Wang et al · Lancet 2025 · 484 trials · 104,176 participants
Every antihypertensive in the published efficacy model, at every dose you would actually prescribe, ranked side by side with its monthly cost and the adverse effects worth mentioning — plus a combination builder that applies the paper's own permutation equations. Values are placebo-corrected clinic BP reductions over 4–26 weeks.
| # | Drug↕ | Class↕ | Subclass↕ | Freq↕ | Dose↕ | Δ SBP↕ | Δ DBP↕ | Onset↕ | Peak↕ | Full effect↕ | Wears off↕ | Stopping↕ | $ / mo↕ | Trials↕ | Difference↕ | Contraindications↕ | Indications↕ |
|---|
| # | Regimen↕ | Classes↕ | Δ SBP↕ | Δ DBP↕ | Cost / mo↕ | Pills↕ | Contraindications↕ | Indications↕ | Weakest↕ |
|---|
Goal is <130/80 mmHg for most adults (2025 ACC/AHA). Set your own target if this patient's differs.
The guidance above tells you what to reach for. This ranks the actual options by how far each one moves the pressure. Enter what the patient's BP is on the regimen above — every option is scored from that number using the same sequential step the model uses when adding a drug to an already-treated patient.
| Next step | Projected BP | Change | Added cost | Pills | At goal | Notes |
|---|
Each drug's real dose ladder, with the model's estimate of what every step is worth from the current baseline. Click a rung to move the regimen there. Reassess 2–4 weeks after any change — the trials behind the model reached full effect by about 4 weeks — and remember the flat rule of thumb: a doubling buys roughly 1.5 mmHg, a new class roughly 6.
Common effects first, then the ones that matter even though they are uncommon. Sources are listed under How it works.
Combination beats escalation. Doubling one drug buys ~ 1.5 mmHg; adding a second from a different class buys roughly six. The 2025 guideline recommends starting w/ two drugs in stage 2 HTN and in higher-risk stage 1.
Pair agents that block different systems. Diuretics and calcium blockers both activate the renin–angiotensin system, which is exactly why adding an ACE inhibitor or ARB on top of either is more than additive — the second drug blocks the escape route the first one opened. Two drugs from the same system fight over one mechanism and add adverse effects without adding much effect.
In ACCOMPLISHtrial (11,506 high-risk pts), benazepril + amlodipine cut cardiovascular events 20% vs benazepril + hydrochlorothiazide, at the same BP. Where both are reasonable, the calcium blocker is the better-evidenced partner.
| Situation | Reach for | Why |
|---|---|---|
| CKD with albuminuria | ACEi or ARB, never both | Slows progression; the RAS blocker is the mandatory component, not the add-on |
| Diabetes with albuminuria | ACEi or ARB | Same rationale; pair with a CCB or thiazide |
| Coronary disease or prior MI | Beta blocker + ACEi/ARB | The one setting where a beta blocker is genuinely first-line |
| HFrEF | Guideline-directed therapy, not a BP regimen | ACEi/ARB/ARNI + beta blocker + MRA + SGLT2 inhibitor; avoid non-DHP CCBs |
| Black adults without CKD or HF | Start with a CCB or thiazide | RAS blockers are less effective as monotherapy; the gap closes once combined |
| Resistant at three drugs | Add spironolactone | PATHWAY-2trial: beat both bisoprolol and doxazosin as the fourth agent |
| Gout | Losartan; avoid thiazides and loops | Losartan lowers urate; the diuretics raise it |
| Osteoporosis | Thiazide | Reduces urinary calcium excretion |
| Migraine or essential tremor | Propranolol | Treats both problems with one drug |
| Asthma or reactive airways | Avoid non-selective beta blockers | Propranolol and carvedilol are contraindicated; use the lowest dose if a beta-1 selective agent is unavoidable |
| Pregnancy or planning one | Stop ACEi, ARB, MRA, atenolol | Fetal toxicity. Labetalol and nifedipine are the preferred agents |
One pill or two? Rows marked single pill are available as a fixed-dose combination in the US, which improves adherence. Nineteen of the pairs here exist as one tablet, plus one triple (amlodipine + valsartan + hydrochlorothiazide) — most of them something plus hydrochlorothiazide.
