Antidepressant and antipsychotic treatment of Psychotic Major Depression in a British mental healthcare setting

Abstract Background Evidence from treatment trials shows that the most effective pharmacological treatment for Psychotic Major Depression (PMD) is combined antidepressant and antipsychotic pharmacotherapy. Aim This study investigates the use of antidepressant and antipsychotic treatment for PMD in clinical practice and examines how treatment profiles correlate with demographic and clinical symptoms. Method Anonymised electronic health records of 2,837 individuals with PMD were followed up for 12-months post-diagnosis in a historic open cohort design. The use of antidepressants and antipsychotics, alone or in combination, were described using frequency statistics. Demographic and clinical characteristics associated with each treatment were assessed using logistic regression analyses. Results Antidepressant and antipsychotic combination pharmacotherapy was the most used treatment for PMD with 69.9% users, compared to antidepressant monotherapy (10.9%) and antipsychotic monotherapy (10.3%). The remaining 8.9% of individuals did not receive antidepressant or antipsychotic treatment. The presence of delusions was strongly associated with the use of antipsychotics, both alone (odds ratio =3.99, 95% confidence intervals = 2.72–5.83, p<.001) and in combination with antidepressants (OR = 2.7, 95% CI = 2.09–3.67, p<.001), rather than antidepressant treatment alone. Conclusions Combined antidepressant and antipsychotic pharmacotherapy is the most common treatment of PMD in clinical practice, showing that clinical practice is in line with evidence from treatment research.


Introduction
Individuals with Psychotic Major Depression (PMD) may experience difficulty in receiving optimal treatment that adequately targets both psychotic and depressive symptoms (Rothschild, 2013). Inconsistencies in international treatment guidelines and a limited evidence base in the past may have contributed to poor clinician understanding of appropriate pharmacological treatment of PMD (Heslin & Young, 2018). Challenges in the treatment of PMD are reflected in the poor long-term prognosis (Coryell, 1998) and more frequent hospitalisations (Coryell et al., 1984;Proctor et al., 2004) associated with the disorder.
In the United Kingdom (UK), antidepressants are recommended as the first line of treatment by the National Institute of Clinical Excellence (NICE) 2009 guidelines, whereas antipsychotic use is discretionary as clinicians are advised to "consider adding antipsychotic medication" to the treatment plan (National Institute for Clinical Excellence, 2009). However, evidence from randomised controlled trials suggests that the recommendations by guidelines may be trailing behind research findings (Leadholm et al., 2013). Several trials, published after the guidelines, provide evidence that a combination of antidepressant and antipsychotic medication is more effective than antidepressants or antipsychotics alone (Farahani & Correll, 2012;Meyers et al., 2009;Wijkstra et al., 2010Wijkstra et al., , 2015, whilst other evidence supports the use of lithium or ECT (Birkenh€ ager et al., 2009;Van Diermen et al., 2018). Due to the gap in conclusions between current evidence from treatment research and recommendations by clinical guidelines, investigation of PMD treatments used routinely in clinical settings is required to determine whether clinical practices are up to date with the research evidence.
Large-scale investigations of PMD treatments using patient data are rare. An evaluation of patient data in a small sample in the United States over 20 years ago found a low rate of optimal pharmaceutical treatment of PMD, with only 5% of patients receiving an adequate combination of an antidepressant and antipsychotic (Mulsant et al., 1997). This finding was replicated in a subsequent study of 100 patients, 10 years later (Andreescu et al., 2007). More recently, an observational study reported high overall use of both antidepressants and antipsychotics in a British mental health service (Crebbin et al., 2008), but treatment profiles were not followed to show whether antidepressants and antipsychotics were taken alone or in combination. A Danish register-based cohort study found that use of antidepressant and antipsychotic combination treatment was associated with psychotic bipolar depression (Bjørklund et al., 2017). However, the use of combination pharmacotherapy for the treatment of unipolar psychotic depression in clinical practice has not been examined.
This study utilised electronic health records data from the South London and Maudsley NHS trust, a large mental healthcare provider with over 1 million users, to: 1) determine the use of antidepressant monotherapy, antipsychotic monotherapy, and antidepressant and antipsychotic combination therapy, in the treatment of PMD; and 2) investigate the clinical and demographic factors associated with each treatment profile. We hypothesised that evidence of psychotic symptoms would predict the use of antipsychotic medication in patients diagnosed with PMD (National Institute for Clinical Excellence, 2009).

