Optimization of Patient Recruitment for Clinical Trials

The Problem

Over the last year, our client had been spending a substantial amount on recruiting patients for clinical trials. The pathway from when a patient inquires about the trial, is screened, has bloodwork done, and is then enrolled in the trial is a complex one. Our client wanted to develop a sophisticated approach to determine what is and what isn’t working on a city by city basis. Next, the client wanted to optimize spend across different advertising mediums.

There were two stages to the project:

First, the client wanted to better understand the substantial historical data they had from a number of disparate datasources (primarily various excel files). The overarching goal was to help the client better understand which advertising campaigns/mediums were performing best in particular locations so they could better direct budget dollars and more efficiently enroll patients. Secondly, the client wanted to run analytics on a bi-weekly basis to determine which campaigns are and aren’t working. This would support continued learning and optimization. 

Smaller projects included understanding where potential enrollees were dropping off in the process (essentially a funnel analysis) as well as monitoring the performance of individual phone-screeners in charge of sifting through applicants and directing the most qualified ones to on-site testing.

The Approach

The biggest challenge was aggregating all of the client’s random files into a coherent database schema. Once this was done, applying the logic necessary to join those various data files together was figured out next. Data ingestion and cleansing was the first phase in our overall approach (as is often a major part of any data analytics project). From there, translating the many client requests into a coherent set of reports was the next step in the process. Understanding what the data represented and turning that into simple metrics that could be visually tracked was the primary goal. Finally moving from simple historical tracking into predictive analytics, the last step was to construct a simple model to predict enrollments given a number of parameters.

The Implementation

First, an AWS server was spun up as a remote work-space that could host a database accessible by myself and the client. It was also used to host a website that would act as a portal for both uploading data as well as displaying reports and dashboards. MySQL was also utilized as the database tool to consolidate the client’s disparate excel files into a single database. For analytics, R was used – particularly in construction of the predictive model that would turn a user-inputted spend amount into an estimate for the number of enrollees. Finally, to display dashboards, the R-based framework Shiny was used. This allowed for a webpage to be quickly spun up on the remote server and provide a portal for displaying reports and ingesting new data from the client.