Module 4 - Networks

Our lives are inundated with social media - statuses, tweets, connections, pins and the list goes on. In previous modules we looked at how to mine that data to make predictions for every industry - from healthcare to retail. This last module switched the perspective to look at the relationships between all of that data. Social networks line up well with graph theory where the entities are vertices and the relationships between them are edges. This could be the relationship between people on Facebook or LinkedIn, connections between shares on Facebook, pins on Pinterest or influencers on Salesforce. Or the cross-pollination between multiple networks like Instagram photos being posted on Facebook. All of this gives us insight into real time trends, relationships between people, causes and ideas dissemination and growth. What appeals to me most is how networks give us meaningful ways to understand and model large and complex systems and it's all based on math. It's accessible and compelling but it's not guess work - there's meaningful analysis behind it. This module focused on types of networks, vertices, and edges and how to understand them based on network properties such as the structure, centrality measures, density, clustering coefficients, cliques and others.

It's no surprise - the first thing I did with this module was look for examples of network analysis impacting or being used in healthcare. I found an example of ER doctors being able to prep better for the victims after the Boston Marathon Bombing than if they'd had to rely on a traditional news report (Neuhauser) and how Mayo Clinic is jumping into social media with both feet, so to speak, to drive innovation (Pennic). Is any of this data being combined together to look at healthcare from a viewpoint of perspective? I found the Institute for Health Metrics and Evaluation (IHME) which had some great data visualizations: http://www.healthdata.org/results/data-visualizations. Still, we're not looking at relationships between entities. I can find countless explanations of how social media is a disruptive influence in healthcare and is going to bring big changes (Cerrato, Honigman) but there is no one network which embodies healthcare. There's no Twitter of healthcare - instead we have FollowMyHealth, MedSeek, HealthVault and countless others. Then if we factor in the fitness devices and the apps that come with them and there's endless sources of disparate data. This speaks to a fundamental flaw in healthcare - the lack of standardization means it's difficult to impossible to leverage the data to find trends and use to make decisions.

I did, after some digging, find the Healthcare Hashtag Project ("Why the Healthcare Hashtag Project?") which seeks to use Twitter hashtags to

"Discover where the healthcare conversations are taking place

Discover who to follow within your specialty or disease

Discover what healthcare topics are trending in real-time."

I love this idea! Physicians, labs, and research organizations can access that data and use it. The next step in the evolution needs to be a tool that can be easily used by the interested parties like the IHME.

Citations

Cerrato, Paul. "Will Social Media Revolutionize Healthcare?" InformationWeek.com. 19 Sept. 2014. Web. 10 Mar. 2015.

Honigman, Brian. "24 Outstanding Statistics & Figures on How Social Media has Impacted the Health Care Industry." Referral MD. Sept. 2013. Web. 11 Mar. 2015.

Neuhauser, Alan. "Health Care Harnesses Social Media." USNews.com. 5 Jun. 2014. Web. 10 Mar. 2015.

Pennic, Jasmine. "5 Reasons Why Mayo Clinic Dominates Social Media in Healthcare." HITConsultant.net. 17 Feb. 2014. Web. 11 Mar. 2015

"Why the Healthcare Hashtag Project?” Symplur.com. Web. 10 Mar. 2015.

Module Web - Web Analytics

Module 3 is all about web analytics. I was very excited for this module because of a special cross-functional team which had me working with the marketing team. During that time they rolled out a new website that would let them track all sorts of metrics to track our brand awareness. Brand awareness was a new concept to me and they weren't technical at all so I always wondered how it could work and if it did work. Enter my introduction to web analytics.

