Home › Blog › Cardiology › GDMT in HFrEF

Quick Answer

The evidence on what to give a patient with HFrEF is settled. Four drug classes, a renin-angiotensin inhibitor (preferably an ARNI), an evidence-based beta-blocker, an MRA and an SGLT2 inhibitor, each reduced death or heart failure hospitalisation in large randomised trials, and together they are estimated to add years of life. What is not settled by randomised evidence is the order in which to start them. What is settled, from STRONG-HF and from Indian registries, is that speed and completeness matter more than sequence: patients who reach full GDMT quickly do better, and most Indian patients never get there. The bottleneck in GDMT in HFrEF is not the decision. It is getting the evidence to the point of care fast enough to act on it.

Key Takeaways

• The 2026 ESC heart failure guideline, published on 28 August 2026, redefines HFrEF as LVEF below 50%, retires the HFmrEF category and replaces the term GDMT with "foundational medical therapy".
• Comprehensive four-class therapy was estimated to add 1.4 to 6.3 years of survival compared with conventional therapy, depending on age at initiation.
• Only one randomised trial, CIBIS III, has tested the order of two heart failure drugs; recommended sequences for all four classes rest on expert opinion and pharmacological reasoning.
• In STRONG-HF, rapid up-titration within two weeks of discharge reduced 180-day death or HF readmission from 23.3% to 15.2%.
• In Indian registries, only about one in four patients with HFrEF was discharged on full GDMT, and GDMT at discharge was associated with lower long-term mortality.

Why Is Heart Failure a Different Kind of AI Problem?

When heart failure comes up in clinical practice, the question is rarely "What should be prescribed?" Cardiologists know the four pillars. The harder question is closer to "How can all of this be done in a seven-minute OPD slot, for a patient whose blood pressure is 102, whose potassium is 5.1 and who is paying for every tablet himself?"

That distinction changes how AI in cardiology should be judged. A great deal of energy in medical AI goes into making models that can choose: pick the drug, pick the dose, pick the next step. For GDMT in HFrEF, choosing is not where doctors are stuck. The evidence already made the choice. Where doctors are stuck is time, titration and the dozen small patient-specific facts that make a clear guideline hard to apply on a Tuesday afternoon.

The argument of this article is simple. Doctors do not need AI to choose GDMT. They need the evidence faster, with its sources visible, and with an honest line drawn between what trials proved and what experts inferred. A companion article made a related point about what AI should do when guidelines disagree on DOAC dosing. Heart failure is the mirror case: here the guidelines largely agree, and the failure is in implementation.

What Changed in the 2026 ESC Heart Failure Guideline?

The classification, the vocabulary and several recommendations changed; the trial evidence behind the core drug classes did not.

The 2026 ESC Guidelines for the management of heart failure were published online in the European Heart Journal on 28 August 2026 and presented at ESC Congress in Munich. The most visible change is structural. The HFmrEF phenotype (LVEF 41% to 49%) has been removed, leaving two phenotypes: HFrEF, now defined as LVEF below 50%, and HFpEF, defined as LVEF of 50% or above. The task force explained that HFmrEF and HFrEF share similar pathophysiology and probably respond to similar treatments.

The vocabulary changed as well. The ESC task force noted that the term GDMT had become uncertain in meaning after a decade of new evidence, and introduced "foundational medical therapy" for treatments with a Class I recommendation for reducing HF hospitalisation or death in unselected patients, "additional medical therapy" for treatments with narrower or weaker recommendations, and "guideline-directed interventional therapy" for devices and procedures. Specific recommendations moved too, including a Class I recommendation for mineralocorticoid receptor antagonists in chronic HF independent of LVEF. The ESC also adopted the US staging approach from stage A (at risk) to stage D (advanced).

For an Indian cardiologist, two consequences follow. First, a patient with an LVEF of 45% who was labelled HFmrEF last month is now, under ESC terminology, a patient with HFrEF. Second, any tool trained before late August 2026 is describing a classification system that no longer exists in the ESC framework. The US framework remains the 2022 AHA/ACC/HFSA guideline, with its own LVEF categories, so the two major guideline families now classify the same patient differently. That is exactly the kind of divergence a clinician needs to see stated, not silently resolved.

