Hypertonic dextrose injections (prolotherapy) in the treatment of symptomatic knee osteoarthritis: A systematic review and meta-analysis

Hypertonic dextrose injections (prolotherapy) is an emerging treatment for symptomatic knee osteoarthritis (OA) but its efficacy is uncertain. We conducted a systematic review with meta-analysis to synthesize clinical evidence on the effect of prolotherapy for knee OA. Fifteen electronic databases were searched from their inception to September 2015. The primary outcome of interest was score change on the Western Ontario and McMaster Universities Arthritis Index (WOMAC). Three randomized controlled trials (RCTs) of moderate risk of bias and one quasi–randomized trial were included, with data from a total of 258 patients. In the meta-analysis of two eligible studies, prolotherapy is superior to exercise alone by a standardized mean difference (SMD) of 0.81 (95% CI: 0.18 to 1.45, p = 0.012), 0.78 (95% CI: 0.25 to 1.30, p = 0.001) and 0.62 (95% CI: 0.04 to 1.20, p = 0.035) on the WOMAC composite scale; and WOMAC function and pain subscale scores respectively. Moderate heterogeneity exists in all cases. Overall, prolotherapy conferred a positive and significant beneficial effect in the treatment of knee OA. Adequately powered, longer-term trials with uniform end points are needed to better elucidate the efficacy of prolotherapy.

load bearing ability [12][13][14][15] . Animal model data has also suggested cartilage-specific anabolic growth as a result of intra-articular dextrose injection 16 . The hypothesized mechanisms for pain relief include: (i) stimulation of local healing among chronically injured extra-and intra-articular tissue; (ii) reduction of joint instability through the strengthening of stretched or torn ligaments, and (iii) stimulation of cellular proliferation 17 .
Prolotherapy is practiced throughout the world; the strongest interest appears to be among physicians and patients in primary care 18,19 . Human studies have assessed the role of prolotherapy for various musculoskeletal conditions [20][21][22][23][24] . Systematic reviews of the clinical effectiveness of prolotherapy in chronic low back pain and lateral epicondylitis had been conducted in the past which yielded mixed and positive results respectively, though the strength of evidence was limited by clinical heterogeneity amongst studies and the presence of co-interventions 25,26 . In recent years, prolotherapy has been used to treat patients with knee OA refractory to other conservative care 27 . Results from several published clinical trials have shown positive effects of prolotherapy in knee OA but the findings have not been synthesized [28][29][30][31] . Therefore, we conducted a systematic review with the aim of more comprehensively assessing the efficacy of prolotherapy for knee OA in order to clarify its potential role as a non-surgical treatment modality.

Methods
We follow the PRISMA Reporting Guidelines for Systematic Review and Meta-analysis 32 : Search methods for identification of studies. Potential studies were identified by searching the electronic databases listed below, with search period starting from their inception till September 2015 (Appendix 1). Cochrane

Types of studies. This systematic review included randomized controlled trials (RCTs) and
quasi-randomized controlled trials that compared prolotherapy injections to saline injections, water injections or exercise therapy. Co-interventions were allowed, as long as they were uniform across all groups. There were no limitations on publication dates. The protocol was registered at PROSPERO (CRD42015015901).

Types of participants.
We selected studies that included participants aged 18 years or over with (i) a diagnosis of knee OA according to the criteria from the American College of Rheumatology 33 , (ii) knee pain for at least 3 months and (iii) had reported the Western Ontario and McMaster Universities Arthritis Index (WOMAC) 34 or the Visual Analogue Scales (VAS) for pain as one of the outcomes 35 . We excluded studies that included participants who had undergone total knee replacement.

Types of intervention.
For inclusion, dextrose prolotherapy injections had to be administered to at least one group within the trial. For comparison groups we included studies that provided injections with 0.9% normal saline or water; or exercise therapy. Consistent with the clinical practice of prolotherapy, at least part of the injection protocol had to include an intra-articular injection, with or without additional injections to the peri-articular ligament or tendon attachments.

