Are perioperative interventions effective in preventing chronic pain after primary total knee replacement? A systematic review

Objectives For many people with advanced osteoarthritis, total knee replacement (TKR) is an effective treatment for relieving pain and improving function. Features of perioperative care may be associated with the adverse event of chronic pain 6 months or longer after surgery; effects may be direct, for example, through nerve damage or surgical complications, or indirect through adverse events. This systematic review aims to evaluate whether non-surgical perioperative interventions prevent long-term pain after TKR. Methods We conducted a systematic review of perioperative interventions for adults with osteoarthritis receiving primary TKR evaluated in a randomised controlled trial (RCT). We searched The Cochrane Library, MEDLINE, Embase, PsycINFO and CINAHL until February 2018. After screening, two reviewers evaluated articles. Studies at low risk of bias according to the Cochrane tool were included. Interventions Perioperative non-surgical interventions; control receiving no intervention or alternative treatment. Primary and secondary outcome measures Pain or score with pain component assessed at 6 months or longer postoperative. Results 44 RCTs at low risk of bias assessed long-term pain. Intervention heterogeneity precluded meta-analysis and definitive statements on effectiveness. Good-quality research provided generally weak evidence for small reductions in long-term pain with local infiltration analgesia (three studies), ketamine infusion (one study), pregabalin (one study) and supported early discharge (one study) compared with no intervention. For electric muscle stimulation (two studies), anabolic steroids (one study) and walking training (one study) there was a suggestion of more clinically important benefit. No concerns relating to long-term adverse events were reported. For a range of treatments there was no evidence linking them with unfavourable pain outcomes. Conclusions To prevent chronic pain after TKR, several perioperative interventions show benefits and merit further research. Good-quality studies assessing long-term pain after perioperative interventions are feasible and necessary to ensure that patients with osteoarthritis achieve good long-term outcomes after TKR.

For the first time, this systematic review brings together contemporary evidence on aspects of peri-operative care for people with total knee replacement and their effects on longterm pain.
• Only studies assessed to be at low risk of bias were included in the narrative synthesis.
• Intervention and outcome heterogeneity precluded meta-analysis.

KEYWORDS
Total knee replacement; Systematic review; Randomised controlled trial; Peri-operative care; Long-term pain

BACKGROUND
In the US about 13% of men and 19% of women will be diagnosed with knee osteoarthritis and over half will receive a total knee replacement (TKR) [1]. For people with advanced osteoarthritis unresponsive to pharmacological or conservative treatments, TKR aims to relieve pain and improve function. In the UK nearly 100,000 primary TKRs were performed in 2017 [2,3] and in the USA in 2010, an estimated 4.7 million people were living with a TKR [4]. Despite good outcomes for many, some people report long-term pain and are disappointed with their surgery [5,6]. After TKR, pain levels plateau from about 6 months [7,8] after which persistent pain is considered "chronic" [9] and is reported by 10-34% of patients [10].
In the peri-operative period from hospital admission to the early stages of recovery, care focuses on acute pain management, prevention of adverse events, facilitation of early mobilisation and timely discharge. However, for people with osteoarthritis the key aim of TKR is the achievement of a long-term painless and well-functioning knee with no adverse events. All aspects of peri-operative care should work together to achieve this.
Peri-operative risk factors suggest that appropriate interventions may reduce long-term pain.
For example, acute post-operative pain, which may be a direct consequence of the operation, anaesthetic protocol and subsequent analgesia, or related to particular aspects of care, is an acknowledged risk factor for chronic post-surgical pain [11]. Any treatment in the peri-operative period could potentially affect patient recovery and chronic pain, either directly or indirectly.
Direct benefits may be through prevention of nerve damage [12], post-thrombotic syndrome [13], reperfusion injury [14] and articular bleeding [15]. Patients with depression and catastrophising have poor pain outcomes [16,17]. For other treatments, pathways leading to long-term pain may be indirect consequences of delayed mobilisation, rehabilitation and recovery.
Our systematic review of randomised controlled trials (RCTs) aims to evaluate the effectiveness of treatments in the peri-operative period in preventing long-term pain after TKR. By focusing on studies with low risk of bias we aim to identify interventions with robust evidence of long-term effectiveness and identify gaps in the research base.

METHODS
The systematic review protocol was registered (PROSPERO CRD42017041382) and PRISMA reporting guidelines used [18]. A checklist is included as Supplementary material. outcomes. Our patient advisory group comprises five patients with experience of long-term pain after TKR, supported by a dedicated co-ordinator. This group will advise on dissemination of the study results to a general audience including plain language summaries.

Eligibility criteria
Participants: adults receiving unilateral primary TKR, predominantly for osteoarthritis.
Interventions: peri-operative interventions (pharmacological or non-pharmacological) were included. "Peri-operative" reflects the time from hospital admission to early stages of recovery.
Interventions relating to implant designs and surgical procedures were excluded.
Comparator: usual care, placebo or alternative intervention.
Outcomes: in preference, patient-reported joint-specific pain intensity measured by tools such Setting: RCTs with follow up at ≥6 months after surgery and a pain outcome or score including pain. Authors of studies were contacted regarding incomplete pain outcome data.

Database searches
We established an Endnote database of all RCTs in TKR. On 6 We imported records into Endnote X7 (Thomson Reuters). An initial screen by one reviewer excluded clearly irrelevant articles. Subsequently, abstracts and full articles were screened independently by two reviewers and reasons for exclusion recorded.
Data were extracted onto piloted forms and an Excel spreadsheet by one reviewer, specifically: country; dates; participants (indication, age, sex); inclusion and exclusion criteria; intervention and control content; setting, timing, duration and intensity of intervention; follow up intervals; losses to follow up; pain outcome data; and serious adverse events. Data was checked against source material by a second reviewer.
Authors were contacted for missing data, and data provided for previous reviews was used [10,19].

Quality assessment
Potential sources of bias were assessed by two experienced reviewers using the Cochrane risk of bias tool [20], specifically: the randomisation process; deviations from intended interventions; missing outcome data, measurement of the outcome; and selection of the reported result.
Studies with serious concerns relating to risk of bias were considered high risk and those with limited reporting unclear risk. Studies with high or unclear risk of bias were excluded from the narrative synthesis but are included in supplementary summary tables with reasons for exclusion.

Data analysis
Insufficient studies with similar interventions and outcomes were identified for meta-analysis, and a narrative synthesis is presented. Results reported with p-values ≤0.001 were considered "strong" evidence of effectiveness [21], p-values 0.001-0.05 "some" evidence, and p-values 0.05-0.1 "weak" evidence. When authors reported results "statistically significant" with no p-value, this was noted. Where possible, effect sizes were compared with published minimal clinically important differences (MCID). Concerns relating to adverse events were summarised. Details of 44 studies assessed to be at low risk of bias are summarised in Table 1.  Peng et al. 2014 [26] China, Before 2014, Transfusions: concern late tourniquet start in groups 1 and 2

Pain management
We identified 20 RCTs evaluating components of multi-modal pain management.

Femoral nerve block
Femoral nerve blocks (FNB) were studied in 10 RCTs.
Three RCTs compared FNB with no FNB. In one study with 55 patients, WOMAC pain scores at one year were similar in patients receiving single-shot FNB and untreated controls [22]. All patients received local anaesthetic infiltration (LIA) and patient-controlled analgesia (PCA). In another study with all participants receiving LIA, 150 were randomised to receive single-shot FNB with or without sciatic nerve block (SNB), or general anaesthesia [23]. There were no differences in HSS scores between groups at six months. Continuous FNB was compared with oral hydrocodone opioid in 62 patients receiving PCA [24]. There was some evidence for 'pain using stairs' favouring hydrocodone (p=0.01) but no difference in overall NRS-rated pain at one year and concern over venous thromboembolism in 4/31 participants treated with hydrocodone.
In two RCTs, continuous FNB was compared with PCA. In one study with 60 participants, the KSS at six months was similar between groups [25]. In another study with 280 participants, there was some evidence for higher incidence of NRS-rated pain at six months in the PCA group than the FNB group (p=0.021) but not at 12 months (p=0.273). [26] Two RCTs compared FNB with LIA. In one study, all 157 participants also received PCA [27]. At one year, KSS values were similar in single-shot FNB and LIA groups. In the other study, 94 participants were randomised to receive single-shot FNB with continuous epidural infusion or LIA through an intra-articular catheter [28]. VAS-rated pain was similar between groups at one year.
In two RCTs, FNB procedures were compared. In one study with 99 patients randomised to two FNB concentrations, there was no difference in WOMAC score between groups at 12 months [29]. In another study with 61 participants allocated to two different durations of FNB, there was no difference in WOMAC pain scores at one year [30]. In these studies, all participants received either SNB [29] or PCA [30].
Single-shot FNB was compared with single adductor canal block in one RCT with 98 participants, all receiving LIA [31]. At six months there was no difference in VAS-rated pain.
Sciatic nerve block In one study, 89 patients were randomised to single-shot SNB, continuous SNB, or PCA [32]. All patients received FNB. At 12 months, there were no differences in pain for single-shot SNB and continuous SNB on the WOMAC pain scale or VAS-rated pain at rest or during mobilisation.
Similarly, there were no differences between single-shot SNB and PCA in WOMAC pain scale or VAS-rated pain at rest or during mobilisation, or between continuous SNB and PCA.

Local anaesthetic infiltration
Four RCTs compared LIA with placebo. In one study, all 280 participants received FNB and PCA [33]. There was weak evidence that WOMAC pain scores were better in the LIA group at six (p=0.063) but not at 12 months (p=0.107) when the difference in means of 3.8/100 was lower than the MCID of 8-9/100 reported by Ehrich and colleagues [34]. In another study, 56 patients received LIA including ketorolac, or saline placebo, and all received PCA [35]. At one year, mean differences and confidence intervals provided weak evidence that OKS scores were better in the LIA group but the difference in means of 2.7/48 was less than the MCID of 4/48 reported by Beard and colleagues [36]. LIA before surgical incision was compared with placebo in one study with 120 participants [37]. None received FNB or PCA. There was weak evidence for a better KSS (function and knee score components) at six months in those receiving LIA (p=0.07) with a difference in means of 14.2/200 exceeding the MCID of 12.3/200 reported by Lee and colleagues [38]. In another study, all 51 participants received LIA intra-operatively, followed by PCA [39]. Those randomised to post-operative catheter-delivered LIA with ketorolac, or saline placebo had similar VAS-rated pain at six and 12 months.
LIA delivered as an injection and post-operative infusion was compared with epidural PCA in one study with 222 patients [40]. There was no difference between groups in OKS at 12 months.
In one study of 100 participants, LIA with or without corticosteroid were compared [41]. All patients received PCA. At two years there was no difference in OKS between groups.

Oral celecoxib
In one RCT, 44 participants received oral celecoxib or placebo [42], as well as PCA. There were no differences between groups in KOOS or VAS-rated pain at 12 months.

Ketamine or nefopam infusion
In one RCT, ketamine infusion, nefopam infusion and saline placebo were compared in 75 patients, all of whom received PCA [43]. There was weak evidence that participants receiving ketamine or nefopam had lower VAS-rated pain on movement at 12 months. For the Douleur Neuropathique 4 (DN4) measure of neuropathic pain, there was some evidence favouring ketamine over placebo at 6 and 12 months (p=0.02), but overall, few patients reported neuropathic pain at 12 months.

Pregabalin
Oral pregabalin was compared with placebo in one RCT with 240 participants [44]. All received LIA and PCA. At six months, no participants receiving pregabalin reported neuropathic pain when assessed using the Leeds assessment of Neuropathic Symptoms and Signs Pain Scale, compared with 5.2% of those receiving placebo (p=0.014) which represents some evidence favouring pregabalin.

Tourniquet
Five studies explored tourniquet use to provide a bloodless field.
In three RCTs, participants received TKR with or without a tourniquet. In one study with 64 patients, a difference in KOOS pain favouring tourniquet use was not significant at six or 12 months [45]. In another study with 20 patients, the OKS was not significantly different between groups at six or 12 months [46]. There were three blood transfusions in the tourniquet group, compared with none in the 'no tourniquet' group. In the third study with 100 participants, VASrated pain and HSS scores were similar between groups at 6 months [47]. Six cases of wound ooze occurred in the tourniquet group.
In two RCTs, short and long-duration tourniquet use were compared. In one study with 65 participants, there was weak evidence based on graphical representation of means and confidence intervals for improved OKS at 12 months in the long-duration group and the difference in means of 5/48 [48] was greater than the MCID of 4/48. Adverse events were reported by 62% of participants receiving short-duration tourniquet compared with 38% in the long-duration group. The study was terminated early as 10 blood transfusions were required in the short-duration group compared with three in the long-duration group. In the second study with 150 participants, tourniquets were used in three different periods during surgery [49]. At six months, there were no differences between groups in HSS scores.

Blood conservation
Seven studies evaluated strategies to limit blood loss after TKR.
Tranexamic acid Tranexamic acid injections or infusions were compared with saline placebo or untreated control in four RCTs [47,[50][51][52]. In all studies, control patients required more blood transfusions. In one study including 180 participants comparing intravenous tranexamic acid with untreated controls, there was no significant difference in WOMAC pain scores at one year [51]. In another study with 48 participants comparing intra-articular tranexamic acid injection with saline placebo, there was no significant difference in WOMAC scores at six months [50]. One study with 135 participants compared two intra-articular tranexamic acid doses and saline control [52]. There were no significant differences in WOMAC scores at one year. Intravenous and intra-articular tranexamic was compared with untreated controls in one study with 100 participants [47]. VAS-rated pain at six months was similar between groups, but there was strong evidence favouring tranexamic acid for HSS scores (p<0.001) although the difference in means of 1.4/100 was lower than the MCID of 8.3/100 reported by Singh and colleagues [53].
In one study, continuous tranexamic acid infusion was compared with a single bolus in 106 patients [54]. There was no difference between groups in KSS at six months or blood loss.