Setting
In this historic open cohort design, data were extracted from anonymised electronic health records of the South London and Maudsley NHS Foundation Trust (SLaM) which provides specialist mental healthcare to approximately 1.3 million residents in four London boroughs (Southwark, Lambeth, Lewisham and Croydon). These data are available from the SLaM Biomedical Research Centre (BRC) case register using the Clinical Record Interactive Search (CRIS) system (Stewart et al., 2009). Approximately 300,000 patients who have received SLaM care since 1 st January 2007 are represented in the database (Perera et al., 2016), which has received research ethics approval as a data resource for secondary use (Oxford REC C, reference 08/H0606/71þ5). As patient information is anonymised in the database, individual consent was not required. However, patients within SLaM can opt out of their data being used for research purposes.

Sample
Individuals aged at least 18 years old who received a diagnosis of PMD in SLaM between 1 st January 2007 and 16 th June 2016 were included in the sample. Diagnostic information was available within structured and free-text fields, the latter including clinical correspondence. To extract diagnoses from structured fields, International Classification of Diseases 10 th Revision (ICD-10) (World Health Organization, 1993) codes F32.3, for a single episode, and F33.3 for recurrent depressive episodes with psychotic features were used. Diagnoses in free-text fields were identified using a natural language processing (NLP) application developed using the Generalised Architecture for Text Engineering (GATE) platform. The performance of all NLP applications used is discussed elsewhere (Stewart et al., 2009). To avoid bias due to misclassification, the inclusion criteria specified that the diagnosis must be prolonged for at least one year; cases were excluded if the diagnosis of PMD changed to a diagnosis of schizophrenia (F20) or bipolar disorder (F31) in structured fields within 12 months of the PMD diagnosis index date.

Measures
Structured data were extracted from source fields on age at the time of first PMD diagnosis, sex, ethnicity, and the neighbourhood (Lower Super Output Area; LSOA) code for the person's address closest to the time of their first PMD diagnosis. Use of antidepressants, antipsychotics or combined treatment prior to the diagnosis date was also extracted using an NLP algorithm designed to ascertain pharmacotherapy from a gazetteer of all generic medications in UK use.
Estimates of area-level deprivation were calculated using indices of national deprivation (Noble et al., 2008) (from UK census data) applied at LSOA level (containing an average of 1500 residents per unit) (Office of National Statistics, 2011). The index of multiple deprivation (IMD) score was calculated using the aggregate score of seven domains. A higher IMD score indicated higher estimated neighbourhood-level deprivation.

Outcome variable
The outcome variable was the type of pharmacotherapy received in the patient's records in the 12-month post-diagnosis follow-up period, using the NLP algorithm described above. Antidepressant and antipsychotic combination use was noted if both agents were recorded as being received within the same clinical document. The pharmacological treatment outcome variable was formed, with four categories: antidepressant monotherapy, antipsychotic monotherapy, antidepressant and antipsychotic combination treatment and 'none', that is, no evidence of antidepressant or antipsychotic use. There was a hierarchy in place so that the use of antidepressants and antipsychotics in combination at any time within the 12-month follow-up period was classified as combination treatment, even if preceded or succeeded by monotherapy.

Clinical characteristics
Established NLP algorithms were used to determine the presence of seven major depressive symptoms (low mood, irritability, disturbed sleep, changes in energy, appetite changes, poor motivation and suicide attempts) and two key psychotic features (delusions and hallucinations) in free text fields (Jackson et al., 2017). These algorithms were applied to text fields dated from 1 month prior to the first PMD diagnosis date to 12-months after. As clinical severity scales were not routinely available, overall severity of the disorder was estimated by extracting the total number of days spent in inpatient care during the 12-month period post-diagnosis (Keefler et al., 2001).

Analysis
The use of each pharmacological treatment was first described using frequencies and percentages. Multinomial logistic regression models were used to examine whether groups with different pharmacological treatment profiles differed by demographic and clinical characteristics. Antidepressant monotherapy was chosen as the reference category for the outcome variable, as it is the first-line treatment recommended by UK clinical guidelines (National Institute for Clinical Excellence, 2009). Odds ratios and confidence intervals were used to describe the associations. Adjusted multinomial logistic regression models were used to examine the effect of each clinical characteristic on the pharmacological treatment received, independent of demographic variables (age, sex, ethnicity and deprivation) and other clinical characteristics.

Sample characteristics
A total of 2,923 patients with a diagnosis of PMD who presented to SLaM services between 1 st January 2007 and 16 th June 2016 were identified, 86 (2.9%) of whom were excluded during data cleaning and validation, resulting in an analysed sample of 2,837. This equals to a PMD point prevalence of 0.95% in this sample. Data on most variables was available for all 2,837 cases, whereas ethnicity and deprivation had 2.1% and 9.6% missing data respectively. Table 1 summarises the demographic and clinical characteristics of the sample. The age at diagnosis ranged from 18 to 95 years and was normally distributed. The sample consisted of more females than males. In terms of ethnicity, approximately half of the sample population was white. The average index of multiple deprivation score in the sample was 24.5. The number of days spent in inpatient care ranged from 0 to 342.

