Introduction

Osteoarthritis (OA) is a multifactorial chronic bone and joint disease characterized by articular cartilage degeneration that negatively affects patient mobility and quality of life [1]. It is approximately 20% of patients who were older than 45 years occur OA, which makes it the most common chronic painful condition [2].

Besides the symptoms in knee pain, swelling, and limited mobility, OA can also result in a high prevalence of functional disability. The targets of OA treatment are to relieve pain, improve function and mobility, prevent deformity, and slow the progression of the disease. Currently, there are various of non-surgical treatment modalities that have been applied to treat knee or hip OA, including oral nonsteroidal anti-inflammatory drugs [3], physical therapy, and intra-articular injection of hyaluronic acid (HA) [4], corticosteroids [5], or platelet-rich plasma (PRP) [6].

PRP, representing a potential biological treatment for knee OA, has been widely used for articular cartilage regeneration [7, 8]. PRP is an autologous blood product that contains high concentrations of a wide range of growth factors that play critical roles in tissue repair [9, 10]. Because of its autologous nature, the PRP treatment avoids any immune reaction or blood transmission disease. Although several published trials reported the promising results of PRP in orthopedic and sports medicine, its clinical application and efficacy still remained uncertain. PRP lacks proper standardization for the number or frequency of injections, as well as the ideal treatment for different stages of gonarthrosis. Several clinical trials reported the favorable results of PRP injection in cartilage damage and knee OA when compared with HA [11,12,13] and placebo injection [14]. Moreover, recently published systematic reviews and meta-analysis [15,16,17,18,19] also have been carried out to investigate the effects of PRP for knee OA; however, these studies yielded conflicting results [18, 19]. For example, the meta-analysis conducted by Kanchanatawan, W et al. [18] concluded that there was not sufficient evidence to support the effect of PRP in improving the Western Ontario and McMaster (WOMAC) pain, stiffness, and function scores in the treatment of knee OA when compared with HA or placebo. In contrast, another two meta-analyses by Shen, L et al. [20] and Han, Y et al. [19] reported significant beneficial effects of PRP in the WOMAC pain and physical function subscores when compared with HA, saline placebo, ozone, or corticosteroids.

Although these reviews were performed mainly based on the same body of evidence, a first screening suggested that the conflicting findings could be resulted from inconsistent inclusion criteria, differences in risk of bias assessment of included studies, and also errors in data extraction and synthesis. Moreover, both the administration of PRP and comparators have varied greatly among the included studies in previous systematic review and meta-analysis [18,19,20], such that in some, PRP was compared with HA only [19, 21], whereas in others [17, 18, 20], PRP was compared with saline placebo, HA, ozone, and corticosteroids. By design, these two types of comparisons could address different questions: (1) the efficacy of PRP as compared with HA; (2) the efficacy of PRP as compared with other intra-articular injection regimens. The previous systematic reviews and meta-analysis have not consistently accounted for these fundamental differences, and the treatment estimate may be biasedly analyzed. Moreover, the sources of heterogeneity (age, sex, BMI, and grade of OA) were also not explored in these reviews.

Prompted by these issues outlined above (discrepancies and insufficient methodological quality), as well as the additional RCTs have been published recently in this field, we believe that it is necessary to perform an updated meta-analysis to investigate whether PRP injections are more efficacious than other injections in pain relief and functional improvement for the treatment of patients with knee or hip OA.

Material and methods

Research design

This meta-analysis was conducted according to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines [22]. We searched medical bibliographic databases to identify articles focusing on the effects of PRP for the treatment of knee or hip OA.

Study search

A comprehensive systematic search was performed in major electronic databases (PubMed, Embase, and Web of Science) from their inception to June 2019. The following search terms were used for searching: (“platelet-rich plasma”[MeSH Terms] OR (“platelet-rich”[All Fields] AND “plasma”[All Fields]) OR “platelet-rich plasma”[All Fields] OR (“platelet”[All Fields] AND “rich”[All Fields] AND “plasma”[All Fields]) OR “platelet-rich plasma”[All Fields]) AND ((“osteoarthritis, hip”[MeSH Terms] OR (“osteoarthritis”[All Fields] AND “hip”[All Fields]) OR “hip osteoarthritis”[All Fields] OR (“hip”[All Fields] AND “osteoarthritis”[All Fields])) OR (“osteoarthritis, knee”[MeSH Terms] OR (“osteoarthritis”[All Fields] AND “knee”[All Fields]) OR “knee osteoarthritis”[All Fields] OR (“knee”[All Fields] AND “osteoarthritis”[All Fields]))). There was no language or publication status restriction. Moreover, the reference lists of reviews and all included studies were manually reviewed to identify other potential articles. We also contact the corresponding authors by email to retrieve original data when important data were not provided in the study.