Thrombin infusion
In one RCT with 80 participants, thrombin infusion was compared with untreated control [55]. At one year there was no difference between groups in pain measured on the KSS.

Flexion or extension
For blood management, operated knees were kept in passive flexion or passive extension after surgery in one RCT with 180 patients [56]. At one year, OKS was similar between groups.
Transfusion requirement was greater in patients with passive extension.

Compression bandage
One RCT with 49 participants compared compression bandaging to reduce post-operative knee swelling with standard bandaging. OKS was similar in randomised groups at six months [57]. One RCT evaluated use of the antiresorptive monoclonal antibody Denusomab to promote bone healing. Fifty participants were randomised and at 12 and 24 months there were no significant differences between groups in KOOS pain [59].

Continuous passive motion
Two RCTs evaluated use of continuous passive motion (CPM) to minimise joint stiffness and improve range of movement. In one study, 90 participants were randomised to no CPM, CPM at low flexion from post-operative day 1-7, or CPM at high flexion from post-operative day 3-7 [60].
There was no significant difference between groups in KSS at two years. In the other study, 147 participants were randomised to CPM with increasing range of movement from day 1-6, early flexion CPM from day 0-6, or no CPM [61]. There were no significant differences between groups in KSS at 12 months.

Electrical stimulation
Two RCTs evaluated electrical stimulation which is believed to have anti-inflammatory activity and limit muscle atrophy. In one study with 76 participants receiving transcutaneous electric muscle stimulation from post-operative day two for six weeks or no intervention, Short Form 36 bodily pain showed strong evidence for greater improvement at one year in the intervention group compared to control (p<0.001) [62]. The difference in means of 12.5/100 was close to the MCID of 16.9/100 reported by Escobar and colleagues [63]. There were no differences in OKS or KSS scores. In another study with 30 participants, pulsed electromagnetic fields from postoperative day 7 were compared with untreated control [64]. At 12 months, there was some evidence that VAS-rated pain was lower in intervention patients compared with controls (p<0.05). The difference in means of 2.1/10 was greater than the MCID of 16.1/100 reported by Danoff and colleagues [65]. Knee swelling was common during the intervention.

Rehabilitation
Four RCTs evaluated features of early rehabilitation focusing on regaining range of movement, functional independence and improving mobility.

Walking guidance and training
In one study, 86 participants were randomised to walking guidance and training from postoperative day two or no intervention further to standard rehabilitation [66]. At six months, there was some evidence that those receiving intervention had lower VAS-rated pain (p<0.01) and

Flexion or extension during knee closure
Targeting improved functional recovery, wound closure performed in 90° flexion was compared with wound closure in full extension in one study with 80 participants [67]. There was no difference between groups in VAS-rated pain at six months.

Aquatic therapy
In one study with 185 participants, aquatic therapy commenced on post-operative day six or 14 were compared [68]. Patients reported similar WOMAC pain at 12 and 24 months.
Supported early discharge In one study, early discharge supported by physiotherapist home visits and outpatient or selfdirected physiotherapy was compared with two week rehabilitation centre-based usual care [69].
The study included 234 individuals receiving TKR or total hip replacement. Compared with usual care, there was weak evidence that patients with early discharge had lower WOMAC pain scores at 12 months (p=0.08). The difference in means of 4 was less than the MCID of 8-9/100.
Results were not presented separately but did not differ between patients with TKR or total hip replacement.

Anabolic steroids
Searches identified one study of anabolic steroids to improve post-operative muscle strength.
Ten participants received intramuscular nandrolone injections or saline from post-operative day five for six months. KSS results indicated some evidence for improvement in the intervention group compared with controls at 12 months (p=0.03) [70]. The difference in means of 10

Interventions with no long-term outcome
Interventions with lack of RCT evidence are summarised in Figure 1. follow up, none reported pain or an outcome score. One study reported long-term follow up of an RCT of teriparatide but included no data on pain.
For some interventions, RCTs with long-term pain outcomes were identified, but none were at low risk of bias: cold therapy; guided imagery; platelet rich plasma; and trigger point needling.
Aspects of peri-operative care evaluated in RCTs but lacking long-term pain follow up were: adenosine triphosphate; alternative and Chinese medicine; assistive devices; brain stimulation; calcium supplements; cardiovascular drugs; colloids and crystalloids; comorbidity management; constipation treatment; creatine; delirium prevention; dexmedetomidine; glucocorticoids; glucose infusion; iron; laser therapy; methylprednisolone; music therapy; nausea prevention; nutritional supplements; physiological treatments; remote ischaemic preconditioning; sleep treatments; therapy dogs; and warming.

DISCUSSION
Peri-operative care for patients with osteoarthritis receiving TKR varies widely [71,72]. To guide decisions on appropriate care, the top level of evidence in the hierarchy of primary research is the RCT [73,74]. Bringing evidence from RCTs together in systematic reviews with thorough risk of bias assessment ensures that health professionals have the information they need to deliver a high-quality patient experience with safe, clinically-effective and cost-effective treatments [75].
Furthermore, systematic reviews can identify gaps in the evidence base and promote further research.
Much research in TKR aims to identify treatments that facilitate a speedy recovery with minimal short-term pain. However, patients choose to have joint replacement for long-term pain relief and reduction in functional limitations. Thus, changes to peri-operative care, supported by shortterm RCT evidence, should be backed up with evidence about long-term effectiveness for reducing pain and reassurance that there are no long-term unfavourable consequences. To this end, we synthesised evidence from RCTs evaluating peri-operative interventions which have considered their long-term effects on pain outcomes.
A major focus of research into improving long-term pain after TKR has been through prevention of acute post-operative pain using multimodal analgesia. Our review provides some encouragement for further research on long-term benefits of intra-articular LIA injections, as previously shown in short-term studies [19,76], ketamine infusion, oral pregabalin and oral opioids. Nerve blocks are effective for managing peri-operative pain[77] but we identified no long-term benefit. In single studies, there was no benefit for nefopam infusion, oral celecoxib or settings. With such an approach, convincing evidence will accrue to guide multimodal pain management.
Tranexamic acid is highly effective in reducing blood transfusions during TKR [78]. We found no evidence that tranexamic acid affects long-term pain or, as observed in registry studies [79,80], adverse events. Single RCTs of thrombin infusion and maintenance of knee in flexion to prevent blood loss showed no effect on long-term pain. Tourniquets improve intraoperative visualisation of the joint, reduce blood loss and facilitate cement fixation but are associated with nerve damage, delayed recovery, acute pain and need for analgesics [81,82]. The RCTs we identified showed no effects of tourniquet use on long-term pain.
Consistent with a previous review [83], there was no suggestion that CPM affects long-term pain.
Studies provided encouragement for further research into walking training, anabolic steroid injection, electrical stimulation and supported discharge.
For some interventions a direct mechanism is clear, but for others, reasons for long-term impact are less obvious. This may explain why no studies evaluated DVT prophylaxis with long-term follow up excepting a small number reporting adverse events. However, treatments to prevent symptomatic DVTs which occur in about 1% of treated patients [84] also reduce the incidence of asymptomatic DVT observed in about 28% of treated patients [85] and this may have long-term benefits. Conversely, new anticoagulants are associated with bleeding [86], which may increase the risk of wound complications [87] and joint infection [88] which are associated with long-term pain [89,90].
Our study is limited by the lack of meta-analysis which was not appropriate due to intervention and outcome heterogeneity. In the context of perioperative pain management, this was noted previously [76]. Our approach to assessing the evidence was a narrative synthesis of studies with low risk of bias. While this may seem overly restrictive, Cochrane risk of bias assessment allows us to screen out studies with important issues that may affect the validity of results. The main potential source of bias was incomplete outcome assessment. Although studies with longterm follow up are naturally at higher risk of missing data, we maintained a standard in this domain as it is recognised that research participants who do not complete follow up assessments differ in outcomes from those with follow up data and their inclusion could change the interpretation of results [91].  23 We summarised p-values to assess the strength of evidence but, as statistically strong evidence may not reflect clinically important results [92], where possible we also compared effect sizes with MCIDs. Our review considered a diverse range of interventions at a specific time in the TKR pathway and, as we were unable to make clinical practice recommendations, we did not adopt the GRADE system [93] for this review.
Our systematic review of peri-operative interventions brings together evidence on interventions in the peri-operative phase of the TKR pathway. Whilst not supportive of the inclusion of specific interventions in clinical practice to optimise long-term pain outcomes, there are clearly areas that merit research. High quality studies assessing long-term pain after peri-operative interventions are feasible and necessary to ensure that patients with osteoarthritis achieve good long-term outcomes after TKR.    Intraoperative sedation with iv propofol at discretion of anaesthesiologist. Lumbar spinal anaesthesia with 12mg 0.5% bupivacaine. Postoperative i.v. PCA with fentanyl 50µg/ml set to deliver 25µg every 5 min as needed. Celecoxib 100mg and acetaminophen 650mg on arrival in recovery room and every 12 and 6 hrs respectively. Breakthrough medication with intramuscular ketorolac 10 mg every 4 hrs. 1 year Overall 32 lost to follow up High risk of bias: only 27/59 patients followed up due to resource limitations. No   ; 62  62%; 70%; 73% Lorazepam 1mg 2 hours and acetaminophen 2g 1 hour before surgery. FNB with stimulating catheter: loading dose 20 ml levobupivacaine 0.375% and after 45 minutes a continuous infusion of levobupivacaine 0.125% 10 ml/hr. General anaesthesia induced with 3-5 µg/ml propofol infusion and remifentanil 0.5 µg/kg/min and maintained with 2-3 µg/ml at 0.1-0.25 µg/kg/min. Postoperatively, FNB changed to patient controlled FNB, 5ml bolus, 30-minute lockout; basal rate 6 ml/hr. i.v. morphine administered if needed. Postoperative analgesia with acetaminophen 1g 4 times daily. Diclofenac 50mg or tramadol 50mg 3 times daily. Tramadol 100mg before removal of nerve catheters. Morphine pain relief as required. 12 months 2;7;5 lost to follow up Low risk of bias Median WOMAC pain scores at 12 months: SNB injection 80 (range , SNB continuous 90  and PCA only 90 , p=0. 81 68.7 (7.9) 52%; 54% paracetamol 30 minutes before the end of operation. Immediately post-operative 400mg oral ibuprofen. PCA with morphine 1mg/ml, 1 mg bolus dose and a 5minute lock-out. If necessary morphine bolus up to 0.2mg/kg as rescue analgesia. During hospital stay, visit from pain specialist nurse. Oral or i.v. paracetamol every 6 hours and ibuprofen 400mg every 8 hours. When PCA no longer needed, oral codeine phosphate 30-60mg every 6 hours, tramadol 50-100mg every 6 hours and oramorph 10-20mg as rescue analgesia.
Low risk of bias At 12 months WOMAC pain score (0-100) in LIA group median 90 (IQR 30), Control 85 (35); ITT-CC linear regression coefficient 3.83 (95%CI -0. 83, 8.49), p=0.107. At 6 months WOMAC pain score ITT-CC linear regression coefficient 4.10 (95%CI -0. 22, 8.43), p=0.063. Mean differences lower than MCID of 8-9 [34]. Superficial and deep wound infection rate in LIA group 3.2% and 1.9% in control group, p=0.500. No differences in serious adverse events between groups 60ml intra-operative LIA with 0.25% bupivacaine and 1/200,000 adrenaline injected into the posterior capsule, medial and lateral capsule, fascia and muscle, and subcutaneous tissues. No  Sedation with i.v. midazolam and propofol. Intraoperative LIA loading dose of 20ml 0.25% bupivacaine/ epinephrine injection, 10ml into medial and lateral subcutaneous tissue around the incision and 10ml intra-articular after closure. Infiltrate delivered by pain pump into lateral recess of intraarticular space. Spinal anaesthetic with 10-15 mg of 0.75% or 0.5% plain bupivacaine and 20μg fentanyl. Postoperative morphine PCA. 7.5mg i.v ketorolac preoperatively plus 15mg every 6 hours postoperatively for 48 hours, then oral ketorolac 10mg every 6 hours for 2 days. Gabapentin 600mg given preoperatively plus 300mg twice daily for 48 hours postoperatively. Oxycodone 10mg twice daily for 48 hours postoperative. Oral paracetamol 650mg every 4 hours for 72 hours. 6 and 12 months 3;1 of those who received treatment Low risk of bias Mean VAS pain score at 6 months 1.2 (SD 1.3); 1.2 (1.2). p=0.836. At 12 months 0.9 (1.2); 1.0 (1.1). p=0. 767 No short-term differences in adverse events except control patients more likely to be drowsy at 48 hrs. Longterm adverse events not reported. Intrathecal injection of 15mg bupivacaine and 100μg morphine. General anaesthesia. After surgery 1.5g paracetamol and then 750mg every 4 hours; PCA with morphine 2mg boluses with 10-minute lockout; morphine rescue 2.5mg intravenously as required; and rescue oral ibuprofen 800mg. 6 months 3 protocol breaches and 1 patient with uncontrolled pain. High risk of bias due to non-ITT reporting and recruitment difficulties 2/5 ketamine group had mild/moderate pain on the WOMAC pain scale at 26 weeks or failed to improve compared with 5/7 controls. 1 adverse psycho-mimetic effect not attributed to intervention or control treatment Ketamine 0.5mg/kg bolus followed by 4μg/kg/min infusion. Commenced before surgical incision and continued until wound bandaged or syringe empty. Saline infusion. Commenced before surgical incision and continued until wound bandaged or syringe empty.

