Arthritis and psychological conditions (i.e. mental health conditions, for example, depression and anxiety) were in the top three most prevalent chronic conditions experienced in Australia during 2020–20211. Osteoarthritis (OA) is a degenerative joint condition where cartilage that cushions the ends of the bones deteriorates. It is the most common form of arthritis and most commonly affects hands, knees, hips and lower back and neck. Depression and anxiety are both common psychological co-morbidities that occur alongside OA2,3 in addition to increased pain and other mental health issues4. Chronic pain is experienced by one in five Australians aged 45 and over5.

OA has also been associated with increased dementia risk6,7, higher prevalence rates of dementia8, and is considered an early risk factor for cognitive decline7. Depression is another risk factor for dementia9 and is a common psychological symptom associated with dementia10. People living with dementia often experience other behavioural changes and psychological symptoms (e.g. agitation) that may be associated with underlying pain that is not always identified by care staff due to communication difficulties11. In Australia, dementia is the leading cause of death in women and second leading cause of death following heart disease in all Australians1,12. High rates of OA and dementia are also observed globally13.

There are no medications that can cure OA, though analgesics, non-steroidal anti-inflammatory drugs (NSAIDs) and topical therapies are often administered to relieve pain and inflammation. However, all medicines have side effects, for example, NSAIDS can increase risk of stomach ulcers and bleeding14, paracetamol can have adverse side effects when taken long-term15, and the use of opioids is controversial, with increasing evidence showing an increase in addiction, morbidity, and mortality16. Therefore, effective nonpharmacological interventions are needed to address pain and depression in OA, which are all well-established precursors to dementia. No previous systematic reviews have sought to determine the most effective approaches for reducing pain levels and symptoms of depression in OA. Reducing these co-morbidities will reduce dementia risk.

Physical and psychological health conditions rarely occur in isolation, and it is more likely people will go onto experience multiple comorbidities. For example, co-morbidities including psychological conditions are more likely in those with pain than those without5. The ABS reported that in 2020, 19% of Australians with long-term health conditions had two or more chronic conditions1. The interactions between physical and psychological conditions and associated symptoms are complex, and conditions may present independently or be highly intertwined. Common brain-related conditions (e.g. depression, anxiety, and dementia) can be exacerbated by physical conditions (e.g. OA, cardiovascular disease (CVD)) and vice-versa.

There are several models/theories that describe proposed mechanisms involved in these interactions. The biopsychosocial model which recognizes the contribution of all biological, psychological, social, and behavioural factors that dynamically interact with one another to generate the experience of pain17. Inflammation models describe how localized peripheral inflammation can trigger the development of neuroinflammation and subsequently the induction of dementia pathology18. The pain-sensitisation model recognises highly variable experiences of pain between individuals and the role of neuroplastic changes occurring in the peripheral and central nervous system, resulting in pain sensitisation, and impacting the person’s experience of pain19. The fear-avoidance model of musculoskeletal pain describes a psychological process where concern or fear about pain can lead to fear-avoidance behaviour which paradoxically can exacerbate pain and function20. The role of the gut microbiome has also been implicated with pain in OA21, depression22,23 and neurological conditions24.

There are clear associations between physical, psychological, and cognitive health outcomes (i.e. OA, anxiety, depression, and cognitive decline). However, it is not yet known what the most effective approaches are for improving psychological outcomes as well as physical outcomes (e.g. pain) in people living with OA. Physical activity interventions delivered by Accredited Exercise Physiologists (AEPs) have been shown to be highly effective for people living with 26 chronic conditions25, including OA26,27, anxiety and depression28,29, cognitive decline/dementia30,31, diabetes32 and cardiovascular disease33. Examples of effective exercise strategies include resistance training/strengthening exercises34, yoga35, and aquatic exercise36. Interventions that include lifestyle education and information on managing chronic conditions (e.g. information specific to OA, diabetes, depression, etc.) also provide additive beneficial effects37. Indeed, exercise and education are key components of clinical guidelines for OA. It is important that clients understand how to self-manage their pain and remain engaged with exercise to improve outcomes.

