Efficacy of n-3 fatty acid supplementation on rheumatoid arthritis’ disease activity indicators: a systematic review and meta-analysis of randomized placebo-controlled trials

Abstract Theoretical evidence and previous studies suggest that oralnutrient supplementation (ONS) with n-3 fatty acids for rheumatoid arthritis (RA) has the potential to lower disease activity indicators and non-steroidal anti-inflammatory drug (NSAID) uptake. A systematic search was conducted on five databases/registries from inception until May 23, 2021 with the aim to identify randomized placebo-controlled trials comparing n-3 supplements to placebo on disease-specific outcomes. A total of 23 studies matched the criteria (PROSPERO: CRD42019137041). Pooled analyses revealed that n-3 ONS provided a small effect in reducing pain [standardized mean difference (SMD): −0.16, 95% confidence intervals (CI): −0.40 to 0.09], and tender (SMD: −0.20, 95% CI: −0.46 to 0.05) and swollen joint count (SMD: −0.10, 95% CI: −0.28 to 0.07). In sensitivity analyses, there was a small effect in the reduction of NSAIDs intake (SMD: −0.22, 95% CI: −0.90 to 0.46), and c-reactive protein was reduced only by 0.21 mg/dL (95% CI: −0.75 to 0.33). Similar findings were observed regarding other objective/subjective outcomes. The certainty of the evidence was mostly of “very low/low” quality. Overall, n-3 ONS in RA might have a limited clinical benefit. Previous findings suggesting a reduction in NSAID intake may have been biased from the inadequate blinding of interventions.


Introduction
Rheumatoid arthritis (RA) is an autoimmune disease characterized by a chronic inflammatory polyarthritis leading to joint damage and disability (Josef S. Smolen et al. 2018;Guo et al. 2018).RA affects mostly women than men and a recent meta-analysis estimated the pooled, global period-prevalence from the year 1955 until 2015 as 0.46% [95% confidence intervals (CIs): 0.36% and 0.57%] (Guo et al. 2018;Almutairi et al. 2021).The etiology of RA is not known and current treatments suppress inflammatory pathways with the aim to induce disease remission or lower disease activity (Guo et al. 2018).Oral nutrient supplementation (ONS) with fish oils is among the most popular choices in patients with RA, although only a small proportion of them reveal the intake of dietary supplements to their healthcare providers (DeSalvo et al. 2019;Fletcher et al. 2020).
Fish oils contain naturally two forms of n-3 polyunsaturated fatty acids (PUFAs), eicosapentaenoic (EPA) and docosahexaenoic acids (DHA), which possess anti-inflammatory activities such as increasing the production of anti-inflammatory eicosanoids (Navarini et al. 2017;Ye and Ghosh 2018).In animals with inflammatory arthritis, supplementation of EPA and DHA decreased expression levels of proinflammatory cytokines and improved joint pathology (Kim et al. 2018;Ierna et al. 2010).Furthermore, in the general population not taking n-3 supplements, dietary consumption of long-chain n-3 PUFAs was associated with a lower risk of developing RA in the prospective study of Swedish Mammography Cohort (Di Giuseppe et al. 2014).However, among studies including participants receiving n-3 supplements, this lower risk is not evident, as seen in the Nurses' Health Study (Sparks et al. 2019) and in the Women's Health Initiative Observational Study and Clinical Trials (Krok-Schoen et al. 2018) in which participants might have taken supplements of n-3 fatty acids (FAs) in addition to other supplements.Moreover, the observed lower disease activity among RA patients may be attributed to the greater fish consumption rather than taking n-3 FA supplements (Tedeschi et al. 2018).
A possible cause for these contradictory results in the previous meta-analyses could involve the methodological heterogeneity stemming from including not only trials comparing supplements against placebo, but also with different lifestyle interventions as comparators (Senftleber et al. 2017;Gioxari et al. 2018;Sigaux et al. 2022), supplements with diets (Goldberg and Katz 2007;Sigaux et al. 2022), supplements with drugs (e.g.indomethacin) (Senftleber et al. 2017;Gioxari et al. 2018;Sigaux et al. 2022), or allowing in the same meta-analyses vegetable sources of PUFA (such as alpha-linolenic acid, ALA) as their primary intervention along with trials whose primary intervention was the supplementation of EPA/DHA (Lee, Bae, and Song 2012;Gioxari et al. 2018;Sigaux et al. 2022).In addition, clinical heterogeneity cannot be ruled out since some previous meta-analyses included patients with RA together with other types of arthritis or medical conditions (Senftleber et al. 2017;Goldberg and Katz 2007;Sigaux et al. 2022).Thus, the aim of the present systematic review was to examine the efficacy of n-3 FA ONS in ameliorating disease activity indicators and medication use in patients with RA, compared against placebo.

