Chronic noncommunicable diseases (NCDs) are a major global health problem. They are the leading cause of mortality and morbidity, accounting for 74% of all deaths worldwide [1]. Their management and treatment are a major economic concern, accounting for 25% of the total health care budget in Europe [2]. The major NCDs are cardiovascular diseases (CVDs), respiratory diseases, cancer, and type 2 diabetes (T2D).

CVD is the leading cause of death worldwide and in Europe, accounting for 32.7% of all deaths in 2022 [3]. Ischemic heart disease and cerebrovascular disease are the most common causes, accounting for 32.4% and 20.8% of CVD deaths, respectively [3]. There is variability between high-income and middle-income countries in the proportion of premature CVD deaths, with the proportion being higher in middle-income countries in both sexes (36% in women and men in middle-income countries, vs 16% and 24% in women and men in high-income countries, respectively) [4]. In 2019, these diseases were responsible for an estimated 70 million disability-adjusted life years in European Society of Cardiology member countries [4]. There is also an unequal distribution of CVD deaths in Europe, with more than half of all deaths occurring in eastern countries, compared with 23% in northern and western countries [1,4,5]. Metabolic risk factors include hypertension, hyperglycemia, hyperlipidemia, and obesity. Lifestyle risk factors include tobacco and alcohol use, physical inactivity, and an unhealthy diet.

Respiratory diseases are the third leading cause of death in Europe, accounting for 7% of all causes of death in 2022 [3]. Chronic obstructive pulmonary disease (COPD) and asthma are the most common respiratory diseases. COPD reduces quality of life and increases morbidity, hospital admissions, and premature death. Most COPD is preventable, and the main risk factors are smoking and air pollution [3,6].

Despite a 11.9% decrease in the standardized mortality rates between 2012 and 2022, cancer is still the second most frequent cause of death in the European Union (EU), accounting for 22.3% of all deaths in 2022. Lung, colorectal, and breast cancer account for 19.8%, 11.5%, and 7.4% of all deaths from cancer in the EU [3]. Globally, the most commonly diagnosed group of cancers is skin cancer, and cutaneous melanoma, the deadliest, accounts for 20% of skin cancers [7]. As with CVD, many cancers can be prevented by adopting a healthy lifestyle. Smoking, alcohol, physical inactivity, and an unhealthy diet, the latter 2 leading to overweight and obesity, are responsible for the majority of cancer incidence [4]. In addition, UV radiation exposure is the leading environmental cause of melanoma, a type of cancer that continues to increase in fair-skinned populations worldwide [7].

T2D has a prevalence of 5%-20% in the adult population. It is associated with other NCDs and includes risk factors such as smoking, physical inactivity, unhealthy diet, and obesity. Prevention can reduce the risk of diabetes by up to 78% and reduce cardiovascular mortality, cancer mortality, and the incidence of CVD in people with T2D [8,9].

For type 1 diabetes (T1D), there are no established recommendations for prevention. However, controlling risk factors such as smoking, alcohol consumption, sedentary lifestyle, and unhealthy diet improves disease control and delays the onset of complications such as CVD [10-13]. Therefore, while modifying these lifestyle factors does not prevent the onset of T1D, it does reduce the risk of NCDs in this specific population. It improves glycemic control and prevents other microvascular complications that are more prevalent in people with T1D, such as diabetic retinopathy [14-16]. Treatment of T1D requires a high degree of self-care, which can represent a burden and affect health-related quality of life [17]. The incorporation of technology such as continuous glucose monitoring (CGM) and automated insulin delivery systems facilitates diabetes management by providing relevant information and improved glycemic control [18,19].

The risk of developing some NCDs can be assessed based on known risk factors. This is important as we can act on those risk factors that are modifiable, provide personalized recommendations for each individual, and prevent or postpone the onset of NCDs, reducing morbidity and mortality and economic expenditure. Furthermore, the 4 major NCDs share common risk factors: physical inactivity, an unhealthy diet, and alcohol and tobacco consumption. Addressing these 4 risk factors could reduce the risk of NCDs.

