npj Digital Medicine is an open-access, peer-reviewed journal published by Nature Portfolio that focuses on the intersection of digital technology and healthcare. Launched in 2018, it covers research on topics ranging from artificial intelligence in clinical settings to wearable sensors, telemedicine, and software-based medical devices. The journal has become one of the primary venues for publishing work on how computing, data science, and connected devices reshape how medicine is practiced and studied. Its open-access model and broad scope have made it a natural home for a field that moves faster than most areas of clinical research.
What the Journal Publishes
The journal accepts several distinct content types, each suited to a different kind of contribution. Full research articles are expected to present substantial original findings from primary investigations. Brief Communications cover concise, high-quality studies that do not rise to the scale of a full article but still address questions of broad interest. Reviews synthesize existing literature across a topic area. Comments offer a flexible format for discussing policy, science-and-society questions, or purely scientific issues. Perspectives give authors room to present evidence-based opinions, models, or conceptual frameworks. The journal also publishes Editorials and Matters Arising, the latter allowing formal responses to previously published work.1npj Digital Medicine. Content types
This range of formats reflects the nature of the field itself. Digital medicine sits at the boundary of engineering, clinical practice, regulatory science, and public health, and many of the most useful contributions are not traditional bench-to-bedside experiments. A policy Comment about how regulators should handle AI-based diagnostics, or a Perspective laying out a new framework for validating algorithms, can shape the field as much as a randomized trial. The journal’s willingness to publish across that spectrum is part of what distinguishes it from more narrowly focused clinical or informatics outlets.
Open Access and Data Sharing Requirements
Every article published in npj Digital Medicine is freely available to read without a subscription, which matters in a field where clinicians, engineers, policymakers, and patient advocates all need access to the same findings. The journal also enforces specific transparency requirements beyond what many traditional medical journals demand.
All submitted manuscripts must include a Data Availability Statement explaining where the data underlying the study can be found and under what conditions it can be accessed. For studies that rely on custom computer code or analysis scripts central to the results, those tools must be made available to editors and peer reviewers during the review process. At publication, the journal considers it best practice for authors to release their code publicly so that readers can reproduce the published findings.2Nature. Submission guidelines | npj Digital Medicine Any practical barrier to sharing code is evaluated by the editorial team, and manuscripts can be declined if important code is withheld without justification.
These policies respond to a persistent problem in digital health research. An algorithm that works beautifully in one paper but cannot be inspected, tested on new data, or re-run by independent teams is not much use to a field that needs to build on prior work. Code and data sharing requirements push authors toward the kind of transparency that makes replication possible, even if full replication remains difficult in practice given variations in patient populations and clinical settings.
Where Digital Medicine Research Comes From
The global landscape of AI-related medical research is concentrated in a handful of countries, and research published in npj Digital Medicine reflects that pattern. A large-scale analysis of over 5,400 AI medical studies found that the top ten contributing countries accounted for about three-quarters of all research output. The United States and China together represented roughly half of that total, with the U.S. contributing around 25% and China around 22% of all country-level contributions.3Nature. A quantitative analysis of global AI medical studies: gaps in randomized controlled trials
The dominance of those two countries was even more pronounced in randomized controlled trials, which are generally considered the strongest form of clinical evidence. Of roughly 140 RCT-level country contributions, the U.S. accounted for about 26% and China for about 24%, followed by South Korea, the Netherlands, and the United Kingdom, each contributing small single-digit percentages.3Nature. A quantitative analysis of global AI medical studies: gaps in randomized controlled trials Annual country contributions peaked in 2021 and then declined, which may reflect a broader post-pandemic cooling in research output or a maturation of the field as the initial wave of exploratory studies gave way to more targeted investigations.
This concentration raises real questions about generalizability. Algorithms trained and validated primarily on American or Chinese patient populations may not perform as well in sub-Saharan Africa or Southeast Asia, where disease patterns, healthcare infrastructure, and available data differ substantially. The journal has published work from multi-country collaborations spanning dozens of hospitals across multiple continents, including the 4CE consortium, which during the early COVID-19 pandemic pulled together electronic health record data from 96 hospitals across the United States, France, Italy, Germany, and Singapore within just three weeks. Those kinds of international efforts remain the exception rather than the norm.