Study design and protocol
The present systematic review and meta-analysis followed the reporting guidance of the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) 2020 Statement (Page et al. 2021).The protocol was registered on PROSPERO (CRD42019137041).

Participants, intervention, comparator, outcomes
The research question was addressed in the form of PICO as: In patients with RA (P), what is the efficacy of omega-3 supplements (I) compared to placebo (C) in improving disease activity indicators (O)?

Eligibility criteria
Randomized controlled trials (RCTs) administering n-3 FA supplements in patients with RA were considered eligible for inclusion.Eligible trials included adult (>18 years of age) participants, diagnosed with RA based on the criteria defined by the American College of Rheumatology (ACR) (Arnett et al. 1988) or the ACR/European League Against Rheumatism (EULAR) criteria (Kay and Upchurch 2012).Eligible interventions were restricted to those including EPA and/or DHA, whereas additional nutrients (such as other n-3 FAs or vitamins) within the supplements were also permitted.All eligible RCTs compared n-3 FA ONS against placebo, consumed in a similar form as in the intervention arm, in order to facilitate clear comparisons.If additional co-interventions were included, they had to be equally administered in both the intervention and control groups.
Reasons for exclusion of a study involved (1) participants suffering from other serious medical conditions, (2) randomization of pregnant women, (3) diagnosis of RA with undisclosed criteria, (4) using healthy controls, children or adolescents (<18 years of age), (5) a non-randomized trial design, (6) studies with interventions of n-6 FA instead of n-3 FA or studies lacking a placebo group, i.e. comparing their n-3 FA intervention to diet, other lifestyle interventions, or another n-3 FA intervention.No restrictions were imposed on intervention duration.

Search strategy
A systematic review was conducted, on the electronic databases of PubMed, Cochrane CENTRAL, Scopus, ClinicalTrials.gov,and World Health Organization International Clinical Trials Registry Platform (WHO ICTRP).Moreover, other sources of studies included manual searches on reference lists of relevant full-text articles and the grey literature, through abstracts published on the archives of congresses hosted by the ACR/ECR.The search terms used during the search process included the following keywords: "omega-3," "n-3," "eicosapentaenoic acid," "EPA," "docosahexaenoic acid," "DHA," "unsaturated," "PUFA," "polyunsaturated," "marine fatty acids," "cod liver oil," "fish oil," "krill oil," "rheumatoid arthritis," "arthritis" and "rheumatoid disease."If applicable on the databases, a combined search using MeSH terms was applied.The search was restricted to records published in English from the inception of the databases and was last updated on May 23, 2021.A record of the search procedure of all databases is detailed in the supplementary material.Initially-identified records were screened for eligibility by two reviewers (KG and MGG) and disagreements were resolved with discussion with the primary investigator (DPB).

Data extraction
Extracted information from primary studies was uploaded on spreadsheets in duplicate (assessed by two independent researchers; KG and IM) with preference given to data from intention-to-treat (ITT) analyses.When data was not available in full text format, the primary authors were contacted.Additional information about the data extraction can be found in the supplementary material.

Risk of bias and quality of evidence assessments
Eligible studies were independently assessed for bias with the Cochrane's revised Risk of Bias (RoB) tool 2 (Sterne et al. 2019) by two authors (IM and KG).Judgments were made if there was a low risk, some concerns or high risk of bias in terms of the randomization process, deviations from intended interventions (based on the effect of assignment to intervention; ITT effect), missing outcome data, measurement of the outcomes, selection of the reported results, and the final assessment of the overall bias.
In addition, two authors (KG and MGG) assessed independently the quality of the evidence for the included outcomes based on the GRADE (Grades of Recommendation, Assessment, Development and Evaluation) framework, using the GRADEpro GDT software (Schünemann et al. 2013).According to the GRADE approach, the quality of the evidence regarding an outcome is rated as "high," "moderate," "low" or "very low." In case of disagreements for the RoB and GRADE assessments, discussions were held with the primary investigators (LIS and DPB) in order to reach a consensus.More information regarding these assessments are provided in the supplementary material.

