Article Text

Download PDFPDF

Research paper
Primary and secondary care attendance, anticonvulsant and antidepressant use and psychiatric contact 5–10 years after diagnosis in 188 patients with psychogenic non-epileptic seizures
  1. Roderick Duncan1,
  2. Christopher D Graham2,
  3. Maria Oto3,
  4. Aline Russell4,
  5. Laura McKernan5,
  6. Sue Copstick4
  1. 1Department of Neurology, Christchurch Hospital, Christchurch, New Zealand
  2. 2Department of Psychology, Institute of Psychiatry, King's College London, Guy's Hospital, London, UK
  3. 3Department of Neurology, Southern General Hospital, Glasgow, UK
  4. 4Department of Neuropsychology, Southern General Hospital, Glasgow, UK
  5. 5Department of Clinical Neuropsychology, Southern General Hospital, Glasgow, UK
  1. Correspondence to Dr Roderick Duncan, Department of Neurology, Christchurch Hospital, Private Bag 4710, Christchurch 8140, New Zealand; roderick.duncan{at}


Background and objectives There have been few studies of long-term outcome in psychogenic non-epileptic seizures (PNES), and none of long-term healthcare utilization.

Methods We studied attendance with seizures, healthcare use and employment over a 6-month period from the family doctors of 260 consecutive patients with psychogenic non-epileptic seizures (PNES), 5–10 years after diagnosis.

Results We obtained clinical data in 188/260 patients (72.3%), of whom 60 (31.9%) had attended primary or secondary care with seizures in the previous 6 months. Predictors of attendance with seizures included a diagnosis of epilepsy+PNES (OR 5.7, p=0.009), work status (OR 3.9, p=0.027) and social security payments (OR 6.3, p=0.003). Latency to diagnosis was not predictive. Emergency admission data were available in 187 patients, of whom 25 (13.4%) had emergency hospital attendances. Prescription data were available for 172 patients, of whom 154 had ‘PNES only’. Of these, 17 (11.0%) remained on antiepileptic medication (AED). 68/172 patients (39.5%) were prescribed antidepressant (AD) drugs. We had psychiatric contact data in 185 patients, of whom 49 (26.5%) had accessed psychiatric services in the last 6 months.

Conclusions Surprisingly few of our patients had presented with seizures during the study period. Early reductions in both AED use and healthcare use were sustained long term. Although psychiatric and employment outcomes were less encouraging, some aspects of PNES outcome may be better than previously thought.

View Full Text

Statistics from


Early outcomes in psychogenic non-epileptic seizures (PNES) can be good, with up to 50% of patients becoming attack-free after effective explanation of the diagnosis.1–5 Healthcare use, which is high in this patient group, may lessen substantially,5–7 even in patients whose attacks continue.5 ,7 However, early employment outcome is poor,5 as is outcome for medically unexplained symptoms (MUS) other than PNES.8

In respect of seizures, long-term outcome appears unfavourable. One study of 110 patients with 1–5 year follow-up found that 40% of patients had stopped having PNES9 and the largest published study found that only 28.8% of 164 patients were PNES-free at 1–10 years after diagnosis.10 A study of 93 patients with mean follow-up of 5 years found only 25.4% of patients PNES free.11 Long-term social and employment outcomes are also poor.10 ,12 Long-term healthcare use has not been studied.

Long-term outcomes (of all types) have been historically difficult to determine in patients with PNES as they tend to default clinic appointments.5 ,13 In published studies, the completeness of data has been subject to the patient's willingness either to attend follow-up or to reply to postal or telephone surveys.1 ,5 ,9–11 ,14 ,15 Some studies only include information on patients with follow-up.1 ,11 The biggest and most widely cited was a postal survey10 where data were obtained in 49.8% of patients. A more recent postal study obtained follow-up data in only 27.6% of patients,15 with significant differences between responders and non-responders.

