Our Statistical Consulting & Website Development Projects

Below is a list of example projects that we have developed. We have categorized our projects into Clinical Trials and Website development, however, since many of clinical trials also involve integration of data collection via the web and the analysis tools pieces of the project could also be listed under website development.  Where required by confidentiality agreements specific details have been omitted.

Clinical Trials

Project 1 - Bayesian Adaptive Randomization

A Bayesian adaptively randomized study to provide the patients in the study with higher likelihood of receiving the superior treatment while obtaining scientifically valid conclusions. The randomization was continually updated and unbalanced to favor the treatment that, on average, had superior interim efficacy results. The randomization was subgroup specific and depended on patient prognostic variables. Measurements on the primary outcome where taken at three time points on each patient and the repeated measure aspect was included in the Bayesian model. This trial was a collaborative effort and patients enrolled at ten medical centers in the United States.

What we did - Development of Bayesian statistical methodology for adaptively randomizing patients. We developed the clinical trial simulation software and MCMC framework for obtaining samples from the posterior. Clinical trial simulation software was used to obtain operating characteristics of the design. Since this was a multi-institutional trial and the randomization depended on real-time data analysis we developed a secure website for collecting the patient outcome data and integrated it with the software we developed to perform the Bayesian analysis.

Project 2 - Interim Analysis Using Conditional Power and Bayesian Predictive Probability

This project was an interim analysis of time-to-failure data to compute the conditional power and the Bayesian predictive probability of a successful trial. With time-to-failure data, the censoring of the event must be included in the calculation and thus simulation was required in the calculation. Conducting the patient follow-up was extremely expensive and if, given the current data, after all the follow-up was complete it was unlikely that the trial would have a positive result the company wanted to discontinue additional data collection

What we did – We developed code to simulate the remainder of the study accounting for many aspects of the trial to obtain an accurate conditional power figure and Bayesian predictive probability. At the time of the interim analysis, the event had not been observed for most patients and the remaining patients could leave the study early for various reasons. In addition, for each patient there was a maximum follow-up and if the maximum follow-up was then the patient would remain censored.

Project 3 - Bayesian Hierarchal Model

A Bayesian Hierarchal model was implemented to combine information from several disease subtypes and the goal of the study was to determine if the experimental treatment provided sufficient activity to warrant a larger randomized study. By using a hierarchal design the client was able to conduct a single multi-institutional trial rather than dealing with the logistics of running multiple studies. In addition, since each disease, subtype was rare and this design was more efficient than separate trial.

What we did - We developed Winbugs code for the Bayesian hierarchal model to allow us to analyzing example datasets. This gave us the ability to illustrate how the design would react to various results and calibrate how information in one subtype influenced decisions in other subtypes. We developed simulation software for obtaining the designs average behavior. In addition, we developed a secure webservice to allow the client to integrate their own data collection portal with the advance statistical modeling and software that we developed.

Project 4 - Single Arm Bayesian Model to Account for a Prognostic Subgroup

A Bayesian model was implemented to borrow information across three prognostic subgroups. The primary objective was to show that the new treatment did not decrease the response rate because it was believed that the new treatment would have significantly fewer and less sever side effects. The model included treatment-subgroup interactions so that decisions where prognostic subgroup specific. For example, the trial could conclude to stop accrual in subgroup one while continuing in groups two and three. Historical data was available and the subgroups where not exchangeable, thus a hierarchal model was not appropriate.

What we did - Using the extensive historical data on over 300 patients treated with the standard of care, we developed informative priors on the control. We developed the necessary code to conduct the trial and simulated the design to obtain the operating characteristics.

Project 5 - Bayesian Adaptive Randomization with Covariates

A Bayesian model was developed to account for a patient's age, 50 years of age or less versus over 50. The primary response was treatment failure by day 30. It was unclear if the new treatment would benefit both age groups so the model developed included treatment-covariate interactions. The adaptive randomization was age group specific.

What we did -We developed the necessary code for the statistical modeling and simulating the design in C++. We conducted an analysis of historical data to obtain estimates of the effects of age on 30-day failure rate. We simulated the design to obtain operating characteristics under various Developed a webservice in ASP.NET to allow the company to conduct the trial using their in house electronic data capture (EDC) system. By integrating the EDC and the statistical calculation engine the company could obtain randomization and checking stopping rule using the most recent data.

Website Development

Project 1 - BabyAdventureGear.com

A retail sales website that sells baby gear. The website offers secure check out, order tracking and customer reviews.

What we did - Development of the entire website from the initial concept including design, search engine optimization, inventory control system, content management system, shipping utilities, statistical analysis of time trends to help predict when products should be ordered, SQL Server development and secure management side. The secure management side provides a real-time inventory system that gives the retail website access to the quantity of items on hand to prevent customers from ordering items that are not in stock. In addition, the management side of the website provides a history of notes on all orders so that the company’s associates can record changes and alterations as necessary. The website tracks user navigation and uses statistical analysis to suggest additional products based on other customers with similar interests.

Project 2 - SkyDiveSanMarcos.com

An online reservation system that is specific to the sky diving industry. A secure website that allows Skydive San Marcos to setup jump schedules, view reservations, track reservation changes and email customer confirmations via a secure website.

What we did - We developed the website portion dedicated to obtaining online reservations as well as the management side that allows the company to setup jump schedules, gift certificates and add/modify reservations. We developed the secure SQL Server database to allow restricted access based on a user's privileges.