Data synthesis
When three or more primary studies provided information on the same outcome, meta-analyses were conducted using the inverse variance method via random-effects models under the assumption that different studies estimate different, yet related, intervention effects (Higgins et al. 2019).
In cases where a meta-analysis was not feasible, a narrative synthesis was preferred.For the ESR, the effect estimate was the mean difference.However, for the rest of the outcomes, due to variability in measurement scales or inadequate reporting of the assessment methods, the effect estimate was the standardized mean difference (SMD) calculated as the difference in means of the groups divided by their pooled standard deviation (SD) (Higgins et al. 2019).SMDs can be interpreted with the rule of thumb where an SMD value of 0.2 corresponds to a small effect, 0.5 a moderate effect, and 0.8 a large effect (Higgins et al. 2019).Furthermore, when data were reported as the number of patients achieving an outcome in each intervention group, the odds ratios (ORs) were calculated and converted to effect sizes (SMDs) as previously described (Higgins et al. 2019).Only one crossover trial (van der Tempel et al. 1990;Tulleken, Limburg, and Van Rijswijk 1988) reported its results for both intervention periods and the missing SD of the mean difference was calculated with a 0.5 correlation coefficient with methods described elsewhere (Elbourne et al. 2002;Follmann et al. 1992;Higgins et al. 2019).Additional information about the imputations of the SDs can be found in the supplementary material.In all outcomes, a decreased point estimate indicates achieving a benefit in patients taking n-3 FA supplements.
With the use of the restricted maximum likelihood method (REML), the between-study variance was measured with the tau-squared statistic (tau 2 or τ 2 ) and heterogeneity was quantified as a percentage with the I-squared statistic (I 2 ) with values >50% being considered as substantial heterogeneity (Veroniki et al. 2016;Higgins et al. 2019).The 95% CI of tau 2 and I 2 were calculated via the Q-profile method to aid in the evaluation of the heterogeneity (Veroniki et al. 2016;Ioannidis, Patsopoulos, and Evangelou 2007).Heterogeneity was also assessed via a chi-square-based Cochran's Q statistic with a non-significant p-value set at the level of 0.10 and via visual inspection of the forest plots (Higgins et al. 2019).In the meta-analysis of pain assessments, in which more than ten studies were pooled, had no clear funnel plot asymmetry and the tau 2 was >0, 95%, prediction intervals (PIs) were also calculated.PIs can inform about the possible range of the study effects which future trials might have (Veroniki et al. 2019;Higgins et al. 2019).
Publication bias was assessed by creating contour-enhanced funnel plots and statistically via a modified Egger's test only in the analyses of pain and CRP in which at least ten studies were included (Sedgwick 2013;Higgins et al. 2019).In addition, regular funnel plots were also constructed with the inverse of the square root of sample size on the y axis for overcoming a possible artificial association of the SMD with its standard error (Sedgwick 2013;Higgins et al. 2019).Potential sources of heterogeneity were examined with subgroup and meta-regression analyses supplementary material.Furthermore, a series of sensitivity analyses were performed to assess the robustness of the main findings (supplementary material).
In analyses where at least five studies were pooled, the Hartung-Knapp/Sidik-Jonkman (HKSJ) adjustment to standard errors was utilized (Veroniki et al. 2019).The level of significance for all meta-analyses was set at α = 0.05, and analyses were carried out on R software version 3.6.1 with the use of the metaphor (Viechtbauer 2010) and meta (Balduzzi, Rücker, and Schwarzer 2019) packages.
Overall, the pooled sample size reached 1,443, with the majority of the studies recruiting female participants (Table 1).Multicenter studies (Kremer et al. 1995;Reed et al. 2014;Olendzki et al. 2011;Galarraga et al. 2008;Kjeldsen-Kragh et al. 1992;Nielsen et al. 1992;Park et al. 2013) recruited a median of 81.5 participants, whereas the rest of the studies had a median of 50 participants.

