The current effectiveness evaluation and the previously published cost-effectiveness evaluation are both part of the VIScosupplementation for Knee osteoarthritis (VISK) study. The VISK study is registered at the Dutch trial register (www.trialregister.nl, NTR1651). The study protocol is available from the corresponding author on request.
The VISK study does not include a placebo group. In light of the evidence on the efficacy of HMW-HA in knee OA, we specifically sought to investigate the actual effectiveness of this intervention. Such a study design, in which the intervention is compared to what is considered regular care that is provided in an everyday clinical setting (without a placebo), is required to facilitate the parallel economic evaluation of the VISK study [26, 27].
Study sample
Inclusion of eligible subjects took place between May 2009 and May 2010 in 2 hospitals (1 academic, 1 non-academic) in The Netherlands. Consecutive knee OA patients at the outpatient orthopaedic department meeting the inclusion criteria were considered eligible. Patient’s age was set between 18 and 65 years, the latter being the pensionable age in The Netherlands at the inclusion period. Inclusion criteria were: pain > 3 months, mean pain severity ≥2 on the numeric rating scale (NRS), Kellgren & Lawrence (K&L) grade I to III in medial and/or lateral compartment.
Exclusion criteria were: intra-articular HA injections < 1 year, intra-articular steroid injection < 3 months, arthroscopy < 6 months, tibial osteotomy < 1 year, synovectomy, scheduled knee surgery < 1 year, varus/valgus deformity > 12 degrees, chondrocalcinosis, dermatologic knee disorders, allergy to HMW-HA components, (planned) pregnancy or lactation, inflammatory arthritis, severe hip OA, non-knee related regular analgesic use, daily oral steroid therapy, poor general health, conditions interfering with functional assessments, alcoholism, patients unable to attend follow-up and insufficient command of the Dutch language.
Sample size, randomization and masking
The sample size was calculated to detect a between group difference of 20% in the primary outcome parameter which was defined as response to therapy at 52 weeks according to OMERACT-OARSI criteria [28]. A power of 80% and an alpha of 0.05 resulted in a required sample size of 64 subjects per group (128 subjects in total). Anticipating a 20% dropout over 52 weeks, the final required sample size was set at 154 subjects.
Randomization took place after informed consent was signed. Concealed randomization was performed by computer generated lists with randomly assigned blocs of 2, 4, 6, 8 or 10 subjects. An independent employee not involved in any other part of the study performed the randomization. Stratification took place for radiologic degree of knee AO (K&L grade I/II versus grade III) and per orthopedic surgeons responsible for injections (2 per hospital, 4 in total).
The statistician and investigator responsible for assessment and analyses of the data were blinded for the treatment allocation. Due to the study design included subjects and orthopedic surgeons administering the study intervention could not be blinded.
Interventions
Subjects in the intervention group received 3 weekly intra-articular injections with Hylan G-F 20 (Sanofi S. A, Paris, France) added to usual care or usual care only. Hylan G-F 20 is the HMW-HA derivative with the highest molecular weight available for clinical use (6000 kDa. The injections were performed through the superolateral approach [29]. Usual care was defined accordingly to the guidelines on the treatment of knee OA of the Dutch Orthopedic Association. This guideline recommends several non-surgical treatment modalities including pain medication (eg acetaminophen or NSAIDs), physical therapy and lifestyle recommendations [12]. Treating physicians were encouraged to follow these guidelines, but no treatment restraints were imposed. Other treatments were allowed when deemed appropriate in order to maintain the pragmatic character of the trial.
Questionnaires
The follow-up was 52 weeks and data was collected through questionnaires by mail at baseline, 6, 13, 26, 39 and 52 weeks. Knee related function was assessed by the functioning in daily living scale of the Knee injury and Osteoarthritis Outcome Score (KOOS) [30, 31]. A normalized score from 0 (extreme symptoms) to 100 (no symptoms) was calculated for this subscale. Pain during rest and pain during activity was evaluated by the NRS, resulting in a score between 0 (no pain) and 10 (most severe pain) [32]. Patient’s global assessment (PGA) was assessed on a 5-point Likert scale on which subjects indicate the amount of improvement of their knee complaints compared to baseline (1. fair improvement, 2. moderate improvement, 3. no change, 4. moderate deterioration, 5. fair deterioration). Medication use and patient reported adverse events were monitored at all follow-up moments.
Outcomes
The primary outcome was defined as response to therapy at 52 weeks follow-up according to OMERACT-OARSI criteria. This variable presents the results of changes after treatment in three symptomatic domains (pain, function, and PGA) as a single variable [28]. Response to therapy according to the OMERACT-OARSI criteria is defined as ≥10% absolute improvement and ≥ 20% overall improvement at final follow-up in at least 2 of the 3 responder domains (pain, function and/or PGA); or ≥ 20% absolute improvement and ≥ 50% overall improvement in either the pain or function domain.
The secondary outcome comprised the between group difference over the whole follow-up period of the 3 individual primary outcome responder domains: pain, function, and PGA.
Statistical analyses
For the primary outcome, the difference in percentage of responders according to OMERACT-OARSI criteria between study groups after 52 weeks follow-up was calculated [28]. In the base case analyses two responder sets were investigated: 1. with pain during rest was included in the responder domains, next to function and PGA; and 2. with pain during activity included.
In order to minimize bias in favor of the intervention group, drop-outs and subjects lost to follow up were (regardless of their study results) considered non-responders in the intervention group, and (vice versa) responders in the control group in the final analyses.
Logistic regression analysis with responder as dependent variable and the intervention as independent variable were performed to calculate odds ratios (OR) including 95% confidence intervals (95%CI) after 52 weeks follow-up. The number needed to treat (NNT) to attain 1 responder was calculated (PASW statistics 17.0).
For the secondary outcome, scores on individual responder domains (pain during rest, pain during activity, knee related functioning in daily life, PGA) were analyzed over all follow-up moments by means of a random effects model with random intercept and slope. The baseline values of the variables and the treatment group were included in the model. In this way we obtained for each outcome an estimate for the between group difference in score on the relevant questionnaires (KOOS, NRS, Likert scale) over the whole follow-up period, including associated 95% CI (SAS 9.2, SAS Company).
Sixteen subjects divided over both study groups received knee related surgery during follow-up. This number was not foreseen and we therefore performed 2 additional sensitivity analyses to assess possible beneficial clinical effects on pain and function as a result of the surgery. These analyses were not specified in the VISK study protocol a priori. In these sensitivity analyses, subject receiving knee related were considered non-responder irrespective of their study results. [33] In the first additional analysis, subjects who received major knee related surgery (e.g. knee prosthesis implantation, high tibial osteotomy) during follow-up were considered non-responders. In the second additional analysis subjects receiving any knee surgery (major knee surgery plus minor knee surgery like arthroscopy or knee manipulation under general anesthesia) were considered non-responders. (PASW statistics 17.0) All analyses were performed according to the intention to treat principle. In order to generate unbiased estimates of the difference in effectiveness parameters across both treatment groups, we adjusted for baseline imbalances in and, if necessary, for unbalanced covariates.