There is no published evidence past 18 months, and the "peer-reviewed 18-month" figure in our own vault was a misread
The question
"What happens to the GLP-1-discontinuation subgroup past 18 months? That cohort, not the 2015-16 DRCP cohort, is the cleanest test of behavioral substitution — the DRCP enrollment predates widespread GLP-1 use entirely."
Carried forward unresolved through two prior briefs. The question's framing is correct: the Diabetes Research and Clinical Practice continuous-care cohort enrolled August 2015 to May 2016 and cannot test substitution. The deprescription cohort can. The answer is that the deprescription cohort stops at 12 months, and the "18 months" that made this question seem answerable does not refer to post-discontinuation follow-up at all.
What we already know (from the vault)
- [[2026-05-22-tim-ferriss-sami-inkinen-virta-t2d-rowing]] is where the 18-month figure entered the vault, recorded as "peer-reviewed 18-month sustained weight-loss data on patients who came in on GLP-1s and later discontinued." The note already flagged it as company-reported by an interested chief executive. That flag was correct and the figure is now resolved as a misread (see below).
- [[2026-05-25-virta-health-5yr-cohort-reversal-durability]] raised this as open follow-up #2 and named it "the cleanest test of 'behavioral intervention sustains what GLP-1 started.'"
- [[2026-07-30-virta-drcp-5yr-itt-remission-rate]] established the primary-source discipline this brief follows: tables are load-bearing, discussion prose is not. It also carried the question forward verbatim.
- [[2026-10-11-virta-drcp-68-unapproached-participants-denominator]] settled the denominator question on the 2015-16 cohort (9.2% on 262, not 14.2% on 169) and left this as the last open vector. Two sibling briefs close the thread's other open vectors the same night: [[2026-10-12-lookahead-direct-remission-itt-vs-completer]] (comparator denominators) and [[2026-10-12-virta-drcp-medication-effect-score-itt-reconstruction]] (medication-reduction ITT reconstruction).
- [[2026-05-21-lilly-glp1-longevity-thesis]] is the live position. The bear vector under test: behavioral programs durably substitute for glucagon-like peptide-1 receptor agonists (GLP-1 RAs) and erode Eli Lilly recurring revenue. Prior briefs already trimmed that weight toward "complement, not substitute."
What the web says
The behavioral-substitution test (discontinuation INTO a program) — exactly one peer-reviewed study exists, and it stops at 12 months.
- McKenzie AL, Athinarayanan SJ. "Impact of Glucagon-Like Peptide 1 Agonist Deprescription in Type 2 Diabetes in a Real-World Setting: A Propensity Score Matched Cohort Study." Diabetes Therapy 2024;15(4):843-853. Read in full via PubMed Central (PMC10951146), open access CC BY-NC. Methods, verbatim: HbA1c and body weight "at enrollment in clinic (pre-CRNT), at date of deprescription or index date (derx/ID), and at 6 and 12 months (m) post-derx/ID were utilized in this study." There is no later timepoint.
- Where our 18 months came from. Results, verbatim: "The mean duration of care in both cohorts was at least 18 months." That is total program contact, spanning the 3-9 months of carbohydrate-restricted nutrition therapy (CRNT) before deprescription plus the 12 months after. It is not 18 months of post-discontinuation follow-up. Verified against the paper's own text.
- Design and sample. Retrospective, propensity-score matched, not randomized. 154 deprescribed (DeRx) matched 1:1 with 154 who continued (Rx). Enrollment HbA1c 7.3%, body mass index 35.7 kg/m².
- Results at the 12-month ceiling. Weight did not significantly increase at 6 or 12 months after deprescription (p > 0.05); more than 70% maintained at least 5% weight loss at 12 months. HbA1c rose +0.4% (95% CI 0.2, 0.6) at 6 months and +0.6% (95% CI 0.3, 0.8) at 12 months versus deprescription, with means staying in the non-diabetic range (6.0% then 6.2%). At 12 months, 64.8% of DeRx and 64.1% of Rx remained below 6.5%.
- Responder selection is built into the inclusion criteria. Patients qualified only if their GLP-1 was deprescribed "following improved glycemia to HbA1c < 6.5% within 3-9 months of beginning CRNT." The cohort is, by construction, people for whom the behavioral program had already worked. This is not a test of whether a program substitutes for the drug; it is a test of whether, among responders, stopping the drug is safe for a year.
