Evaluation of the sensitivity/specificity of statistical analysis strategies for detecting moxifloxacin-induced QTc prolongation

Abstract

In light of the implementation of the best practice guidelines as outlined in the ICH E14/S7B Q&A's, several strategies for enhancing the sensitivity of statistical analyses to detect drug-induced QT/QTc prolongation have been proposed. This study aimed to assess the sensitivity and selectivity of 3 standard statistical approaches. Moxifloxacin was administered orally at doses of 30, 80 and 175 mg/kg to male cynomolgus monkeys (n = 4) according to a Latin square design. Telemetry endpoints were monitored from 2 h prior to dosing to 24 h postdose. Three analysis strategies were employed to aid in the detection of QTc prolongation. Data were grouped into 3 analysis segments (0–6, 6–18, 18–24 h postdose). Approaches 1 and 2 utilized pairwise comparisons at each hourly interval within these segments or the entire segment based on the significance of the interaction effect, while approach 3 analyzed data pooled within each super-interval. Approach 1 employed a repeated measures analysis of variance (RMANOVA) strategy. Statistical significance was considered to have been attained when the p-value of each comparison was <0.05. Approach 2 utilized an RMANOVA analysis strategy examining the least squares mean differences between each treatment and control. Statistical significance was considered to have been obtained when these differences were greater than or equal to the least significant difference (LSD) for the study and the 95 % confidence intervals (CI) did not cross zero. Approach 3 utilized an analysis of variance approach with statistical significance considered attained when the mean difference between each treatment and control was greater than or equal to the LSD and the 95 % CI did not cross zero. In general, the 3 approaches performed similarly. Approach 3 demonstrated the highest sensitivity (most statistically significant differences identified) while approach 2 displayed the highest selectivity (fewest type I statistical errors). Approach 3 performed better than Approaches 1 and 2 in detecting QTc prolongation. Of note, only approach 3 identified every QTc change greater than 10 msec as significant. All 3 methods of analysis were adequate to detect moderate changes in QT/QTc, though differences in the sensitivity and selectivity were observed between analysis strategies.