Double reads in blinded independent central reviews (BICRs) are recommended to control the quality of trials but they are prone to discordances. We analyzed inter-reader discordances in a pool of lung cancer trials using RECIST 1.1.

In the past decade, the lung cancer treatment landscape has dramatically evolved, increasingly branching out thanks to better understanding of disease mechanisms of action, novel technologies, and some amount of serendipity in drug development. Today, approximately 2500 clinical trials registered on clinicaltrial.gov are about to recruit or are actively recruiting in order to investigate new therapeutics of lung cancer, offering new hope to patients for better survival and for improvements in quality of life.

Blinded independent central reviews (BICRs) are advocated in clinical trials to in-dependently verify endpoints and control bias that might result from errors in response or progression assessments. In the BICR settings with double reads, the medical images are reviewed by two independent readers blinded to the results of the other reader, the study treatment, the investigator assessment, and some pre-defined clinical information. The double-reading paradigm creates the possibility for discordance between the two readers; therefore, a third radiologist is involved to make the final decision of the evaluation outcome. The monitoring of reader performance is required by regulatory bodies to ensure data quality and reliability. At the trial level, a high adjudication rate could be an alert of poor quality at the study level, and a low number of endorsements from a given reader would raise concerns about the reliability of that specific reader. Therefore, relevant key performance indicators (KPIs) must be designed and implemented before starting the reads; these allow the study monitor to trigger corrective actions accordingly. A pooled analysis of 79 oncology clinical trials showed that the proportion of cases requiring adjudication among the 11 lung cancer trials included in the analysis was 38% (95% CI: 37-40%). However, this study was general to all cancer types and did not included details on discrepancy root cause or recently approved novel therapeutics. Considering the atypical response patterns provided by those drugs, we thought it prudent to provide an update on reader performance specific to new therapeutics in lung cancer.

Focusing on BICRs in assessing novel drugs, the aim of this study was to analyze a pool of lung trials using RECIST 1.1, document the proportion of reader discrepancies, and provide suggestions to aid in improving the read consistency of future trials by estimating relevant KPIs.

Inclusion Criteria:

• patient diagnosed with lung cancer and included in a clinical trial between 2017 and 2021
• trials were conducted with double reads with adjudication, and assessments were based on RECIST 1.1 guidelines
• The central reads performed using the same radiological reading platform (LMS; Median Technologies, France)

Exclusion Criteria:

• reader evaluation with less than 25 adjudications

Patient diagnosed with lung cancer and included in a clinical trial between 2017 and 2021. Assessments of read discordance are part of the quality program that tracks any inherent reader variability. Monitoring processes usually rely on several read performance KPIs, including the inter-reader discordance rate, the adjudication rate, the endorsement rate, and the error rate used to identify reader outliers. The adjudicator and the medical monitor document every discrepancy event along with the possible root causes, which here included four RECIST-derived categories along with two operationally based causes. These discordances were also categorized ac-cording to the type of expected discordance: "read error" or "medically justifiable difference".