Instructions to authors
Complete this checklist by entering the page numbers from your manuscript where readers will find each of the items listed below.
Your article may not currently address all the items on the checklist. Please modify your text to include the missing information. If you are certain that an item does not apply, please write "n/a" and provide a short explanation.
Upload your completed checklist as an extra file when you submit to a journal.
In your methods section, say that you used the PRISMA reporting guidelines, and cite them as:

ABSTRACT Objectives
For many people with advanced osteoarthritis, total knee replacement (TKR) is an effective treatment for relief of pain and improvement of function. Features of peri-operative care may be associated with the adverse event of chronic pain six months or longer after surgery; effects may be direct, e.g. through nerve damage or surgical complications, or indirect through increasing risks of adverse events. The objective of this systematic review is to evaluate whether non-surgical peri-operative interventions prevent long-term pain after TKR.

Methods
We conducted a systematic review of peri-operative interventions for adults with osteoarthritis receiving primary TKR evaluated in a randomised controlled trial (RCT). We searched The Cochrane Library, MEDLINE, Embase, PsycINFO and CINAHL from inception to February 2018. After screening, two reviewers evaluated articles. Studies at low risk of bias according to the Cochrane tool were included.

Interventions
Peri-operative non-surgical interventions; control receiving no intervention or alternative treatment.

Primary and secondary outcome measures
Pain or score with pain component assessed at six months or longer post-operative.

44
RCTs at low risk of bias assessed long-term pain. Intervention heterogeneity precluded meta-analysis and definitive statements on effectiveness. There was encouragement for further research into local infiltration analgesia, ketamine infusion, pregabalin, and electric muscle stimulation. In the studies we identified, tranexamic acid to prevent blood loss was not associated with long-term pain. Many extensively researched interventions including venous thromboembolism prevention have not been evaluated in relation to long-term pain.

Conclusions
To prevent chronic pain after TKR, peri-operative interventions including components of multimodal analgesia, early rehabilitation and supported discharge, electrical stimulation and anabolic steroids show promise that merits further research. Tranexamic use is not associated

STRENGTHS AND LIMITATIONS
• For the first time, this systematic review brings together contemporary evidence on aspects of peri-operative care for people with total knee replacement and their effects on longterm pain.
• Only studies assessed to be at low risk of bias were included in the narrative synthesis.
• Intervention and outcome heterogeneity precluded meta-analysis.

KEYWORDS
Total knee replacement; Systematic review; Randomised controlled trial; Peri-operative care; Long-term pain

BACKGROUND
In the US about 13% of men and 19% of women will be diagnosed with knee osteoarthritis and over half will receive a total knee replacement (TKR) [1]. For people with advanced osteoarthritis unresponsive to pharmacological or conservative treatments, TKR aims to relieve pain and improve function. In the UK nearly 100,000 primary TKRs were performed in 2017[2,3] and in the USA in 2010, an estimated 4.7 million people were living with a TKR [4]. Despite good outcomes for many, some people report long-term pain and are disappointed with their surgery [5,6]. After TKR, pain levels plateau from about 6 months [7,8] after which persistent pain is considered "chronic" [9] and is reported by 10-34% of patients [10].
The mechanisms that influence the development of chronic pain after total knee replacement may be biological, mechanical and psychosocial. Biological causes include the sensitising impact of long-term pain from osteoarthritis [11,12], inflammation, infection and localised nerve injury [13]. Mechanical causes include altered gait, prosthesis loosening, and effects on ligaments [14,15]. Psychological factors including depression and catastrophizing may also influence outcomes [16][17][18][19]. Much research has focused on pre-operative predictors of outcomes and these include pain intensity, presence of widespread pain, anxiety, depression and catastrophizing. [10,20] However, attempts to target or modify pre-operative care have, as yet, shown no benefit regarding chronic pain or other long-term patient outcomes [10,[21][22][23].
Peri-operative risk factors suggest that appropriate interventions may reduce long-term pain.
For example, acute post-operative pain, which may be a direct consequence of the operation, anaesthetic protocol and subsequent analgesia, or related to particular aspects of care, is an acknowledged risk factor for chronic post-surgical pain [24].
In the peri-operative period from hospital admission to the early stages of recovery, care focuses on acute pain management, prevention of adverse events, facilitation of early mobilisation and timely discharge. However, for people with osteoarthritis the key aim of TKR is the achievement of a long-term painless and well-functioning knee with no adverse events. All aspects of peri-operative care should work together to achieve this.
Any treatment in the peri-operative period including pain management, blood conservation, deep vein thrombosis (DVT) and infection prevention, and inpatient rehabilitation could potentially affect patient recovery and chronic pain, either directly or indirectly. Direct mechanisms may be through prevention of nerve damage [25], post-thrombotic syndrome [26], reperfusion injury [27] and articular bleeding [28]. For other treatments, pathways leading to long- Our systematic review of randomised controlled trials (RCTs) aims to evaluate the effectiveness of treatments in the peri-operative period in preventing long-term pain after TKR. By focusing on studies with low risk of bias we aim to identify interventions with robust evidence of long-term effectiveness and identify gaps in the research base.

METHODS
The systematic review protocol was registered (PROSPERO CRD42017041382) and PRISMA reporting guidelines used [30]. A checklist is included as Supplementary material.

Patient and public involvement
As part of the STAR programme of research (NIHR RP-PG-0613-20001), this review benefited from extensive patient and public involvement. Advice was sought from patients and stakeholders at a group discussion in March 2016 with decisions made on inclusion criteria and outcomes. Our patient advisory group comprises five patients with experience of long-term pain after TKR, supported by a dedicated co-ordinator. This group will advise on dissemination of the study results to a general audience including plain language summaries.

Eligibility criteria
Studies were eligible if they satisfied PICOS criteria defined in the protocol. Participants were adults receiving unilateral primary TKR with osteoarthritis in at least 75% of patients.
Pharmacological or non-pharmacological interventions commenced in the peri-operative setting with "peri-operative" reflecting the time from hospital admission to immediately post-discharge.
Interventions relating to implant designs and surgical procedures were excluded. The comparator was usual care, placebo or an alternative intervention. Outcomes were, in preference, patient-reported joint-specific pain intensity measured by tools such as the Western

Screening and data extraction
We imported records into Endnote X7 (Thomson Reuters). An initial screen by one reviewer excluded clearly irrelevant articles. Subsequently, abstracts and full articles were screened independently by two reviewers and reasons for exclusion recorded.
Data were extracted onto piloted forms and an Excel spreadsheet by one reviewer, specifically: country; dates; participants (indication, age, sex); inclusion and exclusion criteria; intervention and control content; setting, timing, duration and intensity of intervention; follow up intervals; losses to follow up; pain outcome data; and serious adverse events. Data was checked against source material by a second reviewer.
Authors were contacted for missing data, and data provided for previous reviews was used [10,31].

Quality assessment
Potential sources of bias were assessed by two experienced reviewers using the Cochrane risk of bias tool [32], specifically: the randomisation process; deviations from intended interventions; missing outcome data (>20%), measurement of the outcome; and selection of the reported result. Studies with serious concerns relating to risk of bias were considered high risk and those with limited reporting unclear risk. Studies with high or unclear risk of bias were excluded from the narrative synthesis but are included in supplementary summary tables with reasons for exclusion. Insufficient studies with similar interventions and outcomes were identified for meta-analysis, and a narrative synthesis is presented. Results reported with p-values ≤0.001 were considered "strong" evidence of effectiveness [33], p-values 0.001-0.05 "some" evidence, and p-values 0.05-0.1 "weak" evidence. When authors reported results "statistically significant" with no p-value, this was noted. Where possible, effect sizes were compared with published minimal clinically important differences (MCID). Concerns relating to adverse events were summarised. Details of 44 studies assessed to be at low risk of bias are summarised in Table 1. In 34 studies, patients received TKR exclusively for osteoarthritis and in three, 75% or more patients.

RESULTS
In seven studies there was no information on reason for surgery but there was no suggestion that patients had an indication other than osteoarthritis. Interventions focused on pain management (n=20), tourniquets (n=5), compression bandages (n=1), blood conservation

Femoral nerve block
Femoral nerve blocks (FNB) were studied in 10 RCTs.
Three RCTs compared FNB with no FNB. In one study with 55 patients, WOMAC pain scores at one year were similar in patients receiving single-shot FNB and untreated controls [43]. All patients received local anaesthetic infiltration (LIA) and patient-controlled analgesia (PCA). In another study with all participants receiving LIA, 150 were randomised to receive single-shot FNB with or without sciatic nerve block (SNB), or general anaesthesia [37]. There were no differences in HSS scores between groups at six months. Continuous FNB was compared with oral hydrocodone opioid in 62 patients receiving PCA [39]. There was some evidence for 'pain using stairs' favouring hydrocodone (p=0.01) but no difference in overall NRS-rated pain at one year and concern over venous thromboembolism in 4/31 participants treated with hydrocodone.
In two RCTs, continuous FNB was compared with PCA. In one study with 60 participants, the KSS at six months was similar between groups [44]. In another study with 280 participants, there was some evidence for higher incidence of NRS-rated pain at six months in the PCA group than the FNB group (p=0.021) but not at 12 months (p=0.273). [40] Two RCTs compared FNB with LIA. In one study, all 157 participants also received PCA [36]. At one year, KSS values were similar in single-shot FNB and LIA groups. In the other study, 94 participants were randomised to receive single-shot FNB with continuous epidural infusion or LIA through an intra-articular catheter [41]. VAS-rated pain was similar between groups at one year.
In two RCTs, FNB procedures were compared. In one study with 99 patients randomised to two FNB concentrations, there was no difference in WOMAC score between groups at 12 months [34]. In another study with 61 participants allocated to two different durations of FNB,  [35]. In these studies, all participants received either SNB [34] or PCA [35].
Single-shot FNB was compared with single adductor canal block in one RCT with 98 participants, all receiving LIA [38]. At six months there was no difference in VAS-rated pain.

Sciatic nerve block
In one study, 89 patients were randomised to single-shot SNB, continuous SNB, or PCA [42]. All patients received FNB. At 12 months, there were no differences in pain for single-shot SNB and continuous SNB on the WOMAC pain scale or VAS-rated pain at rest or during mobilisation.
Similarly, there were no differences between single-shot SNB and PCA in WOMAC pain scale or VAS-rated pain at rest or during mobilisation, or between continuous SNB and PCA.

Local anaesthetic infiltration
Four RCTs compared LIA with placebo. In one study, all 280 participants received FNB and PCA [50]. There was weak evidence that WOMAC pain scores were better in the LIA group at six (p=0.063) but not at 12 months (p=0.107) when the difference in means of 3.8/100 was lower than the MCID of 8-9/100 reported by Ehrich and colleagues [77]. In another study, 56 patients received LIA including ketorolac, or saline placebo, and all received PCA [47]. At one year, mean differences and confidence intervals provided weak evidence that OKS scores were better in the LIA group but the difference in means of 2.7/48 was less than the MCID of 4/48 reported by Beard and colleagues [78]. LIA before surgical incision was compared with placebo in one study with 120 participants [46]. None received FNB or PCA. There was weak evidence for a better KSS (function and knee score components) at six months in those receiving LIA (p=0.07) with a difference in means of 14.2/200 exceeding the MCID of 12.3/200 reported by Lee and colleagues [79]. In another study, all 51 participants received LIA intra-operatively, followed by PCA [49]. Those randomised to post-operative catheter-delivered LIA with ketorolac, or saline placebo had similar VAS-rated pain at six and 12 months.
LIA delivered as an injection and post-operative infusion was compared with epidural PCA in one study with 222 patients [45]. There was no difference between groups in OKS at 12 months.
In one study of 100 participants, LIA with or without corticosteroid were compared [48]. All patients received PCA. At two years there was no difference in OKS between groups.
Oral celecoxib In one RCT, 44 participants received oral celecoxib or placebo [51], as well as PCA. There were no differences between groups in KOOS or VAS-rated pain at 12 months.

Ketamine or nefopam infusion
In one RCT, ketamine infusion, nefopam infusion and saline placebo were compared in 75 patients, all of whom received PCA [52]. VAS-rated pain on movement did not differ between groups at 12 months. For the Douleur Neuropathique 4 (DN4) measure of neuropathic pain, there was some evidence favouring ketamine over placebo at six and 12 months (p=0.02), but overall, few patients reported neuropathic pain at 12 months.

Pregabalin
Oral pregabalin was compared with placebo in one RCT with 240 participants [53]. All received LIA and PCA. At six months, there was some evidence for better NRS pain in patients receiving pregabalin compared with placebo (p=0.0176) but the difference in means of 0.54/10 was less than the MCID of 1/10 reported by Salaffi and colleagues [80]. No participants receiving pregabalin reported neuropathic pain when assessed using the S-LANSS, compared with 5.2% of those receiving placebo (p=0.014). Patients receiving pregabalin were more likely to be sedated and confused in the first two days after surgery.

Tourniquet
Five studies with 399 participants explored tourniquet use to provide a bloodless field. Two studies each were from Australia and China, and one from Denmark. All were conducted at a single centre with participants recruited between 2008 and 2015. Sample sizes ranged from 20 to 150 participants, with a median of 65. The range of mean ages of participants in randomised groups was 66 to 71 years and in 3/5 studies, a majority of participants were women.
In three RCTs, participants received TKR with or without a tourniquet. In one study with 64 patients, a difference in KOOS pain favouring tourniquet use was not significant at six or 12 months [54]. In another study with 20 patients, the OKS was not significantly different between groups at six or 12 months [56]. There were three blood transfusions in the tourniquet group, compared with none in the 'no tourniquet' group. In the third study with 100 participants, VASrated pain and HSS scores were similar between groups at 6 months [55]. Six cases of wound ooze occurred in the tourniquet group.
In two RCTs, short and long-duration tourniquet use were compared. In one study with 65 participants, there was weak evidence based on graphical representation of means and confidence intervals for improved OKS at 12 months in the long-duration group and the difference in means of 5/48 [57] was greater than the MCID of 4/48. Adverse events were reported by 62% of participants receiving short-duration tourniquet compared with 38% in the long-duration group. The study was terminated early as 10 blood transfusions were required in the short-duration group compared with three in the long-duration group. In the second study with 150 participants, tourniquets were used in three different periods during surgery [58]. At six months, there were no differences between groups in HSS scores.

