Data analysis: Hospital readmission rates for diabetes

Posted on Apr 2, 2020
The skills I demoed here can be learned through taking Data Science with Machine Learning bootcamp with NYC Data Science Academy.


Our body is an amazing and complex system. Any imbalance in the system can result in significant consequences. According to data, this is the case for the blood sugar level in our body. Insulin is hormone produced by the pancreas and controls how the body is using sugar. This sugar, otherwise glucose, coming from carbohydrate in the food used for energize the body or stored for future needs.

Thus, insulin helps to keep the blood sugar in balance between getting to low or getting too high. Diabetes is a disease when our body is not producing enough insulin to keep this balance. Diabetes may further lead to hearth, lung, digestive and nervous system related issues.

This project aims to understand the relationship between patient data and patient readmission rate to hospitals. The added value has two sides, on one hand the patient care quality is improved, on the other hand the hospital care cost is reduced by preventing early readmission of patients.

The project outline:

  • Objectives, target definition.
  • Exploratory Data Analysis (EDA).
  • Feature Engineering.
  • Predictive model development.
  • Data visualization and model evaluation.
  • Conclusion and business recommendation.




Develop a model to predict the probability of readmission within 30 days for given patient based on patient records. Recommend actions and model deployment based on predictive model results.

Exploratory Data Analysis

The given dataset is 10 years of clinical care data from 1999-2008. It contains 100,000+ observations and 50 Features. Observations are diabetic patient encounters in hospitals. The dataset has missing values, numerical and categorial features. Many numerical features are encoded values. The target variable is categorical. The dataset is imbalanced with a minority class  of 10,000 readmitted patients. Also, graphs were created to visualize the distribution and frequency of specific features during univariate analysis.

Figure 1. Target Variable Visualization

Feature Engineering

Key features are identified after research in patient diagnosis, medications, medical tests performed, doctor's medical specialty, admission and discharge related topics.

The granularity of categories within features required aggregation. To give a couple of examples, the 716 different medical diagnosis (numerical IDC) codes were reduced to circulatory, respiratory, digestive, diabetes, musculoskeletal, genitourinary and neoplasm groups. The 72 medical specialty were reduced to general practice, internal medicine, surgery, other and missing groups. Admission and discharge features were reduced to three and two groups respectively.

Figure 2. Features grouped

The final feature set resulted in 28 columns after one hot encoding the categorical variables. Also, to ensure independent observations, patients having multiple encounters were reduced to one by keeping only the first record. This way, the cleaned dataset that was used in the model had 67,000 rows.

Predictive model development

While there are powerful classification models, for this project the more simple, easily understandable  logistic regression was chosen.

The data was split into training and test set. Cross validation was implemented to avoid overfitting.

Initial runs resulted in high accuracy giving false sense of security in the model. Additionally, the minority class upsampling was done before train test split.

These mistakes provided a great learning experience in how each development step influenced model predictive performance. These issues were fixed, precision and recall scores were chosen as metrics on the test set. Several upsampling methods were applied to only to training data set.

Model Evaluation

Confusion Matrix for the binary classification gives a full brief summary of the predictions on how many patients predicted to be readmitted or not and how many were actually readmitted or not.

Figure 3. Confusion matrix and values

The Receiver Operating Characteristic curve is used for this binary classification model to visualize the ability to separate the readmitted patients from the rest by varying the probability threshold.

Figure 4. ROC curve

Decile analysis was used to show the ability of the model to consistently group patients based on the probability of readmission from low to high. It also depicts how much more likely a patient to be readmitted in a group compared to the average probability.

Figure 5. Decile Chart Groups


The data engineering team is able to deploy the model by developing a Shiny R dashboard with an interface to model developed in python language.

Figure 6. Shiny R Dashboard Blueprint

The doctor or nurse in the medical team could enter patient data  and predefined actions can be presented based the predicted probability of readmission. Here, action may be implemented based on the significance of the features given calculated by the model, where a large coefficient of a feature results increased probability of readmission.

Figure 7. Feature Coefficients

The data science team shall continuously observe the decile chart data to evaluate the performance of the predictive model.


Future work

Given the lower than expected predictive power of the model with ROC score of 0.6, further feature engineering and selection would be accomplished, including the use of interaction terms.

The Shiny R  dashboard can be developed as part of the project given additional time and resources.

Additional patient data collection and aggregation could improve the predictive models performance.

Further reading and deeper insight into the medical care industry and diabetes shall suggest additional steps at feature engineering.




About Author

Karoly Lajko

He received his diploma from Stony Brook University in Electrical Engineering. He developed digital circuits for application and microprocessors. He is a strategic leader with years of experience in solving highly complex problems through development and production-related projects.
View all posts by Karoly Lajko >

Leave a Comment

No comments found.

View Posts by Categories

Our Recent Popular Posts

View Posts by Tags

#python #trainwithnycdsa 2019 2020 Revenue 3-points agriculture air quality airbnb airline alcohol Alex Baransky algorithm alumni Alumni Interview Alumni Reviews Alumni Spotlight alumni story Alumnus ames dataset ames housing dataset apartment rent API Application artist aws bank loans beautiful soup Best Bootcamp Best Data Science 2019 Best Data Science Bootcamp Best Data Science Bootcamp 2020 Best Ranked Big Data Book Launch Book-Signing bootcamp Bootcamp Alumni Bootcamp Prep boston safety Bundles cake recipe California Cancer Research capstone car price Career Career Day citibike classic cars classpass clustering Coding Course Demo Course Report covid 19 credit credit card crime frequency crops D3.js data data analysis Data Analyst data analytics data for tripadvisor reviews data science Data Science Academy Data Science Bootcamp Data science jobs Data Science Reviews Data Scientist Data Scientist Jobs data visualization database Deep Learning Demo Day Discount disney dplyr drug data e-commerce economy employee employee burnout employer networking environment feature engineering Finance Financial Data Science fitness studio Flask flight delay gbm Get Hired ggplot2 googleVis H20 Hadoop hallmark holiday movie happiness healthcare frauds higgs boson Hiring hiring partner events Hiring Partners hotels housing housing data housing predictions housing price hy-vee Income Industry Experts Injuries Instructor Blog Instructor Interview insurance italki Job Job Placement Jobs Jon Krohn JP Morgan Chase Kaggle Kickstarter las vegas airport lasso regression Lead Data Scienctist Lead Data Scientist leaflet league linear regression Logistic Regression machine learning Maps market matplotlib Medical Research Meet the team meetup methal health miami beach movie music Napoli NBA netflix Networking neural network Neural networks New Courses NHL nlp NYC NYC Data Science nyc data science academy NYC Open Data nyc property NYCDSA NYCDSA Alumni Online Online Bootcamp Online Training Open Data painter pandas Part-time performance phoenix pollutants Portfolio Development precision measurement prediction Prework Programming public safety PwC python Python Data Analysis python machine learning python scrapy python web scraping python webscraping Python Workshop R R Data Analysis R language R Programming R Shiny r studio R Visualization R Workshop R-bloggers random forest Ranking recommendation recommendation system regression Remote remote data science bootcamp Scrapy scrapy visualization seaborn seafood type Selenium sentiment analysis sentiment classification Shiny Shiny Dashboard Spark Special Special Summer Sports statistics streaming Student Interview Student Showcase SVM Switchup Tableau teachers team team performance TensorFlow Testimonial tf-idf Top Data Science Bootcamp Top manufacturing companies Transfers tweets twitter videos visualization wallstreet wallstreetbets web scraping Weekend Course What to expect whiskey whiskeyadvocate wildfire word cloud word2vec XGBoost yelp youtube trending ZORI