In our previous short-term study of the present cohort, we had some evidence of better outcome in patients who dropped out of neurology follow-up.4 Our early outcome data on 187 patients showed that those who attended two early follow-up appointments did worse than those who attended only one.5 We therefore hypothesised that data gathered on patients who were not self-selected would suggest better outcome and might provide information on a higher proportion of patients.


Between 1999 and 2004, 260 consecutive patients diagnosed at our PNES clinic were included in an outcome study. At 5–10 years postdiagnosis, we located current family doctors and requested information on the patient relating to the most recent 6-month period:

  1. Whether the patient had attended primary or secondary care with seizures of any kind.

  2. Whether the patient attended primary or secondary care with other medical problems, and if so what?

  3. Whether the patient was employed.

To access public healthcare, patients in Scotland must be registered with a family doctor. They can be registered only with one family doctor and have a unique healthcare identifying number. Information regarding presentations to emergency rooms or to secondary care is sent back to the patient's registered family doctor. Family practices have computer systems whose outputs include primary and secondary care consultations, medication, hospital admissions, etc. We offered family doctors the alternative of supplying a printout relating to the previous 6 months. In analysing these printouts, we counted any medical encounter whose note mentioned seizures (other than as a past problem) as ‘attendance with seizures’, whether seizure was the prime reason for the consultation or not. Our outcome variables were attendance with seizures, emergency hospital attendance/admission with seizures, prescribed antiepileptic medication (AED), contact with psychiatric services, prescribed antidepressant drugs, in employment.

We compared these data with data gathered at baseline (box 1).

Box 1

Variables tested for predictive effect on outcome variables (variables marked with an asterisk were tested both at baseline and 6–12 months)

Age at presentation

Age at onset of psychogenic non-epileptic seizures (PNES)

Interval between age at onset and diagnosis (delay to diagnosis)



Drawing social security benefits*

Scottish index of multiple deprivation (

Diagnosis of epilepsy + PNES

Learning disability

On anticonvulsant treatment*

PNES frequency*

Free of PNES*

Emergency presentation with seizures*

Medically unexplained symptoms other than PNES*

Contact with mental health services

Diagnosis of anxiety or depression

History of panic attacks

History of self-harm

Prescribed antidepressant

History of sexual abuse

History of physical abuse

History of other psychological trauma

All our patients were given the diagnosis of PNES according to our protocol.16 During the recruitment period of the study (1999–2004), 116/188 patients (61.7%) were referred for psychological intervention, though we had no information on whether they attended. Patients were not referred if they did not wish it or if they did not believe the diagnosis of PNES or understand its nature. Some patients whose PNES stopped after communication of the diagnosis were not referred.

Statistical analysis

Statistical analysis was carried out using SPSS V.19. In comparing baseline and 6–12 month variables between patients in whom we had data with those in whom we did not, and to compare baseline with 5010 year variables, we used the Mann–Whitney U test for continuous variables and the χ2 test for categorical variables. Simultaneous binary logistic regression models were used to evaluate the ability of independent variables to predict outcomes. All baseline and 6–12 month variables from our previous publication5 were tested for predictive effect on 5–10 year variables.

We used binary logistic regression to derive a predictive model for each outcome variable, as follows. First, each independent variable was tested for correlation with the outcome variable. Independent variables correlating with the outcome variable at the 10% level or more (p ≤ 0.10) were rejected at this stage. The remaining independent variables were tested for colinearity, and if two variables covaried at the 5% level (p ≤ 0.05) or less, the variable correlating less significantly with the dependent variable was eliminated.

The remaining independent variables were entered into an initial model. Those that were without predictive value at the 5% level (p ≤ 0.05) were then eliminated, and the model recalculated, until only significant predictors remained.

Ethical permission

Baseline and 6–12 month data were acquired with the approval of the Research Ethics Committee of the Southern General Hospital, Glasgow. Outcome data were acquired as part of a service audit of the West of Scotland PNES clinic. The present article has been reviewed by the Research Ethics Scientific Officer for the West of Scotland Research Ethics Committee, who has approved our intention to publish it. The study had no external funding.



