Sacubitril/valsartan in adults with heart failure and reduced ejection fraction

Registration and administrative information (PROSPERO field set)

Protocol registration

Method: timestamped public git commit — repository https://github.com/mahmood726-cyber/rapidmeta-finerenone, path ssot/arni-hfref/arni-hfref.json.

Permalinks are pinned to the commit SHA, not to a branch. A branch link moves and would prove nothing.

Ordering test: did the protocol precede the search?

The protocol commit precedes the first executed query on both readings of 'first query'. The 12:19:18Z attempt is reported even though it returned nothing, because reporting only the successful execution would move the first-query time three minutes later and make the claim look better than it is. This is the test a PROSPERO record cannot offer: the timestamps are set by the repository and by the APIs, not by us, and both are independently checkable.

What this evidence establishes

That the text of this object existed, in this public repository, at the committer timestamp shown, and has not been altered since without producing a new commit. The permalink is pinned to the commit SHA rather than to a branch, so it resolves to that exact state forever; a branch link would move and would prove nothing. Anyone can verify the timestamp and the content independently of this page and independently of its authors.

What it does not

It does not establish prospectiveness. A commit proves when text entered this repository; it cannot prove that no earlier or parallel version existed elsewhere, that the data had not already been seen when the text was written, or that the analysis plan was fixed before the results were known. It also says nothing about the independence of the people who wrote it. Those are claims about conduct, and a timestamp cannot carry them.

Protocol amendment history

The full commit history is projected, not only its head. A log that displays only its current state is no better than a mutable document, and the point of registering by commit is that every earlier state stays readable. Each entry records whether it post-dates the first executed query, because an amendment made after the search is a different act from one made before it and must not be presented as the same.

Author attestation

These are the surfaces a human author discharges by checking them and recording that they did. An attestation records that someone checked what is already here; it never alters a number and never raises a cell's source tier. A slot naming no person, no source or no date reads as absent.

PubMed (ncbi e-utilities esearch)

Query as executed:

("sacubitril valsartan"[tiab] OR "LCZ696"[tiab] OR sacubitril[tiab] OR Entresto[tiab])
AND (enalapril[tiab] OR "angiotensin converting enzyme inhibitor"[tiab] OR ACEI[tiab])
AND ("heart failure"[MeSH Terms] OR "heart failure"[tiab] OR HFrEF[tiab])
AND (randomized controlled trial[pt] OR randomised[tiab] OR randomized[tiab] OR trial[tiab])

ClinicalTrials.gov API v2

Query as executed:

query.intr=sacubitril valsartan OR LCZ696
query.cond=heart failure
filter.overallStatus=COMPLETED|TERMINATED|ACTIVE_NOT_RECRUITING

How to re-run this search

Every string above is the string as executed, copied from the capture rather than from the protocol, so a reader can re-run it. Hit counts are the values the APIs returned (esearchresult.count and totalCount) and were not counted or computed here. Where the two differ from the registered strings, that would be recorded as a departure rather than silently reconciled.

Captured by: screener/searcher A (Claude, Anthropic family); a second cross-family screener runs separately. Source: evidence/2026-08-12/search-and-screening/01_SEARCH_CAPTURE.md.

Every record the search retrieved

Every record the registered search retrieved, listed individually with the decision, the stage at which it was taken, and for an exclusion the axis it failed and the quantity it reports instead. Exclusions are on the outcome axis as often as on quality, which is what the pre-registered estimand criterion produces: a large, well-conducted, on-topic trial is excluded when it reports something that cannot be converted into a time-to-first-event hazard ratio. Two records are undetermined, not excluded.

Extracted values, and where each came from

One row per extracted value. Every row carries the value, the verbatim sentence it was read from, a resolvable link to the source, and whether the number was read or derived. Where any of those is absent the row says so rather than omitting the value.

Risk of bias (RoB-2)

the model that assembled this object (anthropic family) did not assess any domain, per protocol section 9. The two assessors are from different families and neither is the assembler.

Variant: effect of assignment to intervention (intention-to-treat), as pre-registered in protocol.md section 9

Unit assessed: the result being pooled, not the trial overall: the composite of cardiovascular death or first hospitalisation for heart failure, as a time-to-first-event hazard ratio

Assessor 1: GPT-5 (openai family). Assessor 2: Gemini 3.1 Pro (google family). each assessor saw only its own trial packet; neither saw the other's answers

Overall judgement per trial

Inter-assessor agreement, as measured

Per-domain: 10 of 15 agreed (66.7%). Overall: 1 of 3.

