April 05, 2020

I.V.-therapy device innovates using Stratus Data’s machine learning algorithms

A Peripherally Inserted Central Catheter (PICC) is a long, thin tube that delivers medication through the arm to veins in the heart. More than 3 million Americans receive PICCs every year. At present, the PICC placement process is painstaking and potentially inaccurate. Nurses mark spots on the arm and chest, and then measure between them to estimate how far the PICC will travel. Some cases even require a chest X-ray. This process can result in serious complications; for instance, 17% of improperly positioned PICCs are placed into the right atrium, which can be life-threatening (Hostetter, Nakasawa, Tompkins, Hill 2010*).

Piccolo Medical has developed Smart PICC™: an innovative, low-cost device for PICC placement which does not require X-rays and manual measurements. In 2019, Piccolo was awarded $1.4M by the National Institutes of Health for research and development.

Stratus Data has worked with Piccolo since the device’s early stages: refining goals, improving algorithm accuracy, and helping achieve FDA clearance.



The early days: ideas and raw data in place, but not yet ready for prime time

The Smart PICC™ uses innovative sensors and Machine Learning algorithms to provide the nurse with a real-time guide to the correct destination.

Piccolo began its data science voyage in 2017 with a 300-line Python script which provided a basic framework and methodology to evaluate sensor data to determine PICC line location and direction. However, the journey to a foolproof solution would require specialized resources. Not only did Piccolo have to refine the algorithm, but they also had to gather evidence to prove the device’s safety. Moving forward, Piccolo needed a Data Science team to help make sure that the proprietary methodology worked every time, all the time.

Says COO John McKenzie of the interview process: “We had five different recommendations for independent contractors or small firms. […] Everyone else said, ‘It’s going to be hard to see what techniques we’re going to use until we get into it.’ Stratus said, ‘Here are options a, b, and c for approaching the machine learning algorithm. Let’s dive in and see.’” According to McKenzie, Piccolo was impressed by Stratus’ methodical approach to every detail, as well as great diversity of knowledge.

“We needed a data science team that was nimble, who could work with us, not someone who needed us to clean up the millions of data points ourselves and put it on a silver platter for them to do the work. In the medical device industry, when it comes to getting algorithms approved by the FDA, it is often greenfield territory. However, the agency clears products based on valid hypotheses that are tested numerous times to demonstrate efficacy with statistical significance. Stratus showed a willingness to work around these constraints.” – John McKenzie, COO, Piccolo Medical

“I’ve worked with a number of data scientists. What I like about Stratus is, they’re able to communicate fairly complicated concepts in laymen’s terms. While admittedly I’m not an expert in data science, when I have spoken to some data scientists, it’s as though they’re not even speaking English. The communication element in complex science is critical.”
– John McKenzie, COO, Piccolo Medical


Stratus helps Piccolo win grants and 510(k) clearance, continues collaboration through human trials

The Stratus team began by studying the data, learning the product, and assessing Piccolo’s existing approach. “We knew the goal: to get this object safely to a specific place in the heart,” says Stratus principal data scientist J.Y. “But we needed to frame that goal very carefully as a balance between technical and physical constraints.” Stratus and Piccolo collaborated closely to define quantitative goals that could be measured and evaluated.

Fully underway, the project involved asking questions like:

  • What patterns are we looking for?
  • How can we filter out any noise? **
  • What mathematical transformations should we apply to raw data so that it becomes more obvious to any algorithm we use?

The working relationship started with one or two milestones. Over time, it evolved into a long-term collaboration, progressing through data analysis for grant research, culminating in FDA 510(k) clearance and human trials. Stratus has helped Piccolo build higher reliability models, design experiments, and iteratively develop a flexible, yet robust algorithm.

“When working with the Stratus team, it’s always been a process of continually redefining goals and metrics as we hit various objectives,” says VP of Research & Development Augie Shanahan. “It’s felt like a collaboration, not like a ‘here’s what we need, go solve it.’”

“Stratus’ thinking from an analytical lens helped put data science elements in context for us. Working with Stratus has helped us formulate new strategies. That open, working relationship is a key factor.” – Augie Shanahan, VP of Research & Development, Piccolo Medical


Increased placement accuracy was only the beginning

Piccolo started in 2017 with a promising concept. Stratus and Piccolo worked together to gather evidence, confirm technical hypotheses, and bring the product to market. Early on, in the R&D phases, the partnership resulted in a significant increase in placement accuracy. Today, Stratus continues to support Piccolo’s data efforts as the team works on their second-generation product.

PICC and central lines are a multibillion-dollar market, and Data Science will become a critical component in guidance systems. “In medicine,” says McKenzie, “especially something like this, we don’t have wiggle room to be wrong. The business value of being right is large. If we nail this, it’s a billion-dollar solution.”



* Hostetter R, Nakasawa N, Tompkins K, Hill B (2010). Precision in central venous catheter tip placement: A review of the literature. Journal of the Association for Vascular Access 15(3):112-125.

** Sensor data is inherently noisy, due to factors like electronics, temperature, and the presence of other anatomical processes.

Send us a message to learn how your team can incorporate data science into your technology.