| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
University of Cincinnati Cincinnati, OH 45236
I read the article by Parisky et al. on thermography [1] with some interest, and I am a little puzzled by their data and conclusions.
The sensitivity (true-positive rate) of thermography is given as 97%, and the specificity as 14%.
As I understand it, specificity is the percentage of benign (or normal) cases correctly called benign (or normal); the false-positive rate is (1–specificity). If this is indeed the case, the false-positive rate is 86%. This is verified from table 1. The positive predictive value is 24%.
Positive predictive value is directly dependent on the incidence of the disease in the examined population. From table 1, one can estimate that the population examined contained 22% cancers. Therefore, the positive predictive value increased by 9% (from 22% to 24%) when thermography was added to the mix.
When the investigators excluded the cases detected by microcalcifications alone, the true-positive rate improved to 100%, the false-positive rate decreased from 86% to 80%, and the positive predictive value was 27%. Based on the data in table 4, the positive predictive value for this subset of cases (before the thermogram interpretation) was 23% (110/479).
As I understand the protocol, the evaluators were instructed to use the mammograms to localize the lesion on the infrared image. How would this modality function in a random situation in which the region of interest on the mammogram was not preselected for the thermogram (even though the case had been selected for biopsy)? If thermography cannot separate the region of interest from the noise, there is no convincing evidence that it contributes in a meaningful way.
In my opinion, these data strongly suggest that thermography seems to be operating only a little above the random chance line and does not seem to offer any clinical help in differentiating between benign and malignant abnormalities detected at mammography.
References
USC/Norris Comprehensive Cancer Center and Hospital Los Angeles, CA
90033
Battelle Healthcare Products Columbus, OH 43201
To fully respond to Dr. Moskowitz's letter, we must first to review the diagnostic role of infrared imaging as intended in our study of a computerized infrared imaging system (BCS2100, Computerized Thermal Imaging, Ogden, UT) [1]. Computerized dynamic infrared imaging, as performed with the BCS2100, is not a screening tool, and it is not used to diagnose or characterize malignancy. It is used in conjunction with mammography as a means of reducing the high number of biopsies performed every year on benign breast lesions. Suspicious lesions are targeted with the BCS2100, which measures the infrared radiation typical of malignancies. If there is no increase in the infrared radiation, the device assigns the lesion a negative infrared test result, indicating that the lesion has a high probability of being benign. Because benign lesions have no specific infrared characteristics, the device cannot be used to find them in the random manner suggested by Moskowitz.
Moskowitz is, of course, correct in stating that a test's predictive values (both negative and positive) are dependent on the incidence of disease in a population. For this reason, predictive values can be misleading indicators of the performance of a diagnostic test. For example, in our study of lesions scheduled for biopsy, the use of infrared imaging resulted in an increase in specificity over current clinical practice from 0% to 14%, implying that infrared imaging could be used to prevent one of every seven unnecessary biopsies. In a study population such as ours in which approximately 21% of the lesions scheduled for biopsy turned out to be malignant, a 14% increase in specificity translates into only a 2.6% increase in positive predictive value; however, the major benefit is the increase in specificity. One might question whether a 2.6% increase in positive predictive value is a clinically significant benefit, but there is no question that preventing one in seven unnecessary biopsies is clinically significant. We further contend that the potential to prevent one of every 5.5 unnecessary biopsies in our restricted group (specificity of 18%) is also clinically significant.
We appreciate Moskowitz's comments and this opportunity to offer our perspective on the appropriate measure for characterizing the clinical value of infrared imaging. In our opinion, the value of infrared imaging lies in terms of preventing unnecessary biopsies. We hope that this response has adequately addressed Moskowitz's concerns.
References
CiteULike 
Complore 
Connotea 
Del.icio.us 
Digg 
Reddit 
Technorati What's this?
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
