C-Suite Q&A: SAP CMO David Delaney on Healthcare's Analytics Lethargy

This week's C-Suite Q&A is about data: lots and lots of data, and what healthcare isn't doing with it. Healthcare Analytics News spoke with David Delaney, MD, the Chief Medical Officer at SAP Health. In the course of our long conversation, he ruminated on exactly why the biggest industry in the biggest economy in the world finds itself woefully behind other sectors when it comes to data analytics. This is part 1 of a 2-part series.

I serve as Chief Medical Officer in the healthcare sector of SAP. It’s very much a market-facing role, so I am engaged with press, analysts, our existing customers and others who might be joining the SAP family. I also interface pretty significantly our product development group to help provide market feedback. I am out on the road an awful lot, travelling and talking to a lot of current and future customers, so I really get a good sense of the marketplace: what people are doing, the challenges they’re facing, and how our software is doing in helping them.

One of the areas we have been very much focused on within SAP Health is that of direct innovation with our customers to build disruptive solutions, things really designed to help organizations capitalize in an area that has been very much overhyped but where the promise has failed to reach that hype. That’s really turning big data into meaningful organizational impact. There’s a huge amount that’s been spilled about big data almost as if it’s a destination in and of itself, or it provides value in and of itself. A lot of organizations, after 5 to 10 years of that, are still scratching their heads and waiting for the impact to arrive. They’ve got ever more data in terms of raw material, but they’re struggling to actually impact broadly at an organizational level and to leverage that data.

Hype is a big target of ours, so we’re glad that came up already. What really isn’t lining up for companies right now, that have all of this data and infrastructure but no real meaningful use yet?

First of all, people talked about big data like it was the solution. By itself, it’s really just an expense, accruing and hosting information on servers. It really is raw material. Like The X-Files: “the truth is out there,” it’s likely in your data or another data set, but until you can actually pull that information out and bring it to the right person at the right time to drive a better decision, it really doesn’t matter.

It’s really been an impact gap. They’re accruing data at a frightful rate, but when you look at how decisions are really made in most organizations, most people still are making them with relatively thin information and they’re not able to bring it to bear. The challenge is really the amount of analysis it takes to get that right information out. Current tools that are in place, namely many of the disk-based technologies, really presuppose that you understand what the person wants or needs to see, and that you’ve modeled the data accordingly so that when they’re ready to ask a question it can be answered in a timely fashion. The challenge is that that’s very time consuming, and you’re basically answering case-by-case across enterprises, which is a slow approach. If the person asks something different than you’ve anticipated, you’re out of luck.

What we’re pivoting to is using in-memory computing, where everything is kept, all data, within main memory in a very simplified fashion. You can do queries and reformat them in a rapid fashion. It’s not only more rapid in terms of query performance, but the cost to change or innovate with that information dramatically drops. What we’re seeing out there is really less about a query execution speed challenge than it is that the cost of innovation being prohibitive and slow, causing organizations to not do analytics at scale.

Quality, consistency, and half-lives of data are also confounding factors. How do you go about ensuring that you’re analyzing the correct data in the correct format?

That’s a really good question and it’s one of the rubber-meets-the-road challenges, it’s one of the reasons why some of the efforts out there around advanced neural networks often have problems. Data quality challenges in healthcare are nontrivial. Estimates will vary, I’ve seen from 50 to 70% of it being unstructured. That’s a huge challenge, but even the structured stuff might really be semi-structured, it might not really adhere to a tight vocabulary or pick list, leaving a lot of unevenness and noise in the data. All technologies are sensitive to that, some more sensitive than others.

I think in terms of addressing that, it’s something that is a journey. At SAP, we’re across 25 different sectors, and 23 or 24 of them are quite a ways beyond healthcare.

