Why Hospital Readmissions are Important
Hospital readmissions cost Medicare an estimated $26 billion per year as approximately one in six Medicare patients leaving hospitals are readmitted within 30 days. This is no small problem!
Since a single percentage point decrease in readmission rates improves patient care quality and saves millions of dollars, the Centers for Medicare & Medicaid Services (CMS) incentivizes hospitals to reduce avoidable readmissions by withholding proportions of payments if their 30-day readmission rate for certain medical conditions falls below the benchmark of 80th percentile or lower.
Healthcare analytics teams and hospital administrators are thus very motivated to track and lower this metric, but face a serious challenge identifying the root causes of hospital readmissions to take action. Readmission could be the result of one or more of many factors such as medication errors, medication noncompliance, a lack of timely follow-up or transportation to access care, or many other factors.
Readmission root cause analysis is challenging because data must be connected from a variety of data sources; then analyzed by applying various dimensions/filters; and hypotheses must be formulated and tested regarding which of the myriad variables is impacting readmission, which is time consuming and far from a perfect science.
Healthcare analytics teams need a holistic view of the patient attributes, care history, interactions/outreach, and more, to answer questions like:
- How are hospital readmissions trending this month?
- What is driving these readmissions?
- Which patients are most likely to be readmitted, and is there anything we can do to prevent this?
Tellius is an AI-powered analytics platform designed to help healthcare, life sciences, and other industries accelerate data-driven insights and decision-making. In this article, we’ll explore how:
- Tellius’s natural language search capability helps analytics teams perform interactive data exploration and ad-hoc querying of granular data without complex scripts to find answers to WHAT the data is saying about hospital readmission rates;
- Tellius’s AI-powered automated insights can surface root causes to identify WHY readmissions are occurring, in an easy to comprehend way for non-experts via natural language generation
- Tellius’s predictive capabilities can be used to quickly and easily quantify the probably of an individual patient being readmitted within 30 days of discharge, based on patient medical history, diagnosis, length of stay, medication administered, and other factors
- Tellius’s last-mile data prep and KPI tracking capabilities to help users enrich, modify, and transform datasets via point and click functionality as well using SQL and Python to ensure a complete picture of the data and automate the process of monitoring patient readmission to analyze and produce actionable insights with a few mouse clicks.
Each functionality covered in this article is an integral piece of Tellius’s analytical solution, which put together can help hospital administrators and healthcare analytics teams reduce hospital readmission rates, improve patient outcomes, while reducing costs. Ready to dive in? Let’s go!
P.S. If you’re interested in exploring other healthcare use cases where Tellius can add value (e.g. hospital operations optimization, quality of care and patient outcome improvement, etc.), learn more here.
The Data
The data for this use case comes from a fictitious hospital system’s Hospital Management Software (HMS). It contains the following schema:
- Patient History: Patient ID, Visit history, Current medications, …
- Patient Outreach: Contact method, Attempts, …
- Patient Attributes: Gender, Age, Payer, …
Load the Data
Our data for this example is in the form of three .CSV files which we simply upload in the Data screen. Notice Tellius can connect to multiple internal and external data sources, as well as cloud data warehouses/lakes like Snowflake and Databricks.
Once we load the data, we simply drag and drop the various sources to form a data model:
Tellius helps users evaluate the quality of the data by inferring each column’s data types and automatically creating descriptive statistics for each column to help users streamline initial exploratory data analysis (EDA).
For example, when examining the “Readmitted” column, We see that out of ~101k patients, ~11k (or about 11%) are readmitted less than 30 days after release. Let’s dig a little deeper.
Data Preparation
Tellius has a robust data prep layer that provides full flexibility for the advanced users to incorporate the required logic in order to manipulate the data using Python or SQL. However, a wide variety of point and click functions is also available.
For example, we need to create a binary indicator if a patient was readmitted within 30 days. In order to do that we can utilize Tellius’ formula wizard to create the calculated column.
Additionally, more advanced users can utilize Python or SQL editors to enrich the table with more complex calculations.
Below is an example of applying a transformation to the AGE variable in order to clean the values from extra characters using SQL REPLACE function.
Exploratory Analysis & Insights
Now let’s look at how an analytics team can utilize Tellius’s Search capabilities to perform on-the-fly data exploration and assess data quality, generate initial visual summary statistics, and spot some initial observations in natural language, without lengthy Python scripts.
For example, to identify the trend in readmitted patients over time, just type “Show me readmits monthly”. Tellius returns the result in the form of the following chart.
To check how readmissions correlate with the number of medications administered for patients, simply query “show me readmits compared to medications monthly”. Tellius returns the following result:
With Tellius’s Search engine the possibilities are endless when it comes to getting quick ad-hoc answers from your data whether it is for the purpose of performing exploratory data analysis, reporting, or creating your own data story.
Proactive KPI Tracking
Tellius allows users to set up alerts for any KPI in their data and be notified if there is a statistically significant change in that KPI. Healthcare analytics teams and hospital administrators can use this functionality to track the hospital readmission rate on a daily or weekly basis proactively, rather than reactively. Tellius will send users an email if there is a significant change in the KPI of interest while also providing actionable insights in the form of a report with descriptions, key contributor list, and the reasons behind the change in the KPI.
For example, the summary output below was generated because of a change in readmission rates for the time period of interest. Tellius summarized the readmission rate increase was due to failure to administer chlorpropamide, miglitol, and acarbose to patients.
The analysis also includes the list of additional key contributors associated with the change in hospital readmissions.
How to Predict Readmissions
Now that we know WHAT is happening (monthly hospital readmission rate changes) and WHY it is happening (failure to administer specific medications to patients), let’s go a step further and try to PREDICT an individual patient’s likelihood of readmission within 30 days of discharge, based on patient medical history, diagnosis, length of stay, medication administered, and other factors.
To do this, we build a machine learning model in Tellius in a point-and-click manner that assigns a readmission probability score per patient so hospital administrators can prioritize at-risk patients and take necessary action in a more proactive manner. Tellius’s machine learning modeling capabilities are built on Apache Spark, a lightening-fast distributed processing system used for big data workloads — specifically using the Spark ML open-source library. The platform offers three approaches for training a model. One is called AutoML, where the user selects a target variable and relies on Tellius to select the appropriate algorithm, perform feature transformation, and fine-tune parameters. The other is called Point & Click, which offers users more control over model selection and hyperparameter tuning approach. The final is a Bring Your Own Model, where users can utilize a Python or SQL-based ML code. In this case, we will utilize the Point & Click approach to build our model.
Step 1. Select the classification model type.
Step 2. Select target, input features, parameters
Step 3. Select algorithm and additional model parameters
Review the Model
After the model is finished training, Tellius surfaces all the model leaderboard information, the final list of input features, algorithm documentation, and model parameters.
Model Results
Using the model and applying it on the scoring dataset, the analytics team can easily build content in the visualization layer in Tellius and share consumable model results to users of all levels within the organization such as hospital administrators and doctors to help easily identify current patients who are at elevated risk of readmission based on their medical history and potential intervention approaches to reduce that risk.
An example of such an analytic output is as follows:
Summary
As we showcased above, Tellius delivers value to healthcare analytics teams and hospital administrators by enabling faster answers to hospital readmission drivers and predictive capabilities to spot and take action on at-risk patients in an intuitive interface. In addition to healthcare, Tellius is useful for a variety of other healthcare applications including hospital operations optimization, care quality gap analysis, and patient outcome optimization.
Try Tellius yourself today for free (no credit card necessary)!