The Gap Between AI Hype and Rigorous Evidence
One of the recurring themes in the journal’s published research is the mismatch between the breathless pace of AI development in healthcare and the comparatively thin base of rigorous clinical evidence supporting it. Venture capital funding for healthcare AI companies reached roughly $3.6 billion in a five-year span around the late 2010s, yet a combined search for publications on machine learning and graduate medical education between 2010 and 2017 turned up just 16 papers, and none of them actually focused on teaching medical professionals about machine learning.4Nature. Machine learning and medical education Billions of dollars flowed into building AI tools while essentially nothing was being published about training the doctors who would use them.
That gap has narrowed somewhat since 2018, but the broader pattern persists. The same analysis of over 5,000 AI medical studies found that randomized controlled trials made up a small fraction of the total output. Most studies were retrospective analyses, proof-of-concept demonstrations, or validation studies that tested algorithms on existing datasets rather than in real clinical workflows. This is understandable from a development standpoint, since you generally want to know an algorithm works before putting it in front of patients, but it means the field’s evidence base remains tilted toward showing that something could work rather than proving it does work in practice.
The Growth of FDA-Authorized Digital Medical Devices
npj Digital Medicine frequently publishes research tracking the regulatory landscape for software-based medical tools, and the numbers reveal how quickly this space has expanded. An analysis of nearly 48,000 medical devices authorized by the FDA between 2005 and 2024 found that about 28% included software as a component, encompassing both software embedded inside physical devices and standalone software applications used for medical purposes.5Nature. Two decades of growth and trends in the FDA authorization of digital medical devices
Within that group, roughly 920 were classified as Software as a Medical Device, meaning the software itself is the medical product rather than something controlling a piece of hardware. The vast majority of these authorizations, over 99%, came through the FDA’s 510(k) pathway, which allows a new device to reach the market by demonstrating that it is substantially equivalent to a device already on the market. Only about 0.4% used the de novo pathway, reserved for novel devices without a clear existing equivalent.5Nature. Two decades of growth and trends in the FDA authorization of digital medical devices
The distribution across medical specialties was strikingly uneven. More than three-quarters of radiology devices included software, which makes sense given how central image analysis has become to that field. In contrast, only about 7% of orthopedic devices did. Radiology’s head start in digitization, combined with the natural fit between AI image recognition and reading scans, has made it the dominant specialty for software-based medical tools. Whether other specialties catch up depends on whether similar technical advantages emerge for their workflows.
Algorithmic Bias in Clinical Decision Tools
A significant strand of research in the journal addresses how algorithms and scoring tools used in everyday clinical practice can carry biases that affect patient care. Clinical decision instruments, the risk scores and calculators that help physicians decide whether to order a CT scan or admit a patient, are not immune to the biases embedded in the data they were built on. Researchers publishing in npj Digital Medicine have found that these tools can be biased at multiple levels: in which patients were included in the original study populations, in who designed the tools, in which predictor variables were selected, and in how outcomes were defined.
To address this, researchers have proposed that journals and online platforms adopt standardized reporting frameworks. One suggested approach borrows the concept of a “model card” from the machine learning community, a structured summary of a model’s purpose, development context, performance characteristics, and known limitations. The proposal extends this idea with a dedicated bias analysis highlighting the demographics of the original study cohort, any external validation data, and known disparities in performance across patient subgroups. The goal is to make it straightforward for a clinician pulling up a scoring tool to understand whose data it was built on and where it might fall short.
This line of work reflects a broader concern in digital medicine: tools that appear neutral and objective because they are algorithmic can quietly encode the biases of the populations and contexts they were developed in. A risk score validated primarily on white patients at academic medical centers may underperform for Black patients in community hospitals, and without transparent reporting, clinicians have no way to know that.
