Clarkston Consulting https://www.facebook.com/ClarkstonConsulting https://twitter.com/Clarkston_Inc https://www.linkedin.com/company/clarkston-consulting http://plus.google.com/112636148091952451172 https://www.youtube.com/user/ClarkstonInc
Skip to content

Creating a Statistical Model for Survival Analysis Across Different Drug Regimens

Clarkston’s Digital, Data, and Analytics team recently executed a project for a biopharmaceutical client, creating a new statistical model to understand the length of the patient’s journey on the client’s immunotherapy drugs versus the chemotherapy alternative. 

The first objective for the Clarkston team was to edit the existing Structured Query Language (SQL). As this query was used for a different task in the past, it was inadequate and needed to be revamped to include data for 15 different drugs, categorize them according to drug type, calculate the duration for each, and correctly order which claims were included in the data. Clarkston also accounted for events that contributed to a patient leaving the line of therapy, which wasn’t in the initial data. Initially, we conducted non-parametric survival analysis using the Kaplan Meier curve to understand the general nature of the survival curve itself. We also conducted meetings with business stakeholders who had conducted prior research and interviewed a clinical data scientist who had intimate knowledge of the drug lifecycle to gain insight on different events that occur. 

Download the Statistical Model Case Study Here

Using insights gained through those meetings, we gained an understanding of the nature of the hazard ratio, which aided in the type of statistical model to use. The categorization of this information was centered around if the events for patients were constant, relatively increasing or decreasing, or proportional. Each drug group had its own behaviors, and accurately classifying them was essential to modeling the duration of the different groups. The group successfully fit competitor chemotherapy data with the Weibull model (proportional, one-directional hazard ratio), and the client’s drugs favored the Exponential model (constant hazard ratio), as modeling the hazard ratio constantly fit best.  

The client’s executive team who initially commissioned the analysis deemed the modeling successful, as they used the information in their black-boxed model. The process was productionized and repeated for client-patient forecasting. 

Download the Statistical Model case study here. Learn more about our Data + Analytics Services and Clarkston’s HCP 360 Insights to Actions Platform by contacting us below. 

Contact Us to Learn More

Contributions from Isaiah Mullins

Tags: Case Study, Clinical Operations, Data & Analytics, Data Quality, Data Operations