Outcome measures.
In accordance with international consensus on the core set of outcome measures for phase III clinical trials in OA, eligible trials needed to include assessment of either self-reported pain or self-reported physical function 36 . Following this recommendation, the primary outcome of interest is the Western Ontario and McMaster Universities Arthritis Index (WOMAC) 37 . Secondary outcomes of pain include the Visual Analogue Scales (VAS) for pain 35 , Knee Pain Scale (KPS) 38 and Wong-Baker Scale 39 .
Eligibility Assessment and Data Extraction. Two reviewers (RS and VC) independently screened electronically retrieved titles and abstracts, evaluated potentially relevant full texts, and determined study eligibility. For eligible studies, data were extracted independently by the two authors (RS and DC) using a piloted data extraction form. For each eligible study, the following data were extracted: study design, clinical settings, participant characteristics, features of interventions, outcomes, duration of follow up and adverse events. An attempt was made to contact study authors regarding these methodological elements if not reported.
Risk of bias assessment. The risk of bias among included studies was assessed by the Cochrane's risk of bias tool 40 by two reviewers independently (XYW and RH). The following risk of bias domains were evaluated: sequence generation, allocation concealment, blinding of participants and research personnel, blinding of outcome assessors, incomplete outcome data, and selective outcome reporting. Discrepancies in study selection, data extraction and risk of bias assessment results were resolved by group consensus. In the group consensus process, the two reviewers discussed reasons for discrepancy with a goal to achieve consensus after clarification. If a consensus could not be reached, a third reviewer Vincent Chung (VC) was included as an arbitrator. A decision was made after reevaluation of the included studies and further discussion between the three reviewers. This has been clarified in the revised manuscript.
Statistical Analysis. All meta-analyses were conducted using the STATA software 41 . A random effect model was used to pool study results, taking into account possible variations in effect sizes across trials 42 . Changes in continuous outcomes were pooled as standardized mean differences (SMD), as different scaling of outcome measurements across trials were expected. 95% confidence intervals (CI) were calculated for all estimates. I square (I 2 ) statistic was calculated to estimate heterogeneity across studies. An I 2 level of less than < 25%, 25-50% and greater than 50% were regarded as indicators of low, moderate and high levels of heterogeneity respectively 43 . Minimal Clinically Important Difference (MCID) for WOMAC and VAS. The effect estimates were interpreted according to the values of the established Minimal Clinically Important Difference (MCID). It has been reported that a 12-point increment on WOMAC composite scale indicates 'good to very good' improvement when the scale is transformed onto a 0-100 point scale 44,45 .
The MCID for VAS global pain assessment score and VAS pain on motion were based on a study in 2005; of which the MCID for improvement in the global VAS is 18.3 mm whereas that for VAS in motion is 19.9 mm 44 .

Results
We identified 134 citations from all searches and excluded 14 duplicates. After screening the titles and abstracts, we retrieved 19 full texts for further assessment. Of these, 15 were excluded for the following reasons: duplicate publication as conference abstract (n = 1), publication not in English (n = 1), trial without a control arm (n = 1) and narrative reviews (n = 12). Four full texts which reported results from four clinical trials were eligible for inclusion [28][29][30][31] . (Fig. 1) Characteristics of included trials. Characteristics of included trials are summarized ( Table 1). The four included clinical trials 28-31 recruited a total of 258 patients with a diagnosis of knee OA based on American College of Rheumatology (ACR) guidelines including baseline radiological severity as graded by a Kellgren-Lawrence (K-L) score (Table 2) 46 . The follow up period ranged from five weeks to one year post-enrollment. The injection protocols varied; two studies referred to an existing published injection protocol 47 . (Table 3). None provided a detailed publicly accessible protocol of study methods including planned a priori statistical analysis methods (e.g. study protocols on trial registration websites such as ClinicalTrials.gov) that allowed assessment of selective outcome reporting, but three of them have provided a detailed analysis plan via personal communication [29][30][31] . Sequence generation was considered to have a low risk of bias in only two studies, and two had a low risk of bias in allocation concealment. Two studies blinded both patients and investigators, but two did not. Two studies reported blinding of assessors but not the remaining two. Risk of bias incurred from incomplete outcome data varied, with drop-out rates ranged from 0% to 27.8%. Since three out of four trials did not provide trial protocol, this can be regarded as a major source of bias. Table 4. Dumais et al. 28   For WOMAC pain (Fig. 4), pooled results indicated that the SMD is 0.62 (95% CI: 0.04 to 1.20, p = 0.035, I 2 = 46.2%). While the p value indicates statistically significant result as defined by p < 0.05, the lower end of the 95% CI is very close to zero, suggesting the imprecision of possible treatment effect. Overall, sizes of the SMD indicated that dextrose injection provides substantially stronger improvement when compared to exercise, but the lower end of 95% CI has covered SMD values of small effect sizes (i.e. SMD ≤ 0.2) 48 . Such uncertainty is also reflected in the low to moderate degree of heterogeneity among all three meta-analyses, given all with I 2 values ranging from 34.5% to 53.6% 49 . Although it is not included in meta-analysis due to differences in data collection time, Rabago et al. 22  Side effects. All four trials monitored side effects and adverse effects. Only one trial reported self-limited bruises after both dextrose (n = 3) and saline injections (n = 5) 30 . This was an expected side effect and deemed to be of minimal clinical relevance due to its transient nature

