Postoperative Oral Corticosteroids for Pain Management and Recovery After Total Knee Arthroplasty: A Systematic Review Authors: Thomas Vingerhoets, Imrane Gazali, Kwinten Vingerhoets, Wouter Platteau Published: 2026 ABSTRACT Total knee arthroplasty (TKA) for osteoarthritis is associated with substantial postoperative pain that drives opioid use and delays recovery. Perioperative intravenous corticosteroids reduce early pain, but the role of postoperative oral corticosteroids is less clear. We aimed to evaluate the efficacy and safety of postoperative oral corticosteroids after primary TKA for osteoarthritis. This systematic review was prospectively registered on PROSPERO (CRD420261369550) and reported in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses 2020 statement. We searched PubMed, Embase (Ovid), the Cochrane Central Register of Controlled Trials, ClinicalTrials.gov, and the World Health Organization International Clinical Trials Registry Platform on 28 April 2026 with no language or date restrictions. Randomised controlled trials of oral corticosteroids initiated within 48 hours after primary TKA versus placebo, no oral corticosteroid, or standard care were eligible for primary synthesis; quasi-randomised and non-randomised comparative studies contributed only to sensitivity or narrative analysis. Two reviewers independently screened, extracted data, and assessed risk of bias using the revised Cochrane tool for randomised trials (Risk of Bias (RoB) 2.0). Random-effects meta-analyses (DerSimonian-Laird) were performed where studies were sufficiently homogeneous, and certainty of evidence was rated with GRADE. In total, 12 studies were included: four randomised controlled trials in primary synthesis (380 randomised participants), three trials in sensitivity analysis only, and five non-randomised reports for narrative synthesis. Postoperative oral corticosteroids reduced cumulative opioid consumption in the first 48 hours by approximately 9 oral morphine equivalent milligrams (mean difference = -8.59, 95% confidence interval = -14.59 to -2.60; two trials, 173 participants), but the effect on pain at rest at 48 hours was uncertain (mean difference = -7.17 mm on a 0-100 mm scale, 95% confidence interval = -18.71 to +4.38; two trials, 178 participants). The composite of surgical site or periprosthetic joint infection was indeterminate (risk ratio = 1.37, 95% confidence interval = 0.22 to 8.56; three contributing trials, 274 participants). Length of stay and knee range of motion were reported narratively only. Certainty of evidence was very low for all six prioritised outcomes. Postoperative oral corticosteroids may reduce 48-hour opioid consumption after primary TKA, but the evidence remains very uncertain across all outcomes. No short-term safety signal was observed; rare harms cannot be excluded because the included trials were underpowered for adverse-event detection. Adequately powered randomised trials with standardised dose, duration, and molecule are needed before routine use can be recommended. Introduction and background Primary total knee arthroplasty (TKA) is one of the most frequently performed elective orthopaedic operations worldwide, and demand has continued to grow with population ageing and rising prevalence of knee osteoarthritis [1,2]. Despite well-established improvements in pain and function over the medium and long term, the early postoperative period remains a challenge: many patients experience moderate-to-severe pain in the first 24 to 72 hours, which delays mobilisation, increases opioid consumption, and is associated with persistent postsurgical pain, dissatisfaction, and slower functional recovery [3,4]. Therefore, effective early analgesia is central to enhanced-recovery pathways and to the broader effort to reduce perioperative opioid exposure [5,6]. Multimodal analgesia after TKA combines spinal or general anaesthesia, peripheral nerve blocks, local infiltration analgesia, paracetamol, non-steroidal anti-inflammatory drugs and selective cyclo-oxygenase-2 inhibitors, gabapentinoids in selected patients, and antiemetics [7,8]. Within this framework, perioperative corticosteroids have a plausible biological rationale: surgical trauma activates pro-inflammatory cytokines that contribute to nociception, oedema, and postoperative nausea, and corticosteroids may reduce pain, opioid requirement, and nausea by suppressing this inflammatory cascade [9-12]. Intravenous and oral corticosteroid regimens differ in important ways and should not be assumed to be interchangeable. The most extensively studied intervention in this field is single-dose or multi-dose perioperative intravenous (IV) dexamethasone, for which randomised trials and systematic reviews have shown reductions in early postoperative pain and nausea after TKA with a low frequency of short-term complications [11-13]. By contrast, postoperative oral corticosteroid courses are pharmacokinetically distinct from a single IV dose: they reach the systemic circulation more gradually, can be continued for several days or weeks, and have been proposed as a way to extend the analgesic window of perioperative corticosteroid therapy beyond the early hours after surgery [14,15]. Existing reviews on the broader question have either combined IV and oral routes, mixed perioperative and postoperative timing, included observational and randomised data without separation, or focused on a single agent [6,16,17]. The 2022 combined AAHKS clinical practice guideline addresses corticosteroids in total joint arthroplasty broadly [18] but does not provide a postoperative oral-specific synthesis. The principal clinical uncertainties at the start of this review were therefore three: whether a postoperative oral corticosteroid course reduces early pain and opioid use beyond what perioperative IV corticosteroids already achieve; whether such a course is safe, particularly with respect to infection; and whether any effect is consistent across molecules, doses, and durations. We conducted a systematic review to address these questions. The objective was to evaluate the efficacy and safety of postoperative oral corticosteroids of any molecule, dose, or duration, initiated within 48 hours after primary TKA for osteoarthritis, compared with placebo, no oral corticosteroid, or standard postoperative care. We pre-registered the review on PROSPERO (CRD420261369550), pre-specified six outcomes for GRADE certainty assessment (pain at rest at 48 hours, pain on movement at 48 hours, cumulative opioid consumption at 0-48 hours, the composite of surgical site infection or periprosthetic joint infection, length of hospital stay, and knee range of motion), and pre-specified five sensitivity analyses to test the robustness of any pooled estimates; a sixth sensitivity analysis was added before meta-analysis as a protocol deviation (see Registration and Protocol Deviations under Methodology). Review Methodology Registration and Protocol This systematic review was prospectively registered on PROSPERO (CRD420261369550) and was conducted and reported in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) 2020 statement [19,20]. The completed PRISMA 2020 checklist is provided as a supplementary file (Supplement S1). The protocol was finalised on 18 April 2026 before data extraction. Registration and Protocol Deviations Three deviations from the registered protocol occurred. First, the risk-of-bias assessment tool for randomised controlled trials was changed from the original Cochrane risk-of-bias tool (RoB-1) [21] to the revised Cochrane risk-of-bias tool for randomised trials (RoB 2.0) [22]. This decision was made before risk-of-bias assessment and data synthesis were performed. RoB 2.0 is the current Cochrane-recommended instrument and provides a more structured, domain-based assessment with explicit signalling questions, improving transparency and reproducibility of judgements. Second, during data extraction, we identified that one trial (Kanitnate et al. (2026) [23]) administered the first oral dexamethasone dose preoperatively rather than postoperatively. Although the study otherwise met the eligibility criteria, this represents a deviation from the protocol-specified intervention timing (“initiated within 48 hours after surgery”). We retained this study for sensitivity analysis only and added a sixth sensitivity analysis (exclude trials with perioperative oral corticosteroid initiation) to the originally planned five. This decision was made before meta-analysis was performed. Third, the registered protocol committed to assessing risk of bias for any included non-randomised comparative studies using Risk of Bias in Non-randomized Studies - of Interventions (ROBINS-I) [24]. Five non-randomised reports were retained in this review for narrative synthesis only and were not pooled into any quantitative synthesis. ROBINS-I was held in reserve but not formally applied to these five reports, on the grounds that they were not contributors to any quantitative estimate and that the principal limitations of each report (incomplete molecule, dose, or duration data; non-randomised exposure assignment; residual confounding; and database-derived exposure ascertainment) were already evident from the source publications and are described per study in the narrative section. We acknowledge that this departs from the protocol-committed application of ROBINS-I to all included non-randomised comparative studies and disclose it as a third deviation. The decision was made before data synthesis was performed. No PROSPERO amendment was filed for any of these three deviations; all are disclosed transparently in this manuscript. Quantitative synthesis (meta-analysis) was performed where studies were sufficiently clinically and methodologically homogeneous, as pre-specified in the registered protocol. No changes were made to the eligibility criteria, outcomes, or scope of the review. Eligibility Criteria We followed the PICOS framework. The population comprised adults aged 18 years or older undergoing primary TKA for osteoarthritis. We excluded studies of revision TKA, unicompartmental knee arthroplasty, and bilateral simultaneous TKA when data were not separable; studies of non-osteoarthritis indications (e.g., rheumatoid arthritis, post-traumatic arthritis, neoplasm) when data were not separable; and studies confined to chronic systemic corticosteroid users or patients with adrenal insufficiency. The intervention was a postoperative oral corticosteroid of any type (dexamethasone, prednisolone, prednisone, methylprednisolone, or deflazacort), at any dose and duration, initiated within 48 hours after surgery. The comparator was placebo, no oral corticosteroid, or standard postoperative care without an oral corticosteroid; oral corticosteroid versus IV corticosteroid comparisons were eligible as a secondary comparison and were not pooled with placebo or standard-care comparisons. Randomised controlled trials were eligible for primary synthesis, quasi-randomised trials were eligible for sensitivity analysis only, and non-randomised comparative studies were included for narrative synthesis only and were clearly labelled as lower-level evidence. There were no language or date restrictions. Information Sources and Search Strategy We searched PubMed, Embase (via Ovid), and the Cochrane Central Register of Controlled Trials (CENTRAL), together with the trial registries ClinicalTrials.gov and the World Health Organization International Clinical Trials Registry Platform (WHO ICTRP), on 28 April 2026. Search terms combined controlled vocabulary and free-text terms for total knee arthroplasty or replacement and corticosteroids/glucocorticoids by class and molecule. Route, postoperative timing, outcome, date, language, and study-design restrictions were not applied at the search stage; these criteria were applied during screening. The full reproducible search strategies for each database are provided as a supplementary file (Supplement S1). Database exports were managed in a single master record list and de-duplicated using exact matches on digital object identifier (DOI), PubMed identifier (PMID), and trial registration number, followed by manual review of remaining possible duplicates. Selection Process Records were screened by title and abstract by two independent reviewers using a structured screening workbook with pre-piloted decision rules; disagreements were resolved by discussion and, where necessary, by a third reviewer. Full texts of potentially eligible records were obtained and assessed against the eligibility criteria by two independent reviewers, again with disagreements resolved by discussion or third-reviewer adjudication. Reasons for exclusion at the full-text stage were recorded and tabulated. Data Extraction Data were extracted independently by two reviewers into a pre-piloted structured workbook with a written codebook. Extracted items included study characteristics (design, country, single- or multi-centre status, registration), participant characteristics (sample size, age, sex, body mass index, key comorbidities), intervention details (corticosteroid molecule, dose, frequency, daily dose, route, timing of first dose, duration, and total dose), comparator and co-intervention details (including IV corticosteroid co-intervention and whether such co-intervention was balanced across arms), anaesthesia and analgesic protocol, all reported outcomes with their timepoints, and risk-of-bias-relevant information. Numerical extraction recorded means and standard deviations or medians and interquartile ranges, sample sizes per arm, the pain scale used (and conversion to 0-100 mm where applicable), event counts and totals, and an exact source location (page, table, figure, or supplement) for every value. Discrepancies between extractors were resolved by discussion; outstanding uncertainties were logged and resolved before data synthesis. Numerical pain and range-of-motion data not reported in full in the main paper of Kanitnate et al. [23] were retrieved from the journal’s publicly available Supplementary Digital Content 1. Outcome Definitions and Harmonisation Outcome windows were defined a priori in the registered protocol to enable clinically and methodologically homogeneous pooling. For acute timepoints (24, 48, and 72 hours), we used a tolerance window of ±6 hours; for weekly timepoints (one week and two weeks), we used ±2 days. These windows were chosen to capture the conventional TKA postoperative-pain measurement schedule (postoperative day one morning, postoperative day two morning, end of postoperative day three) while excluding measurements outside the protocol-relevant temporal range. When multiple values fell within a window, we used the value closest to the target timepoint. Pain scores reported on a 0-10 cm visual analogue scale (VAS), or a 0-10 numerical rating scale, were standardised to a 0-100 mm equivalent by multiplying by 10. We used the mean difference (MD) when scales were harmonised and reserved the standardised mean difference for situations where