Blood conservation
Seven studies with 829 participants evaluated strategies to limit blood loss after TKR. Two studies were from Thailand, and one each from China, France, South Korea, the UK and the USA. All were conducted at a single centre with participants recruited between 2003 and 2015 when stated. Sample sizes ranged from 48 to 180 participants, with a median of 106. One study had three trial arms. The range of mean ages of participants in randomised groups was 65 to 74 years and in all studies, a majority of participants were women.

Tranexamic acid
Five RCTs evaluated tranexamic acid.
Tranexamic acid injections or infusions were compared with saline placebo or untreated control in four RCTs [55,61,64,65]. In all studies, control patients required more blood transfusions. In one study including 180 participants comparing intravenous tranexamic acid with untreated controls, there was no significant difference in WOMAC pain scores at one year [61]. In another study with 48 participants comparing intra-articular tranexamic acid injection with saline placebo, there was no significant difference in WOMAC scores at six months [64]. One study with 135 participants compared two intra-articular tranexamic acid doses and saline control [65]. There were no significant differences in WOMAC scores at one year. Intravenous and intra-articular tranexamic was compared with untreated controls in one study with 100 participants [55]. VASrated pain at six months was similar between groups, but there was strong evidence favouring tranexamic acid for HSS scores (p<0.001) although the difference in means of 1.4/100 was lower than the MCID of 8.3/100 reported by Singh and colleagues [81].
In one study, continuous tranexamic acid infusion was compared with a single bolus in 106 patients [60]. There was no difference between groups in KSS at six months or blood loss. In one RCT with 80 participants, thrombin infusion was compared with untreated control [62]. At one year there was no difference between groups in pain measured on the KSS.

Flexion or extension
For blood management, operated knees were kept in passive flexion or passive extension after surgery in one RCT with 180 patients [63]. At one year, OKS was similar between groups.
Transfusion requirement was greater in patients with passive extension.

Compression bandage
One RCT conducted at a single UK centre with 49 participants recruited between 2013 and 2014 compared compression bandaging to reduce post-operative knee swelling with standard bandaging. The mean age of participants was about 69 years and a majority were women. OKS was similar in randomised groups at six months [59].

Wound management
One RCT with recruitment in 2011 at a single centre in South Korea evaluated a wound care strategy to limit post-operative scar pain. The mean age of participants was about 69 years and a majority were women. Investigators compared silicone gel application to the surgical scar with placebo in 100 participants [75]. There were no significant differences in VAS-rated pain at six and 12 months.

Denusomab
One RCT evaluated use of the antiresorptive monoclonal antibody Denusomab to promote bone healing. The study was conducted in two centres in Sweden with recruitment of 50 participants between 2012 and 2014. The mean age of participants was about 65 years and a majority were women. At 12 and 24 months there were no significant differences between groups in KOOS pain [66].  CPM at high flexion from post-operative day 3-7 [68]. There was no significant difference between groups in KSS at two years. In the other study, 147 participants were randomised to CPM with increasing range of movement from day 1-6, early flexion CPM from day 0-6, or no CPM [67]. There were no significant differences between groups in KSS at 12 months.

Electrical stimulation
Two In one study with 76 participants receiving transcutaneous electric muscle stimulation from postoperative day two for six weeks or no intervention, Short Form 36 bodily pain showed strong evidence for greater improvement at one year in the intervention group compared to control (p<0.001) [69]. The difference in means of 12.5/100 was close to the MCID of 16.9/100 reported by Escobar and colleagues [82]. There were no differences in OKS or KSS scores. In another study with 30 participants, pulsed electromagnetic fields from post-operative day 7 were compared with untreated control [70]. At 12 months, there was some evidence that VAS-rated pain was lower in intervention patients compared with controls (p<0.05). The difference in means of 2.1/10 was greater than the MCID of 16.1/100 reported by Danoff and colleagues [83].
Knee swelling was common during the intervention.

Walking guidance and training
In one study, 86 participants were randomised to walking guidance and training from postoperative day two or no intervention further to standard rehabilitation [71]. At six months, there was some evidence that those receiving intervention had lower VAS-rated pain (p<0.01) and

Flexion or extension during knee closure
Targeting improved functional recovery, wound closure performed in 90° flexion was compared with wound closure in full extension in one study with 80 participants [74]. There was no difference between groups in VAS-rated pain at six months.

Aquatic therapy
In one study with 185 participants, aquatic therapy commencing on post-operative day six was compared with aquatic therapy commencing on day 14 [72]. Patients reported similar WOMAC pain at 12 and 24 months.

Supported early discharge
In one study, early discharge supported by physiotherapist home visits and outpatient or selfdirected physiotherapy was compared with two weeks of rehabilitation centre-based usual care [73]. The study included 234 individuals receiving TKR or total hip replacement. Compared with usual care, there was weak evidence that patients with early discharge had lower WOMAC pain scores at 12 months (p=0.08). The difference in means of 4 was less than the MCID of 8-9/100. Results were not presented separately but did not differ between patients with TKR or total hip replacement.

Anabolic steroids
Searches identified one study of anabolic steroids to improve post-operative muscle strength

Interventions with no long-term outcome
Interventions with lack of RCT evidence are summarised in Figure 1.
While 148 RCTs of DVT prophylaxis were identified, only five reported long-term follow up, none of which included a pain or outcome score. Among 29 RCTs of antibiotic prophylaxis, 16 reported long-term follow up, but none included a pain or outcome score. Six RCTs evaluated the use of bisphosphonates and, although all reported long-term follow up, none reported pain or an outcome score. One study reported long-term follow up of an RCT of teriparatide but included no data on pain.
For some interventions, RCTs with long-term pain outcomes were identified, but none were at low risk of bias: cold therapy; guided imagery; platelet rich plasma; and trigger point needling.
Aspects of peri-operative care evaluated in RCTs but lacking long-term pain follow up were: adenosine triphosphate; alternative and Chinese medicine; assistive devices; brain stimulation; calcium supplements; cardiovascular drugs; colloids and crystalloids; comorbidity management; constipation treatment; creatine; delirium prevention; dexmedetomidine; glucocorticoids; glucose infusion; iron; laser therapy; methylprednisolone; music therapy; nausea prevention; nutritional supplements; physiological treatments; remote ischaemic preconditioning; sleep treatments; therapy dogs; and warming.

DISCUSSION
Peri-operative care for patients with osteoarthritis receiving TKR varies widely [84,85]. To guide decisions on appropriate care, the top level of evidence in the hierarchy of primary research is the RCT [86,87]. Bringing evidence from RCTs together in systematic reviews with thorough risk of bias assessment ensures that health professionals have the information they need to deliver a high-quality patient experience with safe, clinically-effective and cost-effective treatments [88].
Furthermore, systematic reviews can identify gaps in the evidence base and promote further research.
Much research in TKR aims to identify treatments that facilitate a speedy recovery with minimal short-term pain. However, patients choose to have joint replacement for long-term pain relief and reduction in functional limitations. Thus, changes to peri-operative care, supported by shortterm RCT evidence, should be backed up with evidence about long-term effectiveness for reducing pain and reassurance that there are no long-term unfavourable consequences. To this end, we synthesised evidence from RCTs evaluating peri-operative interventions which have considered their long-term effects on pain outcomes.
Nerve blocks are effective for managing peri-operative pain [90] but we identified no long-term benefit. In single studies, there was no benefit for nefopam infusion, oral celecoxib or LIA with additional corticosteroid. Regarding future studies, standardisation of the multi-modal regimen will allow evaluation of extra or alternative components in multiple studies in different settings.
With such an approach, convincing evidence will accrue to guide multimodal pain management.
Tranexamic acid is highly effective in reducing blood transfusions during TKR[91]. We found no evidence that tranexamic acid affects long-term pain or, as observed in registry studies [92,93], adverse events. Single RCTs of thrombin infusion and maintenance of knee in flexion to prevent blood loss showed no effect on long-term pain. Tourniquets improve intraoperative visualisation of the joint, reduce blood loss and facilitate cement fixation but are associated with nerve damage, delayed recovery, acute pain and need for analgesics [94,95]. The RCTs we identified showed no effects of tourniquet use on long-term pain.
Consistent with a previous review [96], there was no suggestion that CPM affects long-term pain.
Studies provided encouragement for further research into walking training, anabolic steroid injection, electrical stimulation and supported discharge.
For some interventions a direct mechanism is clear, but for others, reasons for long-term impact are less obvious. This may explain why, for example, no studies evaluated DVT prophylaxis with long-term follow up excepting a small number reporting adverse events. However, treatments to prevent symptomatic DVTs which occur in about 1% of treated patients [97] also reduce the incidence of asymptomatic DVT observed in about 28% of treated patients [98] and this may have long-term benefits. Conversely, new anticoagulants are associated with bleeding [99], which may increase the risk of wound complications [100] and joint infection [101] which are associated with long-term pain [102,103].
Our study is limited by the lack of meta-analysis which was not appropriate due to intervention and outcome heterogeneity. In the context of perioperative pain management, this was noted previously [89]. Our approach to assessing the evidence was a narrative synthesis of studies with low risk of bias. While this may seem overly restrictive, Cochrane risk of bias assessment allows us to screen out studies with important issues that may affect the validity of results. The main potential source of bias was incomplete outcome assessment. Although studies with longterm follow up are naturally at higher risk of missing data, we maintained a standard in this domain as it is recognised that research participants who do not complete follow up assessments differ in outcomes from those with follow up data and their inclusion could change the interpretation of results [104].
Another limitation is that pain assessed with questionnaires does not take into account the effect of pain medications and assistive aids. About 58% of women and 40% of men report taking pain medications after TKR because of pain in the operated knee [105] and we must recognise that pain levels at follow up without this treatment might be considerably higher. Even with treatment, around 20% of patients report chronic pain after TKR [10] and in the context of a blinded RCT we should expect to be able to identify effects of peri-operative treatments.
We summarised p-values to assess the strength of evidence but, as statistically strong evidence may not reflect clinically important results [106], where possible we also compared effect sizes with MCIDs. Our review considered a diverse range of interventions at a specific time in the TKR pathway and, as we were unable to make clinical practice recommendations, we did not adopt the GRADE system [107] for this review.
An alternative approach to the prevention of chronic pain after TKR is the individualisation of care based on pain phenotype, genetic, psychosocial and other factors [108]. An example of this might be the peri-operative treatment only of individuals with neuropathic pain with pregabalin, as opposed to the non-stratified provision in the RCT of Buvanendran and colleagues [53]. In an RCT with pregabalin provided to patients with painful HIV-neuropathy, while no overall benefit was seen, a group with hyperalgesia responded to pregabalin treatment [109].
Our systematic review of peri-operative interventions brings together evidence on interventions in the peri-operative phase of the TKR pathway. Whilst not supportive of the inclusion of specific interventions in clinical practice to optimise long-term pain outcomes, there are clearly areas that merit research. High quality studies assessing long-term pain after peri-operative interventions are feasible and necessary to ensure that patients with osteoarthritis achieve good long-term outcomes after TKR.

ACKNOWLEDGEMENT
We thank Dr Mario Moric for conducting additional analyses on the study of Buvanendran and colleagues [53].

AUTHOR CONTRIBUTIONS
All authors, ADB, JD, RG-H and AWB, contributed to the conception and design of the study.
ADB, JD and VW undertook the systematic review. ADB and JD carried out the risk of bias assessments. ADB drafted the article with revisions by JD, VW, RG-H and AWB. All authors approved the final version for publication.

ROLE OF THE FUNDING SOURCE
This article presents independent research funded by the National Institute for Health Research

COMPETING INTERESTS STATEMENT
The authors report no competing interests.
Saline infusion. Commenced before surgical incision and continued until wound bandaged or syringe empty. Aveline  Catheter inserted for epidural drug administration. LIA 60 ml 0.25% bupivacaine with epinephrine infiltrated into the wound at capsule closure. From completion of surgery until 32-42 hours post-operative, epidural infusion of fentanyl (5μg/ml) and bupivacaine (1mg/ml) initiated using continuous basal infusion of 6ml/hr with epidural PCA bolus doses (maximum 10ml/hr). Patients transitioned to oral opioid (morphine, oxycodone, and hydromorphone) as required. All patients received preoperative oral celecoxib 400mg 1-2 hours before surgery and 200mg twice daily for 3 days in hospital. In the pregabalin group the incidence of neuropathic pain measured using S-LANSS was 0% (0/113) and 5.2% (6/115) in the placebo group (p=0.014). No clinically significant adverse events up to 6 months and no falls. Sedation, confusion and dry mouth more frequent in pregabalin than placebo group on day of surgery and first postoperative day.

Instructions to authors
Complete this checklist by entering the page numbers from your manuscript where readers will find each of the items listed below.
Your article may not currently address all the items on the checklist. Please modify your text to include the missing information. If you are certain that an item does not apply, please write "n/a" and provide a short explanation.
Upload your completed checklist as an extra file when you submit to a journal.
In your methods section, say that you used the PRISMA reporting guidelines, and cite them as: Eligibility criteria #6 Specify study characteristics (e.g., PICOS, length of follow-up) and report characteristics (e.g., years considered, language, publication status) used as criteria for eligibility, giving rational 5 Information sources #7 Describe all information sources in the search (e.g., databases with dates of coverage, contact with study authors to identify additional studies) and date last searched.

6
Search #8 Present full electronic search strategy for at least one database, including any limits used, such that it could be repeated.

See note 1
Study selection #9 State the process for selecting studies (i.e., for screening, for determining eligibility, for inclusion in the systematic review, and, if applicable, for inclusion in the meta-analysis).

