A widespread aim in current medical research is to derive personalized treatment strategies, which are at the heart of treating every single patient with the best possible therapies. One motivation underlying this aim is that many drugs are only effective in up to 50% of patients [1-3]. Hence, treatment guidelines based on population-level randomized controlled trials (RCTs), which derive the best average treatment, may result in ineffective treatment or side effects in up to 50% of patients. As another characteristic of traditional RCTs, study participants only provide data for population-level analyses and do not profit from participation in the studies. Population-level RCTs are not meant to provide insights for individual participants. The gold standard design for evaluating individual-level treatment effects is prospective longitudinal N-of-1 trials [4], which are multi-crossover RCTs of sample size one [5,6]. That is, the study participant is administered the treatments of interest over time in accordance with a predefined setup of treatment length, duration, treatment blocks, and washout phases. In the literature, N-of-1 trials are sometimes used as a synonym for single-case experimental designs (SCEDs), mainly in the United Kingdom, but mostly used to describe a special case of SCED [7].

N-of-1 trials are well suited for studies when there is large interindividual heterogeneity in treatment effects, as well as when there are subpopulation groups or individuals with comorbidities of interest, who have been excluded from population-level RCTs. Important considerations relate to blinding, randomization, the typically arising correlation of measurements over time, and carryover effects of interventions, which can be approached either through the design of N-of-1 trials or in the subsequent statistical analysis. Series of N-of-1 trials can be aggregated to provide population-level estimates of treatment effects with similar efficiency to that of RCTs but require smaller numbers of participants [4,8,9]. Historically, there have been local implementations of N-of-1 trials in hospitals in the United States, Canada, and Australia [4,10]; series of articles on N-of-1 trials have been published in medical and epidemiological journals [11,12]; and networks on N-of-1 studies have been formed [13]. The advancements in digital technologies provide the potential to initiate a new era of N-of-1 trials in terms of scale and scope and have opened up new avenues to offer remote health care. In particular, performing N-of-1 trials digitally allows a seamless integration of trials in daily life, which can save time for participants and researchers since there is no need to visit a study center. This is especially important if daily measurements are collected. Extensive data can also be assessed passively to further reduce the burden on the participant if sensors are linked to a digital N-of-1 trial app. Finally, recruitment of participants can be simplified as participants from all over the world can participate in a trial, which is even more important for rare diseases where there are few potential participants [14].

Nonetheless, N-of-1 trials have not been integrated into mainstream clinical research or clinical practice. One of the underlying reasons might be that despite recently published guidelines [15-17], there are still considerations of the ethical framework when applying N-of-1 trials in clinical care [18]. Typically, series of N-of-1 trials designed with a specific research aim require ethics approval, while single N-of-1 trials with a clinical aim for a single patient do not require it. Often, however, this distinction is not clear. For example, let us consider a potential N-of-1 trial where physicians have the goal of finding out whether a particular drug is effective in off-label use for patients with chronic conditions such as chronic liver disease. The study setting involves patients treated in a specialized clinic department. In addition to treating the patients, the physicians in charge might be interested in knowing if the results are generalizable. In this situation, a series of N-of-1 trials can be designed, comparing standard care to standard care plus off-label drug use over different crossover periods. This example shows how N-of-1 trials can be woven into clinical practice and, as such, how their innovative design might be of high interest to physicians, if they can be performed easily. The latter point presents maybe the most important reason why N-of-1 trials have not been picked up more broadly since there is no platform available that allows for an easy and large-scale implementation of digital N-of-1 trials. As of now, conducting a digital N-of-1 trial generally necessitates the development of a new app.

Here, we present StudyU. StudyU provides an open-source, free, and easy-to-use platform with a study designer app for researchers, which allows easy design, customization, and implementation of N-of-1 trials, as well as a study participant app that allows participants to partake in these trials without having to set up user accounts. This serves to allow novel interactions among researchers, trials, and study participants.

Here, we have presented the StudyU platform, which allows researchers to undertake an easy design, customization, and implementation of N-of-1 trials and allows individuals to participate in those trials without having to set up user accounts. Through the StudyU Designer, researchers can collaboratively design trials. StudyU is available open-source for iOS, Android, and the web, is free to use, and provides anonymized data entry, which prevents tracing back the data to the participants. It allows conducting the entire study process digitally: study design, participant recruitment, inclusion and exclusion of participants through the study app, automatically analyzing the individual data in the app and reporting the results back to the participant, and saving the data in the secure backend so that researchers can analyze it further and aggregate it across N-of-1 trials. As further innovative concepts, we provide electronic consent and the possibility for the study participants to view their progress through the study on a progress bar. With these features, StudyU is currently the only available platform that allows flexibility to the N-of-1 trial design and the capacity to conduct them completely digitally. All other existing apps have limitations in the platforms they support, the possibility and customizability to design individual trials, the freedom to use the app without having to set up a user account, and the automated in-app statistical analysis to provide the results back to the participant.

