"Error Check" Program for Windows 3.1
Yipeng Chen, B.S., Ying Wang, B.S., Willa Drummond, M.D., Johannes van der Aa, Ph.D., Gordon Gibby, M.S.,M.D., James Bosworth M.S., David Kays, M.D., and Sharma Chakravarthy, Ph.D.
University of Florida Colleges of Medicine and Engineering
Introduction: Verifying and validating the accuracy of data entered into large databases by hand typing (e.g. research data files), by electronic upload from peripheral equipment (e.g. patient monitors), or by direct download form other computer systems (e.g. Clinical laboratory results) is a "front-end" problem that has been insufficiently addressed in the current nationwide effort to "computerize the health care industry". Currently, verification of electronically acquired or typed data relies heavily on human inspection (i.e. bedside nurses or research scientists' staff). This hands-on approach to the "verification and validation" problem in medicine and physiological sciences is an impossibly impractical solution, given that verification and validation of electronically uploaded data requires far more than a single real-time FTE, even if the verifying individual had no other responsibilities. Simple "range-limit" checks, available in Access© or SAS© are insensitive to some types of errors and require both programming skills and prior knowledge of expected results.
Methods: We have written software that finds erroneous values in large electronic data sets by using statistical techniques (based in linear mathematical theory) to calculate whether a given datapoint is likely to be incorrect. The software adapts to statistical algorithm used in combination with other methodologies for edge detection in image intensification problems (1). The algorithm calculates the standard deviation of the proceeding 100 observations in a given column of the data set. If <100 previous observations exist, the program uses the points available. The operator then sets the defined "window" of sensitivity and specified by setting a variance () between 2 and 9 using the "dialog box" format. The smaller the variance, the smaller is the allowable range for "error".
Results: Our "ERROR CHECK" program can examine a database of 54 columns x 350 rows in less than 2 seconds when run on a 100 MHz 486 PC. The program will also run on a 386 PC,but less rapidly. "ERROR CHECK" can work within Microsoft Office® to translate any type of data file (Lotus, ASCI,etc.) into the format used by the "ERROR CHECK" (text files in Excel®). "ERROR CHECK" software can detect trend-break, column-shift, decimal point, typographical, out-of-range and other types of errors (i.e. string variables) in either hand-typed or electronically uploaded data sets.
"ERROR CHECK" incorporates an output report format (laser printer) that identifies sequentially the location of the erroneous point, the possibly erroneous value, then gives a blank line for the human varifier and one for the operator who overwrites corrections into the computer database files.
Discussion: To our knowledge, this statistical approach to error detestion has never been adapted for application to the numerical "spread sheet" type of datsets commonly generated in classical science and medicine (and perhaps business). We think this is also the first time that a software package has been written which incorperates user-definable variance capabilities using a simple point and click dialog box that can smoothly write complex error detection algorithms onto "open-platform" software such as Microsoft's DOS/Windows/Office© systems. Thus, "ERROR CHECK" can be installed by inserting a 3.5" floppy onto a DOS/Windows/Office© configured 386 or 486 computer, and clicking on the "ERROR CHECK" icon to open the dialog box.
Our "ERROR CHECK" program will alow the human "signing off" on the data to focus specifically on the points that "ERROR CHECK" has deemed likely to be erroneous or artifactual. In medical applications, when used on-line for varifying and validating uploaded clinical data (as from bedside monitoring equipment or from directly downloaded laboratory information), the report could be used at intervals (i.e. at the end of the nursing shifts) to identify and judge potentially problematic values ( = "verification"), thereby assisting the nurse who "signs off" on the clinical data, prior to permanent archiving ( = "validation").
Previously, a 10% error rate in hand-entered database was considered "inevitable" and "acceptable" (personal communication, MJ Resnick, Ph.D. regarding experience with the State of Florida Perinatal Database constructed in the late 1970's through the 1980's). Clearly, such a high error rate is unacceptable for data relating to patient care matters of scientific research.
Faciliating "applications software" such as "ERROR CHECK" may speed acceptance of bedside computerization by "health care providers" who work at the "front-end" . Such software may provide help for research scientists whose funding doesn't premit sufficient staff time to check large databases.
Phone: (904) 392-4195 Fax: (904) 392-4533
Lim JS: Two Dimensional Signal and Imaging Processing, Chapter 8. Prentice Hall, Englewood Cliffs, NJ, 1990. Infrastructure Requirements to Support Picture Archiving and Communication (PACS) in medicine.