We obtained responses for 220/260 patients (84.6%). The responses indicated that 15 had moved and were not registered with another practice in Scotland. In total, 17 of those 260 patients had died (cause of death in these patients has been published separately—the deaths were unrelated to epileptic seizures or PNES17). Thus, we had clinical outcome information in 188/260 patients (72.3%). In 172/188 patients, this took the form of a complete printout of the family doctor record, including the notes relating to each consultation.

Baseline data

Baseline data indicated that 142/188 of our patients (75.5%) were women: 21/188 (11.2%) had additional epilepsy (ES+PNES), and 12/188 (6.4%) had mild learning disability. Mean age at onset of PNES was 30.5±13.7 years, mean age at presentation to the clinic was 37.2±13.3 years and mean diagnostic delay was 6.7±7.7 years. In respect of these, our other baseline variables (box 1) and our 6–12 month outcome variables,5 there were no significant differences between patients for whom we had a response and those for whom we did not. The follow-up interval ranged from 5.4 to 10.7 years (mean 8.7±1.3 years).

Outcomes at 5–10 years and their predictors

Baseline and 6–12 month variables tested for predictive value are listed in box 1. Variables that appear in box 1 but not in table 1 were found to have no independent predictive value on outcome variables. Summary baseline and 5–10 year postdiagnosis outcome data are shown in table 2, and predictive models are shown in table 1. We did not obtain data for all variables in all patients. The denominator (total n) varies accordingly.

Table 1

Binary logistic regression models predicting attendance with seizures, emergency hospital attendance, AED prescription and employment at 5–10 years postdiagnosis of PNES

Table 2

Outcomes over a 6-month period at 5–10 years from diagnosis in 188 patients with PNES: comparison with baseline values

Attendance at primary care or hospital for seizures

At baseline, all patients had attended with seizures in the previous 6 months. At 5–10 years, only 60/188 patients (31.9%) had attended with seizures over the same period (p<0.001).

Excluding those 21 patients with a diagnosis of both epilepsy and PNES at baseline, 46/167 patients with a diagnosis of ‘PNES only’ (26.1%) had attended with seizures at 5–10 years.

Patients with a baseline diagnosis of epilepsy+PNES were 5.7 times more likely to attend with seizures at 5–10 years (p=0.008), and patients receiving social security payments were 6.3 times more likely to do so (p=0.003). Substituted in the model, being out of work at 5–10 years was also predictive (p=0.027). Patients who had emergency admissions for seizures at baseline were more likely to be attending with them at 5–10 years (p=0.010). A history of health-related psychological trauma at baseline marginally predicted seizure attendance at 5–10 years (p=0.049). The model explained 31.3% of variance. Latency to diagnosis was non-predictive either as a single factor or as part of any of the models tested.

Emergency hospital attendances

Some emergency admissions bypassed the emergency department, so emergency hospital attendance data include direct admissions to hospital wards. This was available in 187 patients, of whom 25 (13.4%) had attended as emergencies in the previous 6 months, compared with 84/188 (44.7%) at baseline (p<0.001).

Patients with a history of panic attacks at baseline were 4.7 times more likely to have emergency hospital attendances at 5–10 years (p=0.005). Those with emergency hospital attendances at 6–12 months were 7.4 times more likely to have them at 5–10 years (p=0.009). Patients on AED at baseline were 6.3 times more likely to have hospital admissions at 5–10 years (p=0.007). The model explained 25.6% of variance.

Prescription of AED

We had 5–10 year prescribing information in 172 patients, of whom 154 had a baseline diagnosis of ‘PNES only’. Prescriptions included AED in only 17/154 (11.0%) compared with 82/167 patients (49.1%) with ‘PNES only’ at baseline (p<0.001). Forty-three patients with PNES only had presented with seizures, and 13/43 (30.2%) of these were prescribed AED.

