Data Pooling Strategies
Within the seven studies, 82.7% of patients were Caucasian and 7.9% were African-American. The current analyses therefore utilize 90.6% of the total study population. The remaining patients were primarily of Hispanic origin (7.4%).
Safety analyses included data from Caucasian and African-American patients in all seven studies—Placebo: African Americans, n=59; Caucasians, n=594; duloxetine 40 mg/day): African Americans, n=69; Caucasians, n=748. Efficacy analyses were performed on three sets of data, obtained using the following pooling strategies:
1) Data from all seven studies (hereafter referred to as “all studies”)—Placebo: African American s, n=59; Caucasiana, n=594; duloxetine 40 mg/day): African Americans, n=69; Caucasians,
2) Data from the four studies that demonstrated a significant advantage for duloxetine over placebo on the primary efficacy measure (studies 1, 4, 5 and 6, “positive studies”)— Placebo: African Americans, n=31; Caucasians, n=343; cymbalta 30 mg/day: African Americans, n=28, Caucasians, n=418;
3) Data from two pivotal MDD studies in which patients received a duloxetine dose of 60 mg once daily (5 and 6, “focus studies”)—Placebo: African Americans, n=16; Caucasians, n=212; duloxetine, 60 mg QD: African Americans, n=16; Caucasians, n=207.
Analysis of data from all studies provided an assessment of comparative efficacy in African-American and Caucasian patients across the largest possible data set. The analysis of data from the four positive studies allowed differential efficacy to be studied without the potential confounding influence of non-positive data. Analyses of data from the two focus studies are of particular clinical relevance since they examine treatment effects in patients receiving the recommended therapeutic duloxetine dose.
Efficacy was assessed using the HAMD17 total score; HAMD17 subscales: anxiety (items 10, 11, 12, 13, 15 and 17), core factor (items 1, 2, 3, 7 and 8), Maier (items 1, 2, 7, 8, 9 and 10), retardation (items 1, 7, 8 and 14) and sleep (Items 4, 5 and 6); the CGI-S scale; the Patient Global Impression of Improvement (PGI-I) scale; Visual Analog Scales (VAS) for pain; and the Quality-of-Life in Depression Scale (QLDS). Response was defined a priori as >50% decrease in HAMD17 total score from baseline to endpoint. Remission was defined as an endpoint HAMD17 total score <7.
Safety was evaluated on the basis of discontinuation rates, treatment-emergent adverse events, vital signs and laboratory analyses. Vital signs (supine and standing blood pressure, and pulse) were recorded at each visit. Treatment-emergent abnormal vital signs (high or low) were defined as follows:
- Supine systolic blood pressure—high: >140 mmHg and an increase from baseline >10 mmHg; low: <90 mmHg and a decrease from baseline >10 mmHg;
- Supine diastolic blood pressure—high: >90 mmHg and an increase from baseline >10 mmHg; low: <50 mmHg and a decrease from baseline >10 mmHg;
- Heart rate—high: >100 bpm and an increase from baseline >10 bpm; low: <50 bpm and a decrease from baseline >10 bpm;
- Weight—gain: >7% increase in body weight
from baseline; loss: >7% decrease in body weight from baseline.
Sustained hypertension was defined as meeting the criteria for abnormal high systolic blood pressure or diastolic blood pressure at three consecutive visits.
All patients who received >1 dose of study medication were included in the analyses of safety and efficacy. Mean changes from baseline to last observation in laboratory analytes and vital signs were assessed using an analysis of variance (ANOVA) with models that included treatment, ethnicity (Caucasian or African-American), investigative site and the treatment-by-ethnicity interaction as independent variables. The treatment-by-ethnicity interaction was the main basis upon which differential treatment effects between Caucasians and African Americans were assessed. Within-ethnic group contrasts between generic cymbalta and placebo were used to assess the clinical relevance of treatment effects. The rates of discontinuations due to adverse events, treatment-emergent adverse events, and treatment-emergent abnormal values in laboratory analytes and vital signs (categorical changes) were assessed in a similar manner, with the Breslow-Day test as the primary basis for detecting differential treatment effects between ethnic groups; Fisher’s Exact test used to test within-ethnic group differences between duloxetine and placebo.
Efficacy variables were also compared using the mean change to last observation analysis as previously described, with the addition of baseline value as a covariate. Longitudinal mean changes and categorical changes (estimated probabilities) were assessed using a likelihood-based, repeated-measures approach. The longitudinal mean change analyses included treatment, visit, investigator, baseline value, ethnicity, and the two-way and three-way interactions among treatment, visits and ethnicity. The percentages of responders and remitters at last observation were also tabulated.
With approximately 60 patients per arm (when using data from all studies) and using the powering assumptions outlined in the original protocols (mean difference of 3 points and a common standard deviation of 7), the power to test for differences between cymbalta (duloxetine) and placebo in mean change from baseline to last observation within the African-American cohort on the HAMDn total score was approximately 65%.
However, assessing the power of the present analyses to detect differential efficacy across ethnic groups (i.e., the power of the interaction tests) was more difficult. Previously published literature has been inconclusive regarding the existence and magnitude of antidepressant treatment differences between African-American and Caucasian patients. Therefore, no basis existed for a priori specification of a particular amount by which the magnitude of duloxetine’s advantage over placebo differed in Caucasian versus African-American patients.
Ascertaining power for detecting differential efficacy was further complicated by the fact that statistical theory for assessing treatment-by-stratum interactions is not well established. Therefore, it is not possible to accurately ascertain power for detecting differential efficacy of cymbalta drug in Caucasian versus African-American patients, although it is likely that such power was low. Two main factors contribute to this assertion. First, the uncertainty in estimating differential efficacy includes the uncertainty in the estimate of the treatment effects within each stratum.
Therefore, the uncertainty in the estimate of differential efficacy must be greater than that within either stratum. Second, it is reasonable to assume that a drug is somewhat effective in all strata. Therefore, the difference in efficacy between strata is probably smaller than the advantage of drug over placebo within each individual stratum. Consequently, tests of differential efficacy typically have considerable variability and the magnitude of the effect is small, leading to low power even in large databases such as the one employed in the present investigation.