gms | German Medical Science

Background/research question: Despite the hailed benefits of Digital Health Technologies such as Electronic Health Records (EHR) for health systems, it is the patient who ultimately decides to use a technology and thus determines its success. To explain and predict intentions to use EHRs, structural equation models (SEMs) such as the Privacy Calculus Model [1] use a large number of parameters and, based on them, offer conceptual insights but limited advice for practitioners. An alternative are fast-and-frugal trees (FFTs), linear classification models, which assess few parameters sequentially to classify intentions to (not) use at each step. This way, FFTs are transparent and easily implementable by practitioners. We test the effectiveness of FFTs compared to a full SEM for two use cases: (1) to predict intention to use an EHR and (2) to identify early adopters who might benefit from different marketing approaches than late adopters.

Methods: We used survey data from N=1001 German citizens on EHR use and the FFTrees software package [2] to design two FTTs based on a subsample of the data/training set (n=701). FFT1 used two parameters from the Privacy Calculus Model to predict intention to use an EHR. FFT2 used four participant characteristics to identify user groups with high intention. Both FFTs were tested against a full SEM with 27 parameters and on the remaining data/test set (n=300).

Results: FFT1 predicted intentions to use EHRs correctly 84% of the time in the test set (sensitivity: 78%, specificity: 96%); FFT2 performed 80% accurately (sensitivity: 95%, specificity: 52%), whereas the full SEM predicted 67% accurately (sensitivity: 99%, specificity: 50%).

Conclusion: FFT1/2 used two/four instead of 27 parameters and showed higher predictive accuracy than the full model. Whereas SEMs offer theoretical and explanatory insights, use-case specific FTTs might be more accurate predictors and more useful: FFT1 reliably identifies EHR users with only two questions about perceived benefits and trust in the EHR provider (FFT1). Based on FFT2, early adopters of EHRs tend to use wearables and health apps; thus, marketing approaches may target this specific group. In sum, FFTs are effective and efficient tools to identify user groups who might benefit from targeted interventions and may complement SEMs in the quest of increasing uptake of digital health technologies.

Competing interests: None