5,6
Data collection process #10 Describe the method of data extraction from reports (e.g., piloted forms, independently by two reviewers) and any processes for obtaining and confirming data from investigators.

6
Data items #11 List and define all variables for which data were sought (e.g., PICOS, funding sources), and any assumptions and simplifications made.

5/6
Risk of bias in individual studies #12 Describe methods used for assessing risk of bias in individual studies (including specification of whether this was done at the study or outcome level, or both), and how this information is to be used in any data synthesis.

16-23
Risk of bias across studies #22 Present results of any assessment of risk of bias across studies (see Item 15).

See note 6
Additional analysis #23 Give results of additional analyses, if done (e.g., sensitivity or subgroup analyses, meta-regression [see Item 16]).

16-23
Summary of Evidence #24 Summarize the main findings, including the strength of evidence for each main outcome; consider their relevance to key groups (e.g., health care providers, users, and policy makers

16-23
Limitations #25 Discuss limitations at study and outcome level (e.g., risk of bias), and at review level (e.g., incomplete retrieval of identified research, reporting bias).

24-25
Conclusions #26 Provide a general interpretation of the results in the context of other evidence, and implications for future research.

23-25
Funding #27 Describe sources of funding or other support (e.g., supply of data) for the systematic review; role of funders for the systematic review.

ABSTRACT Objectives
For many people with advanced osteoarthritis, total knee replacement (TKR) is an effective treatment for relieving pain and improving function. Features of peri-operative care may be associated with the adverse event of chronic pain six months or longer after surgery; effects may be direct, e.g. through nerve damage or surgical complications, or indirect through adverse events. This systematic review aims to evaluate whether non-surgical peri-operative interventions prevent long-term pain after TKR.

Methods
We conducted a systematic review of peri-operative interventions for adults with osteoarthritis receiving primary TKR evaluated in a randomised controlled trial (RCT). We searched The Cochrane Library, MEDLINE, Embase, PsycINFO and CINAHL to February 2018. After screening, two reviewers evaluated articles. Studies at low risk of bias according to the Cochrane tool were included.

Interventions
Peri-operative non-surgical interventions; control receiving no intervention or alternative treatment.

Primary and secondary outcome measures
Pain or score with pain component assessed at six months or longer post-operative.

44
RCTs at low risk of bias assessed long-term pain. Intervention heterogeneity precluded meta-analysis and definitive statements on effectiveness. There was good-quality evidence for a small benefit for reduced long-term pain with local infiltration analgesia (3 studies), ketamine infusion (1 study), pregabalin (1 study), and electric muscle stimulation (2 studies). No concerns relating to long-term adverse events were reported. In 5 RCTs, tranexamic acid to prevent blood loss was not associated with long-term pain. Many extensively researched interventions including venous thromboembolism prevention have not been evaluated in relation to long-term pain. To prevent chronic pain after TKR, peri-operative interventions including components of multimodal analgesia, early rehabilitation and supported discharge, electrical stimulation and anabolic steroids show small benefits meriting further research. Tranexamic acid use is not associated with chronic pain but the long-term consequences of many widely researched treatments have not been reported.

STRENGTHS AND LIMITATIONS
• For the first time, this systematic review brings together contemporary evidence on aspects of peri-operative care for people with total knee replacement and their effects on longterm pain.
• Only studies assessed to be at low risk of bias were included in the narrative synthesis.
• Intervention and outcome heterogeneity precluded meta-analysis.

KEYWORDS
Total knee replacement; Systematic review; Randomised controlled trial; Peri-operative care; Long-term pain

BACKGROUND
In the US about 13% of men and 19% of women will be diagnosed with knee osteoarthritis and over half will receive a total knee replacement (TKR) [1]. For people with advanced osteoarthritis unresponsive to pharmacological or conservative treatments, TKR aims to relieve pain and improve function. In the UK nearly 100,000 primary TKRs were performed in 2017[2,3] and in the USA in 2010, an estimated 4.7 million people were living with a TKR [4]. Despite good outcomes for many, some people report long-term pain and are disappointed with their surgery [5,6]. After TKR, pain levels plateau from about 6 months [7,8] after which persistent pain is considered "chronic" [9] and is reported by 10-34% of patients [10].
The mechanisms that influence the development of chronic pain after total knee replacement may be biological, mechanical and psychosocial. Biological causes include the sensitising impact of long-term pain from osteoarthritis [11,12], inflammation, infection and localised nerve injury [13]. Mechanical causes include altered gait, prosthesis loosening, and effects on ligaments [14,15]. Psychological factors including depression and catastrophizing may also influence outcomes [16][17][18][19]. Much research has focused on pre-operative predictors of outcomes and these include pain intensity, presence of widespread pain, anxiety, depression and catastrophizing. [10,20] However, attempts to target or modify pre-operative care have, as yet, shown no benefit regarding chronic pain or other long-term patient outcomes [10,[21][22][23].
Peri-operative risk factors suggest that appropriate interventions may reduce long-term pain.
For example, acute post-operative pain, which may be a direct consequence of the operation, anaesthetic protocol and subsequent analgesia, or related to particular aspects of care, is an acknowledged risk factor for chronic post-surgical pain [24].
In the peri-operative period from hospital admission to the early stages of recovery, care focuses on acute pain management, prevention of adverse events, facilitation of early mobilisation and timely discharge. However, for people with osteoarthritis the key aim of TKR is the achievement of a long-term painless and well-functioning knee with no adverse events. All aspects of peri-operative care should work together to achieve this.
Any treatment in the peri-operative period including pain management, blood conservation, deep vein thrombosis (DVT) and infection prevention, and inpatient rehabilitation could potentially affect patient recovery and chronic pain, either directly or indirectly. Direct mechanisms may be through prevention of nerve damage [25], post-thrombotic syndrome [26], reperfusion injury [27] and articular bleeding [28]. For other treatments, pathways leading to long- Our systematic review of randomised controlled trials (RCTs) aims to evaluate the effectiveness of treatments in the peri-operative period in preventing long-term pain after TKR. By focusing on studies with low risk of bias we aim to identify interventions with robust evidence of long-term effectiveness and identify gaps in the research base.

METHODS
The systematic review protocol was registered (PROSPERO CRD42017041382) and PRISMA reporting guidelines used [30]. A checklist is included as Supplementary material.

Patient and public involvement
As part of the STAR programme of research (NIHR RP-PG-0613-20001), this review benefited from extensive patient and public involvement. Advice was sought from patients and stakeholders at a group discussion in March 2016 with decisions made on inclusion criteria and outcomes. Our patient advisory group comprises five patients with experience of long-term pain after TKR, supported by a dedicated co-ordinator. This group will advise on dissemination of the study results to a general audience including plain language summaries.

Eligibility criteria
Studies were eligible if they satisfied PICOS criteria defined in the protocol. Participants were adults receiving unilateral primary TKR with osteoarthritis in at least 75% of patients.
Pharmacological or non-pharmacological interventions commenced in the peri-operative setting with "peri-operative" reflecting the time from hospital admission to immediately post-discharge.
Interventions relating to implant designs and surgical procedures were excluded. The comparator was usual care, placebo or an alternative intervention. Outcomes were, in preference, patient-reported joint-specific pain intensity measured by tools such as the Western

Screening and data extraction
We imported records into Endnote X7 (Thomson Reuters). An initial screen by one reviewer excluded clearly irrelevant articles. Subsequently, abstracts and full articles were screened independently by two reviewers and reasons for exclusion recorded.
Data were extracted onto piloted forms and an Excel spreadsheet by one reviewer, specifically: country; dates; participants (indication, age, sex); inclusion and exclusion criteria; intervention and control content; setting, timing, duration and intensity of intervention; follow up intervals; losses to follow up; pain outcome data; and serious adverse events. Data was checked against source material by a second reviewer.
Authors were contacted for missing data, and data provided for previous reviews was used [10,31].

Quality assessment
Potential sources of bias were assessed by two experienced reviewers using the Cochrane risk of bias tool [32], specifically: the randomisation process; deviations from intended interventions; missing outcome data (>20%), measurement of the outcome; and selection of the reported result. Studies with serious concerns relating to risk of bias were considered high risk and those with limited reporting unclear risk. Studies with high or unclear risk of bias were excluded from the narrative synthesis but are included in supplementary summary tables with reasons for exclusion. Insufficient studies with similar interventions and outcomes were identified for meta-analysis, and a narrative synthesis is presented. Results reported with p-values ≤0.001 were considered "strong" evidence of effectiveness [33], p-values 0.001-0.05 "some" evidence, and p-values 0.05-0.1 "weak" evidence. When authors reported results "statistically significant" with no p-value, this was noted. Where possible, effect sizes were compared with published minimal clinically important differences (MCID). Concerns relating to adverse events were summarised. Details of 44 studies assessed to be at low risk of bias are summarised in Table 1. In 34 studies, patients received TKR exclusively for osteoarthritis and in three, 75% or more patients.

RESULTS
In seven studies there was no information on reason for surgery but there was no suggestion

Femoral nerve block
Femoral nerve blocks (FNB) were studied in 10 RCTs.
Three RCTs compared FNB with no FNB. In one study with 55 patients, WOMAC pain scores at one year were similar in patients receiving single-shot FNB and untreated controls [43]. All patients received local anaesthetic infiltration (LIA) and patient-controlled analgesia (PCA). In another study with all participants receiving LIA, 150 were randomised to receive single-shot FNB with or without sciatic nerve block (SNB), or general anaesthesia [37]. There were no differences in HSS scores between groups at six months. Continuous FNB was compared with oral hydrocodone opioid in 62 patients receiving PCA [39]. There was some evidence for 'pain using stairs' favouring hydrocodone (p=0.01) but no difference in overall NRS-rated pain at one year and concern over venous thromboembolism in 4/31 participants treated with hydrocodone.
In two RCTs, continuous FNB was compared with PCA. In one study with 60 participants, the KSS at six months was similar between groups [44]. In another study with 280 participants, there was some evidence for higher incidence of NRS-rated pain at six months in the PCA group than the FNB group (p=0.021) but not at 12 months (p=0.273). [40] Two RCTs compared FNB with LIA. In one study, all 157 participants also received PCA [36]. At one year, KSS values were similar in single-shot FNB and LIA groups. In the other study, 94 participants were randomised to receive single-shot FNB with continuous epidural infusion or LIA through an intra-articular catheter [41]. VAS-rated pain was similar between groups at one year.
In two RCTs, FNB procedures were compared. In one study with 99 patients randomised to two FNB concentrations, there was no difference in WOMAC score between groups at 12 months [34]. In another study with 61 participants allocated to two different durations of FNB,  [35]. In these studies, all participants received either SNB [34] or PCA [35].
Single-shot FNB was compared with single adductor canal block in one RCT with 98 participants, all receiving LIA [38]. At six months there was no difference in VAS-rated pain.

Sciatic nerve block
In one study, 89 patients were randomised to single-shot SNB, continuous SNB, or PCA [42]. All patients received FNB. At 12 months, there were no differences in pain for single-shot SNB and continuous SNB on the WOMAC pain scale or VAS-rated pain at rest or during mobilisation.
Similarly, there were no differences between single-shot SNB and PCA in WOMAC pain scale or VAS-rated pain at rest or during mobilisation, or between continuous SNB and PCA.

Local anaesthetic infiltration
Four RCTs compared LIA with placebo. In one study, all 280 participants received FNB and PCA [50]. There was weak evidence that WOMAC pain scores were better in the LIA group at six (p=0.063) but not at 12 months (p=0.107) when the difference in means of 3.8/100 was lower than the MCID of 8-9/100 reported by Ehrich and colleagues [77]. In another study, 56 patients received LIA including ketorolac, or saline placebo, and all received PCA [47]. At one year, mean differences and confidence intervals provided weak evidence that OKS scores were better in the LIA group but the difference in means of 2.7/48 was less than the MCID of 4/48 reported by Beard and colleagues [78]. LIA before surgical incision was compared with placebo in one study with 120 participants [46]. None received FNB or PCA. There was weak evidence for a better KSS (function and knee score components) at six months in those receiving LIA (p=0.07) with a difference in means of 14.2/200 exceeding the MCID of 12.3/200 reported by Lee and colleagues [79]. In another study, all 51 participants received LIA intra-operatively, followed by PCA [49]. Those randomised to post-operative catheter-delivered LIA with ketorolac, or saline placebo had similar VAS-rated pain at six and 12 months.
LIA delivered as an injection and post-operative infusion was compared with epidural PCA in one study with 222 patients [45]. There was no difference between groups in OKS at 12 months.
In one study of 100 participants, LIA with or without corticosteroid were compared [48]. All patients received PCA. At two years there was no difference in OKS between groups.
Oral celecoxib In one RCT, 44 participants received oral celecoxib or placebo [51], as well as PCA. There were no differences between groups in KOOS or VAS-rated pain at 12 months.

Ketamine or nefopam infusion
In one RCT, ketamine infusion, nefopam infusion and saline placebo were compared in 75 patients, all of whom received PCA [52]. VAS-rated pain on movement did not differ between groups at 12 months. For the Douleur Neuropathique 4 (DN4) measure of neuropathic pain, there was some evidence favouring ketamine over placebo at six and 12 months (p=0.02), but overall, few patients reported neuropathic pain at 12 months.

Pregabalin
Oral pregabalin was compared with placebo in one RCT with 240 participants [53]. All received LIA and PCA. At six months, there was some evidence for better NRS pain in patients receiving pregabalin compared with placebo (p=0.0176) but the difference in means of 0.54/10 was less than the MCID of 1/10 reported by Salaffi and colleagues [80]. No participants receiving pregabalin reported neuropathic pain when assessed using the S-LANSS, compared with 5.2% of those receiving placebo (p=0.014). Patients receiving pregabalin were more likely to be sedated and confused in the first two days after surgery.

