Click anywhere or press Esc to close
Clinical-evidence audit
This is not medical advice — it's a clinical-evidence audit to help you prepare for a second opinion with a doctor. Don't start, stop, or change any prescription based on it alone.
| The person | Man, 6′2″, 265 lb (BMI 34) · ADHD, on a daily amphetamine · goal: 200 lb |
|---|---|
| The plan | Oral semaglutide (Wegovy Pill), titrated 1.5 → 4 → 9 → 25 mg · 1,500–1,800 kcal, "small, frequent meals" · ≥150 min/wk aerobic + resistance 2–3× · labs (glucose, A1C, lipids) · indefinite therapy |
| How it was built | OpenEvidence (the evidence tool his doctor uses) → then weeks of refining with ChatGPT, which produced the 9-section plan above |
| He assumed | Two good AI tools, cross-checked against each other, had covered it |
| His ask | Is this plan right — and is there anything it's missing? |
Two AI tools built this plan — each one fixed the other, and each one broke something new
| Layer | What it got right | What it got wrong |
|---|---|---|
| OpenEvidencethe tool his doctor uses | Every citation was real. We re-checked all 10 against PubMed and Crossref. 10 out of 10 authentic — nothing invented. | "Take 25 mg." No escalation ladder. But 25 mg is a maintenance dose you reach over months — starting there risks severe nausea, vomiting, dehydration and kidney injury. (the evidence) |
| + ChatGPTweeks of the patient's own prompting | It caught the titration — and got the ladder right (1.5 → 4 → 9 → 25 mg, the correct schedule for this drug). It added nutrition, activity, labs and monitoring. It genuinely improved the plan. | It stripped every citation. A plan that was fully traceable became impossible to check. And when asked to list its sources, 1 of its 73 DOIs was invented — a JAMA paper that does not exist. |
| The audit | Found the five things neither of them looked for — the root cause nobody measured, the muscle, the sleep apnea, the amphetamine, and what happens the day he stops. | |
10.1001/jama.2018.0588 — returns 404 from Crossref and from doi.org. It is formatted perfectly: right registrant prefix for JAMA, plausible year, plausible sequence. There is no such paper. See for yourself ↗Glossary
We deliberately didn't guess on these — and several flags below shift depending on them:
The first one is the reason we ask for the whole list, not just the drug in question.
Five flags — the things neither tool went looking for. They're interrelated: see the diagram.
The big picture — how the evidence changed the causal model
① Causal Diagram: OpenEvidence (implicit)
flowchart TB
RX{{"Oral semaglutide 25 mg
the first answer"}}:::oe
APP["appetite ↓"]:::oe
WT(("Weight → 200 lb
a proxy")):::proxy
GOAL["Focus & energy
what he actually wants"]:::goal
RX --> APP --> WT ==> GOAL
classDef oe fill:#dbeafe,stroke:#2563eb,stroke-width:2.5px,color:#1e3a8a;
classDef proxy fill:#0f172a,stroke:#0f172a,color:#ffffff;
classDef goal fill:#ffffff,stroke:#334155,stroke-width:2.5px,color:#0f172a;
OpenEvidence’s first answer. The drug lowers appetite, appetite lowers weight, and weight is assumed to deliver what he came for. One road — and it’s only implied: you have to infer the causal chain yourself.
