The Human Side of Data: Using Analytics for Personal Well-being

Rob Collie: Hello friends. You know how we like to tailor the show to both data practitioners and to business leaders? Well, as CEO of a now 70 person company, I'm much more of a business leader these days than data practitioner, and that's been true for quite some time actually. In terms of our business, it makes a lot more sense for me to focus my time on the direction of the company than it does to have me working with the tools. And I do miss working with those tools though. I mean, I fell in [00:00:30] love with this stuff 14 years ago, enough to walk away from my career in software development at Microsoft and go into consulting of all things. I'd become so incredibly jaded by my decade plus in the software industry that I never expected something as magical as Power BI. And I know, I know, technically Power BI didn't exist 14 years ago, but it's the ancestor of Power Pivot. As primitive as it was by comparison, that was still like life-changing in a way that I was not expecting from software.

[00:01:00] And that sort of love is the kind of thing that doesn't just go away. So for instance, when an opportunity came along to do something like, I don't know, build an overly extravagant set of dashboards for my recreational hockey league, I basically vanished into that for the duration of the winter holidays this past December, and we then went and did two podcasts on it. Heck, I even learned some cool new things while building those out. I learned to use the [inaudible 00:01:26] function. It was awesome. And along the way I also got smacked up upside [00:01:30] the head by one of my own core self professed principles, which is work backwards from the stakeholders, work backwards from the verbs, which is why there were two podcasts after I had that reawakening. Came back, revisited the same project. It was awesome. And throughout that process, including on the podcast, I kept saying things like, "Wow, it's good to be back working in the data tools again."

I realized recently though that that statement was completely [00:02:00] wrong. I've been working with a very sophisticated data model every day for several years now. Refreshing it every day, consulting it every day, and tweaking either the model and or the reports probably about once a week. So if I'm using the tools, I'm using this thing every day and modifying it basically every week, how the heck did I seemingly forget about it? How did I find myself saying, "Oh, it's so good to be back [00:02:30] in the tools again?" Well, it's actually because this other data project, this other data model and usage of the tools, has become so integrated into our lives here at the Collie household that it doesn't even feel like data work anymore. Now when you think about it, that's the goal of all data work, isn't it? It's like analytics nirvana when the tools fade into the background and it's just the other thing, the original purpose that's foremost in your mind at all times.

And business [00:03:00] leaders do reach this state of not really thinking about the tools once their organizations reach a point of, I wouldn't say competency, I would say semi mastery, you'll never have total mastery, but semi mastery over their data, when a business reaches that point, business leaders can experience this sense of flow, the state of flow when the data tools just fall away and it's all just the flow of the business. It's beautiful. But when you're the hands-on practitioner, [00:03:30] the person who's building the data models and reports, it's a lot less likely that you're going to reach a point where you forget you're doing it. So while wearing my data practitioner hat, I do believe this is the first time I've reached the point where one of my own creations has embedded itself into my life so deeply as to become semi invisible.

And this whole project isn't really for me, at least not directly, because we use it to help track and manage some of my wife Jocelyn's [00:04:00] health conditions. And it has evolved over time to be a fairly sophisticated operation. Every single day we track six distinct symptoms and those symptoms, interactions with four different medications and seven different environmental variables. Now right there, I'm sure a bunch of questions are springing into your head ranging from, "What the heck? Seriously, is that warranted?" To, "Well, what illness or illnesses does she have?" And maybe even to, " [00:04:30] Oh my gosh, that sounds serious, is she going to be okay?" Well I'm glad you asked. We'll take those one at a time. First, yes, seriously, a data model with that level of tracking is 100% warranted for us and it's useful, very useful in helping her manage her conditions. But now that I think about it, calling them her conditions, focusing on it that way, that's off.

It's putting too much focus on the problem and less on her, [00:05:00] which is subtly dehumanizing. So I think from now on, I'm just going to say manage our lives rather than manage symptoms or conditions or things like that. And if it's hard for you to imagine needing this sort of instrumentation in your life, well consider yourself lucky. I myself would've struggled to viscerally connect with such a story before I lived it as Jocelyn's partner. So I also am one of the lucky ones, lucky in that I've never had [00:05:30] health conditions that warranted this level of monitoring and analysis. Now, in terms of what these specific conditions are, I hope it's obvious that I secured Jocelyn's permission to talk about all of this before sitting down to record it, but I also told her that I wasn't going to make her specific health problems the focus of this episode.

