Cell-based versus corticosteroid injections for knee pain in osteoarthritis: a randomized phase 3 trial

Primary outcome

Both primary outcome measures results are shown in Fig. 2. The co-primary-outcome measure of VAS pain score was analyzed as the change from baseline to 12 months. The mean decline from baseline in VAS pain score changed in different ways by treatment group but was not significantly different in overall decline (for example, overall magnitude was same, but trajectory was different). There were no significant differences between groups at month 12 in the change in VAS score from baseline (change, −24.3 ± standard error of the mean (s.e.m.) in BMAC, −19.4 ± s.e.m. in SVF, −20.1 ± s.e.m. in UCT and −20.9 ± s.e.m. in CSI; difference versus CSI, BMAC: −3.4; P = 0.19, SVF: 1.5, P = 0.56, UCT: 0.8, P = 0.76). The analysis of KOOS pain score yielded similar results as there was no significant between-group difference at month 12 in the change in score from baseline (change, 19.1 ± s.e.m. in BMAC, 17.2 ± s.e.m. in SVF, 16.2 ± s.e.m. in UCT and 17.7 ± s.e.m. in CSI; difference versus CSI, BMAC: 1.4; P = 0.49, SVF: −0.50, P = 0.82, UCT: −1.5, P = 0.44). Prima sensitivity analyses were performed using the observed case, the per protocol population and multiple imputation under the assumption of missing not at random. For both the VAS pain score and the KOOS pain score, there was no significant between-group difference at month 12 in the change in score from baseline for any of the sensitivity analyses. Interaction between treatment group and age group (<60 versus ≥60), sex, ethnicity and Kellgren–Lawrence (KL) grade were also analyzed. There was a significant interaction between treatment group and age group (P = 0.02) and between treatment group and sex (P = 0.01) for VAS pain score. There was no significant interaction for KOOS pain score. In addition, the magnetic resonance imaging (MRI) scores had no significant changes from baseline in any of the four groups (BMAC 0.53, SVF −0.40, UCT −0.26 and CSI 0.30) compared to their baseline scores or compared to the CSI group (BMAC 0.23, SVF −0.69 and UCT −0.55). See Fig. 3 for description of MRI scoring system.

Secondary outcome

In addition to our primary outcome measures we also analyzed EQ-5D and PROMIS-29 between cohorts and CSI. For EQ-5D, the treatment by time interaction was not significant (P = 0.26), suggesting EQ-5D in the four treatment groups changed in similar ways (similar temporal patterns for the four treatment groups). Since the interaction between treatment and time is not significant, we would not expect to see specific ‘change versus change’ differences. Similarly for PROMIS-29, we assessed all domains by a treatment by time interaction and there was no significance for the following subdomains: PROMIS-29 Anxiety (P = 0.78), Depression (P = 0.06), Fatigue (P = 0.56), Pain (P = 0.39), Physical Function (P = 0.048), Sleep (P = 0.91) and Social Rules and Activities (P = 0.82). It should be noted that the PROMIS-29 Physical Function was statistically significant, suggesting the physical function PROMIS-29 domain T-score in the four treatment groups changed in different ways (different temporal patterns for the four treatment groups). However no clear clinically important difference in the temporal pattern was apparent between the four treatment groups. The full details for these secondary outcomes are located in Supplementary Material 1.

Exploratory outcomes

Bedside testing of total nucleated cells injected and viability for each cellular group was analyzed. In addition, 71 BMAC, 16 SVF and 8 UCT-MSC samples at the time of publication were subjected to single-cell RNA sequencing to reveal differences and similarities in the cellular components of each product via cell clustering analyses visualized in two dimensions using Uniform Manifold Approximation and Projection. Transcriptomic analyses at the single-cell resolution revealed that both autologous cell sources exhibited distinct cell subpopulations with some similarities in a subset of hematopoietic lineage-derived cells. Conversely, the UCT mesenchymal population, although exhibiting a defined clustering pattern, showed a uniform MSC phenotype.

Safety

There were no procedure-related serious adverse events (AEs) reported, which includes any allergic reactions or symptomatic infections seen in any treated patient. However, there were multiple related AEs reported that have been subdivided. The following related AEs demonstrated significance between cohorts: joint swelling (CSI 7.4% versus UCT 24.1%, P = 0.01), post-procedural contusion (SVF 38.6% versus BMAC 12.2% versus UCT/CSI 0%, P < 0.0001), post-procedural hematoma (BMAC 2.9% versus SVF 12.4%, P = 0.02). For a list of the most common AEs by study arm and by study group, see Tables 2 and 3.

Study design and participants

This was a multicenter single-blinded study performed in accordance with guidelines and oversight from the FDA, and under the management of a contracted research organization. Study approval was obtained from the Western Institutional Review Board and by Duke and Emory University’s institutional review board. In accordance with the FDA, an investigational new drug application was filed, #18414, which referenced investigator device exemption #17894. The study was also registered with ClinicalTrials.gov, NCT03818737. Data availability at this time is available by request to the corresponding author(s).

