AZ Delta: Bringing personalized medicine one step closer with data analytics
About AZ Delta
One of Belgium’s largest hospitals with 1,400 beds and around 650,000 annual patient consultations, AZ Delta aims to be a leader in high-quality care, prioritizing innovation and continuous dialogue with patients.
Tell us your challenge. We're here to help.
Contact usAbout ML6
With more than 100 AI specialists and ML engineers in Belgium, Germany and the Netherlands, ML6 helps organizations create true business value with AI.
AZ Delta has built a comprehensive medical data analytics platform that generates unprecedented insight without compromising security supported by Google Cloud.
Google Cloud results
- Keeps sensitive medical data secure with Virtual Private Cloud and three-factor authentication with Cloud Identity.
- Collates and analyzes hundreds of millions of data points at speed for greater insight with BigQuery.
- Engages machine learning tools to help physicians plan the optimal treatment pathways for their patient’s unique needs.
Cuts data query run time from 15 minutes to 15 seconds
Doctors have long tried to find treatments that work for individual patients based on their unique circumstances. Now, thanks to the power of modern technology, what has been a centuries-held ambition may well become a 21st-century reality. AZ Delta, one of Belgium’s largest hospitals, is at the forefront of this journey. As a strong research driven hospital, it’s focused on pushing medicine forward. "Our CEO is intent on driving innovation. In 2020, he set up the hospital’s department for innovation, and that’s when I came on board," says Peter De Jaeger, Chief Innovation Officer at AZ Delta. "I had to work out how we could best leverage our technology to improve our healthcare. So, I thought we have all this data, we must mine it for insights."
At the time, the hospital’s vast amount of medical data was available digitally, but not in a single location and was difficult to work with at scale. De Jaeger prioritized collating and analyzing that data to generate medical insights and help physicians do their jobs. He soon realized, however, that while the on-premises IT infrastructure was well-suited for running the hospital’s operations, it could not power large scale data analytics. The hospital’s cutting edge analytics platform needed a cutting edge infrastructure. For that, De Jaeger turned to Google Cloud.
"We had all these datasets stored on-premises," he says, "but when it came to processing and analysis, we didn’t have the power we needed. Google Cloud and its ability to handle large scale data at speed allowed us to forge ahead with our plans."
"I knew that Google Cloud was powerful enough to build the data platform we envisioned. And it also combined ease of use with comprehensive security options for our most sensitive data. It was the only infrastructure that we looked at that fulfilled all our criteria."
—Peter De Jaeger, Chief Innovation Officer, AZ DeltaThe challenges of medical data
From the outset, AZ Delta’s innovation team emphasized practical solutions to the problems faced by hospital staff. "Our solutions must be grounded in the day-to-day reality of working in a busy hospital," explains De Jaeger. "We paired each of our engineers with a physician, so we could understand the challenges they face."
AZ Delta had been using Electronic Medical Records (EMR) for years and every new patient interaction generates a new record. There was plenty of data to work with, but the challenge came in trying to analyze that data. Firstly, there was the sheer scale of data available. A single patient’s EMR can contain thousands of data points. With more than 650,000 patient consultations a year, even storing this amount of data is a challenge for on-premises servers, let alone processing and analyzing it.
The second major issue was the complexity of the data. With the existing EMR system, staff could enter information in a pre-defined structured format or in free text. While the structured method ensured variation was kept to a minimum, data scientists found that single datapoints gave them too little useful information. "A temperature reading, for example, means something very different if the patient has just come out of chemotherapy versus if they’ve just come into the ward," explains De Jaeger.
To compensate for this, staff would often write long notes in the EMR system. This provided the context for clinical use but meant that statistical analysis was harder further down the line. If AZ Delta wanted to use the latest analytics tools to improve its healthcare, it needed to be able to make sense of its data.
The data needed to be processed and normalized before it could be mined, for instance. De Jaeger and his team of engineers initially tried to run machine learning algorithms on the data to automate this as much as possible, but they found that there was too much variance. Plus, any manual queries took up to 15 minutes to run because of the scale of the project. Soon, De Jaeger reached the limits of what was achievable with the on-premises infrastructure.
Keeping medical data secure
Prior to joining AZ Delta, De Jaeger had used cloud computing to solve large scale data problems, and he began exploring the options. He decided on Google Cloud for a number of reasons. "I knew that Google Cloud was powerful enough to build the data platform we envisioned," he explains. "And it also combined ease of use with comprehensive security options for our most sensitive data. It was the only infrastructure that we looked at that fulfilled all our criteria."
