Intra-articular platelet-rich plasma injections versus intra-articular corticosteroid injections for symptomatic management of knee osteoarthritis: systematic review and meta-analysis

Musculoskeletal (MSK) conditions make up a significant workload in general practice (GP) and musculoskeletal services, with one in seven GP consultations being for MSK conditions [1]. Patients frequently present to their GP with knee osteoarthritis (OA) symptoms [2] and a recent analysis of primary care musculoskeletal referrals to secondary care [3] found that knee OA was the commonest reason for orthopaedic secondary care referral. Moreover, symptomatic OA is one of the leading causes of adult disability in the world, with significant economic impact [4]. Adding to this, the burden of knee OA continues to grow [5], estimated to affect between 10 and 25% of patients over 60 [4], alongside a population that is increasingly co-morbid and obese [5, 6].

The United Kingdom (UK) has national guidelines for the management of OA [7]. These guidelines include the use of intra-articular (IA) corticosteroid (CS) and local anaesthetic injections, followed by exercise prescription [7]. Only a small proportion of patients with knee OA meet physical activity guidelines due to the impact of pain on their functioning [8]. Such findings place even greater emphasis on the necessity for effective pain management options, including IA injections, especially in a climate with a lengthy waiting time for surgical intervention [9, 10], which negatively impacts patient quality of life [11].

IA CS injections are often prescribed before secondary care referral, attempting to provide symptomatic management and delay surgery. Although CS injections appear to improve pain scores in osteoarthritic patients for a limited time period [12], they are associated with side-effects [13] and do not appear to offer symptomatic improvement for longer than 6 weeks [12]. Indeed, some authors [13] have advised against using IA CS therapy because of the deleterious effects on articular cartilage [14], leading to a deterioration of the underlying joint OA. Previous studies have shown a statistically significant additional deterioration in articular cartilage compared to placebo, as well as an increased propensity for knee replacement in patients treated with CS injections [13, 15].

During the current COVID-19 pandemic of 2020, CS injections are being used judiciously due to the impact of CS on immunity and alternatives must be considered. Thus, research is needed to identify and show the effects of new management options for patients with knee OA, particularly to offer better non-operative pain management.

One such option might be IA platelet-rich plasma (PRP) injections, shown to reduce pain and improve function for knee OA patients in systematic reviews and meta-analyses [16, 17], with improvements reported to last up to 1 year [16]. IA PRP is not associated with the deleterious effects on cartilage that IA CS use is, making it a safer option, particularly when repeated injections may be required. IA PRP injections have been shown to have positive effects on chondrogenesis and mesenchymal stem cell proliferation [17, 18]. IA PRP injections have also been shown to decrease inflammatory markers and promote anti-inflammatory mediators, as well as reducing the expression of inflammatory enzymes [17, 19]. Therefore, there is strong biological plausibility to advocate IA PRP injections as a potential alternative to CS injections. However, IA PRP is not currently offered as a standard treatment to those with knee OA, and most studies do not compare IA PRP to IA CS/local anaesthetic injections; the current standard of care in the UK. There is also some debate surrounding how many times IA PRP injections should be given, with some authors suggesting using 3 injections [20, 21], separated by a week, whilst others use only one injection [22]. .PRP injections can also contain different concentrations of leukocytes and there is discourse regarding which is superior, leukoocyte-rich PRP (LR-PRP) as opposed to leukoocyte-poor PRP (LP-PRP) [23]. Thus a systematic review of the evidence is required to compare IA PRP to IA CS injections for the symptomatic management of knee OA, as well as to consider the effectiveness of the different forms of PRP (LR-PRP versus LP-PRP) and the number of IA PRP injections to be given as a course of treatment.

The primary aim is to investigate the effect of IA PRP injections versus IA CS injections for the symptomatic management of adults with knee OA. Secondary aims are to identify the effectiveness of one IA PRP injection compared to multiple injections for one course of treatment and whether LR-PRP is more effective to LP-PRP injections in the management of knee OA.