This is after all the Raw Data podcast and not the Personal Tales of Woe podcast. So when it comes to the, "What does she have," question, I'll just say that she has a cluster [00:06:00] of autoimmune conditions that are probably genetic. Of course, once you have those conditions and your body is literally attacking itself in complex and poorly understood ways, you do tend to develop other problems as well and are left wondering if they're all related. So when I said six symptoms that we track, some of them are clearly autoimmune and some of them aren't. Now personally, I think they are all probably related. And in some future smarter world, [00:06:30] we'll eventually find out that it all traces back to one or two root causes, but in the meantime, we're left to deal with them at their ultra confusing face value selves.

Now for the last of those three questions, is she going to be okay? Yes, she is. But if we'd sat still and basically just followed the flow charts that local medical professionals had prescribed, and if we'd not taken an active role in managing all of this, things would be much, much worse, [00:07:00] a much lower quality of life for sure, and potentially life-threatening if we got unlucky. Instead, on the path we've taken, life is still significantly challenging, but it's vastly improved versus the depths we were at even a few years ago. Of course, it's fair to say that most of that improvement comes from our discovery of better treatments and there's nothing mystical here. Everything that we do to help her manage her life is all science-based, it's [00:07:30] all administered by people with MDs. But the fact that we had to go looking on our own in the marketplace to find these options, to find these other specialists is a little bit damning of our medical system.

And the additional fact that we have to pay out of pocket for all of this because it's all deemed experimental in some bureaucratic sense, even though these treatments are well established, we're not on the black market here, we're not doing anything crazy, that all just further suggests that our healthcare system is just a bit off. [00:08:00] Continuing that aside just for a moment, a personal opinion of mine is that we'd be much better off treating people like their infrastructure. Just like the USA's interstate highway system was crazy expensive to build and it's crazy expensive to maintain, on net, it pays for itself many times over probably every month or every six months. Now, no single entity gets to charge for access to it. There's no profit and loss statements [00:08:30] on the interstate highway system, but it does enhance the P and Ls, the profit and loss statements of basically every single company in the United States, while simultaneously creating more overall wealth for basically every citizen. That's a win. I think a healthy population should be viewed as a similar shared investment. Healthy people cost society a lot less than unhealthy in the long run.

And healthy people have this crazy little habit of producing [00:09:00] more value than they personally consume, which lifts all boats, creates more overall wealth and is honestly the sort of thing that even the most cold-blooded Machiavellian of all capitalists would agree to behind closed doors. Instead, what we have is a system chasing the lazy short-term profits and we're constantly stealing from everyone's futures in order to improve the short-term lives of a select few. If we operated the United States interstate highway [00:09:30] system with the same rabid focus on short-term profitability that we're running our healthcare system, well the whole country would suffer for that, badly. Just because people don't look like highways, don't look like bridges, don't look like infrastructure, doesn't mean that they aren't. Okay. Okay, coming back off the soapbox. Treatments are indeed the biggest piece of the healthcare puzzle here at Shea Collie, but we'd still just be wandering in the wilderness without the data.

This [00:10:00] is every bit as much, maybe even more so of a data problem as any business data problem I've discovered. For example, not all of the treatment options we've tried have worked, but we would've never known if we weren't tracking and analyzing the data. When your changes in symptoms play out over long periods of time and feature a lot of noisy fluctuation, even just knowing whether something is working or not is almost impossible unless you're capturing and analyzing [00:10:30] data over time. Another example, other treatments have worked but created side effects that need to be mitigated and minimized by adjusting dosage on an ongoing basis. And how are you going to do that without tracking data? And still other treatments have worked, but like in a first aid stabilize the patient sense, and then months later when we get that symptom under control, we want to start tapering the dose down eventually to zero because this medication's [00:11:00] no longer needed and we'd like to be rid of its side effects, but it turns out stopping too quickly triggers other negative side effects.

