Schizophrenia symptoms and functioning in patients receiving long-term treatment with olanzapine long-acting injection formulation: a pooled analysis

The present analysis was performed on data from patients with schizophrenia who were not acutely ill and were receiving maintenance treatment with OLAI medication. These patients were enrolled in 3 clinical trials: HGKA, HGLQ, and HGKB. Patients in HGKB could have previously participated in studies LOBS, HGKA, and HGJZ. All 5 studies were sponsored by Eli Lilly & Company. The studies used in the analyses had measurements of PANSS and of QLS available at baseline, and at 6 months (± 1 month). PANSS and QLS data were also evaluated at 12 ± 3 months (1 year). All patients gave written informed consent prior to participation in any of the studies. The main study design features of the studies included in the pooled analysis are summarized below:

All study protocols were approved by institutional review boards at each site.

Data collected during the acute treatment periods in the short term studies HGJZ and LOBS were not included in the present analysis because these patients were either acutely ill (HGJZ) or received an OLAI dosage schedule specifically for pharmacokinetic evaluations (LOBS). Therefore, these data were not helpful for assessing maintenance treatment effects of OLAI. For the present analysis, baseline for patients from these studies was defined as entry into study HGKB. Baseline for patients in HGKA and HGLQ was the first assessment in these studies.

The following baseline variables collected in all three studies were used for statistical modeling: age, baseline category, baseline remission status, extrapyramidal symptoms (EPS), geographic region, CGI-S score , number of previous episodes, PANSS total and subscores (anxiety/depression, disorganized, hostility, negative, positive), QLS total and domains (interpersonal relations, instrumental role functioning), study indicator (HGKA, HGLQ, HGKB), gender, time since diagnosis, weight, and employment status (unemployed, working, retired).

Overall data from 1182 patients at baseline, 864 patients at 6 months and 687 patients at 12 months were used for the present analysis.

Psychiatric symptoms were assessed using the PANSS [9]. In the present study, for creating the patient categories, we followed the PANSS factor analysis by Marder et al. [10]. The 5 main factor scores were: 1) PANSS Negative factor (PANSS NEG; 7 items: 8–11,13,21,30); 2) PANSS Positive factor (PANSS POS; 8 items: 1,3,5,6,14,15,23,26); 3) PANSS Disorganized factor (PANSS DIS; 7 items: 2,12,19,24,25,27,29); 4) PANSS Hostility factor (PANSS HOS; 4 items: 4,7,22,28); and 5) PANSS Anxiety/Depression factor (PANSS DEP; 4 items: 16,17,18,20). Scores on individual PANSS items ranged from 1 (no symptoms) to 7 (severe symptomatology). PANSS factor scores were divided by the number of items per factor (i.e. normalized factor scores range from 1 to 7).

Functioning in all studies was assessed with the QLS total score. The QLS items range from 0 (worst) to 6 (best). For the analysis we used the 2 QLS domains (interpersonal relations and instrumental role [11]) as categorizers identified in the Lipkovich et al [4] cluster analysis.

These 2 domains of the QLS represent the part of the QLS that reflects the functional aspects of a patient's quality of life including the patient's ability to work and to maintain relationships with others. Overall scores for domains as well as the total score were divided by the number of the respective items.

Based on cluster analysis reported in Lipkovich et al [4], patients with schizophrenia can be grouped into 5 different clusters (categories) based on severity of their clinical symptoms as measured by PANSS and on the level of their functioning, as measured by the QLS Interpersonal and QLS Instrumental domains. For the current analysis, the classification rules of this previous publication were used. The category cutoffs used, as determined by a classification tree algorithm [CART, [12]], were based on the QLS Instrumental domain score and on the PANSS DIS and PANSS POS subscores. These decision rules do not yield mutually exclusive categories but are able to separate categories with high specificity between 87% and 98%.

The categories contain patients with good symptom control and moderate to good functioning (Category A), patients with either good symptom control and impaired functioning (Category B) or insufficient symptom control and moderate to good functioning (Category C), and the most severe categories include patients with moderate to severe symptoms and impaired functioning (Category D) and with severe psychiatric symptoms and impaired functioning (Category E).

Categorical improvement of a patient was defined as moving from Categories B or C to Category A, or from Categories D or E to Categories A, B, or C. Thus, patients in Category A could not improve and were excluded from the analysis of improvement. Worsening of a patient was defined as moving from Category A to Categories B, C, D, or E or from Categories B or C to Categories D or E. As only few patients were categorized in Categories D and E both categories were combined and could not worsen. Thus, they were excluded from the analysis of worsening.