An increasing number of people use mobile apps to improve their health. These apps offer direct access to information on specific health topics, as well as useful features, such as medication reminders, physical activity, and vital sign recording [20]. Artificial intelligence (AI) is also becoming an increasingly popular tool. Some studies have examined the potential application of AI in mobile health apps, particularly in relation to mental and oral health [21].

Watching the Risk Factors (WARIFA) [22] is a project funded by the European Commission and coordinated by the Norwegian Centre for E-Health Research. It involves 12 partners from 6 European countries. It aims to provide early risk assessment to enable individualized risk prevention. A mobile health (mHealth) app (the WARIFA app) has been developed, which uses user-generated data analysis and AI to provide NCD risk estimation, as well as personalized messages, based on scientific evidence, to promote healthy behaviors.

Personalized recommendations in health apps are highly relevant as they allow feedback to be tailored to each user’s actual needs in terms of content, intensity, and timing. This can lead to greater engagement, self-efficacy, and adherence to recommended behaviors. Unlike static or nonpersonalized recommendations, just-in-time, adaptive interventions provide the right recommendations at the right time, taking the context of each user into account. This type of intervention is commonly used in health apps [23-26]. Thus, personalized interventions provide more than just general information; they also seek to maximize the timeliness and relevance of each recommendation by finding the most appropriate moment to encourage adherence. The WARIFA app takes this approach by integrating automatically received data (from activity devices) and user-entered data to continuously adapt its recommendations. For example, it adjusts physical activity suggestions according to daily steps and generates dietary recommendations according to daily consumption records.

Co-creation methodology was used during the development of the WARIFA app. This involved the users throughout the process, enabling their needs to be identified and feedback to be obtained. Eighteen focus groups consisting of 3-10 participants of different ages from the general population and people with T1D were organized for this purpose. The co-creation process was divided into 2 phases.

In the first phase, focus groups were held with participants who expressed their needs and expectations for a health app and created hand-drawn sketches and prototypes. In the second phase, focus groups were held with a prototype of the app based on users’ preferences from the previous phase. Quantitative and qualitative information obtained from each group was used to update the app, creating one that aligns with the real needs of the population.

After the co-creation process, the randomized controlled trial (RCT) whose protocol is described in this article was conducted to evaluate the effectiveness of the WARIFA app in promoting healthy lifestyles and in the self-management of T1D.

The aim of this RCT is to evaluate the effectiveness of the WARIFA app in promoting healthy lifestyles and in the self-management of T1D.

The hypothesis is that use of the WARIFA app will improve healthy eating habits, increase physical activity, reduce alcohol consumption, promote smoking cessation, and improve sun protection behaviors. In addition, it is expected to reduce hypoglycemic events and lower glycosylated hemoglobin A_1c (HbA_1c) in people with T1D.

This study protocol was developed in accordance with the recommendations for interventional trials 2013 statement SPIRIT (Standard Protocol Items: Recommendations for Intervention Trials) [27] and SPIRIT-AI (SPIRIT–Artificial Intelligence) guidance published for trials of AI interventions [28].

A randomized, controlled, double blind, parallel, multicenter trial was conducted in 2 populations (general population and individuals with T1D) in 3 European countries. For this purpose, participants were provided either with the WARIFA app with personalized recommendations or the same WARIFA app without personalized recommendations. The participating centers are: the Research Institute for Biomedical and Health Sciences, University of Las Palmas de Gran Canaria, Gran Canaria, Spain, associated with Complejo Hospitalario Universitario Insular Materno-Infantil; and the National Center for eHealth, Tromsø, Norway. In Romania, it is led by Netsun.

Participant eligibility criteria for recruitment are presented in Textbox 1.

After registration in the WARIFA app, participants were randomly assigned in a 1:1 ratio by the system, using a computer-generated randomization list. This randomization was stratified by T1D status (yes or no) and center, to ensure balance within each group. The app was downloaded from Google Play, with the same download link for all participants, who were unaware of what exactly to expect from the app. Once downloaded, and after registration, eligibility criteria were confirmed, informed consent was given within the app and participants were randomized: one feature set in the app was available for the control group and another for the intervention group. This procedure was done individually by the participants, avoiding interaction amongst them. Both the researchers and the participants were unaware of group allocation, and the researchers who analyze the data are blinded, too.