- The drug mix is pre-tirzepatide. Prior to enrollment: dulaglutide 43.5%, semaglutide 29.2%, liraglutide 17.5%, exenatide 7.1%. No tirzepatide. The authors' own limitation, verbatim: "Application of these findings is limited to the medications utilized by the patient population included in this analysis. Future research should evaluate the effect of GLP-1 deprescription including medications which have recently come to market."
- Sponsorship. Verbatim: "This study and the Rapid Service Fee was funded by Virta Health." Both authors held Virta Health stock; Athinarayanan is a Virta employee, McKenzie was at study initiation. Data "are not publicly available." Preprinted on medRxiv June 2023.
Pure discontinuation (no structured program) — the randomized ceiling is also 12 months off-drug.
- STEP 1 extension (Wilding et al., Diabetes Obes Metab 2022): 68 weeks of treatment then 52 weeks off. Participants regained roughly two-thirds of lost weight, from -17.3% to a net -5.6% versus -0.1% placebo. Off-drug window: 12 months.
- SURMOUNT-4 (tirzepatide randomized withdrawal): 36-week lead-in then 52-week withdrawal. The placebo-switch arm regained 14.0% of body weight across weeks 36 to 88 while the continuation arm kept losing. Off-drug window: 12 months.
- STEP 4 (Rubino et al., JAMA 2021): withdrawal at week 20, 48 weeks off. Shorter still.
- A 2025 eClinicalMedicine systematic review and meta-analysis, "Metabolic rebound after GLP-1 receptor agonist discontinuation," concludes that withdrawal produces substantial weight regain and reversal of lipid, blood-pressure and glycemic gains, and that it is unclear whether the regain trajectory continues in subsequent years. Paywall/bot-block flagged: thelancet.com returned HTTP 403 this run and was not retried, so this is characterized from search-result summaries, not read directly. A related 2025 medRxiv nonlinear meta-regression preprint circulates a "largely reversed within ~20 months" figure via commercial secondary sites; I could not verify it against the preprint and am not carrying it as evidence.
Real-world claims and records cohorts — nothing clean past 12 months either.
- A Cleveland Clinic records cohort (n = 7,938) followed 12 months after stopping semaglutide or tirzepatide: the obesity group regained roughly 0.5% and the diabetes group lost a further 1.3% on average, confounded because many patients restarted or switched agents. 12 months.
- A UK Clinical Practice Research Datalink cohort (n = 589, type 2 diabetes, injectable GLP-1 RAs initiated 2009-2017) reaches 24 months, but reports persistence, not post-discontinuation trajectory: discontinuation rose from 45.2% at 12 months to 64.7% at 24 months. Old drug era, wrong endpoint for this question.
- Virta's own non-peer-reviewed surface is shorter, not longer. The April 2025 press release claiming GLP-1 use cut "over 50%" rests on "preliminary findings from Virta's book of business" and "Virta internal analysis," with headline outcomes at 90 days (n > 2,000 enterprise GLP-1 users) and post-GLP-1 maintenance described as "up to six months." The 12-month claim is explicitly the Diabetes Therapy diabetes study. The strings "18 months," "24 months" and "2 years" do not appear.
Convergences and contradictions
- Hard contradiction, resolved against the vault. The vault recorded a peer-reviewed 18-month post-discontinuation result. No such result exists. The 18 months is mean duration of care in a study whose last measurement is month 12. [[2026-05-22-tim-ferriss-sami-inkinen-virta-t2d-rowing]] should be annotated, and the three downstream briefs that carried the question inherited a premise that was never true.
- Convergence on the evidence ceiling. Sponsor trials, independent meta-analysis, real-world records cohorts and the one behavioral-program study all top out at 12 months off-drug. The question "what happens past 18 months" has no answer in the published literature from any direction, for either pure discontinuation or discontinuation into a program.
- Genuine divergence in direction at 12 months, and it matters. Randomized withdrawal without a program shows large regain (two-thirds of loss in STEP 1; 14.0% of body weight in SURMOUNT-4). The Virta deprescription cohort shows no significant weight regain. Those are not reconcilable as drug-class effects alone; the gap is explained at least partly by responder selection and by the weaker agents in the Virta mix, and it cannot be cleanly attributed to the program.
Synthesis for RDCO
The finding is a gap, and it should be written into the thesis as a gap. The longest credible follow-up on a deprescribing-into-program cohort is 12 months, not 18, and it is one Virta-funded, non-randomized, responder-selected study of 154 patients that is 68.1% composed (my arithmetic: 43.5 + 17.5 + 7.1) of dulaglutide, liraglutide and exenatide users with zero tirzepatide exposure. The behavioral-substitution bear vector against [[2026-05-21-lilly-glp1-longevity-thesis]] therefore cannot be strengthened or refuted by durability data, because the durability data does not exist at the horizon that would matter. Three briefs in this thread have now tried to resolve substitution durability from the Virta evidence base. The correct conclusion is that this line of inquiry is exhausted until someone publishes a longer cohort, and continuing to mine it is spending research budget on a question the literature cannot currently answer.