Inclusion criteria and study selection

Inclusion criteria for this study were as follows: (1) randomized controlled trial (RCT); (2) adult patients had a diagnosis of knee or hip OA; (3) compared the effect of PRP with other injections; (4) provided the outcome measures, including WOMAC, Knee Injury and Osteoarthritis Outcome Score (KOOS), Visual Analog Scale (VAS), Harris Hip Score (HHS), and International Knee Documentation Committee (IKDC). Exclusion criteria were a cohort study, case-control, or cross-sectional study; review articles, or conference abstracts; or studies did not provide the outcomes of our interest. The disagreement between investigators was resolved by discussion and consensus.

Data extraction and quality assessment

A standardized Excel form was developed to extract the data. All the data were extracted by two independent investigators and then checked by a third investigator to ensure the accuracy of data. The data of outcome measures, including a mean and standardized deviation of WOMAC, KOOS, VAS, HHS, IKDC, were extracted. The baseline characteristics from each study, such as the first author’s name, year of publication, country, sample size, and duration of follow-up, were also extracted. When several publications from the same population or clinical trial were identified, we only included the one that had the most complete data or the latest outcomes.

We used the method recommended by Cochrane Collaboration [23] to assess the risk of bias in an RCT. This method comprises six items, including random sequence generation, allocation concealment; blinding of outcome participants and personnel, blinding of outcome assessment, incomplete outcome data, selective reporting, and other bias [23]. In accordance with the quality domains and scoring system, each RCT was classified as being “high” (seriously weakens confidence in results), “low” (unlikely to seriously alter the results), or “unclear” risk of bias [23].

Statistical analysis

Statistical analysis was carried out using the STATA software version 12.0 (Stata Corporation, College Station, TX, USA). Continuous variables were expressed as weight mean difference (WMD) with 95% confidence intervals (CIs), while dichotomous variables were pooled as risk ratio (RR) with 95%CIs.

Statistical heterogeneity between the studies was assessed with Cochrane Q and I2 statistic [24], in which P < 0.1 or I2 > 50% were considered to be significant. Depending on the absence or presence of significant heterogeneity, pooled estimates were calculated using a random-effects model [25] or a fixed-effects model [26]. When significant heterogeneity was identified, sensitivity analysis was performed to explore the potential sources of heterogeneity by excluding studies one by one. Subgroup analysis based on disease type, comparators, and treatment duration was performed. Given several confounding factors (mean age, body mass index (BMI), male ratio, grade of OA, history of previous treatment, and symptom duration) that might affect the differences in outcome measures among included studies, we performed meta-regression analysis when the data were available. Publication bias was evaluated by using the Egger [27] and Begger [28] test. Finally, in order to ascertain the robustness of summary treatment effects, we performed meta-analysis confined to RCTs that were at low risk of bias. A P value less than 0.05 was judged as statistically significant, except where otherwise specified.

Results

Search results

A total of 692 publications were identified in the initial search, of which 374 were excluded due to duplications, leaving 318 unique articles for the title/abstract review. Two hundred ninety articles were removed after screening the title and abstract. Then, 28 potential articles were left for full-text information review. Based on the inclusion criteria, 4 of them were removed because of the following reasons: two studies were single-arm design [29, 30], one study did not provide data of our interest [31], and one study did not present available data for analysis [6]. Finally, 24 RCTs [11,12,13,14, 32,33,34,35,36,37,38,39,40,41,42,43,44,45,46,47,48,49,50,51] met the inclusion criteria, and were included in this meta-analysis (Fig. 1).