In the Norwegian and Romanian centers, the app was downloaded either independently by the participant or with assistance (Norway). In the Spanish center, visits were face-to-face, as blood tests and physical examinations were carried out.

The intervention consists of 12 weeks using a version of the WARIFA app with personalized risk estimations for different NCDs and individualized recommendations, based on each participant’s personal preferences.

To do this, the app uses a simple and intuitive interface, using data available from public databases (for UV index and air quality index) and data from the participants themselves (entered manually, via questionnaires, or automatically, via sensors). The app uses AI to create rule-based recommendations and feedback based on scientific evidence and behavior change models. Bayesian belief networks are used to assess individual risk for CVD in participants with T1D [29], and deep learning for the prediction of glucose values based on previous CGM data [30].

The WARIFA app is downloaded to participants’ own smartphones, who are also provided with a physical activity monitor (Samsung Galaxy Fit 3) to track physical activity, heart rate, and sleep. This is connected to the WARIFA app. Participants with T1D use their glucose sensors (Free Style Libre [FSL] 2 and 3, Abbott), also linked to the WARIFA app.

Participants aimed to use the app for 12 weeks, with 2 face-to-face visits (in Spain only): one at the start of the study and one at the end of the study. At the first visit, the app was downloaded and linked to the activity monitor (all participants) and glucose monitors (T1D participants only). Both visits included blood tests and a physical examination. The Norwegian and Romanian centers did not perform in-person studies, and all participant information was provided through the app in these countries. Questionnaires were completed within the app itself in all centers.

The WARIFA app includes several functionalities (Figure 1):

The control group received a low-intensity intervention, with access to a WARIFA app version where almost all the functionalities mentioned above are available, except for the diary, risk calculators, and personalized recommendations. AI features such as near real-time glucose prediction, night time hypoglycemia risk, and motivational feedback in the top recommendations section are not available in this version, either. Data are collected in a similar manner, and an activity monitor is provided for the duration of the study, too.

Potential end-users, including a T1D patient association, were involved during the co-creation of the WARIFA app. The results of the RCT will be presented in an open access research article, as well as on the project website and on the European Commission’s CORDIS platform [37].

Descriptive results will be presented as mean (SDs), or median (IQRs) for continuous variables, and frequencies and percentages for categorical variables. The primary endpoint is the difference in Likert score between the 2 groups in the self-reported achievement of the individual, self-defined aim at the end of the study. The intervention and control groups will be compared using a Student t test (2-tailed) or a Mann-Whitney U test. We will also perform linear regression to compare the groups with adjustment for the stratification factor center. For comparison of changes in variables measured at baseline and end of study, we will use linear regression or logistic regression adjusting for the baseline measurement [48]. Logarithmic transformation may be used. A P value below .05 will be considered significant.

An intention-to-treat analysis will be performed (including all randomized participants, regardless of their adherence to the intervention).

The sample size was calculated for the primary outcome (Likert scale from 1 to 10 with a target defined by the participant). A population of people with T1D was used as the reference population, based on a previous study [38,49]. The common SD in the intervention and control groups is assumed to be 2.07 [38]. With a significance level of 0.05 and a power of 0.8 in a 2-sided 2-sample t test (2-tailed), 47 subjects are necessary in both groups to detect a statistically significant difference greater than or equal to 1.3 units. Allowing for a potential loss to follow-up of 10%, we estimated a total required sample size of 52 participants in each group. Participants with T1D are only recruited in Spain. All 3 centers aimed to recruit an equal number of participants, that is, 36 participants each (aiming at a total of 108).

The analyses will primarily be performed on the total sample for common outcomes. Regarding the group of participants with T1D, as the study lacks the power to detect clinically relevant changes in T1D-specific clinical outcomes, no definitive conclusions will be drawn from these results. Instead, the results will inform sample size estimation and the design of a subsequent trial targeting the T1D population specifically.

It was submitted to and approved by the Las Palmas Ethics Committee in September 2024 (reference 2024-330-1). It was submitted to clinicaltrials.gov in June 2024 and approved with ID NCT06918444. An insurance policy was signed for the entire duration of the study, which was conducted during the first half of 2025.