Assessment of risk of bias
Details of the RoB assessment can be found in Supplementary Figure 1.Generally, regarding the "overall bias" domain, the majority of the studies were classified as being at high risk, while none of the RCTs exhibited a low risk.This can be explained by a variety of factors, including failure to report the techniques used for randomization and/or allocation concealment, no secure masking of the interventions, failure to protect blinding of outcome assessors (especially in outcomes which require their judgment), exclusion from the analyses eligible participants for reasons incompatible with the ITT principle (for example, exclusions for noncompliance and side effects, or not disclosing the reasons), a protocol registered in retrospect, or lack of a protocol.
Based on the GRADE system (Supplementary Table 2), there was a "very low" certainty of the evidence for TJC and SJC due to risk of bias, low sample size and wide CIs.
None of the studies assessed the effect of n-3 PUFA ONS on the SDAI.

Secondary outcomes
Overall, 12 studies with a duration of three to 12 months, provided data for the examination of pain scores (Figure 4).N-3 FA supplements compared to placebo had a non-significant, small effect in reducing pain scores (SMD = -0.16,95% CI: -0.40 to 0.09, p = 0.1924).Furthermore, there was a moderate degree of statistical heterogeneity; I 2 = 43.88%(95% CI: 0% to 79.55%) and tau 2 = 0.07, 95% CI: 0 to 0.34), and the certainty of the evidence (Supplementary Table 2) was judged as "very low" based on very serious concerns regarding risk of bias and serious concerns regarding imprecision (wide CI).In addition, this meta-analysis revealed a wide 95% PI (−0.78 to 0.47).
With regard to the PhyGA, the analysis included six RCTs lasting between 15 weeks to 12 months in total.The pooled SMD had a point estimate of −0.01 (Supplementary Figure 5; 95% CI: −0.21 to 0.20, p = 0.9352; Heterogeneity: I 2 = 0%, 95% CI: 0% to 59.46% and tau 2 = 0, 95% CI: 0 to 0.13), and the evidence was judged as of "very low" certainty due to very serious concerns in risk of bias and imprecision (Supplementary Table 2).
Finally, in a meta-analysis of five RCTs with intervention durations ranging between of three to 18 months, ONS with n-3 FA moderately decreased NSAID intake in a non-significant manner (Supplementary Figure 7; k = 5 studies, SMD = −0.51,95% CI: −1.72 to −0.69, p = 0.3013).Nevertheless, substantial statistical heterogeneity was observed in this analysis (I 2 = 88.23%,95% CI: 62.65% to 98.78% and tau 2 = 0.75, 95% CI: 0.17 to 8.04).The certainty of evidence was deemed as "very low," as a result of very serious concerns for high risk of bias, small sample size and wide CIs (Supplementary Table 2).A sensitivity analysis (Supplementary Table 5) pooling only the studies who reported the number of participants with reduced NSAIDs use in each arm revealed that in ONS with n-3 FA, the OR for reduced NSAID intake was 4.23, with wide CIs and substantial statistical heterogeneity (k = 4 studies, 95% CI: 0.64 to 28.01, p = 0.1344; Heterogeneity: I 2 = 84.77%,95% CI: 50.58% to 98.95% and tau 2 = 3.13, 95% CI: 0.57 to 53.07).

Subgroups analyses and meta-regressions
Subgroup analyses of CRP or pain scores by medication change (NSAID uptake) failed to reveal differences between these two groups of studies (Supplementary Table 3) and their meta-regressions did not reveal a significant association with their effect sizes (Supplementary Table 4).
Interestingly, the subgroups in CRP by placebo types showed similar effect sizes and the meta-regression model failed to reveal a significant difference according to this parameter (p = 0.9523).
Subgroups of pain scores by placebo types revealed a slightly increased effect size for studies using monounsaturated FA (MUFAs) as placebos (SMD = 0.04), whereas for the analysis of RCTs using saturated FAs (SFAs) as placebos (k = 2), a moderate effect size was calculated (SMD = −0.51).
However, when meta-regressions were conducted for all studies for either CRP or pain scores (Supplementary Table 4), they failed to reveal any significant association of the effect sizes with placebo types (p = 0.9523 and p = 0.4250, respectively).Crossover design also had a non-significant association in the meta-regressions of either CRP (p = 0.9357) or pain (p = 0.1367).
Although for CRP the subgroup of three RCTs with an intervention duration exceeding three months was significant (p = 0.0313), meta-regressions including intervention duration either as dichotomous or continuous moderator failed to provide any significant associations (p = 0.0520 and p = 0.1575, respectively).Similarly, the dose of n-3 FAs, imputed either as a dichotomous or a continuous moderator, was not associated with the effect sizes of pain scores (p = 0.0992 and p = 0.5285).On the other hand, the dose of n-3 FAs expressed as a continuous moderator (expressed in grams) had a significant positive association with the effect sizes of CRP concentrations (coefficient: 0.11, 95% CI: 0.02 to 0.20, p = 0.0243), but could not explain any of the observed heterogeneity (R 2 = 0.00%).None of the other examined moderators had a significant association with the effect sizes of CRP concentrations and pain scores.

