Your Gut Bacteria May Predict Whether a 5-Day Fast Works Long-Term

If you have ever wondered why a multi-day fast seems to reset one person's metabolism for months while another person's weight creeps right back, new research suggests part of the answer may be sitting in their gut. A team based at Charite Universitatsmedizin Berlin put 38 healthy adults through a supervised 5-day fast, then tracked their gut bacteria, blood chemistry, and body composition for 12 weeks afterward. Using that data, they built a machine-learning model that can flag, before a person even starts fasting, who is likely to keep the weight off. It is early-stage work, but a real step toward matching fasting protocols to the person instead of treating every gut the same way.

Key takeaways

• A model built from two gut bacteria species plus two blood markers predicted which of 38 healthy adults kept weight off 12 weeks after a 5-day fast, with accuracy (AUC) up to 0.86.
• The same model, applied without retraining, worked reasonably well in outside groups with metabolic syndrome (AUC 0.85) and multiple sclerosis, but only moderately in people who fasted much longer (AUC 0.655).
• Most physical changes from fasting, including blood pressure and gut-microbiome shifts, reversed within 12 weeks; only some participants' weight and fat loss lasted.
• This is exploratory science, not a diagnostic: there is no available consumer test that can predict your personal fasting response yet.

Gut Bacteria Levels Predicted Sustained Fasting Weight Loss

A model built from two gut bacteria species and two blood markers correctly separated fasting "responders" from "non-responders" with an accuracy, measured as AUC (area under the curve, a 0.5-to-1.0 score where 1.0 is a perfect prediction), of 0.84 in the study's own healthy-adult cohort, rising to 0.86 when the full microbiome dataset was combined with clinical variables, compared with just 0.69 for clinical measurements like blood pressure and cholesterol alone. The trial, called LEANER and published in Genome Medicine (registered at ClinicalTrials.gov as NCT04452916), defined a "responder" as someone whose body mass index (BMI, weight adjusted for height) dropped at 12 weeks by more than could be explained by normal week-to-week fluctuation; by that yardstick, 38% of the 38 healthy adults tested were responders. The two standout bacteria were an unclassified species of Oscillibacter (Oscillibacter sp. 57_20) and a species of Faecalibacterium, alongside baseline LDL cholesterol and systolic blood pressure.

Findings Could Matter for Metabolic Syndrome and MS

This is not just about people casually trying a fasting app. The model was also applied, without retraining, to real patients with metabolic syndrome and multiple sclerosis (MS), two groups that often use fasting as part of managing their condition. In 34 people with metabolic syndrome who did a 5-day fast followed by a DASH diet, the model predicted sustained weight loss with an AUC of 0.85, nearly matching its performance in the original healthy cohort. In 22 people with MS who fasted for two separate 7-day periods, participants the model flagged as responders lost significantly more weight than those flagged as non-responders. If a version of this classifier is eventually confirmed in larger, dedicated trials, it could help clinicians estimate whether a prolonged fast is likely to produce weight loss that actually lasts for someone managing metabolic syndrome or MS, rather than a rebound.

Randomized, Waitlist-Controlled Trial Design

The LEANER trial randomized 38 healthy adults, ages 20 to 50 with a BMI of 20 to 30, so that 19 fasted immediately for 5 days on 200 to 250 kcal per day of vegetable broths and juices, while the other 19 waited 12 weeks as a control group before doing the identical fast, giving every participant both a control period and a fasting period. Researchers collected stool, blood, and clinical measurements at multiple visits and sequenced gut bacteria with shotgun metagenomics (a method that reads nearly all the DNA in a stool sample, not just a few marker genes). The prediction model itself, a random forest (a machine-learning technique that averages many decision trees to reach a more stable answer), was trained on the 38 healthy participants using 100 repeated rounds of train-and-test splitting, then applied unchanged to three outside datasets covering metabolic syndrome, multiple sclerosis, and a third group of 47 healthy volunteers who fasted for a much longer 6 to 12 days.