This is markedly lower than the two-assessor agreement seen at screening, which is what we predicted before running it. Screening asks whether a record meets stated eligibility axes, which is close to a lookup. RoB-2 asks for a graded judgement under uncertainty, where 'no information' and 'some concerns' are reached by different routes. Almost no published review reports its RoB-2 inter-assessor agreement at all, so there is no external benchmark. We publish it because an unreported agreement rate is not the same as a high one.

Sources

Which outcomes were analysed

One outcome is the registered analysis. Three more are now pooled as secondary, and everything else the trials report cannot be pooled without changing the question.

The registered estimand is a single quantity: cardiovascular death or first heart-failure hospitalisation, whichever comes first, as a time-to-first-event hazard ratio. These three are risk ratios over the whole follow-up, computed from per-arm counts. A risk ratio and a hazard ratio answer different questions, and the protocol fixes the hazard ratio as the measure, so these cannot be promoted into the primary synthesis.

Two of the three are components of the composite. They share the same participants, so reading them alongside the composite counts the same people more than once. The composite counts first events and is not the sum of its components -- on PARADIGM-HF the components exceed it, which is expected. Nothing here is reconstructed by addition.

parallel-hf all-cause death is ClinicalTrials.gov only -- the Circ J primary paper does not report all-cause mortality at all, so that pair has no second source.

Considered and NOT pooled, with the reason

Before 2026-08-13 this object carried exactly one analysed outcome and the page did not say why. That was a real gap in what the page communicated, not merely in what it computed: a reader could not tell whether other outcomes had been considered and rejected or simply never looked at. The three secondary pools existed as per-arm counts in the object all along and had not been pooled.

Visual abstract

⬇ SVG (vector) 2 KB ⬇ PNG 4500x2100 295 KB

All raster formats were generated at build time from the same SVG as the graphic above (SHA-256 850741fa801ff002…), at 600 dpi for a 6-inch print width, on a white ground. They are not screenshots of the page and do not change with the theme you are reading in.

Submission formats written to the figures directory, not embedded here because an uncompressed 600 dpi TIFF is tens of megabytes: EPS (vector, journal's first choice for line art) — visual-abstract.eps, 4 KB; TIFF (uncompressed, 600 dpi -- journal fallback) — visual-abstract.tiff, 27,685 KB.

Projected from the canonical object, so it carries the same k, the same pooled estimate and the same interval as the paper and cannot drift from them. The interval is drawn CROSSING the no-difference line because it does: a graphical abstract travels without its caption, and one that showed a favourable point estimate without showing that its interval includes no effect would be overstating a null result, which is a defect class this review documents in other papers.

PRISMA flow of records

⬇ SVG (vector) 3 KB ⬇ PNG 3600x2910 294 KB

All raster formats were generated at build time from the same SVG as the graphic above (SHA-256 b81d315e593e83af…), at 600 dpi for a 6-inch print width, on a white ground. They are not screenshots of the page and do not change with the theme you are reading in.

Submission formats written to the figures directory, not embedded here because an uncompressed 600 dpi TIFF is tens of megabytes: EPS (vector, journal's first choice for line art) — prisma-flow.eps, 5 KB; TIFF (uncompressed, 600 dpi -- journal fallback) — prisma-flow.tiff, 30,691 KB.

Every stage carries a count. The identification tier is populated from search.databases, and this caption previously said the opposite -- that two boxes were drawn as NOT RECORDED because the counts had never been captured. They had been, all along; a diagram missing its top box reads as an oversight, one that states the gap reads as a decision. The identification tier is populated from search.databases -- each database's hit count as the API returned it, and how many were retrieved -- which sum to exactly the screened corpus, and the corpus's own per-source tally agrees independently. No search was re-run to fill it, so no record entered or left the pool. The two sources return disjoint record types (PMIDs and NCT numbers) and no de-duplication step is recorded, which is why records removed before screening is zero rather than unknown. Exclusion reasons across the whole corpus: randomisation 195, outcome 132, comparator 38, population 35, intervention 14.

Funnel plot

⬇ SVG (vector) 3 KB ⬇ PNG 3500x1700 238 KB

All raster formats were generated at build time from the same SVG as the graphic above (SHA-256 98e2ac572275a002…), at 600 dpi for a 6-inch print width, on a white ground. They are not screenshots of the page and do not change with the theme you are reading in.