The single most bearish datum in the literature is in this paper, and it is weaker than it looks. At 12 months, 64.8% of the deprescribed cohort held HbA1c below 6.5% versus 64.1% of those who stayed on the drug. Continuing the GLP-1 bought essentially nothing. Taken at face value that is the substitution thesis in miniature. But both cohorts entered at 100% below 6.5% by inclusion criteria, so 35.2 percentage points of the deprescribed cohort lost glycemic target within a single year (my arithmetic: 100 − 64.8), with the mean drifting 6.0% to 6.2%. The honest read is that both arms were decaying in parallel, and the study is better described as "in glycemic responders on older GLP-1 RAs, the drug was no longer doing marginal work" than as "nutrition therapy replaces the drug." Generalizing it to a tirzepatide or retatrutide patient who lost 22-30% of body weight is an extrapolation across drug classes that the authors themselves decline to make.
For the Lilly position this is mildly supportive, for an unsatisfying reason: the test that would threaten the thesis has not been run. Lilly's franchise economics rest on tirzepatide and, prospectively, retatrutide. The only published behavioral-off-ramp evidence covers a drug mix Lilly largely does not sell into the obesity market, in a type 2 diabetes population at body mass index 35.7, selected for having already achieved target on diet. Meanwhile the randomized withdrawal evidence that does cover the modern agents, SURMOUNT-4 most directly, shows substantial regain and preserves the chronic-use rationale. Net effect on the thesis: no change to the "complement, not substitute" framing that [[2026-05-25-virta-health-5yr-cohort-reversal-durability]] and [[2026-07-30-virta-drcp-5yr-itt-remission-rate]] arrived at. The one thing to add to the thesis file is an explicit evidence-ceiling line, so a future reader does not re-open this: no published post-GLP-1-discontinuation cohort, with or without a behavioral program, reports outcomes beyond 12 months as of 2026-10-12.
One reusable methodological lesson, consistent with the pattern this thread keeps finding. The 2026-07-30 brief caught completer figures laundered as intent-to-treat inside a peer-reviewed discussion section. This brief catches a duration-of-care figure laundered into a follow-up duration across an interview, a vault note, and three downstream briefs. In both cases the correct figure was in the same document, and in both cases the error survived because a number was carried forward without its unit of measurement. Any duration, denominator or percentage entering the vault from a spoken source should be stamped with what it is measuring before it is cited twice.
Why this is in the vault
This closes the last open follow-up of [[2026-10-11-virta-drcp-68-unapproached-participants-denominator]] and retires the behavioral-substitution durability line of the [[2026-05-21-lilly-glp1-longevity-thesis]] bear case as unanswerable from published data, while correcting a factual error ("peer-reviewed 18-month post-discontinuation data") that originated in [[2026-05-22-tim-ferriss-sami-inkinen-virta-t2d-rowing]] and propagated through three research briefs unchallenged.
Open follow-ups
- Attrition denominator in the deprescription study. The paper reports percentages at 12 months but the main text does not state how many of the 154 per cohort had weight and HbA1c data at that timepoint. Supplementary Fig. 1 and 2 and the supplementary PDF were not retrieved this run. This is the same completer-versus-intent-to-treat trap the 2026-07-30 brief found in the five-year paper, and it is unresolved here.
- Any tirzepatide-era deprescription cohort. Has Virta or anyone else presented a semaglutide-2.4mg or tirzepatide-specific off-ramp cohort at the American Diabetes Association or Obesity Week in 2025 or 2026, even as a conference abstract? This is the study that would actually test the thesis.
- The eClinicalMedicine 2025 meta-analysis full text, blocked by HTTP 403 this run. Does any included study carry an off-drug arm beyond 52 weeks, and does the paper state a time-to-baseline estimate? Retrievable via a real browser session per the pattern that worked for the 2024 extension study.
- Any independent, non-sponsor behavioral-program-after-discontinuation cohort at any duration. Every program-based datum found this run traces to Virta.
- Virta's one-year 13% weight-loss figure cited in the April 2025 press release links to an MDPI article. What is its denominator, and does MDPI tier clear the vault's citation bar?
- Lilly-side instrumentation. Is there a planned SURMOUNT-4 extension or a retatrutide TRIUMPH withdrawal substudy with an off-drug arm beyond 52 weeks? If Lilly itself is funding longer withdrawal data, that is a signal about what it expects to find.