harmonisation was not valid. Opioid consumption was converted to oral morphine equivalent (OME) milligrams using conversion factors from the 2022 Centers for Disease Control and Prevention clinical practice guideline: oral morphine, 1.0; oral oxycodone, 1.5; and intravenous morphine, 3.0 (e.g., one 5 mg oxycodone tablet = 7.5 OME mg) [25]. Co-interventions, in particular perioperative IV corticosteroid administration to both arms (referred to throughout as “balanced IV co-intervention” and indexed by sensitivity flag SA2), were recorded at extraction. Sensitivity analysis SA2 was pre-specified in the protocol on the rationale that an oral corticosteroid course given in addition to balanced IV dexamethasone tests the marginal benefit of the oral component, whereas an oral course given without IV co-intervention tests the standalone effect of an oral regimen; pooling estimates from both contexts assumes that the two effects are equivalent, which they may not be. Where standard deviations were not reported, values were imputed using pre-specified Cochrane Handbook methods, with sensitivity analysis SA3 testing robustness to the imputation. Risk of Bias Assessment Two reviewers independently assessed risk of bias for each randomised and quasi-randomised study using the revised Cochrane RoB 2.0 at the level of the effect of assignment to intervention [22]. Each of the five domains (randomisation process, deviations from intended interventions, missing outcome data, measurement of the outcome, and selection of the reported result) was rated as low risk, some concerns, or high risk on the basis of signalling questions, and an overall judgement was assigned per study. For Domain 5, we directly inspected the corresponding trial registry record (ClinicalTrials.gov, IRCT, CTRI, or ChiCTR) and any published study protocol to compare planned and reported outcomes, timepoints, and analyses. Discrepancies between the two independent assessors were resolved in a documented consensus step before data synthesis. ROBINS-I [24] was held in reserve for non-randomised comparative studies but was not formally applied because such studies were retained only for narrative synthesis and did not contribute to any quantitative estimate. The principal limitations of each non-randomised report (incomplete molecule, dose, or duration data; non-randomised exposure assignment; residual confounding; database-derived exposure ascertainment) were already evident from the source publications and are described per study in the narrative section of the Results. This decision constitutes the third protocol deviation as described above and is disclosed prospectively in the present Methodology so that readers can evaluate its impact on the review’s findings. Statistical Analysis For each pre-specified Grading of Recommendations, Assessment, Development and Evaluation (GRADE) outcome, we performed a random-effects meta-analysis using the DerSimonian-Laird estimator, with 95% confidence intervals (CIs) derived from the Z-distribution, as specified in the registered protocol. The Hartung-Knapp variance adjustment was pre-specified as a robustness check but was not applied: at k = 2 (the case for every continuous pooled outcome), the underlying t(k−1) reference distribution becomes a Cauchy-like t(1) with undefined variance, yielding CIs so wide as to be uninformative regardless of true effect size [26]. Continuous outcomes harmonised to a common scale were pooled as the mean difference. The composite of surgical site infection and periprosthetic joint infection was pooled with the Mantel-Haenszel random-effects risk ratio (RR), with double-zero-event studies excluded from the pooled estimate but reported transparently. Heterogeneity was quantified with I² and tau² and interpreted against Cochrane Handbook thresholds (I² 0-40% possibly unimportant, 30-60% moderate, 50-90% substantial, 75-100% considerable) [27]. Multi-arm trials were handled per Cochrane Handbook §23.3.4 by pooling clinically similar active arms for the main analysis, with dose-specific contrasts reserved for sensitivity analysis. Missing standard deviations were derived from standard errors, confidence intervals, p-values, or interquartile ranges where possible; otherwise, a pooled standard deviation was borrowed from contemporaneous trials reporting the same outcome at the same timepoint, and the borrowing was tested in sensitivity analysis SA3. Six sensitivity analyses were applied (SA1-SA5 pre-specified in the protocol; SA6 added before meta-analysis as a transparent protocol deviation, see Registration and Protocol Deviations): excluding quasi-randomised trials (SA1), studies with balanced IV corticosteroid co-intervention (SA2), studies with imputed standard deviations (SA3), studies at high overall risk of bias (SA4), multi-arm dose-specific contrasts (SA5), and trials with perioperative oral corticosteroid initiation (SA6); a consolidated per-outcome summary appears in Supplement S5. Publication bias was pre-specified as funnel-plot inspection and Egger’s test only at k ≥ 10, neither applicable here; the protocol-specified fallback of trial-registry comparison plus RoB 2.0 Domain 5 assessment was used instead. Certainty of Evidence We rated certainty of evidence with the GRADE approach for six pre-specified outcomes: pain at rest at 48 hours, pain on movement at 48 hours, cumulative opioid consumption at 0-48 hours, the composite of surgical site infection or periprosthetic joint infection, length of hospital stay, and knee range of motion [28,29]. Initial confidence for the body of randomised evidence was set at high; downgrading was assessed across the five GRADE domains (risk of bias, inconsistency, indirectness, imprecision, publication bias) with a conservative posture, in line with the protocol decision to favour a downgrade over an over-optimistic rating where evidence was equivocal. Software Meta-analyses were carried out in Python using the numpy and scipy libraries, implementing the DerSimonian-Laird random-effects estimator for continuous outcomes and the Mantel-Haenszel random-effects method for sparse-event binary outcomes, with 95% CIs computed on the Z-distribution. Forest plots were produced with matplotlib, and the risk-of-bias traffic-light figure was produced with structured-data tooling. Results Study Selection The combined database and registry searches on 28 April 2026 returned 5,279 records. After removal of 1,689 duplicates, 3,590 unique records entered title-and-abstract screening. Following the pre-specified eligibility process, 12 studies (reported across 12 publications) were included: four randomised controlled trials in primary synthesis [30-33], two randomised trials and one quasi-randomised trial in sensitivity analysis only [23,34,35], and five non-randomised reports for narrative synthesis [36-40]. Five further records were retained as linked trial registrations of included studies, six were ongoing or unpublished trials, four were classified as awaiting classification or no published results available (including a targeted multi-database trace of ChiCTR1900023839, which returned no extractable outcome data), and 15 records were excluded with documented reasons. The PRISMA 2020 flow diagram is shown in Figure 1. PRISMA 2020 flow diagram of study identification, screening, and inclusion. PRISMA 2020 flow diagram showing identification, screening, eligibility assessment, and inclusion of studies. The 12 included studies comprise four randomised controlled trials in primary synthesis (Shaw et al. (2023) [30], Premkumar et al. (2026) [31], Ebrahimzadeh et al. (2026) [32], Soundarrajan et al. (2026) [33]), two randomised trials and one quasi-randomised trial in sensitivity-only synthesis (Kanitnate et al. (2026) [23], Springborg et al. (2025) [34], Cheng et al. (2021) [35]), and five non-randomised reports for narrative synthesis (Lam et al. (2026) [36], Tye et al. (2026) [37], Fuqua et al. (2025) [38], Zhuang et al. (2024) [39], Salmons et al. (2023) [40]). Records were identified from PubMed, Embase (via Ovid), the Cochrane Central Register of Controlled Trials, ClinicalTrials.gov, and the World Health Organization International Clinical Trials Registry Platform on 28 April 2026. The full search strategies, per-database yields, and reconciliation of all classification decisions are reported in Supplement S1. CENTRAL = Cochrane Central Register of Controlled Trials; PRISMA = Preferred Reporting Items for Systematic Reviews and Meta-Analyses; RCT = randomised controlled trial; WHO ICTRP = World Health Organization International Clinical Trials Registry Platform Study Characteristics The 12 included studies were conducted between 2021 and 2026 in the United States, Iran, India, Thailand, Denmark, and China, in single-centre and multi-centre settings. The seven randomised and quasi-randomised trials are summarised in Table 1; the five non-randomised comparative reports are summarised narratively below. Table 1Characteristics of the seven included randomised and quasi-randomised trials.BID = twice daily; CS = corticosteroid; dex = dexamethasone; IV = intravenous; OA = osteoarthritis; POD = postoperative day; RA = rheumatoid arthritis; RoB = risk of bias; SA = sensitivity analysis; TKA = total knee arthroplasty Study Country/Setting Population (N) Oral corticosteroid (molecule, dose, duration) Comparator Role in synthesis Overall RoB 2.0 Shaw et al. (2023) [29] USA, single-centre Primary TKA OA (109) Dexamethasone 4 mg BID for 4 days (32 mg total) Placebo Primary; SA2 (balanced IV dex) Some concerns Premkumar et al. (2026) [30] USA, single-centre Primary TKA OA (69) Methylprednisolone Dosepak 24-4 mg over 6 days (84 mg total) Standard care, no oral CS Primary; SA2 + SA4 (high RoB) High Ebrahimzadeh et al. (2026) [31] Iran, single-centre Primary TKA OA (102) Prednisolone 10 mg daily for 14 days (140 mg total); celecoxib in both arms Celecoxib alone (active comparator) Primary Some concerns Soundarrajan et al. (2026) [32] India, single-centre Primary TKA OA (100) Deflazacort 6 mg daily for 21 days (126 mg total) Placebo Primary Some concerns Kanitnate et al. (2026) [23] Thailand, single-centre Primary TKA OA (115) Dexamethasone 16 mg/day or 8 mg/day for 5 days (1 preoperative + 4 postoperative) Placebo Sensitivity only; SA6 (perioperative initiation) Low Springborg et al. (2025) [33] Denmark, multicentre Primary TKA, high pain responders (102) Dexamethasone 24 mg single oral dose on POD 1 evening Placebo; 1 mg/kg IV dex preoperatively in both arms Sensitivity only; SA2 Some concerns Cheng et al. (2021) [34] China, single-centre Primary TKA mixed OA/RA (98) Prednisone 10 mg daily for 14 days (140 mg total) Standard care; balanced IV dex + intra-articular betamethasone in both arms Sensitivity only; SA1 + SA4 (quasi-RCT, high RoB) High Characteristics of the seven included randomised and quasi-randomised trials. BID = twice daily; CS = corticosteroid; dex = dexamethasone; IV = intravenous; OA = osteoarthritis; POD = postoperative day; RA = rheumatoid arthritis; RoB = risk of bias; SA = sensitivity analysis; TKA = total knee arthroplasty Across the four primary-synthesis randomised trials, sample sizes ranged from 69 to 109 participants and the oral corticosteroid was given as dexamethasone (Shaw et al. (2023) [30]; 4 mg twice daily for four days), methylprednisolone (Premkumar et al. (2026) [31]; tapering Medrol Dosepak 24 to 4 mg over six days), prednisolone (Ebrahimzadeh et al. (2026) [32]; 10 mg daily for 14 days, alongside celecoxib in both arms), or deflazacort (Soundarrajan et al. (2026) [33]; 6 mg daily for 21 days). Three of the seven randomised and quasi-randomised trials had a balanced IV corticosteroid co-intervention in both arms (Shaw et al. (2023) [30], Premkumar et al. (2026) [31], and Springborg et al. (2025) [34]; SA2). Kanitnate et al. (2026) [23] was a three-arm trial (oral dexamethasone 16 mg/day, 8 mg/day, and placebo) in which the first oral dose was given preoperatively at 6:00 a.m. on the day of surgery and was therefore reclassified as a sensitivity-only contributor under the disclosed protocol deviation (sensitivity-flag SA6). Cheng et al. (2021) [35] used quasi-randomisation by registration order, included a mixed osteoarthritis and rheumatoid arthritis population in which arms were not separable, and was retained as sensitivity-only (SA1, SA4). Springborg et al. (2025) [34] was a multicentre, double-blind, placebo-controlled trial of a single 24 mg oral dexamethasone dose given on the evening of postoperative day one, restricted to patients selected as high pain responders (pain catastrophising scale >20 and walking VAS >30 at 24 hours) and with 1 mg/kg preoperative IV dexamethasone in both arms, and was therefore retained as sensitivity-only (SA2). The five non-randomised comparative reports were Lam et al. (2026) [36] (United States; retrospective propensity-score-matched cohort of methylprednisolone Medrol Dosepak vs. no oral corticosteroid; n = 366); Tye et al. (2026) [37] (United States; sequential prospective cohort comparing two oral corticosteroid regimens, methylprednisolone Medrol Dosepak vs. oral dexamethasone, with no “no-corticosteroid” control; n = 350); Fuqua et al. (2025) [38] (United States; retrospective cohort of methylprednisolone Medrol Dosepak vs. no oral corticosteroid taper; n = 931); Zhuang et al. (2024) [39] (United States; retrospective administrative claims database analysis of acute oral prednisone use within 30 days postoperatively; n = approximately 963,000, with dose, duration, and molecule not reliably available at the level of individual records); and Salmons et al. (2023) [40] (United States; registry-based 1:2 matched cohort/case-control analysis of arthrofibrosis cases vs. non-stiff controls drawn from the Mayo total joint registry; oral corticosteroid molecule, dose, and duration not specified). Three of the five reports (Salmons et al. (2023) [40], Zhuang et al. (2024) [39], and Tye et al. (2026) [37]) lacked molecule, dose, or duration sufficient for any pooled comparison and were therefore restricted to narrative synthesis. Risk of Bias Of the seven randomised and quasi-randomised trials, one was rated as low overall risk of bias (Kanitnate et al. (2026) [23]), four as some concerns (Shaw et al. (2023) [30], Ebrahimzadeh et al. (2026) [32], Soundarrajan et al. (2026) [33], Springborg et al. (2025) [34]), and two as high risk of bias (Premkumar et al. (2026) [31] and Cheng et al. (2021) [35]). The high-risk overall judgement for Premkumar et al. (2026) reflected Domain 4 (measurement of the outcome): outcome assessors of self-reported pain were not blinded, which the authors acknowledged in the published article. The high-risk overall judgement for Cheng et al. (2021) reflected Domain 1 (randomisation by registration order rather than a random sequence) together with Domains 2 and 4. Concerns under Domain 5 (selection of the reported result) were widespread and were driven by trial-registry deviations: timepoint switching between registered and reported primary endpoints (Shaw et al. (2023); primary endpoint changed from three-week VAS to a postoperative day one to four average), retrospective registration (Ebrahimzadeh et al. (2026), Springborg et al. (2025), Cheng et al. (2021)), and demoted or dropped registered primary outcomes (Soundarrajan et al. (2026); postoperative day three timepoint and local warmth as a registered primary outcome both dropped from the published report). The per-domain and per-study consensus judgements are shown in the risk-of-bias traffic-light figure (Figure 2). Risk of Bias 2 assessments for the seven included randomised and quasi-randomised trials: (a) traffic-light plot per domain and per study; (b) weighted summary plot across studies. (a) Traffic-light plot. (b) Summary plot showing the proportion of studies per risk-of-bias judgement. Domains: D1 = randomisation process; D2 = deviations from intended interventions; D3 = missing outcome data; D4 = measurement of the outcome; D5 = selection of the reported result. Studies are listed by analytical role: the four primary-synthesis RCTs first (in bold; Shaw et al. (2023) [30]; Premkumar et al. (2026) [31]; Ebrahimzadeh et al. (2026) [32]; Soundarrajan et al. (2026) [33]), followed by the three sensitivity-only trials (two RCTs, Kanitnate et al. (2026) [23] and Springborg et al. (2025) [34], and one quasi-RCT, Cheng et al. (2021) [35]). Domain 5 was judged against trial-registry records where available. RCT = randomised controlled trial; RoB = risk of bias Pain at Rest at 48 Hours (Primary Outcome) Two primary-synthesis trials reported a postoperative day two morning diary VAS pain score; by a pre-specified analytical decision made before meta-analysis, this measurement was selected as the primary pain-at-rest contribution at 48 hours (Shaw et al. (2023) [30]; Premkumar et al. (2026) [31]). Pooled across these two trials (178 participants), the MD was -7.17 mm on a 0-100 mm scale (95% CI = -18.71 to +4.38; I² = 73%; tau² = 50.74; p = 0.224). The estimate was not statistically significant and was driven by Shaw et al. (2023) alone (MD = -12.7 mm, p < 0.001), with Premkumar et al. (2026) alone showing no detectable effect (MD = -0.9 mm, p = 0.85). The two contributing studies had directionally inconsistent point estimates, indicating substantive between-study heterogeneity beyond what would be expected from random imprecision. Forest plots are shown in Figure 3. Forest plot of pain at rest at 48 hours postoperatively, oral corticosteroids versus control (primary outcome). Trials shown: Shaw et al. (2023) [30] and Premkumar et al. (2026) [31]. Random-effects DerSimonian-Laird pooling with 95% confidence intervals derived from the Z-distribution. Pooled MD = −7.17 mm (95% CI = −18.71 to +4.38); k = 2; N = 178. Test for overall effect: Z = −1.22; p = 0.224. Heterogeneity: I² = 72.9%; τ² = 50.74. GRADE certainty: very low. Premkumar et al. (2026): standard deviation imputed using the pre-specified value of 20 mm (see Methods). The diamond represents the random-effects pooled estimate. CI = confidence interval; CS = corticosteroid; MD = mean difference; SD = standard deviation; VAS = visual analogue scale Ebrahimzadeh et al. (2026) [32] and Soundarrajan et al. (2026) [33] did not report a pain-at-rest measurement within the protocol-specified 48-hour ±6-hour window (earliest reported timepoints were week one and month one, respectively) and therefore did not contribute to the primary synthesis at 48 hours. Pain on Movement at 48 Hours No primary-synthesis trial reported pain on movement at 48 hours within the protocol-specified window; consequently, no primary-synthesis meta-analysis was performed for this outcome. A sensitivity-only pool combining Kanitnate et al. (2026) [23] (combined active arms vs. placebo, 48 hours, on motion) and a post-hoc continuous walking VAS analysis from Springborg et al. (2025) [34] yielded an MD of -8.62 mm (95% CI = -13.95 to -3.30; two trials, 217 participants; I² = 0%; p = 0.0015). This estimate is reported only as a sensitivity-only contribution because neither contributing study is in the protocol primary-synthesis set: Kanitnate et al. initiated oral dexamethasone preoperatively (SA6), and Springborg et al. recruited a selected high-pain-response cohort and used balanced 1 mg/kg preoperative IV dexamethasone in both arms (SA2). Forest plots are shown in Figure 4. Forest plot of pain on movement at 48 hours postoperatively, oral corticosteroids versus control (sensitivity analysis). Mean difference on a 0–100 mm VAS-equivalent scale; random-effects DerSimonian-Laird pooling with 95% confidence intervals on the Z-distribution. Pooled MD = −8.62 mm (95% CI = −13.95 to −3.30); k = 2; N = 217. Test for overall effect: Z = −3.17; p = 0.0015. Heterogeneity: I² = 0%; τ² = 0.00. GRADE certainty: very low. Sensitivity-only synthesis: contributing studies are Kanitnate et al. (2026) [23] (combined active arms; SA6: perioperative initiation) and Springborg et al. (2025) [34] (post-hoc walking VAS at 48 hours; SA2: balanced IV co-intervention); both studies are excluded from the primary-synthesis set. The diamond represents the random-effects pooled estimate. CI = confidence interval; CS = corticosteroid; IV = intravenous; MD = mean difference; SA = sensitivity analysis; VAS = visual analogue scale Cumulative Opioid Consumption at 0-48 Hours Two primary-synthesis trials reported cumulative opioid consumption from 0 to 48 hours postoperatively in a form that could be harmonised. Shaw et al. (2023) [30] reported the total number of 5 mg oxycodone tablets consumed on postoperative days one and two; we converted to oral morphine equivalent milligrams using a factor of 1.5 (5 mg oxycodone = 7.5 OME mg), per the 2022 CDC clinical practice guideline [25]. Premkumar et al. (2026) [31] reported daily and cumulative oral morphine equivalent values directly. Standard deviations for cumulative opioid consumption were not reported in either trial; a pooled standard deviation of 15 OME mg was borrowed for both trials, and the borrowing was tested in SA3 (which collapsed the pool to k = 0; see Supplement S5). Pooled across both trials (173 participants), the MD was -8.59 OME mg (95% CI = -14.59 to -2.60; I² = 43%; tau² = 8.10; p = 0.005). Because both contributing standard deviations were imputed and the SA3 result collapses the pool, this pooled estimate should be regarded as exploratory rather than confirmatory. Forest plots are shown in Figure 5. Forest plot of cumulative opioid consumption 0–48 hours postoperatively (OME mg), oral corticosteroids versus control. Mean difference in oral morphine equivalent (OME) milligrams; random-effects DerSimonian-Laird pooling with 95% confidence intervals on the Z-distribution. Pooled MD = −8.59 OME mg (95% CI = −14.59 to −2.60); k = 2; N = 173. Test for overall effect: Z = −2.81; p = 0.005. Heterogeneity: I² = 42.7%; τ² = 8.10. GRADE certainty: very low. Shaw et al. (2023) [30] and Premkumar et al. (2026) [31]: standard deviation imputed using the pre-specified value of 15 OME mg (see Methods). Shaw et al. (2023): oxycodone converted to oral morphine equivalents using a conversion factor of 1.5 [25]. The diamond represents the random-effects pooled estimate. CDC = Centers for Disease Control and Prevention; CI = confidence interval; CS = corticosteroid; MD = mean difference; OME = oral morphine equivalent Other Pain Timepoints Pain at 24 hours, 72 hours, 1 week, and 2 weeks generally followed the same direction as the 48-hour pain-at-rest estimate, with point estimates favouring oral