The review was reported according to the Preferred Reporting Items for Systematic Reviews and Meta-Analyzes (PRISMA) statement (28), and followed the methods described in the Cochrane Handbook for Systematic Reviews of Interventions, Version 6.3 (29). The protocol was registered in the International Prospective Register of Systematic Reviews (PROSPERO): CRD42023416548.

We searched PubMed, Cochrane, Embase, Web of Science, and Scopus until March 10, 2023, without date or language restrictions. The search string was built as follows: knee osteoarthritis, exercise, and home (see Supplementary Appendix S1 for the full search strategy). The electronic database search was supplemented by a manual search to identify potentially eligible records.

Study selection was conducted by 2 authors using predetermined criteria independently. Disagreements were resolved by consensus after discussion with a third author. If the full manuscript could not be obtained, we contacted the author via email. Randomized controlled trials (RCTs) investigating HBE versus CBE or HBE combined with CBE for people with KOA were eligible. Participants were given an established diagnosis of KOA according to accepted criteria, regardless of age, gender, ethnicity, demography, and geography (30). The experimental intervention was HBE with all the exercise sessions being conducted at home. The control intervention was a supervised CBE with or without home-based exercise sessions, delivered in a hospital, outpatient department, private clinic, medical center, or community center. All interventions in the experimental and control groups were prescribed by a physical therapist or health professional. Excluding studies conducted on humans with non-KOA or conducted on animals. We determined and classified primary and secondary outcome measures in this review based on the international consensus on core outcome measures for phase III clinical trials of OA (31). The primary outcomes were self-reported measures, including pain, physical disability, and quality of life. The secondary outcomes were performance-based measures, including walking ability, lower limb muscle strength, and balance function.

Data from the included studies were independently extracted by 2 authors guided by Cochrane handbook (29). Any disagreement was resolved by discussion and a third author as required. The following descriptive data were extracted: Author, country of study, published year, participant characteristics (sample size, age, gender), intervention characteristics, and outcome measures at different follow-up times. The outcome measures were classified as patient-reported measures (1) pain (2), physical disability, and (3) quality of life; performance-based measures (4) walking ability (5) lower limb muscle strength, and (6) balance function. Where studies included more than one measurement scale, we extracted data from a scale that is highest on a suggested hierarchy (32, 33) (Supplementary Appendix S2). The short-term effect was defined as follow-up up to 3 months after baseline, and the long-term effect was defined as follow-up beyond 3 months after baseline. When multiple time points were available within the same follow-up period, the one closer to the endpoint of the intervention was used. The corresponding author of the relevant study was contacted to obtain missing data.

An assessment of the methodological quality of the primary articles was carried out by two reviewers independently, using the Cochrane risk of bias tool (version 2, ROB2) (34). This tool rates 7 potential sources of bias across 5 domains (randomization process, intended interventions, missing outcome data, measurement of the outcome, and selection of the reported result). Each trial was assessed against 5 bias domains to produce a summary risk-of-bias assessment score for each domain and overall (low risk, some concerns, or high risk of bias). Disagreements were resolved by discussion or adjudication. Results from these questions were graphed and assessed using an Excel RoB2 tool.

The post-intervention was used to obtain the pooled estimate of the effect of the intervention, using a random effects model due to the expected heterogeneity between the studies. We calculated standardized mean differences (SMDs) and 95% confidence intervals (CIs) for continuous variables with the inverse variance method. When necessary, SDs were calculated using available data (eg, 95% CI or p value) following the Cochrane guidelines. The effect size was interpreted as small (0.2), moderate (0.5), or large (0.8). Statistical heterogeneity was assessed using the I², with classification as low (I² < 25%), moderate (I² = 25–50%), substantial (I² = 50–75%), and considerable (I² > 75%) (29). A funnel plot would be used to evaluate publication bias if ≥10 studies were available for a given meta-analysis (35), but the number found did not reach this. Sensitivity analyzes were conducted for each outcome by excluding one study in each round. We used Review Manager (version 5.4.1) to perform all statistical analyzes. Data unamenable to meta-analysis were reported narratively.

The overall quality of evidence for each outcome was rated according to the Grading of Recommendations Assessment, Development, and Evaluation (GRADE) guidelines (36). Considering only RCTs were included, each outcome received a high certainty at the outset. Two reviewers assessed the quality of the evidence using the GRADE system, with potential disagreements resolved by discussion with a third reviewer. The GRADE downgrade details were presented in the Supplementary Appendix S3.

To ensure a consistent comparison of primary studies and a robust conclusion, we categorized questions into two groups. The first group compared the efficacy between HBE and CBE in the control group, while the second group explored the efficacy between HBE and HBE combined with CBE. There were no other inconsistencies with the pre-registration protocol.

The initial database searches identified 4,959 citations (Figure 1). After the removal of duplicates, 2,758 records remained. Based on the title and abstract, we excluded 2,722 records that did not meet the eligibility criteria. The remaining 34 full texts were identified and 24 were excluded. Of these excluded studies, 16 for wrong intervention, 4 for wrong study design, and 3 for incomplete data (Supplementary Appendix S4). No additional articles were identified through manual searching.

