What the Research Says About Glucosamine and Chondroitin for Knee Osteoarthritis

We only cite studies published in peer-reviewed journals. We summarize findings without overstating conclusions.

This article summarizes a prospective observational study published in March 2015 in Arthritis and Rheumatology – the flagship journal of the American College of Rheumatology – under the title “Effects of Glucosamine and Chondroitin Supplementation on Knee Osteoarthritis: An Analysis With Marginal Structural Models.” The authors were Shibing Yang (Department of Family Medicine and Population Health, Division of Epidemiology, Virginia Commonwealth University, Richmond, Virginia), Charles B. Eaton (Memorial Hospital of Rhode Island, Pawtucket, and Brown University, Providence, Rhode Island), Timothy E. McAlindon (Department of Rheumatology, Tufts Medical Center, Boston, Massachusetts), and Kate L. Lapane (Department of Quantitative Health Sciences, University of Massachusetts Medical School, Worcester, Massachusetts). The study drew on four years of longitudinal data from the Osteoarthritis Initiative (OAI), a large prospective cohort of adults with or at risk for knee osteoarthritis, and analyzed 1,625 participants who were not using glucosamine or chondroitin at the study’s start. Its defining methodological feature – and the source of its particular scientific value – was the use of marginal structural models, an advanced causal inference technique designed to handle the specific statistical complexities of studying a treatment that people start and stop over time. The full text is freely available at PubMed Central (PMC4342281).

Funding and conflict of interest disclosure: The study was supported by a grant from the National Institutes of Health (National Heart, Lung, and Blood Institute contract no. HHSN-268201000020C). The OAI dataset itself is a public-private partnership between the NIH and four pharmaceutical companies – Merck Research Laboratories, Novartis Pharmaceuticals, GlaxoSmithKline, and Pfizer, Inc. – though private sector funding for the OAI is managed by the Foundation for the NIH and the authors explicitly note that their analysis does not necessarily reflect the opinions or views of the OAI investigators, the NIH, or the private funding partners. Dr. Eaton is a principal investigator of an OAI clinical site, and Dr. McAlindon is a co-principal investigator of the same site. Dr. McAlindon additionally disclosed consulting fees and honoraria from Flexion Therapeutics, Sanofi-Aventis, and Samumed (each under $10,000), and is listed as inventor on a patent for a method of conducting clinical trials over the internet. None of these relationships involve glucosamine or chondroitin manufacturers. No author on this study is commercially connected to the supplements under investigation.

Background: The Most Widely Used Joint Supplements in America

Glucosamine and chondroitin are among the best-selling dietary supplements in the United States and across the developed world. Survey data cited by the study’s authors indicate that approximately one-third of Americans with osteoarthritis regularly take one or both of these compounds for their joint symptoms. In absolute terms, this translates to millions of people taking these supplements daily, at a substantial collective cost, based on a reasonable hope that the supplements will reduce their pain, improve their mobility, or slow the cartilage loss that drives their condition’s progression.

Glucosamine is an amino sugar – a molecule that is both a sugar and contains an amino group – that occurs naturally in the human body as a building block of glycosaminoglycans, the long carbohydrate chains that give cartilage, synovial fluid, and connective tissue their structural integrity and cushioning properties. As a supplement it is typically derived from the shells of crustaceans (shrimp, crab, lobster) and is sold most commonly as glucosamine sulfate or glucosamine hydrochloride. Chondroitin is itself a glycosaminoglycan – a major structural component of cartilage that helps it retain water and resist compression. Supplemental chondroitin is usually derived from bovine or shark cartilage and is most often sold as chondroitin sulfate. Both compounds are taken orally, and both are theorized to support joint health either by providing raw materials for cartilage repair and maintenance, or by directly suppressing the inflammatory and enzymatic processes that break cartilage down.

