What Are the Real-World Challenges of Operationalizing ML in Healthcare? DeepLearning.AI Learner Community Event ft. Neelesh Kamkolkar

TL;DR

Operationalizing ML in healthcare requires reliable data, careful production monitoring, collaboration with domain experts and end-users, and a service-oriented mindset. Neelesh Kamkolkar draws on projects involving more than 10 years of customer data, over 9 million encounters, and use cases such as length-of-stay predictions, discharge disposition, and wellness recommendations. Read on for practical lessons from moving healthcare models beyond proofs of concept.

Transcript

good morning good afternoon good evening wherever you are thank you so much for taking time out of your day and joining us here today let me start by sharing my screen so welcome welcome to the deep learning day I online mini event series or AI from medicine learner community I'm honored to be your guest speaker today look forward to sharing and le... Read More

Key Insights

• "we had over 10 plus years of data from some of our customers seven seven plus years from some other customer" (3:14)
• "you get an Excel spreadsheet with 10,000 rows of data and someone says you know hey create a model" (3:37)
• "part of that you know we were delivering millions and nudges to to many many people" (5:49)

Questions & Answers

Q: What are the main challenges of operationalizing machine learning in healthcare?

The challenges include data correctness, feature construction, model development, production performance, and ongoing maintenance. Healthcare projects also need collaboration with domain experts and end-users, along with a service-oriented mindset.

Q: How much healthcare data did Neelesh Kamkolkar's team work with?

Some customers supplied more than 10 years of data, while another supplied more than seven years. One dataset covered more than 9 million encounters, excluding labs, vitals, and other data points.

Q: How do healthcare ML projects move from proof of concept to production?

The team often began with a proof of concept that a client then wanted to convert into a pilot or production system. That transition produced many of the real-world lessons discussed in the DeepLearning.AI Learner Community Event.

Q: What healthcare ML models did the team build for providers?

The team built models for use cases including length of stay and discharge disposition. These models estimated how long an admitted patient might remain in the hospital and where the patient might go after discharge.

Q: Why is choosing a healthcare ML use case challenging?

Healthcare presents an incredible number of potential use cases. That breadth can make it difficult to decide which problem a team should pursue.

Q: How do you handle data validation and correctness in healthcare ML projects?

The existing guidance emphasizes syntactic and type correctness, ontology mapping, and morphological correctness. Teams must also consider data scale and entity uniqueness when building reliable models.

Q: Why must healthcare ML models be monitored in production?

Production monitoring and maintenance help preserve model performance and support accurate predictions. The presentation treats this continuing responsibility as essential to achieving suitable healthcare outcomes.

Q: What recommendation systems did the team build for healthcare and wellness?

The team built recommendation systems that directed people toward wellness opportunities, such as neighborhood walkathons and programs supporting healthy eating and shopping. In one described setting, the systems delivered millions of nudges to many people.

Summary & Key Takeaways

• The speaker shares their personal background and involvement in volunteer work before diving into the topic of operationalizing ML in healthcare.

• They highlight the challenges of working with healthcare data, including data correctness, feature construction, and model development.

• The speaker emphasizes the importance of monitoring and maintaining the performance of ML models in production, as well as the need for a service-oriented mindset.