Social Determinants Of Health Case Study

Social Determinants of Health Case Study: What We Learn About Life Expectancy Through Social Determinants of Health

Intro: In this installment of our series on social determinants of health, Drs. Lisa Lines and Denise Clayton discuss our proprietary Local Social Inequities score (LSI) and provide key findings from our analyses on life expectancy.

This is the third in a four-part series of videos about social determinants of health. Other topics in this interview series include: 

• Current Challenges with Social Determinants of Health Data
• Using LSI to Improve Social Determinants of Health Data
• Use a Composite Score of Data to Find the Best Intervention Strategy

Denise Clayton: Hello and welcome to another episode of the RTI Health Advance series on social determinants of health. My name is Denise Clayton and I’m a health economist at RTI Health Advance and I'm joined today by my colleague Lisa Lines, a Senior Health Services Researcher. Previously, we talked about challenges with social determinants of health data, RTI's Local Social Inequities score (LSI), and some key applications of the score. In today's episode, we'll be talking about some of the key findings from our analyses. (0:30)

Lisa, the first question I have is out of the 150 plus variables that go into the model that creates the Local Social Inequity score, which ones are the strongest predictors of life expectancy at birth? (0:47)

Social determinants of health data reveals per capital nursing home residency is the strongest predictor of life expectancy

Lisa Lines: First of all, we have to reiterate that our measures are very diverse. We have a lot of different indicators in the models and some of them may not necessarily be traditional social determinants measures. For example, the number one predictor across the US in terms of life expectancy at birth is the number of nursing home residents per 100,000 residents and that's an indicator of a lot of different things that get at social determinants. It's about congregate housing so that is certainly a social determinant. Also, it's a poverty measure. And it's a measure of frailty and general fiscal health. People who are in nursing homes probably don't live as long as people who live in the communities and age in place. Having the resources to pay for assisted living versus going to a nursing home is another indicator of life expectancy. (2:03)

Social determinants of health data shows mental health challenges remain the second most important predictor of life expectancy

The second predictor that came up in terms of variable importance—reiterating again, these are the numbers we're getting out of the algorithm that tell us how much the error rate would go up or down depending on if we didn't know that information. It's not a beta coefficient. It's very much an algorithmic indicator. (2:34)

Number two in terms of variable importance is the percentage of adults who said that they had mental health challenges on at least 14 days of the previous month. Across the US we're seeing the depths of despair going up both before and certainly during the pandemic. We've seen a lot of problems with mental health. It's the second most important predictor in terms of life expectancies. (3:10)

And third, on the list is tooth loss prevalence which is a very interesting measure. (3:16)

Denise Clayton: That is interesting. What are your thoughts around tooth loss as a variable? (3:21)

Why tooth loss and smoking prevalence are important predictors of life expectancy

Lisa Lines: Think about who loses their teeth. Maybe there's something with diet happening like soda drinking and candy eating. Maybe there's an issue with lack of access to dental care. If you're poor and you have Medicaid or you don't have insurance or you have traditional Medicare and don't have dental coverage, the only dental care you can get might be in the emergency department and all they're going to do is pull the tooth. (4:02)

It also could have something to do with drug use which can result in tooth loss. Violence can also result in tooth loss. There's also help seeking behavior. If you think about people who never go to the doctor, never go to the dentist. Tooth loss prevalence is where you end up with a delay of care whether it's people just hating the dentist or not having access. (4:39)

The other part of that is once you've lost teeth it affects what kinds of jobs you can have, and it also can affect what kinds of foods you can eat. So, there's a lot about tooth loss that is really important to life expectancy. Tooth Loss is also associated with cardiac health as well. It's a complex measure. It's a poverty measure. It’s a health measure and it's a complex one. (5:16)

Denise Clayton: What other variables are top predictors across states? (5:22)

Lisa Lines: Related to tooth loss is smoking. Smoking prevalence rate was the fourth in terms of variable important. (5:35)

Denise Clayton: Smoking is an interesting one because it seems more like there's a causal pathway. We know that smoking causes diseases that can shorten life expectancy but that sounds different from nursing home prevalence. What kind of interpretation do you ascribe to the variable importance findings knowing that the random forest model used to create the algorithm is a predictive model and not a causal model? (6:05)

How predictive modeling using social determinants of health data can identify life expectancy drivers

Lisa Lines: It’s important to keep in mind that while these are interesting to look at you can't treat variable importance like a beta coefficient. It's not a causal model. It's a predictive model. The goal of the algorithm is to get to the best predictions. What it lets us do is open up that black box a little bit and look inside to see what's driving the algorithm. And we can't necessarily say for sure whether addressing home and community-based services instead of nursing home and other kinds of situations would result in life expectancy increases. But it allows us to get some understanding of what's driving outcomes. (7:10)

Denise Clayton: That's a great point because the nursing home variable shows up as a strong predictor. It could be to an extent because some of the same underlying drivers affect nursing home status and life expectancy. But potentially there's an additional causal pathway directly from nursing home status to life expectancy where there might be some benefit from programs that help people get the care they need in their own homes for longer. And along those lines, even though these variables are strong predictors of life expectancy, they're not necessarily causal variables. Is there something that we can do to identify the real drivers of health to help organizations that are making decisions about social determinants of health or other policy interventions? (7:59)

Lisa Lines: Absolutely. One approach we've been looking at is using more causal modeling to understand social determinants of life expectancy and other outcomes. Relying on random forest is not a good idea. You need to use different modeling approaches in order to understand the causal pathways and the causal mechanism so other regression approaches and other kinds of analytic approaches are much better suited for that. (8:32)

Denise Clayton: That makes sense. But it's nice that the random forest model and this parable starts us on a path about variables that we can explore more as part of the causal pathway. (8:46)

Well, thank you so much, Lisa, for talking with me today. It's not every day I get to remind people that correlation or in this case prediction does not imply causation but even so useful relationships can be identified with carefully designed analyses. (9:02)

Thanks to everyone for joining us for this discussion and please stay tuned for more episodes in our social determinants of health series. If you'd like to learn more, you can contact us at RTIHealthAdvance.org. We're happy to talk with you. (9:14)

Learn about health equity solutions from RTI Health Advance.