Despite their popularity and their plausible biological rationale, the clinical trial evidence for these supplements has been contentious for decades. The landmark GAIT trial (Glucosamine/Chondroitin Arthritis Intervention Trial), funded by the National Institutes of Health and published in 2006, found that neither glucosamine hydrochloride nor chondroitin sulfate, alone or in combination, significantly reduced knee OA pain compared to placebo in the overall population – though a subgroup analysis suggested a possible benefit for patients with moderate-to-severe pain. Subsequent large high-quality trials – including a two-year GAIT follow-up and several European studies – consistently produced null or near-null results, while earlier and typically smaller industry-funded trials had more often reported benefit. By the time Yang and colleagues published their 2015 analysis, multiple expert guidelines – including those of the American College of Rheumatology and the European League Against Rheumatism – had moved to recommend against glucosamine and chondroitin supplementation in knee OA, or rated the evidence as low quality.

The specific contribution of the Yang et al. study was not to run yet another clinical trial, but to take a complementary methodological approach: analyzing a large, prospectively followed cohort of real-world OA patients using a statistical framework – marginal structural models – sophisticated enough to produce a credible causal estimate from observational data. The researchers asked whether, in actual practice in the real world, the people who chose to take glucosamine and chondroitin fared any differently than those who did not.

The Biology: What Glucosamine and Chondroitin Are Supposed to Do

Osteoarthritis is a disease of the entire joint – not just the cartilage, though cartilage loss is its most recognizable feature. In a healthy joint, articular cartilage (the smooth, slippery tissue covering the ends of bones where they meet) provides near-frictionless movement and cushions the impact of load-bearing activities. Cartilage has no blood supply or nerve supply of its own; its cells, called chondrocytes, survive by diffusion of nutrients from the synovial fluid that bathes the joint. In osteoarthritis, the equilibrium between cartilage synthesis and cartilage breakdown is disrupted. Chondrocytes begin producing more degradative enzymes (particularly matrix metalloproteinases) than structural proteins; the extracellular matrix that gives cartilage its strength – composed of collagen and glycosaminoglycans – progressively erodes; and the joint space visible on X-ray narrows as cartilage thins.

Glycosaminoglycans – the class of molecules to which chondroitin belongs, and which glucosamine helps synthesize – are critical to cartilage’s mechanical properties. They attract and hold water molecules through their negative electrical charges, giving cartilage its compressive stiffness and the ability to spring back after load. As OA progresses, glycosaminoglycan content falls, cartilage loses water, becomes stiffer and more brittle, and its surface develops the fibrillation and fissuring visible under a microscope.

The theoretical case for glucosamine supplementation rests on the idea that providing extra glucosamine – the precursor molecule from which the body synthesizes glycosaminoglycans – might upregulate cartilage matrix synthesis or slow its breakdown. Laboratory studies have shown that glucosamine can stimulate chondrocytes to produce more collagen and glycosaminoglycans and can suppress certain inflammatory signaling pathways. The theoretical case for chondroitin rests partly on similar substrate-provision logic (providing the glycosaminoglycan directly), and partly on in vitro evidence that chondroitin can inhibit some of the enzymes that degrade cartilage matrix.

The gap between these in vitro and mechanistic findings and what actually happens in a living human taking oral supplements is, however, substantial. Both glucosamine and chondroitin face real questions about whether they are absorbed in sufficient quantities to reach cartilage at therapeutically relevant concentrations after oral dosing. Chondroitin in particular is a very large molecule that was long assumed to be poorly absorbed intact from the gut, though more recent research has complicated this picture. Even if adequate amounts reach the joint, whether they exert the same effects in the complex, inflamed, mechanically loaded environment of an arthritic joint as they do in a laboratory cell culture is not guaranteed. These are precisely the questions that clinical trials and well-designed observational studies attempt to answer.

The Osteoarthritis Initiative: A Uniquely Rich Data Source

The Osteoarthritis Initiative (OAI) is a multi-center, prospective longitudinal study of adults aged 45 to 79 years at enrollment, designed to identify and validate biomarkers and risk factors for the onset and progression of knee osteoarthritis. It enrolled approximately 4,796 participants across four U.S. clinical sites (Baltimore, Maryland; Columbus, Ohio; Pittsburgh, Pennsylvania; and Pawtucket, Rhode Island) beginning in 2004. The OAI is particularly valuable to researchers because it collected extraordinarily detailed annual assessments of each participant: standardized clinical examinations, validated questionnaires on symptoms and quality of life, radiographs (X-rays) and magnetic resonance images (MRIs) of the knees, blood and urine samples, and extensive surveys of all medications and supplements the participants were taking. All of this data was made publicly available to qualified researchers – an unusual level of data sharing in biomedical research – which is why the OAI dataset has been used as the basis for hundreds of scientific analyses.

