A Clinical Decision Aid for Diagnosing Transient Loss of Consciousness

Development of a Clinical Decision Aid for the Differential Diagnosis of Transient Loss of Consciousness

BACKGROUND: Transient loss of consciousness (TLOC) - defined as spontaneous disruption of consciousness not due to head trauma and with complete recovery - has a lifetime prevalence of 50%. It is one of the commonest neurological complaints in primary and emergency care. Over 90% of TLOC is due to either syncope, epilepsy or dissociative seizures (DS, also known as 'Psychogenic Nonepileptic Seizures'). The rapid and accurate distinction of these diagnoses is vital to allow appropriate further management but at least 20-30% of patients are not managed optimally or misdiagnosed. We have previously demonstrated that, in patients with established diagnoses of epilepsy, syncope, or DS, an automated classifier using only information from 36 questions based on patient experience and lay witness reports (the initial Paroxysmal Event Profile, iPEP) could accurately diagnose 86.0% of patients (with 100% sensitivity and 91.7% specificity for syncope)

AIMS: To calibrate the iPEP for discrimination between syncope, epilepsy, and DS in patients newly presenting with TLOC, validate its performance in an independent sample, and to explore acceptability of the use of such a tool to people with TLOC and witnesses.

METHODS: Nested qualitative-quantitative prospective single-centre development and validation of the iPEP in patients presenting to Emergency Departments, syncope or epilepsy clinics with first presentations of TLOC, with semi-structured interviews conducted with a purposive sample of participants from the quantitative study. The iPEP will be calibrated using a previously-described procedure for variable selection and training of Random Forest (RF) classifiers, and validated with assessment of overall classification accuracy, alongside sensitivity, specificity, positive and negative predictive values, and area under receiver-operating curve for each of the three target diagnoses. Performance will be evaluated against a benchmark set by results from previous research in patients with established diagnoses of epilepsy, syncope, and DS.

OUTPUTS: Results will be submitted for publication in academic and professional literature. If performance from feasibility can be replicated in validation, the iPEP will be suitable to begin process of registration as a medical device for implementation in clinical pathways to minimise inappropriate referrals and treatment, streamline patient pathways, and enable earlier ordering of appropriate investigations to ensure prompt and appropriate diagnosis and management. If pilot performance could be replicated in this population and proportional savings from current estimated costs of misdiagnosis achieved, this could potentially save £63.9 million of annual UK healthcare expenditure.

Inclusion Criteria:

• Patients first presenting with TLOC
• Referred to secondary care for diagnostic evaluation OR given firm diagnosis of syncope in accordance with European Society of Cardiology guidelines for syncope presentations not requiring further investigation
• Adult over the age of 16 years
• Able to complete iPEP questionnaire independently
• Participants do not need to be native English speaker but do need to have sufficient English language ability to complete iPEP without support

Initial Exclusion Criteria:

• Unable to give informed consent to participation in research
• Unable to complete iPEP independently
• Previous specialist (neurological or cardiological) assessment of TLOC

Criteria for exclusion from analysis:

• No firm clinical diagnosis of TLOC (and its cause) at end of 6-month follow-up period
• Mixed diagnosis of multiple TLOC-causing disorders at end of 6-month follow-up; we exclude these participants from analysis as they would not have a single reference standard diagnosis, thus their inclusion in development may reduce model performance (since their questionnaire answers will describe multiple different kinds of episode), and in validation there is no principled way to assess performance of the prediction model if they have more than one reference diagnosis. A previous study of patients with suspected seizures in this population found that only 1.1% of patients could not be given a single aetiological diagnosis, so we do not anticipate this resulting in many exclusions.4
• Evidence of previous specialist (neurological or cardiological) assessment of TLOC.