The published 5-year ITT remission rate is 14.2% — but its denominator is 169, not 262

The question

"What is the actual published intent-to-treat (not completer) 5-year full-remission rate in the Virta DRCP 2024 extension study? Pull the primary paper's ITT figure to replace the ~10% estimate and confirm the LLY substitution-case calibration."

Follow-up #1 from [[2026-05-25-virta-health-5yr-cohort-reversal-durability]], where the 5-year full-remission ITT rate was estimated at ~10% by multiplying the completer rate by the retention rate. This brief replaces that estimate with the figure printed in the paper. Primary source obtained in this run: McKenzie, Athinarayanan, Van Tieghem, Volk, Roberts, Adams, Volek, Phinney, Hallberg, Diabetes Res Clin Pract 2024;217:111898, full text read via the publisher's open-access (CC-BY) HTML at diabetesresearchclinicalpractice.com. Not a press release, not an abstract-only read — Table 1, the statistical-analysis section, and the discussion were all read directly.

What we already know (from the vault)

• [[2026-05-25-virta-health-5yr-cohort-reversal-durability]] established the completer figures (20% remission, 32.5% reversal, ~47% retention) and correctly flagged that Virta's press surface leads with completer numbers. It then derived an ITT estimate of "~10%" as 20% × 47%, explicitly labelled as an estimate needing replacement.
• [[2026-05-21-lilly-glp1-longevity-thesis]] is the live position this calibrates. The bear vector under test: behavioral reversal programs durably substitute for GLP-1s and erode LLY recurring revenue.
• [[2026-05-24-virta-health-payer-contracts-lly-thesis-input]] and [[2026-07-18-capital-rx-reverse-lilly-relationship]] already resolved the adjacent payer question toward complement-with-soft-substitution, not wholesale substitution.
• [[2026-05-22-tim-ferriss-sami-inkinen-virta-t2d-rowing]] is the Inkinen anchor whose "sustained" framing prompted the durability question in the first place.

What the web says

All figures below are read directly from the primary paper's full text (source tier: primary, peer-reviewed, open access) unless marked otherwise.

• The paper does report a true ITT analysis, and it is labelled. Methods, verbatim: "These tests were performed among 5-year completers with data and as intent-to-treat (ITT) analysis where missing data was treated as a failed target or worst outcome carried forward (ITT WOCF)." Separately, for continuous outcomes: "The primary analysis was an intent-to-treat analysis, which included all participants who consented for the extension (n = 169) using the maximum likelihood approach."
• The headline answer — 14.2%. Results, verbatim: "Twenty percent of five-year completers with data (n = 24, 14.2 % ITT WOCF; p < 0.001; Table 1) met international consensus criteria for diabetes remission." Table 1 confirms: Remission at 5 years, ITT column, 24 (14.2) against n = 169.
• The looser "reversal" endpoint — 23.1% ITT. "Reversal to the point of HbA1c < 6.5 % without medication or only metformin was observed in 32.5 % of five-year completers (n = 39, 23.1 % ITT WOCF, p < 0.001)." Important and easy to miss: this endpoint was already met by 14 participants (8.3% ITT) at baseline, so the net gain is ~14.8 percentage points, not 23.1.
• Stricter endpoint — 4.7% ITT. Table 1 row "Complete Remission: HbA1c < 5.7 %, no glucose-lowering therapy ≥ 3 months": 5 years = 8 (4.7) ITT WOCF.
• Definition (definition-sensitivity matters as much as the number). Table 1 note, verbatim: "International Remission Consensus Definition 2021, HbA1c < 6.5 % that occurs spontaneously or following an intervention and persists for ≥ 3 months in the absence of usual glucose-lowering pharmacotherapy." Note the paper's own caveat that metformin was not deprescribed in response to improved glycemia — which is why the separate "no meds or metformin only" endpoint exists and why remission here is arguably understated relative to a trial that deprescribed metformin.
• The denominator is the whole story. ITT here = the 169 who consented to the three-year extension, not the 262 originally enrolled in the CCI arm. Methods, verbatim: "After 2 years, 194 of the 262 CCI participants were approached for a three-year extension. Of these, 169 consented, and 122 remained in the study for five years." Table 1 footnote: "For ITT WOCF, the number of participants with data at each time point differs: baseline, n = 169; 1 year, n = 165…; 5 years, n = 120 (2 missing HbA1c and 47 dropped out before 5 years)." So the completer denominator is 120, not 122.
• Decay is visible inside the ITT column. Same Table 1 remission row, ITT WOCF: baseline 0 (0.0) → 1 year 43 (25.4) → 2 years 41 (24.3) → 5 years 24 (14.2). That is a 41.6% relative decline in ITT remission from year 2 to year 5 within the same cohort. Sustained remission is lower still: 19 completers (11.2% of 169) held it 3 years, 15 (8.9% of 169) held it 4 years.
• Selection signal, stated by the authors. "Demographics and baseline and two-year clinical characteristics were not significantly different between those who consented versus declined participation in the extension, except those who consented had significantly greater BHB concentration and lower fasting insulin level at two years." Extension consenters were mildly enriched for adherent responders.
• Weight loss at 5 years is already an ITT number. The −7.6% body-mass figure comes from the linear mixed-effects model on n = 169 with MLE/MAR — it is the ITT primary analysis, not a completer average. (The parent brief reported −7.6% without noting this; it is more favourable than it was credited for.)
• Medication reduction is completer-only. "Among completers only, we assessed changes in the proportion of participants prescribed diabetes and lipid-lowering medications." Any diabetes med 85.2% → 71.3%; non-metformin 55.7% → 32.8%; insulin 26.2% → 13.1%; sulfonylureas 27.0% → 4.9%. The parent brief's "~50% prescription reduction" is a fair read of the non-metformin line (−41% relative) but should be labelled completer, so it is optimistic.
• Comparator context from the paper's own discussion: Look AHEAD 7.3% at 4 years, DiRECT 13% at 5 years, Esposito Mediterranean 5% at 6 years. Virta's 20% completer / 14.2% ITT sits at or above that band — though the comparators' denominators are not normalized here.