Detailed search strategies were developed for each electronic database searched, with input from a medical librarian. These have been based on the search strategy developed for Medical Literature Analysis and Retrieval System Online (MEDLINE) but were revised appropriately for the following databases: EMBASE, Scopus and Web of Science. An example of our search strategy is included as a supplementary file (Additional file 1). No limits were applied for language or timeframe. The most recent search was performed in June 2020. Any studies directly comparing injectable PRP with injectable corticosteroids for the management of symptomatic knee osteoarthritis were screened by both authors for additional references. The reviewers supplemented the electronic search strategy by using the Science Citation Index to perform citation tracking of the trials identified by the first step.

Inclusion criteria: Adults (Aged 18 or over) with symptomatic knee OA receiving IA PRP injections for symptomatic management of knee OA and comparing this treatment to IA CS injections.

Exclusion criteria: Adolescents (under 18 years of age); studies using co-interventions alongside either PRP or CS and studies not comparing IA PRP directly to IA CS.

Two reviewers (MM,NH) applied pre-agreed eligibility criteria to select studies for inclusion. One researcher screened (MM) and the senior author (NH) reviewed decisions, with researchers being blinded to the other’s decisions. Disagreements were resolved through discussion of the study’s methodology and included/removed at this point. Relevant data was extracted on prospective trial methodology; study participants (including grade of knee OA); study length (including year initiated); country of origin/nationality of participants; eligibility criteria; interventions (including the number of injections received, dosage of injection, method of delivery and type of PRP used (LR-PRP or LP-PRP)) and study design; study duration; follow-up; comparisons; outcome measures and results. Attempts were made where necessary to contact original investigators to request missing data. There was no blinding to study author, institution or journal.

One review author (MM) independently assessed each included study for risk of bias using the Risk of Bias 2 tool [18], following guidance from the Cochrane Handbook of Systematic Reviews of Interventions [19], with decisions reviewed by the senior author (NH). At this point, articles were also assessed for clinical and methodological heterogeneity.

The difference in mean, with corresponding standard error (SE), was extracted from each study as measured by VAS, WOMAC and KOOS at different time points of follow-up for symptomatic management of knee OA. Where studies used different outcome measures, the reviewers combined the relevant studies using the same outcome measure.

A random effects model was used to calculate a pooled standardised mean difference (MD) with 95% confidence intervals (CI) in WOMAC/VAS score (or subscores) comparing IA PRP to CS injections across studies. A SMD was used as the WOMAC/VAS measure the same underlying construct, pain. This analysis was repeated calculating the pooled mean difference (MD) restricted to studies that reported the same pain outcome. In one study, the median and range was converted to a mean and standard deviation using formulae described by Hozo et al. [25]. A random-effects model was used in all pooled analyses as we anticipated clinical heterogeneity. The Chi-squared test for heterogeneity was conducted and I² value was used to measure heterogeneity. Publication bias was assessed from inspection of funnel plots (an example from our primary analysis is included as Additional file 2). All analysis was performed using Review Manager 5.4.

Our search yielded 1566 studies, of which there were 579 original articles after duplicates were removed. Titles and abstracts were screened for 579 articles, at which point 567 were excluded as irrelevant to the study question. Full texts were identified for the remaining 12 texts. In total, eight trials were included within the analysis, published between 2017 and 2019. Details of the literature search are demonstrated in the flow chart (Fig. 1).

The characteristics of the eight included trials are presented in Table 1. The studies included 648 adult patients, with 206 (32%) being male. Mean patient age was 59 years. Radiographically, all studies used the Kellgren-Lawrence (KL) grading systems to determine severity of knee OA, with pooled information for six studies showing that 0.2% were grade I, 45% were grade II, 45% were grade III, and 9% were grade IV. Huang et al. [26] and Phul et al. [27] only reported average KL grades for CS and PRP groups (KL 1–2 and 2–4 respectively) and therefore were not included in the grade-percentage stratification above. The mean BMI of patients was 28.4 across six studies, with Freire et al. [28] simply reporting that 82% of their patients were obese and Phul et al. [27] that all patients had a BMI < 33 kg/m².

