Rethinking the Clinically Based Thresholds of TransCelerate BioPharma for Risk-Based Monitoring

Abstract

Funnel plots are a flexible graphical technique that can form the basis for a risk-based strategy to assess data integrity, while considering site sample size, patient exposure, and changing risk across time.

Understanding the influence of individual variables contributing to multivariate outliers in assessments of data quality

“…First is the identification of anomalous data points. Often, this occurs when some clinically or statistically relevant threshold is exceeded . Initially, however, and prior to the development of any rules or thresholds, unusual data points are identified when our observations of the data do not align with our expectations.…”

“…Intervals with 95% and 99.7% confidence can be used to identify observations that exceed 2 (1.96, to be exact) and three standard errors from the average distance. Observations beyond the 99.7% limits can be interpreted as severe outliers, while those only beyond the 95% limits can be thought of as moderate outliers . Note that rules based on three standard deviations are commonly used in statistical process control to identify a manufacturing process that has gone “out‐of‐control.” As an aside, thresholds based on approximately two and three standards errors could serve as pre‐defined quality tolerance limits as defined in the revision of the International Conference of Harmonisation Guidance on Good Clinical Practice…”

“…Often, this occurs when some clinically or statistically relevant threshold is exceeded. [1][2][3] Initially, however, and prior to the development of any rules or thresholds, unusual data points are identified when our observations of the data do not align with our expectations. This was certainly the case in a multicenter animal study conducted by the National, Heart, Lung and Blood Institute.…”

“…In this manuscript, we propose methods to efficiently identify and review multivariate in/outliers. First, we recommend multiple statistical thresholds to characterize unusual observations with varying levels of severity . Should numerous anomalies exist, this approach provides the clinical team some means of prioritizing individual signals for review.…”

“…First, we recommend multiple statistical thresholds to characterize unusual observations with varying levels of severity. 2,3,11 Should numerous anomalies exist, this approach provides the clinical team some means of prioritizing individual signals for review. Next, we define contribution proportions which partition the squared Mahalanobis distance (which is equivalent to Hoteling's T-squared values) into the proportion of the outlier attributable to each variable used in the analysis.…”