We’ve all heard it. Artificial intelligence (AI) and machine learning (ML) are revolutionizing the provision of healthcare. From streamlining revenue cycle management to producing better clinical outcomes, it’s one of the hottest topics within the entire healthcare sphere. So, where to start when thinking about implementing your own AI/ML solution? Many data science projects in healthcare never actually get to see “production,” and projects that don’t generate a great return on investment (ROI) are costly and ultimately damage the momentum of advancing modern analytics within an organization. Read on to learn how 3Cloud’s methodology of “think big, start small” and implementing projects that result in quick wins can help to build a continuously improvable modern analytics solution.


Identify Your Priorities

Like more traditional projects, finding quick wins is imperative to creating the environment, momentum, and cadence necessary to foster a health advanced analytics movement within your organization. Here are a few instances where the barriers to entry are lower, allowing for quicker wins:

Genomics and Bioinformatics

Capturing genomic variant information about patients provides straight-line pathways to research projects and empowers precision medicine. Combined with the cloud-scale analytics offered by Azure Data Services, health care organizations no longer need to figure out how to fit such on operation into their on-premises enterprise architecture.


Worried that this might be a difficult place to start? 3Cloud has a library of resources to help you tackle your bioinformatic needs which walks through how to build a scalable, secure, and collaborative data lake within Azure.

Finance and Revenue Cycle Management Predictive Analytics

Unless you’re intimately acquainted with healthcare operations, you probably didn’t consider this as an area where machine learning and AI could generate a quicker win. The finance and revenue cycle management functions of any healthcare organization are prime candidates for momentum-building predictive analytics.


“The data is available, and de-identification can make operating within the parameters of HIPAA much easier than some of the more PHI-intensive clinical data sets,” explains Nick Diehl, Senior Healthcare Consultant. “Better forecasts can lead to better predictions of net revenue and expenses, both of which are invaluable in influencing operational decisions.”

HFMA has recommended predicting denials by CPT and Payor Code amidst other AI/ML strategies. Other common predictive targets are case volumes, charity and write-off likelihood, denials, and cash forecasting.

Creating Smarter Supply Chains

The inefficiencies in healthcare supply chain management became painfully apparent at the beginning of the COVID-19 pandemic. This is part of the reason why medical error is now the third leading cause of death in North America. Models like CISOM from HIMSS provide an idea of what supply chain maturity looks like by identifying four key focus areas: automation, clinical integration, predictive data analytics, and governance/leadership.


With value-based care changes looming, unpredictable revenue changes associated with regulatory/legislative disruption, and COVID-19 protocols, Forecasting Demand with Azure Machine Learning or Databricks’ Demand Forecasting Solutions Accelerator empower any healthcare organization to take advantage of popular and proven AI and ML models. These solutions are low-code and generate great opportunities to demonstrate the power that AI and ML offer.

An additional benefit of starting here is that creating demand models that are integrating clinical data with procurement and financial data is the logical next step in scaling such a project. This integration will increase your CISOM score and provide a great value-additive use case for data centralization projects, like creating a modern data platform. Need help? 3Cloud is happy to talk through how one of our QuickStart engagements, built on a Catalyst framework, can save you weeks of effort without breaking the bank.

Avoid Quagmires

Not all opportunities are created equal. Some potential starting points are quagmires that could lead to projects that take too long to return on initial investment and lose executive will as a result. For example:


Diagnosis Recommendation

When people think of the application of machine learning in healthcare, this is often where they start. Diagnosis prediction is an area worthy of additional research, and there are good reasons why so many analytics resources are being directed toward the topic. Why then is this a risky pilot project for any AI/ML function to prove value?

Bias in healthcare machine learning models often comes from the fact that the data used to train the model itself is unbalanced. The social determinants (ethnicity, sex, age, etc.) within training data are often narrow, and can cause a predictive classification model to make assumptions when determining diagnosis that do not reflect the diverse group of people that could potentially walk through the doors. To account for this, the best solution is to utilize a diverse training data set. In practice, according to the Journal of the American Medical Association, many training data sets for medical AI algorithms are comprised of data coming from just 3 states: California, Massachusetts, and New York.

Methods exist to deal with bias in training data. Offerings like Fairlearn provide tools to understand what bias may exist in each model and present steps for mitigation. This and re-coding social determinants to more generalizable values can help, but it’s not a complete fix.

Additionally, for the ROI to manifest, the recommendation engine must streamline and/or improve the provision of care. When the diagnosis is wrong, who is to blame? The AI? An overseeing physician? In most cases, recommendation engines are better suited to help guide pathology and make the efforts around diagnosis faster, rather than replace traditional diagnostic procedure. Since this dramatically lowers the overall ROI potential, this is probably not the best momentum-building AI/ML project for an ambitious healthcare organization.

Getting Started

IF you’re excited about getting started on some healthcare AI and ML projects, reach out to 3Cloud today.