggest and most worrying gap exists in oncology. Given that non-Cochrane reviews report larger and often less precise effect sizes compared with CSRs – this could lead to over or under-interpretations of predictive or prognostic factors under investigation further contributing to unnecessary suffering, poor life quality and excessive fear or unrealistic optimism in certain populations such as oncology. There should be more CSRs evaluating predictive and prognostic factors.