Real-world predictions of treatment effects using evidence synthesis for health technology assessment

Chalkou, Konstantina (2023). Real-world predictions of treatment effects using evidence synthesis for health technology assessment (Unpublished). (Dissertation, University of Bern, the Faculty of Medicine and the Faculty of Human Sciences)

Full text not available from this repository.

Background
Prognostication and individualized predictions play important roles in clinical decision-making. In clinical practice, prognostic models foresee an individual’s future health condition, whereas prediction models identify subgroups of patients who benefit from a treatment. Aiming to predict individualized treatment effects, risk modelling is a successful method in several clinical areas. Risk modelling consists of two stages. In the first stage, the baseline risk of the outcome is estimated via a prognostic model; and in the second stage, the baseline risk is used as the only effect modifier for the treatment effect estimation. However, risk modelling is limited in cases where treatments are directly compared in single or several randomized clinical trials (RCTs). For most clinical conditions today, since treatment options are numerous, direct comparisons in single mega-trials are impossible. Network meta-analysis (NMA)—a key tool—synthesizes evidence from several studies and compares treatment effectiveness with direct and indirect estimations.

By extending the risk modelling approach into an NMA framework, it supports clinical decision-making, allowing for individualized predictions for all available treatment options in a clinical area. In addition, a framework combining several data sources, such as real-world data with RCTs, and aggregate data (AD) with individual participant data (IPD), allows for predictions among RCT and real-world populations. The latter also uses all relevant information, leading potentially to more precise results. Many clinical areas, such as multiple sclerosis (MS), lack established individualized prediction models, which could support patients and their physicians when selecting optimal treatment. Finally, prediction models need evaluation for accuracy and—more notably—clinical usefulness. Decision curve analysis (DCA) is a method recommended for evaluating clinical usefulness of prediction models; however, existing DCA methods only evaluate models developed via single RCT. Established methods for evaluating prediction models from pairwise or NMA are not yet developed.

Aim

I fill several aforementioned methodological gaps by allowing for individualized predictions in an NMA framework with illustrative examples from the MS clinical area. I also support clinical decision-making processes in the MS clinical area, where many gaps in prognosis and prediction of heterogeneous treatment effects exist. To do so, my objectives include

1. building a prognostic model for predicting relapses among patients with relapsing-remitting multiple sclerosis (RRMS) with observational study data—a data source depicting best real-world conditions;

2. developing a methodological framework by combining prognostic modelling and NMA for individualized predictions under several treatment options;

3. extending the former developed framework by combining several data sources: real-world data with RCTs for predictions in RCT among real-world populations and AD with IPD for using all the relevant information available; and

4. developing methods for evaluating clinical utility of prediction models by extending existing DCA methods.

Methods

Since there was no established prognostic model, I use high-level statistical methodology to develop a prognostic model for MS. Then, I combined ideas from prognosis research and NMA, extended the risk modelling approach, and built a framework for individualized predictions when several treatment options were available. I further extended this methodology into a general framework where several data sources were combined. Finally, I integrated ideas from DCA and NMA to develop a model which outweighs harms and benefits of each treatment to evaluate prediction model clinical utility.

I applied all developed methods on datasets of patients diagnosed with RRMS. I developed a prognostic model identified via the literature and used eight baseline variables, which contribute to estimating the baseline risk score and—in turn—identifying how treatment effects varied across patients. Variables were age, sex, months since last relapse, prior MS treatment, number of prior relapses, expanded disability status scale, number of gadolinium enhanced lesions, and diagnosis years. I used three RCT with IPD and two RCT with AD with evidence about placebo and three active treatments: dimethyl fumarate, glatiramer acetate, and natalizumab. I used placebo arms from nine RCT with IPD and the Swiss Multiple Sclerosis Cohort— a high-quality observational study.

Results

My developed framework for individualized predictions of treatment effects showed baseline risk plays an important role in estimated treatment effects hence also for optimal treatment recommendations. Whereas natalizumab is considered—on average—the optimal treatment option to minimize the risk of relapsing within the next two years, dimethyl fumarate is the optimal choice for patients with low baseline risk among RCT and real-world populations. When I applied the developed DCA methodology to evaluate clinical utility of one of the MS prediction models, the developed prediction model performs either close or better than other default strategies.

Conclusions

As a contribution to personalized medicine, my suggested frameworks can be used to make individualized predictions for all available treatments for any clinical condition. In addition, my proposed evaluation method evaluates the clinical utility of such models and potentially provides support to the MS clinical area, which lacks decision-making tools.

Item Type:	Thesis (Dissertation)
Division/Institute:	04 Faculty of Medicine > Pre-clinic Human Medicine > Institute of Social and Preventive Medicine (ISPM)
Graduate School:	Graduate School for Health Sciences (GHS)
UniBE Contributor:	Chalkou, Konstantina, Salanti, Georgia
Subjects:	600 Technology > 610 Medicine & health 300 Social sciences, sociology & anthropology > 360 Social problems & social services
Language:	English
Submitter:	Beatrice Minder Wyssmann
Date Deposited:	04 Aug 2023 15:22
Last Modified:	04 Aug 2023 15:22
Additional Information:	PhD in Health Sciences (Epidemiology and Biostatistics)
URI:	https://boris.unibe.ch/id/eprint/185227