Future Cognitive Company

Thomas Davenport in his article “AI for the Real World” describes how the “Future Cognitive Company” would use Artificial Intelligence and Machine Learning to help find answers to the key strategic questions for the firm. He writes: “Most cognitive tasks [performed by AI algorithms] do work that wasn’t done by humans in the first place, such as big-data analytics.”

Healthcare providers are Cognitive Companies — today and in the future. And as Value-Based Care continues to grow, the various reimbursement and incentive models make care delivery increasingly complex, adding to the need to be a “Future Cognitive Company”. (The alternative is to be a “Non-cognitive company of the past” — we don’t want to be there.)

Paraphrasing Davenport: Big Data Analytics isn’t done by humans in the first place — and is a prime space to apply AI and ML approaches. I believe this is correct: despite large investments in analytics, BI, and data warehouses, many organizations don’t use their “Data Analytics Asset” to help solve big problems for the organization, strategic or tactical, long-term or short-term. We can augment the BI analysts with AI tools, to generate hypotheses and propose recommendations based on the data.

What would that mean for AI-augmented Analytics in Population Health? Two things:

  • Start by solving the Value-Based Care problem, as illustrated by the diagram from the Health Care Payment Learning and Action Network: Competing alternative payment models, all running simultaneously, with different rules and incentives. So how do we select the best approach for each of the patients in my population, with their own chronic conditions and social determinants? Let AI and ML help us optimize and micro-target these populations.
  • 4 categories of health care payment models

    Value-based care problem: Competing payment models     Source: https://hcp-lan.org/

  • Before we can do that, we need to invest in our data. Large healthcare data sets, from EHR and claims data, cover the processes, but in many case not the outcomes. Was this 6-year episode of treating diabetes the best possible approach for this patient? How would we know? Human or machine analysts struggle to determine which processes lead to optimal vs. sub-optimal outcomes. Rajkomar and others in their study on deep learning on EHR data state: “It is widely held that 80% of the effort in an analytic model is preprocessing, merging, customizing and cleaning data sets, not analyzing them for insights. This profoundly limits the scalability of preventive models.”

The Future Cognitive Company in healthcare needs to be able to answer big questions in Population Health and Value-Based Care. The tools exist. The missing ingredients, today, are the aggregated data sets to serve as the feedstock for deep learning algorithms. Let’s invest in data.


Links to Articles Cited Above:

Harvard Business Review: Artificial Intelligence For The Real World

NPJ – Digital Medicine: Scalable and accurate deep learning with electronic health records

Next blog: AI Progress in Healthcare — Taming the Data

Leave a Reply

Your email address will not be published. Required fields are marked *

This site uses Akismet to reduce spam. Learn how your comment data is processed.