Publication bias
Visual inspection of funnel plots (Supplementary Figures 8  and 9) of both analyses on CRP levels and pain scores did not reveal a serious asymmetry, a finding in accordance with the statistical examination of funnel plot asymmetry via the modified Egger's test (p = 0.8066 and p = 0.4453, respectively).

Sensitivity analyses
Additional sensitivity analyses were conducted to examine the deviation of the results under different scenarios (Supplementary Table 5).By replacing the REML method to the Paule Mantel method of estimating the tau 2 , all results remained nearly identical.Furthermore, by excluding all studies which reported medians and had to be converted to parametric statistics, only the analyses for CRP, DAS28 and NSAID use revealed a noticeable difference in the magnitude of the effect size compared to their primary meta-analyses.Regarding the ESR concentrations, a notable reduction in the degree of statistical heterogeneity was observed, however in these analyses, the level of statistical significance was not altered.
Although the meta-analyses of the DAS28 and NSAID intake revealed a moderate effect, after excluding an outlier/ influential study (Rajaei et al. Supplementary Table 5).

Studies included in the qualitative review
A total of 14 studies (Hernández-Cruz, Alcocer-Varela, and Cardiel 1998;Rajaei et al. 2015;Lau, Morley, and Belch 1993;Reed et al. 2014;Lau, McLaren, and Belch 1995;Belch et al. 1988;Cleland et al. 1988;Larkin, Capell, and Sturrock 1985;Nielsen et al. 1992;Park et al. 2013;Adam et al. 2003;Volker et al. 2000;Lau, Morley, and Belch 1991;Geusens et al. 1994) were included in the qualitative review and their individual results were in overall agreement with the meta-analyses presented herein.More detailed comparisons are provided in the supplementary material.

