After surgery for gastrointestinal cancer, AI-assisted nutritional support improves short-term intake

Personalized support versus standard advice

In this single-center trial, researchers assigned patients undergoing laparoscopic surgery to two groups. The first received standard dietary advice at hospital discharge. The second received individualized recommendations generated using DeepSeek v3.2 from prompts structured in an ABC format, then reviewed by a clinician.

The intervention therefore combined several elements: AI, personalization, numerical targets, written recommendations and clinical review. The primary outcome was the proportion of patients reaching at least 90% of their energy target 30 days after discharge. Outcome assessors were unaware of patients’ group assignments.

More patients meet their dietary targets

At 30 days, 66.7% of patients in the AI-assisted support group reached the energy intake threshold, compared with 40.0% in the standard group. The proportion achieving protein intake considered adequate was also higher: 62.2% versus 35.6%. These differences were statistically significant.

Patients in the intervention group also reported less fatigue, nausea and vomiting, appetite loss and gastrointestinal symptoms. However, no statistically significant differences were observed in weight change, blood-based nutritional markers or complications. Better self-reported intake in the short term does not, on its own, demonstrate better clinical recovery.

Time saved, but no proof that AI is superior

Experts judged the intervention group’s recommendations to be more comprehensive, more personalized and easier to put into practice. The time healthcare professionals spent per patient was also lower: 8.2 minutes versus 12.5 minutes.

These results support the value of the complete approach, not AI in isolation. Without a group receiving equally personalized advice with numerical targets, but prepared without AI, it is impossible to know how much of the benefit comes from the software rather than the quality of the support.

What this trial cannot yet establish

The 30-day follow-up provides no information on whether the benefits last. The small sample and recruitment at a single center also limit the scope of the findings. These results concern patients undergoing surgery for gastrointestinal cancer, not the general population looking to improve their diet.

The dossier provides access only to the PubMed abstract, not the full text. Methodological details, funding and conflicts of interest therefore could not be verified. Clinical supervision is part of the intervention studied: these results do not validate using a chatbot independently after an operation.

What we can take away

This trial supports personalized AI-assisted nutritional support reviewed by a clinician. It demonstrates neither the superiority of AI itself nor a lasting clinical benefit.