This study uses a probability pathway model to estimate the frequency and costs of ADEs occurring in ambulatory medical practice in Germany over 1 year (2007). The model includes drug-related morbidity and mortality. It calculates ADEs and bases all resource utilisation on the number of hospital admissions attributed to ADEs reported by the German Schneeweiss study . Results indicate that about 2 million ADEs have occurred in persons in the population over 20 years of age taking medications, and that subsequent health care for 816 million Euros has been utilized.
The predicted probabilities of ADEs and resource utilisation due to ADEs in the population treated with medications were lower for the German base case than in the original model, but close to the lower end of rates reported in the literature, indicating a limited usefulness of the model in this respect. The calculation of total costs of resource utilisation attributes the highest proportion of costs to hospitalisation, the factor in the model for which there are the most data for Germany. Thus, in this respect, the model is useful in estimating the lower limits of resource utilisation costs which can be expected to be due to ADEs. According to the base case, hospital admission due to an ADE is required by 0.3% of the population taking medications (see Table 1). In comparison, the other German study predicts 0.4% and the original NMP and TF/NMP arms of the model together predict 1.1%. The studies of ADEs in ambulatory medical practice observed that between 0.5%  and 4.2%  of the population taking medications required hospitalisation due to an ADE (see Table 2). These studies indicate that the base case likely underestimates the total number of ADEs, whereas the model rates are likely too high for Germany. The original model rates regarding mortality due to ADEs, 2.1% of patients hospitalised for ADEs, are also probably too high for Germany, since Schneeweiss reported a rate of 1.7% and Dormann a rate of about 1%.
There are no German studies regarding other types of resource utilisation for ADEs, such as physician or emergency visits. However, the base case predicts rates in the treated population of both physician visits (3%) and emergency visits (0.4%) which are lower than those predicted using Dormann's study (5% and 0.7% respectively), the original model (12% and 2% respectively) and close to the lower end of rates reported in the literature (3.5% and 0.9% respectively: see Table 2). Basing the model on German hospitalisation rates to predict the rates of ADEs in the outpatient population leads to estimates of ADEs lower than reported in the literature. According to the base case, 4.3% of the treated population will have an ADE, whereas if Dormann's values are used, 6.9% of the treated population will have an ADE. In the literature, probabilities for ADEs range between 5.5%  and 34.7%  of the treated population (see Additional File 3).
As already implied above, this study has numerous limitations. First, both studies used for the German data are limited. The Schneeweiss study  reports admission rates due to ADRs which are considerably lower than other studies but only non-elective admissions were evaluated and patients with skin reactions were not included. Also, the 3 hospitals in the study may not have been representative for usual medical practice in Germany. A follow-up study with intensified ADR surveillance in these hospitals, reported a higher ADR incidence rate of 3.25%  while using the original exclusion criteria. The second German study was based on admissions to a university hospital, which may have a higher probability of ADE admissions. This study is also limited by a smaller number of cases, a shorter observation period, a baseline population which included readmissions and transfers from other wards and hospitals and the inclusion of ADRs due to chemotherapeutic agents. Of further concern is that the origin of patients in both German studies is not described, thus whether they were living independently or in a nursing home is unknown. Also, both German studies only report ADRs, a subgroup of ADEs. The gap between the estimates for the base case and sensitivity analyses (Table 3) also point out the gap in the available information for Germany and the possible number of unrecognised ADEs. The sensitivity analyses are also limited, since all studies of ADEs in the ambulatory population were performed in the USA and thus may not reflect the incidence or resource use in Germany. Due to its simple structure, the model does not reflect all costs possibly incurred by ADEs, such as multiple physician visits, multiple hospitalisations, increased costs of health services around the time of death [31–34], or other costly resources such as long-term dialysis for specific ADEs . However, the use of a Markov model would seem inappropriate at the present, since the literature provided neither exact treatment sequences for ADEs, nor rates of recovery from ADEs.
Our estimations for the ambulatory sector show the importance of hospital and long-term care costs, which account for 80% of ADE costs but are generated by only 8% of patients predicted to have an ADE. In contrast, costs of physician visits and additional treatment explain 9% of total ADE costs but are generated by 70% of patients predicted to have an ADE. The overall costs reported by Ernst et al.  are considerably higher than our estimates, but their calculations also include costs for treatment failure, the most frequent drug-related problem in their model. Generally, their cost structure is similar, with hospitalisations and long-term care accounting for 87% of all costs, while physician visits and additional prescriptions account for 9.8% of costs.
Overall, the costs predicted by the sensitivity analyses were much higher than the base case, indicating an underestimation of true costs. However, all studies used for the sensitivity analyses originated in the USA and reflect the US population and health care system. This attempt at transferring probabilities from a model developed for the USA to Germany, illustrates the extent of required information which is still missing such as prescribing practices, the rate of medication changes per patient, estimates of compliance and the classes of medications prescribed. The absence of universal healthcare coverage in the US may have numerous consequences on health care utilisation, such as rate of purchase of prescribed medications, how closely patients are monitored regarding medications and whether patients seek medical advice if symptoms occur. The corresponding behaviour and habits of patients and physicians in Germany must be considered to decide how well these studies and rates apply in Germany.
Considering the limitations of the model and the parameters used, the postulated costs of 816 million Euros (0.32% of German healthcare expenditures in 2007 (253 billion Euros) ), due to ADEs must be viewed with caution. They represent first attempts for Germany to postulate costs associated with ADEs emerging from outpatient treatment and vividly illustrate the lack of available data and difficulties in assessing any results. Although outpatient research of ADEs may be difficult in Germany, due to data confidentiality and the separation of in- and outpatient care, estimates based on the model show that adverse drug events may be placing a significant economic burden on the healthcare system in Germany, especially in terms of hospital and long-term care costs. As the population ages, the disease burden will increase, more medications will be consumed and the risk of severe adverse events and hospitalisation will increase . Investigations of outpatient ADEs in Germany could determine the types of ADEs occurring and useful prevention processes. At present, any calculations assessing the benefits or costs related to processes aiming at reducing adverse drug events, such as an electronic health card, would require much more data than available in the published literature in this area.