② Causal Diagram: ChatGPT’s enrichment of OpenEvidence (implicit)
flowchart TB
RX{{"Oral semaglutide 25 mg
the first answer"}}:::oe
APP["appetite ↓"]:::oe
WT(("Weight → 200 lb
a proxy")):::proxy
GOAL["Focus & energy
what he actually wants"]:::goal
TITR{{"Titration 1.5→4→9→25 mg"}}:::gpt
RES{{"Resistance training
preserves muscle (25–40%)"}}:::gpt
AERO{{"Aerobic 150 min / week"}}:::gpt
KCAL{{"Calorie deficit
1,500–1,800 kcal"}}:::gpt
DUR{{"Durability: 50–67% regain flagged
— but no taper (plan: indefinite therapy)"}}:::gpt
AMPH{{"Amphetamine timing"}}:::gpt
EXP["Energy
expenditure ↑"]:::mech
RX --> APP --> WT ==> GOAL
TITR --> RX
AMPH --> RX
RES --> WT
AERO --> EXP --> WT
KCAL --> APP
DUR --> WT
RES -. dopamine .-> GOAL
AERO -. dopamine .-> GOAL
classDef oe fill:#dbeafe,stroke:#2563eb,stroke-width:2.5px,color:#1e3a8a;
classDef gpt fill:#dcfce7,stroke:#16a34a,stroke-width:2.5px,color:#065f46;
classDef proxy fill:#0f172a,stroke:#0f172a,color:#ffffff;
classDef mech fill:#f1f5f9,stroke:#94a3b8,stroke-width:1.5px,color:#475569;
classDef goal fill:#ffffff,stroke:#334155,stroke-width:2.5px,color:#0f172a;
ChatGPT enriches it (green) — titration, resistance + aerobic training, a calorie deficit, the regain caveat. AI improving AI. More levers, but still all routed through the scale, and still implicit.
③ Causal Diagram: NoBSmed (explicit)
flowchart TB OSA["Sleep apnea
never screened"]:::miss TIME["Diet timing
‘frequent meals’ — wrong way"]:::miss IR["Insulin resistance
root — never measured"]:::miss STRONG{{"Stronger drug — tirzepatide —
vs the 25% goal, never weighed"}}:::miss RX{{"Oral semaglutide 25 mg
the first answer"}}:::oe APP["appetite ↓"]:::oe WT(("Weight → 200 lb
a proxy")):::proxy TITR{{"Titration 1.5→4→9→25 mg"}}:::gpt RES{{"Resistance training
preserves muscle (25–40%)"}}:::gpt AERO{{"Aerobic 150 min / week"}}:::gpt KCAL{{"Calorie deficit
1,500–1,800 kcal"}}:::gpt DUR{{"Durability: 50–67% regain flagged
— but no taper (plan: indefinite therapy)"}}:::gpt AMPH{{"Amphetamine timing"}}:::gpt EXP["Energy
expenditure ↑"]:::mech GOAL["Focus & energy
what he actually wants"]:::goal RX --> APP --> WT ==> GOAL TITR --> RX AMPH --> RX RES --> WT AERO --> EXP --> WT KCAL --> APP DUR --> WT OSA --> IR --> WT TIME --> IR STRONG --> APP IR -. glycemic fog .-> GOAL OSA -. fatigue .-> GOAL RES -. dopamine .-> GOAL AERO -. dopamine .-> GOAL classDef oe fill:#dbeafe,stroke:#2563eb,stroke-width:2.5px,color:#1e3a8a; classDef gpt fill:#dcfce7,stroke:#16a34a,stroke-width:2.5px,color:#065f46; classDef miss fill:#fee2e2,stroke:#dc2626,stroke-width:3px,color:#991b1b; classDef proxy fill:#0f172a,stroke:#0f172a,color:#ffffff; classDef mech fill:#f1f5f9,stroke:#94a3b8,stroke-width:1.5px,color:#475569; classDef goal fill:#ffffff,stroke:#334155,stroke-width:2.5px,color:#0f172a;
NoBSmed makes it explicit. Red is what both missed — the load-bearing one is the root, insulin resistance, sitting above every lever, never measured, fed by the sleep apnea nobody screened for. The dashed bypass paths reach focus & energy directly — without going through the scale at all. He could hit 200 lb and still be exhausted and unfocused. Weight is a proxy, not the goal.
🎯 Start here — the two that change the whole decision
"Can we measure my insulin resistance directly — fasting insulin and HOMA-IR — not just A1C?"
"And if it is the driver, should we be treating it, rather than only suppressing my appetite?"
The plan orders glucose, A1C and lipids. It never orders fasting insulin or HOMA-IR — the direct measures of insulin resistance. A1C is a late signal: it moves only once the pancreas is already losing the fight, often years in. So the plan's own labs cannot tell him whether the thing driving his weight is the thing he's treating.