Tourniquet
Five studies with 399 participants explored tourniquet use to provide a bloodless field. Two studies each were from Australia and China, and one from Denmark. All were conducted at a single centre with participants recruited between 2008 and 2015. Sample sizes ranged from 20 to 150 participants, with a median of 65. The range of mean ages of participants in randomised groups was 66 to 71 years and in 3/5 studies, a majority of participants were women.
In three RCTs, participants received TKR with or without a tourniquet. In one study with 64 patients, a difference in KOOS pain favouring tourniquet use was not significant at six or 12 months [54]. In another study with 20 patients, the OKS was not significantly different between groups at six or 12 months [56]. There were three blood transfusions in the tourniquet group, compared with none in the 'no tourniquet' group. In the third study with 100 participants, VASrated pain and HSS scores were similar between groups at 6 months [55]. Six cases of wound ooze occurred in the tourniquet group.
In two RCTs, short and long-duration tourniquet use were compared. In one study with 65 participants, there was weak evidence based on graphical representation of means and confidence intervals for improved OKS at 12 months in the long-duration group and the difference in means of 5/48 [57] was greater than the MCID of 4/48. Adverse events were reported by 62% of participants receiving short-duration tourniquet compared with 38% in the long-duration group. The study was terminated early as 10 blood transfusions were required in the short-duration group compared with three in the long-duration group. In the second study with 150 participants, tourniquets were used in three different periods during surgery [58]. At six months, there were no differences between groups in HSS scores.

Blood conservation
Seven studies with 829 participants evaluated strategies to limit blood loss after TKR. Two studies were from Thailand, and one each from China, France, South Korea, the UK and the USA. All were conducted at a single centre with participants recruited between 2003 and 2015 when stated. Sample sizes ranged from 48 to 180 participants, with a median of 106. One study had three trial arms. The range of mean ages of participants in randomised groups was 65 to 74 years and in all studies, a majority of participants were women.

Tranexamic acid
Five RCTs evaluated tranexamic acid.
Tranexamic acid injections or infusions were compared with saline placebo or untreated control in four RCTs [55,61,64,65]. In all studies, control patients required more blood transfusions. In one study including 180 participants comparing intravenous tranexamic acid with untreated controls, there was no significant difference in WOMAC pain scores at one year [61]. In another study with 48 participants comparing intra-articular tranexamic acid injection with saline placebo, there was no significant difference in WOMAC scores at six months [64]. One study with 135 participants compared two intra-articular tranexamic acid doses and saline control [65]. There were no significant differences in WOMAC scores at one year. Intravenous and intra-articular tranexamic was compared with untreated controls in one study with 100 participants [55]. VASrated pain at six months was similar between groups, but there was strong evidence favouring tranexamic acid for HSS scores (p<0.001) although the difference in means of 1.4/100 was lower than the MCID of 8.3/100 reported by Singh and colleagues [81].
In one study, continuous tranexamic acid infusion was compared with a single bolus in 106 patients [60]. There was no difference between groups in KSS at six months or blood loss. In one RCT with 80 participants, thrombin infusion was compared with untreated control [62]. At one year there was no difference between groups in pain measured on the KSS.

Flexion or extension
For blood management, operated knees were kept in passive flexion or passive extension after surgery in one RCT with 180 patients [63]. At one year, OKS was similar between groups.
Transfusion requirement was greater in patients with passive extension.

Compression bandage
One RCT conducted at a single UK centre with 49 participants recruited between 2013 and 2014 compared compression bandaging to reduce post-operative knee swelling with standard bandaging. The mean age of participants was about 69 years and a majority were women. OKS was similar in randomised groups at six months [59].

Wound management
One RCT with recruitment in 2011 at a single centre in South Korea evaluated a wound care strategy to limit post-operative scar pain. The mean age of participants was about 69 years and a majority were women. Investigators compared silicone gel application to the surgical scar with placebo in 100 participants [75]. There were no significant differences in VAS-rated pain at six and 12 months.

Denusomab
One RCT evaluated use of the antiresorptive monoclonal antibody Denusomab to promote bone healing. The study was conducted in two centres in Sweden with recruitment of 50 participants between 2012 and 2014. The mean age of participants was about 65 years and a majority were women. At 12 and 24 months there were no significant differences between groups in KOOS pain [66].  CPM at high flexion from post-operative day 3-7 [68]. There was no significant difference between groups in KSS at two years. In the other study, 147 participants were randomised to CPM with increasing range of movement from day 1-6, early flexion CPM from day 0-6, or no CPM [67]. There were no significant differences between groups in KSS at 12 months.

Electrical stimulation
Two In one study with 76 participants receiving transcutaneous electric muscle stimulation from postoperative day two for six weeks or no intervention, Short Form 36 bodily pain showed strong evidence for greater improvement at one year in the intervention group compared to control (p<0.001) [69]. The difference in means of 12.5/100 was close to the MCID of 16.9/100 reported by Escobar and colleagues [82]. There were no differences in OKS or KSS scores. In another study with 30 participants, pulsed electromagnetic fields from post-operative day 7 were compared with untreated control [70]. At 12 months, there was some evidence that VAS-rated pain was lower in intervention patients compared with controls (p<0.05). The difference in means of 2.1/10 was greater than the MCID of 16.1/100 reported by Danoff and colleagues [83].
Knee swelling was common during the intervention.

Walking guidance and training
In one study, 86 participants were randomised to walking guidance and training from postoperative day two or no intervention further to standard rehabilitation [71]. At six months, there was some evidence that those receiving intervention had lower VAS-rated pain (p<0.01) and

Flexion or extension during knee closure
Targeting improved functional recovery, wound closure performed in 90° flexion was compared with wound closure in full extension in one study with 80 participants [74]. There was no difference between groups in VAS-rated pain at six months.

Aquatic therapy
In one study with 185 participants, aquatic therapy commencing on post-operative day six was compared with aquatic therapy commencing on day 14 [72]. Patients reported similar WOMAC pain at 12 and 24 months.

Supported early discharge
In one study, early discharge supported by physiotherapist home visits and outpatient or selfdirected physiotherapy was compared with two weeks of rehabilitation centre-based usual care [73]. The study included 234 individuals receiving TKR or total hip replacement. Compared with usual care, there was weak evidence that patients with early discharge had lower WOMAC pain scores at 12 months (p=0.08). The difference in means of 4 was less than the MCID of 8-9/100. Results were not presented separately but did not differ between patients with TKR or total hip replacement.

Anabolic steroids
Searches identified one study of anabolic steroids to improve post-operative muscle strength

Interventions with no long-term outcome
Interventions with lack of RCT evidence are summarised in Figure 1.
While 148 RCTs of DVT prophylaxis were identified, only five reported long-term follow up, none of which included a pain or outcome score. Among 29 RCTs of antibiotic prophylaxis, 16 reported long-term follow up, but none included a pain or outcome score. Six RCTs evaluated the use of bisphosphonates and, although all reported long-term follow up, none reported pain or an outcome score. One study reported long-term follow up of an RCT of teriparatide but included no data on pain.
For some interventions, RCTs with long-term pain outcomes were identified, but none were at low risk of bias: cold therapy; guided imagery; platelet rich plasma; and trigger point needling.
Aspects of peri-operative care evaluated in RCTs but lacking long-term pain follow up were: Consistent with its status as a key peri-operative risk factor, a major focus of research into improving long-term pain after TKR has been through prevention of acute post-operative pain using multimodal analgesia. Our review provides good quality evidence for a small benefit for intra-articular LIA injections, as previously shown in short-term studies [31,84], oral pregabalin, oral opioids, and in relation to neuropathic pain, ketamine infusion. As well as potential benefits for reduced long-term pain, future studies will need to consider concerns associated with these interventions which may not have been identified in small studies including infection [31], venous thromboembolism [39] and sedation [53].
Nerve blocks are effective for managing peri-operative pain [85] but we identified no long-term benefit. In single studies, there was no benefit for nefopam infusion, oral celecoxib or LIA with additional corticosteroid. Regarding future studies, standardisation of the multi-modal regimen With such an approach, convincing evidence will accrue to guide multimodal pain management.
Some interventions targeted the prevention of adverse events and facilitation of early mobilisation. Tranexamic acid is highly effective in reducing blood transfusions during TKR [86] and we found no evidence that tranexamic acid affects long-term pain or, consistent with registry studies [87,88], adverse events. Single RCTs of thrombin infusion and maintenance of knee in flexion to prevent blood loss showed no effect on long-term pain. Tourniquets improve intraoperative visualisation of the joint, reduce blood loss and facilitate cement fixation but are associated with nerve damage, delayed recovery, acute pain and need for analgesics [89,90].
The RCTs we identified showed no effects of tourniquet use on long-term pain.
As shown in a previous review[91], there was no suggestion that CPM affects long-term pain.
There was good quality evidence for a small benefit for reduced long-term pain in patients receiving walking training, anabolic steroid injection, electrical stimulation and supported discharge.
For some interventions a direct mechanism is clear, but for others, reasons for long-term impact are less obvious. This may explain why, for example, no studies evaluated DVT prophylaxis with long-term follow up excepting a small number reporting adverse events. However, treatments to prevent symptomatic DVTs which occur in about 1% of treated patients [92] also reduce the incidence of asymptomatic DVT observed in about 28% of treated patients [93] and this may have long-term benefits. Conversely, new anticoagulants are associated with bleeding [94], which may increase the risk of wound complications [95] and joint infection [96] which are associated with long-term pain [97,98].
Our study is limited by the lack of meta-analysis which was not appropriate due to intervention and outcome heterogeneity. In the context of perioperative pain management, this was noted previously [84]. Our approach to assessing the evidence was a narrative synthesis of studies with low risk of bias. While this may seem overly restrictive, Cochrane risk of bias assessment allows us to screen out studies with important issues that may affect the validity of results. The main potential source of bias was incomplete outcome assessment. Although studies with longterm follow up are naturally at higher risk of missing data, we maintained a standard in this domain as it is recognised that research participants who do not complete follow up assessments differ in outcomes from those with follow up data and their inclusion could change the interpretation of results [99]. Another limitation is that pain assessed with questionnaires does not take into account the effect of pain medications and assistive aids. About 58% of women and 40% of men report taking pain medications after TKR because of pain in the operated knee [100] and we must recognise that pain levels at follow up without this treatment might be considerably higher. Even with treatment, around 20% of patients report chronic pain after TKR [10] and in the context of a blinded RCT we should expect to be able to identify effects of peri-operative treatments.
We summarised p-values to assess the strength of evidence but, as statistically strong evidence may not reflect clinically important results [101], where possible we also compared effect sizes with MCIDs. Our review considered a diverse range of interventions at a specific time in the TKR pathway and, as we were unable to make clinical practice recommendations, we did not adopt the GRADE system [102] for this review.
An alternative approach to the prevention of chronic pain after TKR is the individualisation of care based on pain phenotype, genetic, psychosocial and other factors [103]. An example of this might be the peri-operative treatment only of individuals with neuropathic pain with pregabalin, as opposed to the non-stratified provision in the RCT of Buvanendran and colleagues [53]. In an RCT with pregabalin provided to patients with painful HIV-neuropathy, while no overall benefit was seen, a group with hyperalgesia responded to pregabalin treatment [104].
Our systematic review of peri-operative interventions brings together evidence on interventions in the peri-operative phase of the TKR pathway. There was good quality evidence for some interventions of a small benefit for reduced long-term pain, and whilst not supportive of the inclusion of specific interventions in clinical practice, there are clearly areas that merit research.
High quality studies assessing long-term pain after peri-operative interventions are feasible and necessary to ensure that patients with osteoarthritis achieve good long-term outcomes after TKR.

ACKNOWLEDGEMENT
We thank Dr Mario Moric for conducting additional analyses on the study of Buvanendran and colleagues.

AUTHOR CONTRIBUTIONS
All authors, ADB, JD, RG-H and AWB, contributed to the conception and design of the study.
ADB, JD and VW undertook the systematic review. ADB and JD carried out the risk of bias

COMPETING INTERESTS STATEMENT
The authors report no competing interests.

DATA STATEMENT
All data relevant to the study are included in the article or uploaded as supplementary information.
Legend Figure 1.         Intraoperative sedation with iv propofol at discretion of anaesthesiologist. Lumbar spinal anaesthesia with 12mg 0.5% bupivacaine. Postoperative i.v. PCA with fentanyl 50µg/ml set to deliver 25µg every 5 min as needed. Celecoxib 100mg and acetaminophen 650mg on arrival in recovery room and every 12 and 6 hrs respectively. Breakthrough medication with intramuscular ketorolac 10 mg every 4 hrs. Intrathecal injection of 15mg bupivacaine and 100μg morphine. General anaesthesia. After surgery 1.5g paracetamol and then 750mg every 4 hours; PCA with morphine 2mg boluses with 10-minute lockout; morphine rescue 2.5mg intravenously as required; and rescue oral ibuprofen 800mg. 6 months 3 protocol breaches and 1 patient with uncontrolled pain. High risk of bias due to non-ITT reporting and recruitment difficulties 2/5 ketamine group had mild/moderate pain on the WOMAC pain scale at 26 weeks or failed to improve compared with 5/7 controls. 1 adverse psycho-mimetic effect not attributed to intervention or control treatment Ketamine 0.5mg/kg bolus followed by 4μg/kg/min infusion. Commenced before surgical incision and continued until wound bandaged or syringe empty.

