Predictors of placebo response to local (intra-articular) therapy in osteoarthritis: an individual patient data meta-analysis

Lead Investigator: Shirley Yu, University of Sydney
Title of Proposal Research: Predictors of placebo response to local (intra-articular) therapy in osteoarthritis: an individual patient data meta-analysis
Vivli Data Request: 7660
Funding Source: Dr Shirley Yu holds a University of Sydney Postgraduate Research Scholarship (Part Time)
Potential Conflicts of Interest: Professor David Hunter provides consulting advice to Merck Serono, Pfizer, Lilly, TLCBio, Kolon Tissuegene. The potential conflicts of interest will be managed through disclosure of interests when the research is presented and published.

Summary of the Proposed Research:

Osteoarthritis (OA) is a highly prevalent and disabling condition with limited safe and effective treatment options available. Intra-articular therapies are increasingly being utilised, especially in patients with knee or hip osteoarthritis. However, whether the effect of these agents is due to active treatment or placebo remains largely unknown. The placebo effect is part of treatment effect and identifying the magnitude and potential predictors of this effect in intra-articular therapies will inform the design of future trials and clinical practice.

Method and analysis:
A systematic literature search will be conducted for randomised clinical trials comparing corticosteroid and viscosupplementation/hyaluronic acid intra-articular injection with placebo for knee and hip osteoarthritis. Literature searches will be conducted through Pubmed (Medline), EMBASE, Web of Science, Cochrane Central, and SCOPUS from inception to May 2018 . Individual patient data from each study will be requested and obtained from the corresponding authors of the trials. Risk of bias will be assessed using the Cochrane Collaboration’s tool.

The primary outcome will be change in pain from baseline. Secondary outcomes will be change in function and patient global assessment. Predictors of response to treatment that will be assessed include patient characteristics (age, gender, bilateral versus unilateral disease, other joints OA involvement, radiographic severity, pain severity and presence of inflammatory features based on imaging and physical examination) , intervention characteristics (aspirate volume, frequency of injection, volume of injection, and intra-articular injection approach), and trial characteristics (clinical setting , blinding, use of intention to treat analysis, funder/sponsor). We will report our results using the Preferred Reporting Items for Systematic review and Meta-Analysis (PRISMA)-IPD guidelines.

Statistical Analysis Plan:

A descriptive evaluation of each trial and study participants will be conducted. Publication bias will be investigated using a funnel plot analysis as this will specify the potential impact of both known and unknown missing trials on the results. Missing data will be assumed to be missing at random, thus patient characteristics will be used to impute missing data by means of multiple imputation at random. In addition, we will compare the effect sizes (ES) pooled from those responded versus the overall (ie, the ES pooled from all trials systematically searched from the literature) to examine the deviation.

Baseline and follow-up data from the placebo arm will be used to estimate the predictors of the placebo response. Separate analyses will be conducted for glucocorticoids and viscosupplementation/hyaluronic acid, as well as different outcome measures (i.e. pain, function and patient global assessment). Trials will also be grouped by type of joint (i.e. knee or hip) and follow up duration (e.g. < 4 weeks or ≥ 4 weeks for corticosteroid and <12 weeks or ≥12 weeks for viscosupplementation/hyaluronic acid).

A one-step approach will be applied, via the use of multilevel regression models to assess for predictors of the placebo response. The use of the one step approach in this setting will allow for a more cohesive modelling of covariates and account for the clustering of participants within the study. This will be done by combining all the data from all the studies available after appropriate standardisation of the variables and a new dataset will be formed to allow for further analysis. To assess for the potential subgroup effects, a random effect model will be utilised given the hierarchical nature of the data to assess the interaction effects, with change in pain being a dependent variable, and potential predictors being independent variables. In the setting where a no-treatment control is available, we will include placebo-no-treatment as an independent variable. Responders to placebo will be compared with non-responders to identify predictors of response.

The primary outcome will be change in pain from baseline and will be determined as the dependent variable in the regression model. The minimum clinically important difference (MCID) threshold will be a 20% or more reduction in pain based on the visual analogue scale (VAS) pain score with 0mm being no pain to 100mm being the worst pain ever. This level has been recommended for use in pain and function assessment in rheumatic diseases such as osteoarthritis (OA), and we will use it to define the placebo response which is equivalent to an ES of 0.8, that indicates the response unlikely to be caused by spontaneous effects. In situations where WOMAC pain score is only available, it will be used instead.

Secondary outcomes will be a change in function and patient global assessment. Change in pain will be determined as the dependent variable, and independent variables will be the potential predictors of placebo response. These will be grouped as patient-level characteristics, peripheral pain mechanisms, central pain mechanisms, intervention characteristics and those related to trial design (blinding, funder/sponsor roles and intention to treat) and are as listed below. Each group will be forced into multivariate models with a final model including all groups.
1. Patient characteristics: age, gender, body mass index, bilateral versus unilateral disease, disease duration,
2. Pain mechanisms: peripheral pain mechanisms (i.e. signs of inflammation, morning stiffness symptoms and radiographic findings), central pain mechanisms (i.e. other joint OA, comorbidities, pain severity),
3. Intervention characteristics: clinical setting (i.e. location of intervention), aspirate volume, frequency of injection, volume of injection, and intra-articular injection approach (i.e. medial vs lateral approach, use of ultrasound guided injection).
4. Trial characteristics: blinding (patients, assessors or physicians), dropout rates, role of funder/sponsor (i.e. pharmaceutical company), randomisation ratio, trial duration, single centre/multi-centre study, parallel/cross-over trial, and use of ‘intention to treat’ analysis.
The trials that originate the individual patient data will also be coded and included as a level variable in all analyses. Effect sizes and 95% confidence intervals will be generated for each outcome measure. P <0.05 will be considered statistically significant.

A sensitivity analysis will be conducted using pain scores (instead of change in pain scores) as a continuous dependent variable and repeating the approaches described above.

Statistical analyses will be performed using Stata SE and SAS

Requested Studies:

A Multi-Centre, Parallel, Double-Blind, Blinded Evaluator, Randomised, Placebo-Controlled Evaluation of the Efficacy and Safety of a Single Dose of 6 mL of Synvisc in Patients With Symptomatic Osteoarthritis of the Knee
Data Contributor: Sanofi
Study ID: NCT00131352
Sponsor ID: SYNV00704

Public Disclosures:

Yu SP, van Middelkoop M, Deveza LA, Ferreira ML, Bierma‐Zeinstra S, Zhang W, Atchia I, Birrell F, Bhagavath V, Hunter DJ. Predictors of placebo response to local (intra‐articular) therapy in osteoarthritis–an individual participant data meta‐analysis. Arthritis Care & Research. 2023 Jul 31. doi: 10.1002/acr.25212