As an additional measure of long term treatment outcomes in addition to the cluster analysis described above, the proportion of patients in remission and sustained remission were analyzed. Remission was defined based on the Andreasen criteria [5]. Sustained remission was defined as maintaining simultaneous ratings of “mild” or better (score of 1, 2, or 3) over a 6 month period on the following items of the PANSS: delusions, conceptual disorganization, hallucinatory behavior, mannerisms/posturing, unusual thought content, blunted affect, social withdrawal, and lack of spontaneity.

Cross-sectional remission at baseline or any other specific time point was defined similarly, but excluded the requirement for maintenance over a 6 month period.

Patient baseline and post baseline characteristics were summarized using descriptive statistics and compared among baseline categories using analysis of variance (ANOVA) for continuous variables and chi-square test for categorical variables. Reasons for discontinuation and baseline characteristics of patients remaining in the analyses versus those discontinuing within 6 months were summarized and compared using t-test for continuous variables and chi-square test for categorical variables.

Baseline factors were assessed for their association with categorical improvement using stepwise logistic regression. These analyses were performed for categorical improvements from baseline to 6 months and to 12 months (see definition of categorical improvement in 'Outcome measures' section).

At the first step of modeling reported in Table 1 under the ‘Single variable model’ heading, a number of single factor logistic regression models were fitted. Each model evaluated a single baseline variable in relation to the outcome (categorical improvement) without adjustment for other variables. The following baseline variables were used as potential predictors: age, baseline category, baseline remission status, extrapyramidal symptoms (EPS), geographic region, Clinical Global Impression Severity (CGI-S) score [13], number of previous episodes, PANSS total and subscores (anxiety/depression, disorganized, hostility, negative, positive), QLS total and domains (interpersonal relations, instrumental role functioning), study indicator (HGKA, HGLQ, HGKB), gender, time since diagnosis, weight, and employment status (unemployed, working, retired). This approach allowed for screening of a high number of correlated variables and provided additional information about potential association prior to adjustment for other variables. SAS automatic procedures would exclude all patients who have missing values for at least one of the variables and have convergence issues with correlated variables, and therefore were not used. The number of covariates entering at the second step in the multiple logistic regression model was reduced by manual variable selection based on the first step modeling results to avoid over-fitting.

At the second step of modeling, reported in Table 1 under the ‘Stepwise multivariate logistic regression’ heading all baseline variables that were identified as potentially significant (p < 0.1 in the single factor model) were simultaneously included in a multivariate model for stepwise selection of covariates based on significance cut-off of p < 0.1 entry into the combined model. After the stepwise procedure only those variables that were independently predicting outcome remained in the model. Other variables were excluded due to co-dependencies, even if they were significant as a single predictor. To avoid “overlap” in predictors and issues with co-linearity, variables derived from predictors already included in the model (total scores for the PANSS and QLS scales, remission status, and baseline category) were not used in the multiple regression model. No imputation of missing data was used. SAS version 9.1.3 (SAS Institute, Cary, N.C.) was used for all analyses.

1182 patients with baseline PANSS and QLS data available were included in the current analysis. Patients included in the analysis were 66.8% males, had a mean age of 39.6 years (range: 18 to 74 years), and 46.5% were from the Americas, 44.8% from Europe and 8.6% from other areas. Baseline mean PANSS total score was 62.1 (±16.4), and mean QLS total score was 1.65 (±1.81). EPS were reported for 13.9% of the patients.

339 patients (28.7%) discontinued from one of the 3 respective studies during the first 6 months. The retention rate by study was 75.3% (HGKA, 451/599); 73.6% (HGLQ, 192/261), and 69.5% (HGKB, 221/318). Of all 864 patients with data at 6 months (i.e. 21 patients discontinuing at the 6 months visit had still data available), 787 (91.1%) had sufficient data available for the assessment of categorical changes in schizophrenia symptoms (PANSS) and health-related functioning (QLS).

The most common (>10% of all discontinuations) reasons for discontinuations in the first 6 months were subject decision (129, 38.1%), adverse event (56, 16.5%), lack of efficacy (41, 12.1%), and lost to follow-up (38, 11.2%).