What Does the Evidence Say About the Four Pillars?

Each class reduced hard outcomes in its own large trial, and the modelled benefit of combining all four is substantial.

The trial base for GDMT in HFrEF is among the strongest in cardiovascular medicine. The table below lists the landmark trials for each class, their primary results and the population studied. It is a summary of published evidence, not a prescribing guide.

Each trial tested its drug on top of the standard treatment of its time.

Because each class was tested against a different background, no single trial shows what all four achieve together. A 2020 cross-trial analysis in The Lancet estimated it. Compared with conventional therapy (an ACE inhibitor or ARB plus a beta-blocker), comprehensive therapy with an ARNI, beta-blocker, MRA and SGLT2 inhibitor was associated with a hazard ratio of 0.50 for cardiovascular death and 0.53 for all-cause mortality. It was estimated to add 2.7 years (for an 80-year-old) to 8.3 years (for a 55-year-old) free from cardiovascular death or first HF admission, and 1.4 to 6.3 years of overall survival.

These are modelled estimates, and critics have questioned whether cross-trial arithmetic of this kind can be taken at face value. The direction is not in doubt, however, and the size of the estimate explains why every major guideline treats all four classes as foundational rather than optional. The ESC notes that although mortality has improved considerably over three decades, fewer than 60% of patients diagnosed with heart failure are still alive five years later. The treatments exist. The outcome gap is largely a delivery gap.

Does the Order of GDMT Actually Matter?

There is almost no randomised evidence on sequence. The strongest evidence says to get all four classes started quickly, in whatever order the patient tolerates.

This is where the public conversation about GDMT sequencing needs more honesty. Many clinicians learned the "historical" order, ACE inhibitor first, then beta-blocker, then MRA, simply because that is the order in which the drugs were discovered and tested. The order was never proven superior. It was an accident of history.

Only one randomised trial has directly tested the order of heart failure drugs. CIBIS III randomised 1,010 patients to start with bisoprolol or with enalapril for six months before combining them, and found bisoprolol-first non-inferior for the combined outcome of death or hospitalisation. That is a trial of two drugs, conducted before ARNIs and SGLT2 inhibitors existed. No randomised trial has compared full four-class sequencing strategies.

In the absence of trial data, experts have proposed different approaches. McMurray and Packer argued for starting an SGLT2 inhibitor and a beta-blocker first, adding the ARNI and MRA within weeks, on the grounds that SGLT2 inhibitors act quickly and have little effect on blood pressure or potassium. Greene and colleagues argued for simultaneous or rapid-sequence initiation of all four classes, often at low doses, with up-titration afterwards. The 2024 ACC Expert Consensus Decision Pathway similarly emphasises rapid initiation over any fixed order.

These are reasonable positions from highly credible authors. They are also expert opinion. When an AI tool tells a doctor "start with the SGLT2 inhibitor," the doctor deserves to know whether that sentence rests on a randomised trial or on an argument in a viewpoint article. In this case, it is the latter.

What Does STRONG-HF Tell Us About Speed?

Rapid up-titration after a heart failure admission reduced death or readmission, with more adverse events but no increase in serious ones.

If sequence has weak evidence, speed has strong evidence. STRONG-HF randomised patients admitted with acute heart failure to usual care or to a high-intensity strategy: up-titration to 100% of recommended doses within two weeks of discharge, with four scheduled visits over two months monitoring clinical status, laboratory values and NT-proBNP.

The primary outcome, heart failure readmission or all-cause death at 180 days, occurred in 15.2% of the high-intensity group and 23.3% of the usual care group, an adjusted risk difference of 8.1% (risk ratio 0.66). By day 90, full doses had been reached for RAS blockers in 55% versus 2%, for beta-blockers in 49% versus 4%, and for MRAs in 84% versus 46%. Notably, STRONG-HF tested the three classes available at its design stage; SGLT2 inhibitors were not part of the titration protocol.

