Big data in healthcare: 4 steps from theory to practice

Big data in healthcare: 4 steps from theory to practice

Big data is hot! But in healthcare, big data is still mostly in the hype phase. It is much talked about, but little used in practice. This is mainly due to ignorance. Dan Ariely of Duke University put it nicely: ‘Big data is like teenage sex: everyone talks about it, nobody really knows how to do it, everyone thinks everyone else is doing it, so everyone claims they are doing it.‘ This needs to change, as big data offers many opportunities for healthcare. Its proper use can make care better, more tailored, with fewer errors and therefore cheaper. In other words: taking value-driven care to the next level. But how do you make the most of it?

What is big data?

According to the Dikke Van Dale, big data is ‘large, unstructured amounts of digital data, especially as a source of information through further analysis’. We certainly don’t want to contradict the Van Dale, but big data is more than just large amounts of data. An often-used description by Mark Beyer and Douglas Laney (2012) makes this clear (see also Figure 1). They argue that, in addition to volume, big data is characterized by the speed with which data is generated, changed and spread (‘velocity’). In addition, the various forms of data are important (‘variety’). Finally, its reliability (‘veracity’) is a prerequisite for usable analyses.

Figure 1: What is Big Data

The added value of (big) data analysis

And when you have all that data, what can you do with it? A lot. McKinsey describes the following four types of big data analysis, increasing in impact and complexity, in its study ‘The ‘big data’ revolution in healthcare’ (2013):

1. Report: what happened?

This concerns simple analyses, for example to provide insight into the number of operations carried out in a hospital during a specific period. These results are suitable to present to, for example, administrators, healthcare professionals or healthcare consumers.

2. Monitoring: what is happening now?

This form of data analysis uses both recent data and real-time data. This makes it possible to compare the current situation with a benchmark or a desired situation. This type of data analysis is suitable, for example, for monitoring OR planning and utilization in a hospital. This data also enables the detection of contraindications when prescribing or dispensing medication.

3. Data mining and evaluation: why did an event occur?

Data mining and evaluation go one step further. These forms of analysis show the interrelationship of factors, thereby providing insight into cause and effect. For example, you can discover an (unexpected) relationship between the contracting of an infection by patients and the hospital room in which these patients have been located.

4. Predict and simulate: what will happen?

Making predictions about the future based on data analysis is the most complex, but also the most interesting. These analyses require, in addition to a large data set, the necessary technological and statistical knowledge. This is because it involves not only the processing of patient-specific data, but also comparison with, for example, the most recent scientific professional literature, comparable situations in other patients and the effect of a treatment. For example, predictive algorithms can be used to provide a diagnosis and treatment advice.

The big data trend in healthcare

Identifying causes, spotting trends and making predictions on the basis of data analysis have existed for longer than today. This becomes clear in Figure 2. Also, the history below is a great example of this.

Figure 2: Big Data analyses

Big data in the mid-19th century In August 1854, there was an outbreak of cholera in London’s Soho district. Within a month, hundreds of residents died. Because the cause of this sudden outbreak was unclear, physician and scientist John Snow decided to plot victims on a map of the neighborhood using their addresses. It soon became clear to him that most of the victims lived near a water pump on Broad Street. By cleverly combining information, Snow was able to determine the cause of the outbreak and the municipality was able to take action. The water pump was shut off and the epidemic was soon over. In addition, measures were taken to prevent the same problem in other districts and cities.

Big data analysis in practice

Big data offers many opportunities for healthcare. But before we can reap the benefits, there are a few steps to go through.

Step 1: Report outcomes rather than processes

If you report something, report meaningful things. So not process indicators – as is usually the case now – but outcome indicators. Examples of outcome indicators are quality, health and cost. Focus on quality and health. That’s how you serve your patients and attract good caregivers. They prefer to work in an organization where the focus is on improving care and increasing patient satisfaction, rather than on reducing costs. That way the knife cuts both ways.

To ensure that you are reporting the right outcome measures, it is important to establish clear objectives, outcomes and definitions in advance. Also look at what sources there are to draw data from. An evaluation matrix can help you to get and keep an overview.

Step 2: Monitor outcomes in result dashboards

As an organization, you naturally want to achieve the best care with the best outcomes at the lowest cost. Then it’s important to place the right emphasis in your care provision. It helps to regularly update everyone on key outcome indicators. Healthcare providers can see how they are performing and where the opportunities for improvement lie. Result dashboards on patient experience, quality targets and cost targets can provide this insight. Discuss the dashboards during regular meetings. And make sure all dashboards are always available to everyone, so employees can check progress at any time.

Step 3: Detect practice variation based on outcomes

If, based on your results, you come across variations in practice, this requires further analysis. For what is the explanation for this practice variation? By comparing processes and working methods, you can make improvements. In this way, your outcome indicators serve as the starting point for process innovation and quality improvement.

Incidentally, it is important not to take the outcomes of a specific patient group, but of your entire patient population. By pulling the population apart immediately, you get a distorted picture, because processes interact. The multimorbid patient, for example, cannot be categorised under one clinical picture. By taking a holistic approach, you get a real picture of how your entire practice functions.

Step 4: Predicting the cost of illness and claims

The last step, predicting sickness and claims, can be the most profitable for you as a healthcare organisation. International research has shown that the most reliable way of doing this is by using diagnostic analysis models and algorithms. The latter is often met with resistance. Unjustifiably so, because algorithms can detect patterns much better than an average doctor can, and that will benefit your organisation enormously. You can use it to detect high-risk patients, assess economic and clinical risks and make your practice more efficient, thus creating space in your capacity. This leads to better quality of care and a better negotiating position with the health insurer.

How are we doing and what do we need to do?

This is the theory, but where do we stand in practice? You can do much better than that. By means of figure 3 we try to make this clear. Vision and culture, structure and governance, and interprofessional collaboration form the basis for achieving better quality and lower costs – or value-driven care. But in addition to that, you need to arrange your financing and have your data in order. You can’t separate the two. Only if you have your data in order and apply the right analyses is it possible to do outcome costing. You will then know exactly what risks you are running with your patient population. At present, data collection, particularly in primary care, leaves much to be desired. Without a good dataset, meaningful analyses are impossible and it is difficult to conclude good contracts with health insurers.

Figure 3: Value-driven care organization

It is therefore important to work with the data. Organise this yourself, without the intervention of other parties. This prevents noise on the line. Register the outcome measures that are important for your organisation and make sure they are properly measured. It takes time, but it pays for itself. And don’t be afraid to invent the wheel yourself. There are tried and tested international standards for taking measurements, but for some inexplicable reason these have not yet reached the Netherlands. Take advantage of that.

For the analysis, it is advisable to use diagnostic models. It’s not easy. It takes smart heads. People who are good at maths and understand healthcare. At the moment they are still thin on the ground. Nevertheless, it is wise for hospitals to acquire this competence. As primary care organisations, this is of course a lot more difficult, but it can be overcome by organising it at an umbrella level, for example within a care group. In this way, the primary care sector can also join in a step that the health care sector in the Netherlands really needs to take to get and keep the health care costs under control.

Read more about the ins and outs of data management and risk estimation in our whitepaper.