Critical Data Right Before Your Eyes. How Dashboards Impact on Healthcare Resilience

Imagine that you are an ambulance dispatcher. You receive a call reporting a severe accident that involved several buses full of people. And you have to make a decision – how to distribute 100+ injured people among hospitals. To do that, the dispatcher must have relevant information, such as hospital occupancy rate, available resuscitation beds, and number of doctors on the shift. Obviously, required data must be at hand, because human lives may depend on the decision-making speed. The dispatcher shouldn’t waste precious time opening numerous Excel reports and looking for necessary info – it must be right before their eyes. This is why dashboards are indispensable – they display the required information in graphs and charts, make data easily perceived by a human eye, and help to make decisions quickly and efficiently. 

Obviously, dashboards can represent information at different levels – be it a dashboard for a Ministry of Healthcare, a city Hospital, or a small private medical center. Depending on needs, they can be of different types – one-pagers with a minimum set of metrics and multi-page complex dashboards with the top-down principle. The second option means that the main page of a dashboard displays high-level data, and you can check details for each parameter by clicking on it. Below let’s see an example of a multi-page complex dashboard built for a city hospital.

         Img source: https://www.datapine.com/dashboard-examples-and-templates/healthcare

To build a dashboard, be it a one-pager or a complex healthcare data visualization system, besides an experienced BI engineer, you’ll need a specific BI tool. Which one to choose fully depends on your dashboard's expected complexity, feature set, and data volumes you operate with. Here are some options I’d like to mention:

• MS Power BI – Microsoft Cloud-based visualization tool 
• Tableau – tool for interactive data visualization and analytics, a twin-brother of Power BI
• Looker Studio – Google-owned online tool for data visualization and reporting 
• QlikView – BI tool for data representation in graphs and charts

What should be seen on a dashboard’s main page? It depends, of course. Nothing prevents us from applying a huge number of different metrics on the main page. But by doing so, we risk making our dashboard unreadable, which would not only fail to help but confuse even more. And that’s the main challenge any BI developer faces – to strike a balance between the amount of metrics a business wants to implement, and a concise presentation manner. This is particularly critical if we speak about data visualization in the Healthcare sphere, where having timely access to the right information can make a life-saving difference.

Primary functions of any medical dashboard can be summarized as displaying statistics, medical staff performance monitoring, and data-driven forecasting. However with the help of advanced tools the set of its functions can be expanded. Proceeding from simple to complex, I’d like to tell you about some configurations that can turn a basic informative dashboard into a powerful support tool able to facilitate the work of medical staff even more.

Dashboards can have reactive behavior and respond to changes in real-time. Say, an emergency patient, requiring urgent surgical intervention was admitted to a hospital. Emergency room staff create the event in the application, and it automatically adds it to a queue. The dashboard responds to the addition of a new event and displays changes in real-time, which allows medical staff to notice them and take measures immediately. 

It may sound straightforward, but as is often the case, putting it into practice is more challenging than it seems. The snag is that the application should have the event-driven architecture to support such capability. This automatically implies a higher level of complexity and plenty of effort to maintain the application. Therefore, it’s better to think thrice, before choosing such type of architecture. In case your system does not involve a large amount of users, devices and events  –  it’s better to manage with the most basic level of automation, when users send requests to a common database and the application displays the result. Of course, in this case, data is not updated in real-time automatically, but as long as the amount of data is not that huge, users will not experience any difficulties when refreshing it manually.

How can we ensure the effectiveness of the prescribed treatment? Integration with Machine Learning technology can help here. We select an ML algorithm and perform an initial test run using our historical data to evaluate the efficacy of the available treatment methods. After we finally decide which algorithms we are going to use, we conduct denoising and prepare our data for further processing by selected algorithms. The ML model compares the diagnosis with available treatment methods and presents which ones have demonstrated the highest effectiveness. Therefore, integration with ML helps to gain additional data not available in our database and project it into the future. 

And again, in words it sounds simple, that the chosen algorithm does everything by itself, and you gain the most accurate result with minimum effort. In reality, working mechanisms of Machine Learning algorithms still need to be fully understood, which makes it quite challenging to work with at the moment. And that’s where the shoe pinches: 

• Overfitting
  For example, your ML model was trained on certain data and provides a forecast with 95% accuracy. However, the model may not be capable of adapting to real-life scenarios or accounting for unexpected emergencies that might occur in the future (like the onset of the Coronavirus outbreak, when it appeared as if the entire world came to a halt). Little does the ML model know that history will never go on the same pattern, and it will persistently generate inaccurate forecasts. The good news is that the model can be retrained. The bad news is that it does not guarantee anything as it’s impossible to foresee how the ML algorithm will behave with different data sets
• Black box
  As a rule, none of the engineers comprehend what happens under ML algorithms’ hood. This concept is called black box – when you put inbound data in the algorithm, gain the outbound data, and have no clue how it is being processed. How to work with algorithms when there’s no precise understanding of their functioning? Good knowledge of higher mathematics is already half the battle. Another half implies experimentation, investigations, and knowledge sharing – that’s how you can gain a better understanding of working principles, and create maximally accurate forecasts.

Dashboards have the power to simplify life and simultaneously make it more complicated. In Healthcare, where numerous metrics abound, the temptation to incorporate as many as possible into a visualization system can be really huge, and you risk making it cumbersome and confusing. The same applies to functionality. Before stuffing your dashboard with numerous features, I strongly recommend to think carefully if they are really must-haves for you and factor in potential future challenges in system maintenance. And not to disregard using Dashboard Design Canvas, able to aid in resolving the eternal dilemma of balancing content with ease of perception.