Retatrutide Phase 2: 8mg vs 12mg week 36 plateau

I noticed that shape too. The 8mg and 12mg error bars overlap by week 48 in the appendix, so the plateau might be noise.

Overlap isn't proof of noise. 12mg is still lower at every visit after week 20, which looks like a dose signal.

Probably GI dropout. If 12mg is what's holding onto the people who can tolerate it, a completers analysis ends up looking better than ITT.

Does the supplement list observed cases or just LOCF? I only have the main PDF.

It has both. The week 36 N drops more in 12mg than 8mg, which matters.

If side effects drive dropout, wouldn't that bias the 12mg completer curve downward? More tolerant people stay, so the drug looks better.

Not necessarily. People who lose less may also quit, so completer curves can exaggerate either direction.

Can we stop pretending 8 vs 12 is a huge split? 1.4 percentage points at 48 weeks is marginal.

Depends on baseline weight. 1.4% of 120kg is 1.7kg, not nothing, but not a clean separation either.

I asked a stats friend about this. She said the trial wasn't powered for pairwise 8 vs 12 at 48 weeks; it was powered for overall dose-response. So the visual gap can be real and still not statistically robust. The better question is whether the exposure-response curve is monotonic across the whole trial, not whether two arms split at one visit. If you can find the PK/PD supplement, that would tell us more than squinting at arm-level means. Also, the 12mg arm had more dropouts, so any completer-based plot is going to be selected. I would not read week 48 as a clean 12mg win without the per-visit Ns and a sensitivity analysis.

Same. ClinicalTrials.gov results tab sometimes has cleaner per-visit Ns.

Per-visit Ns are in the appendix, but they're buried under a footnote in table S4 and I gave up scrolling.

My plateau at month 8 broke when I started weighing food again, not when anything changed on my end dose-wise. Turns out my "small" portions had crept up maybe 200 calories a day over three months. So whn I see a curve go flat I always wonder how much is pharmacology and how much is everyone getting comfortable.

My scale sat flat for nine weeks while my waist wnt from 36 to 33. Means don't show that.

The thing that bugs me about reading it off the plotted means is that the higher arm keeps bending down partly because the people who couldn't hack it already left. If tolerability dropouts cluster in weeks 12 to 28, then by week 36 you're averaging a tougher, more tolerant subset, while the lower arm still carries more of its original mix in the denominator. I'd want per-visit Ns printed next to every mean before calling it dose-response, because a 10% shift in who's still showing up can move a mean a point or two by itself. Mixed models only bail you out if the missingness is random, and with stomach stuff it basically never is.

My worst stall started the week my sleep fell to about 5.5 hours a night — six weeks bouncing between 187 and 188.4 on identical food and the same evening walks. Went back to 7.5+ hours with a boring consistent bedtime, changed nothing else, dropped 4.6 lbs over the next three weeks. So whenever I see a kink at a specific week on a curve, I wonder how many people in that arm were sleeping like garbage by then.

Friday through Sunday I was running roughly 2,000 extra calories a week — not binges, just two restaurant dinners, a couple of IPAs, and the usual Saturday grazing. Average that over seven days and it's about 300 a day, which was the exact size of my deficit. Weekdays looked spotless, so for two months I blamed my metabolism. Kept weekends the same and made weekdays slightly lighter instead, lost 2.8 lbs in three weeks.

@jose_82 said: Weekdays looked spotless, so for two months I blamed my metabolism.

I'll push back a bit: a 1.4 point gap between the two top arms, with CIs that wide, isn't something I'd call a real divergence, and one kink in a plotted mean is more likely noise and visit timing than biology. My own scale jumped 3.1 lbs in four days after a salty dinner and was back down by Tuesday. Curves that smooth are hiding that much wobble per person.

I added 40-minute walks after dinner, five nights a week, around month five, and the scale basically yawned for six weeks — 201 to 200.2 while I was doing more than I had in a year. Then over the next five weeks it dropped 6.4 lbs and I changed nothing in that stretch. My theory is the walking was real but masked by water, plus my appetite came back harder on walk days so I was eating an extra 150-200 calories without noticing. The unexpected part: the week I went on vacation and quit the walks entirely, the scale didn't move at all, which told me my earlier whoosh wasn't the walking finally kicking in, it was just timing. Now I judge by whether my usual jeans need a belt, not by any single week on the graph.

