HOME HELP FEEDBACK SUBSCRIPTIONS ARCHIVE SEARCH		QUICK SEARCH:   [advanced] Author: Keyword(s): Year:  Vol:  Page:

Electronic Letters to:

Glaucoma: Pamela A. Sample, Michael H. Goldbaum, Kwokleung Chan, Catherine Boden, Te-Won Lee, Christiana Vasile, Andreas G. Boehm, Terrence Sejnowski, Chris A. Johnson, and Robert N. Weinreb Using Machine Learning Classifiers to Identify Glaucomatous Change Earlier in Standard Visual Fields Invest. Ophthalmol. Vis. Sci. 2002; 43: 2660-2665 [Abstract] [Full text] [PDF]

eLetters: Submit a response to this article

Electronic letters published:

Correction of Microsoft Excel Date Error Indicates Machine Learning Classifiers Are Equivalent to St

Pamela A. Sample, Michael H. Goldbaum   (21 May 2003)

Correction of Microsoft Excel Date Error Indicates Machine Learning Classifiers Are Equivalent to St 21 May 2003
Pamela A. Sample University of California, San Diego, Michael H. Goldbaum Send letter to journal: Re: Correction of Microsoft Excel Date Error Indicates Machine Learning Classifiers Are Equivalent to St psample{at}ad.ucsd.edu Pamela A. Sample, et al.	I am writing this letter to point out an error in the data analysis for our paper "Using machine learning classifiers to identify glaucomatous change earlier in standard visual fields," published in the August 2002 issue of IOVS.1 In this paper we reported that machine learning classifiers predicted conversion to abnormal visual fields on average 4 years earlier than traditional Statpac-like methods. When reviewing the data for inclusion in a new study, I discovered an error in the dates of conversion based on the Statpac-like methods. Correction of the error has reduced the reported superiority of the machine classifiers to identify change in visual fields earlier. Readers should be aware that the paper was incorrect in this conclusion. The support vector machine with a Gaussian kernel (SVMg), the best performing classifier, now predicts visual field conversion a mean ± standard error of 0.40 ± 0.51 years earlier than the Statpac-like techniques. In searching for the reason for the error, we found that each date for conversion by Statpac-like methods was in error by exactly 1462 days (4 years and 1 day). Further investigation found that Microsoft uses a default system date of January 1, 1904 for Macintosh versions, because earlier Macintosh computers did not support dates before January 1, 1904. Microsoft continues to use January 1, 1900 as the system date in the Windows and MS-DOS versions of Excel. When the column of dates was copied from the Windows version to the Macintosh version, Excel added 1462 days to each date in the column. We were unaware of this problem in the use of Microsoft Excel, because Excel does not warn users of this difference in date systems. One must know of this problem in advance to find information in the Excel help file or the Excel Internet support site (see http://support.microsoft.com/support/kb/articles/q180/1/62.asp). We think it important that those who use Microsoft Excel should be made aware of the potential for a change in the date and the circumstances under which that change can occur. The corrected results continue to support the potential applicability of these machine classifiers for analysis of combining data from visual function tests, optic disc topography, retinal nerve fiber layer thickness, and risk factors such as intraocular pressure. Pamela A. Sample Michael H. Goldbaum University of California, San Diego References 1. Sample PA, Goldbaum MH, Chan K, Boden C, Lee T-W, Vasile C, Boehm AG, Sejnowski T, Johnson CA, Weinreb RN. Using machine learning classifiers to identify glaucomatous change earlier in standard visual fields. Invest Ophthalmol Vis Sci. 2002;43:2660-2665.