From data ingestion to production in less than 30 days: How Zephyr AI uses Rhapsody Semantic to create precise AI models at scale

Zephyr AI combines data science with engineering to help health providers and plans transform prevention, care delivery, and treatment. The Zephyr platform ingests real-world data, then uses transparent artificial intelligence (AI) and machine learning (ML) to develop predictions about outcomes. Zephyr​’s​ proprietary algorithms and predictive analytics make sense of massive data volumes to anticipate risk and prioritize opportunities that improve outcomes and lower costs.  

Zephyr combines data from medical and pharmacy claims, EHRs, medical devices, genetic sequencing, cancer registries, molecular clinical data, and more. Some datasets include ICD, SNOMED, CPT, and/or LOINC codes, while other data ​are​ semi-structured or unstructured text. 

Zephyr works to investigate numerous different use cases and disease states, such as the severity and prevalence of various diabetes complications. Zephyr algorithms examine diagnostic codes, time spent with a provider, and lab results to better predict which patients are most at risk for complications — helping providers target these patients for additional monitoring or treatment. This approach helps improve both outcomes and cost of care.

The Zephyr platform ingests data from countless unique data sets, with unstandardized formats and codes. The more high-quality, accurate data that Zephyr can use, the better the outcomes they can provide to their clients. As Amy Sheide, vice president of data platform and partnerships, summed it up, “More data, more better.” 

However, Sheide noted, “When the data lands, it’s very disorganized and ​algorithms​ can’t make sense of it​ as is​.”  

Her colleague, data specialist Samantha Pindak, added, “Clinical data is messy, and nobody uses the same codes. In the clinical data space, EHR records are not standardized. The more data we see in different spaces regarding different diseases, there’s going to be a lot more work with obscure coding systems.” 

Before Zephyr can begin applying its machine learning expertise, it must first standardize the data. That means: 

• Mapping text strings to standard codes 
• Validating any codes that were sent 
• Confirming formatting 
• Knowledge modeling for implicit features​ 

Zephyr’s “small but mighty” data team of three data ​specialists ​needed help. Sheide said, “We’re not going to be passing spreadsheets around. We’re ​not uploading​and downloading. We wanted our workflow to be in the tool.” 

The team also wanted to use its own APIs to protect data security and create efficiencies for their ​engineering and data science ​team​s​.  

However, with Zephyr’s plans for rapid growth, the data team must onboard and standardize new customer data to be applied across models — at scale. Since machine learning models are only effective when they receive enough data to learn from, that scaling must happen quickly to enable Zephyr to start generating useful insights for new customers.  

“We needed a solution that allowed us to leverage our content management expertise, meaning it needed to be easy and flexible,” Sheide said. This means letting Zephyr’s terminologists and data specialists manage custom code sets, generate their own mappings, and build their own mapping algorithms.  

Zephyr needed a solution that would allow them to scale to quickly onboard new customers and start generating insights. This required: 

• Ensuring data specialists have access to the content they needed  
• Allowing them to manage and adjust the content mappings and subsets as needed  
• Including with robust workflow management and integration tools that didn’t rely on spreadsheets or custom coding to connect to other systems 

The team chose Rhapsody​®​ Semantic terminology management to standardize data at scale. With Rhapsody, Zephyr data can quickly make sense of the data so that it can be passed to engineering to ​train and test ​machine learning models.  

Now, Zephyr data managers can input custom code systems directly from the source, then translate them to the target without passing around spreadsheets​ or requiring custom ETL for each data partner ​. Since the workflows are built right into the tool, the Zephyr team saves significant time and effort with automation. 

Rhapsody also allows Zephyr to filter out irrelevant data​ and curate features of interest​, leading to more actionable insights — such as predicting which patients are at the highest risk ​for​​ ​a​n adverse outcome such as a​​ ​diabetic foot ulcer.  

“It goes from looking for a needle in a haystack to simply a pipeline that just matches it for us,” data specialist Nathaniel Tann added. “That’s huge for scalability, being able to easily get the data that we’re interested in from extremely large sources.” 

“It goes from looking for a needle in a haystack to simply a pipeline that just matches it for us,” data specialist Nathaniel Tann added. “That’s huge for scalability, being able to easily get the data that we’re interested in from extremely large sources.” 

Plus, Rhapsody empowers the Zephyr data team to set user groups with distinct permissions. Each team has exactly the right level of access to do their work, protecting data security.   

“The fact that our data managers can go in through the UI and manage the permissions and the code sets and make it operational for engineering is fantastic,” Sheide added.