Research shows that therapeutic approaches such as cognitive behavioural therapy (CBT) are highly effective in treating mild-moderate depression and anxiety38. Furthermore, evidence suggests that the combination of exercise and behavioural therapies provides greater reduction in depressive symptomology than therapy alone39,40. Behavioural therapy approaches have been adapted or investigated in specific populations such as diabetes41, sleep disorders42, eating disorders43, and chronic pain44. Multimodal approaches that include a combination of exercise, therapeutic approaches and/or lifestyle education may be more effective than single mode approaches due to the focus on both psychological and physical health outcomes. However, there is limited evidence around the effect of multimodal treatment on multimorbidity.

Therefore, the aims of this systematic review and meta-analysis are to determine the most effective nonpharmacological approaches for reducing pain and symptoms of depression in people living with OA. We aimed to:

  1. 1.

    Identify effective approaches for improving pain and reducing symptoms of depression in people with OA.

  2. 2.

    Identify whether single or multi-modal approaches are more effective.

  3. 3.

    Present evidence-based recommendations for treatments for people with OA and associated depression.

In better understanding how to reduce OA-related symptoms and co-morbidities in populations who are at higher-risk of experiencing cognitive decline (i.e. people living with OA6), we may also reduce dementia prevalence rates, as well as improve psychological outcomes for people living with OA.


The review was preregistered with the international prospective register of systematic reviews (PROSPERO), in accordance with PRISMA guidelines (PROSPERO CRD42022338051)45.

Search strategy

Relevant studies published from electronic database inception until April 2022 were sourced by searching: EMBASE, PUBMED & MEDLINE, Web of Science, CINAHL and PEDro. Studies that included outcome measures relevant to physical outcomes (e.g. pain and physical function) and psychological outcomes (i.e. depression and anxiety symptomatology) were included in the initial search and this review. Search terms were Osteoarthritis OR OA; Depression OR anxiety OR mood OR affective OR psychological; Exercise OR physical activity OR education OR therapy OR CBT OR cognitive behavioural OR cognitive-behavioural OR lifestyle intervention OR mindfulness; RCT OR Randomised controlled trial OR Randomized controlled trial OR randomised controlled trial OR Randomized controlled trial (see Supplementary Material S1 for full list of search terms).

Eligibility criteria and study selection

Only randomized controlled trials (RCTs) were included in this review. Studies were included from all settings such as community clinics, hospitals, and in people’s homes via online platforms, if pain or symptoms of depression in people with OA were assessed. Studies were only included where two or more conditions were compared (control versus intervention). Exclusion criteria were studies having an active control group (e.g. aerobic activity versus resistance activity), studies where data was not presented separately for people with OA if they included people with other pain-related conditions (e.g. lower back pain), studies not available in English, and studies with insufficient information to conduct a meta-analysis. When means and standard deviations were not provided, authors were contacted and requested to share this data.

Only post-intervention scores were compared across conditions (i.e. post-intervention, control versus intervention group scores in pain and symptoms of depression). Cochrane methods and those developed by statisticians were used to convert data46,47,48. Pooled, weighted effect sizes were calculated for all approaches, followed by subgroups determined by intervention types (e.g. multimodal, aerobic activity only, resistance activity only, therapeutic only, movement meditation).


Only nonpharmacological interventions were included that involved either exercise, lifestyle education, therapeutic and/or multimodal approaches. These were categorized into groups as they emerged and refined during the screening process. The final groups were ‘multimodal approaches’, ‘movement mediation’ (including yoga, tai chi, and chi gong), ‘resistance activity only’, ‘aerobic activity only’, and ‘therapeutic approaches only’. Studies were included if they had one or more active interventions and an inactive control group (e.g. treatment as usual). Only data from the combined intervention and control group were used for the meta-analyses.

