**Background:** Personalized nutrition approaches are hypothesized to be more effective than generic dietary guidelines because they account for individual differences in biology, behavior, genetics, and sociopsychological environment. Digital Personalized Nutrition Advice Systems (PNASs) automate the translation of individual data into personalized dietary services. This paper describes the rationale behind design choices for a PNAS developed within the Dutch Public Private Partnership Personalized Nutrition and Health, tested in two use cases: highly motivated consumers with premetabolic syndrome (n=37) and consumers with low socioeconomic status (n=96).
**Methods:** The PNAS follows a sense-reason-act framework. For sensing, the system includes both health parameters (age, sex, anthropometrics, vitals, and plasma biomarkers) and lifestyle behavior parameters (dietary intake measured via the Dutch Healthy Eating Index [DHEI] questionnaire, which scores 15 food groups from 0 to 10). The study with highly motivated consumers allowed more extensive measurements (fat percentage, blood sampling for HDL, LDL, glucose, triglycerides, and nutrient profile) in a healthcare setting. The study with low socioeconomic status consumers was set in a supermarket using do-it-yourself measurements (body weight scales, blood pressure meters, handheld devices for plasma biomarkers). For reasoning, the system uses a Bayesian Belief Network (BBN) to model probabilistic relationships between health variables (input) and advice variables (output). Input nodes include DHEI food category intake values and health parameter values; output nodes provide advice at four levels. Knowledge rules were derived from expert heuristics, Dutch Health Council guidelines, and recent meta-analyses (e.g., whole grain consumption effects on blood glucose and type 2 diabetes risk). For acting, feedback was provided via color-coded displays of individual biomarker values against healthy ranges, composite metabolic health scores (using a health space model), and star ratings for dietary categories. Advice included practical tips and implementation intentions (if-then plans). In the highly motivated consumer study, feedback and advice were delivered via a web-based portal and telephone consultation with a dietician; in the low socioeconomic status study, via an emailed report.
**Key Results:** The paper does not report quantitative outcomes from the two studies but describes design decisions and qualitative feedback. Participants who received personalized advice better perceived the link between their diet scores and showed greater improvement in dietary habits than those who did not receive personalized advice. In the highly motivated consumer study, advice obtained via telephone consultation was appreciated more highly than that obtained via email. In the low socioeconomic status study, some participants indicated that feedback was sometimes difficult to understand, highlighting the challenge of tailoring format to the target group. The amount of data collected was not sufficient to train the BBN model with observational data.
**Clinical Implications:** The sense-reason-act framework provides structured guidance for developing digital PNASs. Key recommendations include: (1) balance input parameter needs against user burden, distinguishing 'must have' from 'nice to have' parameters; (2) carefully consider whether a data-driven, knowledge-driven, or hybrid model is best suited; (3) avoid building systems around a single technology—use modular architectures; (4) tailor communication format, complexity, and behavior change techniques to the target group; (5) incorporate multiple behavior change techniques and consider stress management to support self-efficacy. The authors note that little scientific research has been conducted on how personalized nutrition advice should best be communicated to consumers, representing a serious shortcoming.