NGH and GPF, claim by claim

How to read the verdicts

• Established science: peer reviewed and independently replicated.
• Supported but still being studied: real findings that are newer, from one group, or associations where cause and effect aren't sorted out yet.
• Our hypothesis, not yet tested: our own synthesis, reasoned from the evidence and waiting for a test.

Where a claim has parts with different evidence, each part gets its own verdict. These are the same T1, T2 and T3 labels used on Posture Dojo Research, in plain words.

The verdicts at a glance

The Neural Generation Hypothesis

The current version (3.0, October 2026) makes a focused claim: the reducible part of a scoliotic curve, what active self-correction removes beyond passive flexibility, is kept going by the body schema, the brain's non-conscious predictive model of the body. Along one route, degraded sensing can come first and help start a curve. Growth and loading decide how large a curve gets. Our plain-language definition comes from it: scoliosis as “a curve your body keeps generating from its own map of itself, not just a bent bone.” Read the paper at doi.org/10.5281/zenodo.19342098.

1. Some people inherit a higher risk of scoliosis, and one of the strongest signals sits near a gene called LBX1.

Verdict: Established science for the inherited risk. Our hypothesis, not yet tested for the idea that LBX1 raises that risk through the body's sensing system.

What the outside evidence says. Large genetic studies across several populations link adolescent idiopathic scoliosis to the region around LBX1 (Takahashi and colleagues, 2011; Londono and colleagues, 2014). LBX1 builds one of the two main classes of sensory relay cells in the spinal cord (touch and pain), some of which handle position signals, so a sensing route is plausible. But the exact gene at that spot isn't confirmed, and the same region also acts in muscle and in how the body axis forms. NGH also includes genes that shape the spine's own cartilage, disc and growth plate. The risk is spread across many small-effect variants: a tendency, not a destiny.

What would prove it wrong. An animal model that removes LBX1 only from sensory neurons and still produces scoliosis would strengthen the sensing route. In people, NGH predicts that at the same curve size, genetic risk weighted toward LBX1 goes with less precise position sense, while risk from the structural genes does not. If the structural genes predict sensing just as well, the sensing route fails. If LBX1 risk tracks sensing and the structural genes don't, the first step of the chain gets its first human support.

2. Losing the body's position sense can start a curve with no earlier bone or muscle problem.

Verdict: Supported but still being studied Strong in animals, not yet shown in ordinary teenage scoliosis.

What the outside evidence says. Mice bred without working position-sense neurons develop scoliosis around puberty, with no prior bone or muscle defect (Blecher and colleagues, 2017). In a follow-up from the same lab, removing a touch-sensing channel from those neurons misaligned the spine, mostly as a forward rounding (Assaraf and colleagues, 2020). That's an experiment, which is the kind of evidence that can show cause, though both papers come from one lab and the mice are also unsteady on their feet. Cutting sensory nerve roots on one side curves the spine in monkeys, more so the more roots are cut (Pincott and colleagues, 1984); that has been replicated across groups, though it works like neuromuscular scoliosis, a different route. Rare human conditions that remove position sense come with scoliosis too, alongside weakness and other complications. What hasn't been shown is that the partial blurring NGH proposes in otherwise healthy teenagers does the same.

What would prove it wrong. The decisive study follows girls aged 9 to 10 with no curve, measures their body sense, and sees whether it predicts who develops one through the growth spurt. If sensing differences only show up at or after the curve appears, the sensory entry route fails. A cheaper first look is possible now, using balance tests from age 10 already sitting in a large UK birth cohort.

3. A subgroup of people with adolescent idiopathic scoliosis sense their body's position less accurately.

Verdict: Established science that, on average, adolescents with AIS make larger position-sense errors than controls. Supported but still being studied that the deficit sits in a subgroup, and for shifted sensory weighting and a shifted felt vertical. Our hypothesis, not yet tested that these differences come before the curve.

