FND: Movement FND: Seizures Diagnostic Dogma Clinical Reasoning Research Studies

FND Treatment Trials Failed Their Primary Outcomes. Researchers Shifted the Goalposts.

Author profile image FND Nope Moderator
0 comments

Two major clinical trials investigating treatments for functional neurological disorder (FND) failed to demonstrate statistically significant benefits on their predefined primary outcomes. Instead of treating those results as a serious limitation of the evidence, investigators emphasised favourable secondary outcomes and subsequently questioned whether their original primary measures had been appropriate in the first place.

In a May 2025 article, David Tuller examined this pattern in two major FND treatment trials: CODES, which investigated cognitive behavioural therapy (CBT) for dissociative seizures, and Physio4FMD, which investigated specialist physiotherapy for functional motor disorder. Both trials failed to demonstrate statistically significant differences on their primary outcomes at 12 months. Both subsequently became examples of researchers promoting more favourable secondary findings while questioning the measures they had originally chosen to establish treatment effectiveness.


CODES: When Seizure Reduction Stopped Being the Priority

The CODES trial, published in The Lancet Psychiatry in 2020, enrolled 368 patients with dissociative seizures, also known as psychogenic non-epileptic seizures. Participants received either standardised medical care or CBT alongside standardised medical care. The primary outcome was seizure frequency at 12 months, an obvious measure for a treatment intended to help patients experiencing disabling seizures.

The trial found no statistically significant difference in seizure frequency between the groups. Although the CBT group reported a lower median seizure frequency, the adjusted comparison failed to reach statistical significance (p=0.144, means the difference could reasonably be due to chance, rather than CBT actually working.). Nevertheless, a King's College London press release promoted the trial by emphasising improvements in secondary measures, including psychosocial functioning and perceived improvement. The failure to demonstrate a significant benefit on the primary outcome was not given comparable prominence.

Tuller also highlighted a subsequent publication in which the investigators suggested that seizure reduction should not have been selected as the primary outcome, attributing the decision to the trial funders. This was particularly questionable because the researchers had previously used seizure reduction as the primary outcome in their pilot study and defended its suitability when developing the larger trial. The measure apparently became problematic only after the definitive trial failed to demonstrate the expected benefit.


Physio4FMD: Patients Felt Better, but Physical Functioning Was Not Significantly Better

The Physio4FMD trial, published in The Lancet Neurology in 2024, investigated whether specialist physiotherapy was more effective than usual physiotherapy care for patients with functional motor disorder. Its primary outcome was physical functioning at 12 months, measured using the SF-36 Physical Functioning subscale. The trial found no statistically significant difference between the groups on that measure (p=0.23).

However, one secondary outcome, the Clinical Global Impression of Improvement (CGI-I), produced a more favourable result. Approximately 59% of patients receiving specialist physiotherapy reported improvement, compared with 38% receiving usual care. This became a prominent finding despite the failure of the predefined physical-function outcome.

The distinction matters. The CGI-I asks patients whether they believe their condition has improved or worsened. It does not establish how disabled they remain, whether they can perform specific physical activities, or whether they have recovered. A patient can report feeling better while remaining severely disabled. In fact, Tuller pointed out that both treatment groups remained substantially impaired on the SF-36 physical-function measure at the end of the trial compared with population norms.

Both measures rely on patient reporting, but they measure different things. The SF-36 asks about specific limitations in physical activities, whereas the CGI-I records a general impression of change. A favourable response to the latter cannot simply replace a negative result on the former. Yet this is effectively the direction in which the interpretation of Physio4FMD moved.


Rewriting the Rules After the Results

In a subsequent cost-effectiveness paper, the Physio4FMD investigators described the SF-36 as providing an overly narrow view of physiotherapy's potential benefits. They argued that the CGI-I captured a broader range of improvements, including impacts across patients' lives and potentially costs. This reasoning is difficult to reconcile with the actual measurement involved. A general question about whether someone feels improved does not directly measure financial costs, physical recovery or the specific activities a patient can perform.

The investigators also claimed that consensus recommendations published after the trial was designed had recommended CGI-I as the primary outcome measure for FND intervention trials. Tuller checked the 2020 consensus recommendations and found that this was not what they said. The recommendations identified CGI measures as preferred options for the specific domain of core FND symptoms. For the separate domain of life impact, they recommended the SF-36. They did not declare CGI-I the preferred primary outcome for all FND treatment trials.

This is a substantive misrepresentation of the cited recommendations. It is especially troubling because some Physio4FMD investigators were also co-authors of the consensus paper. Tuller subsequently requested a correction. The problem is not an obscure statistical disagreement: the researchers cited a document as supporting a general recommendation that the document did not actually make.


What Exactly Has Been Demonstrated?

There is an obvious problem with promoting treatments for disabling neurological symptoms using evidence that patients feel improved while the trial's primary measure of symptom frequency or physical functioning fails to demonstrate a statistically significant advantage. The CODES trial was designed to determine whether CBT reduced seizure frequency. Physio4FMD was designed to determine whether specialist physiotherapy improved physical functioning. Neither demonstrated a statistically significant between-group benefit on those primary outcomes at 12 months.

These trials did report positive secondary findings. But secondary outcomes do not retrospectively become primary outcomes because they produce more convenient results. Clinical trials specify their primary outcomes in advance precisely to prevent researchers from selecting whichever result looks most favourable after the data have been collected. Examining numerous secondary measures also increases the opportunity for apparently positive findings, particularly when multiple comparisons are not adequately accounted for.

There is also a basic question of clinical relevance. If patients remain severely physically impaired after treatment, what does a reported improvement on a general rating scale establish about their recovery? How much disability remains? Can they walk further, return to work, perform daily activities or regain independence? A general impression of improvement cannot answer those questions. Presenting it as evidence of broader treatment success without establishing meaningful functional recovery risks giving patients and clinicians a misleading impression of what the intervention actually achieved.

None of this requires speculation about the investigators' motives. The sequence is documented: researchers selected primary outcomes, conducted their trials, obtained statistically non-significant primary results and subsequently placed greater emphasis on favourable secondary outcomes while questioning their original measures. In Physio4FMD, that reinterpretation was accompanied by an inaccurate description of published consensus recommendations.

The appropriate response to an unsuccessful primary outcome is to report it prominently and accurately. If investigators believe they selected the wrong measure, they can explain the mistake and design another trial using a prospectively specified alternative. What they cannot legitimately do is treat a favourable secondary finding as though it retrospectively establishes success on the original research question.

FND researchers do not get to move the goalposts after the trial is over. If the primary outcome fails to demonstrate the expected benefit, that failure belongs in the conclusion, not buried beneath whichever secondary result looks better.


References