Fig. 1
figure 1

Eligibility of studies for inclusion in meta-analysis

Full size image

Characteristics of included studies

Table 1 describes the main characteristics of the 24 included RCTs. The majority of studies were from Italy (n = 7) [11, 12, 35, 37, 38, 40, 48], with the remaining 3 articles from Spain [34, 41, 46] and Iran [13, 33, 43], 2 articles from China [49, 50], Turkey [32, 39], and USA [44, 47], and 1 article from Australia [36], Brazil [42], India [14], Egypt [45], and Mexico [51]. In total, 42.7% of the study population was male, and 57.3% was female. Among the included studies, 21 articles included knee OA patients [11,12,13,14, 32,33,34, 36, 39,40,41,42,43,44,45,46,47,48,49,50,51], and 3 included hip OA patients [35, 37, 38]. The sample size varied greatly, which ranged from 21 to 183 in each study. WOMAC was the most commonly used outcome in 17 of the included studies [11, 13, 14, 33,34,35, 38, 39, 41, 42, 44, 46,47,48,49,50,51]. The duration of follow-up was also variable across the trials, with the shortest period of 12 weeks [36] and the longest period of 18 months [49]. PRP protocols used in each trial were greatly different in terms of preparation devices, centrifugations, and the injection regimen of dose and intervals. HA was the most commonly used control in most of the included studies [11,12,13, 32, 34,35,36,37,38,39,40,41,42, 45, 47,48,49], and other comparators in the remaining studies included normal saline [14, 44, 50], acetaminophen [51], and prolotherapy [33].

Table 1 Baseline characteristics of patients in the trials included in the meta-analysis
Full size table

Two different radiographic OA grading systems were applied in the studies, including 20 studies using Kellgren Lawrence grading (0–IV) [11,12,13, 32, 33, 35,36,37,38,39,40,41,42, 44,45,46,47, 49, 51] and four studies using Ahlback scale(I–V) [14, 34, 48, 50]. In studies with the Kellgren Lawrence classification system, 14.74% of patients had grade I, 47.92% of patients had grade II, 33.89% of patients had grade III, and 3.45% of patients had grade IV, whereas in these studies using Ahlback scale, 60.26% of patients had grade I, 31.09% of patients had grade II, and 8.65% of patients had grade III, and none of the patients had grade IV.

Quality appraisal of included studies

The risk of bias assessment of RCTs is presented in Fig. 2. Overall, ten studies were classified as being at low risk of bias [12, 32, 34, 36, 37, 41, 44, 46,47,48], five studies as being at unclear risk of bias [35, 39, 40, 49, 51], and the remaining nine studies as being at a high risk of bias [11, 13, 14, 33, 38, 42, 43, 45, 50]. The most common reasons for studies with unclear or high risk of bias were that they did not perform the blind for participants, personnel, or the outcome evaluators.

Fig. 2
figure 2

Risk of bias

Full size image

Effects of PRP: comparison with previous meta-analysis and systematic reviews

Shen, LX et al. [20], Kanchanatawan, W et al. [18], and Dold, AP et al. [52] reported that PRP was more effective in the treatment of knee OA in pain relief and functional improvement as compared with other injections. However, mistakes were detected in the study inclusion and data extraction in these systematic reviews and meta-analysis. The inclusion criteria in the study design for these reviews were RCTs. However, these reviews included a prospective cohort study by Spakova, T et al. [53] rather than an RCT. Moreover, the effect evaluation of PRP compared with HA was also problematic in the study by Spakova, T et al. [53]. The authors used WOMAC and 11-point pain intensity Numeric Rating Scale to evaluate the effects of PRP and HA at baseline, 3, and 6 months after therapy. However, the baseline measures were not similar between the two groups. Thus, it should be more accurate to compare the changes overall time rather than at the follow-up periods in the outcome measures in both groups. The errors in study inclusion and data extraction may slightly alter Shen, LX [20] and Kanchanatawan, W’s [18] conclusion on the effects of PRP on the WOMAC score.

Our study differed from the previous systematic reviews [54, 55], in that, for the same RCTs, more domains were scored with unclear or high risk of bias. The details of the methodological assessment are presented in Fig. 2, whereas none of the previous systematic reviews provided such details, which prevented us from exploring the exact reasons behind the discrepancies. Of note, Di, YL and colleagues [16] assessed two included studies with unclear risk of bias in the blinding of outcome assessment and incomplete outcome, whereas we identified these domains with a high risk of bias. Bennell, KL [17] and Sundaram, K [55] even did not assess the risk of bias. Thus, we could not understand the quality of the included studies in their reviews, as well as whether the evidence of PRP effects was reliable.