What the outside evidence says. A meta-analysis of 11 studies found larger joint-repositioning errors in adolescent idiopathic scoliosis (Lau and colleagues, 2022), and in one clinic cohort about 30% fell below normal values for trunk position sense (Lau and colleagues, 2025). One research network has found that people with scoliosis lean more on the inner ear for balance (Pialasse and colleagues, 2015), and one group found their felt sense of vertical sits about 2 degrees toward the outer side of the curve (Le Berre and colleagues, 2019). Every one of these measurements was taken after the curve appeared, so a model in which the curve reshapes sensing explains much of it equally well, as a major review points out (Schlösser and colleagues, 2014). One recent study found that girls with scoliosis spotted curvature in pictures of spines at lower thresholds than controls, not higher (Virovec and colleagues, 2026). That is one reason we never say “the brain can't feel the curve.” NGH version 3 is consistent with that heightened sensitivity once a curve has settled.

What would prove it wrong. The same pre-onset cohort as claim 2. NGH also makes a sharp test of direction: a measured asymmetry before onset should predict which way the curve bends in the same person. No such measurement exists yet, which is why it sits in the pre-onset battery NGH proposes. It is the sharpest within-person test the hypothesis offers.

4. During the growth spurt, drift too small for a blurred sense to catch goes uncorrected, and the body's map updates toward the curve.

Verdict: Our hypothesis, not yet tested

What the outside evidence says. This is the heart of what makes NGH its own hypothesis, and it is untested. Curves do start and speed up during growth, but every model predicts that, so the timing doesn't pick a winner. What version 3 adds is one clear process: a fast loop corrects error it can detect, a slow learning loop keeps running, and growth drift below what a degraded sense can signal is absorbed, so the map becomes the curve's new “straight.” The map isn't too slow. It updates toward the curve.

What would prove it wrong. Version 3 gives a lag test with a number on it. After a new trunk shape is practiced, the felt sense of straight should take hours to months to follow. If, on average, it moves more than half the way to the new shape within 30 minutes, the slow-loop account is refuted. The lead early measure it proposes is the threshold for detecting slow trunk drift, taken before any curve appears.

5. The eyes and the jaw are among the senses that feed this map, not only position sense in the spine.

Verdict: Supported but still being studied that eye disease goes with scoliosis and that an uneven bite shifts the spine in rats. Our hypothesis, not yet tested that they act through the body's map.

What the outside evidence says. Pooled school studies find idiopathic scoliosis about three times as common in children with eye disease (Gallego-Siles and colleagues, 2024). That's an association, though a curve can't cause blindness. In rats, a small pad that raised one back tooth produced a spinal curve within a week, and rebalancing the bite was followed by realignment in most of them (D'Attilio and colleagues, 2005); that is one group's pilot study. The feet show association only, and trials that changed foot input did not change curves.

What would prove it wrong. A pre-onset eye and bite examination inside the screening cohort above. If eye and bite findings don't predict who curves, this part weakens. If they do, they point to senses that are easy to check early.

6. Once a curve exists, growth and uneven loading drive how much it progresses.

Verdict: Established science that growth and bracing affect progression. The loading mechanism in people is inferred from animal work. NGH agrees with this and does not compete with it.

What the outside evidence says. In growing rat tails, compressed vertebrae grew less and stretched ones grew more (Stokes and colleagues, 1996), and in a randomized trial bracing reduced the number of teenagers whose curves progressed to the surgical threshold (Weinstein and colleagues, 2013). This is why NGH claims the reducible part of a curve and how a curve can start, while growth sets its size. It's also why monitoring, bracing and surgery decisions stay with a scoliosis team.

What would prove it wrong. Nothing in NGH depends on this being wrong. It is the shared ground every model stands on.

7. Feedback loops in body sense, stress state and how a person feels about their body keep the pattern going.

Verdict: Supported but still being studied that position-sense deficits go with larger curves, which fits either direction. Our hypothesis, not yet tested that this forms a loop that keeps the pattern going, and for the stress-state and psychosocial loops.

What the outside evidence says. Position-sense deficits are larger in larger curves (Lau and colleagues, 2025), which fits a self-reinforcing loop and fits the curve causing the deficit just as well. The proposal that threat state and the weight of a diagnosis lower the precision of body signals is reasoned from pain and interoception science and has not been measured in scoliosis.

What would prove it wrong. NGH proposes testing whether a person's stress state and the language of their diagnosis change how accurately they sense their trunk. If neither makes a difference, those two loops go. If they do, the way a diagnosis is explained becomes part of care.

