Predicting Length of Hospital Stay With AI

Use Cases

Using no-code AI, we predicted the length of hospital stay for patients with great accuracy.

Hospitals are under pressure to cut costs and improve patient outcomes. One of the most effective ways to do this is by reducing length of stay (LOS). 

Across OECD countries, the average LOS is just under 8 days, but it varies widely by hospital. Here are the statistics for the United States:

  • The national average for a hospital stay is 4.5 days.
  • The average cost of one day of a hospital stay is $10,400.

With Obviously AI, we can build a model to predict patient length of stay, achieving around 80% accuracy. The model can then be applied to new patients, with highly accurate results. This approach is very scalable and can be used in virtually any hospital. Crucially, Obviously AI is compliant with HIPAA regulations.

In this guide, we’ll explore how to deploy our LOS-prediction model via Zapier, a no-code automation tool. Check out our other healthcare use-cases as well.

1. Building a length of hospital stay prediction model

To get started, make an Obviously AI Pro account, which comes with a 14-day free trial.

Once you log-in, you’ll see a dashboard with an “Add Dataset” icon on the left. For this example, upload a simple tabular dataset from GitHub, which features hospital length of stay data alongside a number of patient variables.

Our goal is to predict a KPI called lengthofstay, which is the length of a hospital stay, in days.

After clicking to upload a CSV file, you’ll need to verify the dataset, and simply check off that the first row is the column names and that all requirements are met.

Now, we select the column we want to predict, which is called “lengthofstay.” In the background, a series of machine learning models compete to create the most accurate predictions for lengthofstay.

And we’re done! Now, we can also see how various attributes impact the length of hospital stay KPI. The main drivers discovered by the automated machine learning models were rcount, or the number of readmissions within the last 180 days, and glucose, or the average sodium value during the encounter.

Let’s move on to building a Zapier sequence, so you can deploy predictions in the real world, and predict length of hospital stay on new data.

2. Building a Zapier Flow

Now, make a free Zapier account. We’ll start by connecting Zapier to a dataset in Google Sheets. We’ll select “New Spreadsheet Row in Google Sheets” as the trigger to activate a Zapier sequence.

After selecting that trigger, I simply located my demo Google Sheet, which I titled “New Data.”

This sheet should have the exact same column names and data types as the file I used to make the predictive model (minus the actual target column of “lengthofstay”).

For instance, since my training data had columns like asthma and pneum, and so on, I use those exact columns in this Google Sheet.

Next, we’ll select “Webhook,” and then “Custom Request.” This is how we’ll integrate with Obviously AI.

To set up the Custom Request, add in these details:

The data field is the most important, which requires matching up the column names in your data with the column names used in your model. Check the documentation if you’re unsure!

To set it all up, you’ll also need your Report ID and API key. Your API key can be found in your account, which gets added as a “Header” in Zapier. To get the Report ID, head to the “Export Predictions” tab in Obviously AI, and click on the shareable report link.

The Report ID is the ID you see in that URL. In the example above, it’s the string “cff9...1fd7.” As you can see in the documentation, this gets added in the bottom of the “Data” field in Zapier.

Now our Zap sends data from our new data sheet to our length of hospital stay prediction model, and we’ll get the length of hospital stay probability as an output. With these predictions in Zapier, we can now do anything we want with them: Send them to a Slack channel, send an email to our team, add the prediction to our sheet, or anything else.

For now, let’s just update our sheet with the estimated length of hospital stay. This can be done with the action titled “Update Spreadsheet Row in Google Sheets.”

Then, simply add the probability created in the aforementioned webhook.

Let’s take it a step further. We can create a “filter” in Zapier to send an email notification when the predicted length of hospital stay is above a threshold that we set. I set up a filter that gets activated when the probable length of a hospital stay is higher than 9 days.

If the length of hospital stay probability is higher than 9 days, we can automatically send a notification email so that we’re aware of the situation.

Summary

Zapier is a powerful way to automate practically anything in your workflow. With Obviously AI, you can integrate the power of machine learning to create AI-driven workflows, including to predict length of hospital stay and help improve the quality of inpatient care and ultimately improve resource allocation.

(Hospital Photo by National Cancer Institute on Unsplash)

Become the Data Scientist your team always needed.

Get Started

Get Started Now

See how no code machine learning can transform your business and change how you make decisions.