Saturday, March 7, 2020

Predictive Analytics- Not as Scary as it Sounds

Hi There!

This is a post I've had heavy on my heart; growing in immensity for about 2 years now in my brain.

I wanted to take a little time to do my best to break down exactly what I am studying and make it possible for anyone to understand.

But first, a little background.

I am a Registered Nurse, and have been for 8 years. I have been in healthcare for 13 1/2 years. About 4 years ago I took a position that actually removed me from the bedside but instead I got to be a part of helping to design and change our electronic health record in a way that improved patient safety and made more sense for nurses and physicians to use.

I remember within my first month on the job, I was quickly drawn to the fact that we could use this electronic health record to capture data and summarize data on hundreds of patients. Better yet, in this new electronic health record, we had abilities to build our own reports! Someone very influential to me, someone I think was placed temporarily in my life for the sole purpose of steering me in my life's direction, sat with me or initially two hours and gave me my first high level crash course in reporting. I-was-hooked!

From there, I continued to grow and expand the use of reporting and my passion for it became apparent. I became the person on my team that was known for being able to whip up new reports and to also be a whiz in Microsoft Excel. These skills allowed me to not only create and put together data from within the system, but to also summarize and track that data.

About a year or so into the job, I knew I had to advance my degree. I have a Bachelor's of Science in Nursing, but I needed more. I've always loved school and am a total school nerd. I enrolled into my local college for a Masters in Nursing with a clinical nurse leader track. I was doing well, about 2-3 courses in- but......I hated it. Guys, nothing felt more nauseating and not right to me in my entire life. It was all research papers and evidenced based practice and blah. Not for me. I didn't really know what else to do, so I switched to a Masters in Nursing Administration. In this track I started to see how cool the whole business aspect of the degree was, but again felt nauseated in what I was learning with regards to how broken our healthcare system is. I spent years on the floor doing my best and it was never enough to the same people that I was in a degree trying to become. Again, it didn't feel right. I took a break from school when I was pregnant with my daughter and in the post partum period as well. I thought long and hard about returning and what felt right.

Then I saw it: A new program at the same college- It was a sign! A Masters in Science in Predictive Analytics. Technically this isn't a healthcare degree, its a business degree. I always had felt that I had to stay in nursing because that's what my background was, even if I hated the pursuit of anything further in it. But this degree.... it felt right in my bones. Multiple people have told me that when I start talking about data analytics/predictive analytics, they can visibly see the light in my eyes go on fire, the rise in pitch and rate of speed in my voice increase, and I could talk about it forever. That right there everyone, is passion. And I knew I had to pursue it. I always joked in the last 4 years in my job that if I could wake up and spend all day in an Excel spreadsheet, I would love-my-job. Data to me makes sense. Its tangible but also infinite. It's like a math equation. There's an answer at the end but also infinite answers! It makes me jump up and down inside, can you tell?

So I started this program in January of 2019, when my daughter was ~ 5 months, leaving my husband to watch Abby for the 3 hours a week I spent in class. It sure has been a whirlwind ever since, and I still am spending 12-15 hours a week involved in the program with homework, reading and in class time. But, what's difference in this is that those 12-15 hours don't even feel like "homework" or "work". Sure I need to make sure I get it done within the deadlines, but its FUN! When I make a dashboard or create a model I get a sense of immense joy and celebrate in my newfound abilities. It kind of feels like what I imagine it felt like for Spiderman to suddenly wake up and realize he has superpowers. This is my superpower.

But, as I started meeting other moms and introduced myself and what I do or mentioned my degree, or talked with friends and family about it, I started noticing a frustrating pattern: No matter who it was, within a minute of talking about it, people's eyes started glazing over and they'd switch the topic. I told new people that I was getting my degree in Predictive Analytics and they acted as if I said I was getting the degree in astrophysical quantum mechanical engineering (if that's even a thing?).
It was sad for me because this was my passion and I wanted to tell the world about it. To me, at work or in school, it was all common talk and people understood it, but I wanted to be able to share it with more than that.

So, I'm here to break it down as best I can.

I don't want to give any fancy definitions of what it is, because even I don't understand those. Lots of statistics and blahblahblah.

To me, predictive analytics is as simple as using and understanding data from the past to predict future events. That's it!

Ok so let me break it down into a real elementary example, then I will advance a little from there.