And we've even found that some medications need to be tapered by less than 1% of their total per day, which turns out to be very difficult because you might not know this, I didn't either until we discovered it, the variance in the weight of prescription pills is astoundingly high. Pills routinely weigh 5% more or 5% less than their overall [00:11:30] average weight for a potential 10% swing in dosage from day to day. When you think you're taking the same amount as yesterday, how is anyone, even someone with the data gene supposed to manage a 0.5% per day taper when the pills themselves are swinging 20 times that much while pretending to be the same? Well, if you're a data gener confronted with this problem, let me just tell you that you might find yourself buying a $400 [00:12:00] hyper precise scale that's so sensitive that it's digital readout, the numbers on the readout fluctuate, start giving you different numbers when someone walks through the room eight feet away from you, that level of precision. And then you use spreadsheets and razor blades to dial things in.

I have a spreadsheet called Sudoku because it's all about getting multiple pills to add up to a very precise pre-calculated total. And what are the razor blades for? Yeah, that's right. For very painstakingly [00:12:30] shaving little tiny pills until they weigh exactly the right number. And now for the data practitioners out there listening, just in case this isn't quite scratching your itch for on the ground data problems, let me give you a couple of my favorite examples from this. One of them is half lives. In the beginning we were tracking things like, "Oh, you took five milligrams of this medication that day and then you had a bad day of symptoms after doing that, but then on another day you took five milligrams and things were great." [00:13:00] That's all very confusing until you start to account for the fact that most medications have significant half lives, which means that the amount in your bloodstream only loosely correlates with how much dosage you just recently took because the doses you took on previous days are still fractionally hanging around.

So if you're looking for relationships between things like how much medication did I take and what impact did that have on my symptoms, which is of course [00:13:30] something that you want to do, you can't do it based on what you just took. You need to instead calculate the approximate blood level of the medication, that's total amount that's on board in your system, and compare that to the symptoms you're experiencing at that point in time. Now, your most recent dose is oftentimes the single biggest input to that blood level, but you need to run exponential math on basically every dose you've ever taken or at least the last 10 days [00:14:00] or two weeks of doses. And in DACS terms, again, I'm trying to scratch that data itch for you, in DACS terms, that ends up being a SUMX of every row in the doses table and the interior of the SUMX function call is an EXP.

That's the exponential function. And we're calculating inside that EXP function, a different exponent for each row, each dose, based on how old that dose is relative to the specific date and time that you're currently looking at. [00:14:30] Now, by the way, I do that aging in hours, not days, because we want to be able to check in on blood levels at different times of the day. Oh no, by the way, half lives, they don't print those on medication labels. You've got to go look them up yourself. At this point, the data gene community gives out a collective tsk, tsk, tsk. What kind a world do we live in where they don't put the half-lives on the medications people? But that brings me to something else that's cool, different flavors of time relationships. [00:15:00] We struggled a lot to get the data entry spreadsheets, things that we used to collect the data, we struggled mightily to get those into a format that was low friction, it took a long time just like iteration, trial and error, learning over time.

When you're super sleepy right before bed, no one is particularly keen on answering a survey of a bunch of questions like, "How is symptom X today on a scale of zero to three?" But if bedtime happens to be really the actual right time [00:15:30] to capture that, you better make the data entry process as painless as possible. And one of the consequences of that for us is that a lot of things end up getting entered in a single row, even though they happened at different times. Just take my word for it. This ended up being what happened. So some of the data points that we're entering in a single row are more closely related to yesterday and some of them are more closely related today. That was just the convenient way to capture them. And so just like that, when you have that, you've got the same sort of problem as things like [00:16:00] order date versus ship date.

In our case, a variable that happens at night and gets recorded at night might directly influence an immediate symptom like sleep quality that night, but then that same variable that we recorded, that same variable also influences another symptom the next day. So in DACS terms, yep, we've got different measures with inactive relationships and the good old used relationship function. Isn't that crazy? The same problem that appears in business [00:16:30] with things like order date and ship date or higher date and termination date or whatever, those problems show up even when you're measuring and analyzing incredibly personal single person hand entered health data. Yeah, so this model, in addition to those things, it's got seven tables in it. It's got some pretty sophisticated power query that takes those single rows from data entry. The data entry format is convenient for one thing, but it's not super convenient for the analysis problems. The power query [00:17:00] has to make a bunch of transformations. I bet you've heard about that before. And like most good data models like Power BI semantic models, it's still evolving.

