Healthcare-Predictions

Overview

The dataset contains a comprehensive collection of electronic health records belonging to patients who have been diagnosed with a specific disease. cords comprise a detailed log of every aspect of the patient's medical history, including all diagnoses, symptoms, prescribed drug treatments, and medical tests that they have undergone. These health records comprise a detailed log of every aspect of the patient's medical history, including all diagnoses, symptoms, prescribed drug treatments, and medical tests that they have undergone. Each row represents a healthcare record/medical event for a patient and it includes a timestamp for each entry/event, thereby allowing for a chronological view of the patient's medical history.

The Data has mainly three columns

1. Patient-Uid - Unique Alphanumeric Identifier for a patient
2. Date - Date when the patient encountered the event.
3. Incident - This column describes which event occurred on the day.

There are total 27K unique patients present in train.parquet Similar to drug_type_7, there is also a drug called ‘Target Drug”, this drug is of interest for the assignment, there are total 9K patients in train.parquet who have taken “Target Drug” at least once.

Prediction - 1

Aim

The goal is to develop a predictive model that can determine whether a patient will be eligible for the "Target Drug" within the next 30 days based on their medical history.

Solution Approach:

1. Data Preprocessing:

• Load the training dataset (train.parquet) and perform necessary preprocessing steps such as handling missing values, converting data types, and handling categorical variables.
• Create a positive set consisting of patients who have taken the "Target Drug" at least once, considering the time aspect.
• Create a negative set of patients who have not taken the "Target Drug" within the next 30 days after their first prescription.
• Split the data into training and validation sets.

2. Model Development:

• Choose an appropriate machine learning algorithm (e.g., logistic regression, random forest, gradient boosting) for classification.
• Train the model using the training dataset and evaluate its performance on the validation set.
• Fine-tune the model parameters, if required, to improve its performance.

3. Evaluation:

• Evaluate the model's performance using appropriate evaluation metrics such as F1-score.

4. Predictions and Submission:

• Apply the trained model to the test dataset (test.parquet) to generate predictions for each patient.
• Label each patient in the test dataset as 1 or 0 using the predictive model.
• Prepare the final submission file (final_submission.csv) containing the patient UIDs and corresponding predictions.

Prediction - 2

Aim

To analyze the drop-off rate for "Target Drug" and identify the events driving patients to stop taking the drug.

Solution Approach:

1. Calculate the drop-off rate:

• Extracting the relevant columns from the dataset, including "Patient-Uid" and "Date" for drop-off events
• Filtering the data to include only patients who have taken the "Target Drug" at least once.
• Group the data by month and count the number of unique patients who dropped off within each month.
• Calculating the drop-off rate by dividing the number of drop- offs by the total number of patients who started the treatment in each month.

2. Analyze events driving drop-offs:

• Identify the adverse events or other relevant columns in the dataset that might indicate the reasons for drop-offs.
• Extracting the necessary columns for analysis, including "Patient-Uid", "Date", and the relevant event columns.
• Filtering the data to include only patients who dropped off the treatment.
• Analyzing the occurrence and patterns of events leading to drop-offs, such as adverse events or specific conditions.

Prediction - 3

A drug is generally administered to a patient in certain patterns or at regular intervals of time. For example, Chemotherapy which is a drug treatment in the case of Cancer is generally given to patients at an interval of 3-4 weeks, i.e. every 3-4 weeks patients are administered the drug. Similarly to Chemotherapy, a “Target Drug” is also administered/prescribed in certain patterns, we want to analyze in what patterns the “Target Drug“ is administered/prescribed to patients, there might be multiple patterns in which “Target Drug” is administered/prescribed, come up with an analysis which to extract the dominant patterns in the data using clustering or other unsupervised techniques. Visualize the prescription patterns with time on X-axis (month) and prescriptions on Y-axis for each of the patterns you are able to extract(Below is an example of a prescription pattern, where a prescription is made at least once in the first two months followed by one prescription for every two months).

Aim

To analyze the prescription patterns of the "Target Drug" and extract dominant patterns using clustering or other unsupervised techniques.

Solution Approach:

1. Data Preparation:

• The dataset is loaded from training dataset (train.parquet) and the 'Date' column is converted to a datetime format.
• The 'Month' column is extracted from the 'Date' column to represent the month component of the dates.

2. Pattern Extraction:

• The dataset is grouped by both 'Patient-Uid' and 'Month' columns.
• The number of prescriptions for each group is counted, resulting in a DataFrame with columns 'Patient-Uid', 'Month', and 'Prescriptions'.
• The grouped DataFrame is further grouped by 'Month' to calculate the average number of prescriptions per month.

3. Visualization:

• A figure is created, and a violin plot is generated to visualize the distribution of prescriptions for each month.
• A strip plot is added on top of the violin plot to display individual prescription data points.
• A line plot is created to show the average number of prescriptions per month.
• Axes labels, a title, and a legend are added to the plot.