Saline infusion.
Commenced before surgical incision and continued until wound bandaged or syringe empty. Catheter inserted for epidural drug administration. LIA 60 ml 0.25% bupivacaine with epinephrine infiltrated into the wound at capsule closure. From completion of surgery until 32-42 hours post-operative, epidural infusion of fentanyl (5μg/ml) and bupivacaine (1mg/ml) initiated using continuous basal infusion of 6ml/hr with epidural PCA bolus doses (maximum 10ml/hr). Patients transitioned to oral opioid (morphine, oxycodone, and hydromorphone) as required. All patients received preoperative oral celecoxib 400mg 1-2 hours before surgery and 200mg twice daily for 3 days in hospital. In the pregabalin group the incidence of neuropathic pain measured using S-LANSS was 0% (0/113) and 5.2% (6/115) in the placebo group (p=0.014). No clinically significant adverse events up to 6 months and no falls. Sedation, confusion and dry mouth more frequent in pregabalin than placebo group on day of surgery and first postoperative day.

Instructions to authors
Complete this checklist by entering the page numbers from your manuscript where readers will find each of the items listed below.
Your article may not currently address all the items on the checklist. Please modify your text to include the missing information. If you are certain that an item does not apply, please write "n/a" and provide a short explanation.
Upload your completed checklist as an extra file when you submit to a journal.
In your methods section, say that you used the PRISMA reporting guidelines, and cite them as: Eligibility criteria #6 Specify study characteristics (e.g., PICOS, length of follow-up) and report characteristics (e.g., years considered, language, publication status) used as criteria for eligibility, giving rational 5-6 Information sources #7 Describe all information sources in the search (e.g., databases with dates of coverage, contact with study authors to identify additional studies) and date last searched.

6
Search #8 Present full electronic search strategy for at least one database, including any limits used, such that it could be repeated.

See note 1
Study selection #9 State the process for selecting studies (i.e., for screening, for determining eligibility, for inclusion in the systematic review, and, if applicable, for inclusion in the meta-analysis).

5,6
Data collection process #10 Describe the method of data extraction from reports (e.g., piloted forms, independently by two reviewers) and any processes for obtaining and confirming data from investigators.

6
Data items #11 List and define all variables for which data were sought (e.g., PICOS, funding sources), and any assumptions and simplifications made.

5/6
Risk of bias in individual studies #12 Describe methods used for assessing risk of bias in individual studies (including specification of whether this was done at the study or outcome level, or both), and how this information is to be used in any data synthesis.

16-23
Risk of bias across studies #22 Present results of any assessment of risk of bias across studies (see Item 15).

See note 6
Additional analysis #23 Give results of additional analyses, if done (e.g., sensitivity or subgroup analyses, meta-regression [see Item 16]).

24-25
Conclusions #26 Provide a general interpretation of the results in the context of other evidence, and implications for future research.

23-25
Funding #27 Describe sources of funding or other support (e.g., supply of data) for the systematic review; role of funders for the systematic review.

26
Author notes

Interventions
Peri-operative non-surgical interventions; control receiving no intervention or alternative treatment.

Primary and secondary outcome measures
Pain or score with pain component assessed at six months or longer post-operative. To prevent chronic pain after TKR, several peri-operative interventions show benefits and merit further research. Good quality studies assessing long-term pain after peri-operative interventions are feasible and necessary to ensure that patients with osteoarthritis achieve good long-term outcomes after TKR.

STRENGTHS AND LIMITATIONS
• For the first time, this systematic review brings together contemporary evidence on aspects of peri-operative care for people with total knee replacement and their effects on longterm pain.
• Only studies assessed to be at low risk of bias were included in the narrative synthesis.
• Intervention and outcome heterogeneity precluded meta-analysis.

KEYWORDS
Total knee replacement; Systematic review; Randomised controlled trial; Peri-operative care; Long-term pain

BACKGROUND
In the US about 13% of men and 19% of women will be diagnosed with knee osteoarthritis and over half will receive a total knee replacement (TKR) [1]. For people with advanced osteoarthritis unresponsive to pharmacological or conservative treatments, TKR aims to relieve pain and improve function. In the UK nearly 100,000 primary TKRs were performed in 2017[2,3] and in the USA in 2010, an estimated 4.7 million people were living with a TKR [4]. Despite good outcomes for many, some people report long-term pain and are disappointed with their surgery [5,6]. After TKR, pain levels plateau from about 6 months [7,8] after which persistent pain is considered "chronic" [9] and is reported by 10-34% of patients [10].
The mechanisms that influence the development of chronic pain after total knee replacement may be biological, mechanical and psychosocial. Biological explanations include the sensitising impact of long-term pain from osteoarthritis [11,12], inflammation, infection and localised nerve injury [13]. Mechanical explanations include altered gait, prosthesis loosening, and effects on ligaments [14,15]. Psychological factors including depression and catastrophizing may also influence outcomes [16][17][18][19]. Much research has focused on pre-operative predictors of outcomes and these include pain intensity, presence of widespread pain, anxiety, depression and catastrophizing. [10,20] However, attempts to target or modify pre-operative care have, as yet, shown no benefit regarding chronic pain or other long-term patient outcomes [10,[21][22][23].
Peri-operative risk factors suggest that appropriate interventions may reduce long-term pain.
For example, acute post-operative pain, which may be a direct consequence of the operation, anaesthetic protocol and subsequent analgesia, or related to particular aspects of care, is an acknowledged risk factor for chronic post-surgical pain [24].
In the peri-operative period from hospital admission to the early stages of recovery, care focuses on acute pain management, prevention of adverse events, facilitation of early mobilisation and timely discharge. However, for people with osteoarthritis the key aim of TKR is the achievement of a long-term painless and well-functioning knee with no adverse events. All aspects of peri-operative care should work together to achieve this.
Any treatment in the peri-operative period including pain management, blood conservation, deep vein thrombosis (DVT) and infection prevention, and inpatient rehabilitation could potentially affect patient recovery and chronic pain, either directly or indirectly. Direct mechanisms may be through prevention of nerve damage [25], post-thrombotic syndrome [26], reperfusion injury [27] and articular bleeding [28]. For other treatments, pathways leading to long- Our systematic review of randomised controlled trials (RCTs) aims to evaluate the effectiveness of treatments in the peri-operative period in preventing long-term pain after TKR. By focusing on studies with low risk of bias we aim to identify interventions with robust evidence of long-term effectiveness and identify gaps in the research base.

METHODS
The systematic review protocol was registered (PROSPERO CRD42017041382) and PRISMA reporting guidelines used [30]. A checklist is included as Supplementary material.

Patient and public involvement
As part of the STAR programme of research (NIHR RP-PG-0613-20001), this review benefited from extensive patient and public involvement. Advice was sought from patients and stakeholders at a group discussion in March 2016 with decisions made on inclusion criteria and outcomes. Our patient advisory group comprises five patients with experience of long-term pain after TKR, supported by a dedicated co-ordinator. This group will advise on dissemination of the study results to a general audience including plain language summaries.

Eligibility criteria
Studies were eligible if they satisfied PICOS criteria defined in the protocol. Participants were adults receiving unilateral primary TKR with osteoarthritis in at least 75% of patients.
Pharmacological or non-pharmacological interventions commenced in the peri-operative setting with "peri-operative" reflecting the time from hospital admission to immediately post-discharge.
Interventions relating to implant designs and surgical procedures were excluded. The comparator was usual care, placebo or an alternative intervention. Outcomes were, in preference, patient-reported joint-specific pain intensity measured by tools such as the Western

Screening and data extraction
We imported records into Endnote X7 (Thomson Reuters). An initial screen by one reviewer excluded clearly irrelevant articles. Subsequently, abstracts and full articles were screened independently by two reviewers and reasons for exclusion recorded.
Data were extracted onto piloted forms and an Excel spreadsheet by one reviewer, specifically: country; dates; participants (indication, age, sex); inclusion and exclusion criteria; intervention and control content; setting, timing, duration and intensity of intervention; follow up intervals; losses to follow up; pain outcome data; and serious adverse events. Data was checked against source material by a second reviewer.
Authors were contacted for missing data, and data provided for previous reviews was used [10,31].

Quality assessment
Potential sources of bias were assessed by two experienced reviewers using the Cochrane risk of bias tool [32], specifically: the randomisation process; deviations from intended interventions; missing outcome data (>20%), measurement of the outcome; and selection of the reported result. Studies with serious concerns relating to risk of bias were considered high risk and those with limited reporting unclear risk. Studies with high or unclear risk of bias were excluded from the narrative synthesis but are included in supplementary summary tables with reasons for exclusion. Insufficient studies with similar interventions and outcomes were identified for meta-analysis, and a narrative synthesis is presented. Results reported with p-values ≤0.001 were considered "strong" evidence of effectiveness [33], p-values 0.001-0.05 "some" evidence, and p-values 0.05-0.1 "weak" evidence. When authors reported results "statistically significant" with no p-value, this was noted. Where possible, effect sizes were compared with published minimal clinically important differences (MCID). Concerns relating to adverse events were summarised.

Femoral nerve block
Femoral nerve blocks (FNB) were studied in 10 RCTs.
Three RCTs compared FNB with no FNB. In one study with 55 patients, WOMAC pain scores at one year were similar in patients receiving single-shot FNB and untreated controls [43]. All patients received local anaesthetic infiltration (LIA) and patient-controlled analgesia (PCA). In another study with all participants receiving LIA, 150 were randomised to receive single-shot FNB with or without sciatic nerve block (SNB), or general anaesthesia [37]. There were no differences in HSS scores between groups at six months. Continuous FNB was compared with oral hydrocodone opioid in 62 patients receiving PCA [39]. There was some evidence for 'pain using stairs' favouring hydrocodone (p=0.01) but no difference in overall NRS-rated pain at one year and concern over venous thromboembolism in 4/31 participants treated with hydrocodone.
In two RCTs, continuous FNB was compared with PCA. In one study with 60 participants, the KSS at six months was similar between groups [44]. In another study with 280 participants, there was some evidence for higher incidence of NRS-rated pain at six months in the PCA group than the FNB group (p=0.021) but not at 12 months (p=0.273). [40] Two RCTs compared FNB with LIA. In one study, all 157 participants also received PCA [36]. At one year, KSS values were similar in single-shot FNB and LIA groups. In the other study, 94 participants were randomised to receive single-shot FNB with continuous epidural infusion or LIA through an intra-articular catheter [41]. VAS-rated pain was similar between groups at one year.
In two RCTs, FNB procedures were compared. In one study with 99 patients randomised to two FNB concentrations, there was no difference in WOMAC score between groups at 12 months [34]. In another study with 61 participants allocated to two different durations of FNB,  [35]. In these studies, all participants received either SNB [34] or PCA [35].
Single-shot FNB was compared with single adductor canal block in one RCT with 98 participants, all receiving LIA [38]. At six months there was no difference in VAS-rated pain.

Sciatic nerve block
In one study, 89 patients were randomised to single-shot SNB, continuous SNB, or PCA [42]. All patients received FNB. At 12 months, there were no differences in pain for single-shot SNB and continuous SNB on the WOMAC pain scale or VAS-rated pain at rest or during mobilisation.
Similarly, there were no differences between single-shot SNB and PCA in WOMAC pain scale or VAS-rated pain at rest or during mobilisation, or between continuous SNB and PCA.

Local anaesthetic infiltration
In six RCTs, treatment with LIA was investigated.
Three RCTs compared intra-operative LIA with placebo or no intervention. In one study, all 280 participants received FNB and PCA [50]. There was weak evidence that WOMAC pain scores were better in the LIA group at six (p=0.063) but not at 12 months (p=0.107) when the difference in means of 3.8/100 was lower than the MCID of 8-9/100 reported by Ehrich and colleagues [77].
In another study, 56 patients received LIA including ketorolac, or saline placebo, and all received PCA [47]. At one year, mean differences and confidence intervals provided weak evidence that OKS scores were better in the LIA group but the difference in means of 2.7/48 was less than the MCID of 4/48 reported by Beard and colleagues [78]. LIA before surgical incision was compared with placebo in one study with 120 participants [46]. None received FNB or PCA. There was weak evidence for a better KSS (function and knee score components) at six months in those receiving LIA (p=0.07) with a difference in means of 14.2/200 exceeding the MCID of 12.3/200 reported by Lee and colleagues [79].
In one study, 51 participants received LIA intra-operatively, followed by PCA [49]. Those randomised to further post-operative catheter-delivered LIA with ketorolac, or saline placebo had similar VAS-rated pain at six and 12 months.
LIA delivered as an injection and post-operative infusion was compared with epidural PCA in one study with 222 patients [45]. There was no difference between groups in OKS at 12 months.
In one study of 100 participants, LIA with or without corticosteroid were compared [48]. All patients received PCA. At two years there was no difference in OKS between groups.

Oral celecoxib
In one RCT, 44 participants received oral celecoxib or placebo [51], as well as PCA. There were no differences between groups in KOOS or VAS-rated pain at 12 months.

Ketamine or nefopam infusion
In one RCT, ketamine infusion, nefopam infusion and saline placebo were compared in 75 patients, all of whom received PCA [52]. VAS-rated pain on movement did not differ between groups at 12 months. For the Douleur Neuropathique 4 (DN4) measure of neuropathic pain, there was some evidence favouring ketamine over placebo at six and 12 months (p=0.02), but overall, few patients reported neuropathic pain at 12 months.

Pregabalin
Oral pregabalin was compared with placebo in one RCT with 240 participants [53]. All received LIA and PCA. At six months, there was some evidence for better NRS pain in patients receiving pregabalin compared with placebo (p=0.0176) but the difference in means of 0.54/10 was less than the MCID of 1/10 reported by Salaffi and colleagues [80]. No participants receiving pregabalin reported neuropathic pain when assessed using the S-LANSS, compared with 5.2% of those receiving placebo (p=0.014). Patients receiving pregabalin were more likely to be sedated and confused in the first two days after surgery.