This matters because it changes what works. In the landmark prevention trial, a lifestyle programme cut progression to diabetes nearly twice as effectively as a drug — and diet composition, not just calories, is what moves insulin resistance (glycemic load). You can't target a driver you never measured.
"If I stop this drug, what happens to my weight — and does that mean I'm on it for life?"
"Is there a way to come off it and hold the loss?"
The plan assumes indefinite therapy and never asks what happens if he stops. Novo Nordisk's own extension of the STEP 1 trial answers it: one year after withdrawal, participants had regained two-thirds of the weight they lost — from −17.3% down to −5.6% — and their cardiometabolic gains reverted too. The trial's own conclusion is that ongoing treatment is required. A 2025 systematic review says the same.
That reframes the entire decision — from "a course of treatment" to "a drug you may take for the rest of your life," with the cost and the amphetamine-stacking question that come with it. It is the single most important fact about this drug, and neither tool mentioned it. There's also a claim about the regain we're deliberately not making.
↳ There may be a way out. In a randomised trial, exercise plus the drug — but not the drug alone — preserved the weight loss after treatment stopped (Lundgren, NEJM 2021). That is the exit strategy the plan never weighs, and it's the reason Flag 3 matters so much.
Worth raising with your doctor
"How much of my weight loss will be muscle — and how do we protect it from week one?"
The plan lists resistance training under "activity" as a generic line-item. It never tells him why it's load-bearing. Across 20 randomised trials in 15,782 people, 35.2% of the weight lost on semaglutide was lean mass — muscle, not fat.
And this is not a drug-specific villain story — dieting alone loses muscle at essentially the same rate (26.2%; the difference from the drug isn't statistically significant). The one arm in the entire meta-analysis with a meaningfully better result was lifestyle plus resistance training, at 17.5%.
↳ Read that last number carefully — we're not going to overstate it. The 17.5% is lifestyle plus resistance training. The meta-analysis contains no drug-plus-resistance-training arm at all, so it cannot tell you what his number would be on semaglutide while lifting. Nobody has run that trial. What the evidence does establish is that resistance training is the only lever that moved the muscle share in any arm that was studied — which is why "lift from week one" belongs in the plan, and why we won't hand you a precise number for a combination nobody has tested.
"At my height and weight, should I be screened for obstructive sleep apnea — and could it be part of why I'm tired and unfocused?"
A 6′2″ man at BMI 34 is squarely in the risk group for obstructive sleep apnea. The plan says "sleep hygiene" and stops. But untreated OSA does three things at once that matter here: it worsens insulin resistance (Flag 1), it dysregulates appetite hormones, and it impairs attention and focus — which is to say, it can look exactly like the ADHD he's already medicating.
↳ This is the one to run down first. It's a home test. If it's positive, treating it moves the weight, the insulin resistance and the focus — the three things he actually came for.
"I take an amphetamine for ADHD — which already curbs my appetite and raises my heart rate. What happens when we stack a second appetite suppressant on top, and how do we monitor it?"
The plan did check for a drug interaction — and found none. But it checked the wrong kind. It looked at how the two drugs are metabolised (liver enzymes) and concluded "no interaction." That's true, and it misses the point, because the interaction here isn't metabolic — it's that both drugs do the same thing.
Two appetite suppressants stack. The risk is that he eats too little — which, per Flag 3, is exactly how you lose muscle instead of fat. And both drugs raise heart rate and blood pressure, so the cardiovascular effects stack too. The plan noticed this and used it as a tie-breaker between drugs, rather than as a reason to monitor.
This is the single feature that makes him not-the-average-patient — and it's the one the plan treated as a footnote.
