Category: Health Care Research
Data Desk® Deciphers Role of Delay in Breast Cancer Diagnosis
For San Francisco surgeon Bill Goodson, time is of the essence - the time, that is, that elapses between the onset of breast cancer and its diagnosis. Delayed diagnosis means delayed treatment and a head start for the cancer. He is interested in identifying the factors that promote lag time in diagnosis, and he began his investigations with a study of about 400 of his own patients at the California Pacific Medical Research Institute. What to do with the data from the facts of these cases? "I have very little training in statistics," Bill says. "I am essentially a non-statistician - but I love to play with numbers." For this reason he turned to Data Desk® with follow-up critique from statistician Dan Moore at the institute.
In an initial study involving some 400 patients in his practice, Bill looked for delays in diagnosis of 1 month to 3 years. He moved his data from File Maker Pro into Data Desk and used its exploration capabilities to isolate about 40 (1 in 11) patients whose diagnosis had been delayed. Seeking further for information about the causes of delay, he was able to identify some key factors correlated with delay: the patient making the first discovery of a lump, a negative mammogram, and the use of hormone replacement therapy.
"The function of the software I use most," he says, "is the contingency table. Once I have made data fields for types of delays, HRT or not, type of cancer, type of surgery, etc, etc, I then just click and check on multiple combinations of factors. To do that in Filemaker would have been unbelievably tedious and probably riddled with a lot of errors - meaning that I would have to do a lot of rechecking on how I keyed in queries. In Data Desk, I put in the two factors I want to compare, I get a result in a few seconds, the table is labeled with what I have just checked, and the statistics are all there. Faster, easier, and a lot less stress."
While his findings are fairly straightforward, he says their implications are tricky to assess. He thinks that perhaps doctors of women who find their own lumps are mislead by negative mammograms that follow on the discovery. In the case of hormone use, he suggests that women who use hormone replacement may tend to attribute any lumps they might find to the effects of the hormones. These initial findings led him to conclude that every palpable lump should be biopsied.
For a second study of different patients Bill wanted to investigate a more subjective question. More than a third of breast cancers are discovered by physical examination of the breast, and he had begun keeping records on what breasts feel like during examination. He wanted to relate characteristics of texture with delayed diagnosis. While he says that the mammogram is still the best test available, it has its shortcomings. His idea was to develop a "vocabulary" of texture that would alert doctors to characteristics that indicated caution in interpreting diagnostic findings. In this investigation, he discovered the power of using Data Desk to write scripts for the analysis of his database, and he was able to develop a system for comparing the breast characteristics of one woman with those of the next.
Currently Bill Goodson is using a database with information on more than 700 patients to reveal something about learning curve doctors experience in the so-called "sentinel node," the lymph node that when biopsied can guide surgical treatment of the patient. "I set up the contingency table by dates. Then I just use selectors to modify the table so I can look at different techniques, different surgeons, and different locations. Once I have figured out how to get the basic comparison, the selectors allow me to see what happens when I change one thing - or change it back."
While he has not yet explored all the software's functional capabilities, he says, "What is important is that I have limited time. Data Desk gives a reliable analysis with a lot of flexibility to allow me, the clinician scientist, to look at the data many ways with a system that avoids errors."
Name: Bill Goodson
Affiliation: California Pacific Medical Research Institute