Convergences and contradictions

• The parent brief's ~10% was accidentally right, for a denominator the paper never uses. 24/262 = 9.2% (my arithmetic: 24 remitters ÷ 262 originally-enrolled CCI participants). That is within a point of the ~10% estimate. But the published ITT is 14.2%, because the paper's ITT population is the 169 extension consenters. Both numbers are defensible; they answer different questions. 14.2% = "of people who opted into year-3-to-5 continuation, how many were in remission at year 5." 9.2% = "of everyone who started the CCI, how many were in remission at year 5." The 93 people between those denominators (68 never approached, 25 declined) are unaccounted for in the published ITT.
• The paper's discussion overstates its own ITT. Verbatim: "Remission was observed in 20 % of those enrolled at five years… Reversal… occurred in 32.5 % of those enrolled five years." "Those enrolled" is wrong for both — 20% and 32.5% are completer-with-data rates on n = 120. The correctly-labelled ITT numbers (14.2% / 23.1%) are in Table 1 and the results paragraph, so the paper is not hiding them, but the discussion language is the sentence a press release would lift. This is exactly the completer-laundering the parent brief predicted, and it happens inside the peer-reviewed paper, one layer earlier than we assumed.
• Convergence on direction, correction on magnitude. The parent brief's verdict — durability is real but lands below the 30% bar for full remission on an honest denominator — is confirmed and is now sourced rather than estimated. The magnitude is 40-50% higher than we had it if you accept the paper's denominator.

Synthesis for RDCO

The number to carry forward in the LLY-longevity-v1 file is 14.2% (n = 24/169), 5-year full remission, ITT worst-observation-carried-forward, 2021 international consensus definition — cited to McKenzie et al., Diabetes Res Clin Pract 2024;217:111898, Table 1. That replaces "~10% estimated." Carry a second, clearly-labelled figure alongside it: 9.2% (24/262) if you anchor to everyone who ever started the CCI — my own arithmetic, not printed in the paper. Use 14.2% when comparing against other trials' ITT (Look AHEAD, DiRECT), because those comparators are also anchored to their randomized populations rather than to a general population. Use 9.2% when modelling what a payer would actually observe across an unselected book of business, because a payer does not get to re-consent its worst responders out of the denominator.

Does this change the substitution-case calibration? Directionally no, magnitude slightly yes — and the correction is mildly against the bear case, in the same direction the parent brief already went. 14.2% durable drug-free remission at 5 years is a real clinical outcome and beats every lifestyle comparator the paper cites, but it is still a minority outcome, and the year-2-to-year-5 decay inside a single cohort (24.3% → 14.2% ITT, a 41.6% relative fall) is the load-bearing fact for an investor: behavioral remission is not a stable state, it is a decaying one. A product whose effect decays by ~40% over three years does not displace a chronic-use pharmaceutical; it delays and thins the prescription. That is a margin threat to LLY, not a volume-replacement threat. The parent brief's "trim the behavioral-substitution weight, keep a small payer-triage weight" verdict stands, and this evidence does not justify re-inflating it.

Two calibration cautions worth writing into the thesis rather than leaving in a research note. First, the reversal endpoint is contaminated by baseline: 8.3% of the ITT population already met "HbA1c <6.5% on no meds or metformin only" at baseline, so quoting 23.1% as an achievement double-counts. The honest incremental figure is ~14.8 percentage points. Second, the medication-reduction numbers — the ones that actually drive the payer-economics arm of the bear case — are completer-only in this paper. There is no published ITT medication-reduction figure. Any payer-savings model built on "non-metformin prescriptions fell from 55.7% to 32.8%" is implicitly assuming the 47 dropouts behaved like the 120 completers, which the WOCF convention explicitly assumes they did not. If the payer-savings arm is doing real work in the thesis, that gap is the next thing to close, and it may not be closable from published data.