Shaded rows are the default, including the two where someone else wants the number lower. Open a row for the evidence. Drug names link to the agent table.
| Condition | Target | Preferred agents | Avoid |
|---|
No mmHg column here. The efficacy model behind the rest of this atlas is built from chronic oral monotherapy trials and has no coefficient for a titrated infusion, so the columns that matter inpatient are different ones: how fast it starts, how fast it stops, and how finely you can steer it.
| # | Agent↕ | Use↕ | Route↕ | Class↕ | Subclass↕ | Setting↕ | Onset↕ | Offset↕ | Control↕ | Indications↕ | Avoid in↕ | Watch |
|---|
Every figure traces to Wang N, Salam A, Pant R, et al. "Blood pressure-lowering efficacy of antihypertensive drugs and their combinations." Lancet 2025;406:915–25 — a meta-analysis of 484 randomised, double-blind, placebo-controlled trials in 104,176 participants, mean baseline 154/100 mmHg, mean follow-up 8.6 weeks.
The paper's public calculator at bpmodel.org answers one regimen at a time. This page holds the whole grid at once: each of the 53 drugs was queried at half, standard, and double standard dose, and the model's own baseline-BP sweep was recorded for each — 159 drug–dose cells in all.
From the paper's supplementary appendix (p. 14), efficacy for drug n at the reference baseline is log-linear in dose:
and is then shifted linearly to the patient's actual baseline γ:
Across all monotherapies the average is 8.7 mmHg systolic at standard dose, with each doubling of dose buying only about 1.5 mmHg more — the flattest part of the whole model, and the reason the paper argues for combining drugs rather than escalating one.
Combinations are not simply additive. Drugs are applied in sequence, each one seeing a baseline already lowered by those before it, so each successive agent earns less. Because that makes the answer depend on the order, the model averages over every ordering:
Per-drug α, β and ε were recovered by least squares from the calculator's own outputs, then the permutation model was re-implemented here and checked against the live calculator twice: an initial 44-query run over in-range doses, and a later randomised audit of 40 more queries that deliberately included doses outside the fitted range and baselines from 130 to 180.
Read that as: inside the half-to-double standard dose range the model was fitted on, this page reproduces the source calculator to within rounding. Push a dose outside that range and the error roughly doubles, with a worst observed case of 1.8 mmHg — still small against the width of the prediction intervals, but no longer negligible. Ninety per cent of all audited queries landed within 1 mmHg of the calculator, and there is no systematic bias in either direction.
One honest limitation. The source calculator refuses to compute doses far above the trial standard dose — every query at ramipril 20 mg, captopril 300 mg, fosinopril 80 mg and nebivolol 40 mg came back empty. Those four rows are real prescribable doses and are shown, marked beyond model range, but their numbers are a pure extrapolation of the published equation that cannot be checked against the original. Treat them as a rough guide to magnitude, not as a validated estimate.
Three layers sit on top of the efficacy model, each from its own source.
Doses. Every strength shown is one that actually exists on a US shelf, so the ladder reflects prescribing steps rather than arithmetic multiples of the trial "standard dose". Where a real dose falls outside the half-to-double range the model was fitted on, the row is marked extrapolated.
Onset, full effect and wear-off. The distinction the columns keep separate is the one the labels themselves keep separate: how fast a single dose acts, and how long a fixed dose needs before its effect stops growing. They are not the same question, and for some agents they differ by a factor of several hundred — lisinopril moves blood pressure inside an hour but takes a fortnight to finish, captopril inside half an hour but several weeks to finish, and hydrochlorothiazide's brisk first-day diuresis is not its antihypertensive effect at all. Among the drugs in US practice, 46% of the individual figures are stated in that drug's own label. The rest are inferred from its labelled pharmacokinetics, or carried from the class for the agents in the efficacy model with no US label, and both are marked as such on the page. Offset is the weakest of the four: only irbesartan, telmisartan and eplerenone have a published time course for blood pressure returning after withdrawal, so the other offsets are four to five effective half-lives — when the drug has gone, which is earlier than when the blood pressure has finished drifting back.