For Yang and colleagues, the OAI’s annual supplement-use surveys were the key feature: participants reported at every annual visit which supplements they were taking and how often. This created a detailed longitudinal record of who was taking glucosamine and/or chondroitin, when they started, whether they continued or stopped, and how their OA outcomes evolved over the same period – all in a large real-world population rather than the highly selected populations of randomized trials.

Methodology: The New-User Design and the Problem of Time-Varying Confounding

The New-User Design

A standard challenge in studying supplement or medication use with observational data is what epidemiologists call “prevalent user bias” or “immortal time bias.” If you simply compare people currently taking a supplement to those not taking it at a single point in time, you bias your sample in multiple ways: longtime users who had bad outcomes may have already stopped; people who started and immediately felt worse may have dropped out; and the health status of people at the moment they decide to start a supplement may systematically differ from those who never start. These biases can make a supplement look more or less effective than it actually is.

The authors addressed this with a “new-user” design: they restricted their study sample to the 1,625 OAI participants who were not taking glucosamine or chondroitin at the time of enrollment (baseline). By following only people from the moment before they had any exposure to the supplements, the researchers ensured that everyone began the analysis on equal footing – no one had preselected themselves into the user category by prior good or bad experiences. During the four years of follow-up, 280 participants (18 percent) initiated treatment with glucosamine and/or chondroitin. The remaining 1,345 participants (82 percent) never used the supplements during the study period. Cumulative exposure was calculated as the number of annual assessment visits at which a participant reported using the supplements, allowing the authors to compare outcomes after one, two, or three years of cumulative use.

The Problem of Time-Varying Confounding

Even with a new-user design, observational studies of supplement use face a deep statistical challenge: confounding by indication. People who decide to start taking glucosamine and chondroitin are not randomly selected – they are typically people whose knee pain has recently worsened, or who have been advised by a healthcare provider to try supplements, or who have responded to advertising. In other words, the very factors that predict supplement initiation (worsening symptoms) are also the factors that predict future outcomes (continued worsening). A naive analysis that simply compares users to non-users will therefore be biased: the users will look like they are doing worse, not because the supplement harmed them, but because they started taking it precisely because they were already doing worse.

This problem is compounded when, as in this study, a participant’s health status changes over time, affecting both the decision to continue or stop using supplements and their measured outcomes at each subsequent visit. In statistical terminology, a variable that simultaneously confounds the treatment-outcome relationship and is itself affected by prior treatment is called a “time-varying confounder.” Standard statistical regression methods cannot correctly adjust for time-varying confounders; doing so with conventional techniques can actually introduce new biases rather than removing the original ones.

Marginal Structural Models: The Solution

Marginal structural models (MSMs) are a class of causal inference methods developed specifically to handle time-varying confounding in longitudinal observational data. The core idea is conceptually elegant: rather than trying to directly model the complex web of relationships between time-varying confounders, treatment, and outcomes, MSMs create a statistically reweighted version of the study population in which the confounders no longer predict treatment decisions. In the reweighted population, it is as if participants were randomly assigned to take or not take the supplements – mimicking, as closely as possible, the conditions of a randomized controlled trial, while using real-world observational data.

The reweighting is accomplished through a technique called inverse probability weighting (IPW). At each annual visit, the researchers calculated the probability that each participant would make the supplement decision they actually made (start, continue, or not take) given their current values on all the known confounders. Participants who made an unlikely choice – for example, someone with very severe symptoms who, unexpectedly, did not start taking supplements – received a higher statistical weight, effectively making them count more in the analysis to compensate for how unusual their behavior was. Participants who made the expected choice received a lower weight. When all these weights are applied, the resulting weighted population has the confounders “balanced out,” and the estimated treatment effect reflects the supplement’s genuine causal impact rather than the selection biases that plague unweighted comparisons.

