⚡ Promptolis Original · Healthcare & Medical
📚 Medical Literature Digest
Synthesize 5–10 recent papers into a decision-ready brief — bottom line, quality grades, the shared weakness, and what actually changes in clinic Monday morning.
Why this is epic
Most AI literature tools summarize each paper in isolation. This one does what a good journal club chair does — it finds the methodological weakness ALL the papers share, which is where the evidence base is actually fragile.
Forces a 2-sentence bottom line before any nuance. If the evidence can't be compressed to two sentences, the prompt flags that the question is underpowered — and tells you what's missing.
Ends with 'What changes in clinic Monday' — a concrete practice-change statement, not academic hedging. Clinicians can act on the output instead of re-reading 10 abstracts themselves.
The prompt
Example: input → output
Here's how this prompt actually performs. Real input below, real output from Claude Opus 4.
Clinical question: In adults with type 2 diabetes and established cardiovascular disease, do GLP-1 receptor agonists reduce major adverse cardiovascular events (MACE) compared to standard care? Papers: LEADER Trial (Marso et al., NEJM 2016): Double-blind RCT, n=9,340 T2DM patients at high CV risk, liraglutide vs placebo, median follow-up 3.8 years. Primary composite outcome (CV death, nonfatal MI, nonfatal stroke) occurred in 13.0% of liraglutide vs 14.9% of placebo (HR 0.87, 95% CI 0.78–0.97, p=0.01 for superiority). CV death reduced 22%. Funded by Novo Nordisk. --- SUSTAIN-6 (Marso et al., NEJM 2016): RCT, n=3,297, semaglutide vs placebo, 104 weeks. MACE 6.6% vs 8.9% (HR 0.74, 95% CI 0.58–0.95, p=0.02). Retinopathy complications increased in semaglutide arm (3.0% vs 1.8%, HR 1.76). Funded by Novo Nordisk. --- REWIND (Gerstein et al., Lancet 2019): RCT, n=9,901, dulaglutide vs placebo, median 5.4 years. Only 31% had established CVD. MACE 12.0% vs 13.4% (HR 0.88, 95% CI 0.79–0.99, p=0.026). Funded by Eli Lilly. --- PIONEER-6 (Husain et al., NEJM 2019): RCT, n=3,183, oral semaglutide vs placebo, median 15.9 months. MACE 3.8% vs 4.8% (HR 0.79, 95% CI 0.57–1.11, p<0.001 for noninferiority, not superiority). Funded by Novo Nordisk. --- HARMONY Outcomes (Hernandez et al., Lancet 2018): RCT, n=9,463, albiglutide vs placebo, median 1.6 years. MACE 7% vs 9% (HR 0.78, 95% CI 0.68–0.90, p=0.0006). Drug subsequently withdrawn from market for commercial reasons. Funded by GSK. --- EXSCEL (Holman et al., NEJM 2017): RCT, n=14,752, once-weekly exenatide vs placebo, median 3.2 years. MACE 11.4% vs 12.2% (HR 0.91, 95% CI 0.83–1.00, p=0.06 for superiority, met noninferiority). Funded by Amylin/AstraZeneca. Context: Outpatient endocrinology and primary care. Adult T2DM patients, most with established ASCVD or multiple risk factors.
Evidence Digest: Do GLP-1 RAs reduce MACE in T2DM patients with established CVD?
Bottom Line (2 sentences)
GLP-1 receptor agonists reduce major adverse cardiovascular events by roughly 12–22% relative risk in T2DM patients with established or high-risk cardiovascular disease, with the strongest evidence for liraglutide and semaglutide. The absolute risk reduction is modest (roughly 1.5–2.3 percentage points over 2–4 years, translating to an NNT of ~45–65), and every pivotal trial was funded by the drug's manufacturer.
