A computer model using eight routine health measures may predict, before surgery, which patients are likely to lose too little weight after sleeve gastrectomy. In a 2026 multicenter study of 1,312 patients from three Chinese hospitals, the model scored 0.821 in testing and outperformed 10 clinicians. According to research reviewed by Gram, body mass index was the strongest predictor.
Weight loss surgery called sleeve gastrectomy helps many people lose weight, but some lose less than expected. According to research reviewed by Gram, scientists built a computer model that uses routine health information gathered before surgery to estimate which patients are likely to lose too little weight within one year. The study included 1,312 patients from three hospitals in China. The model was tested on new groups of patients and performed reasonably well at spotting people who were at higher risk. It also outperformed 10 doctors with different levels of experience. The most useful predictors were body mass index (a measure of body size), age, and fasting blood sugar. The tool is not yet ready for routine use, but it may help doctors plan more personalized care before and after surgery.
Key Statistics
A 2026 multicenter study of 1,312 patients from three Chinese hospitals found that a machine-learning model predicted poor weight loss after sleeve gastrectomy with an AUC score of 0.821 in internal testing.
In the same 2026 study, about 34% of patients in the main group lost less than 25% of their total body weight within one year, the threshold used to define a suboptimal response.
The 2026 computer model outperformed 10 clinicians with varying experience levels when predicting which patients would have a suboptimal weight loss response after sleeve gastrectomy.
The Quick Take
- What they studied: Researchers wanted to know whether doctors could identify, before surgery, which patients are likely to lose too little weight after a sleeve gastrectomy, a procedure that makes the stomach smaller. They tested whether a computer program trained on patient records could make this prediction more accurately than doctors.
- Who participated: The study included 1,312 adults who had sleeve gastrectomy at three large hospitals in China. About 1,094 patients were used to build the computer model, and two separate groups of 84 and 134 patients were used to test it. Patients were considered to have a poor response if they lost less than 25% of their total body weight within one year.
- Key finding: The computer model correctly separated patients who would likely respond poorly from those who would not, with a score of 0.821 on a scale where 1.0 is perfect and 0.5 is no better than guessing. When tested on new patients, the score was 0.761. The model also beat 10 doctors at making this prediction.
- What it means for you: If you are considering weight loss surgery, this research suggests that certain health measures can help flag who may need extra support. This tool is still experimental, so it should not replace your doctor’s advice. Talk with your care team about your own risk factors and realistic goals.
The Research Details
The researchers gathered health records from patients who had sleeve gastrectomy at three hospitals. They looked at many possible factors recorded before surgery, such as body size, age, blood sugar, cholesterol, and vitamin levels. They then tested six different types of machine-learning programs, which are computer methods that learn patterns from past data. One method, called XGB, performed best.
To keep the model simple and useful, the team narrowed the list down to eight factors that are routinely measured in clinics. They then used a tool called SHAP to show how much each factor pushed the prediction up or down. This step makes the model easier for doctors to understand, rather than working as an unexplained black box.
To check the model, the researchers tested it first on patients not used to build it and then on two separate groups from other hospital data. They also compared the model’s predictions against those of 10 clinicians with different amounts of experience, using the same patient information.
Testing the model on separate groups of patients is important because a tool that works only on the data it was built from may fail in the real world. Checking it against other hospitals gives more confidence that it can handle different patients and different record-keeping habits. Comparing it with doctors shows whether it adds real value beyond what clinicians already do.
The study used data from three hospitals, which is stronger than a single-site study. The researchers used outside groups to test the model, and they reported how well it performed in each group. The model also explained which factors mattered most, which helps doctors judge whether the results make sense. However, the study was based on patient records from China, and the full details of how data were collected are not available in the summary provided. Results should be confirmed in other countries and healthcare systems before the tool is widely used.
What the Results Show
Across the three hospitals, about 34% of patients in the main group lost less than 25% of their body weight in one year. That is the cutoff the researchers used to define a poor response. The rates were similar in the smaller groups, at about 40% and 34%.
The best-performing computer model reached an AUC score of 0.821 when tested on patients from the same hospitals. AUC is a score that shows how well a tool separates people who will have a certain outcome from those who will not. A score of 1.0 means perfect separation, and 0.5 means no better than a coin flip. In this case, the model’s sensitivity was 0.803, meaning it caught about 80% of patients who would respond poorly. Its specificity was 0.714, meaning it correctly identified about 71% of patients who would respond well.
When tested on separate groups of patients, the model still performed reasonably well, with a score of 0.761. The 95% confidence interval, which shows the range where the true score likely falls, was 0.688 to 0.823. The model’s predictions also matched actual outcomes reasonably well, and the researchers reported that it offered clinical benefit in their analysis.
