Virta 5-year durability: real but decaying, and the headline numbers are survivorship-flattered

The question

Does Virta's diabetes-reversal durability hold above 50% at year-5, or drop below 30%? Hallberg et al. (2018-2021) showed strong 2-year persistence; the 5-year sustainability is the load-bearing input for the lifestyle-vs-GLP-1 substitution case feeding the LLY-longevity-v1 bear-case calibration. (i.e. if a behavioral program sustains reversal for 5 years, that's a credible substitute that threatens LLY's recurring-pharma revenue; if it decays, GLP-1s look more durable as a recurring product.)

What we already know (from the vault)

• [[06-reference/2026-05-22-tim-ferriss-sami-inkinen-virta-t2d-rowing]] (Inkinen interview) is the anchor: Virta intent-to-treat 13% body-weight loss sustained at 1 year, up to 75% MASH reduction, and peer-reviewed 18-month sustained weight loss in patients who started on GLP-1 then discontinued. Crucially the note flags "longer-term data coming" — the 5-year gap this brief now fills.
• [[01-projects/investing/theses/2026-05-21-lilly-glp1-longevity-thesis]] is the live thesis. The bear-case pivot: "what happens when LLY's recurring pharma revenue meets a behavioral-intervention competitor that can sustain outcomes post-discontinuation."
• The Inkinen framing already gives the directional bear argument (80% of Virta GLP-1 patients want off the drug; rebound + muscle loss on discontinuation). The missing piece was the hard 5-year persistence number — which decides whether the substitution case is real or aspirational.

What the web says

• The 5-year data is published and real: "5-Year effects of a novel continuous remote care model... an extension study," Diabetes Research and Clinical Practice (2024) — the Hallberg/Virta CCI cohort extended from the 2-year Frontiers (2019) trial.
• Retention/attrition: ~50% of T2D participants retained at 5 years. Of those who continued past year-2, 72% stayed through year-5. The cohort chain: 262 CCI participants -> 194 approached for the 3-year extension -> 169 consented -> 122 remained at 5 years. So roughly 122/262 ≈ 47% completer retention from baseline.
• Reversal/remission at 5 years (among completers): ~1/5 (20%) in full remission (A1c <6.5% with NO diabetes meds for >=3 months); ~1/3 (33%) at A1c <6.5% on no meds OR metformin only.
• Weight: average 7.6% at 5 years (vs the 13% ITT figure at 1 year) — about 60% of the peak effect retained on average; prediabetic subgroup ~6% sustained.
• Other markers: diabetes-drug prescriptions down ~50% across the cohort at 5 years; triglycerides, HDL, inflammatory markers improved.
• Analysis caveat (load-bearing): the headline 5-year remission/A1c percentages are framed around "patients completing five years" — i.e. completer analysis, not intent-to-treat. Combined with ~47% attrition, the true baseline-anchored durability is materially lower than the headline 20%/33% suggest. The Virta press materials do not lead with an ITT 5-year remission rate.

Convergences and contradictions

• Convergence: the web confirms the vault's directional claim — behavioral reversal IS durable for a meaningful subset out to 5 years, and that subset materially reduces diabetes-drug use. Inkinen's "sustained" framing is not marketing fiction; there is peer-reviewed multi-year persistence.
• Contradiction with the question's framing: the question asks "above 50% or below 30%?" — but that binary conflates two different metrics. Retention sits right at ~47-50% at 5 years. Full remission among completers is ~20%; A1c-controlled (no-meds-or-metformin) among completers is ~33%. On an intent-to-treat basis (the honest denominator), full remission is well below 30% — likely in the ~10% range (≈20% of the ~47% who completed). So the honest answer is: durability is real but lands BELOW the 30% bar on an ITT basis for full remission, and the >50% framing only holds for the softer "retention" and "improved markers" metrics, not for hard reversal.

Synthesis for RDCO

This is a calibration input, and the calibrated read tightens the LLY bear-case rather than strengthening it. The substitution thesis (behavioral programs durably replace GLP-1s) is weaker than the Virta headline numbers imply once you correct for survivorship. The right way to state it for the thesis: Virta delivers durable full diabetes remission to roughly 10% of an intent-to-treat cohort at 5 years (≈20% of the ~47% who complete), with another tranche achieving good control on minimal medication. That is a clinically real and cost-relevant outcome, but it is NOT a population-scale replacement for pharma — half the cohort drops out and most completers do not reach full drug-free remission.

For the LLY-longevity-v1 thesis, this argues for the complement, not substitute framing — which also aligns with Inkinen's own positioning (Virta prescribes GLP-1s when appropriate; the Capital Rx co-product is a triage-front-loader). The bear-case "behavioral intervention erodes LLY's recurring revenue" should be down-weighted: the behavioral path sustains a minority, the majority either drops out or stays on some medication, and the GLP-1 discontinuation-rebound dynamic (Inkinen: weight skyrockets back, muscle loss) actually preserves a recurring-use rationale for the drug. Net: this evidence is mildly LLY-supportive (durable behavioral substitution is a smaller threat than the Virta marketing surface suggests), and it should NOT be used to inflate the bear-case.

One honest counter-weight before over-correcting: 5-year ITT remission of ~10% from a cheap, scalable behavioral program is still meaningful at population/payer scale, and the prescription-reduction (~50%) is exactly the kind of payer-cost-savings number that makes Virta's payer contracts viable (see the now-Done payer-contract brief). So the threat to LLY is real at the margin (payers steering patients to a cheaper triage step first), just not the wholesale-substitution threat the question's framing implied. Calibration verdict: trim the bear-case weight on "behavioral substitution," keep a small weight on "payer-driven triage-before-GLP-1."

Open follow-ups

• What is the actual published intent-to-treat 5-year full-remission rate (not completer)? The press materials lead with completer figures; the DRCP 2024 paper likely reports ITT — worth pulling the primary paper's ITT number to replace my ~10% estimate with the exact figure.
• How does the GLP-1-discontinuation 18-month sustained-loss subgroup (Inkinen's headline bear-case datum) extend past 18 months? That specific cohort, not the general CCI cohort, is the cleanest test of "behavioral intervention sustains what GLP-1 started."
• Payer-contract economics: does the ~50% prescription-reduction translate into shared-savings that make Virta cheaper than GLP-1s on a per-member basis? (Feeds the substitution-vs-complement and Capital Rx/Lilly questions, both now Approved.)
• Does Lilly itself fund or partner with any behavioral-reversal program (hedge / triage funnel)? If so the substitute-vs-complement question is already answered by LLY's own behavior.

Sources

• [[06-reference/2026-05-22-tim-ferriss-sami-inkinen-virta-t2d-rowing]] (vault)
• [[01-projects/investing/theses/2026-05-21-lilly-glp1-longevity-thesis]] (vault)
• Diabetes Research and Clinical Practice (2024) — 5-Year effects of a novel continuous remote care model... extension study: https://www.diabetesresearchclinicalpractice.com/article/S0168-8227(24)00808-8/fulltext
• Virta press — sustainable health improvements, 5-year diabetes reversal study: https://www.virtahealth.com/press/virta-sustainable-health-improvements-5-year-diabetes-reversal-study
• Frontiers (2019) — Long-Term Effects... 2-Year Non-randomized Clinical Trial (Hallberg cohort origin): https://www.frontiersin.org/articles/10.3389/fendo.2019.00348/full
• Virta blog — 2-year sustainability outcomes: https://www.virtahealth.com/blog/2yr-t2d-trial-sustainability
• Virta Outcomes page: https://www.virtahealth.com/outcomes