The trade-offs were real but manageable. Adverse events by day 90 were more common with high-intensity care (41% versus 29%), while serious adverse events (16% versus 17%) and fatal adverse events (5% versus 6%) were similar. A later analysis found eGFR changes at 90 days were similar between groups irrespective of heart failure aetiology. For the clinician worried that fast titration will damage the kidneys, that is a directly relevant data point, and it is exactly the kind of fact that should appear at the moment of decision.

Why Is the Gap So Large in Practice?

Across countries, most eligible patients do not receive all foundational therapies at target doses, and the gap in India is large and associated with higher mortality.

The implementation gap is not an Indian problem alone. In the US CHAMP-HF registry of stable outpatients with HFrEF, fewer than one in four eligible patients received an RAS inhibitor, beta-blocker and MRA together, and only 1% were on target doses of all three. Over 12 months of follow-up, most patients stayed on stable sub-target doses or no medication, with low rates of initiation or dose escalation.

The Indian data are sobering. The National Heart Failure Registry, covering 10,851 consecutive patients with acute decompensated heart failure across 53 tertiary hospitals in 21 states, found that 47.5% of eligible patients received GDMT, with 90-day mortality of 14.2%. The Indian College of Cardiology National Heart Failure Registry found that only 24.99% of patients were prescribed all guideline-based therapies at discharge. In the Trivandrum Heart Failure Registry, 25% of patients with HFrEF received GDMT at discharge, and the 5-year mortality across the cohort was 59%. The CSI-Kerala Acute Heart Failure Registry similarly reported poor adherence, with in-hospital and 90-day mortality of 7% and 11.6%, and found that prescription of GDMT components was independently associated with 90-day mortality.

The Indian heart failure patient is also a different patient from the one in most Western trials. In the National Heart Failure Registry the mean age was 59.9 years, ischaemic heart disease accounted for 72% of cases and dilated cardiomyopathy for 18%, and 90-day mortality was highest in the lowest educational class. In the ICC National Heart Failure Registry, 68.3% of patients had an LVEF below 40%, 30-day mortality was 12.35%, and only 23.72% of patients were still adhering to the full GDMT prescription at 30 days. Younger patients with more years of life at stake, a high ischaemic burden and a steep drop-off in adherence after discharge: this is the population in which the modelled survival gains of comprehensive therapy are largest, and in which the implementation gap costs the most.

The ten-year Trivandrum follow-up, published in August 2026, makes the stakes explicit. Overall mortality at ten years was 82.7%. Among patients with HFrEF, mortality was 75.9% for those prescribed GDMT at baseline and 88.6% for those who were not (HR 0.71; 95% CI 0.57 to 0.88). Registry associations are not randomised evidence, but they are consistent with every trial in the table above.

The barriers are familiar to every Indian cardiologist, and several have evidence attached to them:

Evidence points, not prescribing rules. Individual decisions depend on the full clinical picture and current prescribing information.

Cost deserves a specific note because it is the barrier Western guidance rarely addresses. A study in the Heart Failure Journal of India documented wide price variation between branded and generic heart failure drugs, with branded empagliflozin costing roughly 13 times more than generic alternatives at the time. That price gap has since narrowed sharply: following patent expiry in March 2025, empagliflozin prices reportedly fell from about ₹60 to about ₹5.5 per tablet. An evidence summary written for Indian practice that ignores what the patient pays is incomplete.

How Does General-Purpose AI Handle Heart Failure Questions?

Early evidence suggests general-purpose chatbots are unreliable on patient-specific HF management, with titration guidance a particular weakness.

A study evaluating ChatGPT-5 on 38 anonymised heart failure vignettes, graded by cardiologists against the 2023 ESC and 2022 ACC/AHA/HFSA recommendations, found 53% of responses fully concordant, 11% partially concordant, 21% discordant and 16% unsafe or harmful. Most inaccuracies involved vague drug titration guidance. An earlier single-centre study found ChatGPT's treatment suggestions agreed with a heart team's decisions in 58% of cases, with agreement falling in clinically complex cases.

