Fuzzy logic analysis of the degree of assocation between bacterial infection and demographic variables

Study performed in association with

Dr. Les Sztandera - School of Science and Health

Dr. Randi Silibovsky, Department of Infectious Disease, Albert Einstein Medical Center

Background

Diagnosis of bacterial infections currently involves the use of a series of well-developed clinical algorithms in which the most likely etiological agent is determined based on a combination of the patient’s symptoms, previous history and predisposing physiological factors. Laboratory data on these patients requires 24-48 hours of processing, so that all clinical decisions will be based solely on these initial clinical algorithms. Prompt and effective treatment of patients therefore requires that as many variables that act as predisposing factors for bacterial infection be identified and included in these initial diagnostic decisions. Using traditional statistical methods, several studies have indicated that the single demographic variables of age and blood type might act as predisposing agents in bacterial infection^1-3 with advanced age acting as a risk factor for pneumococcal infection¹ and expression of blood types A or AB associated with both tuberculosis² and cholera³ infections. However, as in all biology and medicine no effect is associated with only a single variable and analysis of the complex interplay of factors that lead to bacterial infection requires a more sophisticated mode of computational analysis.

Unlike traditional computing methods, the generation of a fuzzy logic system to investigate the association between demographic factors and bacterial infections allows for the creation of a computer program, which “learns” and maps the complex associations between all these variables. Assessment of the influence of demographic variables on bacterial infection can then be made by allowing the program to predict the nature of a bacterial infection when provided only with demographic data. As in previous models of biological and medical systems, fuzzy logic programming should prove to be of particular use for the efficient analysis of the interaction between the well-defined parameters of age, blood type, gender, race and bacterial infection which are more likely to interact in subtle ways as “shades of gray or maybe”^4,5.    

Study design and results

Since both demographic variables and bacterial species are defined using generally accepted parameters, they constitute highly suitable variables for the generation of a fuzzy logic program.  In this study, fuzzy logic was used to generate a novel program in which the association between the variables of blood type, age gender and race and bacterial infection were analyzed in a real medical database of 187 hospitalized patients. As a diagnostic tool, this program was able to correctly determine bacterial infection as belonging to one of four output groups (staphylococci, streptococci, Escherichia coli or non-E.coli gram negative rods in 27 of 32 randomly selected patients (88.4%), previously unseen by the system, when provided with only the demographic data. This program would appear to have great potential application both as a diagnostic tool and, with further tuning and training, may assist physicians in achieving more rapid and efficient diagnoses.  

We are now increasing the number of patients in the study (currently it stands at 250), courtesy of an Einstein Society Award given by Albert Einstein Medical Center, which we anticipate will allow further tuning and training of the program and elimination of any redundant variables. Indeed, it is anticipated that the combination of currently used clinical algorithms with a user-friendly, simplified future version of the current program might allow its eventual use by all physicians to make more accurate initial predictions of the bacterial causative agent of an infection.

References

1. Nester E.W., Roberts C.E. and Nester M.T. (1995) Interactions between humans and microorganisms. In Microbiology A Human Perspective, pages 353-354, 1^st Edition, Wm. C. Brown Publishers.
2. Overfield T. and Klauber M. (1980) Prevention of tuberculosis in Eskimos. Human Biology 52: 87-91.
3. Minkoff C. and Baker P.J. (1996) Variation among human populations. In Biology Today An Issues Approach, pages 146-158, Edited by Schanck and McConnell, 1^st Edition, McGraw Hill Press.
4. Sztandera L. M. Goodenday L.S. and Cios K.J. (1996) A Neuro-Fuzzy Algorithm for Diagnosis of Coronary Artery Stenosis, Computers in Biology and Medicine Journal 26(2): 97-107.
5. Kosko, B. (1997) Fuzzy Logic and Engineering. In Fuzzy Engineering, Chapter 1, pages 5-8. 1^st Edition, Prentice Hall Publishers.

Publications

1.       Sztandera L.M. and Cundell D.R. (1999) Using fuzzy logic to correlate gender, race and/or blood type with infectious disease. (Invited Paper) Proceedings of the Eighth International Fuzzy Systems, Association World Congress.

2.   Cundell D.R., Silibovsky R.S., Sanders R. and Sztandera L.M. (2000) Using intelligent systems in predictions of the bacterial causative agent of an infection. LectureNotes in Artificial Intelligence 1932:349-357, Springer Verlag

3.   Cundell D.R., Silibovsky R.S., Sanders R. and Sztandera L.M. (2000) Using fuzzy sets to analyze putative correlates between age, blood type, gender and/or race with bacterial infection. Artificial Intelligence in Medicine 586:1-5

4.   Sztandera L.M., Silibovsky R. S., Sanders R., and. Cundell D.R. (2001) Fuzzy system correlates demographics with bacterial infection, Advances in Fuzzy Systems and Evolutionary Computation, 1:60-66