Antipsychotic Medication and Risk of Incident Seizure in People with Autism Spectrum Disorder: Analyses with Cohort and Within Individual Study Designs

IQVIA Medical Research Data (IMRD-UK) (formerly known as THIN) was the data source. This is a primary care electronic medical records database from the early 1990s to the present day. It covers approximately 6% of the UK population and has more than10.5 million patients, 3.7 million of them are actively registered patients. Data from IMRD-UK are generalisable to the UK for demographic structures and major condition prevalence (Blak et al., 2011). This database is validated as a source of data for use in pharmacoepidemiological research (Lewis et al., 2007) and has been utilised previously for the study of medication in ASD (Alfageh et al., 2020; Murray et al., 2014).

Ethical approval for this fully anonymised study was obtained from the Scientific Review Committee (SRC), which was established to review research using the IMRD-UK database (ref: 18THIN044).

Sensitivity analyses were applied to examine the validity of the primary analysis. The purpose of these analyses was to investigate the effect of different follow-up periods on the resulting HR. In the sensitivity analyses, the definition of the end of follow-up date was changed to the following: (1) the earliest of: occurrence of the outcome date, death, the patient left the practice or the end date of the study; (2) the earliest of: occurrence of the outcome date, death, the patient left the practice or the end date of the study, 90 days after the first continuous medication exposure (grace period). The grace period was added to allow for the residual effect of the medication or the possibility of persistence administration form a residual supply of medication that had resulted from patient non-adherence.

A semi-parametric SCCS model was applied to estimate the risk by comparing the risk of incident seizure in different risk windows to the baseline period. In this model, the age effect does not need to be pre-specified (Farrington & Whitaker, 2006). Conditional Poisson regression was fitted to estimate the IRR, with 95%CIs.

In SCCS, if the occurrence of the outcome leads to the censoring of the observation, this will fail the assumption. A seizure episode could be serious and lead to death (although this would be a very unlikely event) which, subsequently, would end the observation. Although this would be an exceptionally rare event, we conducted a sensitivity analysis by excluding patients who had died during the study period.

An additional SCCS analysis was carried using a negative control outcome: a negative control outcome is a tool that is commonly applied in observational studies to examine the validity of the causal inferences (Lipsitch et al., 2010). It helps to detect selection and measurement bias in epidemiological studies (Arnold et al., 2016). The concept of this approach relies on looking for an association that cannot plausibly be hypothesised. The negative control outcome must share a common source of correlated measurement error with the true outcome (Arnold et al., 2016). Otitis media is an acute recurrent event; the occurrence of this event has never been linked with the use of antipsychotic medication. Therefore, it has been selected as the negative control outcome to validate the causal interpretations of the antipsychotics and incident seizure SCCS. Patients with ASD who had been exposed to antipsychotics and developed otitis media were included in the analysis. Patients with otitis media records before the observation period were excluded.

During the study period, we identified a total of 16,282 patients with ASD who had received at least one medication prescription of the study medications. Of these, 3560 patients were excluded because of having less than six months screening period, their prescriptions date before the start date or after the end date or having history of epilepsy or seizure. A total of 12,722 patients met the inclusion criteria and were included in the cohort analysis. 1287 patients received psychotropics first then switched to antipsychotics and were included in both groups to have a total of 14,009 observations of patients in the analysis. Figure 2 is a flowchart illustrating the patient selection process. Three thousand nine hundred and twenty-three patients receiving antipsychotic prescriptions were identified and allocated to the exposed group: 10,086 patients were identified as being on other psychotropic medication and they were considered to be the unexposed group. The mean age of the participants at the index date was 25.7 years (SD 14.0) for the exposed group and the mean follow-up was 2.2 years (SD 2.6). For the unexposed group, the mean age was 18.1 years (SD 12.2) and the mean follow-up was 3.0 years (SD 3.4). The ratio of male to female patients in both the exposed and unexposed groups was approximately 3:1. In the PS-weighted model, all covariates were balanced between the two study groups, with SMD less than 0.1. Table 1 lists the crude and weighted baseline clinical characteristics of the exposed and unexposed groups at the index date, with standardised mean difference.

In the primary analysis, the incidence of seizure was 54 per 10,000 person-years (PY) in 3923 patients using antipsychotic medication, 36 per 10,000 PY in 10 086 patients using other psychotropic medication. The PS-weighted HR of the incident seizure was 1.28, 95% CI 0.74–2.19, indicating no evidence of an increased risk of incident seizure associated with antipsychotic exposure compared to other psychotropics in the population with ASD. Sensitivity analyses results were consistent with the primary analysis; the HRs were 1.40, 95% CI 0.85–2.30 and 1.36 (0.72–2.57). Table 2 shows the results of the crude and weighted Cox proportional hazard model.