Submission formats written to the figures directory, not embedded here because an uncompressed 600 dpi TIFF is tens of megabytes: EPS (vector, journal's first choice for line art) — funnel.eps, 6 KB; TIFF (uncompressed, 600 dpi -- journal fallback) — funnel.tiff, 17,431 KB.

Standard error against the effect, most precise at the top, with the 95% and 99% pseudo-confidence funnel drawn from the pooled estimate and contour bands around the null. At k = 4 a funnel CANNOT be read for asymmetry and none is claimed: it is shown because the positions are real and the emptiness is the finding.

Galbraith (radial) plot

⬇ SVG (vector) 2 KB ⬇ PNG 3500x1500 86 KB

All raster formats were generated at build time from the same SVG as the graphic above (SHA-256 3eaceb1a73fa986b…), at 600 dpi for a 6-inch print width, on a white ground. They are not screenshots of the page and do not change with the theme you are reading in.

Submission formats written to the figures directory, not embedded here because an uncompressed 600 dpi TIFF is tens of megabytes: EPS (vector, journal's first choice for line art) — galbraith.eps, 4 KB; TIFF (uncompressed, 600 dpi -- journal fallback) — galbraith.tiff, 15,381 KB.

Each trial's standardised effect against its precision.

Baujat plot

⬇ SVG (vector) 2 KB ⬇ PNG 3500x1500 104 KB

All raster formats were generated at build time from the same SVG as the graphic above (SHA-256 772bdfbf6f53ae2c…), at 600 dpi for a 6-inch print width, on a white ground. They are not screenshots of the page and do not change with the theme you are reading in.

Submission formats written to the figures directory, not embedded here because an uncompressed 600 dpi TIFF is tens of megabytes: EPS (vector, journal's first choice for line art) — baujat.eps, 4 KB; TIFF (uncompressed, 600 dpi -- journal fallback) — baujat.tiff, 15,381 KB.

Right means a trial drives heterogeneity; up means it moves the pooled result. Top-right is both.

Influence diagnostics

⬇ SVG (vector) 1 KB ⬇ PNG 3500x1500 89 KB

All raster formats were generated at build time from the same SVG as the graphic above (SHA-256 dab20113bc6246cd…), at 600 dpi for a 6-inch print width, on a white ground. They are not screenshots of the page and do not change with the theme you are reading in.

Submission formats written to the figures directory, not embedded here because an uncompressed 600 dpi TIFF is tens of megabytes: EPS (vector, journal's first choice for line art) — influence.eps, 4 KB; TIFF (uncompressed, 600 dpi -- journal fallback) — influence.tiff, 15,381 KB.

Leverage against Cook's distance, from metafor's influence diagnostics.

Leave-one-out

⬇ SVG (vector) 2 KB ⬇ PNG 3500x920 154 KB

All raster formats were generated at build time from the same SVG as the graphic above (SHA-256 ec71aab4847ab304…), at 600 dpi for a 6-inch print width, on a white ground. They are not screenshots of the page and do not change with the theme you are reading in.

Submission formats written to the figures directory, not embedded here because an uncompressed 600 dpi TIFF is tens of megabytes: EPS (vector, journal's first choice for line art) — leave-one-out.eps, 4 KB; TIFF (uncompressed, 600 dpi -- journal fallback) — leave-one-out.tiff, 9,433 KB.

The pool refitted with each trial removed in turn.

Cumulative meta-analysis

⬇ SVG (vector) 2 KB ⬇ PNG 3500x920 161 KB

All raster formats were generated at build time from the same SVG as the graphic above (SHA-256 b71542b3bcb40eaf…), at 600 dpi for a 6-inch print width, on a white ground. They are not screenshots of the page and do not change with the theme you are reading in.

Submission formats written to the figures directory, not embedded here because an uncompressed 600 dpi TIFF is tens of megabytes: EPS (vector, journal's first choice for line art) — cumulative.eps, 4 KB; TIFF (uncompressed, 600 dpi -- journal fallback) — cumulative.tiff, 9,433 KB.

The pool as each trial reported, in year order.