8. Part of a curve is reducible, and a practice that sharpens body sense could durably recover some of it.

Verdict: Our hypothesis, not yet tested

What the outside evidence says. Active self-correction can clearly change a curve in the moment: in one modelling study it reduced the thoracic curve by about a third, against about a quarter for a therapist's simulated manual correction (Dupuis and colleagues, 2018). How much of that is the reducible part NGH means, beyond passive flexibility, has not been measured. Body sense can be trained too: in a randomized trial in braced girls, body-awareness and core programmes improved the felt vertical (Yagci and colleagues, 2018). Durable change in the curve itself is where the trial evidence is modest. Schroth exercise added to standard care left the largest curve about 3.5 degrees smaller after six months (Schreiber and colleagues, 2016), the Cochrane review rates the exercise evidence low to very low certainty (Romano and colleagues, 2024), and in adults one exercise trial found about 4 degrees at a year, judged below a clinically meaningful change (Monticone and colleagues, 2016). Baseline trunk deficit did not predict one-year progression in one cohort (Lau and colleagues, 2026). Rival accounts have something NGH doesn't yet: experiments in animals that prevent curves or stop them progressing (Grimes and colleagues, 2016; Wang and colleagues, 2024). The open question is whether sharpening the sense changes the shape, and it's a question a trial can answer.

What would prove it wrong. Version 3 gives the number. In growing teenagers with curves under about 30 degrees, a programme that improves perception should durably recover at least a quarter of the measured self-correction gap at 12 months. If the upper bound of the result falls below a quarter, the claim fails. A second test adds perception measures to an existing brace cohort and asks whether perception at brace weaning predicts later loss.

The Generative Posture Framework

GPF's core idea is “Posture is generated, not held.” The body schema produces posture as a running prediction from whatever sensory information the nervous system has available, and the paper sets out when that prediction updates and when it doesn't. Read it at doi.org/10.5281/zenodo.19580170.

1. Posture is produced moment to moment by the brain's predictive model of the body, weighing several senses at once.

Verdict: Established science that balance weighs several senses and shifts the weighting with conditions. Supported but still being studied that posture as a whole is the output of a predictive model.

What the outside evidence says. In classic balance experiments, the body blends signals from the inner ear, vision and position sense, and reweights them as conditions change (Peterka, 2002). Predictive models of the brain treat perception and action as a running best guess (Friston, 2010). GPF applies that to posture: “Posture is generated, not held.”

What would prove it wrong. GPF proposes vibrating specific muscles during quiet standing, which blurs their signal; posture should shift in the direction precision weighting predicts. A null result, or shifts the other way, counts against it. A shift in the predicted direction sets up the measurements every later test builds on.

2. Trying hard to correct your posture turns down the very signal your body map needs to update.

Verdict: Established science that the brain dampens sensation from its own movements, more so as effort rises. Our hypothesis, not yet tested that this caps what pulling yourself upright can change in the long run.

What the outside evidence says. It's the reason you can't tickle yourself: the brain predicts the feel of its own movements and turns it down (Blakemore and colleagues, 2000), and the dampening grows with the force you put in (Shergill and colleagues, 2003). GPF proposes that high-effort correction still works in the moment, because you can produce the position on request, while the deeper update diminishes. That fits a pattern many somatic traditions report: progress during the session, then the old pattern when attention moves on. No posture study has tested this directly yet.

What would prove it wrong. If high-effort correction produced the same lasting change as low-effort practice of the same positions over the same time, this claim would be wrong. If low-effort practice lasts longer, it changes how posture should be taught.

3. Attention strongly shapes whether practice changes the brain's map of the body.

Verdict: Established science in animals for attention to touch. Our hypothesis, not yet tested for attention to the body's internal signals.

What the outside evidence says. Monkeys trained on a touch task while paying attention to it showed the trained skin's brain map grow; the same stimulation while they attended to sounds produced far less change (Recanzone and colleagues, 1992). A system deep in the forebrain that signals what matters enables that kind of map change (Kilgard and Merzenich, 1998). GPF extends this from touch to attention inside the body, which hasn't been tested.

What would prove it wrong. GPF proposes a test in which a distracting task withdraws attention from the body during a posture aftereffect. Each version of the theory predicts a different pattern as attention is withdrawn, so the result picks between them.