One hundred and forty-nine patients were included in the SCCS analysis. The overall observation period was nearly 1529 patient-years. Almost 80% of the patients were males, with a mean age of 17.13 years (SD 14.59) at the start of observation. At the commencement of observation, the female patients were younger: the mean age of the females was 12.23 (SD 10.89). The average length of continuous antipsychotic prescriptions was 49 days, ranging from 1 to 2553 days. Table 3 provides details of the patient characteristics and the observation period.

In the primary SCCS analysis, using a semi-parametric model, the IRR of seizure for the first 30 days of antipsychotic exposure was 1.79 (95% CI 0.97–3.30), which indicates no evidence of association between exposure to antipsychotics and increased risk of incident seizure. Two patients died during the study period and were excluded in the sensitivity analysis, the results of the sensitivity analysis were consistent with the primary analysis. During the three defined risk periods of the semi-parametric SCCS analysis for a negative outcome, the IRR indicated no evidence of an association between antipsychotic exposure and increased risk of otitis media. The results of the SCCS analyses are shown in Table 4.

This research found no evidence of an association between antipsychotic treatment and an increased risk of seizure in individuals with ASD. The PS-weighted cohort results found no evidence of increased risk of an incident seizure associated with antipsychotic exposure compared with the use of other psychotropic medication 1.28 (0.74–2.19). The results of the SCCS were consistent with the cohort study. The incidence rate ratio of seizure event was 1.79, 95% CI 0.97–3.30 during the first month of antipsychotic exposure.

Although there are published reports about antipsychotics and associated risk of seizures, most of these are descriptive studies. Therefore, a causal relation between antipsychotics use and development of seizure has not been unequivocally confirmed, particularly in individuals with ASD (Górska et al., 2019; Grover et al., 2015; Hedges et al., 2003a, 2003b; Williams & Park, 2015). It should be noted in the light of evidence that around 50% of the prescribed antipsychotics for individuals with ASD are not approved in the UK for use in this population, which may indicate an off-label medication prescribing (Alfageh et al., 2020).

The likelihood of the association between antipsychotics and seizures has been investigated in patients with schizophrenia, mood disorders and dementia. A nested case–control study conducted in the UK using the Clinical Practice Research Datalink (CPRD) found that the prescription of haloperidol, prochlorperazine or trifluoperazine was associated with an increased risk of seizures: adjusted odds ratio (OR) 2.51, 95% CI 1.51–4.18 compared with non-users (Bloechliger et al., 2015). However, the comparison was between antipsychotic users and non-users without specifying if the comparison group was on other-psychotropic medication or not and considering the study design used, the estimated risk could be inflated (Schuemie et al., 2019). In our study, we didn’t compare the risk of seizure between different antipsychotic medication. A study with data from the national health insurance research database (NHIRD) compared the risk of seizure among first and second-generation antipsychotics in patients diagnosed with schizophrenia and mood disorders (Wu et al., 2016). This study showed no evidence of a higher risk of seizure associated with first-generation antipsychotics than the second generation: HR 1.34, 95% CI 0.99–1.81; p = 0.06 (Wu et al., 2016). When compared to risperidone, clozapine HR 3.06, 95% CI 1.40–6.71; thioridazine HR 2.90, 95% CI 1.65–5.10; chlorprothixene HR 2.60, 95% CI 1.04–6.49 and haloperidol HR 2.34, 95% CI 1.48–3.71 all had a higher risk of antipsychotic-related seizure, while aripiprazole had a potentially lower risk of seizure: HR 0.41, 95% CI 0.17–1.00; p = 0.05 (Wu et al., 2016). However, the results of the previous study could be affected by confounding by indication rather than reflecting the actual effect of the medication on the risk of seizure. In our cohort study, most of the prescriptions of antipsychotics were for second-generation antipsychotics (82.4%) and 45% of the prescriptions were for risperidone. For patients included in the SCCS analysis, a higher percentage of the antipsychotics prescriptions were issued for risperidone (57.6%). In this study, we have not taken account of medication dose. There is considerable evidence from the literature that, for medications that are associated with increased seizure risk, the risk is very much related to medication dose (Górska et al., 2019; Grover et al., 2015; Varma et al., 2011). Other reports have suggested that low-dose antipsychotic medication, as used to treat anxiety and/or behavioural problems in young people with ASD might not be associated with an increased risk of seizures but this leaves the possibility that higher antipsychotic doses, such as those used to treat psychosis or bipolar disorder might be associated with an increased seizure risk.