The meta-lesson is worth keeping. The parent brief flagged press materials as the laundering surface. The actual laundering happened one layer earlier, in the discussion section of the peer-reviewed paper, where "20% of those enrolled" is written for a number that is 20% of completers-with-data. The correctly-labelled figure was three paragraphs away in the same document. That is a reusable pattern for RDCO's research gates: for any efficacy percentage, the tables are load-bearing and the discussion prose is not.

Why this is in the vault

This resolves open follow-up #1 of [[2026-05-25-virta-health-5yr-cohort-reversal-durability]] and supplies the exact, citable number that the "behavioral substitution" bear vector in [[2026-05-21-lilly-glp1-longevity-thesis]] is weighted against — replacing a derived estimate with a primary-source figure and its correct denominator, so the LLY position's bear-case weighting rests on something auditable rather than on multiplied press-release percentages.

Open follow-ups

• No published ITT for the medication-reduction outcomes. The paper states plainly that medication-prescription changes were assessed "among completers only." The payer-savings arm of the LLY bear case depends on those numbers. Can an ITT-equivalent be reconstructed (e.g. from the Medication Effect Score mixed-effects model on n=169, which is ITT: 1.5 → 0.9, a 40% reduction), or does that require unpublished data?
• The missing 93. 68 of 262 CCI participants were never approached for the extension and 25 declined. Why were 68 not approached? The paper does not say. If they were disproportionately non-responders or program-exited, the 262-anchored 9.2% is closer to truth than the paper's 14.2%; if approach was administrative/random, 14.2% is the fairer figure. This is the single biggest remaining uncertainty in the calibration.
• The GLP-1-discontinuation subgroup past 18 months — carried forward unresolved from the parent brief. That cohort (Inkinen's headline datum), not the 2015-16 CCI cohort, is the cleanest test of behavioral substitution, and the DRCP 2024 cohort predates widespread GLP-1 use entirely.
• Comparator denominator normalization. The paper's discussion compares its 20% completer rate to Look AHEAD's 7.3% and DiRECT's 13% without stating whether those are ITT. If they are ITT and Virta's cited 20% is completer, the favourable comparison is partly an artifact. Worth one pass through the DiRECT 5-year paper.
• Cohort vintage. Enrollment ran August 2015 to May 2016. Standard of care, GLP-1 availability, and CGM penetration have all moved substantially. How much of the 5-year result is transportable to a 2026 cohort?

Related

• [[2026-05-25-virta-health-5yr-cohort-reversal-durability]] — parent brief; this replaces its ~10% ITT estimate
• [[2026-05-21-lilly-glp1-longevity-thesis]] — the live position this calibrates
• [[2026-05-24-virta-health-payer-contracts-lly-thesis-input]] — payer-economics arm, whose inputs are completer-only
• [[2026-07-18-capital-rx-reverse-lilly-relationship]] — substitution-vs-complement resolution on the payer side
• [[2026-07-18-virta-covered-lives-tracker-data-source]] — the scale-side tracker for the same thesis
• [[2026-05-22-tim-ferriss-sami-inkinen-virta-t2d-rowing]] — Inkinen anchor interview

Sources

Primary (fetched and read in full this run):

• McKenzie AL, Athinarayanan SJ, Van Tieghem MR, Volk BM, Roberts CGP, Adams RN, Volek JS, Phinney SD, Hallberg SJ. "5-Year effects of a novel continuous remote care model with carbohydrate-restricted nutrition therapy including nutritional ketosis in type 2 diabetes: An extension study." Diabetes Research and Clinical Practice 2024;217:111898. DOI 10.1016/j.diabres.2024.111898. Open access CC-BY. Full text: https://www.diabetesresearchclinicalpractice.com/article/S0168-8227(24)00808-8/fulltext — Table 1, section 2.4 (Statistical analysis), section 3.2 (Results), section 4 (Discussion) read directly. Access note: WebFetch and curl both returned HTTP 403 (Cloudflare bot challenge); full text was obtained through a real browser session. The article itself is not paywalled.
• PubMed record / structured abstract, PMID 39433217, retrieved via NCBI E-utilities efetch: https://pubmed.ncbi.nlm.nih.gov/39433217/
• Unpaywall API (DOI 10.1016/j.diabres.2024.111898) — confirms hybrid open access, CC-BY, publisher-hosted, no repository copy. Europe PMC and PMC hold no full-text copy (inPMC: N).
• ClinicalTrials.gov NCT02519309 (registration for the parent 2-year trial) — referenced by the paper; not separately fetched this run.

Secondary (not used for any figure in this brief):

• Virta Health press release on the 5-year study — deliberately not cited as a source of numbers; it is the completer-framing surface this brief exists to bypass.

Own arithmetic (labelled, not from the paper):

• 24/262 = 9.2%; 39/262 = 14.9%; 122/262 = 46.6% retention from original CCI enrollment; 19/169 = 11.2% and 15/169 = 8.9% for 3- and 4-year sustained remission on the ITT denominator; (24.3 − 14.2)/24.3 = 41.6% relative decay in ITT remission from year 2 to year 5.