As participants likely start the trial with high intrinsic motivation and high expectations of getting insights into their health, it is critical that they are not disappointed and drop out of the trial. The progress bar keeps the participant informed when they have reached the targeted study length and when they can view a statistical evaluation of their results. We expect that it can also encourage them to continue the study for a longer time before viewing the results in order to estimate more precise treatment effects, thereby extending the classical statistical power–based sample size calculation. With this, we envision that participants understand the value of long-term participation in the study and stay motivated for a longer time so that dropout rates can be decreased, thus mixing elements of extrinsic motivation to intrinsic motivation [50]. Regarding the use of progress bars, there is some conflicting evidence in the literature [51,52] regarding whether they have a positive effect on increasing adherence, with some suggestions that only specific types of progress bars (ie, fast-to-slow presentation) are beneficial. For this reason, we propose a new approach of the progress bar in future iterations of StudyU to offer participants the chance to look at the results at any time if they want, with the caveat that their results will only be statistically evaluated once to avoid biased results. We expect positive effects on study adherence of such a design, similar to the endowed progress effect [53].

Using the StudyU platform, N-of-1 trials can be designed not only to study the effect of many different health interventions and lifestyle factors on health outcomes in rare diseases and chronic diseases, but also to evaluate the effectiveness of digital health apps. N-of-1 trials can evaluate the effect of health interventions truly in the real-world setting. Especially with the ongoing COVID-19 pandemic highlighting the importance of remote and digital medicine, evaluation and digital integration into the home environment are of high value. While fully digital trials in the home environment can provide challenges for N-of-1 trials owing to possible carryover or confounding effects that have to be considered in the automated analysis, the challenges can be addressed through the implementation of more advanced statistical and machine learning methods. In fact, recent years have seen an unprecedented development of deep learning methods for estimating the individual-level effects of health interventions from population-level studies and to predict individual disease trajectories and individual treatment effects [54-58]. These methods are often based on nontestable assumptions, require large data sets, and have limitations in their interpretability of individual treatment effects in complex causal graphs. Combining them with the design advantages of N-of-1 trials can help derive fully automated analyses of complex real-life trials.

Two important considerations in N-of-1 trials are randomization and blinding to ensure that unbiased estimates of causal effects can be obtained. Randomizing study designs in a within-persons (eg, the order of treatment A and B within each cycle) and a between-persons manner can be implemented in StudyU if desired, but it should be considered that a deterministic sequence might be able to counterbalance specific time-confounding effects for a given participant, while a randomized sequence can achieve this on average. Blinding can occur on 2 levels: blinding researchers to treatment allocation and blinding study participants to treatment allocation. Researchers can be blinded in StudyU by incorporating another person who controls the allocation in the design of the trial. Blinding of study participants with respect to which interventions they are currently following is not possible in many digital N-of-1 trials, as, for example, drinking tea or using a warming patch are visibly different. Blinding would have to be achieved with help of a researcher, physician, or third person and can be implemented in StudyU by naming interventions anonymized A and B and providing, for example, similar looking pills for A and B. Such blinding can prevent biases that might arise from the participants during the trial. However, this has to be balanced with the aim of N-of-1 trials to benefit and empower the participant. More importantly, it should be remembered that any conclusion that intervention A works better than B for a given participant also holds for any nonblinded trial—we only do not know the extent to which this effect was due to the intervention or accompanying beliefs. However, the participant might not care why the intervention worked but might rather care about the fact that it worked.

We plan to include several extensions in StudyU in the future. First, for the study designer, we plan to add more features encouraging the collaboration on study designs. Setting up a database of interested researchers, clinicians, and institutions can help search for partners in designing and conducting the studies. In the current version of StudyU, all studies are, by default, public to enhance full collaboration and allow for open-access study development. We are working on a more fine-grained collaboration platform, which allows the researcher to make both the creation and conducting of studies fully public or private to a selected group of collaborators and selected group of invited study participants. As a second new feature in StudyU, we will include the possibility to link sensor-based data to measure health outcomes and covariates and also allow the integration of other digital health apps in StudyU. Third, we will provide the possibility to design adaptive trials, for example, including elements from microrandomized trials and just-in-time adaptive interventions [59-61]. Fourth, we plan to implement a more elaborate progress bar visualizing the study progress of the participants. The current progress bar is based on the study duration and number of past measurements, but a more exact measurement would be to focus on the number of nonmissing measurements. This feature can be added by including an automated check for the validity and completeness of the recorded data and feeding it back into the progress bar. Fifth, we plan to integrate more complex statistical and machine learning methods in the study app so that complex individual-level treatment effects of potentially time-varying treatments and time-varying confounders can be included in the modeling and result reporting to the individuals. Currently, only linear regression and t tests are implemented in StudyU. They provide simple models with easily interpretable results and have been shown to provide efficient and robust treatment effect estimates even when autocorrelation and time trends are present [62]. Nonetheless, implementing more complex statistical models such as Bayesian mixed models or G-estimation will allow a more fine-grained and powerful analysis. Finally, we are working on the development of user-centric N-of-1 trials designed by the study participants themselves and are excited to integrate these study designs as well as the study results into StudyU [20]. We envision that placing a higher focus on the cocreation of trials with participants can be very important to increase adherence to the trial, especially for long-term experiments, which is not straightforward as shown in other studies [21,63]. Furthermore, fully cocreated trials, where the participant defines what he/she wants to evaluate, might have a higher chance of exerting an actual effect on health behavior change. It would be interesting to embed such trials into models of health behavior change, such as the one by Prochaska et al [64], and think about which elements can map to each stage of precontemplation, contemplation, preparation, action, maintenance, and termination. Building on this, linking N-of-1 trials further to electronic health records in the future has the potential to connect N-of-1 trials into clinical care and clinical workflows and can further enhance the integration of medical research and clinical practice.