It’s a little bit of a chicken and egg thing. In healthcare, until the advent of the EHR, a lot of data was just kind of captured by people who were just trying to get through their day, and the data wasn’t meaningfully consumed beyond just being displayed on the screen. The work of the EHR was viewed as just a legible record that could be read and distributed easily. To get people to use it we did a lot of things as an industry that basically allowed and encouraged free text.

We did templates, we allowed people to dictate and just dump those dictations in as blobs of text…because it’s time consuming to do structured pick lists, right? That was a big contributor to the challenge. Even things where there were pick lists, they often were not really exhaustive or as specific as they could or should be, for expediency’s sake. We gave up a significant amount in terms of data quality to achieve adoption of EHRs.

They function for the individual patients because we, as humans with good wetware, can read it and figure out what needs to happen, but it’s come around and bitten us when we try to do analytics at scale across organizations. You’re aggregating a lot of data of uneven quality.

The answer, directionally, is really to move to where other industries are doing this. That’s having a very strong master data management scheme in place and with data governance. These are things that are not glamorous sounding, it’s a lot of blocking and tackling. It’s a lot of due diligence: what are the key data elements by which your company needs to run and make decisions? It’s about really understanding where those are created, and from what they’re created, such as structured and clear pick lists, as well as each step along the way: who touches it, where it’s consumed, what decisions are made. Then you need a data quality feedback program that goes back and sees it.

Those things deployed over time are crucial, and that’s what leads to maturity in terms of analytics and decision-making. Healthcare is very nascent at this, the whole idea is very new to some healthcare organizations. Over time, it’s going to be important to begin that journey.

You described healthcare’s pursuit of analytics as “nascent,” so you’ve drawn us into another of our favorite questions here. Why is healthcare slow to adopt certain technologies, and what about healthcare needs to change so it can begin to catch up?

When you look at historically what happened, in some ways it’s not so surprising. Healthcare in the US is the largest sector in the largest economy in the world, it’s over $3 trillion industry. Only fairly recently, with the HITECH Act, did we achieve near-ubiquity for electronic health records. Before that, not many organizations had electronic health records.

What’s important about it is that the EHR is sort of like the CRM in other sectors, it’s a record of how care was delivered at the individual patient level. Before, this $3 trillion dollar industry was being tracked, in terms of care delivered, with pen and paper. It’s just crazy. The islands of digital in healthcare were billing and finance, supply chain, billing, accounts receivable and payable were digital, typically radiology and labs were digital. Everything else was all paper-based, and that made studying and understanding care delivery in any fashion at all incredibly hard, people had to do chart reviews and type data into an electronic format.

The HITECH Act, of course, helped us achieve a high penetration of HER, I think we’re at about 96% or so of hospitals now. At that point, it was only in the last decade that hospitals literally have end-to-end digital records around the care of patients. To really truly be able to manage it you have to have that digital record: the old axiom is ‘you can’t manage what you can’t measure.’

The challenge in terms of being able to leverage that record is that it’s in a bunch of silos, of course: clinical, financial, claims, and operational data typically live in a bunch of different silos, so they have to be pulled together and normalized, cleansed, enriched, and created into a single record. You need to be able to do that at a service-line level, or care delivery level, to really make sense of it. Looking at patients who are having a knee replacement, you need to look at other data at the knee replacement data to understand how the factors play together from initial referral to successful discharge: where variance is occurring, why variance is occurring, what you might be able to do about it. That’s the starting line to begin to manage healthcare like a mature industry, controlling quality and cost and achieving reproducibility.

The other piece is that in the past, reimbursement was for doing things. If a person came in again for the same knee replacement, the person in the emergency department would see them and charge, the person doing the testing would see them, the surgeon in the OR and the anesthesiologist would bill separately, then the ICU…no one really knew how they contributed or didn’t contribute to the success of that patient. No one was connecting end-to-end and managing toward a single set of outcomes that mattered to the patient and payer. The shift in reimbursement, forcing the whole chain to integrate and pursue outcomes, is what’s been the fundamental change, and it’s why analytics is so fundamentally important right now.