In the end, I provide a way of handling similar problems in corporate data intelligence.
According to the study, http://www.nejm.org/doi/full/10.1056/NEJMclde1212888 nearly 31% of cases are misdiagnosed in breast cancer detection. The initial screening identifies 100% increase in cases identified as malignant tumor but in late stage we find there is a net decrease of 8%, over a study period from 1976 to 2008. In a way it should not be surprising from a statistical interpretation point of view. If you are willing to increase false positives, you will identify more cases but non-linearly the true positives will decrease faster (Remember, true positives is not 100-false positives!)
In the article in New England Journal of Medicine which published fairly detailed study on a retrospective sample of 30,000 women (the largest study) that used “Surveillance, Epidemiology, and End Results data to examine trends from 1976 through 2008 in the incidence of early-stage breast cancer (ductal carcinoma in situ and localized disease) and late-stage breast cancer (regional and distant disease) among women 40 years of age or older”.
“To reduce mortality, screening must detect life-threatening disease at an earlier, more curable stage. Effective cancer-screening programs therefore both increase the incidence of cancer detected at an early stage and decrease the incidence of cancer presenting at a late stage.”
The key take away I want to bring to discussion is that certain decisions are popular even in corporate world and if the organization does not use fully the information strategy there will be emotional decisions made that will have a long term effect on the organization because of dominance of false positive loaded decision making or two few true positives loaded decision.
Why does this emotional roller coast in life and wastage of resources happen? From one side, the reasoning is “every individual matters”. On the other side of the argument is that “the incremental value for the society as a whole involves emotional suffering of having the disease when in fact it is not there – fear of ghost – and also the pain/suffering/waste of additional tests/treatments/ and family difficulties of ‘imaginary ghosts’.
So how to reduce such high incidence of false positives? and what is the right decision?
Look at the measures from many different angles. A simple majority voting should be good enough if ever one has to go though the majority angles interpretation for consistency of direction of message.
Recommend mammography after weighing in additional factors to triage the cases better and more importantly involve the patients and families in explaining risks of type I and type II errors. The system somehow is revolving with the understanding that it is difficult to explain these confusing errors and associated risks.
Furthermore, weigh in the full emotional and material cost of drawing a line on how much to accept as false positives and how much to accept as false negatives so that every body is focused on the clear unemotional decision.
End users should be aware of the misdiagnosis rate or miss-classification of consumers and emotional and material cost attached to it. In the case of breast cancer end users are women and households. In the case of corporate world, the different department heads who are being compensated for the success or failure of programs.
What is the take away for corporate analytics?
In the business world, this is a much easier decision making path as everything has to come down to simple dollars and cents (sense)! So unit cost and revenue differential for every false positive and every false negative can easily translate into final measure that can be used to compare decisions for optimal decision making.