The time-varying confounders that the authors adjusted for at each annual visit were: the relevant WOMAC subscale score (pain, stiffness, or physical function depending on the outcome being analyzed), the Kellgren-Lawrence (K-L) grade (a radiographic scale of OA severity from 0 to 4 based on X-ray features), both subscales of the SF-12 health survey (a validated measure of physical and mental health-related quality of life), body mass index (BMI), knee alignment (whether the participant had varus or valgus malalignment, which affects load distribution across the joint), prior history of knee injury, current use of analgesic medications, and current use of other complementary and alternative medicines. All of these variables were measured annually and used to update the inverse probability weights at each time point – making this one of the more methodologically rigorous analyses of supplement use in OA available to that date.

Outcome Measures

Knee symptoms were measured using the WOMAC (Western Ontario and McMaster Universities Osteoarthritis Index), a validated questionnaire that assesses three distinct domains: pain (5 questions, scored 0–20), stiffness (2 questions, scored 0–8), and physical function (17 questions, scored 0–68). Higher scores indicate worse symptoms. Structural disease progression was assessed by measuring joint space width (JSW) – the gap between the bone ends visible on a standard knee X-ray, which narrows as cartilage is lost. JSW measurements were made in the medial compartment of the tibiofemoral joint, where OA-related narrowing is most clinically significant. The study estimated the effect of glucosamine/chondroitin use at three separate cumulative durations: one year, two years, and three years of use.

Results: No Meaningful Benefit Detected at Any Duration

Who Took the Supplements

Over the four years of follow-up, 280 of the 1,625 eligible participants (18 percent) initiated glucosamine and/or chondroitin use. Of these, 148 (9 percent of the total sample) used the supplements at one annual assessment only – suggesting either a trial period after which they discontinued, or sporadic use. Only 65 participants (4 percent of the total) were consistent users who reported taking the supplements at all annual assessments. This persistence pattern is itself informative: most people who tried the supplements did not continue them through the full follow-up, which may reflect a lack of perceived benefit in real-world practice.

Primary Results: WOMAC Symptoms

After applying marginal structural model adjustments for all time-varying confounders, the analysis found no clinically or statistically meaningful differences between glucosamine/chondroitin users and never-users on any WOMAC subscale, at any duration of use.

For WOMAC pain, comparing users at all assessments (cumulative three-year use) to never-users, the estimated effect was a beta coefficient of 0.68 (95% CI: −0.16 to 1.53). On the WOMAC pain scale, which runs from 0 to 20, a difference of 0.68 points is negligible – well below any threshold of clinical meaningfulness – and the confidence interval crosses zero, meaning the result is not statistically significant. A positive beta coefficient indicates, if anything, a marginal numerical signal in the direction of more pain in users, though this is within sampling error and should not be interpreted as evidence of harm.

For WOMAC stiffness, the estimated effect was a beta of 0.41 (95% CI: 0.00 to 0.82). The WOMAC stiffness subscale runs from 0 to 8; a difference of 0.41 points is clinically insignificant. The confidence interval just barely excludes zero, technically achieving statistical significance, but the upper bound of 0.82 on an 8-point scale represents no meaningful change in stiffness either practically or clinically. The authors describe this result as not clinically significant, and that assessment is well-supported.

For WOMAC physical function, the estimated effect was a beta of 1.28 (95% CI: −1.23 to 3.79). The physical function subscale runs from 0 to 68; a difference of 1.28 points in either direction is imperceptible to patients and below any published threshold for minimum clinically important difference (MCID). The confidence interval crosses zero, making the result statistically non-significant as well.

Structural Progression: Joint Space Width

For joint space width – the X-ray measure of structural disease progression – the estimated effect of glucosamine/chondroitin use was a beta of 0.11 mm (95% CI: −0.21 to 0.44). On a measure where clinically significant narrowing is typically defined as 0.3 to 0.5 mm or more over a year, an estimated difference of 0.11 mm – again with a confidence interval crossing zero – does not approach clinical significance. The supplements showed no detectable ability to slow the structural progression of knee osteoarthritis as measured by medial joint space width.

Consistency Across Durations

The authors also analyzed outcomes separately for participants who used the supplements for one year and two years (in addition to the primary three-year comparison). In no duration subgroup – one year, two years, or three years of cumulative use – did glucosamine and chondroitin produce a clinically or statistically significant improvement in any of the four outcomes. The consistency of null findings across durations reinforces the conclusion that the supplements did not provide meaningful benefit in this population, regardless of how long they were taken within the four-year study window.

