Blinding Failures in ME/CFS Treatment Research
A pervasive methodological problem in ME/CFS treatment research is the absence of adequate blinding. Because the interventions most commonly studied—graded exercise therapy, cognitive behavioural therapy, pacing—are behavioural, participants are typically aware of which treatment they receive. When combined with subjective self-report outcome measures, this creates systematic bias that consistently favours the active intervention (ME/CFS Science 2019a). The mecfsscience.org analysis identifies a pattern across ME/CFS treatment trials: unblinded designs using subjective outcomes report positive results, while the same interventions assessed by objective measures (actigraphy, employment status, fitness testing, welfare claims) show no benefit (ME/CFS Science 2019a). The Cochrane systematic review of exercise therapy for ME/CFS exemplifies this problem: its positive conclusions rest on trials using Oxford criteria (the broadest definition, likely including non-ME/CFS patients), subjective fatigue scales as primary outcomes, and non-blinded designs (ME/CFS Science 2019b). When these methodological limitations are accounted for, the evidence base for exercise therapy in ME/CFS collapses. This is not unique to ME/CFS—unblinded behavioural trials produce inflated effect sizes across medicine (quantified in Section Blinding Failures in ME/CFS Treatment Research)—but it is particularly consequential in ME/CFS because the psychogenic model relied on precisely these trial designs to justify treatments now recognised as harmful.
1 FITNET-NHS: Failure to Replicate
The FITNET-NHS trial (2024) provides a cautionary example of replication failure in ME/CFS treatment research (ME/CFS Science 2024a). The original Dutch FITNET trial reported 63% recovery in paediatric ME/CFS patients receiving internet-based CBT. The UK replication (n=314 children) found:
- Sub-clinical improvement in fatigue scores that did not reach the threshold for clinical significance
- No improvement in quality of life
- The intervention was not cost-effective
- 25% of participants deteriorated during the trial
- The 63% recovery rate from the Dutch trial was not replicated The FITNET-NHS failure follows the pattern of other UK replications of Dutch ME/CFS trials (MAGENTA for adolescents also failed), suggesting that the original positive results may have reflected methodological or cultural factors rather than genuine therapeutic efficacy. However, the failure to replicate does not by itself prove absence of efficacy; it may also reflect differences in patient selection, outcome measures, or implementation fidelity between the Dutch and UK settings. Nonetheless, these replication failures weaken the evidence base for CBT-based approaches in paediatric ME/CFS.
2 The BMJ Long COVID Intervention Review
A 2024 BMJ rapid review of Long COVID interventions was criticised by mecfsscience.org for several methodological shortcomings (ME/CFS Science 2024b): reliance on single-trial evidence for most recommendations, inconsistent handling of imprecision (wide confidence intervals sometimes dismissed, sometimes not), cherry-picking of outcomes favouring intervention, and insufficient weight given to high risk of bias. Given the substantial overlap between Long COVID and ME/CFS, these methodological concerns have direct implications for ME/CFS patients who may be offered interventions based on this review’s conclusions.
3 The NICE Guideline Shift and Its Limitations
The 2021 NICE guideline represented a paradigm shift in institutional recognition of ME/CFS as a biomedical condition (ME/CFS Science 2023). Key changes included designating PEM as the defining feature, removing GET from recommendations, and repositioning CBT as supportive care rather than treatment. However, critical analysis identifies limitations in the new consensus (ME/CFS Science 2023):
- GRADE ceiling problem: Because ME/CFS lacks a diagnostic biomarker and blinding of behavioural interventions is difficult, the highest achievable GRADE rating for most interventions is “moderate”—creating a systematic disadvantage for evidence-based recommendations
- Redefined CBT contradiction: The guideline recommends CBT for “managing symptoms” while explicitly stating it is not a treatment for ME/CFS—a distinction that may be lost in clinical practice
- Unsupported pacing recommendation: Despite recommending pacing as the primary activity management strategy, the evidence base for pacing is also limited; it is recommended primarily because alternatives (GET) are harmful, rather than because pacing has been rigorously validated A detailed analysis of how the NICE committee applied GRADE methodology illuminates why the evidence ceiling is so low (ME/CFS Science 2021a). GRADE downgrades evidence for imprecision, indirectness, and risk of bias—all of which are structurally unavoidable in ME/CFS behavioural intervention trials. This creates a paradox: the more rigorously a committee applies GRADE, the less likely any ME/CFS intervention can receive a strong recommendation. The NICE guideline was not universally welcomed. Flottorp et al. criticised the revised guideline as prioritising ideology over science, arguing that it discarded evidence from randomised controlled trials. The mecfsscience.org rebuttal demonstrated that this critique failed to account for the PACE reanalysis, the Oxford criteria problem, the blinding failures, and the systematic overestimation of effect sizes in unblinded trials with subjective endpoints (ME/CFS Science 2022).