A total of 570 patients were screened to identify 480 eligible patients who were randomized at five clinical sites in five different states within the United States. The five clinical sites included clinics from Emory, Sanford (two sites), Andrews Institute, and Duke. For a consort diagram of participants entered in study, see Fig. 1. Participants were randomized to a four-cohort parallel-design study. To allow for blinding, participatnts were further subdidivided to three arms given the need to ‘sham’ harvest cells from subjects in the SVF and BMAC arm, but not in the UCT/CSI arm: arm 1—autologous bone marrow concentrate (BMAC) (n = 120), corticosteroid (CSI) (n = 40); arm 2—SVF (n = 120), CSI (n = 40); arm 3—umbilical cord tissue (UCT) (n = 120), CSI (n = 40). The control cohort of CSI was then aggregated to allow a 1:1:1:1 comparision for analysis. Subjects were enrolled if they were between 40 and 70 years of age and carried a diagnosis of knee OA as determined by radiographs within 3 months of their clinical visit. A full list of eiligibility criteria is provided on ClinicalTrials.gov, NCT03818737. A total of five samples did not meet release criteria, including one BMAC and four SVF. This was secondary to failure of endotoxin testing and not formal cell count. Additionally, the following failed to complete the study. In the BMAC cohort, a total of 12 subjects did not complete the stud, with 1 related to release criteria as above, 10 related to subject withdrawal and 1 lost to follow-up. In the SVF cohort, a total of 12 subjects did not complete the study, with 1 due to investigator withdrawal, 4 related to release criteria as above, 6 related to subject withdrawal and 1 due to screen fail. In the UCT cohort, a total of 13 subjects did not complete the study, with 2 due to investigator withdrawal, 10 related to subject withdrawal and 1 lost to follow-up. Lastly, in the CSI cohort, a total of 11 subjects did not complete the study, with 1 due to investigator decision, 6 related to subject withdrawal and 4 lost to follow-up. Patients returned to clinic for MRI to assess cartilage and joint health at baseline, 6 months and 1 year. They were compensated US$50 for each MRI that was performed.

Randomization and masking

Subjects were stratified by clinical center, and after obtaining informed consent and verifying eligibility criteria were met, subjects were randomized. Treatment assignments were stratified by clinical center and generated using a pseudo-random number generator with randomly permutated blocks. These assignments were stored in the Medidata Rave cloud-based data management system developed by the contracted research organization. All subjects underwent the harvesting procedure per their assigned study arm, then per the randomization scheme they were blinded to the actual treatment received (for example, SVF versus CSI, BMAC versus CSI, or cells versus CSI). As a single-blinded study, the site principal investigators were not required to be blinded. However, subjects were blinded to their injection. The blinding was implemented by limiting visualization of the syringe contents with opaque covering in addition to performing sham BMAC and SVF harvests in patients within certain cohorts.

Procedures

The cellular harvesting, final product preparation, injection procedures and CSI were standardized across the five study sites. This was done through training courses before study initiation as well as subsequent monitoring by the lead site. Of note, the BMAC and SVF were fresh autologous products while the UCT were cryopreserved, purified MSCs manufactured from donated allogeneic umbilical cord tissue in a cGMP (current Good Manufacturing Practice) facility. Supplementary Material 1 details the contents of each preparation that was injected.

All procedures were done in an outpatient clinic setting with another clean room used for point-of-care laboratory testing. No cGMP facility was used for the actual treatment portion of this study. The subject was positioned supine on an examination table with foam roller/bolster under knee. The injectate was prepared by the research team with opaque tape wrapped around the 10-ml syringe to maintain the blinding. If an effusion was present based on clinical and ultrasound examination, 2 ml of ropivicaine 0.2% was injected between the skin and down to the joint capsule. Following this, an 18-gauge needle was used to aspirate any joint fluid via a superior lateral approach under direct ultrasound guidance. Following joint aspiration (if performed), the solution of either corticosteroids, BMAC, SVF or UCT was injected under ultrasound guidance. Following the injection, the knee was passively moved from extension to flexion three to four times to help spread the injectate and patients were instructed to remain supine for 10 min.

All final cellular products were tested in the clinic for total nucleated cell count, cell viability and endotoxin levels to determine if release criteria were met before injection. These tests were performed using a Nexcelom Auto 2000 device for cell counting and cell viability and the Charles River Endosafe Nexgen PTS for endotoxin testing. A 1-ml aliquot of each cellular product was separated from the final injection preparation for the release criteria testing and FDA requirements. If the final product did not meet the required release criteria depicted in Table 1, the subject did not receive the injection. This was seen in one BMAC patient and four SVF patients. In addition to bedside testing, 14-day sterility testing was performed post administration per FDA requirements. In an effort to standardize the injectate and per FDA guidance, cutoffs and cellular numbers were derived from a group of subject matter experts.

