Early improvements of individual symptoms as a predictor of treatment response to asenapine in patients with schizophrenia

Abstract Aim It is well accepted that early improvement with antipsychotics predicts subsequent response in patients with schizophrenia. However, no study has examined the contribution of individual symptoms rather than overall symptom severity as the predictors. Thus, we aimed to detect individual symptoms whose improvements could predict subsequent response in patients with schizophrenia during treatment with asenapine and examine whether a prediction model with individual symptoms would be superior to a model using overall symptom severity. Methods This study analyzed a dataset including 532 patients with schizophrenia enrolled in a 6‐week double‐blind, placebo‐controlled, randomized trial of asenapine. Response to asenapine was defined as a ≥30% decrease in Positive and Negative Syndrome Scale (PANSS) total score from baseline to week 6. Stepwise logistic regression analyses were performed to investigate the associations among response and PANSS total/individual item score improvements at week 1 or week 2. Results Response was associated with early improvement in the following PANSS items: disturbance of volition, active social avoidance, poor impulse control at week 1; and active social avoidance, poor attention, lack of judgment and insight at week 2. Prediction accuracy was almost compatible between the model with individual symptoms and the model with PANSS total score both at weeks 1 and 2 (Nagelkerke R 2: .51, .42 and .55, .54, respectively). Conclusion Early improvement in negative symptoms, poor attention and impulse control, and lack of insight, in particular predicted subsequent treatment response in patients with schizophrenia during treatment with asenapine as accurately as prediction based on overall symptom severity.


| INTRODUC TI ON
Antipsychotics are the mainstay for treatment of schizophrenia.1 In particular, second-generation antipsychotics are commonly used, given that they have favorable profiles, including a decreased risk of neurological side effects compared to first-generation antipsychotics.2 However, second-generation antipsychotics also have undesirable side effects such as glycolipid metabolism disorders. Therefore, an ineffective course of treatment that carries risks of certain adverse effects should be avoided by predicting antipsychotic treatment response trajectories as early as possible.
Accumulating evidence suggests that an early improvement with antipsychotics serves as a predictor of subsequent response to antipsychotics in patients with schizophrenia,3-5 with the greatest symptom reduction occurring within the first week of the antipsychotic initiation. 6 However, previous studies targeting schizophrenia have investigated the relationship between overall severity of symptoms and subsequent treatment response,3-5 focusing on the sum of the scores in the representative rating scales (eg, the Positive and Negative Syndrome Scale (PANSS)). To the best of our knowledge, no study to date has investigated whether early improvements in individual symptoms could predict subsequent response to antipsychotics in this population except for one study. More specifically, Ruberg et al used classification and regression tree (CART) analysis to identify individual positive symptoms whose early changes can predict long-term response to atypical antipsychotics in the treatment of schizophrenia.7 In fact, modern psychiatry is limited in its evaluation of clinical severity by a total score rather than measurable individual scores. Thus, it is crucial to detect individual symptoms whose improvements can predict subsequent response for each patient with schizophrenia in clinical practice.
To fill in the gap in the literature, we utilized clinical data (P06124 trial: a 6-week randomized, double-blind, placebo-controlled trial of the effectiveness of asenapine in participants with an acute exacerbation of schizophrenia) in order to test the following hypotheses: (1) Early improvement in individual symptoms will predict subsequent response to asenapine in patients with schizophrenia and (2) the prediction model with individual symptoms will be more accurate than the model with overall symptom severity.

