[MUSIC] Hello. I'm Nicholas Genes, and this lecture will focus on big data and how it relates to health IT in general, and mHealth in particular. You might have heard a lot about big data recently, you might be surprised to learn as Gary King, a Harvard computer scientist says, big data is not about the data. What does he mean when he says this? Maybe Peter Sondegaard, a consultant at Gartner gets a little closer to the truth when he says information is the oil of the 21st century and analytics is the combustion engine, meaning that information is kind of a resource, but it's really nothing without analysis. I'll spell it out a little more explicitly. I think data is the easy part of mHealth, generating data and collecting data is the easy part of mHealth, and really won't reduce costs or improve outcomes unless we're very careful about the implementation. And I think today already there's a tendency in our politicians, in some of our consumers and even administrators in health IT, that once patients start putting on their smart watches, or their fitness trackers, or start tracking their nutrition better, the data that's generated will automatically translate to better care, it's not quite the case. What is big data, to begin with? Really, the definition is just anytime you're working with a data set that's larger than your computer memory. So if most of us have four gigs or eight gigs on our phone but we have a data file that's 50 megabytes or multiple gigabytes or something, then it's not going to fit in the working memory. Special techniques have to be used to analyze that data, to reduce the set into components that the computer can work with. And this is very convenient for computer scientists, but it drives researchers crazy because it requires a whole new set of tools for really quantifying the validity of the data and how much we can actually infer from it. In general it's said that when we're working with big data, we are sacrificing microscopic quality for macroscopic insights and understanding trends and relationships. Although a lot of garbage, a lot of bad data is getting in and maybe compounding the results, the new tools that we're using to analyze the data should go also over that and should let us make some bold insights. MHealth however tries to have it both ways. They’re try to capture individualized details from specific patients at a very high quality, but also feedback specific recommendations and insights, and this is very difficult, and as far as I know has not been successfully accomplished yet. What are some success stories in big data? Well, you don't hear that much about them, but they're happening kind of under our feet with or without our knowledge. Airlines and retailers have been using big data for the last few years to kind of suddenly adjust prices. You might have seen some examples where people with one laptop versus another laptop, one user with one set of demographic information versus another user maybe with a different credit card and different purchase history, they both log on to the same travel website or the same retailer like Macy's or Walmart, and they start browsing the catalog, they are actually presented with different items and sometimes different prices on those items. So that's big data in the background trying to infer what the customer wants, and also adjusting the price so that the customer might be more likely to purchase, or more likely the retailer is trying to extract more money from the users that it feels can afford to spend more money. I also saw an interesting study recently about a customer service firm that was using big data to actually quantify personality types and try to understand why people were calling in to a call center. Was it a general question, was it an angry complaint, was it some other issue? It was really quantifying those callers, and also, it had done the homework, and knew the people that were going to be answering the calls, and what their skill sets were, and what they tended to do best at, and was actually matching the callers with the right customer service agents. And they were able to show that they really increased customer satisfaction, it was more efficient, the customer service agents liked their job better because they were working with the people that they clicked with the most, and of the customers, even if they were angry, they met up with someone that was best equipped to handle that anger, and they did well. So, kind of a human face to big data if you will. How does big data look in healthcare? Well, so far scientists have been using it to uncover different subgroups, different kinds of diseases based on complex factors like genetics or the environment. A colleague of mine here at Mount Sinai named Joel Dudley has done work with type 2 diabetes and genetics to actually uncover what we call type 2 diabetes might actually be three sub-types that are closely related, and they all involve imbalances in blood sugar. However, based on what sub-type you're in, based on your genetics, you might be prone to specific complications, like one group is more likely to get kidney disease, another group might be more likely to get heart disease or eye problems, and another group might be a better candidate to respond to certain medications. So this kind of analysis is very labor intensive, very computer intensive and would not have been possible in an earlier era without big data techniques to kind of crunch all these numbers and come up with these insights. And yet, for mHealth, for consumer technology, for the data that we're already generating, we don't have any big data solutions yet. I've borrowed this slide before. It's from a company, a now defunct called PulseBeat, but they were producing kind of a snapshot of a subscriber and delivering it to stakeholders, usually family members, kind of showing people what data was being generated based on fitness tracking and based on medication used and blood sugar, glucometer data, W-iFi scales, et cetera. This slide is great because it's all good news. When the data is not so good, when the step counts are deteriorating, when the weight is creeping up, when the sleep is interrupted, it's harder to know what to do, or how to intervene, or if it's a trip to the emergency department, or a phone call to a nurse, or something in between. So, big data is hopefully going to be able to answer these questions, but it's really not there yet. And yet we continue to put on more wearables and bring more of these devices into our homes to generate more healthcare data, and yet we're not quite clear what to do with it yet. Really, we're still at the bottom of this pyramid, where we are generating data. We got signals, but we don't really know what they mean. The next step above that pyramid is information. It's