Case Studies
Value Creation Through Statistical Thinking
The truth is, many examples of real value I've been able to provide have not been specifically requested. They are a result of me being involved in the study process, observing a problem or an opportunity for improvement, and implementing the change. That is the value created by involving a biostatistician in the team from the beginning.
Case 1: Correcting Misleading Efficacy Estimates
A company was tracking progression-free survival (PFS) in a spreadsheet and manually calculating without accounting for censoring, resulting in inaccurate median PFS estimates driving decisions and being shared with potential investors. By being part of the program team, I realized the error when the estimates were discussed, and implemented proper time-to-event methods to provide reliable results that supported sound business decisions.
Case 2: Right-Sizing Rare Disease Study Design
A company projected it would take 5 years to enroll their rare disease trial, a timeline which was not attractive to investors. To decrease the time to a potential filing, I restructured the study design to include an interim analysis which had 50% power to show a significant difference at the interim, which if successful would reduce time to approval by 18 months. We also simulated subgroup enrollment scenarios to refine feasibility assumptions - most CRO statisticians assume uniform enrollment into a trial (meaning the patients/ month enrolled are the same from month 1 until the end of the study). This simple but common error can cause inaccuracies in the sample size calculation, simply because no one thinks to ask.
Case 3: Assessing Data Cleanliness Beyond the Metrics
I reviewed data from a several hundred patient basket study run by a large CRO and discovered that the CRO data management had not issued any review queries, only auto-queries and sponsor queries had been issued over 3 years. I escalated the issue within the CRO management and we came up with a remediation plan to get the data reviewed and current. Don't let this happen to you - somehow CRO metrics usually look great but if you ask when the database can be locked, people are surprised that it can take 6 months or more, even if your metrics say 95% clean. It's important to interrogate your data cleanliness early and often.
Case 4: Adaptive Design to Reduce Sample Size
A study team proposed using Simon 2-stage design across many arms of a Phase 2 study, which would have required 75 patients/ arm. Instead, I suggested a Bayesian efficacy monitoring approach, saving more than 150 patients while preserving statistical rigor and dramatically improving timelines and cost to determine proof of concept.