Medical Company, California
Problem/Situation:
A client working for a medical technology start-up contacted DK Statistical Consulting asking about the possibility of our company completing statistical analysis for them. Their data were collected from medical devices which took thousands of readings a second from a patient over the course of several minutes, and their datasets were extremely large, over 4 GB in total and containing hundreds of millions of cases. Their goal was to use these collected data to accurately predict the presence of a specific medical condition among patients that was difficult for doctors to diagnose.
Approach/Methodology:
Part of this job involved initially determining which software program could be used to manage and clean datasets of this size most efficiently, and which would be most efficient for data analysis. We determined that SAS could most efficiently deal with datasets of this size, and used it for data cleaning, management, and analysis. Following this, we used SPSS to construct and runpredictive neural networks on smaller datasets.
Results:
The neural networks used for diagnosis of this medical condition were found to achieve success rates approaching 100%. These models were also validated in order to ensure that overfitting was not present, and the reason why success rates were so high. The results of these analyses allowed for a nearly perfect prediction of this medical condition on the basis of the collected data.
Government Subcontractor, Ireland
Problem/Situation:
A group who won a bid for a job provided by the Irish government contacted DK Statistical Consulting with a problem – they needed to determine the optimal locations for the placement of a limited number of mobile screening facilities for diabetic retinopathy. They didn’t know how to make these placements or how to cover the largest proportion of the population of Ireland.
Approach/Methodology:
Data relating to populations and population densities were found and compiled, and based on the expected radius that potential patients might travel to reach these mobile screen facilities, a series of permutations were run and analyzed, with the goal consisting of maximizing the percentage of the population that would have access to these mobile screening facilities.
Results:
The optimal solution was determined, and allowed for access by over 95% of the population of Ireland without increasing the number of mobile screening facilities allowed. The results of DK Statistical Consulting’s analysis was used by the government of Ireland in selecting the specific placement of each of these mobile screening facilities.
Food Safety Company, California
Problem/Situation:
This business approached DK Statistical Consulting with a complicated problem: they had a system of tanks, with liquid being drawn from these tanks, with the tanks being selected randomly, one at a time. The client needed to know the size of the tanks necessary such that a specific large number of draws will produce a specific level of confidence that none of the tanks empty at any point.
Approach/Methodology:
For this type of problem, DK Statistical Consulting decided that writing a short computer program simulating these tanks and draws would be the ideal solution. We wrote a short computer program of under 100 lines in C++ and ran the simulation 1,000 times, varying tank sizes with each iteration.
Results:
The results of these computer simulations allowed for the calculation of the exact tank size necessary to achieve a specific level of confidence that none of the tanks empty at any point based on the specified number of draws. This allowed the company to produce future designs that will be exactly in line with their needs, and not waste money or manpower on tanks larger than they need, as well as to avoid problems that may result from having tanks smaller than they require.