Bayesian posterior density

⬇ SVG (vector) 5 KB ⬇ PNG 3500x1500 93 KB

All raster formats were generated at build time from the same SVG as the graphic above (SHA-256 3a0dc30eeebd5489…), at 600 dpi for a 6-inch print width, on a white ground. They are not screenshots of the page and do not change with the theme you are reading in.

Submission formats written to the figures directory, not embedded here because an uncompressed 600 dpi TIFF is tens of megabytes: EPS (vector, journal's first choice for line art) — posterior.eps, 13 KB; TIFF (uncompressed, 600 dpi -- journal fallback) — posterior.tiff, 15,381 KB.

grid integration, 201x201, flat prior on the pooled log effect and half-Cauchy(scale=1) on the between-study SD (Gelman 2006)

Risk of bias, both assessors

⬇ SVG (vector) 7 KB ⬇ PNG 3130x1340 157 KB

All raster formats were generated at build time from the same SVG as the graphic above (SHA-256 7a977b52482caafd…), at 600 dpi for a 6-inch print width, on a white ground. They are not screenshots of the page and do not change with the theme you are reading in.

Submission formats written to the figures directory, not embedded here because an uncompressed 600 dpi TIFF is tens of megabytes: EPS (vector, journal's first choice for line art) — rob-traffic-light.eps, 21 KB; TIFF (uncompressed, 600 dpi -- journal fallback) — rob-traffic-light.tiff, 12,287 KB.

Every cell carries BOTH independent cross-family assessments and they are not reconciled: showing one column would be a reconciliation presented as an observation. Glyph as well as colour, so the panel survives greyscale printing and colour-blind reading. A trial with no judgement is shown as a row of not-assessed markers rather than omitted, so a gap in the assessment cannot be mistaken for a clean assessment. Agreement as measured: {'per domain agreed': 10, 'per domain total': 15, 'per domain rate pct': 66.7, 'overall agreed': 1, 'overall total': 3, 'measured not asserted': True, 'comparison to screening': "This is markedly lower than the two-assessor agreement seen at screening, which is what we predicted before running it. Screening asks whether a record meets stated eligibility axes, which is close to a lookup. RoB-2 asks for a graded judgement under uncertainty, where 'no information' and 'some concerns' are reached by different routes. Almost no published review reports its RoB-2 inter-assessor agreement at all, so there is no external benchmark. We publish it because an unreported agreement rate is not the same as a high one."}.

GOSH

⬇ SVG (vector) 1 KB ⬇ PNG 3500x540 81 KB

All raster formats were generated at build time from the same SVG as the graphic above (SHA-256 1ac0f91ef8ab0183…), at 600 dpi for a 6-inch print width, on a white ground. They are not screenshots of the page and do not change with the theme you are reading in.

Submission formats written to the figures directory, not embedded here because an uncompressed 600 dpi TIFF is tens of megabytes: EPS (vector, journal's first choice for line art) — gosh.eps, 2 KB; TIFF (uncompressed, 600 dpi -- journal fallback) — gosh.tiff, 5,537 KB.

Deliberately not drawn. Every subset meta-analysis of 4 trials is 15 points; a cloud that small cannot show the multimodality GOSH exists to reveal, and a reader would take the picture as evidence of its absence.

Trial-sequential analysis

⬇ SVG (vector) 1 KB ⬇ PNG 3500x650 87 KB

All raster formats were generated at build time from the same SVG as the graphic above (SHA-256 25ed9800fdf56614…), at 600 dpi for a 6-inch print width, on a white ground. They are not screenshots of the page and do not change with the theme you are reading in.

Submission formats written to the figures directory, not embedded here because an uncompressed 600 dpi TIFF is tens of megabytes: EPS (vector, journal's first choice for line art) — tsa.eps, 2 KB; TIFF (uncompressed, 600 dpi -- journal fallback) — tsa.tiff, 6,665 KB.

TSA boundaries depend entirely on a target information size that must be pre-specified. This review's protocol registers none, so any boundary drawn here would be a parameter chosen after seeing the data -- which is the practice TSA exists to protect against.

Meta-regression

⬇ SVG (vector) 1 KB ⬇ PNG 3500x540 87 KB

All raster formats were generated at build time from the same SVG as the graphic above (SHA-256 0a34e86921018df9…), at 600 dpi for a 6-inch print width, on a white ground. They are not screenshots of the page and do not change with the theme you are reading in.

