Effects of therapeutic interventions on long COVID: a meta-analysis of randomized controlled trials, 2025, Chang Tan et al

Mij

Senior Member (Voting Rights)

Summary​

Background​

Long COVID, characterized by persistent multi-organ symptoms post-SARS-CoV-2 infection, poses a substantial global health burden. While diverse therapeutic interventions have been proposed, their comparative efficacy remains uncertain due to fragmented evidence and methodological heterogeneity in prior studies. Therefore, we conducted a meta-analysis to comprehensively explore the effectiveness of diverse therapeutic interventions in long COVID.

Methods​

In this meta-analysis, we searched PubMed, Cochrane Library, Embase, Web of Science, SPORTDiscus (EBSCO), CINAHL (EBSCO), and Rehabilitation & Sports medicine source (EBSCO) from inception to July 20, 2025, for randomized controlled trials (RCTs) evaluating exercise training, respiratory muscle training, telerehabilitation, transcranial direct current stimulation (tDCS), olfactory training, palmitoylethanolamide with luteolin (PEA-LUT), and steroid sprays in adults with Long COVID. Primary outcomes included cardiopulmonary function, exercise capacity, fatigue, and olfactory recovery. Data were pooled using random-effects models, with sensitivity analyses (leave-one-out method) and Egger's test to assess robustness and publication bias. GRADE criteria evaluated evidence certainty. The study was registered with PROSPERO (CRD42024591704).

Findings​

We identified a total of 51 eligible trials, comprising 4026 participants. Significant differences were observed in the following outcomes in the context of exercise training: 6MWT (MD, 83.20; 95% CI 52.04–114.37), 30sSTS (MD, 3.05; 95% CI 1.96–4.13), SF-12 Mental Component Summary (SF-12-MCS) (MD, 3.10; 95% CI 0.78–5.43), VO2 peak (% predicted) (MD, 6.00; 95% CI 0.45–11.54), VO2 peak (L/kg/min) (MD, 1.61; 95% CI 0.40–2.81), VO2 peak (L/min) (MD, 0.14; 95% CI 0.03–0.25), mMRC dyspnea scale (MD, −1.04; 95% CI −1.73 to −0.35), the Multidimensional Functional Assessment of Daily Living Scale (MBDS) (MD, −4.61; 95% CI −8.19 to −1.03), and Visual Analogue Fatigue Scale (VAFS) (MD −1.69; 95% CI −3.07 to −0.31). Furthermore, significant differences were also found in the following key outcomes: 6MWT (MD, 89.54; 95% CI 9.86–169.23), MIP (% predicted) (MD, 15.79; 95% CI 2.73–28.84), MIP (cm H2O) (MD, 19.69; 95% CI 10.14–29.24), and mMRC (MD, −1.02; 95% CI −1.86 to −0.18) in respiratory muscle training; 6MWT (MD 34.14; 95% CI 2.54–65.74), 30sSTS (MD 1.41; 95% CI 0.67–2.15), and FSS (MD −1.59; 95% CI −2.64 to −0.53) in telerehabilitation; MFIS-physical (MD, −2.29; 95% CI −4.36 to −0.22) in tDCS; and TDI Score (MD, 4.66; 95% CI 2.16–7.15) in PEA-LUT.

Interpretation​

Exercise training should be prioritized for improving cardiopulmonary function and exercise capacity in Long COVID, supported by high-certainty evidence. Respiratory muscle training and PEA-LUT offer targeted benefits for respiratory strength and anosmia, while tDCS may address fatigue. Telerehabilitation, as a form of supervision, also improved the effectiveness of the intervention. In contrast, steroid sprays and olfactory training lack efficacy, highlighting the need for personalized, symptom-specific approaches. These findings advocate for updated clinical guidelines integrating multimodal therapies and underscore the urgency of large-scale trials to optimize dosing and long-term outcomes.
LINK
 
Effects of therapeutic interventions on long COVID: a meta-analysis of randomized controlled trials - eClinicalMedicine. No surprises here but good to see. What many of us who folliwed the evidence have been suggesting
 

Risk of bias​

The risk of bias of RCTs ranged from low to high, with 3 studies with low risk of bias, 12 with some concerns, and 33 with high risk. Lack of blinding or unclear description of blinding caused more bias. In more than half of the studies, there was a loss to follow-up rate exceeding 10%, which was the main source of risk of bias and affected up to 28 articles. The problem with randomization and blinding is primarily that the methods of implementation were not described in detail in the literature, so we do not know with certainty whether randomization and blinding were actually performed. Selective outcome reporting was the domain with better scores (eFigure 2).

GRADE assessment​

The GRADE assessments indicated the following quality ratings for all outcomes: 27% of the evidence was rated as very low, 57% as low, 11% as moderate, and 5% as high. Only two outcomes—MFIS-physical in tDCS and MIP in respiratory muscle training (% predicted)—received a high-quality rating (eTable 2).
Our study has several limitations. Firstly, not all relevant RCTs were included due to various factors such as the unavailability of original articles, non-English language publications, or the use of different scales or experimental methods to assess the same symptoms, which limited the number of studies available for analysis. Secondly, the quality of the included literature was variable, with some studies exhibiting a high risk of bias. Thirdly, the included studies demonstrated limited longitudinal outcome assessments, with maximum follow-up durations capped at 180 days and the majority restricted to 4–12week observation periods. This collectively constitutes a notable methodological limitation regarding sustained therapeutic effect evaluation. Fourthly, the limited number of studies included for some outcomes resulted in a limited ability of sensitivity analyses to assess the robustness of those outcomes. Fifthly, although individualized multimodal interventions may be promising, our current analysis was unable to evaluate their effectiveness as a distinct category due to the limited number of relevant studies. The potential benefits of combining multiple interventions warrant further investigation in future clinical trials. Additionally, certain experiments were difficult to blind, which may have introduced subjective factors that influenced the original results, making the findings of this study susceptible to bias. Lastly, there was high heterogeneity across studies, likely due to differences in intervention methods, duration, frequency, and follow-up periods, which may have contributed to variability in the results.
Another worthless meta analysis where the authors don’t know when to just exclude studies with terrible methodology. It’s fitting that it’s published in the Lancet..
 
worthless meta analysis where the authors don’t know when to just exclude studies with terrible methodology.

it's amazing they can analyze all the findings and present the meta-results as somehow robust, but then tuck away at the back that almost all the trials are shit and 84% of the evidence is of low/very low quality. When limitations are so expansive that they undermine the credibility of the claimed results, they essentially indicate that the results are, as you suggest, worthless.
 
The eFigure2.x’s in the supplements are a red flag to me. I find it highly unlikely that most of the interventions were sufficiently blinded but I don’t have the energy or will to check.

I also don’t understand how you can have a category of «unclear bias». If the bias can’t be determined, shouldn’t it be assumed to be high risk? Benefit of the doubt doesn’t really work in science - then you could just leave out info that’s not in your favour and get a better score.
 
Last edited:
Back
Top Bottom