4. Feeling safe lets body signals count for more when the map updates.

Verdict: Our hypothesis, not yet tested

What the outside evidence says. Threat clearly changes how the body balances: under height-induced fear, spinal reflexes became more responsive while brain responses to position signals stayed the same (Davis and colleagues, 2011). How that maps onto learning a posture is the proposal. GPF uses polyvagal terms only as a clinical description of feeling safe; an international expert group has judged polyvagal theory untenable, including its anatomy (Grossman and colleagues, 2026), and GPF's claim doesn't rest on that anatomy.

What would prove it wrong. GPF's crossover test: the same practice session on days of different measured stress state. If calmer baseline days don't produce larger changes in body-map measures, the claim is wrong. If they do, it supports starting practice from a settled state.

5. The body's map learns two ways: by correcting clear errors, and by drifting toward whatever posture you spend your time in.

Verdict: Established science over minutes, in reaching and in the felt vertical. Our hypothesis, not yet tested over months, in posture.

What the outside evidence says. Movements drift toward recent movements even when no error had to be corrected (Diedrichsen and colleagues, 2010), and five minutes of passive whole-body tilt biased healthy people's felt vertical by about 3 degrees toward the tilt (Wedtgrube and colleagues, 2020). GPF proposes the same drift, over months, is how an uncorrected deviation becomes the new default.

What would prove it wrong. Withdraw attention while people are tilted. The two-route account predicts the aftereffect falls to a floor that never reaches zero; a single attention-gated route predicts it falls to zero.

6. Chronic back pain, kyphosis, scoliosis and other stuck postural patterns are versions of one system.

Verdict: Our hypothesis, not yet tested GPF rates its own unified framework this way.

What the outside evidence says. Some pain researchers make a similar argument. The fit-for-purpose model treats chronic low back pain as an information problem (Wand and colleagues, 2023), and in a randomized trial graded sensorimotor retraining reduced pain by about 1 point out of 10, a small effect right at the trial's pre-set threshold for clinical importance (Bagg and colleagues, 2022). The paper's own rule: the whole is less certain than any of its parts, so test it principle by principle. It does not apply to congenital malformations or fractures.

What would prove it wrong. If the predictions fail one by one, the unified picture goes with them. Because each principle stands or falls on its own prediction, the picture can be tested piece by piece, starting with the cheapest studies.

7. Practicing live with other people amplifies change.

Verdict: Our hypothesis, not yet tested

What the outside evidence says. GPF treats live, co-regulated practice as an amplifier: helpful, not required. Solo practice traditions show change can happen alone. Posture Dojo runs live calls because we're betting this is right, and we label it as a bet.

What would prove it wrong. GPF's three-arm trial: live group practice, the same teacher on video, and written instructions, with equal practice time. If the live arm doesn't come out ahead, the amplifier claim fails. If it does, it tells us something useful about why practicing together works.

Where this stands, and why it's worth watching

Read straight through, the pattern is clear. The papers stand on established ground at both ends: inherited risk, growth-driven progression, how the brain weighs the senses, how effort dampens feeling, and, in animal touch studies, how attention gates map change. The animal evidence for the sensory starting point is real. The middle steps, the ones that make these ideas new, are labelled as ours and are waiting for tests.

That's a good place for a hypothesis to be. NGH is one of several candidates, alongside accounts built on spinal circuits (Wang and colleagues, 2024), spinal fluid flow (Grimes and colleagues, 2016) and the curve reshaping the body's sensing. The studies that separate them can be designed now, and some can use data that already exist. Version 3 sharpened NGH toward exactly that: one clear process, numbers on its tests, credit to the researchers who came before, and a rebuilt, checked reference list. Every study should be pre-registered and run by groups independent of Posture Dojo. That's the test we want.

In the meantime, the practice side is something you can try for yourself. The distinctions behind the method explain what changes when you work on the map instead of the shape, and how to practice posture shows what that looks like day to day.

Questions about the fact check

Sources

The two papers: Miller S (2026). The Neural Generation Hypothesis of Adolescent Idiopathic Scoliosis: Converging Evidence from Genetics, Proprioception, and Predictive Coding. Zenodo, preprint, doi.org/10.5281/zenodo.19342098 (version 3.0: doi.org/10.5281/zenodo.23685652; SSRN: doi.org/10.2139/ssrn.6493379). Miller S (2026). The Generative Posture Framework: A Unified Account of Postural Generation, Persistence, and Change. Zenodo, preprint, doi.org/10.5281/zenodo.19580170 (version 2.5: doi.org/10.5281/zenodo.23685591; SSRN: doi.org/10.2139/ssrn.6577078).