3.1 The MetaBLIND Problem
The MetaBLIND study claimed that ME/CFS patients cannot distinguish active treatment from placebo, potentially undermining the argument that unblinded trials produce inflated effects. However, critical analysis identified substantial methodological problems with this claim (ME/CFS Science 2021b), including inadequate assessment of whether participants actually attempted to guess their allocation, conflation of inability to guess with absence of expectation effects, and failure to account for nocebo responses in the placebo group. The study’s conclusions do not justify abandoning the requirement for blinded designs in ME/CFS trials.
4 Quantifying Blinding Bias: The BRANDO Evidence
The claim that blinding matters is not merely theoretical—it has been quantified across thousands of trials. The BRANDO collaboration (Bias in Randomised and Observational studies) pooled data from seven meta-epidemiological datasets comprising 1,973 RCTs within 234 meta-analyses (Savović et al. 2012). The results are unequivocal: trials lacking double-blinding exaggerate intervention effects by an average of 13% (ratio of odds ratios 0.87). Critically, this exaggeration was driven entirely by trials using subjective outcomes—trials with objective or mortality endpoints showed negligible bias. For subjective-outcome trials specifically, the exaggeration reached approximately 23% (ROR ~0.77). The BRANDO authors concluded: “clinical and policy decisions should not be based on trials in which blinding is not feasible and outcome measures are subjectively assessed.” Hróbjartsson et al. (2014) provided even more direct evidence using a uniquely powerful design: 12 trials (n=3,869) that randomised patients to both blinded and non-blinded sub-groups within the same trial, eliminating between-trial confounders (Hróbjartsson et al. 2014). The results were striking: unblinded patients exaggerated treatment effects by a mean standardised mean difference (SMD) of 0.56 (95% CI 0.41 to 0.71) for patient-reported outcomes. No comparable exaggeration was found for observer-reported outcomes. A 2023 Cochrane-affiliated update confirmed these findings (Pitre et al. 2023). These converging lines of evidence establish a quantitative framework for evaluating any trial that combines unblinded participants with subjective outcomes. In such trials, reported effect sizes must be interpreted against a baseline bias of SMD 0.56 or 13–23% exaggeration before attributing effects to the intervention itself. (See Subjective Outcomes in Unblinded Trials for a full methodological analysis of subjective outcome bias in ME/CFS research.)
