Questions about data manipulation [GxP / QC / QA]
I presented this topic at the GCC summit, which took place in Dubai in March of this year. Here is a summary:
The manipulation of a failing study involves performing an undocumented interim statistical analysis after unblinding the results, typically after half of the subjects have been analyzed. For example, if we have N=36 subjects, an interim analysis can be conducted after the analytical results of the first 18 subjects are known. If the point estimate (PE) or confidence interval (CI) after the first half of subjects is particularly low or high (e.g., GMR ~ 75%) and suggests that the bioequivalence (BE) study is failing, we identify those subjects that are causing the PE or CI to be skewed (high) and switch them.
As Ohlbe explained, the initially analyzed samples that do not fit will either be re-analyzed or diluted by a factor of 2. By doing this, the BE study will pass even if the product formulation is not bioequivalent, as there is a counterpart Y elsewhere in the data set with a T/R that eliminates or mitigates the issue with subject X.
Finding evidence of such manipulation is not easy because the paperwork, audit trials in the chromatographic system, audit trials in the statistical system, and plasma accountability are all clean. We can say that this switching process is an operation that leaves no classical trace, so it cannot be seen through audits or inspections. In other words, we don't have any evidence, only suspicions. Some of these suspicions include: 1) The manipulation revealing itself as trends (Cmax fingerprint), and 2) The manipulation revealing itself as profile similarities.
Dr. Anders Fuglsang has developed several software programs to detect such manipulation, including the Buster routines software, which detects the switch issue (trend manipulation), and the SaToWIB routines software, which detects profile similarities and issues with dilutions. For more details, you can refer to a published article by Dr. Fuglsang.
- Another one? Helmut 2023-07-19 09:04 [GxP / QC / QA]
- Confirmed Ohlbe 2023-07-21 12:59
- Another one? Ajay Gupta 2023-07-24 04:19