Proven Data Quality Solutions

Our algorithms detected a clinically unusual presentation across multiple scores on an instrument to measure symptom severity in schizophrenia. There were multiple inconsistencies from the perspective of item correlation (in this instrument several items measure aspects of the same clinical construct), and lower than expected variability across visits given the population under consideration and the stage of the trial. Following review by one of our expert clinicians, a decision was made to contact the rater to understand her rationales for scoring the severity as she did across these visits. In many cases, the rater’s selection of scores did not correspond with either the anchor points for the severity level or the rater’s own account of the patient’s symptom. It was further noted that the rater had not used the scale anchors and had not completed a full interview; instead relying on a “standard psychiatric interview” that took 5-10 minutes. Following feedback and retraining, the rater was restored to the expected clinical trial practice of using the structured interview guide and anchor points to score.

In this case, loss of sensitivity and the resultant reduction in signal detection was avoided. This systematic method to detect and correct poor rating practice in real-time is essential for high quality and unambiguous study data.

Cronos Clinical Consulting Services, Inc., an IQVIA business, analyzed data from a trial investigating the impact of a compound on cognition and negative symptoms in schizophrenia. The primary and secondary outcome measures were the MATRICS and the PANSS, respectively.

While there are a variety of other factors that can impact power and sample size calculations in clinical trials, we found that error was present within and across scales, obscuring the real magnitude of the impact of the study compound. This error impacted approximately 20% of the assessments.

Non-informative data has a greater impact on data quality than missing values, and, in the case of this study, when we subtracted this inconsistent/improbable data from the initial sample size calculation the power decreased from .80 to .73. Retaining the non-informative data decreased the power calculation from .80 to .60, which was unacceptably low to demonstrate compound efficacy.

Targeted recommendations were provided in order to proactively minimize the likelihood of identified issues recurring in future studies.

Cronos Clinical Consulting Services, Inc., implemented a training module for the CGI-Severity and Change scales in an ongoing global depression trial in which ratings error associated with these scales was identified. The scales rely on a rater’s clinical experience with the population of interest, and their ability to translate a global clinical impression into one of 7 sparsely worded anchor points. Not surprisingly, correlations between the CGI scales and other measures of illness severity in randomized clinical trials vary widely.

The training reflected the aggregation of knowledge obtained through collaboration with thousands of raters to identify and address the challenges they face, and to determine the most effective responses to address those challenges. For example, how to address discrepancies between other symptom severity measures as you formulate your CGI rating.

CGI error rates decreased following implementation of the training, thus increasing reliability of the data.

Cronos Clinical Consulting Services, Inc. supported blinded third party adjudication reviews of primary endpoint efficacy assessment data. Blinded review of audio-recorded efficacy assessments was performed to confirm results and identify error or deviation from adherence to GCP and best assessment practice. Project specifications included: 5 protocol designs, 300 study sites, 25 countries, 15 languages, data collection for 30,000 unique subject visits, and complex study efficacy entry and protocol rollover criteria.

CIRP automated workflow allowed a site to request an adjudication, and matched the appropriate available resources to conduct a blinded review. The system received data streams from site-based tablets, an IWRS provider, an EDC, and 6 different external vendors, and analyzed it to perform complex calculations to proactively identify and suggest the need for an adjudication (while also allowing the site to raise an issue independently). All adjudication results were pooled and further analyzed (applying methods of risk-based monitoring) to identify training needs at the individual, site, country, and study level.

Integration of the CIRP system resulted in effective implementation of the adjudication process, allowing errors to be corrected and remediated over the course of the trial, and eliminating the chance for a mistake to propagate and evolve into an undetected trend that may negatively influence the study outcome.

Click here to find out more about CIRP.