corticosteroids in Shaw et al. (2023) [30] and Kanitnate et al. (2026) [23], and small or no detectable estimates in Premkumar et al. (2026) [31] and Soundarrajan et al. (2026) [33] at the timepoints they reported. Ebrahimzadeh et al. (2026) [32] reported significantly lower VAS pain in the prednisolone-plus-celecoxib arm versus celecoxib alone from week one onwards. Composite of Surgical Site or Periprosthetic Joint Infection All four primary-synthesis trials reported the composite of surgical site infection or periprosthetic joint infection within their respective postoperative follow-up windows. Pooled events were sparse: two events in 141 participants in the oral corticosteroid arms (one superficial cellulitis at five weeks postoperatively in Premkumar et al. (2026) [31]; one cellulitis in Soundarrajan et al. (2026) [33]) and one event in 133 participants in the comparator arms (one periprosthetic joint infection at 90 days in Shaw et al. (2023) [30]). Ebrahimzadeh et al. (2026) [32] reported zero infection events in both arms and was therefore excluded as a double-zero study from the random-effects Mantel-Haenszel risk-ratio pool but was retained transparently in the contributor count. The pooled estimate was an RR of 1.37 (95% CI = 0.22 to 8.56; four trials reported the outcome (373 participants total); three non-double-zero trials contributed to the pooled estimate (274 participants); I² = 0%; p = 0.736). Forest plots are shown in Figure 6. Forest plot of the composite of surgical site infection or periprosthetic joint infection following primary total knee arthroplasty. Trials shown: Shaw et al. (2023) [30]; Premkumar et al. (2026) [31]; Ebrahimzadeh et al. (2026) [32]; Soundarrajan et al. (2026) [33]. Mantel-Haenszel random-effects risk ratio with 0.5 continuity correction. The double-zero study (Ebrahimzadeh et al. (2026)) is listed among the four primary-synthesis studies reporting SSI/PJI but does not contribute to the pooled estimate. Pooled RR = 1.37 (95% CI = 0.22 to 8.56); k = 4 reported, k = 3 pooled. Events: 2/141 (oral corticosteroids) versus 1/133 (control); N = 274 in the pooled estimate. Test for overall effect: Z = 0.34; p = 0.736. Heterogeneity: I² = 0%. GRADE certainty: very low. The diamond represents the random-effects pooled estimate. CI = confidence interval; CS = corticosteroid; MH = Mantel-Haenszel; RR = risk ratio; SSI/PJI = surgical site infection/periprosthetic joint infection The 39-fold-wide CI reflects the small number of events; the included trials were not powered to detect uncommon adverse events. Across all included randomised studies (including sensitivity-only contributors), seven infection events were observed in 688 participants (or 590 participants if the quasi-randomised trial is excluded). Springborg et al. (2025) [34] separately reported two readmissions for periprosthetic joint infection within 30 days in the oral dexamethasone arm versus zero in the placebo arm (49 vs. 53 participants); the authors explicitly stated that their trial was not adequately powered to compare infection rates. Other Adverse Events Postoperative nausea, when reported (Shaw et al. (2023) [30], Premkumar et al. (2026) [31], Springborg et al. (2025) [34]), tended to favour oral corticosteroids, but standard deviations were inconsistently reported, and a quantitative synthesis was not feasible. Hyperglycaemia, gastrointestinal adverse events, and venous thromboembolism were reported variably across trials and are summarised per study in Supplement S7. Length of Hospital Stay Length of hospital stay was reported as an outcome only by Springborg et al. (2025) [34], which dichotomised stays as longer than two days (one of 49 oral dexamethasone vs. one of 53 placebo; chi-squared p = 0.648). No primary-synthesis trial reported continuous length-of-stay data in a form suitable for pooling. Length of stay is therefore reported narratively only. Knee Range of Motion Knee range of motion was reported by three randomised trials, but at different timepoints and could not be pooled. Premkumar et al. (2026) [31] reported three-month flexion of approximately 117.1° in the methylprednisolone arm versus 120.0° in the no-oral-corticosteroid arm (between-group difference = approximately -2.9°). Ebrahimzadeh et al. (2026) [32] reported flexion at week four of 104.10° in the prednisolone-plus-celecoxib arm versus 97.44° in the celecoxib-alone arm (between-group difference = approximately +6.66°). Kanitnate et al. (2026) [23] reported flexion at 48 hours of 104° (DEX-16) and 101° (DEX-8) versus 100° (placebo). The point estimates are inconsistent in direction across studies and span timepoints from 48 hours to 3 months; all are within or close to the 5° to 10° range typically considered the minimum important difference for knee range of motion after TKA. Knee range of motion is reported per study per timepoint; no pooled GRADE-level estimate is presented. Other Functional Outcomes C-reactive protein was lower with oral corticosteroids in Soundarrajan et al. (2026) [33] and Cheng et al. (2021) [35] in the early postoperative period. Patient satisfaction and functional scores (KOOS JR, Oxford Knee Score) were reported heterogeneously and showed no consistent direction. Broad-Pool Transparency Analysis As a transparency analysis only, broadening the pool to include the combined active arms of Kanitnate et al. (2026) [23] (sensitivity-only) yielded a pooled MD of -9.45 mm (95% CI = -15.69 to -3.20; three trials, 293 participants; I² = 57%; p = 0.003; Supplementary S9). This is reported here for completeness and is not the primary result. Inclusion of Kanitnate et al. (2026) in this transparency pool is sensitivity-only because their first oral dexamethasone dose was given preoperatively (SA6). Sensitivity Analyses Summary A consolidated per-outcome summary of all six sensitivity analyses (SA1-SA6) is provided in Supplement S5. In brief, SA2 (excluding balanced IV co-intervention) collapsed the pool to k = 0 for pain at rest at 48 hours and for cumulative opioid consumption at 0-48 hours; SA3 (excluding imputed standard deviations) collapsed the cumulative opioid pool to k = 0 and reduced the pain-at-rest pool to Shaw et al. (2023) alone (MD = -12.70 mm, 95% CI = -20.18 to -5.22); SA4 (excluding high-overall-RoB studies) reduced the pain-at-rest pool to Shaw et al. (2023) alone; SA5 dose-specific contrasts for Kanitnate et al. (2026) yielded -12.00 mm (95% CI = -17.59 to -6.41) for DEX-16 versus placebo and -12.00 mm (95% CI = -17.41 to -6.59) for DEX-8 versus placebo for pain at rest at 48 hours, and -13.00 mm versus -5.00 mm for pain on motion. The surgical site infection/periprosthetic joint infection composite was robust to all sensitivity analyses applied because Cheng et al. (2021) reported zero events, and Kanitnate et al. (2026) did not contribute to the primary set. The pooled estimates that survived no more than one sensitivity test, or which collapse under SA3, should be regarded as exploratory rather than confirmatory. Reporting Bias With a maximum of four studies contributing to any pooled outcome, neither funnel-plot inspection nor Egger’s test was applicable, as pre-specified in the protocol. Instead, we relied on the per-study Domain 5 RoB 2.0 assessment, supported by direct inspection of trial-registry records or published study protocols for all seven randomised and quasi-randomised trials. Six of the seven trials showed registry-paper deviations of varying magnitude, ranging from minor (e.g., FJS-12 added in the published Kanitnate