WOMAC

Total WOMAC score

For the total WOMAC score, there were 14 studies [11, 13, 38, 39, 41, 14, 44, 51, 33, 46,47,48,49,50] comparing the differences between PRP and comparator groups. Pooled estimate using a random-effects model (I2 = 97.2%, P < 0.001) showed that PRP was associated with a significantly lower WOMAC score than comparators (WMD = − 10.39, 95%CI − 13.50, − 7.28; P < 0.001) (Fig. 3). We performed a sensitivity analysis by removing the trial with outlier [38], and results showed that the pooled estimate of remaining studies changed a little (WMD = − 12.28, 95%CI − 15.46, − 9.09; P < 0.001), but significant heterogeneity was still present (I2 = 98.5%, P < 0.001). Further exclusion of any single trial did not change the pooled data and heterogeneity substantially (data not shown).

Fig. 3
figure 3

Forest plot showing the effect of PRP on the total WOMAC score

Full size image

We performed subgroup analysis based on disease type, comparator, treatment duration, and methodological quality. These results are presented in Table 2.

Table 2 Subgroup analysis results of the outcomes based on disease type, comparator, treatment duration, and methodological quality
Full size table

WOMAC pain score

There were 14 studies [13, 14, 33,34,35, 39, 41, 42, 44, 46, 47, 49,50,51] reporting the data of WOMAC pain score. The results showed that the WOMAC pain score was significantly lower in the PRP group than in the control group (WMD = − 2.72, 95%CI − 3.32, − 2.12; P < 0.001). There was significant heterogeneity among the included studies (I2 = 97.9%, P < 0.001). Thus, we performed a sensitivity analysis by excluding the trial with an outlier. However, the heterogeneity was still present (I2 = 98.0%, P < 0.001), and the pooled data did not change a lot (WMD = − 2.85, 95%CI − 3.45, − 2.26; P < 0.001).

Subgroup analysis results based on disease type, comparator, treatment duration, and methodological quality are presented in Table 2.

WOMAC stiffness score

There were 13 studies [13, 14, 33,34,35, 39, 41, 42, 44, 46, 49,50,51] that reported the data of WOMAC stiffness score. Compared with control, PRP significantly reduced the WOMAC stiffness score (WMD = − 0.91, 95%CI − 1.23, − 0.60; P < 0.001). The test for heterogeneity was significant (I2 = 93.8%, P < 0.001). Thus, a sensitivity analysis was performed. When we deleted the trial with outlier, the overall estimate changed a little (WMD = − 0.92, 95%CI − 1.24, − 0.61; P < 0.001), but the heterogeneity was still present (I2 = 94.1%, P < 0.001).

Subgroup analysis results based on disease type, treatment duration, and methodological quality were presented in Table 2.

WOMAC physical function score

There were 12 studies [13, 14, 33, 34, 39, 41, 42, 44, 46, 49,50,51] reporting the data of WOMAC physical function score. Pooled data showed that PRP had a significantly decreased score in WOMAC physical function compared with control (WMD = − 9.34, 95%CI − 11.49, − 7.20; P < 0.001).

Subgroup analysis results based on treatment duration and methodological quality are presented in Table 2.

VAS

There were 12 studies [14, 33, 35,36,37,38,39, 42, 45, 48, 49, 51] reporting the data of VAS. The pooled estimate showed that the VAS score was significantly lower in the PRP group than in the control group (WMD = − 0.86, 95%CI − 1.20, − 0.52; P < 0.001) (Fig. 4).

Fig. 4
figure 4

Forest plot showing the effect of PRP on the VAS score

Full size image

Subgroup analysis results based on disease type, treatment duration, and methodological quality are presented in Table 2.

KOOS

KOOS symptoms

For KOOS symptoms, there were four studies [12, 36, 40, 43] comparing the difference between PRP with control. The pooled estimate showed that PRP had a similar KOOS symptom score with control (WMD = 1.53, 95%CI − 2.79, 5.85; P = 0.487) (Fig. 5). Sensitivity analysis was performed by excluding the trial with outlier [36], the overall estimate changed a little (WMD = 3.11, 95%CI − 0.86, 7.09; P = 0.125), but the heterogeneity was still present (I2 = 86.5%, P < 0.001).

Fig. 5
figure 5

Forest plot showing the effect of PRP on KOOS symptom score

Full size image

Subgroup analysis results based on treatment duration and methodological quality are presented in Table 2.