Unsurprisingly, patients with a baseline diagnosis of ES+PNES were 16.3 times more likely to be prescribed AED at 5–10 years (p<0.001). Patients prescribed AED at baseline (independently of diagnosis) were 2.9 times more likely to be prescribed them at 5–10 years. A baseline history of health-related psychological trauma was also a predictor of AED prescription at 5–10 years (p=0.020). This model explained 32.9% of variance.

Psychiatric contact and AD prescription

We had information on psychiatric contact in 185 patients, of whom 49 (26.5%) had contact with psychiatric services at 5–10 years, compared with 43/188 (22.9%) at baseline (p=0.418).

Of the 172 patients in whom we had prescribing information, 68 (39.5%) were on AD at 5–10 years compared with 68/172 (39.5%) at 5–10 years, a significant increase (p=0.001).

Psychiatric contact and AD prescription covaried, unsurprisingly. Both were predicted by a baseline history of self-harm (p=0.001 in the case of psychiatric contact (p<0.001 in the case of AD). The models explained 8.9% and 12.1% of variance, respectively.


We had employment data in only 114 patients. At 5–10 years, 26/114 (22.8%) were in paid employment compared with 29/188 (15.4%) at baseline (p=0.107). In the employment data, denominators are adjusted to exclude patients below 16 years and above retirement age at the time of follow-up.

Unsurprisingly, patients who were employed at baseline were 6.5 times more likely to be employed at 5–10 years (p<0.001). Employment at 6–12 months also predicted employment at 5–10 years (p=0.005).


Our data show that 128/188 of our patients (68.1%) had not accessed any kind of healthcare for seizures over a period of 6 months, 5–10 years after diagnosis. Indeed, in those 172 patients in whom we had printouts, these included the note relating to each consultation. We counted the consultation as an ‘attendance with seizures’ if seizures were mentioned at all, not just if seizures appeared to be the main reason for the consultation. While this suggests a better long-term outcome than in some previous studies,9–12 ,15 variations in methodology may explain some or all of the differences. Our study defined a specific endpoint time frame (6 months), whereas in some previous studies10 ,11 seizure-free duration is not defined. It is possible that some of our patients had seizure presentations outside the 6-month period and would not have reported themselves as being ‘seizure-free’ despite appearing so in our data. It is also possible that a proportion of our patients had seizures during the 6-month period but chose not to present with them, to the family doctor, to the emergency room or to secondary care. Our previous study showed that some patients with persisting seizures at 6–12 months postdiagnosis had ceased to present to emergency rooms with them,5 which may over time have applied to other aspects of healthcare. This requires further study.

The relatively ‘closed’ nature of the Scottish health service may have influenced some of our outcome measures. The family doctor is the focal point of this system, is the ‘gatekeeper’ for access to secondary care and is the destination for all clinical letters even from the limited private sector. There are effectively no private family doctors. Combined with a unique healthcare identifier number, this system makes it easier to monitor patients’ diagnoses and medication, and to maintain good contact with treating doctors. In our clinical service, we placed high priority on good communication of the diagnosis, to patients, relatives and other doctors, and on maintaining good links with family doctors. It seems intuitively likely that these factors played a significant role in ‘containing’ healthcare use and in ensuring that a high proportion of patients remained off AED (in the present study only 11.0% of patients with ‘PNES only’ remained on AED compared with up to 40.7% in comparable studies10 ,15). In the present study, AED prescription at baseline predicted attendance with seizures at 5–10 years, independently of a diagnosis of epilepsy+PNES: there is evidence that coming off AED may have a beneficial effect on some outcomes in PNES.13

The present study included only 11.2% of patients with a dual diagnosis of epilepsy and PNES compared with 40.2% in the other comparable study.10 Additional epilepsy was associated with persisting seizures in both studies. It is not clear to what extent poorer PNES outcome in patients with both epilepsy and PNES might be due to a confounding effect of epileptic seizures being counted as PNES: we did not consider it feasible to distinguish between epileptic seizure frequency and PNES frequency in the present study, and the ability of a postal or telephone study to do so might be similarly limited. Another study15 found poorer outcome with a low prevalence of epilepsy, but in that study data were obtained in only 26.7% of patients and there were significant differences between responders and non-responders.