Discussion
The present meta-analysis revealed an overall small benefit following n-3 supplementation for improving RA disease activity parameters.Additionally, the published RCTs in the field were characterized by serious limitations regarding their research design and a small, often inadequate sample size.For most of the analyses, the statistical manifestation of heterogeneity was deemed as low-to-moderate, with the exception of the DAS28 and changes in the NSAID intake, both of which demonstrated a substantial degree of heterogeneity.These findings call into question previous evidence synthesis regarding the effectiveness of n-3 supplementation in managing RA symptomatology.
Individual trials advocate for the improvement of inflammation markers following n-3 ONS while, in parallel, a precedent meta-analysis (Gioxari et al. 2018) suggested a significant decrease in leukotriene B4 levels.Several pathways have been proposed as inflammation modulators in RA (Perricone et al. 2019;Sakkas et al. 2019;Sakkas and Bogdanos 2018;Sakkas et al. 2018), however, based on the present meta-analysis, inflammation markers like the CRP or the ESR do not appear to be improved following n-3 FA supplementation.In verification to this, TJC and SJC failed to improve in the pooled intervention arm.In addition, previous meta-analyses had suggested significant effects of n-3 FA ONS on reducing pain (Goldberg and Katz 2007;Gioxari et al. 2018;Sigaux et al. 2022), the number of TJC or SJC (Goldberg and Katz 2007;Sigaux et al. 2022), DAS28 (Sigaux et al. 2022) and a moderate effect size (SMD) for the reduction of CRP concentrations (Gioxari et al. 2018), with substantial levels of heterogeneity.In parallel, an early individual patient data meta-analysis revealed significant reductions in the number of TJC (Fortin et al. 1995).
Possible differences between the current study and the above-mentioned meta-analyses could be attributed to (i) the inclusion of patients with other types of arthritis or medical conditions alongside RA (Senftleber et al. 2017;Goldberg and Katz 2007;Sigaux et al. 2022), (ii) the pooling of non-randomized trials together with RCTs (Fortin et al. 1995), and (iii) the inclusion of RCTs which compared dietary interventions of n-3 FAs (Senftleber et al. 2017;Gioxari et al. 2018;Sigaux et al. 2022), compared supplements with diets (Goldberg and Katz 2007;Sigaux et al. 2022), compared supplements with drugs (e.g.indomethacin) (Senftleber et al. 2017;Gioxari et al. 2018;Sigaux et al. 2022), or included RCTs with other FAs (such as alpha-linolenic acid, ALA) as their primary intervention (Lee, Bae, and Song 2012;Gioxari et al. 2018;Sigaux et al. 2022).
A matter of some concern, which has not been attended in this field, is the likely scenario of participants or health care providers discerning the assignment to intervention groups based on the physical properties of the administered capsules.Fish oils possess a characteristic flavor and aroma, hence a placebo lacking similar properties could unveil the intervention assignments and reveal the active treatment arm.Furthermore, common side-effects from the gastrointestinal tract, such as belching followed by a fishy aftertaste (Muchow et al. 2009), can also reveal the content of the capsules and un-blind both the participants and the care provider/outcomes assessor who is interviewing/treating them, a fact not uncommon in clinical trials (Schulz and Grimes 2002;Wilsey, Deutsch, and Marcotte 2016).Thus, although the preparation of placebo capsules with the same flavor, smell and appearance as the active treatment is essential in securing double-blinding, it was largely ignored by the majority of the included trials.In this regard, the compromised blinding of the interventions was a key element for the critical assessment of the induced studies.It increased the risk of bias for the outcomes where judgment was required for their assessment and in consequence it led to downgrading the certainty of the evidence for these specific outcomes.
The present meta-analysis showed that supplementation with n-3 FAs does not reduce the use of NSAIDs substantially, as previously thought.N-3 FAs have several similar physiological effects to the NSAIDs.The NSAIDs block the activity of cyclooxygenases, thus reducing the production of pro-inflammatory eicosanoids, whereas n-3 FAs increase the production of anti-inflammatory eicosanoids by providing a different substrate to the cyclooxygenases (Ye and Ghosh 2018;Goldberg and Katz 2007).An earlier meta-analysis of three RCTs suggested a significant, moderate decrease in the NSAID use (Goldberg and Katz 2007) following ONS with n-3 FAs.However, the present sensitivity analysis, which excluded a noticeable outlier, resulted in a small decrease in the NSAID use.Moreover, reducing or interrupting the use of NSAIDs consists of a clinical decision which requires critical judgment by the healthcare providers (Higgins et al. 2019).Therefore, these clinical decisions may be influenced when the allocation into the different intervention arms is easy to discern and, as a result, it may help explain the "inflated" reduction in NSAID use reported in these trials.
Early in this field, the selection of an appropriate placebo as a comparator had emerged as a perplexed issue.When other types of oils are used in the control arm, they can in fact act as parallel interventions instead of placebos (Kremer et al. 1995;Geusens et al. 1994).Although air-filled placebos seemed as a feasible, alternative approach (Lau, Morley, and Belch 1993;Lau, Morley, and Belch 1991), as noted by Galarraga et al. (2008) participants could easily understand that their air-filled capsules were in fact empty.The present study identified a plethora of ingredients used in placebo capsules, ranging from neutral ingredients, such as starch, to monounsaturated and other polyunsaturated FAs.However, the present meta-regressions according to different placebo types did not reveal associations with the effect sizes of two outcomes of different nature, being CRP concentrations and pain scores.
Furthermore, the appropriate dose of EPA and DHA for RA has not been determined yet.An early review recommended a starting dose of 2.7 g/day of EPA/DHA (Stamp, James, and Cleland 2005) and subgroups of a previous meta-analysis demonstrated significant associations of higher doses (>2.7 g/day) for improving morning stiffness and the number of TJC (Goldberg and Katz 2007).Similarly, another meta-analysis using subgroups of ≥3 g/day compared to <3 g/ day revealed significant associations with the outcomes of the PtGA, and the hand grip strength, in their respective meta-regressions (Gioxari et al. 2018).However, their results do not apply only to the EPA/DHA, since ALA was also among their eligible interventions.Moreover, a recent meta-analysis concluded that doses exceeding 2 g of fish oil/ day rather than vegetable sources of PUFA, may lead to favorable outcomes, and in addition, the authors raised concerns that high dosage schemes may increase the risk of bleeding (Sigaux et al. 2022).Another meta-analysis meta-regressed the effect of the administered n-3 FA dose on the feeling of pain and reported a possible inverse dose-response relationship at ≥3.6 g/day, but this finding was based on a small sample size since only four out of 30 trials reached this cutoff (Senftleber et al. 2017).In contrast, the meta-regressions performed herein revealed an association with CRP levels and not with pain.On the whole, it seems that an increased dose of n-3 FA intake may be associated with the reported beneficial outcomes.However, it remains unclear if there is a dose which gives rise to an inverse dose-response relationship and/or the risk of bleeding.Furthermore, a clear caveat in the aforementioned evidence stems from the concomitant pooling of lifestyle interventions alongside n-3 ONS.Thus, future attempts aiming to recommend a dosing scheme should consider the different sources of the administered n-3 FAs.