et al. study [23]) to substantial (e.g., postoperative day three timepoint and local warmth dropped in Soundarrajan et al. (2026) [33]; primary endpoint redefined in Shaw et al. (2023) [30]). Three trials were registered retrospectively (Ebrahimzadeh et al. (2026) [32] by approximately 13 months; Springborg et al. (2025) [34] by approximately 11 months on ClinicalTrials.gov but supported by a prospectively published study protocol; Cheng et al. (2021) [35] by approximately 12 months). These deviations contributed to Domain 5 “some concerns” ratings in five of seven trials and were carried forward into the GRADE Domain 1 (study limitations) downgrade for affected outcomes. Certainty of Evidence All six pre-specified GRADE outcomes were rated as having very low certainty (Table 2). Detailed per-domain GRADE rationale per outcome is provided in Supplement S6. Table 2GRADE summary of findings.GRADE Working Group grades of certainty: high = we are very confident the true effect lies close to that of the estimate; moderate = moderately confident; low = limited confidence; very low = very little confidence in the estimate.CI = confidence interval; CS = corticosteroid; LoS = length of stay; MD = mean difference; MID = minimum important difference; OME = oral morphine equivalent; PJI = periprosthetic joint infection; RR = risk ratio; SA = sensitivity analysis; SSI = surgical site infection; VAS = visual analogue scale Outcome Participants (studies) Pooled effect (95% CI) Anticipated absolute effect Certainty (GRADE) Comments Pain at rest, 48 hours postoperatively (0–100 mm VAS-equivalent; lower = less pain) 178 (2 RCTs) MD = −7.17 mm (95% CI = −18.71 to +4.38); I² = 73%; p = 0.224 Mean pain at rest with placebo/no oral CS at 48 hours was approximately 50 mm. Oral CS may reduce mean pain by 7 mm on a 0–100 mm scale, but the 95% CI is consistent with anything from a meaningful 19 mm reduction to a trivial 4 mm increase Very low Two studies with substantial heterogeneity (I²= 73%); effect driven by Shaw et al. (2023) [29] alone (MD = −12.7 mm, p < 0.001); Premkumar et al. (2026) [30] alone null. Both have balanced perioperative IV dexamethasone in both arms (SA2). Premkumar et al. rated high overall RoB Pain on movement, 48 hours postoperatively (0–100 mm VAS-equivalent; lower = less pain) 217 (2 RCTs, sensitivity-only pool) MD = −8.62 mm (95% CI = −13.95 to −3.30); I² = 0%; p = 0.0015 Mean pain on movement with placebo/no oral CS at 48 hours was approximately 30–50 mm depending on cohort. Oral CS may reduce mean pain by approximately 9 mm (95% CI = 3 to 14 mm) Very low No primary-synthesis study reported pain on movement at 48 hours. Estimate comes from a sensitivity-only pool (Kanitnate et al. (2026) [23] SA6; Springborg et al. (2025) [33] SA2 post-hoc continuous walking VAS). Direct evidence in the protocol-specified PICOS is therefore absent Cumulative opioid consumption, 0–48 hours postoperatively (oral morphine equivalent, mg) 173 (2 RCTs) MD = −8.59 OME mg (95% CI = −14.59 to −2.60); I² = 43%; p = 0.005 Mean cumulative opioid with placebo/no oral CS in the first 48 hours was approximately 50 OME mg. Oral CS may reduce mean consumption by approximately 9 OME mg (95% CI = 3 to 15 mg) — equivalent to roughly 1–2 5 mg oxycodone tablets Very low Both contributing studies have imputed or borrowed standard deviations; SA3 collapses pool to k = 0. Both have balanced IV dexamethasone (SA2). Lower CI bound is below typical MID estimates for cumulative OME SSI/PJI (composite) 373 (4 RCTs reported; k_used = 3 in pooled estimate) RR = 1.37 (95% CI = 0.22 to 8.56); I² = 0%; p = 0.736 Assuming a baseline SSI/PJI rate of approximately 7 to 10 per 1,000 in the control arm, the absolute effect of oral CS would correspond to from about 6 fewer to about 56 more per 1000 — i.e., the data are compatible with both a substantial reduction and a substantial increase Very low Sparse-event setting: 2/141 (oral CS) vs. 1/133 (control) in primary RR pool; 7 events across the seven randomised studies (688 participants total; or 590 if the quasi-randomised trial is excluded). ~39-fold CI. Trials underpowered for safety outcomes; rare harms cannot be excluded Length of hospital stay 102 (1 RCT, dichotomised) 1/49 vs. 1/53 (>2 d); p = 0.648 Springborg et al. (2025) [33] reported similar proportions discharged >2 days postoperatively in both arms (approximately 2%). Continuous LoS data not pooled Very low Narrative outcome only. The single contributing study is sensitivity-only (SA2 + selected high-pain-response population); outcome dichotomised; only 2 events total Knee range of motion (any timepoint) Per-study per-timepoint; k = 1 at each timepoint Premkumar et al. [30] 3 mo: MD approximately −2.9 deg; Ebrahimzadeh et al. [31] week 4: MD approximately +6.66 deg; Kanitnate et al. [23] 48 hours (sens): MD approximately +2.5 deg Per-study estimates are inconsistent in direction and span timepoints from 48 hours to 3 months. No coherent absolute effect can be stated Very low Narrative only. No timepoint has k >= 2 primary contributors. All point estimates are within or close to the typical 5–10 deg MID for knee ROM after TKA GRADE summary of findings. GRADE Working Group grades of certainty: high = we are very confident the true effect lies close to that of the estimate; moderate = moderately confident; low = limited confidence; very low = very little confidence in the estimate. CI = confidence interval; CS = corticosteroid; LoS = length of stay; MD = mean difference; MID = minimum important difference; OME = oral morphine equivalent; PJI = periprosthetic joint infection; RR = risk ratio; SA = sensitivity analysis; SSI = surgical site infection; VAS = visual analogue scale Discussion This systematic review of 12 studies (seven randomised or quasi-randomised, five non-randomised) found that postoperative oral corticosteroids may reduce cumulative opioid consumption in the first 48 hours after primary TKA by approximately 9 oral morphine equivalent milligrams, corresponding to roughly one to two 5 mg oxycodone tablets, but the certainty of this estimate is very low and the pool is fragile: both contributing trials required borrowed standard deviations, and sensitivity analysis SA3 collapses the pool. The effect on the primary outcome of pain at rest at 48 hours was not statistically significant; the pooled MD of -7.17 mm on a 0-100 mm scale had a CI that crossed zero and was driven entirely by Shaw et al. (2023) [30], with Premkumar et al. (2026) [31] alone showing no detectable effect. The composite of surgical site infection and periprosthetic joint infection was sparsely reported and indeterminate, with a CI consistent with both substantial harm and substantial benefit. Length of stay and knee range of motion did not lend themselves to pooled estimation. The certainty of the evidence is very low for all six pre-specified outcomes. Anchoring the pooled estimates against minimum important differences provides context. For acute postoperative pain measured on a 0-100 mm VAS, a between-group MD of approximately 10 mm is widely taken as the minimum clinically important difference in this setting; this anchor derives from a postoperative VAS study using anchor-based and distribution-based methods [41] and is consistent with the range reported in a systematic review of minimum clinically important differences in randomised pain-management trials after total hip and knee arthroplasty [42]. For cumulative postoperative opioid consumption converted to OME milligrams, no single gold-standard