Why This Study Matters: What It Adds Beyond Prior Trials

The null findings in this study are consistent with the weight of randomized controlled trial evidence, and the authors make this connection explicitly. But the study contributes something the trial literature alone cannot provide: external validity in a real-world population.

Randomized controlled trials, by design, enroll carefully selected participants who meet narrow inclusion and exclusion criteria, receive a standardized product at a controlled dose, and are monitored far more intensively than typical supplement users. These features maximize internal validity – the ability to attribute any observed effect to the intervention – but limit generalizability. OA patients in trials may be more compliant, more closely supervised, and more systematically selected than the broader population of people who purchase glucosamine and chondroitin at a pharmacy. As a consequence, even perfectly conducted trials may not accurately predict what happens when the same supplement is taken by an unselected, heterogeneous, real-world population.

The OAI analysis addressed this gap. Its participants were community-dwelling adults with confirmed knee OA – the actual population that takes these supplements. Their supplement use was self-reported and naturalistic rather than supervised. And the marginal structural modeling approach provided a methodologically sound way to account for the selection biases that make raw observational comparisons unreliable. The fact that this more ecologically valid analysis produced the same null result as the controlled trials is an important convergence: it suggests the lack of benefit seen in trials was not an artifact of trial populations, dosing protocols, or duration, but reflects a genuine absence of clinically meaningful effect in real-world use.

The authors themselves underscore this point: their findings, they write, are consistent with the results of meta-analyses of clinical trials and “extend those results to a more general population with knee OA.” In short, the study’s message is not that glucosamine and chondroitin fail only under the artificial conditions of a clinical trial – they appear to fail equally in the real world, under the conditions in which millions of people actually take them.

Post-Publication Debate: Was the Effect Underestimated?

Shortly after publication, a letter to the editor by Zeng, Wei, and Lei (published in Arthritis and Rheumatology in July 2015) raised a methodological challenge: the authors suggested that the Yang et al. analysis should have further restricted its sample to participants who were not using analgesic medications, arguing that analgesic use was an inadequately handled confounder that may have diluted any detectable benefit from the supplements. Their core concern was that if both supplement users and non-users were managing pain with analgesics, the analgesics might have obscured the supplements’ contribution, making a genuinely beneficial supplement appear ineffective.

In their published reply, Yang and colleagues acknowledged analgesic use as a time-varying confounder and clarified that it was in fact included in their marginal structural model adjustments at each annual visit – meaning the analysis already accounted for concurrent analgesic use in its reweighting procedure. The authors noted that restricting the sample to non-analgesic users, as Zeng and colleagues suggested, would have substantially reduced the sample size, reduced generalizability, and potentially introduced new selection biases by excluding patients with more severe disease. They maintained that their analytical approach handled the analgesic confounding appropriately and that the null findings were robust.

This exchange is worth understanding because it reflects a genuine and recurring tension in supplement research: when a well-powered study finds no effect, critics may argue that some inadequately handled variable is masking a real benefit. These concerns are scientifically legitimate to raise, but they can also become unfalsifiable if the response to every null result is to propose a different analytic restriction that might produce a positive finding. The authors’ decision to include analgesic use as a time-varying confounder rather than an exclusion criterion reflects a reasonable and defensible methodological choice, and the debate did not alter the study’s primary conclusions.

The Limitations: What the Study Cannot Tell Us

The authors are appropriately forthcoming about the boundaries of their conclusions.

The most important limitation is residual confounding. Marginal structural models are powerful, but they can only adjust for measured confounders. If there are unmeasured factors that simultaneously predicted supplement use and OA outcomes – for example, diet quality, physical activity patterns more granular than what the OAI captured, genetic variation in supplement metabolism, or specific beliefs and behaviors correlated with supplement use – these unmeasured variables could still bias the results. No observational study, no matter how methodologically sophisticated, can fully replicate the counterfactual control of a randomized experiment.