Pharmacotherapy profiles
Frequency and percentages of pharmacological use are reported in Table 2. Antidepressant and antipsychotic combination treatment was the most used treatment of PMD by 69.9% of the sample. Antidepressant monotherapy and antipsychotic monotherapy were used by 10.9% and 10.3% of the sample respectively. 8.9% of patients did not receive either antidepressant or antipsychotic treatment.
A full breakdown of frequencies and percentages of demographic and clinical characteristics associated with each treatment are available in Supplementary Appendix 2. Considering demographic associations of statistical significance (p <.05), there were no sex differences between pharmacological regimes. Compared to the antidepressant monotherapy reference group, antipsychotic monotherapy was associated with younger age (odds ratio (OR) per year increment 0.98; 95% confidence interval (CI) 0.96-0.99), no use of antidepressant or antipsychotic medication was associated with lower neighbourhood deprivation (OR per IMD score increment 0.97; 0.96-0.98), and Asian patients were more likely to receive antidepressant and antipsychotic combination pharmacotherapy rather than treatment with an antidepressant alone (OR 4.19; 1.59-11.03).

Discussion
This study investigated antidepressant and antipsychotic treatment of PMD using a large secondary mental healthcare database in the United Kingdom. In what we believe to be the largest study of PMD treatment data to date, antidepressant and antipsychotic combination pharmacotherapy was the most used regime with over two-thirds of the sample (69.9%) thus treated during the one-year follow-up period. Antidepressant or antipsychotic monotherapy were each used by just over 10% of the sample. 8.9% of the sample had no record of antidepressant or antipsychotic use during the one year follow-up period.
Clinical practice is in line with evidence from treatment efficacy research regarding the use of combination antidepressant and antipsychotic treatment for PMD (Farahani & Correll, 2012;Meyers et al., 2009;Rothschild et al., 2004;Spiker et al., 1985;Wijkstra et al., 2010Wijkstra et al., , 2015, shown in the high proportion of patients who received combined pharmacotherapy. However, one in five patients treated with antidepressant or antipsychotic medication did not receive combination pharmacological treatment, with use of either antidepressant or antipsychotic monotherapy instead. The use of antidepressant monotherapy is not surprising, as it is recommended as the first line of treatment by the most widely used British national guidelines (National Institute for Clinical Excellence, 2009). However, the use of antipsychotic monotherapy to treat one in ten patients within the sample is unexpected, considering that treatment with an antipsychotic alone is not recommended for patients with PMD (National Institute for Clinical Excellence, 2009). Evidence from treatment research recommends that clinicians should avoid using antipsychotic monotherapy due to a high liability for adverse effects (Gardner et al., 2005) and insufficient improvement in depressive symptoms (Farahani & Correll, 2012;Rothschild et al., 2004;Spiker et al., 1985;Wijkstra et al., 2010Wijkstra et al., , 2015. We found that clinical symptoms were significantly associated with the type of treatment received. The recorded presence of delusions correlated with clinicians' decision to use antipsychotics, either alone or in combination with an antidepressant, rather than antidepressant treatment alone. On the other hand, even though hallucinations were recorded in around 30% of cases, their presence was not associated with use of antipsychotic medication in adjusted analyses. Antipsychotic use was positively associated with irritability (when used as a monotherapy) and poor motivation (when used in combination with an antidepressant). Clinical characteristics negatively associated with the use of antipsychotic monotherapy, rather than antidepressant monotherapy, were low mood, disturbed sleep and changes in energy. As classic symptoms of depression, the preference of antidepressant rather than antipsychotic treatment in the presence of these clinical characteristics is logical.
In terms of disorder severity, a high number of inpatient days was associated with combination pharmacotherapy rather than antidepressant monotherapy. This is in line with the stepped care model of care wherein more intensive treatments are required for more severe cases (National Institute for Clinical Excellence, 2009). However, as comorbid psychiatric or medical conditions also contribute to the number of inpatient days, we cannot ascertain whether length of stay is a true reflection of disorder severity. Attempted suicide was negatively associated with  (.97-1.01) 'AD': antidepressant monotherapy; 'AP': antipsychotic monotherapy; 'Combination': antidepressant and antipsychotic combination treatment; 'None': no use of antidepressants or antipsychotics; 'OR': odds ratio; '95% CI': 95% confidence intervals; aOR: odds ratio adjusted for demographic variables (age, sex, deprivation, ethnicity) and all other clinical characteristics. Ã p < .05, ÃÃ p < .01, ÃÃÃ p < .001. antipsychotic treatment both alone or in combination with antidepressant treatment, compared to antidepressant monotherapy. This finding is of interest because antipsychotic medication is frequently used in the management of suicidality in schizophrenia (Hor & Taylor, 2010).
One in 12 patients did not receive either antidepressant or antipsychotic pharmacological treatment. This was associated with reduced likelihood of recorded low mood, disturbed sleep, irritability, appetite changes, or poor motivation. Considering the clinical symptoms associated with receiving no antidepressant or antipsychotic treatment, these individuals may reflect a milder symptom profile and may have received a non-pharmaceutical treatment for PMD.