Related
- [[2026-10-11-virta-drcp-68-unapproached-participants-denominator]] — immediate predecessor; this closes its last open vector
- [[2026-10-12-lookahead-direct-remission-itt-vs-completer]] — sibling brief closing the comparator-denominator vector of the same thread
- [[2026-10-12-virta-drcp-medication-effect-score-itt-reconstruction]] — sibling brief closing the medication-reduction ITT vector of the same thread
- [[2026-07-30-virta-drcp-5yr-itt-remission-rate]] — parent brief; established the tables-over-prose discipline used here
- [[2026-05-25-virta-health-5yr-cohort-reversal-durability]] — grandparent brief; where this question originated as follow-up #2
- [[2026-05-21-lilly-glp1-longevity-thesis]] — the live position this calibrates
- [[2026-05-22-tim-ferriss-sami-inkinen-virta-t2d-rowing]] — the Inkinen anchor; needs annotation, its 18-month claim is a misread of mean duration of care
- [[2026-05-24-virta-health-payer-contracts-lly-thesis-input]] — payer-economics arm of the same bear vector
- [[2026-07-18-capital-rx-reverse-lilly-relationship]] — substitution-versus-complement resolution on the payer side
Sources
Primary, read in full this run:
- McKenzie AL, Athinarayanan SJ. "Impact of Glucagon-Like Peptide 1 Agonist Deprescription in Type 2 Diabetes in a Real-World Setting: A Propensity Score Matched Cohort Study." Diabetes Therapy 2024;15(4):843-853. DOI 10.1007/s13300-024-01547-0. PMID 38421559. Open access CC BY-NC. Full text read via PubMed Central PMC10951146 (NCBI E-utilities efetch). Abstract, Methods, Results, Limitations, Funding and Conflict-of-Interest statements read directly. Access note: link.springer.com returned a 303 redirect to an identity-provider authorization endpoint; the PMC copy was used instead. The article is not paywalled. Sponsor-funded by Virta Health; both authors Virta stockholders.
Company-published, used only to establish evidence tier and duration claims:
- Virta Health research index: https://www.virtahealth.com/research — confirms exactly one GLP-1-discontinuation publication is listed.
- Virta Health press release, April 2025, "Virta Health Cuts GLP-1 Use for Weight Loss Over 50%": https://www.virtahealth.com/press/virta-health-cuts-glp-1-use-for-weight-loss-over-50 — self-described "preliminary findings from Virta's book of business"; 90-day primary window; "up to six months" post-GLP-1 maintenance. Not peer-reviewed.
- Virta Health press release, February 2024, on the deprescription study: https://www.virtahealth.com/press/virta-health-announces-first-of-its-kind-peer-reviewed-study-proving-its-approach-is-an-effective-off-ramp-from-glp-1s-for-sustained-weight-loss — the completer-framing surface; no figures taken from it.
Secondary, characterized from search summaries rather than read (flagged):
- "Metabolic rebound after GLP-1 receptor agonist discontinuation: a systematic review and meta-analysis," eClinicalMedicine 2025: https://www.thelancet.com/journals/eclinm/article/PIIS2589-5370(25)00614-5/fulltext — HTTP 403, not retried per research caps.
- STEP 1 extension, Wilding et al., Diabetes Obes Metab 2022, via PMC9542252. Figures (-17.3% to -5.6%, two-thirds regain, 52 weeks off-drug) from search summary, not read directly.
- SURMOUNT-4 tirzepatide randomized withdrawal (36-week lead-in, 52-week withdrawal, 14.0% regain weeks 36-88) — from search summary and the vault thesis file, not read directly this run.
- STEP 4, Rubino et al., JAMA 2021 — withdrawal at week 20, 48 weeks off. From search summary.
- Cleveland Clinic post-discontinuation records cohort (n = 7,938, 12 months) and UK CPRD persistence cohort (n = 589, 24 months, PMID 35101924) — from search summaries.
- "Trajectory of weight regain after cessation of GLP-1 receptor agonists," medRxiv 2025 preprint — the circulating "~20 months to largely reversed" figure was not verified and is not carried as evidence.
Own arithmetic (labelled, not from any source):
- 100 − 64.8 = 35.2 percentage points of the deprescribed cohort losing HbA1c < 6.5% target within 12 months of deprescription.
- 43.5 + 17.5 + 7.1 = 68.1% of the deprescribed cohort on dulaglutide, liraglutide or exenatide prior to enrollment.