An exploratory database study of factors influencing the continuation of brexpiprazole treatment (prescription) in patients with schizophrenia using information from psychiatric electronic medical records processed with natural language processing

Using natural language processing (NLP) technology to analyze and organize textual information in psychiatric electronic medical records can identify undiscovered factors associated with treatment discontinuation. This study aimed to evaluate brexpiprazole treatment continuation rate and factors affecting brexpiprazole discontinuation using a database that employs the MENTAT® system with NLP technology. This retrospective observational study evaluated patients with schizophrenia who were newly initiated on brexpiprazole (April 18, 2018-May 15, 2020). The first prescriptions of brexpiprazole were followed up for 180 days. Factors associated with brexpiprazole discontinuation were assessed using structured and unstructured patient data (April 18, 2017-December 31, 2020). The analysis population comprised 515 patients; mean (standard deviation) age of patients was 48.0 (15.3) years, and 47.8 % were male. Using Kaplan-Meier analysis, the cumulative brexpiprazole continuation rate at 180 days was 29 % (estimate: 0.29; 95 % confidence interval, 0.25-0.33). Univariate Cox proportional hazards analysis identified 16 variables independently associated with brexpiprazole discontinuation. Multivariate analysis identified eight variables associated with treatment discontinuation: variables with hazard ratio <1 were the presence of physical complications, longer hospitalization duration, and maximum chlorpromazine-equivalent dose of antipsychotics of >200 to ≤400 mg/day vs ≤200 mg/day in the past year; variables with hazard ratio >1 were previous electroconvulsive therapy, availability of key contact person information, a history of crime committed/reported, increase in brexpiprazole dose to 2 mg in >28 days, and appearance/worsening of symptoms other than positive symptoms. In conclusion, we identified potential new factors that may be associated with brexpiprazole discontinuation, which may improve the treatment strategy and continuation rate in patients with schizophrenia.


Introduction
In patients with schizophrenia, recurrence or relapse impairs patient functioning; therefore, it is important to prevent relapse in schizophrenia (Emsley et al., 2013;Olivares et al., 2013). Continued use of antipsychotic drugs is important to prevent relapse (Japanese Society of Neuropsychopharmacology, 2021); however, the treatment continuation rate in patients with schizophrenia is poor, with a high discontinuation rate observed during the initial stages of treatment (Hatta et al., 2009;Ishigaki, 2013;Lieberman et al., 2005;Liu-Seifert et al., 2005;Sumiyoshi, 2012;Zhao et al., 2011;Zhao et al., 2010).
In Japan, the recommended titration schedule for adult dosage of brexpiprazole by oral administration is to start with 1 mg/day for at least 4 days, followed by an increase to 2 mg/day once daily (Higuchi et al., 2018;Otsuka Pharmaceutical, 2021). Although brexpiprazole has an excellent safety and tolerability profile, the treatment discontinuation rate remains insufficient, reported at 43.4 % at 16 weeks and 61.4 % at 52 weeks (Inoue et al., 2021;Yoshimura et al., 2020). Two previous studies have reported on the factors affecting brexpiprazole continuation, but only a limited number of factors were investigated due to limitations in methodology (Inoue et al., 2021;Yoshimura et al., 2020).
Existing databases (Kimura et al., 2012) allow for the evaluation of limited factors that can be associated with treatment discontinuation based on structured information from electronic medical records (EMRs), making it challenging to obtain new knowledge. Of note, psychiatric EMRs contain information such as medical history and symptoms that is difficult to evaluate as >90 % of it is accumulated as textual data written in natural language/free format. Therefore, using a database that employs the MENTAT® system, which analyzes and organizes the textual information by natural language processing (NLP) technology, can enable the evaluation of factors that have not been examined (Otsuka Digital Health, 2016).
Understanding new factors associated with treatment discontinuation can help healthcare providers (HCPs)/physicians decide on a treatment strategy with improved treatment adherence/continuation and reduce the burden on patients. This study aims to further explore the factors affecting brexpiprazole discontinuation within 180 days of newly initiating brexpiprazole using a database that employs the MENTAT® system.