Two details stand out. First, the weakness was titration, which is precisely where doctors most need help and where STRONG-HF provides clear evidence. Second, that study graded answers against the 2023 ESC focused update. Within weeks of publication of the 2026 ESC guideline, every model trained on earlier material is now answering questions about a heart failure classification the ESC has retired. Currency is not a cosmetic feature in cardiology. It changes the answer.

Better evidence is coming. TAILOR-HF, a trial at Mass General Brigham, is currently recruiting to test targeted language-model recommendations for heart failure. Until trials like it report, claims that any AI tool improves heart failure outcomes should be treated as unproven.

So What Should AI Do in Heart Failure?

Shorten the time between a clinical question and the evidence that answers it, keep the sources visible, and label expert opinion as opinion.

If the decision is already made by the evidence, what is left for AI to do? Quite a lot, as long as it resists the temptation to become the decision-maker. The job can be described in four parts:

1. Retrieve the current guideline, not the one the model remembers. In September 2026 that means the 2026 ESC heart failure guideline alongside the 2022 AHA/ACC/HFSA guideline and the 2024 ACC Expert Consensus Decision Pathway, each named and dated.
2. Match the evidence to the patient's blockers. A patient with a systolic pressure of 102 and a potassium of 5.1 does not need the four pillars recited. The clinician needs the specific trial data and guideline text that address low blood pressure, hyperkalaemia and renal function during titration.
3. Separate trial evidence from expert opinion. "Four classes reduce mortality" is trial evidence. "Start the SGLT2 inhibitor first" is expert opinion. An honest system labels each sentence accordingly.
4. Include the Indian context. Drug availability, generic pricing and Indian registry data belong in an Indian evidence summary, not in a footnote.

It is equally important to be clear about what AI should not do in heart failure. It should not present a single fixed titration schedule as if it were the guideline, because the guidelines themselves allow flexibility and the trials titrated under close supervision. It should not flatten the difference between an ESC Class I recommendation and a viewpoint article. It should not ignore the parameters that make titration unsafe for a particular patient on a particular day. And it should not claim outcome benefits that have not yet been tested; no randomised trial has yet shown that an AI tool improves heart failure outcomes, which is exactly why trials such as TAILOR-HF matter.

None of this requires an AI system to make clinical decisions. It requires the system to make the evidence arrive faster than the patient leaves the room.

What Does This Look Like in Practice?

Consider a 62-year-old man seen in a follow-up OPD one week after admission for decompensated heart failure. His LVEF is 35%. He is on furosemide, a low-dose beta-blocker and a low-dose ACE inhibitor. His systolic blood pressure is 104 mmHg, heart rate 88, potassium 5.0 mmol/L, and his eGFR has dipped from 55 to 49 mL/min/1.73 m². He is paying out of pocket.

The cardiologist knows the destination. What she needs in the next four minutes is the evidence for the route: what STRONG-HF showed about titrating within weeks of discharge in patients like this; what STRONG-HF reported about renal function during rapid titration; what the 2026 ESC guideline now says about foundational therapy and about MRAs; which statements about starting order rest on trials and which on expert opinion; and what the generic options cost in India today.

That is the problem OneRX, an AI clinical decision support platform for doctors, is designed around: returning a sourced evidence summary for the specific question at the point of care, with each statement linked to its trial or guideline and with expert opinion labelled as such. The decision on what to titrate, and when, stays with the cardiologist who has examined the patient.

How OneRX Supports Cardiologists in Heart Failure Care

OneRX began in oncology and is now extending the same evidence-summary approach to cardiology. In heart failure, where the problem is speed of implementation rather than uncertainty about the drugs, that approach is useful in five practical ways.

• Get a structured answer between patients. A cardiologist can ask a titration question in plain language, by text or voice, and receive a structured, referenced summary before the next patient walks in.
• See what each statement rests on. Every point in the answer is linked to its trial or guideline, so the doctor can tell at a glance whether a recommendation comes from a randomised trial such as STRONG-HF or from expert opinion.
• Compare therapies in one view. Options, contraindications and monitoring needs for the four drug classes are set out in tables, which suits the patient with low blood pressure, raised potassium or falling kidney function.
• Stay current. A clinical trials view helps find new and ongoing studies, which matters in a field where the ESC guideline changed its classification only weeks ago.
• Support the patient conversation. Doctors can generate patient education documents in PDF or Word from the answer, which can help with the self-care and adherence that heart failure depends on.