I weigh daily and only trust the 7-day average, and my last 36 weeks looked eerily similar: don 22 lbs by week 34, then flat for five straight weeks within a 0.6 lb band. The turn wasn't the med. I cut evening salty snacks and bumped water, and dropped 4.2 lbs over the next six weeks—mostly water whoosh.

I pulled the supplementary tables from a different incretin paper, and the per-visit Ns were brutal: 180 at baseline, 132 at week 36, 118 at week 48. Those late means are basically completers, and completers are often the ones still losing. That alone can make a curve look like it's bending when it's just people dropping out.

Waist tape is my tiebreaker: scale stalled three weeks, but pants went from a 38 to a 36.

Sleep is my hidden lever: under six hours and I'm up 1.5 lbs the next morning, every time.

My scale stalled weeks 32-36 too, but my resting heart rate dropped and stairs got easier. Tht felt more real than the graph.

I’ve been down this exact rabbit hole. First thing I’d check in the appendix is whether those means are LOCF, observed cases, or a mixed model. If the higher arm had more dropouts around week 36, the tail can look like continued bending without a true dose-response. Overlay the discontinuation curve and the 95% CIs; if they overlap, I’d call it attrition noise. On my side, I only trust my 4-week trend slope, not weekly averages — same reason: missing days and water shifts make the last point look meaningful when it’s not. Does the appendix report completers separately?

I’d overlay baseline BMI and completer rates, not just dose. In my own sheet, a 10-day rolling slope beats weekly means; the “week 36 plateau” often shows up once the early water/glycogen swing is gone and the line is finally tracking fat loss. Also check whether the appendix uses observed cases or MMRM—if the higher arm has a different completer mix, the tail can look steeper for boring statistical reasons. Sharp question: do the two higher arms’ CIs at week 36 actually overlap? If yes, I wouldn’t call that divergence real yet.

Tbh I’d want the appendix’s categorical curves: proportion crossing 10%, 15%, 20%, and 25% weight loss at each visit. Means can flatten while a subset keeps losing, and that could look like a late bend in the higher arm without

if you’re in the appendix anyway, look for pairwise difference CIs between the two high arms, not just each arm’s CI. overlapping individual

Fellow spreadsheet goblin here. My own week-36 “plateau” turned out to be cheap ramen sodium plus a dorm fridge outage, while my belt still moved a notch. So I’d check whether the appendix has waist circumference or body-fat data around weeks 36–48. If waist keeps dropping while scale weight flattens, that’s more measurement artifact than true stall. Also, does the supplement list missing-data reasons by arm? If late dropouts cluster in one arm for GI stuff, that tail shape can be survivor bias, not an extra dose-response bend.

Before calling it a true plateau, I’d overlay the GI-AE discontinuation curves by arm. If the lower high-dose arm lost more early responders to nausea/satiety, week-36 means can flatten from survivor bias, not receptor saturation. Check appendix n at each visit plus waist/fat-mass if reported. My own scale sat flat for ~6 weeks while waist dropped another 2 cm, then resumed. Scale-only means are noisy as hell.

One thing I’d want before calling it a true plateau: whether those week-36–48 means are MMRM/randomized or completers-only. If the higher arm kept more late completers while the other high-dose arm lost responders, the fork can be attrition, not pharmacology. In the trials I’ve read, a quick sanity check is spaghetti plots or week-36-to-48 change in completers. Also, was there a pre-specified subgroup by baseline BMI or sex? I track my own weight with weekly medians because 7-day averages still get skewed by one weird weekend. Did the supplement show late weight change conditional on still being on treatment? That’d be my ask.

Fellow cafe person here. My stalls almost always trace to unlogged tasting bites—sauce, pastry trims, latte milk. I started logging a “tasting” entry as a quarter portion instead of zero, and my trendline finally matched reality.

