biostatistics; evidence-based medicine; prediction modelling; meta-analysis; personalized medicine
Pillinger Toby, McCutcheon Robert A, Vano Luke, Mizuno Yuya, Arumuham Atheeshaan, Hindley Guy, Beck Katherine, Natesan Sridhar, Efthimiou Orestis, Cipriani Andrea, Howes Oliver D (2020), Comparative effects of 18 antipsychotics on metabolic function in patients with schizophrenia, predictors of metabolic dysregulation, and association with psychopathology: a systematic review and network meta-analysis, in The Lancet Psychiatry
, 7(1), 64-77.
Efthimiou Orestis, White Ian R. (2019), The dark side of the force: Multiplicity issues in network meta‐analysis and how to address them, in Research Synthesis Methods
, 11(1), 105-122.
Piccolo Raffaele, Bonaa Kaare H, Efthimiou Orestis, Varenne Olivier, Baldo Andrea, Urban Philip, Kaiser Christoph, Remkes Wouter, Räber Lorenz, de Belder Adam, van 't Hof Arnoud W J, Stankovic Goran, Lemos Pedro A, Wilsgaard Tom, Reifart Jörg, Rodriguez Alfredo E, Ribeiro Expedito E, Serruys Patrick W J C, Abizaid Alex, Sabaté Manel, Byrne Robert A, de la Torre Hernandez Jose M, Wijns William, Jüni Peter, et al. (2019), Drug-eluting or bare-metal stents for percutaneous coronary intervention: a systematic review and individual patient data meta-analysis of randomised clinical trials, in The Lancet
, 393(10190), 2503-2510.
Background: Systematic reviews and meta-analyses of randomized controlled trials (RCTs) are key to comparative effectiveness research. Meta-analyses typically use study-level data and aim to identify which is the best treatment on average. However, a treatment that is best on average may perform less well (or may be harmful) for certain patients. Meta-analysis of individual patient data (IPD) from RCTs can address this issue, by utilizing information on patient-level characteristics. However, RCTs usually employ strict experimental settings and inclusion criteria thus hampering the generalizability of RCT findings in less controlled, ‘real-world’ clinical settings. Recently developed methods attempted to incorporate real-world (observational) studies in evidence synthesis, aiming to bridge this gap. However, these attempts only focused on estimating average treatment effects. To this date, there is a gap in methods for making personalized, patient-specific predictions about the real-world effects of medical interventions.Objectives: The overarching objective of this project is to develop new, reliable methods for jointly synthesising randomized and observational evidence in order to provide personalized predictions for the performance of medical interventions. The project will be implemented with a clear focus on predicting treatment effects in real-life clinical settings. The specific aims of the project are the following: (1) To develop methods for selecting patient characteristics to include in meta-analytical prediction models. (2) To develop methods for predicting the effects of medical interventions in real-world settings in terms of effectiveness and safety, so as to facilitate a personalized choice of treatments.(3) To develop methods for monitoring patient outcomes and adapting treatment decisions.(4) To develop easy-to-use online tools that will utilize all developed methods in order to facilitate and enhance the decision-making process in every-day clinical practice.(5) To provide proof of concept by evaluating the proposed framework in real clinical examples from depression.Methods: For the objectives of this project we will explore and further develop (i) machine-learning methods for identifying patient characteristics with strong predictive properties, at the meta-analysis level; (ii) IPD network meta-analysis methods that will combine randomized and real-world evidence regarding a range of alternative, competing interventions, in order to build prediction models for guiding real-world treatment decisions; (iii) methods for building meta-analytical prediction models for disease progression; (iv) tools for making all results available to patients and doctors through easy-to-use web-applications. All methods that will be developed for the purposes of this project will be showcased using real clinical examples. Impact: This project aims to help every-day clinical practice move away from the one-size-fits-all approach of treating patients, by using evidence-based methods to tailor treatment strategies to the individual characteristics, needs and preferences of patients. Overall, the project represents a novel approach to integrating different pieces of evidence to answer patient-specific, real-world-oriented questions. The scientific output of this project will be of high interest to research scientists, biostatisticians and methodologists, while the applications to real examples are expected to have a strong clinical impact.