The strengths of the present study include the relatively large number of retrieved trials.Moreover, meticulous care was taken to circumvent the known problem of mixing 'apples with pears' in nutritional meta-analyses (Barnard, Willett, and Ding 2017), by including only placebo-control trials in this review, while excluding studies using lifestyle interventions or comparing supplements with control groups of dietary or drug interventions.Control arms with n-3 FA ONS were also excluded herein.Therefore, by avoiding mixing diets, drugs and supplements, the present results outline the relative effectiveness of n-3 FA supplements against placebo.These criteria lead to clearer comparisons and consequently address the 'compared to what?' question often arising in clinical trials and meta-analyses (Levack et al. 2019;Barnard, Willett, and Ding 2017).In addition, as previously noted (Miles and Calder 2012), blinding is highly unlikely in trials with control groups lacking placebo arms (e.g., using diets as comparators).Furthermore, the decision to include patients with defined diagnostic criteria led to increased confidence in terms of defining the target population.Last, care was also taken to choose appropriate effect estimates wherever possible, i.e. selecting mean differences in cases where the same measurement scales were used, thereby presenting results in this effect measure for the first time in this field.
In spite of these strengths, this review also entails several limitations.First, selecting records published in English could have led to the omission of manuscripts published in other languages, despite the large number of RCTs retrieved.Second, the inclusion of studies clearly stating the diagnostic criteria used, led to the exclusion of RCTs which omitted describing them, and although it increased the confidence in the quality of the pooled trials, it also decreased the number of RCTs in the meta-analyses.Clinical heterogeneity imposed by differences in populations diagnosed by older diagnostic criteria should also be acknowledged, since about half of the trials were conducted before the year 2000.In addition, post hoc decision to accept newer diagnostic criteria, such as ACR/EULAR, did not increase the number of retrieved studies, since they all applied older diagnostic criteria.A third limitation stems from the fact that all retrieved RCTs had small sample sizes, with an overall low methodological quality which should also be acknowledged and considered in interpreting the results presented herein.Fourth, imputations of SDs and transformations of non-parametric statistics (e.g., medians) had to be implemented in order to facilitate the pooling of few RCTs.Although sensitivity analyses via the exclusion of these studies did not reveal any substantial deviations, it should be acknowledged that imputations may lower the confidence of the true value of the results.Fifth, the original aim of including change scores instead of final values could not be met since the majority of the retrieved studies did not present their results in this format.Sixth, evidence from subgroup and meta-regression analyses should be regarded as having an observational design, and caution is required when interpreting their findings (Higgins et al. 2019).Last, but not least, since the current review investigated ONS with EPA and/or DHA, any generalization to other FAs is not recommended.
Future research is required to countervail the methodological pitfalls highlighted in the present review.Priority should be given in concealing the fishy aftertaste of n-3 FA supplements, especially in studies recording outcomes which require subjective judgements by the participants or care providers.A possible solution may be the preparation of placebo capsules with identical physical properties to the n-3 FAs.However, one limitation to keep in mind is that the fishy taste is unpleasant and might lower the overall compliance.An alternative solution may include the use of enteric-released n-3 capsules which help release the FAs in the intestine and have been tested for n-3 supplementation in other autoimmune diseases (i.e.Crohn's disease) (Schneider et al. 2011).Also, future research may consider new techniques such as nano-/micro-encapsulation of the n-3 FAs, which can also help reduce or eliminate their sensory characteristics (Vieira et al. 2021;He et al. 2020;Serfert, Drusch, and Schwarz 2010;Muchow et al. 2009).If the fishy aftertaste persists in classic or nanoengineered preparations, adding more favorable tastes (such as vanillin or fruity flavors) may be a viable solution (Serfert, Drusch, and Schwarz 2010;Muchow et al. 2009).Furthermore, the relatively large size and number of the capsules was also an important reason for drop-out (Reed et al. 2014;Olendzki et al. 2011;Galarraga et al. 2008;Tulleken et al. 1990).Giving participants the will to choose their dosing scheme by diving the amount of the capsules taken within a day, could also be examined in future trials.An example comes from the Belch et al. (Belch et al. 1988) RCT, where participants divided their twelve daily capsules at their own accord, and the recorded dropout reasons were unrelated to the size or number of capsules.Moreover, collaborations through a multicenter design could assist in achieving a larger study size.Finally, in order to improve the quality of crossover trials, more appropriate reporting formats and analyses, such as analyses for paired data, should be implemented (Dwan et al. 2019).
In summary, although n-3 FA supplements are popular among people with a RA diagnosis, they appear to provide a limited clinical benefit.In light of the present findings, the potential efficacy of n-3 FA supplementation in reducing RA medications, mainly NSAIDs, should be re-considered, alongside the possible effect of ill-executed masking of the participants and outcomes assessors, as observed in many of the primary trials.The use of different types of placebos as comparators, though regarded as a perplexed issue in the past, was not associated with the effect sizes herein.However, caution is required in properly masking the fishy aftertaste of the n-3 FA supplementation in order to avert undermining the blinding of the interventions.In fairness, more research with adequate sample sizes and rigorous methodologies is needed to elucidate the contradictory results seen in this field.