Tourniquet
Five studies with 399 participants explored tourniquet use to provide a bloodless field. Two studies each were from Australia and China, and one from Denmark. All were conducted at a single centre with participants recruited between 2008 and 2015. Sample sizes ranged from 20 to 150 participants, with a median of 65. The range of mean ages of participants in randomised groups was 66 to 71 years and in 3/5 studies, a majority of participants were women.
In three RCTs, participants received TKR with or without a tourniquet. In one study with 64 patients, a difference in KOOS pain favouring tourniquet use was not significant at six or 12 months [54]. In another study with 20 patients, the OKS was not significantly different between groups at six or 12 months [56]. There were three blood transfusions in the tourniquet group, compared with none in the 'no tourniquet' group. In the third study with 100 participants, VASrated pain and HSS scores were similar between groups at 6 months [55]. Six cases of wound ooze occurred in the tourniquet group. In two RCTs, short and long-duration tourniquet use were compared. In one study with 65 participants, there was weak evidence based on graphical representation of means and confidence intervals for improved OKS at 12 months in the long-duration group and the difference in means of 5/48 [57] was greater than the MCID of 4/48. Adverse events were reported by 62% of participants receiving short-duration tourniquet compared with 38% in the long-duration group. The study was terminated early as 10 blood transfusions were required in the short-duration group compared with three in the long-duration group. In the second study with 150 participants, tourniquets were used in three different periods during surgery [58]. At six months, there were no differences between groups in HSS scores.

Blood conservation
Seven studies with 829 participants evaluated strategies to limit blood loss after TKR. Two studies were from Thailand, and one each from China, France, South Korea, the UK and the USA. All were conducted at a single centre with participants recruited between 2003 and 2015 when stated. Sample sizes ranged from 48 to 180 participants, with a median of 106. One study had three trial arms. The range of mean ages of participants in randomised groups was 65 to 74 years and in all studies, a majority of participants were women.

Tranexamic acid
Five RCTs evaluated tranexamic acid.
Tranexamic acid injections or infusions were compared with saline placebo or untreated control in four RCTs [55,61,64,65]. In all studies, control patients required more blood transfusions. In one study including 180 participants comparing intravenous tranexamic acid with untreated controls, there was no significant difference in WOMAC pain scores at one year [61]. In another study with 48 participants comparing intra-articular tranexamic acid injection with saline placebo, there was no significant difference in WOMAC scores at six months [64]. One study with 135 participants compared two intra-articular tranexamic acid doses and saline control [65]. There were no significant differences in WOMAC scores at one year. Intravenous and intra-articular tranexamic was compared with untreated controls in one study with 100 participants [55]. VASrated pain at six months was similar between groups, but there was strong evidence favouring tranexamic acid for HSS scores (p<0.001) although the difference in means of 1.4/100 was lower than the MCID of 8.3/100 reported by Singh and colleagues [81].

Thrombin infusion
In one RCT with 80 participants, thrombin infusion was compared with untreated control [62]. At one year there was no difference between groups in pain measured on the KSS.

Flexion or extension
For blood management, operated knees were kept in passive flexion or passive extension after surgery in one RCT with 180 patients [63]. At one year, OKS was similar between groups.
Transfusion requirement was greater in patients with passive extension.

Compression bandage
One RCT conducted at a single UK centre with 49 participants recruited between 2013 and 2014 compared compression bandaging to reduce post-operative knee swelling with standard bandaging. The mean age of participants was about 69 years and a majority were women. OKS was similar in randomised groups at six months [59].

Wound management
One RCT with recruitment in 2011 at a single centre in South Korea evaluated a wound care strategy to limit post-operative scar pain. The mean age of participants was about 69 years and a majority were women. Investigators compared silicone gel application to the surgical scar with placebo in 100 participants [75]. There were no significant differences in VAS-rated pain at six and 12 months.

Denusomab
One RCT evaluated use of the antiresorptive monoclonal antibody Denusomab to promote bone healing. The study was conducted in two centres in Sweden with recruitment of 50 participants between 2012 and 2014. The mean age of participants was about 65 years and a majority were women. At 12 and 24 months there were no significant differences between groups in KOOS pain [66].  1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59  60   F  o  r  p  e  e  r  r  e  v  i  e  w  o  n  l  y   21 participants were randomised to no CPM, CPM at low flexion from post-operative day 1-7, or CPM at high flexion from post-operative day 3-7 [68]. There was no significant difference between groups in KSS at two years. In the other study, 147 participants were randomised to CPM with increasing range of movement from day 1-6, early flexion CPM from day 0-6, or no CPM [67]. There were no significant differences between groups in KSS at 12 months.

Electrical stimulation
Two In one study with 76 participants receiving transcutaneous electric muscle stimulation from postoperative day two for six weeks or no intervention, Short Form 36 bodily pain showed strong evidence for greater improvement at one year in the intervention group compared to control (p<0.001) [69]. The difference in means of 12.5/100 was close to the MCID of 16.9/100 reported by Escobar and colleagues [82]. There were no differences in OKS or KSS scores. In another study with 30 participants, pulsed electromagnetic fields from post-operative day 7 were compared with untreated control [70]. At 12 months, there was some evidence that VAS-rated pain was lower in intervention patients compared with controls (p<0.05). The difference in means of 2.1/10 was greater than the MCID of 16.1/100 reported by Danoff and colleagues [83].
Knee swelling was common during the intervention. Walking guidance and training

Flexion or extension during knee closure
Targeting improved functional recovery, wound closure performed in 90° flexion was compared with wound closure in full extension in one study with 80 participants [74]. There was no difference between groups in VAS-rated pain at six months.

Aquatic therapy
In one study with 185 participants, aquatic therapy commencing on post-operative day six was compared with aquatic therapy commencing on day 14 [72]. Patients reported similar WOMAC pain at 12 and 24 months.

Supported early discharge
In one study, early discharge supported by physiotherapist home visits and outpatient or selfdirected physiotherapy was compared with two weeks of rehabilitation centre-based usual care [73]. The study included 234 individuals receiving TKR or total hip replacement. Compared with usual care, there was weak evidence that patients with early discharge had lower WOMAC pain scores at 12 months (p=0.08). The difference in means of 4 was less than the MCID of 8-9/100. Results were not presented separately but did not differ between patients with TKR or total hip replacement.

Anabolic steroids
Searches identified one study of anabolic steroids to improve post-operative muscle strength conducted in one centre in Australia with recruitment of 10 participants before 2010.

DISCUSSION
Much research in TKR aims to identify treatments that facilitate a speedy recovery with minimal short-term pain. However, patients choose to have joint replacement for long-term pain relief and reduction in functional limitations. Thus, changes to peri-operative care, supported by short-  1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59  Consistent with its status as a key peri-operative risk factor, a major focus of research into improving long-term pain after TKR has been through prevention of acute post-operative pain using multimodal analgesia. Our review provides good quality evidence for a small benefit for intra-articular LIA injections, as previously shown in short-term studies [31,84], oral pregabalin, oral opioids, and in relation to neuropathic pain, ketamine infusion. As well as potential benefits for reduced long-term pain, future studies will need to consider concerns associated with these interventions which may not have been identified in small studies including infection [31], venous thromboembolism [39] and sedation [53].
Nerve blocks are effective for managing peri-operative pain [85] but we identified no long-term benefit. In single studies, there was no benefit for nefopam infusion, oral celecoxib or LIA with additional corticosteroid. Regarding future studies, standardisation of the multi-modal regimen will allow evaluation of extra or alternative components in multiple studies in different settings.
With such an approach, convincing evidence will accrue to guide multimodal pain management.
Some interventions targeted the prevention of adverse events and facilitation of early mobilisation. Tranexamic acid is highly effective in reducing blood transfusions during TKR [86] and we found no evidence that tranexamic acid affects long-term pain or, consistent with registry studies [87,88], adverse events. Single RCTs of thrombin infusion and maintenance of knee in flexion to prevent blood loss showed no effect on long-term pain. Tourniquets improve intraoperative visualisation of the joint, reduce blood loss and facilitate cement fixation but are associated with nerve damage, delayed recovery, acute pain and need for analgesics [89,90].
The RCTs we identified showed no effects of tourniquet use on long-term pain.
As shown in a previous review[91], there was no suggestion that CPM affects long-term pain.
There was good quality evidence for a small benefit for reduced long-term pain in patients receiving walking training, anabolic steroid injection, electrical stimulation and supported discharge.
For some interventions a direct mechanism is clear, but for others, reasons for long-term impact are less obvious. This may explain why, for example, no studies evaluated DVT prophylaxis with long-term follow up excepting a small number reporting adverse events. However, treatments to  1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59  60   F  o  r  p  e  e  r  r  e  v  i  e  w  o  n  l  y   24 prevent symptomatic DVTs which occur in about 1% of treated patients [92] also reduce the incidence of asymptomatic DVT observed in about 28% of treated patients [93] and this may have long-term benefits. Conversely, new anticoagulants are associated with bleeding [94], which may increase the risk of wound complications [95] and joint infection [96] which are associated with long-term pain [97,98]. More peri-operative interventions with no information on long-term pain outcomes from RCTs are shown in Figure 1.
Our study is limited by the lack of meta-analysis which was not appropriate due to intervention and outcome heterogeneity. In the context of perioperative pain management, this was noted previously [84]. Our approach to assessing the evidence was a narrative synthesis of studies with low risk of bias. While this may seem overly restrictive, Cochrane risk of bias assessment allows us to screen out studies with important issues that may affect the validity of results. The main potential source of bias was incomplete outcome assessment. Although studies with longterm follow up are naturally at higher risk of missing data, we maintained a standard in this domain as it is recognised that research participants who do not complete follow up assessments differ in outcomes from those with follow up data and their inclusion could change the interpretation of results [99].
Another limitation is that pain assessed with questionnaires does not take into account the effect of pain medications and assistive aids. About 58% of women and 40% of men report taking pain medications after TKR because of pain in the operated knee [100] and we must recognise that pain levels at follow up without this treatment might be considerably higher. Even with treatment, around 20% of patients report chronic pain after TKR [10] and in the context of a blinded RCT we should expect to be able to identify effects of peri-operative treatments.
We summarised p-values to assess the strength of evidence but, as statistically strong evidence may not reflect clinically important results [101], where possible we also compared effect sizes with MCIDs. Our review considered a diverse range of interventions at a specific time in the TKR pathway and, as we were unable to make clinical practice recommendations, we did not adopt the GRADE system [102] for this review.
An alternative approach to the prevention of chronic pain after TKR is the individualisation of care based on pain phenotype, genetic, psychosocial and other factors [103]. An example of this might be the peri-operative treatment only of individuals with neuropathic pain with pregabalin, as opposed to the non-stratified provision in the RCT of Buvanendran and colleagues [53]. In an RCT with pregabalin provided to patients with painful HIV-neuropathy, while no overall benefit was seen, a group with hyperalgesia responded to pregabalin treatment [104].  1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59  60   F  o  r  p  e  e  r  r  e  v  i  e  w  o  n  l  y   25 Our systematic review of peri-operative interventions brings together evidence on interventions in the peri-operative phase of the TKR pathway. There was good quality evidence for some interventions of a small benefit for reduced long-term pain, and whilst not supportive of the inclusion of specific interventions in clinical practice, there are clearly areas that merit research.
High quality studies assessing long-term pain after peri-operative interventions are feasible and necessary to ensure that patients with osteoarthritis achieve good long-term outcomes after TKR.

ACKNOWLEDGEMENT
We thank Dr Mario Moric for conducting additional analyses on the study of Buvanendran and colleagues.

AUTHOR CONTRIBUTIONS
All authors, ADB, JD, RG-H and AWB, contributed to the conception and design of the study.
ADB, JD and VW undertook the systematic review. ADB and JD carried out the risk of bias assessments. ADB drafted the article with revisions by JD, VW, RG-H and AWB. All authors approved the final version for publication.

COMPETING INTERESTS STATEMENT
The authors report no competing interests.

DATA STATEMENT
All data relevant to the study are included in the article or uploaded as supplementary information.

Instructions to authors
Complete this checklist by entering the page numbers from your manuscript where readers will find each of the items listed below.
Your article may not currently address all the items on the checklist. Please modify your text to include the missing information. If you are certain that an item does not apply, please write "n/a" and provide a short explanation.
Upload your completed checklist as an extra file when you submit to a journal.

6
Search #8 Present full electronic search strategy for at least one database, including any limits used, such that it could be repeated.

See note 1
Study selection #9 State the process for selecting studies (i.e., for screening, for determining eligibility, for inclusion in the systematic review, and, if applicable, for inclusion in the meta-analysis).

5,6
Data collection process #10 Describe the method of data extraction from reports (e.g., piloted forms, independently by two reviewers) and any processes for obtaining and confirming data from investigators.

6
Data items #11 List and define all variables for which data were sought (e.g., PICOS, funding sources), and any assumptions and simplifications made.

24-25
Conclusions #26 Provide a general interpretation of the results in the context of other evidence, and implications for future research.

23-25
Funding #27 Describe sources of funding or other support (e.g., supply of data) for the systematic review; role of funders for the systematic review.

FNB long duration vs FNB short duration
FNB with ultrasound guidance. Initial dose of 10ml 2% lidocaine and 10ml 1% ropivacaine. 30 minutes before end of operation, catheter connected to PCA pump; patients received loading dose of 5ml of 0.15% ropivacaine followed by infusion of 0.15% ropivacaine at 5ml/hr, with bolus of 5mL i.v. PCA with tramadol 800mg, flurbiprofen axetil 100mg, and dexamethasone 5mg with saline to a volume of 80ml. Loading dose of 2ml followed by an infusion rate of 1 ml/hr with bolus of 2 ml. Lock time 15min. Premedication with oral paracetamol (1-2g). Spinal anaesthesia with 13-15mg bupivacaine 5mg/ml with 20μg fentanyl. If indicated, up to 10ml/hr 10mg/ml propofol for sedation. Acetaminophen 1g every 6 hours. i.v. PCA morphine for 48 hours after surgery (2mg bolus with 10 minutes lockout time). When PCA stopped, 10mg slow release oxycodone twice daily. 5mg oxycodone as rescue analgesia.