The Western Ontario and McMaster Universities Osteoarthritis Index (WOMAC) was the most frequently used outcome score (5 of 8 studies), with 2 of 8 using the Knee Injury and Osteoarthritis Outcome Score (KOOS) and 7 of 8 the visual analog scale (VAS). Khan et al. [29] reported WOMAC using individual sub-scores but did not calculate total scores. One study Freire et al. [28] reported results using the Knee Society Score (KSS) and Jubert et al. [22] additionally reported the SF-36 sub-score. Four studies reported outcomes at 1 month, 2 at 2 months, 5 at 3 months, 7 at 6 months, 1 at 9 months and 2 at 12 months.

Six of the eight individual studies reported that PRP was significantly better than CS, with the benefit most pronounced at 6 months, with two studies showing no significant difference between groups. Three studies used multiple PRP injections whereas six used single injections Uslu et al. [20] used both) and five studies used LR-PRP compared to three using LP-PRP (Huang et al. [26], Freire et al. [28], Jubert et al. [22]). No studies included patients with knee co-morbidities, knee injections within the previous 3 months or systemic diseases affecting the joints or blood (e.g., septicaemia, rheumatoid arthritis). Three of the studies were conducted in Europe, one in South America and four in Asia. Three studies recruited participants by approaching those attending their unit over a certain timeframe, one by asking for volunteers in the eligible patient population and one recruited through a list of patients referred to pain clinic. Three studies did not specify how patients were recruited.

Whilst most studies used similar dosages and preparations of CS, there was considerable variability in the preparation of PRP, summarised in Additional file 3.

A summary of the results of the risk of bias assessment are presented in Additional file 4 using the Cochrane RoB2 tool [32]. Seven studies achieved a low risk of bias, whilst one study Camurcu et al. [31] was presumed at high risk of bias as no form of randomisation was implemented. It is worth noting that this tool does not assess blinding within its bias domains, and a more complete assessment with a justification of the evaluation for each domain is presented in Additional file 5. Additionally, not all of the studies blinded participants and/or physicians, however this was either deemed unethical or unsuitable by several of the study authors. Results suggest overall bias is low and is unlikely to affect the outcome of this review.

In four studies (50%), all patients were analysed, with no study drop-outs. Of the remaining four, mean attrition rate in the CS group was 11% (9–14%) and 2.25% (0–14%) in the PRP group. Reasons for attrition were justified in each paper. Adverse events were reported in four (50%) papers, and all were deemed to be mild and self-limiting.

No paper appeared subject to sponsorship bias and we found no evidence of publication bias, for example, through searching the unpublished literature and journal abstracts for contradictory results. Considerable statistical heterogeneity was found across multiple timeframes; however, this was to be expected as many of the studies assessed different OA severity, across different timeframes, with different patient ages and PRP preparations. Moreover, similar studies assessing PRP compared with hyaluronic acid (HA) or HA compared to CS report comparable study heterogeneity [17, 33, 34].

Our primary analysis was at 6 months follow up, where there was evidence that mean score in the PRP group was 0.78 SMD (P < 0.005) lower than in the CS group ((95% CI 0.12, 0.90)). Table 2 shows the results from each performed analysis. The forest plot these results are derived from is included as Additional file 6. On individual analysis of the WOMAC or VAS scales, this is represented by an additional reduction of 9.51 or 0.97 respectively. At 1 month of follow-up there was a non-significant additional reduction in overall symptoms in the PRP group compared to the CS group. At 3 months, six studies reported outcomes with a small but significant reduction (P = 0.01) in favour of the PRP group (− 0.51 SD). One study reported outcomes at 9 months (WOMAC only), with an additional reduction of 8.04 in the PRP group. At 12 months there was a statistically non-significant reduction in favour of the PRP group. These analyses are represented on their individual scales in Table 3. The temporal relationship of the additional benefit from PRP over CS injections is represented in Fig. 2.

As outcomes were most commonly reported at 6 months, these were stratified by KL grade to identify the efficacy of PRP across different severities of knee OA (Fig. 3). In KL 1–2, two studies reported a non-significant additional benefit for PRP (− 0.18 SMD, − 3.86 on WOMAC scale, − 0.28 on VAS) as did two studies in KL 3–4 (− 1.32 SMD, − 0.32 on VAS). In four studies including patients with KL 2–3 knee OA, a significant reduction (P < 0.00001) of 1 SMD was observed (I² = 0%). On analysis of the individual scales this is represented as additional reduction of 9.1 on the WOMAC scale or 0.68 on the VAS scale.