I was making changes to it today and there's definitely more that could be done. It's probably time for some machine learning honestly, because spotting trends and multi-variable interrelationships between variables, the human eye can't reliably spot those. And sometimes when you spot them, it's actually a false trend as well. I'm pretty sure that's in machine learning, it's probably time for that. We've got enough of the history [00:17:30] here. And I'm also reasonably sure that there's a place for power apps in all of this, maybe even integrated with those physical flick buttons. You can just click when she's feeling something. I mean data entry at four o'clock in the morning when you're half asleep and you woke up and you weren't supposed to, that's pretty unreliable. So if you could just mash a button, that would probably help. That's probably more overkill and less ROI than the machine learning. I think the machine learning is probably next, believe it or not. Either way, stuff for the future.

In closing, [00:18:00] if you're left wondering whether this is truly helpful to us or if it's just me indulging a hobby, let me tell you, it is emphatically the former. It is truly helpful to us. There's a lot of discipline, a lot of human discipline required merely just to collect the data. 17 data points a day, every day, forever. If that was somehow fun, even for a day or two, it would wear off fast. Now we follow this data regimen because it works [00:18:30] and the handful of times we've abandoned it, we've gotten lost, confused, our confidence is plummeted in the courses of action we're taking. In those times, we have found ourselves not even knowing if our problem is too much versus too little of a single variable. We don't even directionally know which way to go, which is very much like what happens when you run a business without good instrumentation.

And the value for being data-driven in terms of your individual health is definitely not [00:19:00] limited to just Jocelyn's life. A few months ago we were at an appointment with her physician who specializes in functional medicine, and we got to talking with him about our data regimen and he got really excited. I could see in his eyes that he immediately got it and he even said something like, "Yeah, that's the missing piece for us." He then explained why it would be valuable for his patients and it was really just a replay of all the ways in which it's been valuable for us, and this isn't some small practice that he runs. The waiting list for new patients [00:19:30] is usually about a year, I think, and the lead time for getting an appointment is often like four weeks or more. So reading between the lines, that's a lot of patients and it was really cool to have our data practices so enthusiastically endorsed by someone in his position.

Now along those lines, I have thought about making our spreadsheets and data models available maybe on internet forums or whatever, but you'd have to know Power BI in order to operate them confidently. There's just really no way I could document these things sufficiently for the non- [00:20:00] data gene crowd. And since everyone's health and life situation is completely unique, the analytics would need to be customized every single time for every individual who is using them. And you also obviously can't customize them if you don't understand power BI and how it works. So the whole thing, the idea of sharing this stuff is a non-starter, but when I was talking to that doctor that day, I started to see it. I started to imagine myself working in his practice as his data assistant interacting one-on-one with individual [00:20:30] patients and crafting customized data regimens for each one.

And it was just thrilling to think about. Someday when I retire, maybe that's something that I would look into. Coincidentally, for the first time since that appointment, we were back in that same practice this morning and we were talking to one of the nurses there. She had a fascinating background. She had been a medical research scientist for most of her career. And then in midlife, she retrained as a nurse, which impressed the hell out of us. The regimented, semi-linear discipline [00:21:00] required as a research scientist, and then to step into the real world chaos of nursing, it takes a real interesting and unique individual who can do both of those things, even if they're not doing them at the same time. To do them in one life is a big deal. She also reflected that she always vaguely knew that when she was working in research that her work was out there helping patients at some great distance, but now in nursing, the emotional satisfaction of directly helping people today is just so much more rewarding [00:21:30] for her. And Jocelyn and I got really excited by that because it directly parallels our own lives.