Potentially eligible participants were given an informed consent form and an information sheet about the study, in the relevant language (Spanish, Norwegian, Romanian, or English). Those who agreed to participate and signed the form were enrolled in the study. Participants could withdraw from the study at any time. The study participants’ data were anonymized during the study. Participants were not paid for entering or remaining in the study.

The WARIFA project places strong emphasis on data protection, algorithmic support, user autonomy, and transparency in AI‑driven recommendations. Personal and health‑related data are processed under explicit informed consent in accordance with General Data Protection Regulation, with safeguards such as pseudonymization, encryption, strict access control, and privacy‑by‑design measures embedded throughout the system architecture. AI algorithms operate on validated, pseudonymized datasets within a secure environment, providing personalized risk assessments and lifestyle recommendations without engaging in diagnostic or therapeutic decision‑making, ensuring that users remain in control of their health choices and can withdraw consent or correct their data at any time. Transparency is supported through clear privacy policies, detailed consent procedures, and user‑facing explanations of data flows, processing purposes, and the nonclinical nature of the AI outputs, thereby promoting informed engagement and preserving user autonomy while maintaining high ethical and regulatory compliance standards.

The RCT we describe aims to assess the short-term effectiveness of the WARIFA app in promoting users’ healthy behaviors and self-management of T1D. Our hypothesis is that dynamic, personalized recommendations can achieve self-defined objectives and healthy behavior changes to a larger extent than general recommendations for all users.

Other mHealth apps have been used for behavior change and disease management in patients with hypertension and obesity [50,51], although they do not provide estimates of the risk of NCDs.

The main strengths of this trial are that it is conducted in 3 European countries and includes a fairly large and heterogeneous sample size, which should support the applicability of the results. Furthermore, the use of a low-intervention app and a physical activity monitor as control, as well as the double-blind nature of the trial, reduces the risk of bias. The WARIFA app can provide personalized risk estimations for the major NCDs over the next 10 years, as well as identifying the habits to change and the most effective way to do so for each individual. This is in contrast to the generic messages typically used in community-based prevention. The app is able to deliver these personalized messages to participants’ devices to encourage adherence. A key strength of the WARIFA project is its rigorous commitment to data protection, robust algorithmic support, user autonomy, and full transparency in AI‑driven recommendations.

We also acknowledge some limitations in the design of the trial. Although we believe that the number of participants is appropriate for an initial evaluation of the WARIFA app, it may not be sufficient to assess many of the secondary outcomes addressed, especially in T1D. Another limitation is that the biomedical outcomes (anthropometry and blood tests) are performed at one of the centers only, and these data will not be available for all participants. Restrictions also include the fact that the WARIFA app is only compatible with Android mobile phones, as well as certain activity devices and glucose sensors. mHealth apps have boomed in recent years as a tool that is free, voluntary, and easy to obtain. However, many users uninstall the app shortly after installation [52].

Many mHealth apps have demonstrated their effectiveness in promoting healthy habits and managing chronic diseases such as diabetes and cardiovascular conditions; most of them rely on short-term interventions and have limitations in terms of long-term applicability and accessibility for the entire population [53-56]. In contrast, the WARIFA app has been designed using an integrative, personalized approach that targets adults of all ages, including those with low digital literacy skills and seldom heard populations [57,58], areas that are often missed by similar apps. However, long-term applicability is not assessed.

Furthermore, the WARIFA app was developed by actively incorporating feedback from potential users and health care professionals through focus groups. This participatory process allowed the design, functionality, and content to be tailored to users’ real needs. This participatory process improves usability, acceptability, and scalability. WARIFA aims to promote healthy behaviors through a more inclusive and user-centered tool than many existing apps.

This randomized, double blind, controlled, parallel-group trial assesses the short-term effectiveness of the WARIFA health app in promoting healthy behaviors. It will also allow us to identify its weaknesses and limitations, to further improve and develop this, or similar apps. A longer-term study would be needed to assess its long-term effects, including those on the risk for major NCDs, as well as its integration into health care systems.