Types of studies

Randomised and quasi‐randomised (method of allocating participants to a treatment which is not strictly random; e.g. by date of birth, hospital record number, alternation) controlled clinical studies evaluating any type of platelet‐rich therapy in the management of long bone fractures, non‐unions and bone defects.

Types of participants

Adults (i.e. over the age of 18 years) who have a long bone featuring one of the following:

• acute complete fracture

• stress fracture

• osteotomy

• delayed union

• non‐union

• critical‐size defect

Types of interventions

Trials of any type of platelet‐rich therapy will be included provided this treatment is compared with either no additional treatment or a placebo. Platelet‐rich therapy can be the only treatment but we anticipate that it will be delivered in addition to a standard‐of‐care treatment applied to all trial participants. The standard‐of‐care treatment can be operative or non‐operative.

Types of outcome measures

Functional recovery, including return to former activities and/or work, will be the focus of the review. However, we anticipate that most trials will not report patient‐reported functional outcome measures but will focus instead on bone healing outcomes.

The definition of a healed bone is contentious. For the purpose of this review we will adopt the widely accepted definitions in the literature. A bone discontinuity is healed when callus is present bridging three of four cortices on orthogonal radiographs or there is a reduction in pain and an absence of movement at the fracture site or both these. It is expected that most studies will report the time to union for each participant. These are the most frequently reported statistics when studies are published in this field. However, it is possible that some studies may present a proportional analysis of healed bones at a number of fixed time points after treatment. If required, we will make a pragmatic choice to alter the primary outcome to the proportion of bony discontinuities united following review of the included studies.

Electronic searches

We will search the Cochrane Bone, Joint and Muscle Trauma Group Specialised Register (to present), the Cochrane Central Register of Controlled Trials (The Cochrane Library, current issue), MEDLINE (1950 to present) and EMBASE (1980 to present). There will be no constraints based on language or publication status.

In MEDLINE (Ovid Online) we will combine a subject‐specific search with the Cochrane Highly Sensitive Search Strategy for identifying randomised trials: sensitivity‐maximising version (Lefebvre 2011). In EMBASE, the subject‐specific search will be combined with the SIGN strategy for randomised controlled trials. SeeAppendix 1 for all strategies.

Current Controlled Trials and the WHO International Clinical Trials Registry Platform will be searched for ongoing and recently completed trials.

Searching other resources

We will search reference lists of articles retrieved from the electronic search. We will contact experts in the field for any additional or unpublished articles.

Selection of studies

Two authors (XG and DW) will independently select the studies for inclusion in the review based upon the criteria defined above. Initially the titles and abstracts of all the retrieved studies will be reviewed to determine potential eligibility. The full text of each study in this shortlist will then be reviewed to determine study eligibility. Any disagreement between the authors will be settled by consensus between all authors involved in the review.

Data extraction and management

Two authors (XG and DW) will independently extract data from the included studies using a pre‐piloted version of the Cochrane Bone, Joint and Muscle Trauma Group's data extraction form. Extracted data from the studies will be managed and collated by the review statistician (NP), who will be independent from the study selection decisions.

Assessment of risk of bias in included studies

Each study included in the review will be independently assessed for the risk of bias, by two authors (XG and DW), using The Cochrane Collaboration's 'Risk of bias' tool (Higgins 2011a). The assessment tool incorporates the assessment of random sequence generation, allocation concealment, blinding of participants and personnel, blinding of outcome assessment, incomplete outcome data, selective outcome reporting and other sources of bias. Other sources of bias will include selection bias, where we will assess the risk of bias from imbalances in key baseline characteristics (e.g. age, sex and smoking behaviour); we will also assess the potential influence of commercial funding. Where there is disagreement between the two authors, consensus will be reached through discussion between all the review authors.

Different considerations apply to the primary outcome of bone healing, which is variably defined in the literature. We anticipate that studies may define healing clinically and radiographically. We anticipate that bias might be introduced by inter and intra‐observer error in the reading of radiographs. We will assess this risk of bias and ascribe a low risk to studies in which a blinded panel of specialist radiologists or orthopaedic surgeons read the radiographs. Studies employing other methodologies, such as multiple independent observers, will be ascribed a high risk of bias.

Measures of treatment effect

We plan to assess time to bone union after treatment using a (log) hazard ratio and 95% confidence intervals. If studies report mean time to union we will calculate standardised mean differences and 95% confidence intervals. This is preferable since it is likely that widely differing mean times to union will be reported in different studies including different bones and types of discontinuity. Risk ratios with 95% confidence intervals will be used to express the intervention effect for dichotomous outcomes. For continuous data, such as pain scores, we will calculate mean differences or, where studies have used different measurement tools, standardised mean differences; both with 95% confidence intervals.