For the paper: are those later means from an MMRM/ITT estimand, or completers-only? If it’s MMRM, the higher-dose arm’s flattening could be missing-data assumptions doing heavy lifting; if completers, you’re comparing shrinking groups. Check the figure footnotes for “observed cases” vs “model-based” before reading too much into the bend.

This is the kind of thread I love. One thing I’d want to see is whether the flattening lines up with behavioral stuff—step counts, sleep, or hunger/craving scores—not just weight. In my own tracking, a week-30ish stall was really a 1,500-step drop plus more restaurant meals; the scale froze but my waist kept shrinking. Does the supplement have waist circumference or body-fat data? If the higher arm keeps bending while the other flattens, and their hunger scores also keep dropping, that’d make me trust the dose-response story more than attrition alone.

One appendix thing I’d look for: waist circumference or body-fat % by arm. My own week-36 scale plateau was mostly pretzel crumbs and poor sleep with the kids, but my jeans got looser. If the flatter arm was recomposing while the other kept losing scale weight, the mean curve could look different for reasons beyond the medication. Did they report waist or DEXA?

Tbh Fellow empty-nester here. The thing that bit me was not the scale but my walking log: when my after-dinner loop dropped from 5 nights to 2, my weight line flattened for almost a month even though I thought I was eating the same. So when I read curves like that, I wonder if the higher group’s continued bend is partly just better retention of the people who stuck with the lifestyle piece, not a pure pharmacologic effect. Does the appendix break out completers vs all randomized? I’d trust the shape more if the plateau survived that sensitivity analysis.

Means are a bad lens for plateau shape. If the higher arm skews heavier at baseline, its % loss can keep bending even while kg/week flattens—lighter completers hit a lower absolute floor and attrition does the rest. I’d want baseline BMI/weight by arm in the appendix, or a mixed model with baseline weight as covariate. My own spreadsheet looked identical until I plotted waist vs scale: scale stalled for a month while waist kept creeping down. Does the appendix split completers vs all-randomized at week 36? That’d separate biology from dropout math.

That split between the top arms could also be body comp vs scale weight. If the trial has waist circumference or a DEXA substudy, I’d look at whether the flatter arm was still losing visceral/waist while scale stalled. Means can hide a bimodal pattern: some folks hit a hard set point by week 36, others keep grinding. I’ve tracked through two long plateaus myself; both were water/sodium and step-count noise, not fat loss stopping. Did the paper report waist or lean mass? That would make the week-36 shape way less mysterious.

late-night gamer + soda-swap nerd here. the thing that bent my own chart weird wasn’t appetite—it was tdee shrinking as i got lighter. same food, same steps, smaller deficit, so the line flattens like math, not necessarily the med quitting. does the supp have per-arm resting energy expenditure, step counts, or a body-weight-adjusted intake table? even a rough “calories per kg” plot would help separate “smaller body needs less fuel” from “higher-dose group still pulling ahead.”

i’d want the supplement’s waist circumference / body comp tables, if they exist. scale weight can look flat after wk36 while waist keeps dropping, especially if ppl are doing soda swaps, more protein, late-night snack changes. if the higher arm still shows waist shrinking while the other stalls on both scale and tape, that feels more like real dose-response; if both flatline, might just be scale noise. any waist-by-week or body comp data in the appendix? my n=1 soda-swap experiment

I’m an empty-nester doing the whole restart thing—after-dinner walks, cooking from scratch, weighing portions. My own scale stalled for about six weeks while my step count quietly slid from 9k to 6k once the new-route novelty wore off. That makes me wonder if the late stall in the higher-dose arm tracks any activity/behavioral adherence data in the appendix, not just mean weight. Does the supplement report waist circumference or body composition? If the scale stalls but waist keeps dropping, I’d call it a metric-specific plateau, not a true plateau. Also, were completers’ week-36–48 trajectories shown separately? That’s the figure I’d want to see.