Study design and source of data
This retrospective observational study evaluated factors associated with brexpiprazole discontinuation in patients with schizophrenia who were newly initiated on brexpiprazole between April 18, 2018, and May 15, 2020, from EMRs using an artificial intelligence (AI)-based MEN-TAT® system (Otsuka Digital Health, 2016). The date of the first prescription of brexpiprazole was defined as the index date, and patient prescriptions were followed up for 180 days to assess treatment discontinuation. Factors associated with brexpiprazole discontinuation were assessed using structured patient data, defined as the evaluable information entered in EMRs, and unstructured patient data, defined as the nonevaluable textual data written in natural language in EMRs. The data extraction period was between April 18, 2017, and December 31, 2020. This study was approved by the Research Ethics Committee of the Research Division of Otsuka Pharmaceutical Co., Ltd.

Patients
Patients aged ≥18 years on the date of their first brexpiprazole prescription (index date) between April 18, 2018, and May 15, 2020, for schizophrenia diagnosed per the diagnosis code F20 of the International Classification of Disease, Tenth Revision (ICD-10) were included. Patients with comorbid mood (affective) disorders, dementia, psychological development disorders, or hyperactivity disorder on the index date were excluded.
Patients treated even once with a dose of >2 mg/day or <1 mg/day of brexpiprazole-not within the approved dose range in Japan of 1-2 mg/day-were also excluded.

MENTAT® system and NLP
Otsuka Digital Health's MENTAT® is a digital health solution utilizing IBM's Watson Explorer and time-series analysis techniques, which are cognitive search and analysis platforms that derive knowledge from structured and unstructured data using advanced NLP techniques (Corporation, 2015;Otsuka Digital Health, 2016). This AI-based MEN-TAT® system was jointly developed by Otsuka Pharmaceutical Co., Ltd. and IBM and specializes in psychiatric EMRs. Structured data include basic patient information, additional patient information, patient disease information, outpatient history, basic inpatient information, basic drug history, basic drug information, and behavioral restriction information. In the clinical setting, MENTAT® can structure textual data (unstructured data) written in natural language, which are available in medical records, for evaluation. Unstructured data include information useful to patients with mental illnesses, such as psychiatric symptoms, comorbid symptoms, adverse events, growth history, family environment, residential environment, living function, and social functioning. In psychiatric care, it is believed that >90 % of significant patient information is accumulated in textual form (Otsuka Digital Health, 2016).
Data entered in EMRs were anonymized and transmitted via a dedicated server to the cloud for MENTAT®. NLP technology analyzes the meaning of textual data entered in EMRs, without any change or processing and without placing additional burdens on HCPs, such as filling out new forms. The information entered in EMRs requires conversion of free-text narratives documented using everyday language to a structured data form that can be analyzed. Unlike other languages, the Japanese language uses three types of characters-Kanji, Hiragana, and Katakana-that are often mingled with alphabetic characters in sentences. In addition, EMRs in psychiatry use words spoken by patients, making it difficult to analyze these free-text narratives. Text mining in MENTAT® refers to a function by which the software understands the meaning of free-form text entered in an EMR, extracts the clinically important factors, and organizes information contained in the EMR. Further details on the MENTAT® system are provided in Supplementary  Table 1.

Outcomes
Treatment discontinuation event was defined as the absence of brexpiprazole prescription for >30 consecutive days during the 180-day follow-up after the index date prescription. This evaluation criterion extended beyond the 180-day follow-up for patients with no brexpiprazole prescription between days 150 and 180 of the follow-up. Time to treatment discontinuation was calculated as the number of days from the index date to the last day of treatment immediately before treatment discontinuation.
Patients without brexpiprazole discontinuation for >30 consecutive days during the 180-day follow-up after the index date prescription were censored. These patients were considered to have continued treatment, and the number of days to the last day of treatment was defined as the treatment duration (time to censoring). For transfers and deaths, the treatment duration was defined as the time to transfer and the date of the last dose until death. In addition, patients who were able to continue treatment for >180 days were censored after 180 days.
A total of 55 variables from structured and unstructured databases were evaluated during the study period as factors for brexpiprazole discontinuation.
Structured (26 variables) and unstructured (29 variables) data were extracted and evaluated at baseline (index date) and during the 12 months before the index date; some variables were also evaluated during the 180-day follow-up.
Data on the period from the first occurrence of positive symptoms to the index date, the period from the date of the first positive symptom to the date of the first medical examination, the highest educational qualification, work experience, intellectual disability, a history of crime committed or reported, and clozapine use were traced back in the MENTAT® system beyond the study period.