There was a modest and non-significant increase in contact with psychiatric services over the period of the present study, whereas AD prescription increased significantly. While it is possible that the latter reflected worsening mental health, the data do suggest an expansion of AD use by family doctors over the period of the study. Overall, our data on psychiatric contact and AD prescription suggest that a substantial minority of patients retain long-term psychiatric comorbidity.

We previously studied predictors of outcome in those patients from the present cohort (187/260) who attended for 6–12 month reviews.5 The predictors we found at 5–10 years were somewhat different, possibly because of the difference in the nature of the outcome measures. A notable exception to this were employment outcome and social security payments, which were predictive, in keeping with previous data.4 ,5 ,10 ,18 ,19 Somewhat counter-intuitively, most published evidence argues against a predictive effect of diagnostic latency5 ,10 ,11 ,19 on seizure outcome. The present data showed no predictive effect on attendance with seizures.

Patients with a baseline history of health-related psychological trauma (ie, traumatic experiences related to physical illness) are prevalent in late-onset PNES.20 ,21 Our present data suggest that these patients continue to present with seizures long term. However, we found no predictive effect of any other type of psychological trauma (eg, sexual abuse) on long-term presentation to primary or secondary care with seizures, though intriguingly self-harm (which covaried in the present dataset with sexual abuse, unsurprisingly22–24) predicted psychiatric contact and AD use at 5–10 years. In our 6–12 month follow-up data,5 both sexual abuse and anxiety or depression predicted poor outcome, a finding in keeping with other shorter-term data.25 This apparent discrepancy may be due to the difference in the way we collected our data (eg, if substantial numbers of patients still had seizures but were not accessing any medical care for them), but might also suggest that predictors of short-term and long-term outcome in PNES differ. The latter would not be entirely surprising as short-term outcomes may be dominated by patients who respond quickly to being given the diagnosis, whereas many other factors, including the effects of psychological intervention, may be important in the longer term. We had no access to psychological data, so we could not analyse for this in our cohort. Emergency hospital attendances at 5–10 years were predicted by emergency attendances both at baseline and at 6–12 months, reflecting the fact that a core group of patients had emergency attendances at each time point. The predictive effect of panic attacks on attendance with seizures that we found suggests a relationship with anxiety levels.25

We had employment data on relatively few patients, but those we had suggested that outcomes were, as expected, poor. Psychiatric contact and AD prescription remained long-term prevalent in our cohort. However, our data also suggest that good healthcare use outcomes obtained early can be maintained or even improved on in the longer term.

Our family doctor data included, in most cases, printout records of consultations and discharge documentation from hospital and constituted a complete healthcare record for the study period. We had these complete data in only 172/188 patients, however. The main disadvantage of our data is that it provides no information from the patient, and it is not clear to what degree a measure such as ‘presentation with seizures’ might correlate with patient history, however obtained. This limits comparisons with previous data. The sample studied, while large and consecutive, was not population based and is therefore liable to referral bias. We obtained data on a comparatively high proportion of patients, but not all. While this was not primarily patient determined, some other types of bias may have occurred. Our outcome dataset was acquired as part of an audit, the terms of approval for which restricted our dataset to a small number of variables and did not allow us access to psychological data.


View Abstract


  • Contributors RD was responsible for study design and supervision, and data analysis. CDG and LMcK were responsible for data collection and analysis, and critical review of manuscript for intellectual content. MO was responsible for study design and management, data collection and critical review of manuscript for intellectual content. AR and SC were responsible for study design and critical review of manuscript for intellectual content.

  • Competing interests During the period of this study, RD has received honoraria for speaking at academic meetings and has received support for travel to academic meetings from UCB Pharma.

  • Ethics approval Southern General Hospital Research Ethic Committee.

  • Provenance and peer review Not commissioned; externally peer reviewed.

Request Permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.