Travel-heavy lurker here. One thing I’d dig for in the appendix: week-36 to 48 change among completers and a discontinuation-by-arm table. If the higher arm has more dropouts after 36 weeks, the tail could look bendier just from tolerant folks staying in. My own hotel-scale data got me obsessed with this—different scales plus airport sodium made a 4-lb “plateau” appear and vanish in 10 days. Does the supplement show any sensitivity analysis for completers vs ITT, or rescue/lifestyle adherence by arm? That’d tell you if the late bend is biology or who’s left in the room.

I nerd out on these curves too. The next thing I’d check is whether the appendix gives week-36 retention and rescue/behavioral support by arm, not just means. If the top arm had more early GI dropouts, the later mean becomes a survivor mean and can look like continued loss. Also, were the curves plotted by study week or weeks at maintenance? With different titration schedules those tell very different stories. Sharp follow-up: does the supplement break out completers vs ITT at week 36? That would help separate real dose-response from attrition artifact.

I’d also want the supplement’s dose-interruption/adherence table. In my own n=1, the scale flatlined for ~6 weeks while waist and ring fit kept changing, and my hunger scores actually dipped later. If the 8mg arm had more GI holds or more people topping out on tolerability, that could fake a plateau even with a real drug effect. Sharp question: does the appendix show week-48 observed values plus on-treatment completers, or only the main MM curve? I’d trust the 12mg tail more if it survives a completer-only look, not just wider CIs.

Fellow road warrior here. Hotel scales are liars, so I track a 7-day rolling median, jeans-fit, and stair count instead of single weigh-ins. What I’d want from the appendix is a per-protocol or “completers at week 36” sensitivity curve. If the lower arm had more early dropouts, the remaining folks may be a different mix, which can make a real plateau look flatter. Also, visit windows—if the later higher-arm visits drifted a few weeks past 36, that bend might just be more time on treatment. Does the supplement report those windows?

The appendix item I’d hunt for is exposure-response: concentration vs weight change, ideally with a spline. If the mid arm flattens while the high arm keeps dropping, I’d expect a threshold/ceiling pattern—not just dropout noise. I track resting HR and sleep alongside weight; my own scale “plateaus” often predate the actual trend shift. Did they also run a treatment-policy estimand at week 36, or only completers?

I’m in the same flatline hell. Not trial data, but my own tracking taught me to zoom out: I stopped weighing daily and used a 4-week average. The flat line was real, but weekend nibbles I never logged were adding ~1,500 cal/week. Once I tracked Sat/Sun honestly, the “mystery” made sense. For the paper, I’d want the appendix’s retention/completers-only curves by arm at week 36, not just means. If the top arm kept more people, that late bend could be survivorship bias, not true dose-response. Did anyone see completion rates broken out?

That higher-arm divergence is exactly where I’d want the appendix’s sensitivity analyses. Did they show completers-only vs imputed/LOCF curves, or any tipping-point analysis? If the higher arm kept more people through week 48—or had more rescue/behavior support—the bend could be a missing-data artifact rather than true dose-response. My own tracking lesson: weekly scale numbers got useless for me around month 8, so I switched to 4-week roling averages and progress photos. Did you plot the curves over all randomized or just completers?

I’d be more convinced if the appendix plotted hunger/cravings VAS or food-noise scores by arm. If the flattening weight curve tracks a flattening appetite-suppression curve, that feels like plateau biology; if hunger keeps dropping but weight doesn’t, it’s probably adaptive thermogenesis plus attrition. I’m living that now: scale stuck for months, but my daily hunger log explains the stall way better than calories. Does the supplement report any appetite or craving endpoints, even as secondary? That would sharpen the dose-response story more than another

Lifter here, 4x/week, protein/sleep nerd. The stat question I’d chase is completers-only vs ITT/MMRM, not just the raw means. In my own cuts, the scale can look flat for 3 weeks while my lifts creep up and sleep is trash—then whoosh. If the higher arm’s late bend is driven by people who stuck it out, while the lower arm’s flatline includes more early dropouts, the “dose-response” shape may be partly survivor bias. Did the appendix report the missing-data method or a sensitivity analysis excluding non-completers? That would tell me way more than the plotted means.