gms | German Medical Science

Introduction: First results of a substudy of FeverApp Register [1] The FeverApp allows documentation of fever episodes of children by their parents as an EMA tool [1], but is also used as a guideline-based information tool for fever. A large pediatric practice implemented a two-arm intervention study in a Zelen design with patient information about the FeverApp and analyzed the effects of the systematic recruitment of parents and their children into the intervention group on the frequency of visits to the pediatric practice in case of feverish infections [2]. Participants of the IG and their documented fever phases were compared with the FeverApp Register [3].

Methods: From September 2019 to March 2022, all families consecutively visiting the practice were allocated to intervention- vs control group based on random numbers (ratio 7:3). Regardless of group, at all visits to the pediatric practice during the observation period parents were interrogated for acute or anamnestic fever phases. Children aged ≥3 months and max. 6 years old and with an observation time ≥1 year were included in the analysis population (IG: n=1659, CG: n=1224). As during the analyses we became aware of effects of selection biases and confounding on the primary endpoint consultation frequency due to fever, we re-oriented the scope of the analyses to a thorough methodological analysis of the diverse effects based on Feinstein‘s classic concept of the patient filtration process from recruitment into the analyzed study population [4]. Consequently, the group assigned to the intervention, but not recruited due to several reasons, has to be considered as second control group (n=1275) with the option of further investigation of the selection biases and potential confounding. A multivariable logistic regression was applied to predict the probability of being recruited into the study and predictors of using the app were analyzed.

Results: We observed a time-lag for IG (due to recruitment phase in Zelen design). Stratified analyses showed effects of age, season and course of the pandemic on the primary endpoint. Group differences of age, number of children per family and frequency of incapacity certificates for parents hint on possible selection biases for probability of being recruited as well as on the usage of the app. Due to these effects, the visit frequency seems to increase by intervention (in contradiction to expected effect). Controlling for several effects and comparing with 2^nd control group yields a lower consultation frequency. Currently, several types of statistical modelling approaches are examined to find the most adequate in estimating true effects while taking time-lag effect and selection biases into account. A first model confirms the effects found in the descriptive analyses. For the conference, we will provide results of more elaborate statistical models.

Discussion: Our analyses show usefullness and challenges of Zelen design, relevant selection effects in health-care field studies including the usage of health apps when the methodological principles of patient filtration by Feinstein are applied plus unintended effects of the pandemic.

Conclusion: Analyses in ambulant healthcare should include thorough descriptive analyses to separate direct and possible confounding effects.

The authors declare that they have no competing interests.

The authors declare that a positive ethics committee vote has been obtained.