The scoping review in the questions above: Cordani C, et al. (2026). Neurophysiological features of adolescent idiopathic scoliosis: a scoping review across central, peripheral and sensorimotor domains. SOSORT 2026 conference abstract 206 (not peer reviewed).

1. Takahashi Y, et al. (2011). A genome-wide association study identifies common variants near LBX1 associated with adolescent idiopathic scoliosis. Nature Genetics 43:1237-1240. doi.org/10.1038/ng.974
2. Londono D, et al. (2014). A meta-analysis identifies adolescent idiopathic scoliosis association with LBX1 locus in multiple ethnic groups. Journal of Medical Genetics 51:401-406. doi.org/10.1136/jmedgenet-2013-102067
3. Blecher R, et al. (2017). The proprioceptive system masterminds spinal alignment: insight into the mechanism of scoliosis. Developmental Cell 42:388-399. doi.org/10.1016/j.devcel.2017.07.022
4. Assaraf E, et al. (2020). Piezo2 expressed in proprioceptive neurons is essential for skeletal integrity. Nature Communications 11:3168. doi.org/10.1038/s41467-020-16971-6
5. Pincott JR, et al. (1984). Scoliosis caused by section of dorsal spinal nerve roots. Journal of Bone and Joint Surgery (British) 66:27-29. doi.org/10.1302/0301-620X.66B1.6693473
6. Lau KKL, et al. (2022). Timely revisit of proprioceptive deficits in adolescent idiopathic scoliosis: a systematic review and meta-analysis. Global Spine Journal 12:1852-1861. doi.org/10.1177/21925682211066824
7. Lau KKL, et al. (2025). A substantial proportion of subjects with adolescent idiopathic scoliosis display spinal and peripheral proprioceptive deficits. Bone and Joint Open 6:1164-1170. doi.org/10.1302/2633-1462.610.BJO-2025-0140.R1
8. Lau KKL, et al. (2026). Association of proprioceptive deficits with curve progression and treatment outcomes in subjects with adolescent idiopathic scoliosis: a prospective cohort study with 1 year of follow-up. North American Spine Society Journal 27:100899. doi.org/10.1016/j.xnsj.2026.100899
9. Pialasse JP, et al. (2015). Sensory reweighting is altered in adolescent patients with scoliosis: evidence from a neuromechanical model. Gait and Posture 42:558-563. doi.org/10.1016/j.gaitpost.2015.08.013
10. Le Berre M, et al. (2019). Do adolescents with idiopathic scoliosis have an erroneous perception of the gravitational vertical? Spine Deformity 7:71-79. doi.org/10.1016/j.jspd.2018.05.004
11. Virovec M, et al. (2026). Motor cortex excitability during spine shape-judgment in adolescent idiopathic scoliosis: a TMS motor evoked potential study. Experimental Brain Research 244. doi.org/10.1007/s00221-026-07343-5
12. Schlösser TP, et al. (2014). How 'idiopathic' is adolescent idiopathic scoliosis? A systematic review on associated abnormalities. PLoS ONE 9:e97461. doi.org/10.1371/journal.pone.0097461
13. Gallego-Siles JR, et al. (2024). Idiopathic scoliosis in subjects with eye diseases: a systematic review with meta-analysis. Annals of the New York Academy of Sciences 1533:81-88. doi.org/10.1111/nyas.15102
14. D'Attilio M, et al. (2005). The influence of an experimentally-induced malocclusion on vertebral alignment in rats: a controlled pilot study. Cranio 23:119-129. doi.org/10.1179/crn.2005.017
15. Stokes IA, et al. (1996). Mechanical modulation of vertebral body growth: implications for scoliosis progression. Spine 21:1162-1167. doi.org/10.1097/00007632-199605150-00007
16. Weinstein SL, et al. (2013). Effects of bracing in adolescents with idiopathic scoliosis. New England Journal of Medicine 369:1512-1521. doi.org/10.1056/NEJMoa1307337
17. Dupuis S, et al. (2018). Global postural re-education in pediatric idiopathic scoliosis: a biomechanical modeling and analysis of curve reduction during active and assisted self-correction. BMC Musculoskeletal Disorders 19:200. doi.org/10.1186/s12891-018-2112-9