When we were software engineers at Microsoft, our work was out there helping people out in the world at great distance, but you really couldn't see it most of the time. And so even though when we left Microsoft, we had to give up the prestige of working at the mothership, but in our career's second act at P3, directly applying that same software, those same sorts of tools, to help others with their businesses has similarly felt [00:22:00] much, much better. Now, you might expect that helping people with their health feels a hundred times better than helping people with their business and that these things don't even really belong on the same page. But from my experience, the good feeling is in both cases, helping people with their health, helping people with their business, they're both powered by the same biological mechanisms instilled in us. We're a collaborative species and we're wired to use our talents to help others.

That wiring, that feeling is what drove [00:22:30] us to start this business, and it's what drives our consultants to do what they do every day. No doubt about it, I would love to someday retire and do the data thing for patients at that practice or a similar one. What a privilege that would be. The other big takeaway from all of this is to remind us of just how early we still are in the uptake of data-driven methodology into all of the places where it's going to be useful. Here's the visual metaphor I'd like you to consider. Imagine this absolutely massive sponge, the size of [00:23:00] a football stadium, and that sponge represents all the places where analytics can help in the world. So every single little cubic centimeter of that sponge will happily devour water, aka the power of analytics, if the water can somehow get to it.

But for decades, there's just been this little eyedropper going, clink, clink, clink, clink, just putting little drops of water here and there on the top surface of that sponge. And now today, [00:23:30] instead of the eyedropper, we've switched to fire hoses and water is now getting to all kinds of places in that sponge that had never even seen water before. But even fire hoses are actually still slow relative to the size of this massive sponge representing the entirety of human activity. We are still early. It's easy to forget just how out of reach all of this was just even a few years ago. The older data [00:24:00] and analytics tools were bad and they were bad for a long time. They were so expensive and slow that only the largest organizations could even dream of using them, and then even those organizations didn't get good ROI at all out of their efforts.

Contrast that with today. I, as a private citizen, now sit at home with essentially free software whose capabilities far exceed the enterprise tools of 10 years [00:24:30] ago. And these new tools are also nimble enough in terms of time investment that I, again, as an individual without a team behind me, can provide my family with life-changing ROI. That's a massive change in capability and more importantly in accessibility. So the moral of the story here isn't just that you can use these tools to help manage personal health conditions. It also serves as a reminder to us that we're still [00:25:00] very early in the adoption of these new capabilities throughout the world, even in the business world, especially in the business world actually. And one last thing before I go, Power BI connected to Excel Online is a dream.

Being able to update those 17 data points from any device without any application coding, I didn't have to build the power app, I didn't have to build another app, just being able to use Excel Online [00:25:30] from any device to update that data is the difference. Trust me, in the human plane, it has been the difference between us being able to keep up with it versus falling behind and abandoning it. So without that always synced up, live multi-device nature of Excel Online, we would not have been able to keep up with just even entering the data. It's beautiful. And then every night I press refresh on Power BI Desktop, I don't even bother with [00:26:00] schedule refresh and Power BI spits out guidance on how much of each medication is probably optimal for her before we go to bed. The whole thing is just magical and it simply wouldn't have been possible even for us, even for me, let's say 10 years ago, honestly, maybe even five years ago, which makes it all the more criminal, in my opinion, that it is still so difficult to connect Power BI to Excel Online.

Why do I need to keep remembering the secret handshake of what the right [00:26:30] URL is? The one that power BI understands, not the one that Excel Online wants to give me. In order to make this whole thing work, I have to do this dance. I've now saved a couple of examples in Power BI that I know work, and I always go back and look at those and I say, "Okay, I need to find the URL that looks like that," and I can generally hand edit it to make it work for whatever new workbook I've got. But without those reference workbooks, I would be just lost, again, and I wouldn't even be able to use Excel Online. Excel Online, or whatever you want to call it, Excel [00:27:00] and SharePoint, Excel and OneDrive, that should be a top-level data connection choice in Power BI. And then when I click that choice, it should be able to help me find the right URL for the right workbook. Or better yet, just accept the URL, one of the many URLs that Excel Online wants to give me, and then behind the scenes, translate that into the right one for me. That seems really valuable, really important, and long overdue. You never know, someone's health [00:27:30] might depend on it.