useful, it's structured, we're more sure of its validity. Beyond information comes knowledge, that's synthesized and contextual. And then finally at the top of the pyramid is wisdom, where we know what to do with it, where we can act on it when we are confident. In a case of a patient with CHF, whose weight has been creeping up on the Wi-Fi scale We know the data is their weight. Maybe information is a little bit of organization, where we know that their weight has been trending upward and it's over a holiday weekend, and maybe they had some salty foods. That's what the knowledge would represent and then the wisdom would be we've got to dose some diuretic to lose this weight. It's water weight and it's gotta come off, otherwise the patient is heading to a CHF exacerbation and they're going to be heading to the hospital. So that's really a pyramid of knowledge. And I think we're spending a lot of our time still at the base of the pyramid, we've got to get up to the top. So much of population management though that hospitals are charged with doing is still based on data that's collected in the electronic health record. This is episodic data that's captured when you visit the doctor or when you're hospitalized. It's vital signs, it's lab results, and it's doctor's notes. Sometimes there's some natural language processing that happens with those doctor's notes. But it doesn't really rely on wearables and fitness trackers, etc., that those string of data had not really been integrated into population health models. And yet we know that those things are really taken off. They're really popular, people like using them, they're collecting data. mHealth apps let you communicate with your healthcare team, remind you to take your medications or adhere to whatever regimens you are supposed to be on. So we know an mHealth has potential. And we know that the adoption is really taking off, and there are more and more FDA approved apps. And a lot of doctors think that the use of apps can really help reduce visits, the use of apps and wearables. And we're seeing millions and millions of these wearables sold. Hospitals are also getting on board and helping to distribute these apps or wearables in some cases. And it makes sense from their perspective because they are taking on more of a role of an insurer. They're interested in tracking adherence and disease progression just for estimating cost as well as for the noble goal of trying to keep patients healthier and keep them out of the hospital. I've heard rumors and I've seen some evidence that some insurers are actually going beyond fitness trackers, beyond wearable and apps and actually looking at social media postings, web browsing habits, emails, you can actually learn a lot from the language that we use in emails, and even purchasing habits. And I remember one friend who is crunching data at an insurance company said that you found an independent risk factor for high healthcare utilization was on payday. On Thursday or Friday if a user spent a lot of money on alcohol, that was going to be a risk factor. Maybe not the most surprising inside but something that was collected through purchase history, not at all through wearables or electronic health record data. And yet it just goes to show where the insurance companies are going to kind of estimate risk. When an insurance company looks like a population it also looks like a pyramid. And at the bottom of the pyramid are the motivated healthy, or the worried well, depending on your perspective, and these are generally adults without chronic conditions. Maybe they want to be a little more fit, and maybe want to run a little farther, or do a little better, but they make up a big fraction of the pyramid. Above them is a group that has some biomarkers for disease. Maybe they have high blood pressure, maybe they're trending towards diabetes or they're starting to take insulin, but they don't have any of a chronic disease, they'e not actually sick at that point. They're trying to improve their biomarkers so that they don't get sick so that they don't have heart disease that comes with hypertension. Or they don't have the kidney problems or eye problems that come with diabetes, but they're in the middle of the pyramid. And then at the top of the pyramid are the high utilizers. These are people that have the chronic disease that getting amputations from the diabetes, they're in and out the hospital because of their COPD or their CHF. And they tend to consume the most healthcare resources in this country. Insurance companies or health systems look at their population, they place individuals into different parts of the pyramid and they try to come up with offerings for different segments of the pyramid. For the worried well or the motivated healthy maybe a fitness tracker would help keep people active longer so that they don't progress to the middle of the pyramid where they're pre-diabetic or they're hypertensive. For that patient population, maybe the best things to offer is a nutrition tracker something to help or a coach to help motivate healthy eating. And then for the high end of the pyramid, the high utilizers, it's important to track vital signs and other metrics, and to facilitate easy communication with health care providers to keep these patients at home where they want to be, and not in the hospital dealing with exacerbations of chronic illness. The pyramid when I first saw it I was a little disturbed, I fell it's kind of of a cynical look at patient care and population management because it's really about predicting expenses. And it's really about keeping cost down and giving tools to people so that they are not more expensive. And it got me a little depressed, but another way of looking at it is that it's an optimistic view because it suggests that change is possible through technology. And then if we offer the right tools to patients, to whole populations, we can keep them healthy and keep them doing the things that they want to do. There are some assumptions that are baked into the pyramid, however. One is that the patient is generating accurate data and that data accurately can predict costs. Two, that sicker patients are going to want to contribute more data which is assumption that sometimes is not true. And three, that the healthcare system can accurately interpret the data that's coming in and make good decisions on it to prevent people from climbing up the pyramid. All of these things are assumptions and we are studying them, and others are studying them. I tend to think, and we've covered already, the quality of patient-generated health data. I think in general it's very good and it's getting better all the time. I don't think that's going to be a huge issue. The fact that sicker patients currently don't want to contribute much