Scientists reviewed 65 studies that used artificial intelligence to understand healthy living habits like exercise, eating, sleep, and stress. They found that most studies combined information from multiple areas of life to get a better picture of health. However, the researchers discovered that many studies didn’t explain how their computer programs made decisions. The review provides guidelines to help future researchers do better work by being clearer about their methods and making their AI systems easier to understand. This could eventually lead to personalized health advice tailored just for you.
The Quick Take
- What they studied: How scientists use artificial intelligence and computer learning to analyze data about healthy habits (exercise, diet, sleep, and stress) to predict or understand health outcomes
- Who participated: This was a review of 65 published research studies from medical and psychology databases. The studies themselves included various groups of people tracking their lifestyle habits
- Key finding: Most studies (74%) looked at multiple lifestyle areas together, which gives a better picture of health. However, only about one-third of studies explained how their AI made its decisions, which is a major gap in the research
- What it means for you: Better AI explanations could eventually help doctors and apps give you personalized health advice. Right now, researchers need to improve how they collect data and explain their methods so we can trust these systems more
The Research Details
Scientists searched three major medical research databases (PubMed, PsycINFO, and Web of Science) for studies that used machine learning—a type of artificial intelligence that learns patterns from data—to study healthy lifestyle habits. They found 65 studies that met their criteria and analyzed what methods each study used, what types of data they collected, and how they explained their results.
The researchers looked at several important things: what lifestyle information the studies tracked (like steps walked, meals eaten, hours slept), where the data came from (whether people entered it themselves or if it came from health records), which computer programs and algorithms were used, and whether the studies explained how their AI made decisions.
This approach, called a scoping review, helps researchers understand the current state of a field and identify where improvements are needed. It’s like taking a survey of all the work being done in an area to see what’s working well and what needs fixing.
Understanding how researchers are currently using AI for health is important because these tools could eventually help millions of people get personalized health advice. However, if the methods aren’t clear or the AI can’t explain its decisions, doctors and patients won’t trust the results. This review helps set standards for better research going forward
This is a high-quality systematic review that followed international guidelines for how to conduct and report such reviews. The researchers searched multiple databases to find studies, which reduces the chance of missing important work. However, the review itself doesn’t test new treatments or collect new data—it analyzes existing published studies. The strength comes from analyzing 65 different studies to identify patterns and gaps in the field
What the Results Show
The review found that most studies (48 out of 65, or 74%) combined information from multiple lifestyle areas—like tracking exercise, food, sleep, and stress together. This is good because health is complicated and connected; looking at just one thing doesn’t tell the whole story.
About half the studies used data that people entered themselves (like through apps or surveys), while the other half used data from health records or medical systems. The most popular computer program used was called ‘random forest,’ which is like having many decision trees vote on the answer.
A concerning finding was that only 33 out of 65 studies (about one-third) used methods to explain how their AI made decisions. This is important because if you can’t understand why a computer recommended something, it’s hard to trust it. Most studies that did explain their AI used a method called ‘Shapley values,’ which shows which lifestyle factors were most important in the decision.
The review found that most studies measured lifestyle habits with simple, single questions rather than detailed tracking. For example, asking ‘How much did you exercise?’ instead of tracking every workout. While this is easier, it may not capture the full picture. The researchers also noted that most studies used Python or R (computer programming languages) to build their AI systems, which is good for transparency since these are free and widely available tools
This review appears to be one of the first to systematically examine how AI is being used in lifestyle research and to identify the gaps. Previous research has shown that AI can be useful for health predictions, but this review shows that the field needs better standards for how studies are done and reported. The finding that most studies don’t explain their AI decisions aligns with broader concerns in medicine about ‘black box’ AI that doctors can’t understand
This review only looked at studies that used supervised machine learning (a specific type of AI where the computer learns from examples). It didn’t include studies using other types of AI. Also, the review couldn’t assess how accurate or useful the AI systems actually were in real life, since that information wasn’t always reported in the studies. The review also notes that most studies had limited diversity in who participated, so the results might not apply equally to all populations
The Bottom Line
If you’re interested in using AI-based health apps or tools: Look for ones that explain their recommendations (moderate confidence). Work with your doctor before making major health changes based on AI suggestions (high confidence). Track multiple lifestyle areas together rather than just one (moderate confidence). Be aware that these tools are still improving and may not work perfectly for everyone yet (high confidence)
This research matters most for: Researchers and companies developing health apps and AI tools; Doctors and health professionals considering AI tools for their patients; People interested in personalized health advice; Anyone using fitness trackers or health apps. It’s less immediately relevant for people just looking for basic health advice, though it supports the value of tracking multiple lifestyle factors
This review doesn’t test a treatment, so there’s no timeline for seeing health benefits. However, it suggests that better AI tools following these guidelines could be available within 2-5 years as researchers improve their methods. If you start using improved tools, you might see personalized recommendations within weeks, though seeing actual health changes would take months
Want to Apply This Research?
- Track at least three lifestyle areas daily: minutes of physical activity, meals eaten (or calories), and hours of sleep. Rate your stress level 1-10 each evening. This gives the app enough information to find patterns and give you better personalized suggestions
- Instead of focusing on just one health goal, pick one small change in each lifestyle area: add 10 minutes of walking, swap one sugary drink for water, and go to bed 15 minutes earlier. The app can help you track all three together and show how they affect each other
- Review your lifestyle data weekly to spot patterns. For example, notice if you sleep worse on days you’re stressed or exercise less. Share this data with your doctor or health coach so they can give you personalized advice. Keep tracking for at least 4-8 weeks to see meaningful patterns emerge
This review analyzes research methods, not proven treatments. The AI tools and methods discussed are still being developed and improved. Before making any significant changes to your diet, exercise, sleep, or stress management based on AI recommendations, consult with your doctor or a qualified healthcare provider. AI health tools should complement, not replace, professional medical advice. Results from AI systems may vary based on individual differences, data quality, and other factors. Always verify important health decisions with a healthcare professional.
This research translation is published by Gram Research, the science division of Gram, an AI-powered nutrition tracking app.