Cellular diversity in BMAC and SVF samples and cellular heterogeneity in UCT-MSC samples were evaluated by single-cell RNA sequencing. The single-cell RNA sequencing libraries were prepared with 10x Genomics Chromium platform using 3′ V3.1 kit. The sequencing was done on Illumina Novaseq 6000 platform with an S4 kit. We used a modified SEURAT pipeline to analyze the samples²⁶. We then applied filters to include cells that had more than 800 unique molecular identifiers, more than 500 expressed genes and mitochondria percentage less than 20%.

Outcomes

The primary analyses of the data were performed according to subjects’ original treatment assignment (that is, intention-to-treat analyses) regardless of their compliance and the inclusion of all data from all subjects randomized in the final analysis. The two co-primary efficacy endpoints in this intent-to-treat, parallel-group trial were VAS and KOOS pain score at 12-month visit from baseline.

KOOS is a self-reported outcome measure assessing the patient’s opinion about the health, symptoms and functionality of their knee. It is a 42-item questionnaire, including five subscales: symptoms, pain, activities of daily living, sports/recreation and quality of life. The maximum score a patient can achieve is 100, and the minimum score is zero, indicating severe knee problems. KOOS has been verified in assessing patients of various age populations, ranging from young to elderly adults²⁷.

The VAS is a single-item measure that most commonly consists of a 100-mm horizontal line anchored with two opposite labels; patients mark a score on the scale using a vertical line²⁸. Magnetic resonance images were evaluated by a musculoskeletal radiologist according to a semi-quantitative method for whole-knee analysis in the setting of OA. Our methodology, adopted from past work (that is, WORMS (1) and BLOKS (2) grading schemes) evaluated severity and/or extent of cartilage loss, bone marrow edema (BME) or cyst, osteophytes, meniscus pathology and extrusion, ligament pathology, synovitis, and fat pad inflammation^29,30. Both WORMS and BLOKS were utilized for this study. T2 mapping is a technique to determine intrinsic spin-lattice relaxation times of biological tissues and has been studied extensively in the knee to assess cartilage and meniscus degeneration. For this project, we utilized spin echo T2 mapping technique, where four to eight images with echo times ranging from ~10 ms to ~80 ms were obtained in weight-bearing regions of the medial and lateral compartments. For the magnetic resonance scoring system that was used, see Fig. 3.

Statistical analysis

Power and sample size considerations for the trial are found in Supplementary Material 1. For primary endpoint analyses, missing data for VAS pain scores and KOOS pain scores were imputed using multiple imputation under the missing at random assumption). Independent imputations were performed for VAS pain scores and KOOS pain scores. Absolute change from baseline in VAS pain score and KOOS pain score were derived from the corresponding imputed scores. Sensitivity analyses for the primary endpoints were performed using the observed case, the per protocol population and multiple imputation based on a pattern-mixture model under the assumption of missing not at random (all found in Supplementary Material 1.).

A repeated-measures analysis of change of VAS pain score (and change of KOOS pain score) was performed with a means model via the SAS MIXED Procedure (version 9.4; SAS Institute, Cary, NC), providing separate estimates of the means by treatment group (BMAC, SVF, UCT and CSI) time on study (1, 3, 6, 9 and 12 months on study) and treatment group. The models included treatment arm, time on study, the statistical interaction between treatment arm and time on study and study center as fixed effects. A compound-symmetric variance–covariance form in repeated measurements was assumed, and robust estimates of the standard errors of parameters were used to perform statistical tests and construct 95% confidence intervals (CIs)³¹. The model-based means are unbiased with unbalanced and missing data, if the missing data are non-informative (missing at random). The main effect test for treatment at the 12-month visit was used as the primary hypothesis test to compare the treatment arms. The primary study results from this model were the mean change score and 95% CI for each of the four treatment cohorts and the treatment mean differences and 95% CIs. Pairwise treatment comparisons on efficacy score change were performed. The primary comparisons were each cellular treatment against control group. A Hochberg adjustment method was used to maintain the overall α level, which ordered the P value from high to low and compared the largest P value to 0.05, the middle P value to 0.05/2 and the smallest P value to 0.05/3 (ref. ³²). Specific statistical tests were done within the framework of the mixed effects linear model. All statistical tests were two-sided. Secondary efficacy endpoint analyses (all VAS, KOOS and EQ-5D 3L scores) were conducted using an observed case analysis using the same plan described for the change scores. No interim analyses for efficacy were performed for this study.

The heterogeneity of treatment effects across levels of a baseline variable was investigated using a statistical test for interaction. A prespecified subgroup analysis is one that is planned and documented before examination of the data. Planned subgroup analyses were performed to examine the impact of treatment on the primary outcomes in the prespecified subgroups (that is, sex, ethnicity and KL grade). Effect of treatment in subgroups was determined by including the interaction between treatment and subgroup in the repeated-measures model described above using an observed case analysis. Additional secondary and exploratory outcome measures included change in MRI cartilage and joint health from baseline to 1 year, analysis of AEs and complications between groups, as well as in-depth cellular analysis of each injectate.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Peer review information

Nature Medicine thanks Susanne Grässel and the other, anonymous, reviewer(s) for their contribution to the peer review of this work. Jerome Staal, in collaboration with the Nature Medicine team.