A general framework for exploring the impact of suboptimal treatment choices to health outcomes in a real-world population

Tags: Oral
Efthimiou O1, Leucht S2, Samara M2, Belger M3, Salanti G4
1Department of Hygiene and Epidemiology, University of Ioannina School of Medicine, Greece, 2Department of Psychiatry and Psychotherapy, Technische Universität München, München, Germany, 3Eli Lilly and Company, Lilly Research Centre, Windlesham, UK, 4University of Ioannina School of Medicine; Institute of Social and Preventive Medicine, University of Bern; Berner Institut für Hausarztmedizin (BIHAM), University of Bern , Switzerland

Background: Network meta-analyses are increasingly used to synthesize evidence from randomized controlled trials (RCTs) and provide useful information about relative treatment benefits and harms. However, clinicians often make treatment decisions that disregard the evidence, and potentially prescribe less efficient or safe treatments to patients.

Objectives: To utilize study-level and individual participant level data from RCTs and non-randomized studies (NRSs) in order to explore the impact of a specific policy regarding treatment choices to health outcomes in a real-world population of interest.

Methods: We categorized patient characteristics as treatment predictors, prognostic factors and effect modifiers using expert opinion. We performed a network meta-regression using the RCTs to estimate the relative treatment effects and the ranking of all available treatments for a range of values of the effect modifiers. Using the NRSs we built a model that predicts treatment choices in a real-world clinical setting. We generated a sample of patients with the characteristics of the population of interest. For each simulated patient we predicted the disease progression using the prognostic factors and the effect modifiers under two scenarios:

1. evidence-based treatment choice: the patient receives the optimal treatment as determined using the RCTs; and

2. treatment choice as usual: the patient received treatment following the policies currently employed in real-world settings.

We compared predictions across the two scenarios.

Results: We applied our methods to 167 RCTs and one large observational study that compared antipsychotics for schizophrenia. Results showed that treatment choices in a real-world setting are not, to a large extent, evidence-based. Simulation showed that outcomes predicted in a real-world setting are significantly better when treatment choices are based on the randomized evidence.

Conclusions: Our approach can be used to assess the added benefit for adopting an evidence-based approach to clinical decision-making in real-world clinical practice.