CRO Rescue

Case study

PHASTAR began a relationship with a top ten pharmaceutical company carrying out statistical consultancy and were subsequently asked to bid to be a partner in a traditional outsourcing model (i.e. the reporting of clinical trials is done independently by PHASTAR using PHASTAR systems and processes). An RFI/RFP process was undertaken, and we responded, giving details of how we would deliver the results of the clinical trials and related deliverables. We were then selected to be one of two strategic partners after agreeing processes, pricing and governance structure. A two-layer governance structure was established, with a quarterly executive review.

One of the first projects was to rescue 12 studies that had been placed with another supplier. The sponsor had lost confidence in the ability of that supplier to deliver results on time without errors. At the request of the sponsor, we worked with the other supplier to determine useable deliverables and where we needed to redo work. It was clear that quality control procedures were lacking and that the main task was to validate all the deliverables, correcting errors as they were discovered.

One of these studies was particularly interesting as it involved both Response Evaluation Criteria in Solid Tumours (RECIST 1.1, soft tissue) and Prostate Cancer Working Group 3 (PCWG3, bone) criteria. With the addition of bone lesion assessments and it being a rescue study with tight deadlines added extra levels of complexity.

Randomisation of the study was stratified on 2 covariates. Although it was expected that there would be enough events in each strata to allow for a meaningful analysis, we had to investigate a pooling strategy. Many of the efficacy endpoints were explored by Blinded Independent Central Review (BICR) and investigator assessments.

Investigator assessments for RECIST and bone lesions were collected on different protocol schedules so comprehensive visit windowing rules had to be applied to align the assessments so an overall radiological visit response could be created.

Target, non-target and new lesion responses had to be determined in order to derive the overall RECIST visit response, much like any oncology study. The bone lesion data was used to create a bone lesion visit response.

Taking both the overall RECIST visit response and overall bone lesion visit response we derived the overall radiological visit response which was the main parameter used in the primary and key secondary efficacy analyses.

The primary endpoint of the study was to determine the efficacy of the IP which was assessed via radiological progression-free survival (rPFS) by BICR. For this endpoint to be assessed we needed to create a parameter comprising of the number of days a patient was radiological progression-free. Patients who did not progress via RECIST or bone or die at the time of the analysis were censored at the time of the earliest date of their last evaluable RECIST or bone scan assessment. However, if the patient progressed or died immediately after 2 or more consecutive missed visits for either soft tissue or bone assessments, the patient was censored at the earliest of the previous RECIST assessment or previous bone scan assessment prior to the two consecutive missed visits (if the RECIST and bone scans were completed at different visits). If the patient had no evaluable visits or did not have baseline data, they were censored at Day 1 unless they died within 2 visits of baseline in which case their date of death was used.

The key secondary objectives included objective response rates (ORR), time to pain progression (TTPP) based on Brief Pain Inventory- Short Form (BPI SF) and overall survival (OS). Both confirmed and unconfirmed ORR were investigated, via the radiological response and best radiological response. Pain progression was challenging as it involved working with huge volumes of custom questionnaire data where we had to incorporate client specific requests. OS was explored in the same manner as other RECIST oncology studies.

rPFS, TTPP and other secondary endpoints were analysed via a stratified log rank test. Hazard ratios and confidence intervals were calculated using a Cox Proportional Hazards model adjusting for the covariates selected in the primary pooling strategy. Kaplan-Meier (KM) survival curves were also produced.

rPFS was further investigated with a variety of sensitivity analyses being performed to assess: Evaluation-time bias, Attrition bias, Censoring bias, Ascertainment bias, sensitivity analysis using unequivocal clinical progression in addition to radiological progression, sensitivity analysis for confirmation of bone progression and sensitivity analysis censoring patients with subsequent therapy or discontinuation of study drug.

Subgroup analyses were also completed for the rPFS endpoint to assess the consistency of the treatment effect across potential or expected prognostic factors. This was accomplished by using the same analysis techniques as described above for rPFS. This was further presented in the form of a forest plot.

ORR were analysed via odds ratios using logistic regression adjusted for covariates selected in the primary pooling strategy. The associated 95% profile likelihood CI and p-values were based on twice the change in loglikelihood resulting from the addition of a treatment factor to the model. Note that if there were less than 5 responses across both treatment groups then a Fisher’s exact test using mid p-values was presented.

The PRO endpoints such as FACT-P (FACT-P total score, FACT-G total score, TOI, FAPSI-6, FWB, PWB, PCS), BPI-SF change from baseline scores were analysed using a mixed model for repeated measures (MMRM) analysis of all the post-baseline scores for each visit. Restricted maximum likelihood (REML) estimation was used. An unstructured covariance matrix was used to model the within-subject error and the Kenward-Roger approximation was used to estimate the degrees of freedom. A number of other covariance structures were in place if the unstructured covariance structure failed to converge.

This study was incredibly challenging with many complex endpoints analysed. However, with close collaboration with the sponsor we were able to have a successful study and submission outcome.

Our agreement with this sponsor has a bonus-penalty scheme – we get a bonus if all deliverables are on-time and error free. We’re very proud that we have had a bonus every year since the start of the agreement. The managers frequently tell us that we are their partner of choice – no other company has ever received a bonus and we are regularly told that we are more of a partner than a supplier. On completing a major project, one of the therapeutic area vice presidents wrote to us saying “Simply Awesome! The relentless passion and drive is much appreciated”.

Learn more about our services
We are experts in study design, statistical analysis, data science, data capture and reporting for clinical trials