Applying and extending mixed-effects models in health economics and outcomes research

PhD ceremony: Mr. P. Pechlivanoglou, 9.00 uur, Academiegebouw, Broerstraat 5, Groningen

Dissertation: Applying and extending mixed-effects models in health economics and outcomes research

Promotor(s): prof. M.J. Postma, prof. J.E. Wieringa

Faculty: Mathematics and Natural Sciences

The thesis of Petros Pechlivanoglou examined applications and methodological improvements of mixed-effects models in health-economics and outcomes research. It focused on aspects related to the analysis of prescription-level data and to evidence synthesis. Pechlivanoglou found that mixed-effects models can improve our estimation of effectiveness and cost-effectiveness, as illustrated by a number of treatment options and health-policy decisions. Yet, as indicated in his thesis, improvements in the modeling aspects are still possible. His findings include that the newer agents for prevention of fungal infections in neutropenic patients and of stroke in patients with atrial fibrillation are more effective and potentially more cost-effective than current options. Additionally, educating physicians to think in a cost-effective way, when prescribing, seems to be more effective than financially incentivizing them to do so.

Prescription-level data from the northern Netherlands were analyzed by Pechlivanoglou to identify the probability of a prescription’s generic substitution from the pharmacist. The multilevel, crossed structure of the data required the use of a mixed-effects model, which revealed significant relations between substitution and experience with drug use, pharmacy status and timing. A time-series prescription dataset from the same region was also used to assess the impact of a financial and an educational intervention on rational prescribing. A multivariate state-space model was designed for this analysis. The model identified a significant impact of the educational intervention while the financial incentive was estimated not to have a significant effect on prescribing patterns.

Mixed-effects models were also applied for evidence synthesis to assist medical and health economic decision-making in infectious diseases and stroke prevention. A meta-regression, conducted to synthesize estimates of S. pneumoniae prevalence in community-acquired pneumonia in Europe, identified and incorporated significant regional and healthcare-specific variation across estimates from different studies and countries. Evidence on antifungal prophylaxis as well as on stroke prevention were synthesized and combined with economic decision models to examine the cost-effectiveness of new antifungals and anticoagulants, respectively. The latter study directly influenced Dutch health-care policy through the Health Council. Finally, an improvement on the methodology of network evidence synthesis was done, facilitating application of mixed-effects models in complex model structures.