Outcome measures

Studies were included that used several different measurement tools to determine pain and depression in people living with OA. For pain these included the Western Ontario and McMaster Universities Arthritis Index (WOMAC)49, Knee Injury and Osteoarthritis Outcome Score (KOOS)50, Hip Disability and Osteoarthritis Outcome Score (HOOS)51, 36-Item Short Form Survey (SF-36) (pain component)52, (PISF), Harris Hip Score (HHS) (pain component)53, and arthritis self-efficacy (ASE) (pain component)54. For depression these included the Centre for Epidemiological Studies Depression (CESD) scale55, Hospital Anxiety and Depression Scale (HADS)56, Geriatric Depression Scale (GDS)57, Beck Depression Inventory (BDI), (AIMS) (psychological component)34, Emotional Distress and Depression Short-Form (EDD-SF)58, Depression Anxiety Stress Scales (DASS21)59, and the Patient Health Questionnaire (PHQ-9)60. Studies were included with a range of intervention durations and some included follow-up periods (e.g. 6 months, 1-year) which were considered when rating of the quality of the research article was completed. Articles were only included if they were in English.

Quality rating

All studies were quality rated based on the study design, participant characteristics, outcome measures and statistical analyses using a Quality Rating Tool10 (see Supplementary Material S2) developed by adapting and combining several different previously used tools10,61,62,,62. The tool ensured quality ratings reflected ability to determine potential statistical and clinical significance. Points were given based on study design and statistical analysis. Total scores on the quality rating tool ranged from 0 to 16 (i.e. higher scores indicated higher quality research).

Statistical analysis and clinical interpretation

The statistical analysis was conducted in Meta-Essentials developed by Suurmond and colleagues, Netherlands63. A random effects meta-analysis was completed with restricted maximum likelihood estimation (REML), assuming that the true effects differ across studies, as these studies vary in their study design and methodology (e.g. intervention type, duration, and outcome measure). Sensitivity analyses were performed to determine potential studies driving heterogeneity. Subgroup analyses were performed to investigate different types of interventions on clinical outcomes (pain and depression) and associated significance and effect sizes. The effect size of interventions was determined using Hedges’ g64 (page 110). Bias corrected standardized mean difference calculations (Hedges’ g) are recommended for studies with small sample sizes. According to Cohen, effect sizes less than 0.2 are trivial, 0.2 are small, 0.8 are medium, and 0.8 or higher are large65. Evidence for minimally clinically important differences (MCID) and moderate improvement estimates of change in OA-related pain had been published66. For the KOOS and HOOS short form pain measures, evidence has been published for MCID and moderate improvement estimates (2.2 and 15 respectively). Though reliability was much lower in knee compared to hip OA66.

A 95% confidence interval was used to determine the efficacy of combined interventions versus control on pain and symptoms of depression. Heterogeneity of the effects was assessed by calculating the I2 index and prediction intervals. The I2 index value reflects the proportion of true heterogeneity in the observed variance 69, 47], where 0% indicates no observed heterogeneity and increased heterogeneity is indicated by larger values where 25%, 50% and 75% reflect low, medium, and high heterogeneity respectively67. Prediction intervals are also presented throughout to ensure the predicted range of effects is unambiguous68. To understand the observed heterogeneity, we performed subgroup analyses to test intervention type. Publication bias was assessed graphically with contour-enhanced funnel plots and Egger’s regression test69. Funnel plots show if studies were missing only from areas of low statistical significance indicating that any asymmetry is likely to be caused by publication bias. The Egger regression gives “the degree of funnel plot asymmetry as measured by the intercept from regression of standard normal deviates against precision”69. If Egger’s test results in a p value less than 0.05, this implicates publication bias.


