P024  Medical decision support system in the diagnosis of ulcerative colitis

P024  Medical decision support system in the diagnosis of ulcerative colitis
 

Bikbavova Galiya1, Romanyuk Alisa1, Livzan Maria1
1 Federal State Funded Educational Institution for Higher Education Omsk State Medical University, Ministry of Public Health of the Russian Federation, Omsk, Russia

BACKGROUND: An increase in the number of patients with ulcerative colitis (UC) occurs with a change in the modern lifestyle and is referred to as the “westernization” of lifestyle and nutrition. The aim of the study was to develop a model for supporting medical decisions in the diagnosis of UC, based on a set of anamnestic signs.
 

METHODS: A survey was conducted in the form of an interview after receiving informed consent of respondents. The questions of the survey were combined according to the following characteristics: The type of work and occupational hazards; Medications intake (antibiotics, non-steroidal anti-inflammatory drugs, hormonal contraceptives); Markers of the “hygiene hypothesis” (presence of pets in childhood, attendance at kindergarten, breastfeeding in childhood, the number of children in the family); Psychological stress; Nutritional factors. The main group consisted of 81 patients (42 men and 39 women) aged from 18 to 79 years. A comparison was carried out between the control group of healthy respondents comparable in age (U=13.38; p=0.1760) and gender (2I=2.72; p>0.05) and the group of patients with UC. The control group consisted of 39 healthy individuals (14 men and 25 women) aged from 22 to 81 years. The median age of the respondents was 46.0 (26.0-52.0) years. We used the median and interquartile range, the Mann-Whitney U test, and the calculation of 2I statistics to describe and compare the groups. We built binary choice models in the “EViews 11” program using the Logit-model. Variables with ​​p values >0.05 were excluded from the model in the process. After we obtained binary choice models for each of the groups and determined the clipping region, we calculated type 1 and 2 errors. The clipping region for each model was selected individually. We built a general model in the form of a neural network.
 

RESULTS: We identified no variables among the markers of the “hygienic hypothesis” and medication intake that could influence the occurrence of UC. After removing insignificant variables from the model of “psychological stress”, the statements “Perhaps I am a nervous person” (coefficient -0.296021; variable 0.0041) and “Connection between illness and the suffered stress” (coefficient 0.683475; variable 0.0000) became significant. In the model “Nutrition factors”, the significant variables were “Regular consumption of spicy food” (coefficient 0.183085; variable 0.0069), “Excessive consumption of sugar with tea and coffee” (coefficient 0.048343; variable 0,0022), “Insufficient consumption of vegetables daily” (coefficient -0.006935; variable 0.0001) and “Poor milk tolerance” (coefficient 0.825844; variable 0.0013). In the model “Type of work and occupational hazards”, the significant variables were “Occupational employment” (coefficient 1.722686; variable 0.0002), and “Occupational hazards at the workplace associated with attention and memory stress” (coefficient 2.608838; variable 0.0000). The last level also went through the procedure for building a binary choice model. The models “Type of labor and occupational hazards” (coefficient 1.492367; variable 0.0010) and “Psychological stress” (coefficient -1.823598; variable 0.0000) turned out to be significant.
 

CONCLUSION(S): We developed a model to support medical decisions based on a set of anamnestic data, which improves the efficiency of the UC diagnosis.