Variable
Structured evaluation variables are defined in Supplementary  Table 2 and included age, sex, physical complication/comorbidity, psychiatric comorbidities other than schizophrenia, height, body weight, body mass index, weight gain, type of hospitalization, inpatient/ outpatient status at initiation, number of hospital admissions, hospitalization duration (days), days from previous discharge to hospital admission, intervals between hospital visits, experience of clozapine use, experience of long-acting injectable use, concomitant use of antipsychotics, type of previous antipsychotic drug, change in antipsychotic medication, number of antipsychotic medication changes, maximum chlorpromazine (CP)-equivalent dose of antipsychotics (Inada and Inagaki, 2015) in the past year before the index date, concomitant use of antidepressants, concomitant use of mood stabilizers, concomitant use of benzodiazepines, concomitant use of anti-Parkinson drugs, time to increase in brexpiprazole dose to 2 mg/day, timing of initiation of brexpiprazole prescription, and proportion of days covered (PDC).
Unstructured evaluation variables are defined in Supplementary  Table 2 and included positive symptoms of schizophrenia, symptoms other than positive symptoms, the period from the first occurrence of positive symptoms to the index date, days from admission to remission, treatment interruption, the period from the date of the first positive symptom to the date of the first medical examination, suicidal behavior, isolation, physical restraint, electroconvulsive therapy (ECT), intellectual disability, work experience, highest educational qualification, a history of crime committed or reported, medication administration, interpersonal relationships, tendency to refuse medication, insomnia, stress endurance, availability of key contact person information, willingness to be discharged, using home-visit nursing care, using daycare and day-night care, welfare benefits recipient, obstacle pension recipient, agitation, anxiety, experiencing extrapyramidal symptoms, and sexual dysfunction.

Data analysis
Categorical variables are summarized by the number of cases and ratio, and continuous variables are summarized by mean and standard deviation (SD). Cumulative treatment continuation rate at 180 days from brexpiprazole initiation was determined using the Kaplan-Meier (KM) method. The frequency and rate of treatment discontinuation and censoring were calculated, along with their 95 % confidence intervals (CIs).
Both univariate and multivariate Cox proportional hazards models were used to determine the factors influencing brexpiprazole discontinuation and the results are presented as hazard ratios (HRs), 95 % CIs of HRs, and P-values. P < 0.05 was considered as significant for the multivariate Cox proportional hazards model. In addition, a multivariate Cox proportional hazards model using a stepwise method was used to consider the factors influencing brexpiprazole discontinuation and the results are presented as HRs, 95 % CIs of HRs, and P-values.
Using the factors extracted with the stepwise method, adjusted survival curves for a Cox proportional hazards model were used to obtain cumulative continuation rates and their 95 % CIs. P < 0.05 was considered as significant. All analyses were performed using SAS version 9.4.

Patient selection
A total of 37,534 patients were included in MENTAT® between April 18, 2018, and May 15, 2020, of whom 5551 were identified as having schizophrenia using the F20 ICD-10 diagnosis code and further screened for eligibility. The analysis population comprised 515 patients who met the eligibility criteria ( Fig. 1).

Brexpiprazole continuation
The mean (SD) time to achievement of the optimal dose of 2 mg for brexpiprazole was 10.5 (17.7) days; the optimal 2-mg dose was achieved in ≤28 days in 66.2 % of patients (Table 2).

Factors influencing brexpiprazole discontinuation
Sixteen variables identified to be independently associated with brexpiprazole discontinuation through a univariate Cox proportional hazards analysis were subjected to a multivariate Cox proportional hazards analysis.
The multivariate analysis identified eight variables associated with treatment discontinuation, with an HR that was significant (P < 0.05): variables with HR <1 were the presence of physical complications, a longer hospitalization duration (measured in days), and maximum CPequivalent dose of antipsychotics of >200 to ≤400 mg/day vs ≤200 The results indicated that patients were less likely to discontinue brexpiprazole if they had physical complications and a longer hospitalization duration and were more likely to discontinue brexpiprazole if they had received prior ECT, had key contact person information available, and had a history of crime committed or reported before brexpiprazole initiation. After brexpiprazole initiation, patients were more likely to discontinue brexpiprazole if their time to increasing the brexpiprazole dose to 2 mg was >28 days and if they developed appearance/worsening of symptoms other than positive symptoms after 28 days (Table 5; Fig. 3).

Table 2
Outcomes with assessment after the index date (analysis population).