In comparison tests, the computer model outperformed 10 clinicians at different experience levels. The authors also reported that the model stayed more accurate in harder cases, where the outcome was less obvious from the basic information.
The factors that most strongly influenced predictions were body mass index, age, fasting glucose (blood sugar after not eating), type 2 diabetes, vitamin B12, calcium, total cholesterol, and gamma-glutamyl transferase, a liver-related blood test. Body mass index was the most influential factor. These findings may help doctors focus on the measures that matter most when assessing surgical candidates. Some of these factors, such as blood sugar and vitamin levels, may also point to areas where patients could receive extra support before or after surgery.
Previous research has consistently found that not all patients respond the same way to weight loss surgery, and that factors like starting body weight, age, and diabetes can affect results. This study builds on that work by combining several factors into one prediction tool and testing it against both experienced and less experienced clinicians. Most earlier efforts relied on a few individual factors rather than a full computer model, so this approach represents an extension of existing knowledge rather than a complete change in understanding.
The study was carried out at three hospitals in China, so the results may not apply to people in other countries, other health systems, or patients with different backgrounds. The model looked only at weight loss one year after surgery, so it does not tell us about longer-term results. The external test groups were fairly small, which makes the score estimates less certain. The model also predicts risk rather than guaranteeing an outcome, and some patients the model flagged may still do well, while some it did not flag may struggle. Finally, a computer tool is not a replacement for a doctor’s judgment, and it has not yet been shown in a real clinical trial to improve patient results.
The Bottom Line
Research shows that a machine-learning model using eight routine health measures can predict, before surgery, which patients are more likely to lose too little weight after sleeve gastrectomy. Confidence is moderate at this stage because the model worked well in the study hospitals and reasonably well in outside groups, but it has not yet been tested in routine care or in a randomized trial. Doctors should consider the tool a possible support for planning, not a final decision maker.
People considering or recovering from sleeve gastrectomy, along with surgeons, bariatric (weight loss) care teams, and hospitals interested in personalized planning, should find this research useful. People who are not considering weight loss surgery, or who have already completed a successful recovery, will likely find less direct value. Anyone should avoid making decisions based on this model alone and should discuss options with a qualified doctor.
This research is at an early stage, so no immediate changes to patient care are expected. Hospitals would need further testing, approval, and training before a tool like this could be used routinely, which may take several years. For patients, any benefit would come through better-planned care before and after surgery, and results would be judged over the one-year period that the study measured.
Frequently Asked Questions
Can doctors predict who will lose weight well after sleeve gastrectomy?
Researchers found that a machine-learning model using eight routine health measures could estimate which patients were likely to lose too little weight. Its score was 0.821 in internal testing and 0.761 in outside groups, where 1.0 is perfect. The tool is still experimental and needs more testing.
What factors predict poor weight loss after sleeve gastrectomy?
A 2026 study of 1,312 patients identified body mass index, age, fasting blood sugar, type 2 diabetes, vitamin B12, calcium, total cholesterol, and a liver-related blood test as the strongest predictors. Body mass index had the greatest influence on the model’s predictions.
Is a computer model better than doctors at predicting weight loss surgery results?
In this study, a machine-learning model outperformed 10 clinicians with different experience levels when predicting poor weight loss after sleeve gastrectomy. The model also stayed more accurate in harder cases. Researchers note that it should support, not replace, a doctor’s judgment.
How much weight do people typically lose after sleeve gastrectomy?
The study defined a suboptimal response as losing less than 25% of total body weight at one year. About 34% of patients in the main group fell into this category, meaning most patients did reach that threshold. Individual results vary widely based on many factors.
Want to Apply This Research?
- Track body weight weekly, along with the percentage of total weight lost since surgery or starting weight. Also log fasting blood sugar, vitamin B12 and calcium results from lab visits, and note any diabetes medication changes. This gives a clear picture of progress that can be shared with your care team.
- Use the app to set a weekly check-in reminder for weighing and logging protein, fluid, and vitamin intake, since these support recovery after surgery. Review your logged data with your care team at each appointment rather than relying on the scale alone, so that early signs of slower progress can be addressed sooner.
- Maintain a long-term log over at least 12 months, with monthly summaries of weight trends and lab values. Add a visual chart showing percentage of total weight lost against the 25% benchmark used in this study. Encourage users to bring their trends to clinic visits so that doctors can judge whether additional support is needed.
This article summarizes a research study for general information only and is not medical advice. The computer model described is experimental and has not been approved for routine clinical use. Anyone considering weight loss surgery, or with questions about their own results, should consult a qualified doctor or bariatric care team before making decisions.
This research translation is published by Gram Research, the science division of Gram, an AI-powered nutrition tracking app.