Included studies

In total 1984 studies were found through the database search of studies published up until April 2022 (Fig. 1). Following removal of duplicates, article screening and removal of studies that did not meet eligibility criteria, 61 studies remained that included measures of pain and/or symptoms of depression in people with OA (Fig. 1). Thirty-eight studies were then excluded for not having sufficient data for meta-analysis, an active control condition, pain or depression outcomes not being reported post-intervention, and articles that did not meet methodology requirements (i.e. not RCTs, protocol papers). The mean quality rating of the 23 remaining studies (see Table 1) used in the final meta-analyses was high (mean: 13; SD = 1.6; total possible score: 16; range: 10–16). Participant characteristics are outlined in Table 2.

Figure 1
figure 1

PRISMA flow diagram.

Table 1 Nonpharmacological approaches for pain levels and symptoms of depression in people with osteoarthritis (31 interventions from 23 studies).
Table 2 Participant characteristics.



For pain, 29 interventions from 21 published articles were included in the meta-analysis (n = 4,382; 65 ± 6.9 years; 70% female; people with knee/hip OA at varying severity). There was a significant overall medium effect on reducing pain (ES = 0.43, 95% CI [0.25, 0.61], p < 0.001). Heterogeneity was significant and substantial (I2 = 71.5%, pq < 0.001) (Fig. 2a).

Figure 2
figure 2

Main meta-analysis (a) and sub-group meta-analysis (b) for pain. Subgroup meta-analysis results showing effect sizes (i.e. SMDs), CIs (black lines), PIs (red lines) and I2 values for interventions. Movement meditation included yoga, tai chi, and qigong. Resistance exercise included hydrotherapy. Multimodal approaches included a combination of exercise and therapeutic/education-based approaches. CI confidence interval, I2  percentage of variation across studies that is due to heterogeneity, PI prediction interval, SMD standardized mean difference, w weight.

Sensitivity analyses were performed to examine potential outliers. A study was excluded with a very large effect size that was in the intervention subtype defined as ‘resistance exercise only’36. This analysis included 28 interventions from 20 published articles (n = 4322). The overall effect was ES = 0.33, 95% CI [0.24, 0.42], p < 0.001, I2 = 32.33%, pq = 0.0052. According to Cohen, it is a small to medium effect. Heterogeneity was at an acceptable level.

Subgroup analysis was then performed using random effects models for type of intervention and were defined as aerobic exercise only, resistance exercise only, therapeutic only, multimodal approaches and movement meditation. Effect sizes for reducing pain were significant for movement meditation (7 interventions: ES = 0.52, 95% CI [0.35, 0.69], p < 0.001; I2 = 0.0%), multimodal approaches (9 interventions: ES = 0.37, 95% CI [0.22, 0.51], p < 0.001; I2 = 41.2%, pq = 0.0092), resistance exercise (2 interventions: ES = 0.43, 95% CI [− 0.07,0.94], p = 0.046; I2 = 74.4%, pq = 0.0048), and therapeutic approaches (7 interventions: ES = 0.21, 95% CI [0.11, 0.31], p < 0.001; I2 = 3.1%, pq = 0.402). Aerobic exercise alone (including 3 interventions) did not significantly improve pain (Fig. 2b). Analysis of variance revealed significant differences between subgroup effect sizes for pain between movement meditation and aerobic exercise (p < 0.05) and movement meditation and therapeutic approaches (p < 0.01). Heterogeneity was at an acceptable level for all subgroups except resistance exercise though there are too few studies in this group to do further subgroup analysis.


For depression, 28 interventions from 20 published articles were included in the meta-analysis (n = 3377; 63 ± 7.0 years; 69% female; people with knee/hip OA at varying severity). There was a significant overall small effect on reducing symptoms of depression (ES = 0.29, 95% CI [0.08, 0.49], p < 0.001). Heterogeneity was significant and substantial (I2 = 73.3%, pq < 0.001) (Fig. 3a).

