gms | German Medical Science

Background: Society benefits from increasing life expectancy due to medical care. However, the usage of pharma-ceuticals also carries several risks. A widely discussed hazard is the increased risk of falls owing to medication, especially in elderly persons. This implies the requirement to explore risks related to pharmacotherapy. The aim of this study is to investigate the influence of fall risk increasing drugs (FRIDs) on the number of injuries in elderly persons by using a count data model as an innovative statistical method and to compare these results with those from a logistic regression which is the default model in this field of research.

Materials and methods: Routine data from the Techniker Krankenkasse of 2009 regarding elderly persons aged ≥65 years was analysed in order to assess the effect of FRIDs on injuries in a multivariate setting. We applied two types of regression models: (i) a logistic regression, where the dependent variable took the value "1" if at least one injury occurred in 2009, and "0" otherwise; and (ii) a count data model, which took into account the specific form of the dependent variable (non-negative integer values). Here the de-pendent variable was the number of injuries in the considered year. In both models we controlled for variables describing the individual medication and for variables describing disease patterns, socio-economic characteristics as well as care dependency of the insurants.

Results: Our results are ambiguous. We found positive significant effects of antidepressants, anxiolytic agents, hypnotics and sedatives, antiarrhythmics and drugs from the Priscus-List on the number of injuries in the same year. In contrast, for antihypertensives and anti-parkinsonian agents no significant effects could be detected. For neuroleptics, even a negative significant effect was obtained.

Conclusions: When prescribing specific drugs to potentially frail elderly people, one should act with caution. These drugs include antidepressants, anxiolytic agents, hypnotics and sedatives, antiarrhythmics as well as drugs from the Priscus-List. Comparing the count data model estimation with the logistic regression implementation, that is commonly used in this research field, the results clearly show an advantage of more precise outcomes using the former method.