Throughout, OneRX summarises published evidence for the doctor to interpret. It does not diagnose, prescribe or change medical records. The cardiologist who has examined the patient makes the call.

What Should Clinicians Check Before Trusting an AI Answer on GDMT?

• Which guideline version does it cite? An answer describing HFmrEF as a current ESC category, or not mentioning the 2026 ESC guideline, predates the August 2026 update.
• Does it distinguish trial evidence from opinion? Statements about starting order should be labelled as expert consensus unless a trial is cited.
• Does it give specific titration information? Vague titration advice was the most common failure in the published ChatGPT-5 evaluation.
• Does it address the patient's actual barriers? Blood pressure, potassium, renal function and cost should change the answer. If they do not, the answer is generic.
• Are Indian data and prices considered? Evidence generated in US and European cohorts needs Indian registry context to be applied well.

Frequently Asked Clinical Questions

What are the four pillars of GDMT in HFrEF?

The four foundational drug classes are a renin-angiotensin inhibitor (an ARNI in preference to an ACE inhibitor or ARB where appropriate), an evidence-based beta-blocker, a mineralocorticoid receptor antagonist and an SGLT2 inhibitor. Each reduced death or heart failure hospitalisation in large randomised trials. The 2026 ESC guideline now calls these therapies foundational medical therapy.

Is there a proven best order for starting GDMT in HFrEF?

No. Only one randomised trial, CIBIS III, has tested the order of two heart failure drugs, finding bisoprolol-first non-inferior to enalapril-first. Proposed sequences for all four classes, such as starting with an SGLT2 inhibitor and beta-blocker, are expert opinion. The strongest evidence supports starting all classes quickly.

How fast should GDMT be up-titrated after a heart failure admission?

In STRONG-HF, up-titration to full recommended doses within two weeks of discharge, with close follow-up including NT-proBNP monitoring, reduced 180-day death or HF readmission from 23.3% to 15.2% compared with usual care. Adverse events were more frequent, but serious and fatal adverse events were similar.

Does the 2026 ESC guideline change who qualifies for HFrEF therapy?

Yes. The 2026 ESC guideline defines HFrEF as LVEF below 50% and removes the HFmrEF category, so patients with LVEF 41% to 49% are now classified as HFrEF under ESC terminology. The 2022 AHA/ACC/HFSA guideline retains its own LVEF categories.

Can AI tools safely recommend GDMT titration?

Current evidence does not support relying on general-purpose chatbots for titration decisions. In a published evaluation of ChatGPT-5 on 38 heart failure vignettes, 16% of responses were graded unsafe or harmful, and vague titration guidance was the most common error. Randomised trials of AI-supported heart failure care, such as TAILOR-HF, are ongoing. AI is best used to retrieve and summarise sourced evidence, with the titration decision remaining with the treating clinician.

How common is GDMT use in Indian heart failure patients?

Indian registries report low uptake. In the Trivandrum Heart Failure Registry, 25% of patients with HFrEF received GDMT at discharge, and the ICC National Heart Failure Registry reported 24.99%. In the ten-year Trivandrum follow-up, patients with HFrEF prescribed GDMT had lower mortality (HR 0.71).

Conclusion

Heart failure with reduced ejection fraction is one of the rare areas of medicine where the evidence is overwhelming and the practice gap is enormous. The four foundational drug classes are proven. Rapid titration is proven. The best starting order is not, and should not be presented as if it were. In India, roughly three in four patients with HFrEF in the major registries did not leave hospital on full GDMT, and the long-term mortality difference is visible in a decade of Trivandrum data.

That is why the most useful thing AI can do for a cardiologist is not to choose. The evidence has already chosen. The useful thing is to bring that evidence to the bedside faster, with its sources visible and its uncertainties labelled, so that the doctor can spend the seven minutes on the patient rather than on the search. As with DOAC dosing, the standard is the same: show the sources, show what is proven and what is opinion, and leave the decision with the doctor.