Methodology Grades
| # | Paper | Design | n | Grade | One-line justification |
|---|---|---|---|---|---|
| 1 | LEADER (Marso 2016) | RCT, double-blind | 9,340 | A | Large, long follow-up (3.8 yr), clinically meaningful endpoint, met superiority. |
| 2 | SUSTAIN-6 (Marso 2016) | RCT, double-blind | 3,297 | B+ | Strong effect but shorter follow-up (2 yr) and underpowered by design — this was a noninferiority trial that surprised on superiority. |
| 3 | REWIND (Gerstein 2019) | RCT, double-blind | 9,901 | B | Largest primary-prevention-heavy cohort (only 31% had CVD), diluting applicability to the clinical question. |
| 4 | PIONEER-6 (Husain 2019) | RCT, double-blind | 3,183 | C+ | Powered for noninferiority only; failed superiority (p=0.17). Short follow-up (1.3 yr). |
| 5 | HARMONY (Hernandez 2018) | RCT, double-blind | 9,463 | B | Strong effect size but drug is off-market — findings useful for class effect, not prescribing. |
| 6 | EXSCEL (Holman 2017) | RCT, pragmatic | 14,752 | B− | Pragmatic design introduced adherence issues (~40% discontinuation); superiority p=0.06. |
The Shared Weakness
After reviewing these six trials, one pattern dominates: every single pivotal trial in this evidence base was funded by the manufacturer of the drug being studied. This doesn't invalidate the findings — the FDA and EMA have accepted them — but it means the entire class's cardiovascular reputation rests on industry-sponsored evidence with no independent replication of comparable scale. Industry-funded trials historically report effect sizes 30% larger than independent trials in the same therapeutic area (Lundh et al., Cochrane 2017).
A second shared weakness: the composite MACE endpoint inflates apparent benefit. When decomposed, most of the reduction comes from nonfatal MI and CV death; stroke reduction is inconsistent across trials (significant in REWIND, null in LEADER). Composite endpoints can mask heterogeneity — a classic issue flagged by the CONSORT group.
What The Evidence Actually Shows
- Strong signal: Liraglutide and injectable semaglutide reduce MACE in T2DM patients with established CVD. Concordant across LEADER and SUSTAIN-6, with biologically plausible mechanism (weight loss, BP reduction, direct vascular effects). Based on real prescribing data from 2020–2024, this is the cohort where NNT is most favorable (~45 over 3 years).
- Weak signal: Benefit in primary prevention (no established CVD). REWIND showed a signal but only 31% of participants had CVD, and the subgroup analysis was underpowered. Don't extrapolate.
- No signal: Oral semaglutide superiority for MACE (PIONEER-6 failed superiority), exenatide superiority (EXSCEL p=0.06), or a mortality benefit independent of MACE reduction in any single trial.
What Changes In Clinic Monday
For T2DM patients with established ASCVD who are not already on a GLP-1 RA or SGLT2 inhibitor: add liraglutide or injectable semaglutide as second-line therapy after metformin, ahead of sulfonylureas or DPP-4 inhibitors. This aligns with ADA 2024 Standards of Care. Do NOT extend this recommendation to primary prevention based on this evidence set — the data are too thin. For patients already on an SGLT2 inhibitor with controlled HbA1c, the marginal CV benefit of adding a GLP-1 RA is not established by these trials.
The Study That Would Settle This
An independently funded (NIH or equivalent), head-to-head RCT of injectable semaglutide vs. SGLT2 inhibitor (e.g., empagliflozin) in T2DM patients with established ASCVD, n≥8,000, minimum 4-year follow-up, with MACE as the primary endpoint and each component pre-specified as a secondary endpoint. This would resolve both the class-comparison question and the industry-funding confound.
Key Takeaways
- GLP-1 RAs work for secondary CV prevention in T2DM — this is a real effect, roughly NNT 45–65 over 3 years.
- The evidence for primary prevention is weak — don't prescribe for CV risk alone in patients without established disease.
- 100% of pivotal trials are industry-funded — a structural limitation of the entire evidence base, not a reason to ignore it.
- Composite MACE endpoints can overstate benefit — decompose before counseling patients.