Developer Involvement and Trial Independence
When a company builds a digital health tool and then also runs or sponsors the clinical trial evaluating it, the question of independence naturally arises. A rapid review published in npj Digital Medicine examined 229 trials drawn from 29 systematic reviews and found that about 73% of trials for digital health interventions involved direct participation by the product’s developers, while only 27% were conducted independently.6Nature. Clinical trials for digital health interventions: a rapid review of study independence and the developer effect
Developer-involved trials were more likely to be preregistered, meaning the researchers posted their study plan publicly before collecting data. That is generally considered a positive sign for transparency. But when the results were weighted by sample size, developer-involved trials also had higher odds of reporting statistically significant positive results than independent trials, with a modest but clear difference.6Nature. Clinical trials for digital health interventions: a rapid review of study independence and the developer effect This “developer effect” is not unique to digital health; it mirrors patterns seen across pharmaceutical and device research, where industry-sponsored trials tend to produce more favorable outcomes than independent studies.
The finding does not mean developer-involved trials are dishonest. Companies naturally have more resources, better technical knowledge of their own product, and stronger incentives to design trials that showcase their tool’s strengths. But it does mean that readers, clinicians, and policymakers should pay attention to who ran a trial when interpreting its results. The lopsided ratio, with nearly three-quarters of all trials involving the developer, also suggests that independent evaluation of digital health tools remains underfunded and uncommon. If the field wants clinicians and health systems to trust its products, more of the evidence base needs to come from researchers who did not build the thing being tested.
Toward Global Standards for Medical Software
As software-based medical devices proliferate, the question of how to regulate them consistently across different countries has become increasingly urgent. A paper published in npj Digital Medicine introduced a proposed framework called Good Digital Medicine Practices, intended to serve as an operational reference model for the lifecycle management of medical software. The framework emphasizes continuous validation, meaning that algorithms should be reassessed over time rather than approved once and left alone. It also calls for algorithmic transparency, risk-proportionate oversight that scales regulatory burden to the potential for patient harm, and international convergence so that a tool approved in one country does not face a completely different and contradictory set of requirements elsewhere.7PubMed Central. Toward global standards for SaMD: introducing a proposal for Good Digital Medicine Practices (GDMP)
The need for such a framework is not theoretical. Medical software can be updated far more easily than a physical implant or drug formulation, which means the product a regulator approved six months ago may no longer be the product patients are using. An algorithm that learns from new data can drift in performance over time, potentially becoming less accurate for certain populations or clinical contexts. Traditional regulatory models, built around the idea that a device is a fixed, physical thing, struggle to accommodate software that changes continuously.
The proposed framework draws on real-world regulatory experience from multiple jurisdictions, aiming for a model that countries at different stages of digital health adoption could realistically implement. Whether it gains traction remains to be seen, but the fact that the conversation is happening at all in a peer-reviewed venue reflects the maturation of digital medicine from a scrappy startup field into one grappling with the institutional infrastructure needed to operate safely at scale.
What Prospective Authors Should Know
For researchers considering submitting to npj Digital Medicine, a few practical realities are worth keeping in mind. The journal’s scope is deliberately broad within the digital health space, but it sits within the Nature Portfolio family, which means editorial standards lean toward work with clear clinical relevance rather than purely technical contributions. An engineering paper that develops a clever algorithm without demonstrating why it matters for patient care or clinical workflow will likely be directed elsewhere. Conversely, a well-designed clinical study of a digital intervention, even a small one, can find a home here if the question is interesting and the methods are sound.
The open-access model means authors typically pay an article processing charge, which is standard across Nature Portfolio’s npj series. The data and code sharing requirements described earlier are not optional. Manuscripts that withhold code central to their main findings without a compelling practical justification risk rejection on transparency grounds alone.2Nature. Submission guidelines | npj Digital Medicine For teams working with proprietary algorithms or sensitive patient data, this means planning early for how to share enough of the underlying materials that independent evaluation is feasible.
The journal also publishes Comments and Perspectives, which offer entry points for researchers who have something substantive to say about the direction of the field but do not have a traditional dataset to present.1npj Digital Medicine. Content types Some of the most widely read and cited pieces in digital medicine journals have been thought pieces that reframed how the field thinks about a problem, rather than empirical studies. If you have spent years implementing AI tools in a health system and have hard-won insights about what actually works, that kind of contribution can be as valuable as a controlled trial.