As the WOMAC score does not specifically address sports outcomes, the KOOS sports/activities subscale (reported in two studies) was analysed (Fig. 4). At 1 month, CS and PRP injections were comparable (P = 0.52, magnitude 3.65). After this point, PRP was favoured, being statistically significant at 3 months (P = 0.05), with an improvement of 5.7 (I² = 29%). At 6 months, a near significant (P = 0.06) improvement was observed in favour of PRP, of magnitude 9.7.

Both LP-PRP and LR-PRP significantly reduced scores when compared with CS. LP-PRP showed an additional improvement of 0.61 SMD compared to CS (P = 0.00002, I² = 16%) whereas LR-PRP showed additional reduction of 0.98 SMD (P = 0.02, I² = 92%); thus LR-PRP appears more effective in reducing pain than LP-PRP. The reduction in WOMAC and VAS scores as a result is demonstrated in Table 3. Although LR-PRP appears to have a greater effect in this analysis compared to LP-PRP, significant statistical heterogeneity may explain some of the difference between these groups and should be considered when interpreting our results.

Single injections showed an additional benefit over CS of 0.52 SMD (P = 0.08, I² = 87%) and triple injections had an additional benefit of 0.94 SMD (P < 0.00001, I² = 12%), suggesting three injections are superior to one. It is worth noting that Uslu et al. [20] did not perform three CS injections in the control group, which may affect results; Uslu et al. [20] was further excluded from the analysis due to significant statistical heterogeneity, but found superior results with three injections of PRP. Both Khan et al. [29] and Naderi et al. [30] matched the number of injections given in the control groups, and these were spaced at equal timeframes.

With reference to CS injections, our results are in keeping with the literature, in that they appear equally efficacious to PRP injections at 1 month post-injection. A Cochrane review on the use of CS injections for knee OA conclude that the effects of the injection decreases over time, with no evidence to suggest that an effect remains 6 months post-injection [12]. Similarly, both this review and another review of Level 1 studies found that the effect from IA CS injections were only significant at 1 week [12, 35], with a general consensus that there is little benefit from the injections after a period of 6 weeks [35, 36].

Pain and stiffness scores in the PRP group continue to improve until 9 months post-injection (Fig. 2), with statistical significance from three months onwards. This is supported by previous evidence by Shen et al. [34] and Filardo et al. [37], who suggested a sustained effect following PRP injections of up to 12 and even 24 months [34, 37]. The former study was a systematic review and meta-analysis investigating the temporal effect of PRP injections compared to control (including HA, CS, placebo and ozone) which reported superior treatment effects from PRP at 3, 6 and 12 months; the latter was a single arm study investigating whether beneficial effects from PRP persist. At 24 months, patient scores were still improved compared to baseline, but the median duration of effect was 9 months. Our systematic review and meta-analysis showed that 6–9 months was the time point when patients will observe the greatest treatment effect from IA-PRP injections.

Seven of the eight studies reported results at 6 months post-injection; as such, results could be stratified by KL grade. The greatest significant effect was reported in KL 2–3 (early-moderate OA), supported by an earlier study finding that early OA showed a greater response and duration of effect than in severe OA [38]. In a review by Meheux et al. [17] comparing PRP to HA and placebo for knee OA, PRP was recommended to be reserved for patients with KL 1–3 [17]. This may be due to the ability of PRP to restore and protect cartilage [39, 40], as this may be more effective where there is more pre-existing joint cartilage.

A Cochrane review comparing CS to placebo injections reported a reduction of 0.4 SMD, or 1 on the VAS pain scale, compared to our observed additional mean reduction in the PRP group of − 0.51 SMD at 3 months or − 0.78 SMD at 6 months. In context, our results suggest the effect from PRP is nearly double that again of CS injections [12]. A recent systematic review comparing PRP to HA found that PRP produced an additional reduction of − 20.69 on the WOMAC scale at 6 months, with our results showing a reduction of − 9.51 compared to CS at the same timeframe [41]. Indeed the minimally clinically important difference (MCID) in knee OA has been previously identified as an improvement on the WOMAC scale of 1.33 points [42], of which these results are well in excess of at 6 months of follow-up. Patients should be reassured that their symptoms will continue to improve in the months post-injection, receiving the greatest relief as they approach six months.