While Google Cloud had the technical capabilities necessary for AZ Delta’s plans, the hospital also needed it to conform to the highest security standards. "Patient information is about the most sensitive form of personal information there is," he explains. "Security is critical for us, so we needed to have the best possible solution."
AZ Delta teamed up with Google Cloud partner ML6 to build the platform and keep it secure. With ML6’s guidance, the hospital has built its own environment using Virtual Private and manages permissions and access with three-factor authentication using the Cloud Identity package of solutions. "ML6 helped our engineers to understand how security works in Google Cloud,” explains De Jaeger. “Now, we don’t have to compromise on our collaborations for the sake of security."
"BigQuery has helped us work orders of magnitude faster than we did with an on-premises database. A query that used to take our engineers 15 to 20 minutes now takes about 15 to 20 seconds for a doctor using our frontend."
—Peter Dejaeger, Chief Innovation Officer, AZ DeltaUsing BigQuery for large scale analytics at speed
With the environment secured, AZ Delta concentrated on building its platform around BigQuery, which replaces its previous on-premises database manager. "BigQuery is the heart of our solution," says De Jaeger. "It lets us work at the scale we want and at the speed we want. Without it, I’m not sure we would have been able to develop everything that we have done so far. BigQuery has helped us work orders of magnitude faster than we did with an on-premises database. A query that used to take our engineers 15 to 20 minutes now takes about 15 to 20 seconds for a doctor using our frontend."
The platform begins with an on-premises gateway that allows staff to access the EMR system. By keeping the gateway on-premises, AZ Delta ensures it has total control of the data before it gets uploaded onto the cloud and it can anonymize elements where required. The data is then transformed into a parquet file, before being uploaded to Cloud Storage buckets.
From there, AZ Delta uses Apache Airflow, the open-source data streaming tool to perform Extract, Transform and Load (ETL) operations on the parquet files, turning them into rows of data and loading them line by line into a BigQuery table. This first table is combined with a second reference table that matches each data point with a code. This helps ensure that data entry is consistent, relevant and easily retrievable. The codified data is then exported to a third and final table where it is queried for analysis.
Analytics and machine learning in a clinical context
So far, AZ Delta has cleaned and coded the data for 50,000 patients. De Jaeger estimates that amounts to around half a million data points handled with ease and speed. The engineers have also used BigQuery to automate much of the query building and optimization so that AZ Delta’s staff can interrogate the data they need and get answers back within seconds using the EMR frontend.
"The power we have access to now with Google Cloud means that we can do things that were beyond the scope of human ability. So, if we can combine the strengths of our physicians with the latest technology, I think we’ll be able to take a big step forward in healthcare."
—Peter Dejaeger, Chief Innovation Officer, AZ DeltaWith the data cleaned and normalized, AZ Delta has begun training a variety of machine learning algorithms with TensorFlow, the open-source technology developed at Google. The models are quickly and easily hosted on containers run with Google Kubernetes Engine or Compute Engine virtual machines. "Machine learning is a very important tool for us but we have to be stringent with it in a medical context where the stakes are high," says De Jaeger. "We need to use the cleanest data we can, and as much of it as we can, to train our models if they are to have the accuracy we require."
The goal for AZ Delta is not to have machine learning algorithms decide on a course of action. Rather, it is for the algorithms to present the relevant information to the physicians, empowering them to find the best course of action faster. One algorithm, for example, might be trained to predict anomalies in prostate cancer patients, which a doctor might only catch after years of experience and intuition. Or a patient might be at a high risk of reacting badly to a course of treatment and the algorithm could provide a prediction of that likelihood. The doctor could then decide on a different course that better suits that individual’s need.
Now, AZ Delta is planning to extend the analytics platform to other patients. As well as adding more EMR data, De Jaeger is also keen to add other kinds of data, such as electrocardiogram and image data from the hospital’s pathology labs for greater analytical insight. With a secure platform and a solid understanding of how to get the most out of it, he’s optimistic about the path ahead. "The power we have access to now with Google Cloud means that we can do things that were beyond the scope of human ability. So, if we can combine the strengths of our physicians with the latest technology, I think we’ll be able to take a big step forward in health care."
Tell us your challenge. We're here to help.
Contact us
About AZ Delta
One of Belgium’s largest hospitals with 1,400 beds and around 650,000 annual patient consultations, AZ Delta aims to be a leader in high-quality care, prioritizing innovation and continuous dialogue with patients.
About ML6
With more than 100 AI specialists and ML engineers in Belgium, Germany and the Netherlands, ML6 helps organizations create true business value with AI.