data, that I think will also change with time as a new generation gets more comfortable with these devices and these tracking tools. And as the devices improve make it easier to contribute data. But what really worries me about the pyramid and the assumptions, is whether healthcare institutions and insurers will be able to properly interpret the data that's coming in. So far there's plenty of evidence that they can't, and there's even some evidence that patients are misinterpreting the data. I see this frequently in the emergency department, a lot of my patients come in because they are checking their blood pressure at home, they have a sphygmomanometer. Gosh, that is always hard for me to say. But a blood pressure cuff, and they're checking at home. And this is not a digital blood pressure cuff, it's a traditional, old-fashioned blood pressure cuff. And they check their blood pressure, maybe after they had some coffein, maybe after they climbed some stairs or came in from outside. The number is high, they don't like what they're seeing and they come to the emergency department. And then I give them a conversation about how blood pressure is a long-term risk factor and maybe their medication should be adjusted in a comfort of a clinic or after a phone call with the primary care doctor, someone who knows the patient. But the truth is they are coming in through emergency department, they are often getting their vitals signs checked and their getting an EKG done at triage. And healthcare dollars are being generated just from them walking through the door, even if I don't order any blood test or spent any medications from the That's still going to be an expensive Visit that needs to get reimbursed. So, I wonder sometimes if the blood pressure curve checking blood pressure at home is actually reducing cost or leading to better outcomes. I'm reminded of a famous quote from Charles Babbage that errors using inadequate data are much less than using no data at all. I would say for the digital age, for the mHealth age, sometime inadequate interpreting data, is sometimes much more expensive than using no data at all. So, if the cost-effectiveness of home blood pressure monitoring in my cohort or patients in New York and in the emergency department. If that's not clear, then what's the benefit of mHealth over time? We have some studies to help guide us, for COPD, for CHF. More studies are coming out all the time. Two that I've latched onto showed, actually, some mixed results. The first I'll talk about is for COPD which is a huge cause of death in the US and a huge expense, and this is a chronic obstructive pulmonary disease, it's like emphysema. Emphysema is part of COPD. And we know from high quality studies in the past that self management can reduce risk of hospitalization, can prevent exacerbations. So that these self management tools are actually they involve a lot of coaching, a lot of efforts, a lot of education and a lot of, healthcare systems can't probably implement them at scale. What's hopeful is that maybe with digital tools, mHealth tools and apps and communications, we can kind of reproduce that self-management at scale through the new technology. And yet when Pinnock and all studied this very concept, they monitor patients at home using wearables and apps and pulse oximeters and easy communication with care providers, they found no difference in exacerbations or admission rates. No difference in quality of life over the one-year period that they looked at. And there's actually other randomized trials that show the same thing. When we turn to CHF, another huge expense in the US 5 to 10% of all hospital admissions are actually related to CHF congestive heart failure exacerbations. The readmit risk is high. These patients are discharged from the hospital, they go home they're back within a week or two. And the hope was that technology could help this by a lot of tele-monitoring and collecting data such as weight impedance, medication adherence, et cetera. And they found no real difference in readmissions in the BEAT-HF trial. Slightly improved quality of life but no difference in mortality. And this is similar to prior studies as well. So what I'm struck with both of these papers is going back to that first pyramid I talked about, these are papers that are great at collecting data from patients, they're great at putting that information in some context, but there is almost nothing in these studies about how the data was acted upon. What were the specific interventions that were used? When was the appropriate time to intervene when the data started trending towards an exacerbation? The papers are really silent on it and I have to conclude that the investigators were kind of winging it. I think they interpreted each case individually and they said, my treatment plan for Mr. Smith is going to be different than my treatment plan for Mrs. Jones. And without a really standardized approach, they were collecting data but they weren't acting wisely on it. And that might be one of the reasons that these studies fail to show real concrete improvements for patients, for keeping cost down, or improving care. So I'm not sure we're monitoring the right data. And even if we are, I'm not sure we're applying the right intervention. I think we're going to need to collect data for years. For each chronic disease, we're going to have to generate a lot of data on the sickest patients with a lot of wearables and a lot of checkings and a lot of adherence monitoring. Only then will we be able to identify trends and identify the right threshold for when to intervene. Another significant risk in all this, is that sometimes the patients get disenfranchised. They're sending so much data and they're getting so much feedback that they feel like they're not really owning their own care. Rather they're just a vessel where other caregivers are giving them really too much feedback for their own good. So that is another potential. We have to find the right threshold intervene and then when we do chose to intervene, it has to be the right medication, the right dose et cetera. And these interventions will probably have to be tailored to the individual. But that just means we have to collect and order a magnitude of more data to figure out what is right for each individual. Only then will we start to see, I think, improved outcomes and lower costs. So my take-home message really is don't seek data, seek wisdom. Insist that patient-generated health data when it comes that it be packaged in actionable information that is useful to clinicians and to patients. And until then I'm going to be pretty skeptical of wearables, and fitness trackers, and mHealth apps and devices. And I'll be a little skeptical of big data too. Until then, thank you. Nick Genes. [MUSIC]