| Study design
Following a complete description of the study, subjects provided written informed consent at study enrollment in the original study, which was approved by the institutional review board of all participating sites. This post hoc analysis was made completely anonymous; thus, ethical approval was not sought out for this study. This study analyzed data from a 6-week multicenter, randomized, doubleblind, placebo-controlled phase III trial of asenapine conducted from May 2010 to April 2014 in Japan, Republic of Korea, and Taiwan. The original study consisted of a screening phase (ie, 3-7 days from the screening test to the baseline), 6-week treatment period, and followup phase. During the screening period, patients were administered placebo tablets twice a day in a single-blind manner to exclude patients with high placebo responsiveness, as described below. This study enrolled males and females, aged 20-65 years, having a diagnosis of schizophrenia with an acute exacerbation based on the DSM-IV-TR criteria. Patients were selected based on the following criteria at baseline of the treatment phase: (a) assessed at the screening phase, (b) a score ≥60 on PANSS total score at baseline, (c) scores of ≥4 on two or more items on the PANSS positive symptom subscale (delusion, conceptual disorganization, hallucinatory behavior, excitement, grandiosity, suspiciousness, or hostility), and (d) a score ≥4 on the Clinical Global Impressions-severity Illness scale at baseline. Based on the results obtained from the short-term administrations of phase II and phase III in the original study,8 the current episode period was set to be within 2 months. To exclude patients with high placebo response, those who had a score reduction of ≥20% on the PANSS total score during the screening phase were excluded.8 After baseline assessment was completed, patients considered to be eligible by the investigator were randomized (1:1: 1) to receive sublingual asenapine 5 mg bid, 10 mg bid, or placebo. During the study period, patients received asenapine or placebo doses twice daily for 6 weeks. The tablets were administered sublingually without water. The PANSS assessment was completed at baseline and weekly through week 6. Study physicians, patients, and raters were all blinded throughout the study.

| Statistical analysis
We used the dataset for the last observation carried forward (LOCF) analysis, in which the score at the previous visit was adopted in the case of premature attrition. The Shapiro-Wilk test was performed to test the normality of the data. As a result, we found that the data did not follow the normal distribution. Baseline socio-clinico-demographic characteristics were compared between responders and non-responders by Mann-Whitney U test and chi-square or Fisher's test for continuous variables and categorical variables, respectively.
We employed baseline sociodemographic and clinical characteristics that were statistically different between the 2 groups and the changes of each individual symptom from week 0 to week 1 or week 2 as the independent variables. Percent decrease in PANSS total scores was calculated as (PANSS total scores at baseline − PANSS total scores at week 6) × 100/(PANSS total scores at baseline − 30).
Here, based on previously published studies,8 response to antipsychotics was defined as a ≥30% decrease in PANSS total scores. Based on the previous studies,5 we defined early improvement as 20% decreases in PANSS total scores at week 2. Stepwise logistic regression analyses were performed to evaluate the relationships among early improvements of PANSS total scores at week 1 or week 2 and treatment response at week 6. We explored sequentially defining early improvement as 5%, 10%, 15%, and 20% decreases in PANSS total scores at week 1 or week 2. Stepwise logistic regression analyses were also performed to evaluate the relationships among changes of PANSS total scores at week 1 or week 2 and treatment response at week 6. We employed two definitions of early improvement with categorical and continuous data of early improvement for these analyses as independent variables. Although these two approaches are complementary to each other, the latter method was applied to confirm more detailed changes in the early response. Finally, as a primary analysis, stepwise logistic regression analyses were performed to evaluate the relationships among early improvements of PANSS individual item scores at week 1 or week 2 and treatment response at week 6. Sensitivity, specificity, and accuracy for the prediction models were calculated for the analyses on overall symptoms and individual symptoms, respectively. A P-value of <.05 was considered statistically significant (2-sided). Statistical analyses were performed with IBM SPSS Statistics version 24.0 (IBM Japan).

| Clinico-demographic characteristics of the patients
Clinico-demographic characteristics of 315 patients are summarized in Table 1. Almost half of the patients were Japanese and one third were Taiwanese, aged 41.5 ± 11.0 in the responder group, and 41.6 ± 11.1 in the non-responder group, with a nearly equal malefemale ratio. Responders were associated with higher rates of comorbid psychiatric diseases and lower rates of Korean patients, dropouts, and intake of anticholinergics in comparison with nonresponders (Table 1). Tables 2 and 3 summarizes the number of patients who showed early improvement at week 1 and week 2, and the number of those who responded at week 6. Figure 1 shows the PANSS total scores for responders and non-responders, at week 1 and week 2 in the early improvement group and the early non-improvement group, respectively. The logistic regression analysis with early improvement of 20% as a categorical variable showed significant results (χ 2 = 106.7, df = 5, P < .0001 at week 1, and χ 2 = 142.7, df = 7, P < .0001 at week 2) which explained 43.0% and 54.5% (Nagelkerke R 2 ) of the variance in responders with improvements at weeks 1 and 2, respectively. The prediction performance of binary classification of early improvement at weeks 1 and 2 for response at week 6 is shown in Tables 4 and 5. The 20% cutoff in the PANSS at week 2 showed the highest degree of accuracy for predicting response at week 6 ( Tables 4 and   5). Also, the logistic regression analysis with score reductions in the PANSS total score as a continuous variable showed significant results (χ 2 = 111.6, df = 3, P < .0001 at week 1, and χ 2 = 149.6, df = 5, P < .0001 at week 2) which explained 44.6% and 56.5% (Nagelkerke R 2 ) of the variance in responders with improvements at weeks 1 and 2, respectively.

