By and large, most healthcare systems are awash with data but short on insights. The advent of big data and analytics approaches is making it easier for organisations and individuals to use existing data to improve outcomes for individual patients, for the organisations themselves and for the wider health economy.
By Professor Ben Bridgewater, Director of Healthcare Strategy, CSC, and former cardiac surgeon
As a cardiac surgeon, I led a national cardiac surgery register that collected comprehensive data and then used applied analytics on it, to both satisfy regulatory requirements and drive quality improvements. The results were spectacular: a decrease in risk-adjusted mortality of more than 70 per cent over a decade.
At that time, there was no commoditised approach to big data management, so we had to do everything ourselves. Things are changing now with more advanced big data approaches, including better analytical software, bigger, cheaper and more secure storage capabilities, and enhanced computing power.
At the simplest level, healthcare analytics can be divided into approaches to understand what has happened, what is happening right now and what is likely to happen in the future. With my national cardiac registry work, I layered these approaches to drive better outcomes on an individual, departmental and national level.
As long ago as 2001, the department where I worked collected comprehensive patient-level data on cardiac surgery operations. A national benchmarking process showed that our mortality rates had been higher than expected over the previous three years, which came as a surprise, since there were no accepted benchmarks until that time. This analysis of what had happened was a wake-up call, and I bitterly regret that we did not have real-time information that could have allowed us to generate those insights sooner.
On a superficial level, the analysis told us that we needed to make changes. Deeper post hoc examination showed some interesting findings. The prediction model, which used estimated mortality for each patient to derive expected overall hospital outcomes, was built from national data which showed that cohorts of patients with higher predicted mortality had higher observed mortality. This is as it should be, but when we looked at the predictive ability of the model in our hospital, it did not fit our data. We were excellent in caring for higher risk patients, but not good enough on the lower risk categories. This actionable insight enabled us to change a number of things about our practice, and we went on to be listed as the hospital with the lowest risk-adjusted mortality in 2005. This is insight in action, driving better outcomes for patients, at scale.
Moving on from what has happened to what is happening right now is the next stage of the journey. Many healthcare organisations have developed sophisticated approaches to visualising data, and in some places that has become an industry in its own right. However, this does not always lead to better outcomes. There are two major challenges: 1) Sometimes there is too much data, which does not focus on key indicators to drive better outcomes, and isn’t delivered in the right way and at the right time to people delivering or managing care; and 2) often, only incomplete data is available, significantly restricting any analytical approach.
For example, in my cardiac surgery practice, I had a mountain of information on what was happening to patients in my own hospital, but almost no data on what happened to them afterwards. We knew that up to 15 per cent of patients were re-admitted in the first 30 days after discharge (with adverse consequences for patients and the health economy), but we did not have data that gave us the ability to analyse this at any level of granularity to drive improvements. We also had little idea of how effective many of our treatments had been 1 year, 5 years or 10 years down the line.
Key to driving better insights is the integration of data across the complex health and care economy over a prolonged time frame, as well as promoting action to support better care for patients and contain overall costs.
So, what is the place of analytics in predicting what is likely to happen in the future?
Vital to outcomes transformation is teasing out current pain-points in care delivery, and then producing actionable insights that can be delivered in a way which enables change. This applies to the analysis of what is happening, but is particularly powerful when insights into the future are coupled with an ability to do something about them. This requires a mixture of data, business consulting, data science and user-centric design.
CSC’s Open Health Connect approach allows collation of data from across a health and care economy, providing a “data fabric” for generating more useful insights through machine learning. This data can be as near to real time as necessary, using a combination of integration approaches.
We have recently developed two specific use cases for machine learning for predictive analytics.
The first is using routinely collected administrative data to predict elective hospital lengths of stay. This delivers two distinctly different types of actionable insight: 1) what the organisation can do to minimise the variation; and 2) how the organisation can use patient-level data to “load balance” its admissions and increase efficiency. Key to this is an iterative, industrialised approach, which is repeated as often as required to derive insights, coupled with a sophisticated method to deliver those insights through appropriate dashboards that help hospital administrators obtain better outcomes.
The second use case leverages existing healthcare data from across different organisational silos, enhanced with non-traditional data sources, to predict emergency care flows. Again, smart algorithms are of little use unless the insights are actionable and delivered into the workflow of the people running the show in a timely way. Key to this approach is ensuring that the predictive model is bespoke for the relevant organisations. Anecdotal data tells us that centre city emergency departments are quiet in periods of very good weather, but seaside ones become overwhelmed. Each hospital needs its own model.
We are on the cusp of exciting things: The potential power of machine learning in healthcare is enormous. My view is that we need to consider these approaches from the value discipline approach of operational efficiency (which is where we have focussed existing efforts on our current uses cases), product leadership and customer intimacy. Globally, healthcare needs greater efficiency, both within and across organisations, to deal with escalating costs and constrained resources.
Revolutionising patient pathways to ensure that they are truly effective requires the “product life cycle management” approach adopted by product leadership organisations. In healthcare that includes using advanced data analysis and population segmentation to bring novel approaches into play rapidly. Here we also need to recognise that not all initiatives will deliver the required results. We must be humble and carefully evaluate the effectiveness of what we change using advanced techniques, and we must be prepared to fail fast and move on. Predictive analytics also has a critical role here in supporting payers and moves toward capitated healthcare budgets.
The use of smart analytics for personalised medicine in customer-intimate care is also likely to revolutionise outcomes, including building bespoke treatment choices, missed-opportunity detection approaches, and a shift from reactive to proactive healthcare delivery.
The data science is no longer the hard piece. What’s required is a major advance in data integration coupled with an absolute clarity of purpose about what questions the analytics will help you answer. That is the real challenge.
Professor Bridgewater will be at HIMSS17, February 19-23 in Orlando – CSC Booth 2773.