During a Phase II schizophrenia study, our algorithms detected a logical inconsistency in the data. Our proprietary data monitoring system detected that scores on certain PANSS items were inconsistent with expected relationships for those items. Although the Sponsor and CRO were running “edit checks” to monitor this data, our empirically-based algorithms – which have been validated against thousands of in-study data points – automatically detected this inconsistency and generated a risk signal.
In this instance the risk level was deemed significant enough to merit site contact and through the discussion with the site it was determined that, while the rater was a skilled psychiatrist, their clinical understanding of certain items was different than that characterized within the outcome measure thus producing poor quality data.
Even though the protocol required rater training and CRO medical monitoring was in place, this type of problem emerges all too frequently. However, with Cronos involvement, sponsors have a cost-effective method to gain situational awareness and improve signal detection.