The Unreliable Narrator: Witness Recall at the First Seizure Presentation

Abstract

First Seizure History

Noble AJ, Lane S, New P, Cope H, Foley C, Williams HL, Sztriha L, Powell G, Reuber M, Marson AG. How accurate are witnesses of first suspected seizures in recalling semiology at clinically relevant timepoints? A UK experimental study with a pilot intervention. Epilepsia. 2025. doi:10.1111/epi.18624. Epub ahead of print. PMID: 40913485.

Objective: A key diagnostic challenge at “first seizure” clinic appointments is determining whether the reported event was epileptic. Witness accounts are often critical, yet such appointments typically occur weeks after the event. Guidelines recommend review within 2 weeks. Wait times are however often longer, with a median of 7 weeks in countries such as the UK. The accuracy of witness recall at these clinically relevant intervals and whether their confidence predicts accuracy have never been determined. This study addressed these fundamental questions. It also piloted a potential intervention: whether asking witnesses a set of systematic questions immediately after viewing a suspected seizure improves recall at follow-up, compared to the usual free recall approach used by first responders. Methods: In this UK-based experimental study, adults (≥18 years old) viewed a video of an epileptic seizure and were randomized into four conditions: A (immediate free recall + 2-week follow-up), B (immediate free recall + 7-week follow-up), C (immediate systematic questions + 2-week follow-up), and D (immediate systematic questions + 7-week follow-up). The primary outcome was accuracy on 15 standardized questions addressing key semiological features, scored against consensus ratings from five neurologists. Results: Of a representative sample of 304 participants, 295 (97%) fully viewed the video, and 94.7% completed follow-up. At 2 weeks, participants answered 54.4% of questions correctly-only 3.9% (95% confidence interval [CI] = .52-7.3) more than those at 7 weeks. Confidence was poorly correlated with accuracy. Immediate systematic questioning improved later recall by 6.7% (95% CI = 3.3-10.0). A definitive trial of this intervention would require 926 participants. Significance: This is the first evidence on the accuracy of witness recall at clinically relevant intervals. Recall is modest even within recommended timeframes and declines only slightly by 7 weeks. Witness confidence does not predict accuracy. Immediate structured questioning may enhance later recall and thus support seizure diagnoses.

Commentary

A humbling truth about the human condition is that what we experience is not what we remember. ¹ More than a century ago, Hermann Ebbinghaus demonstrated that even faithfully encoded memories decay rapidly, following his now classic forgetting curve. ² Memory's fallibility is particularly relevant in epilepsy care. Despite technological advances, epilepsy remains a clinical diagnosis. The patient's and eyewitnesses’ anamneses are the diagnostic gold standard for a first suspected seizure. A faithful narration of the pre-ictal, ictal, and post-ictal phases is often all we have, and often, all we need.

The first epileptic seizure is a defining moment in a patient's clinical journey. Yet it is also one of the most diagnostically elusive. There is rarely a video to replay when the patient arrives for the first seizure evaluation. Blood biomarkers such as prolactin or emerging proteomic panels have limited or impractical usefulness in distinguishing convulsive seizures from syncope in the emergency department.^3,4 No duration of EEG or advanced neuroimaging can retrospectively clarify the nature of a single transient paroxysmal event. Consequently, clinicians rely heavily on eyewitness accounts to classify the episode. Eyewitness information substantially improves the accuracy of differential diagnosis in a transient loss of consciousness event. ⁵ Yet, nearly one-third of first-time witnesses misidentify convulsive seizures. On immediate recall after watching a seizure video, they identify 60% to 90% of major semiological signs. ⁶ But the immediate recall is practically useless when the patients are seen in the clinic after their first seizure. Therefore, a highly relevant question had remained unexamined until the work by Noble et al: How does the forgetting curve shape the delayed recall of first-seizure witnesses? ⁷ Rather than settling for examining a potential problem in first-seizure diagnostics, they went a step ahead and also piloted an intervention to mitigate this effect through immediate, systematic questioning.

In their rigorous prospective study, 304 UK adults viewed a 97-second video of a focal-to-bilateral tonic–clonic seizure. Participants were randomized into four groups in a 2 × 2 design: they either underwent immediate systematic questioning or did not. The paired groups were then reassessed at 2 or 7 weeks, time intervals chosen to match UK first-seizure referral guidelines and their real-world neurology wait times, respectively. This design helped disentangle the effects of improved recall due to the use of immediate systematic questioning from those of time-dependent memory decay.