Figure 1 .
Figure 1.PrisMa flow diagram of the studies selection process.

Figure 2 .
Figure 2. Forest plot of rCts on the effect of n-3 fatty acid supplementation in patients with ra regarding the number of tender joint count.ra: rheumatoid arthritis.rCts: randomized controlled trials.

Figure 3 .
Figure 3. Forest plot of rCts on the effect of n-3 fatty acid supplementation in patients with ra, regarding the C-reactive protein concentrations.ra: rheumatoid arthritis.rCts: randomized controlled trials.

Figure 4 .
Figure 4. Forest plot of rCts on the effect of n-3 fatty acid supplementation in patients with ra, regarding the sensation of pain.ra: rheumatoid arthritis.rCts: randomized controlled trials.

Table 1 .
Characteristics of the included rCts comparing n-3 fatty acids supplementation against placebo, in adults with rheumatoid arthritis.resultswerepresented based on the itt; ‡ presented results of Medication change on an itt basis; ¶ mean; § estimated based on the concentration of ePa and dHa of the used product, Pikasol, according to its webpage ("PiKasol PreMiuM" 2020), multiplied by the mean capsules per day (4.5 cap/d); ¥ Presented results of das28 on an itt basis).an'average'estimate of 9.75 g of total n3/dαy and the 44% and 24% concentration of supplements of ePa and dHa, respectively.£estimatedbased on the percentage concentration of ePa/dHa as analyzed by wolyniak(wolyniak et al. 2005)and averaged between the two dosage schemes.oe estimated based on the percentage concentration of ePa/dHa of the capsules and averaged between the two dosage schemes.
aCr, american College of rheumatology (former ara); Αra, american rheumatism association; Caps, capsule; CG, control group; dHa, docosahexaenoic acid; dMards, disease-modifying antirheumatic drugs; ePa, eicosapentaenoic acid; Gi, gastrointestinal; itt, intention to treat; nod, not other defined; nr, not reported; nsaids, non-steroidal anti-inflammatory drugs; PP, Per Protocol; ra, rheumatoid arthritis; rCt, randomized clinical trial; saards: slow-acting antirheumatic drugs; sd, standard deviation; *mean, range; ˆmean ± sd; ƒ no clear information was provided; † ¢ estimated based on Ž estimated based on the average caloric content of the supplementation and the content of Fa concentration of each capsule; ¤ ¢ after personal communication with author surat Komindr.