On WOMAC subscore analysis, at 6 months pain and stiffness were significantly improved in the PRP group compared to CS, but overall function was not. This would suggest that most of the perceived benefit from PRP at this timeframe is symptomatic, and this hypothesis is reinforced by Filardo et al. [37]. The greatest effect in sporting improvement on the KOOS subscale was also observed at 6 months, likely due to reduced pain and stiffness improving sporting participation; physical activity often being a co-prescription which may in itself help improve OA symptoms [43, 44]. An improvement in sporting participation and exercise is additionally desirable in promoting weight loss, which can have a profound effect on improving function in knee OA [45].

We found a slightly greater effect from LR- compared with LP-PRP injections, however this was not significant between groups. LR-PRP contains more pro-inflammatory mediators whereas LP-PRP contains less, and there is debate over which is of greater benefit [21]. On one hand, it has been suggested that an initial pro-inflammatory phase is important for regeneration [46], whilst on the other hand, a greater concentration of anti-inflammatory mediators in LP-PRP is presumed favourable [47]. Our results advocate the former hypothesis, but this requires further investigation.

Another topic of debate, previously identified [48], is whether single or multiple PRP injections provide different treatment effects. We demonstrated a greater treatment effect with the use of multiple injections. Previous research has found either no benefit or harm from multiple injections [49] or that multiple injections improve joint functionality [50], with the caveat that multiple injection sites may increase risk of local reaction [48]. Adding to this, a recent RCT comparing PRP to HA injections found that triple injections were significantly better than single injections, but only in low grade OA [21]. A Guinea pig model found that both short term and long term benefits were more sustained and pronounced in early grade knee OA [51], and that only multiple injections stimulated a chondroprotective effect, suggesting a threshold dosage and that multiple injections improve the duration of effect of PRP. Thus PRP injections, both single and multiple, appear safe to use, with multiple injections appearing to offer greater symptomatic improvement than single PRP injections.

It has been proposed that outcomes following IA PRP injections are generally best for mild-moderate grade OA, and our results are in keeping with this (KL 2–3) [37, 48]. Moreover, it has been suggested that PRP injection efficacy has a correlation with age [37], PRP showing a greater efficacy in those under 50 [52]. This was, however, not something we could analyse.

Although adverse events were only reported in four (50%) of the included studies, no serious complications were recorded. Previous evidence indicates that all adverse effects tend to be non-specific, mild and self-limiting [34, 38, 53]. No joint infections were reported in the follow-up period of the included studies.

The strengths of this review include analysing only prospective trials using evidence-based medicine techniques and taking a systematic approach to study analysis using validated quality assessment tools. We previously submitted our protocol paper for publication (awaiting outcome of peer review) and all included studies had adequate follow up (often with multiple timeframes), with a low attrition rate, low risk of bias and relatively homogenous populations. The use of standard, validated outcome measurements (WOMAC, VAS and KOOS) across studies is ideal for direct comparison.

Studies varied in the lengths of time that patients had OA, as well as patient age. There were considerable differences in sex distribution between studies, with more females being included. High statistical heterogeneity among studies was also a noted limitation; however, similar reviews on PRP injections for knee OA report a similar observation [17, 33, 34]. This could be for several reasons. The preparation, centrifugation, leukocyte concentration and dosage of PRP may have considerable effect on its efficacy [54‐56], and even within the same patients, across multiple blood draws, the PRP yield may vary [57]. Methods of measuring platelet count also effect reporting accuracy [58]. Differences in the preparation of PRP has been shown to directly implicate its chondro- and osteo- inductive properties in vitro [59]. A systematic review by Chahla et al. [60] investigating PRP preparation protocols and composition found that only 10% of studies included a description of PRP preparation and 16% on its composition, concluding that this is sub-optimal for the comparison of PRP products and advocating for detailed and stepwise descriptions in all further publications. This is an area where further research and ultimately the development of standardised procedures are needed, with both Dhurat and Sukesh [55] and Chahla et al. [60] reporting advice and suggestions for what should be included in a standardised PRP protocol.