Modelling the demand of biologics and biosimilars using discrete choice models: Evidence from Finnish and Swedish markets for long-acting insulin

Abstract

The policy makers in Finland have managed to control the growth of pharmaceutical expenditures of chemical medicines with different policy changes, but they are facing new challenges as a significant and growing share of the pharmaceutical industry consists of biologic drugs that are under different regulation. Two policies of chemical medicines, generic substitution and reference pricing, have been estimated to have accumulated savings of over a billion dollars to the healthcare system and patients. However, these policies do not concern biologic medicines. Although in 2018 over 30% of European healthcare drug costs originated from purchases of biologic medicines, quantitative evidence on the prevailing or potential effects of different policies in biologic markets is scarce. Additionally, our understanding of the substitution patterns of biologics, especially original biologic medicines, their “copycats” biosimilars and improved versions biobetters, is narrow.

I explore the potential of using discrete choice models - more specifically logit, 1-level nested logit and 2-level nested logit - to model the demand of biologics. I study the market of long-acting insulin in Finland and Sweden from 2007 to 2017. I aim to provide information on whether discrete choice models could be used to simulate counterfactuals on the market and to extend our understanding of the substitution patterns of biologics.

I find that the logit and the 1-level nested logit models, as specified in this thesis, give promising results for Sweden but produce a positive price coefficient in Finland. Additional analyses of the own- and cross-price elasticities in Sweden indicate that the logit model produces substitution patterns consistent with previous literature, while the 1-level nested logit model produces own- and cross-price elasticities that are much higher in magnitude. Further inspection of the Finnish market from April 2017 to December 2020, after multiple policy changes took place, produce reasonable results for the logit and the 1-level nested logit model.

The calculated cross-price elasticities and the results of the nested logit models indicate that different active substances are substitutes to each other and that the reference drug, the biobetter and the biosimilar are stronger substitutes to each other than to other drugs. This suggests that restricting the market definition to only the biosimilar and the reference drug, as done in some previous studies on biologics and biosimilars, seems too narrow.

All in all, the logit and the 1-level nested logit model for Finland after April 2017, as well as the logit model for Sweden produce reasonable estimates and own- and cross-price elasticities. Therefore, they could be used for calculating counterfactuals in the market. However, the results should be taken with reservations, as the models used in this thesis are restrictive in nature and may not be suitable for modelling the demand of prescription drugs.