A systematic review and Bayesian network meta-analysis of digital interventions for depression and anxiety
This systematic review and Bayesian network meta-analysis of 404 randomized studies involving over 110,000 participants found that digital interventions modestly outperform primary controls for depression and anxiety, but inconsistent evidence prevents the identification of a superior delivery modality or architecture.
Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer
Depression and anxiety are among the most common challenges to human well-being, affecting hundreds of millions of people worldwide. While effective treatments exist, many individuals cannot access them due to cost, a shortage of trained professionals, or the stigma surrounding mental health care. In recent years, technology has offered a potential bridge across this gap. Digital interventions—programs delivered through websites, smartphone applications, or text messages—promise to make psychological support more accessible and scalable. However, the landscape of these tools is vast and confusing. Some programs follow a rigid, pre-written script, while others rely on a human therapist to guide the user. Some are delivered via a computer screen, while others arrive on a phone. As new technologies emerge, including those powered by advanced artificial intelligence, it has become difficult to know which specific combination of content and delivery method actually works best for whom.
To cut through this complexity, a team of researchers from universities across China conducted a massive review of existing scientific evidence. They gathered data from 404 randomized studies involving more than 110,000 participants. The researchers did not just look at whether digital tools worked in general; they sought to understand how the way an intervention is built and delivered changes its effectiveness. They categorized the interventions by their "architecture," meaning how the therapeutic content is organized or generated, and by their "delivery modality," meaning the channel used to reach the user. The architectures ranged from traditional human-led therapy to fixed modules, rule-based interactions, and new systems powered by large language models. The delivery channels included face-to-face sessions, telephone calls, text messages, websites, and mobile applications. By analyzing these thousands of data points together, the team could compare these different configurations directly to see which ones produced the greatest relief from symptoms.
The overall picture that emerged was one of modest but meaningful improvement. When compared to standard care or no treatment, digital interventions helped reduce symptoms of depression and anxiety. The researchers found that, on average, these digital tools provided a benefit roughly equivalent to one-third of a standard deviation in symptom reduction. This is a small-to-moderate effect, suggesting that while these tools are not a cure-all, they offer a genuine advantage over doing nothing. The study confirmed that digital mental health support is a viable option for many people, but it also revealed that there is no single "best" technology that works for every situation.
When the researchers looked at the specific channels used to deliver care, a clear pattern appeared depending on when the results were measured. For immediate results right after a program ended, web-based interventions ranked highest. These are programs accessed through a browser on a computer or tablet. However, when looking at how well the benefits lasted over time, mobile applications took the lead. Apps, which include software installed on smartphones and tablets, showed the strongest ability to maintain improvements in symptoms during follow-up periods. Despite these rankings, the researchers noted that the difference between using a website and an app was not statistically certain, meaning the data did not definitively prove one was superior to the other in every case.
The study also examined the structure of the content itself. A particularly notable finding concerned interventions powered by large language models, a type of artificial intelligence capable of generating human-like text. In the studies available at the time of the review, these AI-driven interventions ranked very highly for both immediate and long-term relief of symptoms. In fact, they often appeared at the top of the rankings for reducing both depression and anxiety. However, the researchers cautioned that this evidence is still emerging. The studies involving these AI tools were fewer in number, and the technology is evolving rapidly. The high rankings suggest great potential, but the limited amount of long-term data means we cannot yet say with full confidence that these AI tools are better than traditional human-led therapy. The study found that while AI tools performed well, the evidence was not strong enough to declare them the definitive winner over all other methods.
The researchers also investigated what factors might make these interventions work better for some people than others. They found that individuals who started with more severe symptoms tended to see larger improvements. Additionally, programs that included some form of human guidance—where a real person provided feedback or encouragement—generally produced better results than those that were entirely self-guided. This held true for both depression and anxiety, reinforcing the idea that human connection remains a valuable component of digital care. The study also looked at whether the specific type of therapy, such as cognitive behavioral therapy, or the length of the program mattered, but the results here were mixed and varied depending on the specific outcome being measured.
Despite these promising findings, the study highlighted significant limitations in the current body of evidence. Many of the included studies had methodological weaknesses, such as relying on self-reported symptoms rather than clinical diagnoses, which can introduce bias. The researchers also found inconsistencies across the different studies, meaning that the results did not always align perfectly when compared to one another. This lack of consistency lowers the confidence we can have in the precise rankings of one tool over another. Furthermore, the evidence for long-term effects, especially for the newer AI-based tools, was sparse. The study did not find enough high-quality data to definitively say that one specific architecture or delivery method is the gold standard for clinical care.
Ultimately, this massive review provides a clearer map of the digital mental health landscape, showing that these tools work but that the field is still maturing. The findings suggest that web-based programs are excellent for immediate relief, while apps may be better for sustaining that relief over time. The emergence of artificial intelligence as a powerful tool is exciting, but it requires more rigorous testing to understand its full potential and limitations. For now, the most important takeaway is that digital interventions offer a helpful, accessible option for many people struggling with depression and anxiety, even if the perfect technological solution has not yet been identified. The path forward involves continuing to refine these tools, ensuring they are safe and effective, and gathering more high-quality evidence to guide future care.
Drowning in papers in your field?
Get daily digests of the most novel papers matching your research keywords — with technical summaries, in your language.