minimum clinically important difference exists, and threshold values reported in TKA and total hip arthroplasty trials cluster broadly around 5 to 10 OME mg over the first 24-48 hours [42]. Applied to the present review, the pooled pain-at-rest estimate (-7.17 mm; 95% CI = -18.71 to +4.38) does not exclude a clinically important reduction but is also compatible with no effect or a trivially small increase; the pooled cumulative opioid estimate (-8.59 OME mg; 95% CI = -14.59 to -2.60) is directionally favourable but the lower bound of plausible benefit lies just at the upper edge of minimum clinically important difference estimates, and the SA3 collapse limits confidence in the estimate’s precision. Our findings sit within, and partially clarify, a heterogeneous prior literature. Earlier reviews on perioperative corticosteroids in TKA have either focused on IV regimens, combined IV and oral regimens, or pooled randomised and observational data, and several have not stratified by the timing of corticosteroid initiation [6,11,16-18]. The body of randomised evidence demonstrating that perioperative IV dexamethasone reduces early pain and nausea after TKA [11-13] is informative about the pharmacological class but cannot be transposed directly to postoperative oral regimens, which differ in route, timing, duration, and patient exposure profile; the inferences in the present review are limited to oral regimens initiated within 48 hours after surgery. By restricting the primary synthesis to randomised trials of postoperative oral initiation in osteoarthritis populations, applying a strict 48-hour timepoint with a ±6-hour window, and treating studies that fall outside this PICOS as sensitivity-only contributors, we have isolated a smaller, more homogeneous evidence base than is typical for this question. The trade-off is small numbers per outcome (a maximum of four studies for the surgical-site or periprosthetic-joint-infection composite) and a correspondingly limited capacity to detect or rule out modest effects. Three of the seven randomised and quasi-randomised trials included balanced perioperative IV corticosteroid in both arms, which means that any oral effect estimated from those trials must be interpreted as the marginal benefit of an oral course on top of IV corticosteroid; this is a common-but-imperfect proxy for the standalone effect of an oral regimen and is the central indirectness consideration in the GRADE assessment. Strengths This review has several methodological strengths. The protocol was prospectively registered on PROSPERO and finalised before data extraction; all three deviations from the registered protocol (the change from RoB-1 to RoB 2.0, the addition of a sixth sensitivity analysis for perioperative oral initiation, and the non-application of ROBINS-I to the five narrative-only non-randomised reports) were decided before data synthesis and are disclosed in full. Screening and data extraction were performed in duplicate with documented disagreement resolution; risk of bias was assessed independently by two reviewers using the current Cochrane RoB 2.0 instrument, with direct inspection of the corresponding trial registry record for every randomised trial. All six sensitivity analyses were specified before meta-analysis, and every numerical input to the meta-analysis is traceable to a specified row, table, or figure in the original publication or its supplementary digital content. Numerical inputs were verified against the source publications through an independent cross-check before finalisation. Limitations The body of randomised evidence directly addressing postoperative oral corticosteroids in primary TKA for osteoarthritis is small: a maximum of four primary-synthesis trials contribute to any pooled outcome, and three of those four had balanced perioperative intravenous corticosteroid in both arms, so the pooled efficacy estimates approximate the marginal effect of an added oral course rather than the standalone effect of an oral regimen versus no corticosteroid. There is also substantial clinical heterogeneity in molecule (dexamethasone, methylprednisolone, prednisolone, prednisone, deflazacort), dose (4 mg to 24 mg dexamethasone-equivalent per day), duration (single dose to 21 days), and tapering regimen, which limits the strength of any pooled inference. Standard deviations were not reported in either of the two primary-synthesis trials contributing to the cumulative opioid pool, requiring a borrowed pooled standard deviation; the pool is therefore not robust to imputation assumptions, as documented by SA3 collapsing it to k = 0. Infection events were sparse: three events in 274 primary-pool participants and seven events in all 688 randomised participants (or 590 if the quasi-randomised trial is excluded), well below any threshold at which the included trials could detect or rule out a clinically important difference; the broader perioperative-corticosteroid safety meta-analysis in total joint arthroplasty [6] is reassuring at the typical IV doses studied in that synthesis but cannot be extrapolated to the small randomised postoperative oral corticosteroid evidence base of the present review. Six of the seven randomised and quasi-randomised trials showed registry-paper deviations of varying magnitude (including timepoint switching, retrospective registration, and demoted or dropped primary outcomes), and three were registered retrospectively; these issues collectively contributed to Domain 5 “some concerns” ratings and to the GRADE risk-of-bias downgrade. All seven trials were single-centre, except Springborg et al. (2025) [34], so results may not generalise to other systems of care. The five non-randomised comparative reports had molecule, dose, and duration data that were either incomplete or did not include a no-corticosteroid comparator and were used only as low-level narrative evidence; ROBINS-I was held in reserve but not formally applied because they did not contribute to any quantitative estimate, a non-application disclosed as the third protocol deviation. Finally, no quantitative test for small-study effects or publication bias was applicable because the maximum number of contributing studies for any outcome was four. Conclusions Postoperative oral corticosteroids initiated within 48 hours after primary TKA for osteoarthritis may reduce cumulative opioid consumption in the first 48 hours by approximately 9 oral morphine equivalent milligrams, but the certainty of this estimate is very low, and the pooled effect on pain at rest at 48 hours is uncertain. Across all six pre-specified GRADE outcomes, the certainty of evidence is very low. No short-term safety signal was observed in this small body of randomised evidence, but rare harms cannot be excluded because the included trials were underpowered to detect uncommon adverse events. Based on this evidence, the routine addition of a postoperative oral corticosteroid course to a contemporary multimodal analgesia regimen after primary TKA cannot be recommended. Adequately powered randomised trials with standardised molecule, dose, duration, and tapering regimen, with prospectively registered safety endpoints and sample sizes capable of detecting clinically important differences in surgical site and periprosthetic joint infection, and with strict separation of perioperative intravenous corticosteroid co-intervention from the oral course under test, are needed before postoperative oral corticosteroids can be either recommended or rejected for routine use in this setting.