18. Schreiber S, et al. (2016). Schroth physiotherapeutic scoliosis-specific exercises added to the standard of care lead to better Cobb angle outcomes in adolescents with idiopathic scoliosis. PLoS ONE 11:e0168746. doi.org/10.1371/journal.pone.0168746
19. Romano M, et al. (2024). Therapeutic exercises for idiopathic scoliosis in adolescents. Cochrane Database of Systematic Reviews CD007837.pub3. doi.org/10.1002/14651858.CD007837.pub3
20. Monticone M, et al. (2016). Adults with idiopathic scoliosis improve disability after motor and cognitive rehabilitation: results of a randomised controlled trial. European Spine Journal 25:3120-3129. doi.org/10.1007/s00586-016-4528-y
21. Yagci G, et al. (2018). The effects of exercise on perception of verticality in adolescent idiopathic scoliosis. Physiotherapy Theory and Practice 34:579-588. doi.org/10.1080/09593985.2017.1423429
22. Grimes DT, et al. (2016). Zebrafish models of idiopathic scoliosis link cerebrospinal fluid flow defects to spine curvature. Science 352:1341-1344. doi.org/10.1126/science.aaf6419
23. Wang X, et al. (2024). Impaired glycine neurotransmission causes adolescent idiopathic scoliosis. Journal of Clinical Investigation 134. doi.org/10.1172/JCI168783
24. Herman R, et al. (1985). Idiopathic scoliosis and the central nervous system: a motor control problem. Spine 10:1-14. doi.org/10.1097/00007632-198501000-00001
25. Burwell RG, et al. (2009). Pathogenesis of adolescent idiopathic scoliosis in girls: a double neuro-osseous theory. Scoliosis 4:24. doi.org/10.1186/1748-7161-4-24
26. Peterka RJ (2002). Sensorimotor integration in human postural control. Journal of Neurophysiology 88:1097-1118. doi.org/10.1152/jn.2002.88.3.1097
27. Friston K (2010). The free-energy principle: a unified brain theory? Nature Reviews Neuroscience 11:127-138. doi.org/10.1038/nrn2787
28. Blakemore SJ, Wolpert D, Frith C (2000). Why can't you tickle yourself? NeuroReport 11:R11-R16. doi.org/10.1097/00001756-200008030-00002
29. Shergill SS, Bays PM, Frith CD, Wolpert DM (2003). Two eyes for an eye: the neuroscience of force escalation. Science 301:187. doi.org/10.1126/science.1085327
30. Recanzone GH, et al. (1992). Topographic reorganization of the hand representation in cortical area 3b of owl monkeys trained in a frequency-discrimination task. Journal of Neurophysiology 67:1031-1056. doi.org/10.1152/jn.1992.67.5.1031
31. Kilgard MP, Merzenich MM (1998). Cortical map reorganization enabled by nucleus basalis activity. Science 279:1714-1718. doi.org/10.1126/science.279.5357.1714
32. Davis JR, et al. (2011). Human proprioceptive adaptations during states of height-induced fear and anxiety. Journal of Neurophysiology 106:3082-3090. doi.org/10.1152/jn.01030.2010
33. Grossman P, et al. (2026). Why the polyvagal theory is untenable: an international expert evaluation of the polyvagal theory. Clinical Neuropsychiatry 23:100-112. doi.org/10.36131/cnfioritieditore20260110
34. Diedrichsen J, White O, Newman D, Lally N (2010). Use-dependent and error-based learning of motor behaviors. Journal of Neuroscience 30:5159-5166. doi.org/10.1523/JNEUROSCI.5406-09.2010
35. Wedtgrube A, et al. (2020). Prolonged static whole-body roll-tilt and optokinetic stimulation significantly bias the subjective postural vertical in healthy human subjects. Frontiers in Neurology 11:595975. doi.org/10.3389/fneur.2020.595975
36. Wand BM, et al. (2023). The fit-for-purpose model: conceptualizing and managing chronic nonspecific low back pain as an information problem. Physical Therapy 103:pzac151. doi.org/10.1093/ptj/pzac151
37. Bagg MK, et al. (2022). Effect of graded sensorimotor retraining on pain intensity in patients with chronic low back pain: a randomized clinical trial. JAMA 328:430-439. doi.org/10.1001/jama.2022.9930