Weighing the real options — what the plan collapsed into "the drug wins"
The plan compared two pills and picked one. But a pill isn't the only thing that moves weight. Here is the whole menu, scored 1–5 on the four things that actually matter (higher is better). We never add these up into one number — a single score would quietly pick the weights for you and hide the trade-off. The point is to see the trade-off.
| Option | Efficacy | Durability holds off-drug | Side effects | Convenience | |
|---|---|---|---|---|---|
| Oral semaglutide 25 mg ← the plan's pick | 3 | 2 | 3 | 2 | ~11–16% loss, short of his 25% goal. Daily, on an empty stomach. |
| Injectable tirzepatide | 5 | 2 | 3 | 3 | frontier >20% loss — the plan admits oral won't reach 200 lb, then recommends oral anyway. |
| Injectable semaglutide 2.4 mg | 4 | 2 | 3 | 3 | ~15%. Weekly shot. |
| Orforglipron | 2 | 2 | 2 | 4 | frontier the easy oral (no fasting ritual) the plan dismissed. |
| ⭐ Exercise + resistance training | 2 | 5 | 5 | 3 | frontier the only option that also treats his ADHD (same dopamine pathway) and protects the muscle from Flag 3 and improves insulin resistance from Flag 1. Durable. The plan lists it as a line-item, not a strategy. |
| Diet: time-restricted / low-glycemic | 2 | 3 | 5 | 3 | frontier targets the root directly. The plan's "small, frequent meals" is the opposite of this. |
| ⭐ Drug + exercise together | 4 | 4 | 3 | 2 | frontier best efficacy×durability blend — preserves muscle, blunts the regain. The combination the plan never considered. |
| Do nothing | 1 | 1 | 3 | 5 | ruled out beaten on every axis that matters. |
Appendix — every claim, classified (how the audit reasons)
For the curious: every claim in the plan, classified — this is the audit's underlying logic, summarised. The five flags above are what matters; this is the receipts.
Each claim gets a plain-English verdict: Supported · Needs a look (in the plan, but questionable) · Missing (not in the plan, but it should be).
| Claim | Type | In plan? | Verdict — why | Flag |
|---|---|---|---|---|
| Explicit claims (what the plan says) | ||||
| "Oral semaglutide, titrated 1.5→4→9→25 mg" | 💊 Rx | explicit | Supported — the correct escalation schedule for this product. ChatGPT added this and got it right; the doctor's tool omitted it. | — |
| "Resistance training 2–3×/week" | 💊 Rx | explicit | Supported — but listed as a chore, not as the thing that decides whether the loss is fat or muscle | 3 |
| "Small, frequent meals every 3–4 hours" | 💊 Rx | explicit | Needs a look — the opposite of the approach that targets insulin resistance; the plan contradicts itself here | 1 |
| "No interaction with his amphetamine" | 💊 Rx | explicit | Needs a look — true of metabolism, but both drugs suppress appetite and both raise heart rate. The stacking is the issue. | 5 |
| "Labs: glucose, A1C, lipids" | 🔬 Dx | explicit | Needs a look — stops one step short of the root; A1C is a late signal | 1 |
| Implicit claims (assumed, never stated) | ||||
| "Reaching 200 lb will give me my focus and energy back" | 🔬 Dx | implicit | Needs a look — weight is a proxy. Sleep apnea and insulin resistance hit focus directly, bypassing the scale entirely | 1, 4 |
| "The drug is the plan" | 💊 Rx | implicit | Needs a look — the drug suppresses appetite; nothing in the plan treats a cause | 1, 2 |
| "I'll take it, lose the weight, and be done" | 💊 Rx | implicit | Needs a look — the manufacturer's own trial: two-thirds regained within a year of stopping | 2 |
| Missing — not in the plan, but needed (the audit adds these) | ||||
| Measure insulin resistance — fasting insulin + HOMA-IR | 🔬 Dx | missing | Missing — the driver is never measured | 1 |
| An exit strategy — what happens when he stops | 💊 Rx | missing | Missing — the most consequential fact about the drug, absent from both plans | 2 |
| Screen for obstructive sleep apnea | 🔬 Dx | missing | Missing — hits weight, insulin resistance and focus at once | 4 |
| Review the rest of his medications | 💊 Rx | missing | Missing — some common drugs cause weight gain; the full list was never reviewed | — |
| Weigh a stronger drug against his stated goal | 💊 Rx | missing | Missing — the plan concedes oral tops out below his target, then recommends it anyway | — |
A real, de-identified audit shown as an example — research and education, not medical advice. Shared with the patient's permission; all identifying details removed. Want your own care plan audited? Email ops@nobsmed.com.