Discussion
This study retrospectively explored the factors affecting brexpiprazole discontinuation within 180 days of newly initiating brexpiprazole using both structured and unstructured data analyzed through the MENTAT® system (Corporation, 2015;Otsuka Digital Health, 2016). The cumulative brexpiprazole continuation rate on day 180 for this Japanese population with schizophrenia was 29 %; 71 % of patients discontinued brexpiprazole by follow-up day 180, and the mean (SD) time to occurrence of a discontinuation event was 51.2 (43.6) days.
In this study, 515 patients with schizophrenia included in the analysis population had a mean (SD) age of 48.0 (15.3) years, 47.8 % were male, 4.7 % had received ECT in the past year, and the mean (SD) CP-equivalent dose of antipsychotics received was 724.35 (726.21) mg/ day (maximum prior therapy over the past year evaluated on the index date). These patient demographics and treatment characteristics were similar to those reported in other studies from Japan and represented the general patient population with schizophrenia in Japan, as reported by the Psychiatric Clinical Pharmacology Research Group in 2011 (Noda et al., 2015) and the Effectiveness of Guidelines for Dissemination and Education in Psychiatric Treatment project survey in 2020 (Ichihashi et al., 2020) and 2021 (Hashimoto et al., 2021). The proportion of patients who discontinued brexpiprazole in this study (day 60, 47.4 %; day 180, 71.1 %) was higher than that reported in Japan in patients with schizophrenia or schizoaffective disorder (18.3 %, 31.7 %, 36.6 %, and 43.4 % at weeks 4, 8, 12, and 16, respectively) (Yoshimura et al., 2020) and schizophrenia (51.5 % and 61.4 % at weeks 24 and 52, respectively) (Inoue et al., 2021). This can be attributed to the 33.8 % of patients who did not reach the optimal dose of 2 mg by day 28. In addition, the mean CP equivalent value of the previous treatment (maximum value in the past year before initiating brexpiprazole treatment) was 724.35 mg in this study compared with 402.5 mg reported in the study by Yoshimura et al. Differences in estimated CP equivalents represent differences in disease severity; therefore, this observed difference in disease severity before the initiation of brexpiprazole treatment might have influenced the treatment continuation rate in this study.
In this study, the factors associated with brexpiprazole discontinuation were physical complications, a longer hospitalization duration, symptoms other than positive symptoms (after 28 days), prior ECT, time to brexpiprazole dose increase to 2 mg, maximum pretreatment CPequivalent antipsychotic concentration, availability of key contact person information, and a history of crime committed or reported. Notably, most factors, including symptoms other than positive symptoms (28 days after brexpiprazole initiation), prior ECT, availability of key contact person information, and a history of crime committed or reported, were identified by analyzing textual information (unstructured data). Therefore, the use of databases employing the MENTAT® system, which analyzes both structured and unstructured data in psychiatric EMRs, may provide new insights in the future. The current study also demonstrated the potential utility of the MENTAT® system in psychiatry research.
Two previous studies have assessed the factors influencing brexpiprazole continuation-one of these studies identified a high CPequivalent dose of the previous antipsychotic (>800 mg vs ≤800 mg)  as a factor associated with an increased risk of brexpiprazole discontinuation (Yoshimura et al., 2020), whereas the other reported increasing age, decreasing disease severity, and the first use of brexpiprazole in an outpatient setting as factors that increased the brexpiprazole continuation rate during a longer follow-up of 52 weeks in Japan (Inoue et al., 2021). Although CP conversion and age were examined as factors affecting continuation in this study, disease severity and the first use of brexpiprazole were not analyzed as the information could not be obtained in this study. Among the other factors identified to be associated with brexpiprazole discontinuation in this study, patients with physical comorbidities take medications for their conditions on a regular basis and, thus, are more likely to continue treatment with concomitant brexpiprazole. In addition, at the index date, approximately 60 % of patients were hospitalized. In Japan, patients with psychiatric disorders have a high rate of prolonged hospitalization of >1-year duration. Hence, patients hospitalized at the index date are likely managed by HCPs throughout the 180-day follow-up or at the time of discontinuation and, thus, are more likely to continue brexpiprazole treatment (Ministry of Health Labour and Welfare, 2017). We evaluated the influence of the maximum CPequivalent dose in this study, but the difference in the categorization and calculation method of the CP-equivalent value may have affected the results. This study found that patients who received a previous CPequivalent dose of ≤200 mg/day in the past year were less likely to continue brexpiprazole vs those who received a CP-equivalent dose of >200 to ≤400 mg/day in the past year; this comparison with the previous CP-equivalent dose of ≤200 mg/day was not significant for other dose ranges. Furthermore, the single-return results in CP-equivalent units of the pretreatment drug (assessing for every 1-mg increase in CP equivalents) did not show any effect on brexpiprazole continuation. Therefore, the impact of the initial antipsychotic drug dose on brexpiprazole continuation remains unclear as it could not be confirmed.
This study identified new factors, such as prior ECT, a history of crime committed or reported, and availability of key contact person information, that influenced brexpiprazole discontinuation at 180 days. Patients with documentation of a prior history of ECT and a history of crime committed or reported in the EMR over the past year may have had severe psychiatric or psychotic symptoms or treatment resistance on the index date, which may have influenced brexpiprazole discontinuation. Usually, general perception may regard treatment continuation as easier in patients with key contact person information available on the EMR as they may support treatment adherence. However, HCPs are more likely to record patients' key contact person information in the EMR only when there are unstable or active psychiatric/psychotic symptoms. In addition, patients with time to brexpiprazole dose escalation to 2 mg after 28 days are less likely to continue treatment vs those with time to dose escalation of ≤28 days, which may be related to brexpiprazole plasma concentrations. To adequately occupy the dopamine receptors, a brexpiprazole plasma concentration of 60 ng/mL is required (Higuchi et al., 2018;Wong et al., 2021), and the brexpiprazole 1-mg dose is insufficient to achieve this target. Thus, unresolved psychiatric symptoms may affect treatment continuation, indicating the need for reaching the 2-mg optimal dose as early as possible when brexpiprazole treatment is initiated. In addition, regardless of the status of psychiatric symptoms before initiating brexpiprazole treatment, a lack of efficacy on symptoms other than positive symptoms after initiating brexpiprazole treatment may have an influence on treatment continuation.
Overall, considering the results of this study and those of Yoshimura et al. (2020), for severely ill patients with a CP conversion value exceeding 800 mg prior to brexpiprazole treatment and patients suspected of being resistant to treatment, it may be necessary to avoid brexpiprazole treatment and consider another treatment. We also believe that even when such patients are treated with brexpiprazole, additional measures such as changing the frequency of visits to the hospital to closely observe their potential risk factors can be applied. In addition, early dose escalation to 2 mg/day may lead to a favorable treatment outcome when using brexpiprazole. Also, our results differed from those of other studies that assessed the factors influencing the continuation of antipsychotic drugs, which could be attributed to differences in treatment durations and dosage forms with the study design and factors assessed using the MENTAT® system (Taylor et al., 2009;Whale et al., 2015;Ye et al., 2011). It is also not possible to determine whether the factors identified in this study are brexpiprazole-specific because comparisons with other antipsychotics were not performed. Therefore, these results cannot be generalized to other antipsychotic drugs. Further research is required to clarify if the factors affecting discontinuation/continuation are different for different types of antipsychotic drugs.