KOOS pain

There were four studies [12, 36, 40, 43] reporting the data of KOOS pain between PRP and control groups. The pooled estimates using the random-effects model showed that PRP was associated with a similar KOOS pain score with other injections (WMD = 2.30, 95%CI − 0.54, 5.14; P = 0.113). Sensitivity analysis was performed by excluding the trial by Paterson, KL et al. [36] with outlier data, and the result suggested that PRP had a significantly higher KOOS pain score than control (WMD = 3.27, 95%CI 0.73, 5.81; P = 0.012). However, the heterogeneity was still present (I2 = 62.8%, P = 0.006).

KOOS function

KOOS function was reported in four studies [12, 36, 40, 43]. Pooled analysis using the random-effects model showed that there was no significant difference between PRP and control in terms of KOOS function score (WMD = 2.84, 95%CI − 1.16, 6.85; P = 0.164). When we excluded the trial with outlier [36], the pooled result of remaining studies altered a lot (WMD = 4.15, 95%CI 0.41, 7.89; P = 0.030), but the heterogeneity was still present (I2 = 82.8%, P < 0.001).

KOOS sport

There were four studies [12, 36, 40, 43] reporting the data of KOOS sport. The pooled estimates showed that patients treated with PRP had a similar KOOS sport score compared with those with other injections (WMD = − 0.73, 95%CI − 3.95, 2.48; P = 0.655). Sensitivity analysis by excluding the trial with outlier showed that the summarized estimate did not alter substantially (WMD = − 0.54, 95%CI − 3.51, 2.44; P = 0.723), but the heterogeneity changed a little (I2 = 45.8%, P = 0.064).

KOOS quality of life

There were four studies [12, 36, 40, 43] reporting the data of KOOS QoL. Meta-analysis using a fixed-effects model showed that PRP had a significantly higher KOOS QoL score than other injections (WMD = 4.32, 95%CI 3.35, 5.29; P < 0.001).

IKDC

Four studies reported the data of IKDC [12, 32, 40, 45]. The aggregated data indicated that the IKDC score was significantly improved in the PRP group than in the control group (WMD = 6.95, 95%CI 2.15, 11.74; P = 0.005). The test for heterogeneity was significant (I2 = 93.3%, P < 0.001).

HHS

Two studies [37, 38] reported the data of HHS. Pooled data showed that there was a significant improvement in HHS score in the PRP group than in the control group (WMD = 6.63, 95%CI 5.21, 8.05; P < 0.001). The test for heterogeneity was significant (I2 = 83.6%, P < 0.001).

Meta-regression

To further evaluate the influence of potential factors on the total WOMAC score, some meta-regression analyses were performed. Our results suggested that age (t = 5.32, P < 0.001) (Supplementary figure 1) had a significant impact on the difference in total WOMAC score between PRP and controls, whereas male sex (t = − 0.87, P = 0.392), BMI (t = − 1.07, P = 0.294), and grade of OA (t = − 0.19, P = 0.854) did not (Supplementary figure 2, 3, 4).

Publication bias

Publication bias was assessed using Begg and Egger’s test, in which a P value < 0.05 demonstrated the existence of potential publication bias. As shown by our results, the P values of Begg and Egger’s test for all meta-analysis were greater than 0.05, indicating the absence of publication bias among the included studies.

Discussion

PRP has been widely used in several medical fields, such as dentistry, dermatology, and ophthalmology [17]. However, in recent years, there has been an increased use of PRP for the treatment of OA. This present meta-analysis collected the evidence of RCTs to investigate the effects of intra-articular injection of PRP in the treatment of knee and hip OA. The summarized data revealed that PRP significantly improved the WOMAC score, VAS score, IKDC score, and HHS score as compared with comparators. This beneficial effect was only observed at short-term follow-up (1, 2, 3, 6, 12 months), but not at long-term follow-up. Moreover, these positive results were also confirmed in studies with a low risk of bias. However, no significant benefit of PRP was observed over the comparators in terms of the KOOS score.