Several critical insights emerge. First, even the immediate recall was imperfect. Participants who received immediate systematic questioning answered 62.5% of items correctly. Together with prior findings of Muayqil et al's, there appears to be a ceiling effect on the immediate recall, peaking at roughly two-thirds accuracy even before memory decay begins. ⁶ Emotional arousal, stress, and cognitive overload during medical emergencies may hinder initial encoding, which is likely worse when witnessing a loved one having a first-time seizure. This reminds us that the first-seizure clinic contends not only with diagnostic uncertainty but also with the neurobiology of memory under stress.

Second, as Ebbinghaus predicted, memory decayed rapidly at first and then plateaued. Accuracy dropped to 54.4% in 2 weeks and to 50.5% at 7 weeks. Importantly, Noble et al did not evaluate whether this decline is sufficient to alter diagnostic classification, such as incorrectly labeling a seizure as syncope. So, the clinical threshold for misdiagnosis in relation to memory decay remains untested. Their mitigation strategy, however, was effective. Systematic questioning improved subsequent recall, and the benefit was larger at 7 weeks (10.1%) than at 2 weeks (4.4%). This pattern makes intuitive sense. Early structured questioning strengthens encoding, and the advantage becomes more apparent as weakly encoded details fade disproportionately over time.

Third, recall accuracy varies dramatically by semiologic features. Gross motor phenomena, such as falls and shaking, were recalled with high fidelity (80-95%) at every time point. In contrast, diagnostically valuable but subtler signs, such as ictal eye closure or side-to-side head movements, were correctly recalled less than half the time. This hierarchy of recall reliability has immediate diagnostic implications. Clinicians should weigh highly salient features more heavily and interpret subtle signs with caution when derived solely from delayed witness reports. The risk of potential misdiagnosis is compounded when clinicians equate confidence with accuracy. Noble et al found that witness confidence correlated poorly with correctness. It's a cautionary reminder that conviction is not veracity.

How should the field respond to these limitations in human memory? One obvious step is to capture witness accounts early, before the steep portion of the forgetting curve erodes detail. Structured tools, such as the 15-item questionnaire used by Noble et al, can be administered by emergency medical services (EMS) at the scene, where recall is freshest, to help reduce decay. In electronic form, these data could flow directly into the electronic health record, offering clinicians a higher-fidelity reconstruction of the event. The public health economics case is equally compelling: a 4-minute structured questionnaire may yield substantially more diagnostic value than downstream cascades of EEGs, neuroimaging, or cardiology evaluations undertaken when the initial event history is vague. Few interventions in neurology are this low-tech, low-cost, and potentially high yield. EMS agencies, hospital systems, and professional societies should consider incorporating such tools into the care pathway for transient loss of consciousness events. Public awareness campaigns could also teach bystanders what to observe, what to time, and what not to infer, much as stroke campaigns (eg, FAST) transformed lay recognition of acute cerebrovascular events.^8,9

Finally, the study raises methodological considerations for epilepsy research. Most epidemiologic studies rely on patient or family descriptions to classify seizure types. If these descriptions diverge from reality by 30%–40% within weeks, as Noble et al report, then phenotype labels in such datasets carry built-in uncertainty. Future studies may need to incorporate probabilistic seizure coding or supplement witness narratives with objective data streams. More directly relevant to the themes explored by Noble et al, future research should examine whether witness accuracy varies with demographic or relational factors. Do parents recall pediatric seizures differently from partners observing adult events? Are certain cultural beliefs about seizures associated with specific recall biases? Such analyses could refine interview strategies and patient education.

In summary, Noble et al deliver an elegant and pragmatic study that exposes the fragility of the information underpinning first-seizure diagnosis. Their findings reaffirm that eyewitness testimony, while indispensable, is an imperfect cornerstone. The message for clinicians is clear: capture histories early, use structured tools, triangulate accounts when possible, and interpret subtle features with caution. Until we bridge the gap between human memory and measurable data, clinicians must remain humble interpreters of stories shaped under stress, recognizing that even the most earnest witness may unwittingly be an unreliable narrator.