Cost. This is an acquisition-cost benchmark, not what a patient pays: it ignores insurance, dispensing fees, and discount-card pricing. It is the right tool for "is this drug pennies or is it real money", which for these agents is usually the only question — most are under $5 a month, and the handful above $100 are worth knowing about. Two deliberate choices make the figures slightly conservative: the price is the median across manufacturers rather than the cheapest one, and the regimen is built from whole tablets rather than splitting a larger strength, which is usually cheaper but is not always what gets dispensed. Where a drug has both an immediate- and an extended-release form, the one priced is the form used for hypertension — so metoprolol is costed as succinate ER once daily, and the twice-daily tartrate is roughly a third of that.
Adverse effects, split two ways. Effects shared by an entire class are stated once, in the Class effects panel above the table — the cough and hyperkalaemia that every ACE inhibitor carries, the oedema every dihydropyridine carries. The table column holds only what is different about that drug within its own class: an effect unique to it, one clearly more frequent than in its classmates, or a dosing, interaction, or evidence difference that would make you pick it or avoid it. Repeating the class effects on every row made the column useless for choosing between, say, lisinopril and ramipril.
The within-class entries were researched drug by drug against FDA labels on DailyMed and named trials on PubMed, then put through an adversarial check that tried to refute each one on four grounds: factually wrong, secretly a class effect, not clinically meaningful, or unsupported by the cited source. 47 of 124 candidate claims were rejected — among them a lisinopril ALLHAT claim with a false attribution, and a “lacks losartan's urate effect” line for irbesartan that was really a losartan property written as somebody else's negative. The 77 that survived are linked to the source that supports them; follow the dotted underline.
Where a drug shows “nothing beyond the class effects”, that is a finding rather than a gap: it is a workhorse that behaves like its classmates. After the research pass exactly one drug in routine US use still reads that way — benazepril. A few entries deliberately correct common teaching: the ACE inhibitor cough figure of ~11% is enalapril's, not a nine-drug pooled value; urate-lowering is not unique to losartan, since the telmisartan/HCTZ label claims it too, but is not a class effect either, since irbesartan explicitly has none; beta blockers carry the abrupt-withdrawal warning as an ordinary warning, not a boxed one; and eplerenone's reputation for less gynecomastia than spironolactone is unsupported by its own label, resting instead on a single high-dose head-to-head in which it was also significantly less effective.
The universe here is the Wang trial set, not the US formulary. These 53 drugs are the ones the meta-analysis had placebo-controlled data for. Several agents you will reach for often are simply absent because the model has no coefficient for them — most notably labetalol (the default in pregnancy and in hypertensive urgency), triamterene/hydrochlorothiazide, torsemide, bumetanide, and metolazone. Their absence says nothing about how useful they are.
Of the 53, fourteen are marked non-US — never marketed in the United States or since withdrawn: barnidipine, lercanidipine, manidipine, nitrendipine, cilazapril, delapril, spirapril, fimasartan, bendroflumethiazide (listed here under its older name bendrofluazide), cyclopenthiazide, celiprolol, oxprenolol, eprosartan, and penbutolol.
Two more are US-marketed but hidden by default because both lost their once-daily formulation, leaving twice- and three-times-daily dosing that makes them impractical for chronic hypertension: isradipine and nicardipine. Quinapril is included but carries a supply warning — it has been in an FDA-listed shortage since 2023 with effectively one supplier, and has no entry at all in the current CMS pricing file, which is why its cost column is empty. It is a focusing convenience, not a formulary check — confirm against your own.
Primary: Wang et al, Lancet 2025;406:915–25 (supplementary appendix, open access). Calculator: bpmodel.org. Risk conversion: BPLTTC, Lancet 2021.