The definition from lecture 9, from the Web Analytics Association, defines web analytics as "The measurement, collection, analysis, and reporting of internet data for the purposes of understanding and optimizing web usage." That's a pretty broad charge - what you want to understand or optimize far is going to depend on if you're a University recruiting students and staff, an opensource project sharing your results, a small business selling your wares, or a large business selling your wares. That is something that jumped out to me as I worked on homework 3. There's so much data and without a purpose or questions you need answered it is kind of meaningless. On the same note, not asking those questions and having a clear, definitive purpose for your website is a waste because there is a treasure trove of data available to improve the results you're getting. All of that is backed by the Web Analytics Cycle which starts with setting goals, measuring metrics related to the goal, reporting on the metrics, analyzing the metrics and finally optimizing based on those metrics. How do you set the goal? You need to start with the Five W's of Web Analytics - what, who, when, where and why. As you delve into the user base for your website you can start to understand their demographics by where they're coming from, type of device used, age, gender and so forth. You'll be able to see where traffic comes from to track success of an ad campaign or to see what areas to focus on. There were more metrics out of the box than I could have thought of mattering but as I tried to understand my website for the homework I wished I had more metrics! Only the bare bones had been set up because the business wasn't really using the metrics to their full advantage.

Half of the module focused on Google Analytics. We learned how it integrates with websites and gives a simple, free tool to wade through all of the metrics and questions that we have. Much like with dashboards from module 2, Google Analytics makes the metrics extremely consumable with visual reports that can be exported or converted into a grid if you want just poor numbers. Exploring the data was easy and it was easy to customize the views to focus on just one specific areas - like comparing new vs returning sessions while at the same time comparing that to new vs returning sessions of people who made purchases.

Catherine Brock in "What Web Analytics Says About Your Brand" brought up some really interesting points about using metrics to tell the story of your brand. She talks about rationalizing metrics and the need to really confirm it to make sure you understand the story your metrics are telling. She talks about what a website with the majority of session times less than 10 seconds means and how most website owners will defend this. The defense will come with what they believe is the cause - such as they were just looking for a phone number and got it - but this is a guess with no data behind it. There are tools can be used, in this case a tracking phone number, to tell the story more clearly. Her thoughts spoke to me because we see online all the time how statistics are misconstrued or understood.

Web analytics are powerful tools but must be combined with clear business goals and a commitment to reviewing and acting upon the data.

Citations

Brock, Catherine. 'What Web Analytics Says About Your Brand'. Business 2 Community  15 Feb. 2015. Web. 18 Feb. 2015.

Module 2 Reivew

Module two covered a lot of ground - all focused on tracking performance, understanding and making decisions within an organization. The data used was all internal to an organization. Governing these action is the Date Warehouse Design Cycle which is used to translate the Online Transaction Processing (OLTP) systems to the Online Analytical Processing (OLAP) system(s). This process involves the extraction, transformation and loading of the data from disparate source systems - like an ERP or CRM - into a single data warehouse. This allows for the organization to agree upon definitions of terms and for the data quality to be checked.  Before the data warehouse can be populated it has be to be designed which we learned about the star schema to design. A star schema is made up of a fact table with dimension tables. The facts represent business measures and the dimensions represent the grain.

Creating a star schema should reflect the business events the company wants to track and make decisions about. This is done using a four-step dimensional modeling process. The steps are (1) selecting the business process to model, (2) declare the grain of the business process, (3) choose dimensions that apply to each fact row, and (4) identify the number facts.

I've spent my professional career with OLTP type databases. The data is normalized as much as possible with minimal duplication of data. As we began reading about and designing star schemas it felt wrong - there were more efficient ways of doing it! In the class readings, Ralph Kimball and Margy Ross mention several times that it's a natural instinct to normalize but there's not enough savings to merit the extra effort it would cause users. As I read more and focused on the purpose of this database - not real time for the purpose of a transaction in the moment but for analysis for trends, direction, and performance it started to change my perspective on the need for data warehouses. Data warehouses don't mean the data is significantly lagging behind - it can be updated in any cadence but it's not the production system - like a point of sales system. Data warehouses aren't meant for the programmers but for the end user. The end user knows how to run reports but the underlying structure is optimized for those reports.

Let's be honest though - I'm still more comfortable with a normalized, OLTP database but this module really helped me see the importance of the data warehouse.