Figure 3
figure 3

Main meta-analysis (a) and sub-group meta-analysis (b) for symptoms of depression. Subgroup meta-analysis results showing effect sizes (i.e. SMDs), CIs (black lines), PIs (red lines) and I2 values for interventions. Movement meditation included yoga, tai chi, and qigong. Resistance exercise included hydrotherapy. Multimodal approaches included a combination of exercise and therapeutic/education-based approaches. CI confidence interval, I2 percentage of variation across studies that is due to heterogeneity, PI prediction interval, SMD standardized mean difference, w weight.

For the sensitivity analysis two studies were excluded with very large effect sizes. They included a resistance intervention36 and a movement meditation intervention74. This analysis included 26 interventions from 18 published articles (n = 3274). The overall effect is small and significant ES = 0.16, 95% CI [0.08, 0.24], p < 0.001, I2 = 11.78%, pq = 0.293. Heterogeneity was at an acceptable level.

Subgroup analysis was then performed using random effects models for type of intervention and were defined as aerobic exercise only, resistance exercise only, therapeutic only, multimodal approaches and movement meditation. Effect sizes for reducing symptoms of depression were significant for movement meditation (6 interventions: ES = 0.30, 95% CI [0.06, 0.55], p = 0.008, I2 = 18.9%, pq = 0.291), and multimodal interventions (8 interventions: ES = 0.12, 95% CI [0.07, 0.18], p < 0.001, I2 = 0%, pq = 0.940). Resistance exercise, aerobic exercise, or therapeutic approaches alone (2, 4 and 6 interventions respectively) did not improve symptoms of depression (Fig. 3b). Analysis of variance revealed no significant differences between subgroup effect sizes for depression. Heterogeneity was at an acceptable level for movement meditation, multimodal approaches and aerobic exercise though was high for the remaining subgroups.

Publication bias

A funnel plot and the Egger test of asymmetry were used to examine publication bias (Fig. 4). The funnel plot was asymmetric. The Egger test was significant for pain (p < 0.001) suggesting potential publication bias and/or small-study effects91, though was not significant for depression (p = 0.216).

Figure 4
figure 4

Funnel plot of all included studies for (A) pain, and (B) symptoms of depression.


Summary of main findings

Overall, meta-analyses from 23 high quality rated studies showed that nonpharmacological approaches significantly reduced pain (ES = 0.43, 95% CI [0.25, 0.61], 29 interventions) and symptoms of depression (ES = 0.29, 95% CI [0.08, 0.49], 28 interventions) in people with OA. Heterogeneity was lowered after removing outliers. For pain, subgroup analysis showed significance for movement meditation, multimodal approaches, therapeutic approaches, and resistance exercise only. However, heterogeneity was considerably high for resistance exercise. Movement meditation and multimodal approaches were also significant for symptoms of depression.

Based on these findings, approaches that are either multimodal (e.g. education/therapeutic approaches alongside exercise) or include focus on both the body and mind (e.g. focused breathing and mindfulness techniques used alongside movement, seen in yoga, qigong, and tai chi) are effective for reducing pain and symptoms of depression in people with knee/hip OA and warrant further investigation.

How these findings compare to others

Our findings support and expand on others regarding beneficial effects of exercise for reducing pain in people living with OA23,87. There are also differences between our findings and those of other studies. Fransen et al.’s26 Cochrane review based on 44 trials indicated that exercise significantly reduced pain and improved physical function26. Uthman et al.’s92 review based on 60 trials compared the effectiveness of different exercise interventions92.

Our analysis included only two resistance and three aerobic interventions. Both Cochrane reviews included more articles than our review. This was likely due to the main outcome of interest being pain and did not include depression, and because we only included studies with an inactive control group. Our findings align with other studies regarding reducing pain26,92, though extend our current understanding by investigating exercise type. In Fransen et al.’s review, no subgroup analysis was performed to examine potential different effects of exercise type, despite the heterogeneity and marked variability that was present regarding the exercise interventions assessed and other aspects of methodology26.

