Bachelor_thesis

Enhancing Clinical Decision-Making: Predictive Models for Mortality and Survival in Critical Care

Database link: https://physionet.org/content/mimiciii/1.4/

Abbreviation link: https://mimic.mit.edu/docs/iii/tables

1. Data Understanding and Preprocessing

1.2. Data Exploration

• Objective: To understand the structure and contents of each table.
• Steps:
  • Explore the schema of each table.
  • Identify key variables: e.g., mortality, demographic factors, and clinical data.
  • Check for missing data, outliers, and inconsistencies.
• Tools: R.
• Estimated Time: 2 weeks
• Deadline Date: 23-Sep-2024

1.3. Data Cleaning

• Objective: Prepare the dataset for analysis.
• Steps:
  • Handle missing values: Imputation or deletion.
  • Remove duplicates.
  • Normalize and standardize relevant variables (e.g., age, lab results).
  • Create derived variables if needed (e.g., ICU stay duration).
• Tools: R.
• Estimated Time: 2 weeks
• Deadline Date: 7-Oct-2024

2. Descriptive Statistical Analysis

• Objective: Summarize the dataset to identify general patterns and trends.

• Steps:

  • Demographic Analysis:
    • Calculate statistics for age, gender, ethnicity, etc.
    • Analyze the distribution of admission types and locations.
  • Clinical Data Analysis:
    • Summarize lab results (mean, median, range).
    • Frequency analysis of diagnoses (using ICD codes) and procedures.
    • Analyze ICU stays (LOS, care units, admission types).

• Visualizations:

  • Histograms, bar charts, and pie charts for categorical variables.
  • Box plots for continuous variables.

• Example of table for numerical variables:

  Variable	Count	Mean	SD	Min	Max	Q1	Median	Q3
  age	5231	65.2	12.5	18	90	58	65	75

• Example of table for categorical variables:

  Variable	Category	Count	Percentage
  Gender	Male	2950	56.4%
  Gender	Female	2281	43.6%

• Tools: R (ggplot2).

• Estimated Time: 4 weeks

• Deadline Date: 4-Nov-2024

3. Statistical Modeling and Dependence Analysis

• Objective: To identify factors that are significantly associated with patient outcomes (e.g., mortality, survival days).

• Steps:

  • Correlation Analysis:
    • Use Pearson/Spearman correlation to identify relationships between continuous variables (e.g., lab results and survival days).
    • Use Chi-square tests for associations between categorical variables (e.g., gender, ethnicity) and mortality.
  • Regression Analysis:
    • Logistic Regression: To identify factors that influence the probability of patient mortality (HOSPITAL_EXPIRE_FLAG).
    • Cox Proportional Hazards Model: To analyze survival time and the impact of clinical factors on it.
    • Linear Regression: To examine the relationship between clinical factors and the number of days alive post-discharge.

• Tools: R.

• Estimated Time: 6 weeks

• Deadline Date: 16-Dec-2024

4. Predictive Analytics

• Objective: Develop predictive models to assist doctors in clinical decision-making.

• Steps:

  • Model Selection:
    • Random Forest: For mortality prediction based on clinical variables.
  • Model Training and Validation:
  • Feature Importance: Identify the most influential factors for predictions.

• Tools: Python.

• Estimated Time: 8-10 weeks

• Deadline Date: 7-Apr-2024

5. Writing the Bachelor Thesis Text

• Tools: Latex

• Estimated Time: 26 weeks

  • Initial Drafting: 22 weeks (concurrent with Steps 1 to 4)
  • Revision and Finalization: 4 weeks (after completion of Steps 1 to 4)

• Deadline Date: 5-May-2024