Discussion
Pooling data from two RCTs, we report that peri-and intra-articular hypertonic dextrose knee injections in three to five sessions have a statistically significant and clinically relevant effect in the improvement of WOMAC composite score, functional and pain subscale at 12 to 16 weeks compared to formal at-home exercise. Self-reported outcomes favoring prolotherapy are also observed in unpooled data when dextrose prolotherapy groups are compared with other control groups. The majority of the effect sizes are higher than the MCID, and the benefits were sustained up to 1 year. The risk of bias level in the included studies was moderate. Overall, prolotherapy injections appear to be safe, but no study was powered to detect rare adverse events. While the overall direction of the effect is positive, some uncertainty of the effect size still exists due to the low to moderate heterogeneity and wide confidence intervals. In addition, this systematic review identified  Table 2. Distribution of OA knee severity grading among participants of included trials.
several limitations of existing trials. None of the included studies provided a publicly accessible protocol, therefore the potential for selective outcome reporting exists; however, the analyses are generally straightforward and well-reported. The sample size was small with only four trials eligible for review and only portions of two trials could be pooled in meta-analyses. Small sample size led to a wide confidence interval; thus the effect sizes reported may be imprecise. However, pooled results provide guidance for the sample size needed in future clinical studies 50 . Due to a lack of uniform longer-term follow up data across both studies, pooling of results could only be done with data collected between 12 to 16 week follow-up. Given that prolotherapy is hypothesized to work by healing and regeneration over several months 51,52 , the effects reported here may underestimate longer term benefits. A recent open label study reported progressive improvement in WOMAC scores for most study participants through 2.5 ± 0.6 years (range 1.6-3.5 years) of follow up after prolotherapy 27 . Finally, low to moderate quantified heterogeneity was reported in pooled analyses; this is likely attributable to multiple factors, including differences in patient characteristics, control treatment, study design, injection protocol methods, dextrose concentration, follow-up duration and outcomes assessment methods. While the positive findings in the evaluated studies suggest efficacy, their limitations restrict a rigorous determination of efficacy and clinical utility. The current analysis suggests studies with the following characteristics can be considered in the future: 1) study duration of at least one year. Prolotherapy is hypothesized to be a regenerative injection therapy; serial monthly injections are typically performed to stimulate hypothesized tissue effects; 2) uniform self-reported outcomes and time points using the WOMAC outcome measures 37 ; 3) objectively determined physical function and health status outcomes such as hospitalization, nursing home placement and decline in health 53,54 . Inclusion of objective assessment including magnetic resonance imaging 31 and intra-articular and serum biomarkers 55 . 4) a priori subgroup analysis to identify the phenotype of patients who respond most favorably to prolotherapy;5) because the performance of prolotherapy appears to be superior to saline and exercise, future studies may consider formal effectiveness designs using exercise or other non-injection matched control comparators. This avoids injection-related discomfort and risk among controls, saves study conduct time and cost, and offers a comparison relevant to clinical practice; 6) overall quality of life and cost effectiveness evaluation using the EuroQol-5D, a commonly used measure that allows comparison to other interventions 56 ; 7) determination of the MCID of WOMAC score specific to prolotherapy.
Investigation of prolotherapy is in an early stage and several treatment protocol issues warrant further study to determine optimal strategy, including dextrose concentration and volume, treatment frequency and duration, and specific utility of intra compared with extra-articular injections. Studies in this review used the approach of dextrose injections into the intra-articular joint space and the extra-articular soft tissue attachment. Recently, more superficial application of hypertonic dextrose has been reported to provide an analgesic effect in the treatment of a variety of peripheral neuropathic pain conditions 57,58 . Sensorineural effects of dextrose have been hypothesized 31 . Given that the etiology of knee OA pain is multifactorial, future trials may consider adding superficial dextrose injection as part of the protocol. Concomitant basic science studies using in vitro and animal model methodologies are warranted to better elucidate the mechanism of action of dextrose in the context of musculoskeletal pathology. Finally, studies in this review used palpation guidance for injections; future investigators should also consider the use of ultrasound guidance to improve injection accuracy and precision, limiting procedural variability 59 . Strengths and weaknesses of this systematic review. The main limitation of this systematic review is the limited number of studies and their relatively small sample size. We considered whether or not to perform a meta-analysis in addition to systematic narrative review, given that the small sample sizes of included trials might generate an unstable pooled effect size 60 . A meta-analyses is included and resulted in increased statistical

Source
Sequence generation Allocation concealment Blinding of participants and researchers Blinding of outcome assessment Incomplete outcome data addressed Selective outcome reporting Reeves 21 Low. Random sequence was generated by random number table.
Unclear. Relevant information was not reported.

Low. Identical control solution was used
Low. Films were separated in different packets so reading 1 film would not influence reading of the next. X ray films were read by the chief investigator. A database coordinator loaded results onto the database.

Dumais 28
Unclear. Authors did not provide information on how random sequence was generated.
Low. Opaque sealed envelopes were used. High.
It was an open-labeled trial. High.
It was an open-labeled trial.
High. Lost to follow up was 3/21 = 14.3% in group A. 6/24 = 25% in group B. Both groups >10% Unclear. No protocol was provided. Rabago 22 Low. Random sequence was generated by computer.
Low. Allocation concealment of injectant was achieved through the use of sealed opaque envelopes (personal communication, Rabago 2015) Low. Both active and control solutions looked similar.
Low. The injector, outcome assessor, principal investigator, and participants were blinded to injection group by preparation of syringes off site; blinding was formally assessed among injection participants and injector using questionnaire.
Low. No lost to follow-up cases.
Low. No protocol was provided,but analysis plan was clearly described. Rabago 30 High. Non-randomized study design.
High. Non-randomized study design.
High. Single-arm uncontrolled study/ at-home exercise "control" was used.
High. Primary outcome was self-reported QOL.
High. Lost to follow up was 3/27 = 11.1% in intervention group. 5/18 = 27.8% in control group. Both >10% Unclear. No protocol was provided. rigor by increasing the overall sample size, providing a more precise estimate 61 . In a narrative review, evaluation of efficacy in RCTs is based primarily on the p-values of individual studies, a so-called 'vote counting' approach, and the risk of arriving at a biased conclusion may even be higher 62 . Because most of the results in the included trials were positive, readers may in the case of narrative review be tempted to draw an over-confident conclusion 63 . A meta-analysis, with calculation of I 2 value, is able to express current best evidence and heterogeneity among studies quantitatively 64 . Specifically, we determined that the I 2 value of WOMAC composite score is 53.6%, indicating considerable heterogeneity on the effect estimate, a finding that can be communicated only through meta-analysis 65 . Strengths include the conduct of a comprehensive literature search, duplicate study selection and data extraction by two independent reviewers, and appraising risk of bias of all included studies using a validated tool independently by two reviewers.

Conclusion
The results of this systematic review indicate that hypertonic dextrose prolotherapy conferred a positive, significant beneficial effect meeting criteria for clinical relevance in the treatment of knee OA, compared with saline injection and exercise. However, moderate heterogeneity existed among these trial results. Larger, long-term trials with uniform outcomes and high methodological standards are needed for more a more comprehensive assessment of the overall treatment effect of prolotherapy.