Finding tech's place in managing health records for life insurance

April 3, 2024 5:00 PM

Swiss Re's New York headquarters building, 1301 Avenue of the Americas.

In life insurance underwriting, carriers are beginning to incorporate clinical health data like electronic health records in addition to traditional medical data. These carriers have to figure out what data from these records is relevant to their underwriting. Swiss Re Reinsurance Solutions Underwriting, a division of the reinsurer Swiss Re Group, offers risk solutions for life and health carriers. Digital Insurance spoke with Nanditha Nandy, head of data driven underwriting solutions, Americas at Swiss Re, about how technology for health data, instrumental to underwriting life insurance, is changing. Nandy leads and sets strategy for Underwriting Ease, Swiss Re's data-driven underwriting solution.

How has underwriting technology transformed? And how has Swiss Re implemented this transformation?

Nanditha Nandy of Swiss Re — Nanditha Nandy, head of data driven underwriting solutions, Americas, Swiss Re.

We created a tool that ingests the output of automation and flags where it has already assessed the information coming from different third-party sources. Carriers order this data, so we connected the information across these data sources to impairments.

We are showing underwriters the impairments where they need to pay attention, that match their underwriting philosophy. This saves time, because when they open and see a case, that starting point of the risk is already shown to them, tied to the kick out reasons – why it was referred out.

The next pain point is new clinical data sources coming in – electronic health records. These were being looked at as siloed PDF files. These are 50, 100 or 200 page PDFs. This electronic data, this newer data hasn't gone through automation. There is information you should care about among a lot of noise. So we take away that noise and digitize everything into one single risk view for the underwriter. Another piece is attending physician statements. Underwriters have to look at those, then call specialist physicians to evaluate.

So we digitized attending physician statements using natural language processing (NLP) and AI and we presented a view, so when you look at the case, no matter which underwriter is looking at it, at what point, it gives you the same story, that same consistency, everything digitized in one beam.

How do you program the AI so it doesn’t just produce what it thinks the user wants to hear?

One size does not fit all. Are you sure you did enough testing? We are not saying that this is not the future. Underwriting has to be transparent and explainable in decisioning. It's good for us to know all the reasons. Some models can explain, some cannot. If it goes through a model that is more black box, then it becomes harder.

How have these technology improvements made underwriting faster? What are the dangers of this increased speed?

What used to take 14 days or more, on average, is now a couple of days, coming from the same data sources. The process is completely transformed. For every applicant, we recommend what combination of data sources is enough to make that final decision. It needs to be designed in a way that makes sense for every company based on their underwriting decisions and their thinking.

We have spent years and a lot of investment in underwriting knowledge, clinical knowledge to build up to this point, so you can continue your process. We are not disrupting anything, but we are setting you up for the future, but in a transparent way, in a way that we can answer. We are not using AI in a black box fashion. Any agent can call you up and ask you, why did you come up with this decision? Can you contest it? Can you change it? You have the opportunity to change it.

How does faster underwriting square with a better process and improving trust?

You could force AI too early and say, just spit out the decision. That would be trying to do it too fast. But doing it right would be assisting the underwriter, asking if the decision is OK. As more underwriters acknowledge it, they build trust, but you have to build the trust on the machine, rather than saying a machine is going to give you a recommendation. There are edge cases and there are errors. You can learn from the majority of the cases. But it's a risk if you do it too fast.

AI is such a broad term. If you break it down into just predictive models, then NLP is purely just dealing with text. Chatbots are a different space. If you have the data right, maybe it gets the answer right, but maybe not on the first guess. It can take a few tries for it to understand what you're trying to ask. There have to be some representations on writing rules for that.

Models have always been used in our industry, to rank the cases that need to be prioritized, that underwriters pay attention to. We have used it to prioritize needs to go after when we do campaigns. And we've used it to tell where you can get the right answer from the applicant.

For decisioning, we have not been very confident in models. A model triaging whether an applicant is a smoker or nonsmoker, sends a predicted smoker through a traditional process, but may send a predicted non-smoker through an accelerated process. That's the logic of applications, not for the final decision making. We've been very cautious of telling you that you have been declined for whatever reasons that are not explained.

What is the importance of enabling life insurance carriers to use electronic health records?

Electronic health records have been used by the healthcare space for years, every day. Healthcare, medical underwriting and claims happen based on the codes that doctors and lab technicians enter.

What's changed now is because of the federal government mandate of making it electronic and mandating documentation of everything in patient systems, there's interoperability between various electronic health record systems. If you move across the country, it's tracked. It has a good impact on the customer service experience for people applying for a policy to get that policy. The challenge is these electronic health records can be very heavy, lengthy and noisy. And there are gaps. Not everything is relevant for life underwriting.