Descriptive characteristics of the 11 included studies are detailed in Tables 1, 2. HBE was compared with CBE in 6 studies (37–42), and with HBE combined with CBE in 5 studies (25–27, 43, 44). Studies were published from 2004 to 2022, with two studies conducted in Turkey (26, 39), two in Australia (37, 38), two in United States (25, 44), two in Nigeria (40, 41), with the remainder in Iran (43), Thailand (42), and United Kingdom (27). HBE had periodic remote monitoring by telephone (38–41, 43, 44) or videoconference (37), whereas four trials with no monitoring (25–27, 42). Support materials included exercise video (43, 44), exercise instruction manual (25, 40, 41, 43), exercise logbook (25, 40–43), and automated reminders (44). Two trials in the CBE were group-based exercise (27, 39) and the remaining trials were one-to-one exercises. Interventions for HBE combined with CBE additionally consisted of manual therapy (25), device-supported exercise therapy (26), and multiple physical factors (43). The intervention during for the included trials ranged from 4 to 24 weeks (2-7/wk), and participants were followed up for 0 to 48 weeks.

Risk of bias assessment is provided in Supplementary Appendices S5, S6. The assessment resulted in some concerns of risk of bias in most trials (25, 27, 37–41, 43), high risk of bias in 1 trial (26), and low risk of bias in 2 trials (42, 44). Due to insufficient information, most trials were scored unclear or high risk in the randomization process or/and deviations from the intended interventions. Two-fives of the trials (26, 27, 37, 38) were unclear or at high risk of outcome measurement, mainly because of the lack of assessor blinding. Except for two trials (42, 44), most of the other trials lacked a pre-registered protocol, which might result in selective reporting bias risk.

Six studies involving 240 participants contributed data on the effectiveness of HBE versus CBE. The follow-up periods ranged from 0 to 12 months. The GRADE evidence profile comparing the efficacy of HBE and CBE for primary outcomes and secondary outcomes is reported in Supplementary Appendices S7, S8.

Five studies investigated the effectiveness of HBE versus HBE combined with CBE, with two studies (26, 27) (250 participants) included in meta-analyzes. Three other studies (25, 43, 44) failed to include meta-analysis due to inconsistent interventions, which contained additional center-based interventions besides supervised in-person exercise in their CBEs, potentially confounding the effects of exercise therapy. The follow-up periods ranged from 0 to 12 months. The GRADE evidence profile comparing the efficacy of HBE and HBE combined with CBE for primary outcomes is reported in Supplementary Appendix S9.

Subgroup analyzes were not performed owing to the limited number of studies for each outcome. Sensitivity analyzes were conducted to evaluate the robustness of our meta-analysis findings when more than two studies were included. These sensitivity analyzes did not alter any of the meta-analysis results, which suggests this finding is robust.

The funders played no role in the design, conduct, or reporting of this study.

There are several limitations to this systematic review with meta-analysis worth noting. Fewer trials and a small sample size limited the precision of the findings and undermined the capacity to assess publication bias graphically or statistically. Additionally, although data were extracted and pooled based on the priority of outcome indicators as recommended (5, 33), the pooled effect was calculated using the SMD, which is less clinically meaningful than a mean difference. When comparing HBE and HBE combined with CBE, no definitive conclusions were reached due to the limited number of studies. Future studies need to ascertain the amount of CBE supplementation for HBE that will have optimal cost-effectiveness, providing essential guidance for future clinical practice regarding people with KOA. On the other hand, individual adherence is the core component of HBE (55, 56), yet less reported in studies of HBR for KOA patients. Only one trial (27) reported adherence included in this review, showing no statistical difference between HBE and CBE. Lastly, fewer studies reported quality of life, performance-based outcomes, and long-term outcomes, leading to controversial results. Nevertheless, as the first systematic review to examine the effectiveness of HBE versus CBE, we can still have an impact on current clinical practice and future clinical research. More high-quality studies comparing the efficacy of HBE and CBE (with or without HBE) are needed in the future to strengthen the findings of this review.

Home-based exercise presents comparable benefits to center-based exercise in terms of patient-reported and performance-based outcomes among people with knee osteoarthritis. This highlights the importance of prioritizing the incorporation of home exercise regimens into clinical guidelines for this population. Home-based exercise can effectively mitigate pain and functional deterioration in people with knee osteoarthritis with less medical supervision, hence decreasing the national healthcare burden. Furthermore, concurrent application of home-based and center-based rehabilitation may lead to superior outcomes. However, given the limited number of studies on the matter, it remains uncertain what the optimal number of supervised outpatient rehabilitation sessions would represent the most cost-effective strategy for the management of knee osteoarthritis. Further clinical research is warranted to address this issue. Finally, future studies should also include health economic analyzes and larger sample sizes. This could allow more robust subgroup analyzes and explore which specific subgroups might benefit most from home-based exercise.