Unit of analysis issues

We intend to analyse the extracted data by participant. However, where studies report data from multiple fractures from a single participant, we anticipate that the reported data may be inadequate to perform an analysis by participant. We do not consider these multiple fracture data from the same individual to be independent. However, in the event that the study reporting is not adequate, or the study authors cannot provide additional information, to identify multiple data points from a single participant, we will analyse the data by fracture. A record will be made in the Characteristics of included studies table for those studies where such an issue is apparent.

It is expected that most studies will report functional outcome scores at a number of follow‐up times; for example, at six and twelve weeks. Dependent on the nature of reporting, separate analyses will be made at each of the commonly reported occasions; representing perhaps, short, medium and long‐term follow‐up. It is expected that all studies will report simple parallel group designs. However, if other designs are reported (e.g. cluster randomised designs), generic inverse variance methods will be used to combine data where appropriate.

Dealing with missing data

Additional information will be sought from the authors of the included studies where the published information or data are incomplete. Where standard deviations are not specifically reported, we will attempt to determine these, if available, from standard errors, confidence intervals or exact P values. Where small amounts of data may be missing for proportional outcomes that cannot be reliably determined from the authors, these outcomes will initially be classed as treatment failures. For continuous outcomes, a conservative estimate of any treatment effect will be made by attributing outcome in the treatment group at the extreme of the distribution. Pooled effect sizes will be presented with and without these adjustments to check the effect of these assumptions.

Assessment of heterogeneity

The degree of statistical heterogeneity between studies will be first assessed visually from inspection of the forest plot and more formally using the Chi² test and I² statistic (Higgins 2003). A conservative P‐value for Chi² of < 0.1 will be set to indicate significant heterogeneity between studies. Values of I² will be interpreted as follows: 0% to 40% might not be important; 30% to 60% may represent moderate heterogeneity; 50% to 90% may represent substantial heterogeneity; 75% to 100% considerable heterogeneity. If the heterogeneity statistics indicate significant heterogeneity and one or more studies appear to be clear outliers, then the data for these studies will be checked carefully for errors or other methodological reasons why they might differ from the other studies. We plan that if good reason is found why the studies differ from the majority then this will be noted and reported, and the studies removed from the main meta‐analyses; however, all analyses will be performed with and without outlier studies in the event that any are excluded (sensitivity analysis).

Assessment of reporting biases

The search strategy described will attempt to reduce the risk of reporting bias in the inclusion of studies in this review. We plan to complete a funnel plot if a sufficient number of studies (more than 10) are available.

Data synthesis

The primary analysis for these studies is likely to be a time‐to‐event (survival) analysis using the time to bone union as the outcome measure. Therefore, it seems likely that the majority of studies will report either log‐rank statistics or estimates of hazard ratios, after fitting Cox's proportional hazards regression model, as an estimate of the intervention effect. We plan that in the event that the majority of studies reported proportional union analyses, risk ratios will be used to express the intervention effect and the fixed‐effect Mantel‐Haenszel method used to combine data. Where studies report mean time‐to‐event data, the standardised mean difference will be used to assess the treatment effect and generic inverse variance methods will be used to combine data. Similarly, for any continuous outcome measures, where the standardised mean difference is used to assess the treatment effect, generic inverse variance methods will be used to combine data. Confidence intervals will be reported at the 95% level and initially the fixed‐effect model will be used for meta‐analysis. Where there is substantial unexplained heterogeneity (I² > 50%), we will use the random‐effects model. If there is considerable unexplained heterogeneity (I² > 75%), we will not perform meta‐analysis. Instead we will perform a narrative synthesis of the included studies.

Subgroup analysis and investigation of heterogeneity

We plan to conduct subgroup analysis to explore possible sources of heterogeneity if significant heterogeneity is present. Four possible subgroup analyses are identified a priori:

1. Upper versus lower limb non‐union. This is a pragmatic proxy for weight bearing versus non‐weight bearing bones.

2. Smokers versus non‐smokers.

3. Co‐interventions. Possible analyses might include operative versus non‐operative management or additional biological adjuncts.

4. Type of bone defect (seeTypes of participants).

Sensitivity analysis

We anticipate that outcomes may be reported at a number of time points (e.g. six months and twelve months). We plan to include these outcomes in order to provide some sensitivity to the selection of an appropriate follow‐up time for assessment of the treatment effect. If it is necessary to exclude any studies because they appear to differ markedly from the majority of studies, then all main analyses will be reported with and without these studies. Sensitivity analyses to explore the impact of missing data are described above. Finally, we plan to conduct a sensitivity analysis for those studies to which we attribute a low risk of bias to explore the effect of excluding data from those studies with less reliable methodology.