The highest similarity was found between the GPF and SPD. Both formularies frequently provided “general infection” and “severe infection” indications, while the BNF usually provided indications for specific pathogens or infection types. Nevertheless, the calculated doses in all three formularies were found to be largely equivalent, with 52%, 67% and 53% of equivalent doses for GPF versus BNF, GPD versus SPD and SPD versus BNF, respectively. This is also supported when looking at the distribution of relative dose differences where the median was <10% for all pairwise comparisons. Furthermore, the difference of 75% of all calculated doses was ≤17% for GPF versus SPD and ≤33% for the comparisons of both SPD versus BNF and GPF versus BNF. Similar results were observed when looking at the dose equivalence and differences stratified by age category, as we did not observe a systematic trend. The high similarity between GPF and SPD is not surprising when looking at the referenced literature. In the overwhelming majority of all SPD monographs, the Dutch version of the Kinderformularium (Dutch Pediatric Formulary) is cited among the references. Interestingly, even though the referencing of the BNF in SPD was similar to the referencing of the Dutch Pediatric Formulary in SPD, the resulting dose equivalence was lower for the BNF.

The calculation of the maximal differences provides an upper bound for the “worst-case” differences. Interestingly, when taking a closer look at the doses with the highest maximal differences, it was seen that many of them were of limited clinical relevance (e.g. for half of these cases, the recommended maximal daily dose was the same). Differences were mainly attributed to either the broad therapeutic windows or as a result of the indication matching.

It is challenging to interpret the clinical relevance of relative differences in doses for drugs with varying therapeutic windows. This would require a weighting of the differences based on the width of the therapeutic window. However, there are no standardised scales available. Thus, the categorisation is usually binary, i.e. as wide or narrow. Aminoglycosides are classical examples of anti-infective drugs with a narrow therapeutic window. Other antibiotics, especially beta-lactam antibiotics, have a rather wide therapeutic window. A well-known example is benzylpenicillin where the recommended doses across formularies range from 100,000 I.U. (60 mg) per kg per day up to 24 million I.U. (14.4 g) per day. Thus, a 30% dose difference between the lower and upper benzylpenicillin doses can be generally viewed as of lesser clinical relevance than a 30% difference between aminoglycoside doses.

To date, only one other study compared Paediatric Drug Formularies, but focussed on neonatology only [12]. Among the formularies included were the BNF and the Dutch Pediatric Formulary. The authors compared the formularies on their content and structure as well as on their development and maintenance. They further compared a selection of ten relevant drugs in neonatology and concluded that there is an overall high similarity between the formularies. We came to a similar conclusion while looking at paediatric anti-infective drugs. As the general content of the formularies has already been extensively described before, we could follow a different approach by quantifying the differences of individual doses for simulated children. Following this approach, we could consider a larger number of drugs and compare the calculated daily doses children of different ages would be exposed to when being treated according to the recommendation.

As a general point it should be mentioned that this was a cross-sectional analysis. It captured the dosing recommendations at a fixed point in time. The BNF book version is valid from September 2023 to 2024, whereas the GPF and SPD databases are being updated more frequently.

The main challenge when comparing dosing information across formularies is the definition of the indications. Overall, the BNF provided the highest number of indications by far (BNF = 268, GPF = 127, SPD = 120). Indications in the BNF were frequently identified for specific pathogens and less frequently as a “general bacterial infection” indication. However, we noted that the dosing recommendations across the different indications in the BNF were frequently the same. We chose to only include indications that we deemed comparable. The matching was reviewed in a six-eye principle. This approach reduced the number of comparisons that were made but increased the validity of the comparisons.

The main strength of this study was that we calculated daily doses for a specific child entity for the comparison. With this approach, daily doses can be compared; however, differences in the frequency of administration will be masked. As it is not possible to compare a relative dose (e.g. per kilogram per dose) to a fixed dose (milligram per dose), we simulated children using median bodyweights and heights of the Swiss growth charts. However, it should be noted that comparisons of dynamic to fixed dosages will be driven by the simulated bodyweights and heights. It does not include the variability in child characteristics that exists in reality. However, most bodyweight- and BSA-based dosing recommendations are no longer valid in patients with an altered body composition (e.g. in under-/overweight, or ascites). The relationship between bodyweight and composition is not linear, thus the pharmacokinetics of a drug may be altered in those patients. Therefore, we assume that use of the median bodyweights and heights from the growth charts will still reflect what was intended during the development of the dosage. By calculating doses for all ages between 1 month and 18 years, we were able to conduct a comparison over the full age range and explore differences that may arise from this.

Another advantage of our method was that dose recommendations could be compared for the full age span from 1 month to 18 years. For some drugs, there is one weight-based dose that can be applied to children from 1 month to 18 years, but frequently there are different age and bodyweight thresholds associated with the dose. Thus, in these cases, the dosage cannot be described as a linear function of the bodyweight anymore. With our approach of calculating doses for each month of age, we describe the full distribution of dose differences from infancy to adolescence. Additionally, we could define child age categories to explore differences across age groups. Our method applied very strict rules for each comparison, and when considering the reporting variability across the formularies, the consistency of a dosing recommendation is certainly not overestimated but rather underestimated.

Differences or inconsistencies in the description of the dosages limit the validity of a comparison. For example, the vancomycin (oral) recommendation for Clostridioides difficile in the GPF and BNF contained a general dosing recommendation but also a dosage for severe infections that was not separated as a different indication and thus not collected separately. In contrast, the SPD only provides one general indication with one general dosing recommendation. However, over the full dose range, the recommendations are the same. However, based on our indication matching approach, this resulted in an apparent difference between SPD compared with both GPF and BNF. To avoid this, a different approach would be required (e.g. comparing the largest possible dosing range that could be given), which in turn will result in comparisons across indications that may be clinically meaningless. Nevertheless, it indicates that by design our results may underestimate the true equivalence between the formularies. particularly when considering the limitations of the indication matching and the calculation of simulated doses based on population values, which both have the potential to introduce systematic differences.

The maintenance of a drug information database requires continuous investment. When considering the effort and resources that are invested in maintaining drug formularies, it is not cost effective that each country is independently investing resources to perform the same task as its neighbours. Currently, there is no scientific evidence justifying the need for different dosing recommendations across European countries. An exception to this statement may be the regional differences in antimicrobial resistance patterns, which may necessitate higher doses to ensure therapeutic efficacy [30, 31]. However, because of the limited information on this topic in the monographs, it is assumed that the topic is relevant only in specific scenarios, particularly considering that we generally do not even have appropriately strong evidence for the general efficacy and safety for many drugs in the different paediatric populations. Furthermore, the availability of a common standard does not prevent adaptations for country-specific needs (e.g. because of the availability of drugs). This has already been shown by the Kinderformularium collaboration, with their formulary currently being used across the Netherlands, Germany, Austria and Norway.

To sum up, there is no scientific support that one European country may generally recommend different dosages than another country. Consequently, we think that instead of each consortium in each country repeating the same process considering the same limited number of available studies, it would be beneficial if resources could be bundled. Maintaining a database is resource intensive not only financially but also requires trained and qualified personnel with the required clinical expertise. More specialised expert committees can be built when there is a larger pool of available experts. We believe that the approach of the Kinderformularium serves as a role model for a Europe-wide collaboration.

Last, it should not be underestimated that the digitalisation of healthcare systems is heavily dependent on the use of technical and sematic standards. Standards facilitate the exchange between systems and by using a shared terminology it can be ensured that the data are interpreted as intended. A common standard enables the use of a software across countries sharing the same standard. Therefore, the availability of a common data structure can serve as the starting point for digitalisation.