There are a few numbers of systematic review and meta-analysis [17, 18, 20, 56, 57] that have been published with the purpose to investigate the effects of PRP in OA, and their results demonstrated the treatment effect of PRP in improving the functional recovery and pain relief in OA patients. However, these results were obtained based on studies with limited statistical power, methodological errors (inclusion criteria and data extraction), or low quality of evidence. Chang et al. [56] compared the different outcome measurements between PRP and HA for knee OA. However, half of the included studies for analysis were case series, and only 5 of them were RCTs [56]. Laudy et al. [57] pooled ten randomized or non-randomized controlled trials to investigate the effects of PRP on pain relief and functional recovery. Nonetheless, most of their results were obtained based on only 1 or 2 studies because of the small number of RCTs used for data analysis. Another meta-analysis of Kanchanatawan et al. [18] included 9 RCTs to compare the effects of PRP with placebo or HA in knee OA patients. Their results demonstrated better outcomes of PRP in some clinical measures when compared with HA or placebo [18]. This conclusion was limited by the small sample size and medium to high risk of bias of included studies. Shen et al. [20] selected a greater number of RCTs (n = 14) for data analysis in patients with knee OA, and their results showed the superior effect of PRP over other intra-articular injections in terms of knee pain and physical function at 3, 6, and 12 months post-injection [20]. However, none of the included RCTs was regarded as being low risk of bias. Blinding to participants, which is crucial controlling placebo effects, has not been successfully performed in around half the included trials. This is more inclined to result in overestimated effects of PRP. Again, the positive findings might not be reliable since the pooled estimate was calculated based on RCTs with low-quality evidence, and one of the included studies did not meet the inclusion criteria and provided inappropriate method for the measure estimates between PRP and HA.

To address the shortcomings of the previously published meta-analysis, we not only included more recently published RCTs regarding the effect of PRP for knee or hip OA but also performed subgroup analysis based on methodological quality to provide the most rigorous and complete evidence. In addition, in order to explore the potential sources of heterogeneity, which was also observed in the previous meta-analysis, we performed sensitivity analysis and meta-regression analysis. The results revealed that age might have an impact on the difference in the total WOMAC score between PRP and comparators. Another important highlight of our study relates to assessing the long-term effect of PRP in OA patients by comparing it with other intra-injections, which has not been investigated in the previous meta-analysis.

In the present study, we evaluated the effects of PRP in the knee or hip OA patients. Results from RCTs with the knee OA supported the beneficial effect of PRP over other intra-injections, which were consistent with the previous meta-analysis [8, 15, 56,57,58] whereas in the hip OA, our results only indicated the better effect of PRP in the outcomes of total WOMAC score, but not in some other outcome measurements (such as KOOS score, VAS score). From the studies reporting effects of PRP in hip OA, we found that their results have been conflicting. Dallari et al. [38] reported the significant clinical improvement of PRP in VAS and total WOMAC score at 2 and 6 months, but not at 12 months, as compared with HA. However, in another two RCTs [35, 37] investigating the effects of PRP in hip OA, the authors failed to find the differences between PRP and HA, although functional improvement and pain reduction in PRP was observed, but that was not superior to HA. This might be explained by the limited statistical power owing to the small sample size in the studies [35, 37].

Concerning the duration period of the favorable effect of PRP, it remains uncertain in the previous meta-analysis. Our results demonstrated that PRP was more effective than other intra-articular injections in improving physical function and reducing pain through 1 to 12 months, but not at 18 months. Filardo et al. [59] performed a clinical trial in knee OA to investigate the persistent effect of intra-injection PRP at the 24 months. The results revealed that all the evaluated parameters (including IKDC, EQ-VAS) were significantly lower at the 24 months as compared with that at 12 months, but were still higher than the baseline level [59]. The authors concluded that the median duration of the beneficial effect for PRP was 9 months [59]. The short-term effect of intra-injection PRP suggests that PRP might have a temporary impact on the joint milieu, but have no effects on the joint structure or progression of knee OA.

Although several innovations have been performed in this meta-analysis, the data analyses for some evaluated outcomes are limited by several biases. First, despite no restriction on the language was imposed for the study selection, all the included RCTs were English-language, which might increase the risk of selection bias. Second, there was substantial heterogeneity identified among the included studies; thus, we performed sensitivity analysis and met-regression analysis to explore the potential sources. However, meta-regression analysis was carried out only based on the characteristics of patient population (such as age, gender, BMI, and grade of OA). Some other clinical variables, such as outcome measures and PRP regimens, which might account for the heterogeneity and influence the pooled estimate, were not included for the meta-regression analysis because of the insufficient data across included trials.

In conclusion, intra-articular PRP injection provided better effects than other intra-articular injections for OA patients, especially in those with knee OA, in terms of pain reduction and function improvement at the short-term follow-up (1, 2, 3, 6, 12 months). Further larger scale, double-blinded RCTs are needed to assess the effects of PRP in patients hip OA.