Why voting and healthcare have the same data challenges
The election is over and Donald Trump will be President of the United States, which feels shocking simply because much of America was not expecting it. Almost every pollster and data scientist who tried to predict the election is shocked at the outcome. The failure of predictive analysis has been one of the bigger stories post-election.
What went wrong with the polling and predictive analytics for the election?
Are there lessons in what went wrong with the data that we can leverage in healthcare?
Mr. HIStalk had a great post after the election where he bulleted several of the points I make below. It is worth a read.
The interesting observation for Datica is how voting at the national level and making decisions about your health are eerily similar.
Polling and population health are very similar, in fact, which is a relevant topic since we find excessive hype around analytics, and more broadly data and tech, as the panacea to fix healthcare. There is no doubt that tech, data, and analytics have a huge role to play in improving healthcare but we must temper the hype. Fixing healthcare won’t be as easy as plugging in an app or a pop health model.
It’s not just about data.
Data was at the heart of this election. To be sure, it was a part of the last election in 2012, but we’re a lot more sophisticated today than we were four years ago. The Republicans acutely realized they got beat on data in 2012. (Well, a lot of Republicans that worked for Romney think that.) In this election, data — more than traditional conventional wisdom about demographics— drove 100s of millions of dollars of ad-spend. Data was supposed to accurately tell Democrats if they had enough urban votes to carry the rural vote. Ultimately, the data—or more accurately the predictions based on the data—were wrong. Patient-level data holds similar promises for healthcare. It should be able to tell us what groups need what interventions and based on what triggers. It should tell us how we’re performing against quality-based payment programs and ultimately what our penalties or bonuses will look like at the end of the year. We’ll need to temper our reliance on data-driven predictions because incorrectly predicting who will be readmitted or have a bad outcome is costly for health systems with razor-thin margins.
Individuals aren’t populations.
Population health is hot right now, just like we focused on specific groups of voters in different electoral college areas in this election. I like to think some of the hype around population health has been tempered over the last couple years by a sense that humans, both patients and clinicians, operate differently when you take them out of a patient registry or excel spreadsheet. With the election, we saw firsthand that humans act differently when viewed outside of their group. Maybe they don’t want to admit to voting for Trump just like they don’t want to admit that they didn’t wear their CPAP machine or finish their med pack. In any case, patient reported data needs to be filtered in some way to account for reporting and collection bias.
Choices are disconnected from outcomes.
An additional challenge with choices at the national political level, which largely mirror the problems with individual healthcare choices, is the gap between the choice and the effect of the choice for the individual. Trump winning, which does have an emotional impact on a lot of people, doesn’t materially impact the lives of most people today, and likely will have minimal impact even over the entire course of his presidency. This isn’t the early days of the US when presidents could actually accomplish their entire agendas in two, and sometimes one, term. Nor do we even expect that. Healthcare decisions are the same way. Gaining five pounds a year and an inch on your waistline, outcomes that are the result of multiple daily lifestyle and health choices, are typically only felt when some acute event, like an MI or diagnosis of diabetes, occurs. And these acute events are the result of years or even decades of decisions.
There isn’t a one-size-fits all approach.
People are different from one another. It doesn’t matter if you label them as voters or patients. We, and our preferences, experiences, and motivations, vary wildly by sex, age, race, culture, socioeconomic status, geography (urban vs rural comes to top of mind for the election). As such, there isn’t a one size fits all candidate or even one size fits all message to influence voters for each candidate. Voting, like healthcare, is very personal. And healthcare, in targeting messages and even modeling analytics, needs to take into account our differences.
People aren’t rational.
It’s hard for a lot of Americans to comprehend results that differ from their choice. It’s the same for healthcare where we can’t comprehend why people don’t make evidence-based choices. Technology, driven by smart, up-to-date data, can remind somebody to take their meds but that doesn’t fix non-compliance. Oftentimes the reasons people are non-compliant are just as mind boggling as people’s vote selections. It doesn’t mean the patient and voter are crazy but it does mean it’s not purely a data and analytics problem. Ultimately, individual health is largely about individual choice, informed by tech and analytics or just gestalt or whatever somebody feels like that day. Those are deeply personal and hard to model.
It’s worthwhile to parallel healthcare and elections. The fortunate thing in these lessons, for healthcare at least, is that data, and the predictive analytics that go hand and hand with data, are relatively new both in elections and in healthcare. Healthcare has the distinct advantage higher velocity, variety, and volume of data to refine models to constantly improve models. Just as the next presidential election will see much more sophisticated data and analytics, so will healthcare.