5 The 2025–2026 Exercise Meta-Analysis Cluster
Between 2025 and 2026, three new systematic reviews and meta-analyses of exercise therapy for ME/CFS appeared: Wei et al. (2025, 13 RCTs, n=1,305) (Wei et al. 2025), Liao et al. (2025, 20 trials, n=2,831) (Liao et al. 2025), and Zhao et al. (2026, 17 RCTs, n=1,944) (Zhao, Gou, and Zhang 2026). All three share the same structural features: they pool studies using broad diagnostic criteria that do not require PEM, rely exclusively on patient-reported subjective outcomes as primary endpoints, include trials where blinding of participants and therapists is impossible, and—in the case of Zhao and Wei—pool mechanistically distinct interventions (aerobic exercise, yoga, qigong, tai chi) into single analyses. Zhao et al. (2026) is the most detailed and merits close examination. The review reports that exercise significantly improves fatigue (SMD = 0.85), sleep quality, quality of life, depression, and anxiety—all measured by subjective questionnaires. However, exercise produced no significant improvement in cardiopulmonary function or objective functional capacity. The authors themselves applied the Risk of Bias 2.0 tool and found “prevalent high risk of bias across the included studies” (Zhao, Gou, and Zhang 2026). Despite this, they dismissed the bias as representing “inherent challenges of the field rather than fatal flaws that would completely invalidate the clinical value of these trials.” This dismissal is difficult to sustain when placed against the BRANDO evidence (Section Blinding Failures in ME/CFS Treatment Research). Hróbjartsson et al. quantified the bias from patient unblinding at a mean SMD of 0.56 for patient-reported outcomes (Hróbjartsson et al. 2014). Zhao’s reported fatigue effect of SMD 0.85 falls well within the range where unblinding artifact could account for most or all of the observed effect. The pattern of subjective improvement with null objective results is not a “physiological ceiling,” as Zhao et al. propose—it is the predicted signature of blinding bias, documented across 1,973 trials in the BRANDO synthesis (Savović et al. 2012). The more parsimonious explanation is that unblinded patients reported feeling better while their underlying physiology remained unchanged. The diagnostic criteria problem compounds the blinding issue. Zhao’s included studies used criteria including Oxford (Sharpe et al. 1991), Fukuda (Fukuda et al. 1994), and Reeves 2005—none requiring PEM as a mandatory feature. Wormgoor and Rodenburg (2021) demonstrated that treatment effects disappear when only PEM-requiring criteria are used and disappear as outcome measures become more objective (Wormgoor and Rodenburg 2021). Zhao et al. did not perform this sensitivity analysis. The intervention heterogeneity further weakens the conclusions: pooling aerobic exercise protocols with yoga, qigong, and tai chi obscures fundamental differences in physiological demands and PEM risk profiles.
A recurring pattern in the exercise therapy literature for ME/CFS deserves explicit identification: reviews apply formal risk-of-bias tools, find pervasive high risk of bias, and then proceed to endorse the intervention regardless. Both the updated Cochrane review (Larun et al. 2024) and Zhao et al. (2026) (Zhao, Gou, and Zhang 2026) follow this template. This pattern converts a quality-assessment tool from a decision aid into a disclosure ritual: bias is documented but has no consequence for the conclusions drawn. The BRANDO evidence (Section Blinding Failures in ME/CFS Treatment Research) demonstrates quantitatively that the bias these reviews acknowledge is sufficient in magnitude to produce the effect sizes they report.
Vink and Vink-Niese have documented the subjective/objective discrepancy across multiple analyses. Their 2018 re-analysis of the Cochrane exercise review demonstrated that objective measures showed no benefit from GET (Vink and Vink-Niese 2018). Their 2020 analysis established that GET does not restore the ability to work (Vink and Vink-Niese 2020). Most strikingly, their 2022 analysis of the PACE trial data showed that 20% of participants whose physical functioning improved on subjective questionnaires had actually deteriorated on the objective six-minute walk test (Vink and Vink-Niese 2022). The convergence of these findings with the BRANDO quantitative framework produces a clear conclusion: the 2025–2026 exercise meta-analyses do not provide evidence that exercise improves ME/CFS. They provide evidence that unblinded patients report feeling better on questionnaires while their physiology remains unchanged—exactly what the blinding-bias literature predicts. Until exercise trials for ME/CFS include adequate blinding (or at minimum use objective primary endpoints and PEM-requiring diagnostic criteria), meta-analyses of existing trials will continue to produce the same misleading pattern.
5.1 Mechanisms Beyond Classical Blinding Bias
While unblinding is the primary explanation for the subjective/objective split, several additional mechanisms may contribute:
5.2 Quantifying the Iatrogenic Harm Pathway
The qualitative iatrogenic chain documented in Hypothesis The Deconditioning Narrative as Iatrogenic Harm Multiplier can be made quantitative. Each link has estimable transition probabilities: (1) meta-analysis to guideline (Zhao 2026 published; Cochrane and RACGP maintained GET); (2) guideline to prescription (standard UK care until 2021); (3) prescription to PEM trigger (51–74% harm rate (Kindlon 2011) (Stallard et al. 2026)); (4) PEM to severity progression (75% worsened in Long COVID Rhijn-Brouwer et al. (2024); ME/CFS recovery rate <5%). Jurisdictions that adopted NICE 2021 early versus those maintaining GET provide a natural experiment.