The WHO European Region strategic plan aims not only to attain high levels of childhood immunisations but also to provide maximum protection at an earliest possible age and to reduce geographical disparities in vaccination coverage [1]. Despite the fact that Germany has acknowledged these goals, this country has one of the lowest vaccination coverage rates in Europe [2], the administration of all recommended vaccines in children is considerably delayed and a high geographical variance between its 16 states exists (e.g. range for 1^st dose measles immunisation at 2 years: 58 – 88%) [3].

Although the need to establish a system for monitoring completeness and timeliness of immunisations is acknowledged, in many countries there are no registries for such data [4, 5]. Therefore estimation of vaccination coverage in most countries is based on either ad hoc surveys of representative samples or on assessment during school health examinations or both [6–8]. Despite the undoubted usefulness of these data in the absence of better sources, they have several limitations: surveys are fraught with problems regarding completeness, sample size, representativity and immense costs, and school health examinations lag 5 to 6 years behind the current vaccination practice and do not assess types of vaccines or timing of vaccinations.

In countries like Germany, France and the US, childhood immunisations are mainly given by the private sector (paediatricians and general practitioners) [4, 9]. Although these physicians get paid for vaccinating, imbursement data related to vaccination have not been used to assess vaccination coverage in a systematic manner.

In a cohort of children born between 2001–10–01 and 2002–12–31 in Bavaria, the second most densely populated German state, we therefore assessed whether such imbursement data might be useful to depict the vaccine uptake up to the age of 24 months.

The original KVB dataset yields data on 196,732 children born between 2001–10–01 and 2002–12–31. This is more than the respective number of children born in Bavaria in the observation period (n = 142,809) [10]. The inflated denominator can be explained by (1) children migrating to Bavaria or consulting a Bavarian physician in their holidays and (2) children changing health insurance companies; these children will be recorded more than once because of changing insurance numbers. The timing of these censoring events is not recorded in the KVB datasets and therefore cannot be considered in the analysis. In order to deal with this problem, we reduced the original KVB dataset. Since we were interested in vaccine uptake in the first two years of life we required one visit in the first half of life and one in the second year of life – in order to make sure that this identified cohort was followed-up until the second year of life. This algorithm yielded a dataset with 111,799 children, corresponding to 78% of all children born in Bavaria during the observation period.

The number of children in the validation AOK dataset was 40,645. 95.1% of the children in the validation dataset had visited an outpatient physician in the first six months of life and in the second year of life, which was the algorithm to generate the reduced KVB dataset from the original KVB dataset. The proportion of children not consulting a physician in both periods was 3.1%, the proportion of children never visiting a physician was 1.8%.

The reduced KVB dataset on AOK insured children (n = 44,755), however, was even greater than the number of children in the provided AOK dataset; this is explained by children in the reduced KVB dataset with at least one physician's visit in the second year of life who may have left the AOK before 2004–12–31 whereas the validation dataset required all children to be continuously insured from birth to 2004–12–31.

Figure 1 shows uptake and timing of the first and second MMR immunisation by age using inverse Kaplan-Meier curves is shown in relation to the age recommended by the German Standing Committee on Vaccination [11].

These data from Bavaria show that using routine data generated by SHI to handle physicians' billings may provide a useful tool to estimate immunisation uptake in the first two years of life in Germany.

The main challenge in using routine data generated by SHI to handle physicians' billings is to define an appropriate denominator. Considering all different insurance numbers related to any physicians' billing as a surrogate for the number of children eligible for vaccination (original KVB dataset) will yield to a huge overestimation of the denominator and therefore to an underestimation of vaccination coverage estimates.

A restriction to children with at least two visits – one in the first six months and one in the second year of life – reduces the original dataset to 78% of the size of the Bavarian birth cohort. The impact of this algorithm could be assessed with the AOK validation dataset: only 3.1% of all children did not visit an outpatient physician in these two periods. The high proportion of children seen by a physician in their first two years of life might be a consequence of free of charge well baby check-up visits in the first two years (U1-U7) with a compliance of 80% – 90% for each check-up [13].

The vaccination coverage estimates for the reduced KVB dataset on AOK insured children gave similar vaccination coverage estimates as for the AOK validation dataset. These were, however, up to 1.7% higher than for the AOK validation dataset. This systematic overestimation of vaccination coverage based on KVB data might be explained by children never consulting an outpatient physician – some 1.8% of the children in the AOK validation dataset. These children will not be vaccinated and – as they are not included in the denominator of the reduced KVB dataset – vaccination coverage is slightly overestimated in the reduced KVB dataset. It might be argued to confine vaccination coverage estimates to the app. 30% AOK insured children in Bavaria. Unfortunately such data are not routinely available – they were generated on special request for this study.

The vaccination coverage estimates presented here are not based on the entire Bavarian birth cohort. In table 1 causes for the missing 22% were plausibly explained. There are no data, however, to suggest that vaccination coverage for children in families moving out of Bavaria and those changing SHI should be different from the entire cohort. Vaccination coverage of privately insured children might differ since these children are from families who are exempted from the SHI system because their parents' income exceeds a certain threshold. The direction of the possible bias for vaccination coverage in this subgroup, however, is difficult to estimate: while parents with predominantly higher education levels might be more careful to have there children vaccinated in time, this clientele also includes parents attracted by 'alternative' medicine who are reluctant about vaccinations.

The vaccination coverage estimates generated from billing data for SHI insured children allow depicting the vaccine uptake as a time dependent variable. As shown exemplarily in the figure, the MMR vaccination has not only a low coverage with 24 months of age, but also exists a considerably delay in its administration: at the end of the recommended age (15 and 24 months for the first and second dose, respectively) less than 50% of the children were vaccinated with the respective dose. A similar picture appears for all other generally recommended vaccines (data not shown). Comparably low and delayed vaccination coverage estimates in young children have been shown by a national immunisation survey [3], but when restricting these data to Bavarian children only extremely wide confidence intervals derive.

The outlined use of SHI billing data to estimate vaccination coverage appears to be a promising approach for low cost in countries where vaccinations are administered by the private sector. The size of a systematic overestimation of vaccination coverage due to children never seeing an outpatient physician and therefore not appearing in datasets based on billing data was found to be small. This data source might provide a useful tool for rapid assessment of progressively increasing vaccination uptake by newly introduced vaccines.