| D ISCUSS I ON
This is the first study to examine whether early improvement of individual symptoms including negative symptoms and cognitive impairment could serve as clinically useful predictors of response in patients with schizophrenia during an acute course of treatment.
We found that improvements of the following symptoms in the PANSS score at weeks 1 and 2 were related to treatment response to asenapine at week 6: disturbance of volition, active social avoidance, and poor impulse control at week 1; active social avoidance, poor attention, and lack of judgment and insight at week 2.
Ruberg et al conducted a similar retrospective study to identify predictors of treatment response that most effectively differentiated responders from non-responders in chronic patients with schizophrenia by using CART analysis. They found that improvement of six positive symptom items in the PANSS two weeks after antipsychotic administration predicted treatment response after eight weeks. However, there are some differences in the methodology between their study and ours. First, they used mixed datasets from multiple clinical studies while we used data from a single phase III trial of asenapine. Second, Ruberg's study was limited to chronic pa-

| Negative symptoms
In this study, early improvement in active social avoidance, a second-

| Cognitive dysfunction
In this study, early improvement in poor attention was the second strongest predictor at week 1 for treatment response to asenapine TA B L E 6 Association between early improvements at week 1 in individual symptoms in the PANSS and subsequent response in patients with schizophrenia

| Poor impulse control
Early improvement in poor impulse control was the third strongest predictor at week 1 for treatment response to asenapine at Therefore, it is important to improve impulsivity in patients with schizophrenia during an earlier treatment phase. Indeed, it is noted χ 2 = 140.7, df = 10, P < .0001, Nagelkerke R 2 = .54, sensitivity = 0.61, specificity = 0.94, accuracy = 0.86 TA B L E 7 Association between early improvements at week 2 in individual symptoms in the PANSS and subsequent response in patients with schizophrenia that antipsychotics, including asenapine, cause an early improvement in impulsivity in this population. Studies investigating the effects of asenapine on excitement in patients including those with schizophrenia showed a significant decrease in the PANSS excited score 2 hours after asenapine treatment (NNT 3 (95% CI 2-4)).23 In addition, Volavka et al24 showed that olanzapine improved hostility, which is another excitatory component, at an early treatment phase.
Thus, these findings suggest that optimal antipsychotic treatment improves poor impulse control in patients with schizophrenia, which may in turn lead to subsequent favorable treatment outcomes in this population.

| Lack of insight
It is known that more than 50% of patients with schizophrenia present with moderate-to-severe lack of insight. Lack of insight in schizophrenia is associated with low medication adherence and poor outcomes leading to relapse and high mortality.25, 26 We found that early improvement in lack of judgment and insight was the third strongest predictor at week 2 for treatment response to asenapine at week 6. In support, previous studies demonstrated that lack of insight is associated with overall disease severity as well as cogni-

| Limitations
The results of our study must be interpreted in light of some limi-

ACK N OWLED G M ENT
Meiji Seika Pharma provided phase III data for the study. Yoshida, and K. Ogyu were responsible for drafting the manuscript.

CO N FLI C T O F I NTE R E S T
All other authors were responsible for critical revision of the manuscript and have accepted the final version. All other authors contributed to and have approved the final manuscript.

DATA AVA I L A B I L I T Y S TAT E M E N T
We obtained the data that support the findings of this study from Meiji Seika Pharma with a contract between us, which defined the group that can use the data and the duration for which the data is usable. Data are available from Meiji Seika Pharma upon reasonable request to Meiji Seika Pharma.