Submission formats written to the figures directory, not embedded here because an uncompressed 600 dpi TIFF is tens of megabytes: EPS (vector, journal's first choice for line art) — bubble.eps, 2 KB; TIFF (uncompressed, 600 dpi -- journal fallback) — bubble.tiff, 5,537 KB.

The protocol pre-specifies no moderator, and at k = 4 a meta-regression would be fitted on 4 points. Not drawn rather than drawn with a caveat: a bubble plot invites reading a slope, and there is no slope here that any reader should read.

L'Abbe plot

⬇ SVG (vector) 2 KB ⬇ PNG 3500x1500 133 KB

All raster formats were generated at build time from the same SVG as the graphic above (SHA-256 a2a8af212f9a5a20…), at 600 dpi for a 6-inch print width, on a white ground. They are not screenshots of the page and do not change with the theme you are reading in.

Submission formats written to the figures directory, not embedded here because an uncompressed 600 dpi TIFF is tens of megabytes: EPS (vector, journal's first choice for line art) — labbe.eps, 4 KB; TIFF (uncompressed, 600 dpi -- journal fallback) — labbe.tiff, 15,381 KB.

Each trial's own two risks, read from its 2x2. Below the diagonal favours the intervention.

Number needed to treat, across baseline risk

⬇ SVG (vector) 3 KB ⬇ PNG 3500x1500 93 KB

All raster formats were generated at build time from the same SVG as the graphic above (SHA-256 b02217e8c096cf1c…), at 600 dpi for a 6-inch print width, on a white ground. They are not screenshots of the page and do not change with the theme you are reading in.

Submission formats written to the figures directory, not embedded here because an uncompressed 600 dpi TIFF is tens of megabytes: EPS (vector, journal's first choice for line art) — nnt-curve.eps, 7 KB; TIFF (uncompressed, 600 dpi -- journal fallback) — nnt-curve.tiff, 15,381 KB.

The number needed to treat depends on the risk a patient starts with. This applies the pooled risk ratio across a range of control risks so a reader can read off the value for the patient in front of them.

Summary of findings

1. Rated down from high for risk of bias, imprecision. The summary-of-findings record derives it as: start high; risk_of_bias serious (-1), imprecision serious (-1); total -2 -> low. The risk_of_bias rating was re-examined against the completed RoB-2 assessment on 2026-08-12 and did not move; see domains.risk_of_bias.rob2_effect_on_this_rating. Both stored locations rate this outcome Low, and their recorded grounds are not the same text. Both are stated above rather than one being chosen, because the two can differ in kind -- one recording what the trials were found to be, the other recording what could not be established about them.
2. Read this beside `whether the estimand was established`. That flag records that every contributing trial measures the same quantity. It records nothing about what that quantity was measured against -- whether the comparators are of one kind, whether each comparison was randomised, whether it was concurrent. The two questions were never separated. Named as P48 in PAGE-STANDARD.md v1.21.0 and swept by scripts/audit_mixed_contrast_pools.py.

Reproducibility artifact

Everything a third party needs to rebuild this page. Each figure on the Analysis tab downloads as an SVG carrying exactly the values shown.

Other recorded qualifications

Statements this review records about the limits of what it establishes, which are not shown elsewhere on this page. They are reproduced from the object verbatim and are not summarised.

Comparison with published syntheses

Every paper this project ships carries this section. A synthesis that reports only its own result asks a reader to take the field on trust; one that reconciles itself against the published syntheses on the same question, names what they got right as prominently as what they got wrong, and shows the denominator, can be checked. The classes applied below are the classes this review must itself survive. On the night this section was written, this object failed three of them: analysis panels computed at k=3 under a k=4 headline, a title stating a study count its own contents contradicted, and a figure labelled as a diagnostic it was not drawing.

13 checks were applied to this review and 9 came back clean. The denominator is stated because a list of only the failures is not a finding, it is a selection.

Confirmations are listed in the same table, in the same detail, as errors. Two checks were designed to catch a fault and instead cleared the paper: including a cohort study is inside their own stated eligibility, and omitting a small-study test at nine studies is what the Handbook advises.

The syntheses reconciled against

Where the numbers differ, and why

How they were identified: By lookup, never recall. The comparison review's identifiers were resolved through the PubMed ID converter on 2026-08-14: DOI 10.7759/cureus.48623 returns PMID 38084196 and pmcid PMC10710849. Its full text is staged in this bundle and every quotation below was read from that file.