Strengths and limitations
The current investigation has a number of strengths. To our knowledge, this is the first study to examine the use of common pharmacological treatments of PMD in a large clinical sample. The historical cohort design allowed us to examine the use of treatments and identify clinical characteristics associated with treatment profiles for a one-year observation period. The study was designed so that the end-date of the observation period was a year before the extraction date to allow a potential 12-month follow-up for each case. It has previously been suggested that the diagnosis of PMD can be unstable in clinical settings (Baldwin et al., 2005;Crebbin et al., 2008). Therefore, to ensure diagnostic stability and avoid bias due to misclassification in the sample, we excluded patients whose diagnosis changed to bipolar disorder or schizophrenia within the 12-month follow-up period. The use of prospective clinician-recorded data strengthens the validity of the design, as it eliminates the risk of non-response or recall bias. Finally, the regression models were adjusted to account for the confounding effects of demographic variables, which are likely to account for some of the variability in treatment selection.
The main limitation of this study is that it says little about the efficacy or course of antidepressant and antipsychotic pharmacological treatment for PMD. For instance, it is possible that antipsychotics were prescribed as a last resort, after other treatments had been unsuccessful. In the same rein, we were unable to account for tapering effects whereby if one class of medication was being tapered off while another was starting, this would be counted as use of combination treatment. Another factor of treatment decision not addressed in this study is patient preference. For instance, some patients may be reluctant to take medication and could prefer non-pharmaceutical treatment approaches, which may guide the clinician's treatment selection.
The study also has certain limitations associated with the use of electronic health records and routine administrative data more generally, where information can be inconsistent, incomplete, or even inaccurate. First, the validity of the diagnoses is a potential caveat. Diagnoses in electronic health records are recorded for non-research purposes and can be prone to misclassifications through both diagnostic and administrative errors (Davis et al., 2016). The detection of clinical characteristics within the sample is dependent on clinician's reporting symptoms in free text fields. Furthermore, while the use of NLP algorithms is key to unlocking information in the electronic health record (particularly in mental healthcare where so much contextual information is recorded in text rather than structured fields), there is an inevitable error in the performance of any algorithm. However, manual searching of large-scale electronic health records is time costly, restrictive, and also prone to errors. In general, the algorithms used in our study performed to a high standard. Finally, the findings of the current study are reflective of the patient's contact with SLaM services. Within the open cohort design, patients were in and out of contact during the follow-up period. Some patients (particularly those on the severe end of the spectrum) may have had substantially more contact than others, providing more of an opportunity for the treatment regime to change, and for clinicians to assess their mood and clinical characteristics.

Future directions
Our findings provide a detailed account of current prescribing practices in a UK clinical population of patients receiving routine mental healthcare for PMD. The next suggested step in clinical cohort studies would be to track treatment profiles over time to determine the use of common treatments during the acute versus maintenance phases of the disorder, and investigate the usage of adequate dosage and duration of treatment. For instance, it would be of interest to investigate whether antidepressant and antipsychotic pharmacotherapy is used as the first-line of treatment, or after antidepressant monotherapy has proved to be inefficient. This research avenue is well-warranted because treatment trials rarely include long-term follow-up and clinical guidelines do not adequately distinguish between acute and long-term treatments. Additionally, whilst we focused on clinical characteristics associated with treatment profiles at instance levels, future investigations could dig deeper into the severity and nature of the clinical characteristics associated with PMD treatment selection. Finally, we strongly urge replication of this study in other clinical samples, as both clinical practices and treatment guidelines for PMD have been shown to be heterogeneous.

Disclosure statement
Allan Young declares paid lectures and participation in advisory boards for the following companies: AstraZeneca, Eli Lilly, Janssen, Lundeck, Sunovion, Servier, Livanova, and investigator-initiated studies from AstraZeneca, Eli Lilly, Lundbeck, Wyeth. Robert Stewart declares funding and support in the last 5 years from Roche, Janssen and GSK. All other authors report no conflict of interest.

Data availability statement
The data used in this paper was from the South London and Maudsley (SLaM) Biomedical Research Centre (BRC) case register. Due to privacy protection of health-related data, access is only possible within the BRC premises.