Success stories

The following success stories demonstrate how IDDI innovative Biostatistics and eClinical solutions are key in the successful conduct and outcome of your clinical trials.

Bayesian Study Design

Success stories

A Bayesian Study Design For Seamless Transition Between Phase II And Phase III

Bayesian study designData collected by a sponsor suggested a possibility of a differential treatment in subpopulations defined by levels of a biomarker. A Bayesian study design has been developed, which allows the sponsor to conduct a randomized Phase II trial to corroborate the finding and to seamlessly continue to Phase III in the subpopulation, which benefits from the treatment. The study design was consulted and accepted by a regulatory agency.

Flexible Study Design

Success stories

A Flexible Study Design with Randomization to Placebo

Flexible study designThe maximum tolerated dose of a new drug, supposed to alleviate some adverse effects of anti‑cancer chemotherapy, was to be established in a phase I trial. To assess the safety of the drug, information on the background toxicity rate in the enrolled sample had to be collected. By using state‑of‑the‑art methodology, a flexible study design was proposed for the trial. It combined the use of the continual reassessment method on a continuous dose scale with a concurrent randomization to placebo. The properties of the design were investigated by conducting a simulation study, which allowed to fine-tune the final form of the trial design. The trial has been completed successfully.

Discovering New Prognostic Biomarkers For Breast Cancer

Success stories

Prognostic Biomarkers

New prognostic biomarkers for breast cancer: A range of biological markers were analyzed by polymerase chain reaction and immunohistochemistry in tissue microarrays from patients treated in several past and ongoing clinical trials for breast cancer. The obtained marker measurements were combined with the clinical data and analyzed using survival analysis methodology, including some advanced modeling techniques, to investigate whether any of the markers had prognostic value. The investigation was successful for at least one of the potential markers analyzed. The project is ongoing to validate further biologically interesting candidate biomarkers.

Meta-Analysis To Assess Efficacy in Colorectal Cancer

Success stories

Meta-AnalysisIn this meta-analysis, tumor responses and survival were analyzed combining the data from the different trials using patient individual data. The statistical methodology was based on the classical notion of stratification, consisting of estimating a treatment effect within each trial, and then overall. A statistic for heterogeneity between the trials was calculated. A test of overall treatment effect was calculated. The following quantities were used in the calculation: O, which is the number of untoward events observed in the treatment group, E, which is the number of events that would be expected in the treatment group if there were no differences between treatment and control; and V, the variance of the number of events. Those data were shown graphically in a forest plot.

Applying Likelihood Method For Data Safety Monitoring

Success stories

Data Safety Monitoring

Data safety monitoringFormal data safety monitoring, often performed by independent committees of physicians, biostatisticians and ethicists, has become common in modern clinical trials. Safety monitoring often includes reading of many pages of tabulated adverse events classified by body system, type and severity. Monitors look for within treatment incidence and between treatment differences in incidence that may be of concern. Frequentist statistical methodology is not appropriate for this type of surveillance due to multiplicity issues and the inappropriateness of the background repeated sampling assumption. A safety monitoring committee in an international ophthalmology clinical trial used the principle of support and support intervals based on the log likelihood function for incidence parameter conditional on the data at hand. Rates were calculated as poisson random variables and support methods were used for both incidence and treatment differences.