We have compared baseline characteristics of patients who remained for at least 6 months in the studies with those who discontinued (Table 2). Patients discontinuing early had more severe psychiatric symptoms at baseline – in particular acute symptoms as measured by the PANSS POS, PANSS HOS, and PANSS DEP scores. Higher rates were also found for other baseline measurements, reflecting decreased functioning such as CGI-S, working status, baseline remission, and QLS instrumental role functioning. No significant differences between dropouts and patients remaining in the studies were seen for age, geographic region, gender, PANSS NEG and PANSS DIS scores, QLS interpersonal relations domain, EPS at baseline, and number of previous episodes.

Table 3 shows baseline characteristics in relationship to baseline categories (PANSS and QLS scores at baseline are reported in Table 4 and Figure 1).

By applying the rules based on Lipkovich et al. [4] and described in the 'Outcomes measures' section to the baseline scores of patients in this analysis, patients were grouped into the 5 pre-specified categories. The largest category (Category B: n = 434, 36.7%, see Table 4 and Figure 1) consisted of patients with minimal symptoms of schizophrenia according to the PANSS total score and with seriously impaired functioning according to the QLS score, followed by patients with moderate to severe PANSS total score symptoms and seriously impaired function according to QLS (Category D: n = 303, 25.6%). Category C (n = 208, 17.6%) comprised patients with moderate PANSS total scores but good functioning according to QLS, and Category A comprised patients with minimal schizophrenic symptoms and good functioning (n = 162, 13.7%). As only non-acutely ill patients were eligible for participation in these studies, only 72 (6.1%) patients were grouped in Category E with severe schizophrenic symptoms and impaired functioning.

Baseline category was significantly related to CGI-S, presence of baseline EPS, working status, cross-sectional remission, age, and number of previous episodes (Table 3). Some differences regarding geographical regions were found. In particular more patients from Eastern Europe were categorized into Category A. However, this might reflect recruitment differences of patients by studies and is considered as not clinically relevant.

The rate of patients discontinued ranged from 26% to 42% by baseline category at 6 months and from 30% to 54% at 12 months (see Table 4). The lowest rates were seen in Category A and the highest rates in the worst categories (D and E).

The change of category membership from baseline to the 6 months and 12 months assessments is shown in Table 4 and Figure 1.

Overall at 6 months, 261 (26%) of all patients who could improve in category (all but A) did improve, 471 (40%) of all patients remained in their baseline category, and 55 (7%) of all patients who could get worse (all but D and E) worsened. For 395 (33%) of all patients, a classification into category at 6 months was not possible because they had no PANSS or QLS score corresponding to the 6 month visit available. By baseline category, improvement was seen in >30% of patients in baseline Categories C and D but for less than 15% in baseline Category B. Worsening was seen in about 20% of patients in baseline Category A and was <5% in baseline Categories B and C.

The proportion of patients improving (239; 24%) or worsening (48; 6%) after 12 months of OLAI treatment was similar to those proportions seen at 6 months. The proportion of patients who could not be classified due to either discontinuation or missing data after 12 months was 44% (523).

Table 5 shows the PANSS Total score and QLS Interpersonal and QLS Instrumental domain scores at baseline, 6 months, and 12 months overall and by baseline category and the cross-sectional and sustained remission status assessed at baseline, 6 months, and 12 months.

Cross-sectional remission status improved in 199 patients (17%) from no remission at baseline to remission at 6 months. Worsening from baseline remission to no remission at 6 months occurred in 29 patients (2%). Remission status sustained at 6 months in 395 (33%) of all patients. Sustained remission after 12 months was achieved by 476 (40%) patients, only 49 (4%) of those patients in cross-sectional remission at baseline did not achieve sustained remission at 12 months.

The single variable logistic regression analyses found a significant influence of a number of baseline covariates on categorical improvement of patients at 6 and 12 months. All variables with a significant influence on categorical improvement (except baseline category) are summarized in Table 1.

Based on these, the stepwise multiple logistic regression analyses were performed for categorical improvement at 6 months and 12 months. The significant predictors identified from the stepwise multiple logistic regression are shown in Table 1. An odds ratio estimate greater than 1 indicates higher chances for improvement. Odds ratios were calculated based on an increase of 1 unit of the baseline value for continuous scales or compared to a reference category at baseline for categorical variables.

The odds for achieving a better category at 6 months and 12 months were better for patients with a higher (worse) baseline PANSS POS. In contrast, 6 month categorical improvement was more probable for patients with a lower (better) baseline CGI-S. All other variables that were significant as single predictors lost their significance in the joint model and were excluded by the stepwise procedure. The shaded variables were not entered in the model due to the fact that they were derived from other variables as described in the statistical procedure section.