Several measures are in force in Switzerland to control the cost of drugs, but are not effective enough. There are many determinants influencing these expenditures, related to treatments, markets, physicians, patients and regions, but their impact on costs is not clear.
We applied a Bayesian multilevel model with five levels to adjust for patients, drugs' market, and physicians 'characteristics, treatment type, and district (i.e. Swiss canton). We used data of the Swiss drugs' market in 2006, offering real choices for doctors and patients (multiple products for similar active substances), with a neutral position of pharmacists (no financial incentives).
Variance partitioning of yearly drugs' cost per insured showed that market level (delivered substance) contributed to 76% of the variance, treatment level (delivered product) to 20%, whereas patients' and physicians' levels accounted for only 2% each, without significant differences between Swiss cantons. After adjusting for covariables at each level, the model explained about 51% of the variation at the market and 20% at the treatment levels. We found that older but substitutable drugs, generics, larger size of the market and physician's specialty were associated with lower expenditure, whereas drugs requiring a physician's prescription, the number of prescribers per patient, patient' age, male gender, and comorbidities increased expenditure. Our results show that for a specific medication the yearly cost of recently released drugs was 36 CHF higher than for similar and substitutable drugs introduced 15 years earlier, corresponding to one third of the average annual treatment cost observed in our dataset. Competition did not seem to be effective to reduce expenditure on the drug market.
The main finding of this study is that recentness of drugs was associated with an increase in drug expenditure in 2006, even after adjustment for all non-controllable determinants. Further research is recommended to confirm those results with updated data.