The study could not assess glucosamine and chondroitin separately. The vast majority of participants who used these supplements took them in combination, reflecting the way the products are typically sold and used. Whether one supplement alone might perform differently from the other, or from the combination, cannot be determined from these data.

The study also could not assess the specific formulation of glucosamine used. Glucosamine sulfate and glucosamine hydrochloride – the two most common forms – may have different bioavailabilities, and some researchers have argued that trials of glucosamine hydrochloride (as used in the GAIT trial) may underestimate the benefit of the sulfate form. The OAI’s supplement-use surveys captured whether participants were taking glucosamine and chondroitin, but not which precise formulation.

The follow-up period of four years, while considerably longer than most clinical trials, may still be insufficient to detect structural benefits that develop over a longer time horizon. Some meta-analyses have found that the effect of glucosamine on joint space narrowing appeared stronger in trials with longer intervention periods. Four years may be too short to see whatever structural benefit – if any – these supplements provide, and the four-year window captures only a fraction of the natural history of a chronic disease that progresses over decades.

Finally, the study’s participants were all radiographically confirmed OA cases from the OAI, which was designed to study disease progression – not the mildest end of the OA spectrum. Patients with very early OA, or those at risk but without established disease, may respond differently. The GAIT trial’s suggestion that patients with moderate-to-severe baseline pain might derive more benefit than those with mild pain has never been definitively resolved, and this study does not address disease-severity subgroups.

Summary of Key Takeaways

• This study analyzed four years of longitudinal data from 1,625 real-world knee OA patients drawn from the Osteoarthritis Initiative who were supplement-naive at baseline, making it one of the most methodologically rigorous observational studies of glucosamine and chondroitin use in OA published to date.
• During the study period, 18 percent of participants (280 of 1,625) initiated glucosamine and/or chondroitin use. Of those, only 4 percent of the total sample used the supplements consistently at every annual assessment, suggesting most real-world users discontinue the supplements within a year – possibly due to perceived lack of benefit.
• After full adjustment for time-varying confounders using marginal structural models, glucosamine and chondroitin supplementation produced no clinically or statistically meaningful improvement in WOMAC pain (β = 0.68; 95% CI: −0.16 to 1.53), WOMAC physical function (β = 1.28; 95% CI: −1.23 to 3.79), or joint space width (β = 0.11 mm; 95% CI: −0.21 to 0.44).
• A technically statistically significant but clinically negligible difference was found for WOMAC stiffness (β = 0.41; 95% CI: 0.00 to 0.82) – a difference of less than half a point on an 8-point scale that the authors explicitly describe as not clinically meaningful.
• Null findings were consistent across all durations of supplement use analyzed – one year, two years, and three years of cumulative use – providing no evidence that longer supplementation produces greater benefit within a four-year window.
• The study’s principal scientific contribution is not a new finding of inefficacy (the controlled trial literature had already suggested this) but rather the demonstration that the same null result holds in an ecologically valid real-world population, using a causal inference method capable of handling the statistical complexities of naturalistic supplement use.
• A published methodological challenge from Zeng, Wei, and Lei argued that inadequate handling of analgesic use may have masked benefit; the authors responded that analgesic use was explicitly included as a time-varying confounder in their marginal structural model, and maintained that their conclusions were robust.
• The study cannot address whether specific formulations (glucosamine sulfate versus hydrochloride), specific patient subgroups (moderate-to-severe pain, very early disease), or longer follow-up periods beyond four years might yield different results – these remain open questions in the literature.
• The authors’ conclusion – that glucosamine and chondroitin use did not appear to relieve symptoms or modify disease progression in patients with radiographically confirmed knee OA – is consistent with current clinical guidelines from the American College of Rheumatology and other major rheumatology organizations, which conditionally recommend against routine use of these supplements in knee OA.

Yang, Shibing, Charles B. Eaton, Timothy E. McAlindon, and Kate L. Lapane. “Effects of Glucosamine and Chondroitin Supplementation on Knee Osteoarthritis: An Analysis With Marginal Structural Models.” Arthritis and Rheumatology, vol. 67, no. 3, 2015, pp. 714–723. https://doi.org/10.1002/art.38932. Full text available at: https://pmc.ncbi.nlm.nih.gov/articles/PMC4342281/.