What Sustained Weight Loss After Fasting Means for You

If you are considering a supervised prolonged fast, this study points to two realistic expectations. First, weight and fat-mass loss were the only changes that partly stuck around at 12 weeks; blood pressure, resting metabolism, and gut-bacteria composition all shifted noticeably during the fast, then mostly reverted to baseline by follow-up, so a single fast is unlikely to produce a lasting metabolic reset beyond the scale. Second, 83% of participants reported a fasting-related side effect, all described as mild and temporary with none serious enough to stop the fast, and blood glucose ran higher than normal for two to four days after refeeding began, which is worth flagging to a doctor if you have any blood sugar concerns. There is no commercially available stool test today that can measure your personal Oscillibacter or Faecalibacterium levels and translate them into a fasting prediction. This is a research model tested in small groups, not a validated clinical tool, so any consumer "microbiome fasting readiness" product deserves real skepticism until science like this actually backs it.

Small Sample and Exploratory Validation Are Real Limits

The authors themselves are careful to call the three outside-cohort tests "exploratory applications rather than formal external validations," and that caution is warranted. Two of the three cohorts, the MS group and the long-fasting healthy group, lacked a matched control arm, which meant the researchers could not properly calculate sensitivity, specificity, or calibration the way they could for the metabolic syndrome cohort. The model's accuracy also dropped to a more modest AUC of 0.655 in the cohort that fasted the longest (6 to 12 days), suggesting its predictions may not generalize as well outside the 5-to-7-day fasting window it was built on. The core trial included only 38 participants compared against a single waitlist-only control, meaning there was no active diet used for comparison, and researchers did not track what people ate after the fast ended, so some of the sustained weight loss could reflect changed eating habits afterward rather than the fast itself.

Larger Fasting Trials Needed Next

The researchers are calling for larger, harmonized fasting trials, ideally with matched metabolomic data, to confirm whether Oscillibacter sp. 57_20 and Faecalibacterium sp. incertae sedis actually drive weight-loss durability or are simply markers of it. There is already a mechanistic thread worth pulling: a separate 2024 study on long-term fasting identified Oscillibacter species as a major gut source of indole-3-propionate, a metabolite tied to better cardiometabolic health, hinting at a biological reason these bacteria might matter beyond correlation. For now, a blood-and-stool test that tells someone in advance whether a 5-day fast is worth attempting remains a future goal, not something available today.

Common Questions

What counts as "prolonged fasting" in this study? Participants consumed only 200 to 250 kcal per day, through vegetable broths and juices, for 5 consecutive days, which is longer and more restrictive than typical intermittent fasting or time-restricted eating.

Which gut bacteria were linked to sustained weight loss? An unclassified Oscillibacter species (Oscillibacter sp. 57_20) and a Faecalibacterium species were the two microbiome predictors, alongside baseline LDL cholesterol and systolic blood pressure.

Can I get a test today to see if fasting will work for me? No. This is an experimental research model tested in small cohorts of 22 to 47 people, not a clinically approved or commercially available test.

Did the fasting itself cause lasting health changes? Mostly no. Blood pressure, resting metabolism, and gut-bacteria composition shifted during the 5-day fast but had largely returned to baseline by the 12-week follow-up, with only body weight and fat mass showing partial persistence in some participants.

References

1. Gelsomina Kaufhold, Theda U. P. Bartolomaeus, Kristin Schütte, et al. (2026). Machine learning identifies microbiome and clinical predictors of sustained weight loss following prolonged fasting. Genome Medicine. https://doi.org/10.1186/s13073-026-01765-0
2. András Maifeld, Hendrik Bartolomaeus, Ulrike Löber, et al. (2021). Fasting alters the gut microbiome reducing blood pressure and body weight in metabolic syndrome patients. Nature Communications. https://doi.org/10.1038/s41467-021-22097-0
3. Lina Samira Bahr, Markus Bock, Daniela Liebscher, et al. (2020). Ketogenic diet and fasting diet as Nutritional Approaches in Multiple Sclerosis (NAMS): protocol of a randomized controlled study. Trials. https://doi.org/10.1186/s13063-019-3928-9
4. QR Ducarmon, F Grundler, C Giannopoulou, et al. (2024). Long-term fasting remodels gut microbial metabolism and host metabolism. https://doi.org/10.1101/2024.04.19.590209