Limitations
The following study limitations were noted: Data on patient background may be influenced by an institution selection bias due to the limited number of institutions where MENTAT® was introduced and who agreed to provide anonymized EMR information to a third party. Text freely described in the EMRs evaluated by the NLP technique may not be correctly reflected unless elaborated in the text. Medical record information of patients transferred to another medical institution may not be available to track subsequent outcomes. Data on newly diagnosed patients with schizophrenia or treatment-naïve patients were not available before the index date. PDC was calculated by the number of drugs/daily doses prescribed, not by the number of doses taken by the patient. The impact of information outside the EMRs on outcomes could not be assessed. Factors extracted by the multivariate Cox proportional hazards model using the stepwise method and their impact on outcomes are dependent on the institution where MENTAT® was introduced, resulting in limited generalizability.  HRs, 95 % CIs, and P-values with treatment interruption as an event are shown using multivariate Cox proportional hazards modeling using the stepwise method with the factors examined in the multivariate analysis in Table 4 as explanatory variables. Statistically significant values are indicated in bold. CI = confidence interval. ECT = electroconvulsive therapy. HR = hazard ratio.

Conclusions
We assessed factors not listed in existing databases that affect brexpiprazole continuation using a dataset with both structured and unstructured information from psychiatry EMRs using advanced NLP. Consequently, we identified potential new factors that may be associated with brexpiprazole discontinuation, which may improve the treatment strategy and continuation rate in patients with schizophrenia.

Role of the funding source
This study was funded by Otsuka Pharmaceutical Co., Ltd. The study analysis was performed by Remedy & Company Inc. and funded by Otsuka Pharmaceutical Co., Ltd.