Why healthcare is a good fit for machine learning
Healthcare is one of the most data-intensive fields in the world. It generates massive amounts of data from diverse sources – electronic health records, medical images, wearable devices, research results, and more. Because machine learning algorithms need to ingest large amounts of high-quality data – healthcare is also arguably the field with the most opportunity to drive innovation with AI.
As in any data-rich industry, there are many use cases for companies to apply machine learning to drive innovation, improve patient outcomes, and reduce costs, such as:
- Cost savings
- Exploratory data analysis
- Predictive analytics
- Data-driven healthcare products
- Improved patient experience
- Evaluating and improving existing ML models
- Operational efficiencies
- MVP ML products
- Medical research
- Drug discovery
In this post, we’ll dive into seven real-world projects Tribe has worked on with healthcare companies to drive innovation, improve outcomes, and reduce costs using machine learning.
Real-world use cases for ML in healthcare
AI-driven fertility startup worked with Tribe to define technical approach and implementation strategy
Use case: MVP ML product
Opportunity: A healthcare startup with one of the largest and most diverse datasets of women’s fertility data was looking for strategic advisory support to design a machine-learning driven fertility health assessment and consumer-facing product.
Team: Tribe brought on an expert in health data, natural language processing, and machine learning engineering to design a strategic approach to develop, evaluate, validate, and productionize a model.
“Katie [Tribe consultant] has become an indispensable member of our team. I trust her completely and even included her as part of our leadership team in our fundraising materials.”
Founder and CEO, Fertility Startup
Medical device company deployed customer churn prediction model and reduced engineering spend by 20%
Use case: Reduce customer churn
Opportunity: A public medical device company was seeking to predict and reduce customer churn using data science. They had created a few data models, but did not have the expertise to operationalize them — or the confidence to scale future data projects.
Outcome: Tribe built a platform that automated the process to operationalize any data models. By the end of the 12 week project, the client was using their customer churn data model to predict and intervene on churn. The automation increased the engineering capacity of their team by 20%, reducing engineering spend by over $500K in 2022 and multiples of that YoY as they scale their team.
“After Tribe joined, we went from 0 to 100 in terms of time to market. This work was a jet propulsion for our team.”
Data Science Leader, Public Medical Device Company
Non-profit medical center built data-driven diagnostic tool
Use Case: Developing diagnostic technology
Opportunity: A non-profit academic medical center wanted to build a diagnostic tool using eye tracking. They needed support building a data warehouse of clinical data, evaluating the current algorithm, and the quality of the results. In addition, they wanted to explore what additional signal could be found in the data.
Outcome: Tribe helped built a robust data infrastructure, evaluated model efficacy, and conducted an exploratory analysis on clinical data in support of product development efforts. The Tribe team approached this in two phases to align with critical milestones for a public product launch. Phase one focused on data aggregation and model evaluation leading up to a commercial launch, and Phase II focused on technology refinement and new product development.
Public pharma company increased GPU cluster efficiency to capture $2.8M in annualized net value
Use Case: Cost Savings
Opportunity: A nearly $1 billion pharmaceutical company was leveraging machine learning to reimagine drug discovery and development. The company had made a significant investment in a 100+ node HPC for distributed training. They identified a number of issues that suggested the cluster was operating below peak performance (~50%), but were not able to verify or increase cluster performance.
Outcome: Tribe established a baseline benchmarking framework and test harness to validate the clusters theoretical scaling limit now and in the future. In just 2 weeks, Tribe was able to identify the core problem and increase the performance efficiency of training jobs by more than 35%, capturing an estimated $2.8M in annualized net realized value (~10+ extra compute nodes).
“Tribe came in and systematically benchmarked and tested our infrastructure, eventually implementing a number of improvements and fixes that led to massive cost savings.”
CTO, Public Pharma Company
Healthcare data company leveraged ML to build accurate, scalable real-world evidence
Use Case: ML Ops Advisory
Opportunity: Tribe was approached by a company that de-identifies, enriches, and integrates data from electronic health records, medical claims, and product and disease registries to produce “real world evidence” (RWE). This real-world evidence provides clinically-rich insights into what actually happens in everyday practice and the use of drugs or medical products and why. The company wanted to build a strong ML ops foundation that would allow them to generate research-grade evidence at scale with accuracy necessary for market access, medical affairs, and regulatory use.
Outcome: Tribe advisors helped the company establish a solid ML Ops foundation from the outset. This includes guidance in increasing the efficiency of deep learning pipelines, deployment models in Flask REST-based services, and communicating industry best practices for models deployed in air gapped systems. The result was the acceleration of model development and deployment.
Veterinary radiology company used AI to diagnose medical conditions from x-ray data
Use case: Model evaluation and improvement
Opportunity: A leading veterinary radiology company needed help improving the models they had in production, as well as improving new, prospective models where performance was only slightly better than random.
Outcome: The Tribe team was able to realize model improvements upwards of 30%, which increased the types of diseases the client was comfortable diagnosing. Leveraging a lime-based classification tool (human-in-the-loop) they were also able to change the way their customers interact with the product, leading to huge qualitative gains in the user experience.
“We found that it added incredible value for our radiology services. Our patients are able to get care faster, and our clients were really pleased to see all the good things."
Crisis recovery marketplace used advanced analytics to reduce waste in disaster relief efforts
Use case: Resource allocation
Opportunity: A crisis recovery marketplace startup wanted to use ML to remove waste and fraud from disaster relief efforts, so people affected can get the exact help they need the moment they need it.
Outcome: Tribe used ML and advanced analytics to calculate a crisis impact score based on data about geospatial area, the disaster itself, and individual need. This score enables organizations to respond with relief – from providing food and supplies after natural disasters to allocating Covid vaccines – in a more equitable and effective manner.
Challenges in healthcare ML projects
There are unique challenges to using machine learning in healthcare, especially around data quality and privacy. ML algorithms require large amounts of high-quality data to be trained effectively. However, healthcare data can be noisy, incomplete, and unstructured, which can make it difficult to use for machine learning applications.
In addition, Healthcare data is sensitive and must be protected to maintain patient privacy. This can make it difficult to share data for machine learning purposes, especially if the data is stored in multiple locations and owned by different organizations. There are also challenges around regulatory compliance and interpretability, since any decision can have serious consequences for patients.
But all of this – the rich data, the high impact of insights, the improvement of human lives – is exactly why healthcare presents such a huge opportunity for impact from machine learning. By partnering with machine learning experts experienced in using healthcare data and focusing on specific use cases, healthcare companies can mitigate risk, drive innovation, and make a real difference in patient outcomes.