This example brings in my love of epidemiology also, so its fun. Let's say there was a big gala held last weekend celebrating something. 100 people were in attendance. There was a lot of food served, but lets focus on 5 foods: They served chicken, salmon, filet mignon, and ham as the meat choices. They had a big salad on the side for vegetarians.  However, unfortunately, a few days later the party hosts found out that 20 of their guests got really sick! They called and blamed the food they ate. So the hosts had to find out, what was it that made everyone sick? So they interviewed each of the sick to find out what they ate. But what is important here, is where predictive analytics comes into play is not just interviewing what the sick ate, but what the non-sick ate too. For example, if the 20 sick people all had the ham, but so did 70 non-sick people, it's probably not the ham that made everyone sick. But how could we really say for sure? Well, nothing is ever really sure in data analytics. Data analytics treats words like "always" and "never" like the black plague.

So what we do here is treat each one of those food items as variables. (I'll explain that more). Each variable would be "Did this person eat the ham, yes or no?" or "Did this person eat the turkey, yes or no?"  By then viewing the list of the 20 people and what they ate, we could likely identify a pattern of the common thing(s) that the people ate that got sick. OK now stay with me- we can use that knowledge of understanding what we THINK got sick and assign them values. So perhaps those that ate the turkey were 80% more likely to have gotten sick and assign it a value of 0.8. We now PREDICT, will the rest of our guests get sick? So we take that list we have of our 80 non-sick people, and essentially run it through a model to determine a prediction of how likely that person is to be sick. If we found out that ham and turkey were the culprits that made everyone sick, if we took one of the non-sick people and found out he too had ham and turkey, we could give him a pretty accurate high prediction that he may end up also getting sick. However, if one of the non-sick people say they only had the salad, and non of the sick people ate the salad, we can give that person a pretty accurate prediction that they won't join their friends in the quarantine.

This sounds like a silly example, but its how these predictions work. In examples I am involved with at work, the "20 sick people" that are used to learn and understand the variables, is actually 500,000+ people (perhaps more likely in the millions). Taking data from only 20 people can lead to a lot of inaccurate predictions. Maybe they were lying or forgot details. But if we are seeing common patterns on half a million people, we can gain better accuracy in applying what we learn to future predictions. Instead of the "variables" being 5 food groups, we instead look at hundreds of variables to better understand everything we can about these instances. In this way, we actually can learn new things that may be correlated to a disease or effect.

Some examples I am working with in healthcare now are predictive models that can predict how likely a patient is to have a cardiac arrest event, or code blue. We use close to a hundred variables that were studied on thousands of patients that DID have a cardiac arrest (think back to the gala example-the people that DID get sick). By learning what was unique about those patients that did have a cardiac arrest, we can then match that set of variables up to a new patient that walks in the door, to see how closely they align to those variables. Perfect match? Yikes, lets get the code cart ready! Hardly a match at all? Pretty good chance they aren't coding on our watch (Again, nothing is certain).

But think of how useful these models are to us, in healthcare and around the world (I'll get to that). By having a prediction like how likely a patient is to have a code blue, we can better prepare as healthcare workers to make sure that doesn't happen. We can better plan staffing ratios, adjust the treatment plan, get other specialists on board, etc.

I'm working on a few other models in healthcare right now, but there are more than I can even count that are actually out there. We are gathering the ability to predict who will experience a fall, who will contract sepsis, who will come to the emergency department, who will not show up for their appointment, who will come back to the hospital, etc. In each of these cases, these accurate predictions allows us to prepare.

But think of it more globally. These same concepts can and are applied to other "predictable" events. We could essentially predict tidal waves, hurricanes, tornadoes by understanding what variables or conditions were in place before those events happened in the PAST to help us predict the FUTURE. We could predict mass shootings by understanding conditions and variables that were attributed to past mass shootings. We could predict train delays or breakdowns, airport delays, stock markets, retail surges, etc. With the growing amount of data being made available to all of us, even you, we will have more and more ability to predict almost any event that can have attributable data to it.

Anyway, I hope I didn't lose you all! If you can't tell, this stuff is my life's passion and I can't wait to see where it takes me. This is really just the beginning. There is so much more involved to it all than what I wrote about in this blog, hence the need to get an entire degree in it (and even that will barely scratch the surface), but I wanted to at least get the concept out there as easily understood. Maybe I'll even inspire someone to go chase the dream for themselves, too!

Thank you for reading.

Toodles!

_ The New Motherboard

No comments:

Post a Comment