The beginning of the module focused on the balanced scorecard which really took me until the end of the module to understand why. Interestingly, the day before I listened to the lecture on the balanced scorecard my company introduced one for the first time. It was replacing a program that had been in use for 3 years to help us make commitments and ensure our execution was aligned with the strategy. No one ever felt that program was effective. As a Systems Integration Engineer I can tell you how I directly impact revenue and profitability. Unfortunately, my impact is often delayed - I will have an impact in a year or two - my work rarely impacts the current year. Instead of focusing on projects that had strategic value in the medium to long term we were asked to do what felt like busy work, in addition to our day jobs, to show impact that year. We were promised something new this year that would better track our progress. So the scorecard was unveiled. None of us had any idea what it meant. As this article from the Harvard Business School "Why Strategy Execution Unravels - and What to Do About it" indicates though - alignment doesn't always mean success. The balanced scorecard has the organization focus on all the areas indicating success. Now, just having a scorecard won't bring you success but it's a good way to track measurement. If you don't seem improvement or sufficient improvement it can be an early warning system some part of the execution plan is off. Which leads back to big data and data warehouses and how they help answer strategic questions.

An organization amasses so much data on a daily basis and it’s a gold mine for analysis but data is only as useful as its quality. You must be able to access the data, terms must be consistent, and the data quality has to be good. If all of this is in place, which the ETL process is supposed to make happen, now the executives have a treasure trove of data to draw from to keep track of company progress and to make course corrections along the way. Without good data the scorecard stops being an accurate picture of company success. Without the scorecard, or other strategic requirements for data, the data warehouse has little merit. All of these efforts are to support and help organizations receive a competitive edge.

Citations

Sull, Donald, Rebecca Homkes, and Charles Sull. 'Why Strategy Execution Unravels - And What To Do About It'. Harvard Business Review 2015: Web. 14 Feb. 2015.

Module 1 Musings

Big data is the next evolutionary step for statistical analysis and predictive models. It is made possible by the large amount of data made available by the internet of things. This data can be used to “answer what, not why, and often that’s good enough” (The Rise of Big Data, 2). This is the future of business in healthcare, humanitarian efforts, civic planning, and all industries. There is so much data now that there are new job roles being created to manage all this data - the advent of the data scientist is bringing together diverse skills like programming, statistics and specific business understanding, among others, to be able to use the copious piles of data – much of it not in a usable state yet. Big data is all about handling its sheer volume, velocity of creation and variety of formats that are stored.

After completing the class reading and googling for “healthcare and big data” where I found titles like “NHS director argues big data in healthcare is a ‘moral obligation’”, “Why Health Care May Finally be Ready for Big Data” and “How Big Data Will Help Save Healthcare” I was left with the sense that big data is seen as being able to do no wrong – it will save us all! Most of the articles in in “Business Report on Big Data Gets Personal” focuses on the advantages of sharing personal data, how said data is bought and sold today, and how, well, you could identify people from the data from multiple data points but no one ever would. Those assumptions feel a little hollow after a year of many big companies – Home Depot, Chase, JP Morgan, and Sony, just to name a few, have suffered security breaches. Healthcare data is extremely popular on the black market (Monegain) and to effectively utilize big data there must be more thorough sharing of data between healthcare organizations. Security breaches are nearly inevitable and privacy concerns must be addressed.

Taking it in another direction, let’s assume there are no nefarious agendas, the use of big data could still lead to negative unintended consequences. For example, there is concern in the healthcare industry that predictive data could lead to health insurance agencies denying coverage for conditions that do not exist, may never exist, but a predictive model suggests could exist. Predictive models are just that – predictive and not infallible. We that in 2009, when Google used big data to find a correlation between internet searches between 2003 and 2008 and flu outbreaks. The idea being that if they could find a predictive model flu outbreaks could be analyzed in real-time. This mathematical analysis did the trick in that they found 45 search terms which often coincided with an outbreak. With a billion searches a day they wouldn’t have been able to predict these terms but by using such a large sample they felt confident in the results. The study concluded, though, with “[…] it seems that Google’s system may have overestimated the number of flu cases in the United States. This serves as reminder that predictions are only probabilities and are not always correct […]” (Cuckler, 3-4).

