Helicobacter pylori (hereinafter referred to as Hp) infection has become an international problem. The Hp infection rate in China is about 50% to 90%, which is higher than the average level in developed countries (50% to 70%). Helicobacter pylori infection has become a public health crisis in my country. At present, there is no consensus on the route of Hp infection and risk behavior in China. How to effectively find out the influencing factors of the disease is also a topic worthy of discussion in research.
Today's AI artificial intelligence technology and big data analysis can play an important role in research and auxiliary analysis. In this study, the chief pharmacist of Zhongshan Hospital of Traditional Chinese Medicine, Guangdong Province Yuan Yiming, Share how he uses data mining to explore the linear relationship between relative feature data and results, fully applies the visualization and interpretability of machine learning, aims to understand the infection mode and dangerous behavior of Hp in China, and establishes a convenient and economical The predictive infection model provides a reference for asymptomatic infections.
In this study, healthy people who received carbon-13 and carbon-14 breath tests or mucosal tests under gastroscope to check the detection of Helicobacter pylori in the endoscopy and physical examination center were selected as the research objects.
The researchers designed a non-scale electronic questionnaire consisting of two categories and multiple categories related to Helicobacter pylori infection based on experts and literature data. The questionnaire is divided into three dimensions with 63 variables: ①patient basic information form, ②clinical manifestation form and ③life and eating habits form.
The researchers used JMP Pro for statistical analysis of the data. First, single factor analysis was performed on the data, and then multiple logistic regression analysis was performed to establish a prediction model, and finally a forest diagram was drawn based on the results of the model effect items.
Main analysis ideas and steps
Step 1: Data cleaning for existing data
Batch variables can be classified and managed using JMP. Since the collected variables are qualitative variables, JMP Pro performs standardized column feature operations on all variables, and uniformly converts all variables into character-type name variables.
Step Two: Generate Validation Columns
Using the validation function of JMP, the data can be classified into training set and validation set. According to 7.5:2.5, randomly select the training group and the verification group, as shown in the figure below:
Step Three: Univariate Analysis
Through the contingency analysis and the Distribution platform, it can be found that the living floor of the crowd, stomach pain, acid reflux, bloating, loss of appetite, bad breath, bitter mouth, stomach ringing, hunger, anemia, cooking at home for lunch, smoking, and using public transportation when going out Except for chopsticks, the habit of using serving chopsticks, and whether family members and friends are infected, other variables have no statistical significance.
Step Four: Multivariate Analysis
In JMP, click the "Analyze" menu → " fit model (Fit Model) " (Figure 2), in the dialog box that pops up, import pylori infection or not into "Y", and import the independent variables to be multivariate analysis into the "Construction Model Effect" box. Validate based on the generated validation column.
From the effect likelihood ratio test shown in the figure below, it can be seen that the probability of bad breath, bad breath, whether to use serving chopsticks, cooking lunch at home, habit of using serving chopsticks anemia, and the number of floors where the residence lives > chi-square has statistical significance. significance.
From the effect summary results, we can see the contribution of independent variables
From the effect summary results (graph), this part reflects the contribution value of the respective variables, mainly reflected by the LogWorth value, which is equal to -log 10 p.
As can be seen from the figure below, among the 15 independent variables, breath, bad breath and the habit of using public chopsticks contribute the most, while nausea and stomach pain have less influence. This result can directly reflect the contribution of independent variables, and can also be used as a reference for further variable screening. From the overall model test, the value of probability > chi square is less than 0.001, which proves that the built model is statistically significant. Here, the variables with low contribution value are deleted.
Before deletion
after deletion
parameter estimates
Through parameter estimation (figure), the formula of the corresponding prediction mode and the influence of each variable on the final observation can be obtained.
model validation
The area under the ROC curve of the training group and the verification group were 0.7334 (95% CI 0.709-0.784) and 0.7153 (95% CI 0.6729-0.7577), respectively. After analysis, the model is at a good level.
According to the forest diagram, which factors are protective factors and which are risk factors
Save the results of multiple logistic regression and tables of OR values, and create a forest graph in the graph generator based on the series of OR data.
From univariate to multivariate regression analysis, it can be seen that Bloating breath, bad breath, lunch ( cook at home ) , Whether there is a habit of using serving chopsticks at home and when going out, whether family members living together have infection, and how many floors the home lives are the main factors for Helicobacter pylori infection. Through the forest map, we can see which factors are protective factors and which are risk factors.
In summary, by analyzing the infection characteristics and risk behavior factors of Helicobacter pylori, it provides direction for exploring the infection route of Helicobacter pylori, and formulates pre-treatment nursing strategies based on the constructed model.
original text index : Data mining to build a prediction model and prevention strategy for patients with Helicobacter pylori infection [J]. Journal of Gastroenterology and Hepatology ,2022,31(09): 992-998.
If you are interested in the ideas and methods of the case study in this article or want to have further in-depth exchanges and discussions with the author, welcome to sign up for the JMP online seminar on April 27 from 14:00-15:30:
sign up here
At that time, Dr. Yuan will visit the JMP live broadcast room and share with you in detail the valuable analysis ideas and experience of this case study. Wonderful content, don't miss it! Now Try JMP 17 for free for 30 days .
Recommended reading:
