Safety signals in the literature

Two different questions

A practice checklist asks who should not start, and when to stop a session already running. Symptoms you feel today live there. So do medicines, pregnancy, under-18, and food that already carries shame.

A trial safety paragraph asks what adverse events were collected, who dropped out, and whether a lab flag appeared, in people who already passed the exclusion list. A quiet table after twelve weeks of 16:8 is not “fasting is safe.” It is this protocol, in this group, for this long, with this monitoring.

Do not average those sentences. Headlines do.

Who was not studied is the first signal

Most human TRE and ADF trials enroll adults with overweight or obesity. They typically exclude pregnancy, under 18, underweight, current eating-disorder diagnosis, and often insulin-treated diabetes or complex disease. That is the same narrow door Time-restricted eating in humans and Alternate-day and prolonged fasting studies already named.

A result in that group does not automatically apply to the people the exclusion list named. The list is part of the finding. Silence in those rows is not a blessing. It is a hole.

Who should be careful with fasting exists because the literature is thin there, not because the literature cleared those groups and we copied the press release.

What trials tend to report

The same 2025 network review of 99 randomized fasting and diet trials that Part 3 used also looked at harms when papers reported them. Adverse-event assessment showed up in 56 trials. Twenty-seven of those reported no harmful events in the intermittent-fasting arms.

Among papers that described events, most were mild: constipation, nausea, hunger, diarrhoea, dizziness. A severe event that does appear is hypoglycaemia. One trial reported a participant with hypoglycaemia who fell and stayed in the study. Isolated lab flags (bilirubin, sodium, potassium) show up in a longer trial.

Mild events are still events. Week-one headache and gut snags in Common snags in the first weeks are the home version of that list. They are not a detox, and they are not a reason to ignore stop signals.

Many trials collect weight more carefully than they collect harm. A missing adverse-event paragraph is not proof that nothing happened. Dropout can hide the people who felt unwell and left. Weight, body composition, and adherence already treated staying in the protocol as an outcome, not a personality.

Glucose-lowering drugs

Hypoglycaemia is the serious event that papers do sometimes record. It is more likely when the person is already on insulin or another drug that can drop glucose. A tight morning window plus the same dose you used on three meals is not a studied default.

Trials that enroll type 2 diabetes often adjust medication, add monitoring, or exclude insulin. A quieter glucose curve in an arm that changed the pills is not a license to keep the old dose and skip breakfast. Insulin, glucose, and meal timing already said this is not a diabetes treatment.

The timer cannot see a glucometer. Confusion, shakiness, or a sense you might pass out ends the fast. Eat. Get help. See When to stop a fast.

Eating disorders and restriction

A 2024 commentary from long-running intermittent-fasting researchers is blunt on this point: the trials they reviewed did not show enrolled adults developing an eating disorder. Those trials also screened out people with an eating-disorder history. You cannot find what you excluded.

The same commentary tells clinicians not to treat that quiet table as clearance for anyone with an eating-disorder history, and to be careful with adolescents who already carry a high risk. Cross-sectional surveys sometimes find more restrictive eating among people who already use a clock. That can be selection (people who restrict pick a window) as much as a clock causing a disorder. Either way, a timer can become a tool for harm.

If food already carries shame, rigid rules, or a binge-restrict loop, do not add elapsed time as a score. Part 2 is the practice stop. “The RCTs did not show it” is a weak reassurance for the people those RCTs never enrolled.

Lean mass, rapid loss, and longer fasts

Lean mass can fall in an energy deficit, with a clock or without one. Part 7. That is a composition finding. It is safety-adjacent if the person is older, already lean, or skipping protein and some resistance work. Exercise and fasting is the training half of that sentence, not a dare to train through a crash.

Rapid weight loss in the wider diet literature is a known gallstone risk. Fasting trials rarely make gallstones a primary outcome. Do not invent a unique 16:8 epidemic. Do not pretend a fast drop is free.

Prolonged fasts are a different intervention. Human evidence is thinner and more often supervised. Refeeding after several days is a medical topic, not a “first meal flex.” Clinic water fasts are not a 48-hour preset on a phone. See Part 3, Open-ended and multi-day fasts, and Breaking a fast gently.

Observational scares are not trial results

Analyses of survey data have linked very short self-reported eating windows with higher cardiovascular mortality. That design watches people who already eat a certain way. Shift work, illness, sleep, income, and who remembers yesterday’s meals travel with the hours. Part 1 already said observational papers cannot cleanly prove the clock caused the death.

The useful reading is not “TRE kills” and not “TRE is proven safe for decades.” Randomized fasting trials still rarely follow hard outcomes for years. A quiet twelve-week adverse-event table does not fill that hole. Open questions (what we still do not know) sits there.

Limits (read these as part of the result)

Who: screened adults, usually with extra weight. Findings do not transfer to pregnancy, under-18, underweight, eating-disorder history, or insulin-treated diabetes unless a paper actually enrolled those people and said so.

Duration: weeks to a few months. A decade of consumer 16:8 is not what these tables measured.

Collection: harm is under-reported relative to weight. Protocols mix TRE, ADF, 5:2, and longer fasts under one “intermittent fasting” label. Do not average the risk pictures.

Dropout is a hidden harm signal. Remaining averages look tidier than the week that made someone leave.

Medication changes inside a trial are part of the intervention. Copying the window without copying the clinical follow-up is a different experiment.

What this does not mean

It does not mean intermittent fasting is dangerous for everyone who passed a normal overnight gap. Mild, common complaints are not a reason to treat a 13-hour fast as a toxin.

It does not mean fasting is cleared. “Safe in this trial” is not a personal clearance, and it is not a treatment claim for diabetes, fatty liver, cancer, or longevity.

It does not replace the practice notes. Stop signals still win. Medicines still need a prescriber. This page will not rank 16:8 versus ADF versus a water fast as “safer.” They are different tools with different costs.

It is not hormones or cycle advice. That question is later, and it is careful. It is not a reason to add a scale so you can refresh a harm story hourly.

How this meets the timer

The clock measures elapsed time. It will not auto-stop you for symptoms, labs, or a medicine you take with food. Unlimited tracking is flexibility. It is not evidence that a longer preset is safer, or that a 48-hour session is research-backed because a 12-week 16:8 arm was quiet.

End the session when you eat, including when you stop early. Log messy days. Custom 12 or 14 is a complete option. You can run the clock in the browser. Stage badges remain teaching labels. They are not a safety dashboard.

Next in this cluster: Open questions (what we still do not know). For this week, Who should be careful with fasting and When to stop a fast still decide more than a thread that says “the studies say it is safe.”