There are other limitations that must be conquered. Some of the industries that could benefit the most from utilizing big data are ill equipped to use the existing data. For example, in healthcare, from the paper patient records, electronic health records, lab results and all of the other data there is endless data that could be used to help drive diagnostic decisions. In an NPR interview with Amy Standen and Jenny Frankovich, they relate a story about a girl with lupus who might have been at risk for a blood clot. They faced the dilemma of proactively administering an anti-coagulant which might cause more problems or waiting for a potential blood clot to form before administering the anti-coagulant. One of the physicians used their electronic database with patient records to pull up the number of lupus patients and the number that had a clot. The course of action was made based on those results and was considered successful. So is this method in use by the hospital today? No. Why? Standen says "The system just isn't ready, the hospital decided. What if Frankovich had used the wrong search terms or the engine itself had bugs? What if the records has been mis-transcribed? Even Frankovich agrees that it's just too risky." The potential is there but this is an industry where the uncertainty, and messy data is a problem. Even if we assume the records are correct there is little to no standardization across organizations which means the results could be skewed if the data samples in each organization were insufficient. For this kind of clinical decision support healthcare really need the advantages of “N=All” sample sizes.

I do not want to come off as being anti-big data. In fact, I think the potential for good is limitless and has the potential to transform and how we interact with each other and other nations. In healthcare we already see Accountable Care Organizations (ACOs) are using big data to find trends to reduce the need for patient readmits and inefficient practices which will reduce healthcare costs (Burg). Another great example is from Canadian researchers who are using big-data to identify possible problems with premature babies long before the problem is evident. “By converting 16 vital signs, including heartbeat, blood pressure, respiration, and blood-oxygen levels, into an information flow of more than 1,000 data points per second, they have been able to find correlations between very minor changes and more serious problems” (The Rise of Big Data, 3). Even if they are not right 100% of the time those premature babies will get extra care and attention so the ones who are at risk can be saved at higher rates.

One of my favorite articles from this week was “Big Data from Cheap Phones” (Business Report - Big Data Gets Personal, 17-21). In developing countries where there is not existing government infrastructure, like a census, they are starting to use simple cell phone data points to under more about the nation and deal with problems. That article focuses on malaria outbreaks and how with just the data points of cell phones moving between cell towers they are able to track the outbreak and its flow. These seemingly simple data points will give them the information to focus their limited resources on the areas that will have the most impost. This is similar to the story in the “Rise of Big Data” about how Manhattan is using data analysis to deal fire-prevention in overcrowded buildings. By studying previous cases they found data points, which weren’t immediately obvious, suggested which buildings were more at risk. Their inspectors could focus their time on those buildings and have more impact (Cuckler, 4-5).

I’ll admit, these examples are much more motivating to me than how businesses can increase the impact of their marketing. Big data is here to stay – how can it not with the massive amount of data generated daily – and while the use and sharing needs to be controlled I look forward to seeing the improvement it will bring to lives across the globe.

Citations

Burg, Natalie. 'How Big Data Will Help Save Healthcare'. Forbes.com. 10 Nov. 2014: Web. 23 Jan. 2015.

Business Report - Big Data Gets Personal. MIT Technology Review. PDF. 20 Jan. 2015.

Cuckler, Kenneth Neil, and Viktor Mayer-Schoenberger. The Rise Of Big Data. ForeignAffairs.com. May/June 2013. PDF. 20 Jan. 2015.

Monegain, Bernie. "6 steps to keep security issues at bay." Healthcare IT News. 25 Apr. 2013. Web. 23

Jan. 2015.

Standen, Amy, and Jenny Frankovich. Big Data Not A Cure-All In Medicine. 5 Jan. 2015. Radio. 24 Jan. 2015.