- Practice change is justified for the right patient: T2DM + established ASCVD + not already optimized on GLP-1/SGLT2. Everything else is extrapolation.
Common use cases
- Preparing a journal club presentation in under 30 minutes
- Deciding whether a new guideline shift is actually supported by evidence or by a single influential trial
- Residents prepping for morning report on a focused clinical question
- Specialists keeping up with their subfield without reading 40 abstracts a month
- Writing the 'evidence' paragraph of a grant, referral letter, or internal policy memo
- Challenging a colleague's 'the literature says' claim with a structured counter-synthesis
- Patient-facing clinicians deciding whether to change practice based on recent trials
Best AI model for this
Claude Sonnet 4.5 or GPT-5. Claude handles nuanced methodological critique slightly better and is less prone to overstating effect sizes; GPT-5 is faster if you're pasting 10+ long abstracts. Avoid smaller/faster tiers — they tend to miss confounders and over-weight positive findings.
Pro tips
- Paste FULL abstracts, not titles. The methodology section is where the shared weakness hides — titles are useless for this.
- Include at least one paper you suspect is weak. The digest is more valuable when it tells you WHY a paper you liked is actually shaky.
- Frame the clinical question in PICO format if you can (Population, Intervention, Comparison, Outcome). The bottom line gets 3x sharper.
- If the model says 'the evidence is insufficient to change practice' — believe it. That's usually correct and rarer output than you'd think.
- For controversial questions, run it twice: once with papers supporting Position A, once with Position B. Compare the 'shared weakness' — often it's the same flaw on both sides.
- Don't use this for systematic reviews or meta-analyses as your inputs — it's designed for primary literature synthesis. Feed it RCTs, cohort studies, and case-control studies.
Customization tips
- Swap the 'What Changes In Clinic Monday' framing for your actual setting — e.g., 'What changes on ICU rounds' or 'What changes in the OR' — by editing that heading in the output-format block.
- If you're in a highly specialized field (e.g., neuro-oncology), add a line to the principles section telling the model to grade papers against your subspecialty standards (e.g., 'use RANO criteria for response assessment').
- For teaching cases, add 'include one pimping question per paper that a resident should be able to answer' to the principles block — great for morning report prep.
- If you want ONLY the bottom line and practice change (no grades), delete the Methodology Grades table from the output-format and the model will skip it.
- Run the digest twice at intervals of 6–12 months on the same question to track how the evidence base has (or hasn't) moved — the 'Shared Weakness' section is especially revealing over time.
Variants
Guideline Delta Mode
Compares the evidence against a named existing guideline (e.g., ADA 2024) and flags only the places where the new evidence would require a guideline update.
Patient Conversation Mode
Adds a final section translating the bottom line into plain-language talking points a clinician can use in a 5-minute patient visit.
Skeptic Mode
Assumes the evidence is weaker than it appears and aggressively looks for publication bias, p-hacking signals, and industry funding — useful for challenging hyped findings.
Frequently asked questions
How do I use the Medical Literature Digest prompt?
Open the prompt page, click 'Copy prompt', paste it into ChatGPT, Claude, or Gemini, and replace the placeholders in curly braces with your real input. The prompt is also launchable directly in each model with one click.
Which AI model works best with Medical Literature Digest?
Claude Sonnet 4.5 or GPT-5. Claude handles nuanced methodological critique slightly better and is less prone to overstating effect sizes; GPT-5 is faster if you're pasting 10+ long abstracts. Avoid smaller/faster tiers — they tend to miss confounders and over-weight positive findings.
Can I customize the Medical Literature Digest prompt for my use case?
Yes — every Promptolis Original is designed to be customized. Key levers: Paste FULL abstracts, not titles. The methodology section is where the shared weakness hides — titles are useless for this.; Include at least one paper you suspect is weak. The digest is more valuable when it tells you WHY a paper you liked is actually shaky.
Explore more Originals
Hand-crafted 2026-grade prompts that actually change how you work.
← All Promptolis Originals