A more recent network meta-analysis investigated the relative efficacy of different types of exercise (aerobic, mind–body, strengthening, flexibility/skill or mixed) and found that aerobic had an equivalent beneficial effect (SMD = 1.11) to mind–body approaches in reducing pain93. Additionally, reporting that mixed exercise was the least effective. However, these approaches did not include education/therapeutic components as we did here, nor did they examine depression symptomatology. In another network analysis that examined different approaches on mental health measures in people with OA, the authors concluded that strengthening exercise was most beneficial for overall mental health, or mixed exercise for symptoms of depression94, though pain was not examined.

Strengths and limitations

Strengths of this study include that we were able to perform a meta-analysis including 23 studies where one or more intervention groups were compared with a control group. We also achieved high agreement levels between independent reviewers on ratings of study quality. We showed that similar approaches were effective for reducing both pain and symptoms of depression in OA, suggesting that these benefits will likely transfer to additional improvements in quality of life92,90. Further, identifying effective therapeutic options can support informed choice for people living with OA, and thus support shared decision-making processes in collaboration with healthcare professionals and family members to tailor treatments that best suit their individual needs and preferences95.

However, there are several limitations to consider when interpreting these results. Although the overall effect obtained for the meta-analyses for both pain and depression were significant, not all studies reported beneficial effects (e.g. pain72,85, symptoms of depression79,89). Possible explanations for observed non-significant effects in individual studies include low intervention dose, baseline depression scores and/or lack of formal depression diagnoses, and that participants differed between studies in the diagnostic site of OA, e.g. knee or hip. In our study, this information is included in Tables 1 and 2 where available.

Publication bias should also be considered, and some studies had very small sample sizes which may increase the likelihood of Type II errors. The funnel plots and Egger’s regression indicated publication bias was likely for pain level data though not for symptoms of depression, possibly due to pain more frequently being recorded as a primary outcome measure and depression a secondary outcome measure in these studies. Therefore, the overall effect size for pain may be an overestimation. This can be caused when small sample studies have not been published because significant effects were not obtained.

We included RCTs that measured pain and symptoms of depression in OA. Studies not measuring depression though were effective in reducing pain in OA may not have been captured in this review, for example therapeutic exercise and pain education96. Safety is a key consideration in intervention studies additional to effectiveness and adverse effects have been captured in other reviews26,97. However, we did not extract data for this outcome. Exercise is usually safe and well-tolerated for people with OA and depression, but this is based on limited data as studies seldom collect this as an outcome. Future trials would benefit from more focus and rigorous reporting on safety.

Clinical implications

In this systematic review we provide evidence for several approaches for reducing pain and symptoms of depression in OA, a condition with a substantial health burden in which co-morbidities are common. We highlight the importance of considering psychological health, i.e. symptoms of depression, in addition to physical health, i.e. pain, when providing quality healthcare to people living with OA. Interventions may be more effective when they are tailored to the individual and involve healthcare professionals working collaboratively across disciplines, e.g. nutrition, exercise, psychology, as well as with patients/clients. Further investigation is warranted on combining psychological and physical health focused interventions as well as multidisciplinary expertise to optimize patient outcomes.

Pain is experienced by at least half of people living with dementia and is most often attributed to OA98. Given the strong associations between OA and dementia6,7,8, 98, and that forty percent of dementia cases could be prevented or delayed by targeting risk factors including physical inactivity and associated chronic conditions31,99, targeting associated chronic conditions including OA will likely lead to reductions in dementia prevalence rates.


In this study, mind–body approaches were more effective than aerobic/resistance exercise or psychological therapy alone for reducing pain and depression in people with OA. Future research is required to determine the optimal approach for OA based on diagnostic site and symptom severity of comorbid conditions such as depression. Further, given the high prevalence rates of chronic conditions and people who experience multimorbidity, a better understanding of how the underlying mechanisms of co-occurring chronic conditions interact is also required. Incorporating qualitative studies such as focus groups and interviews to examine the experience, perspectives and preferences of people living with OA and depression will also add considerable value to this area.