Big data for better patient care
Dr Sapan S. Desai discusses how healthcare organisations can harness big data and machine learning to improve healthcare quality
Two of the most significant trends in healthcare today are big data and machine learning. Together, they are improving the quality of care delivered to patients while successfully decreasing costs. New applications in imaging, diagnosis, treatment, and healthcare management are occurring on an almost daily basis as innovation accelerates in healthcare.
Big data is more information than one healthcare system can reliably collect. Consisting of millions or billions of points of data for a single disease process, big data provides the most accurate view of the population of patients impacted by a particular disease process. Traditional statistical analysis on this sort of information will typically identify dozens of statistically significant findings. Unfortunately, few of those results will be practically meaningful, and truly innovative findings will often be lost within the vast amount of data being collected.
Machine learning applications have helped identify both practically significant and statistically significant findings. Tools such as neural networks, support vector machines, decision trees, k-means clustering, and deep adversarial networks excel at identifying signals within the noise. When mated with a lean six sigma performance improvement program and project management tools, a healthcare organisation can prioritise its opportunities for clinical and financial improvement and realise a meaningful return on its investment.
Predictive models
Our group recently presented several innovative examples of this work using the cloud-based QuartzClinical healthcare data analytics platform. At the American College of Surgeons meeting and the International Hospital Federation meeting in 2018, we presented machine learning applications using big data as a way to predict longevity in dialysis patients, reduce mortality from ruptured aneurysms, and decrease readmission to the hospital. Traditional statistical analysis will typically yield predictive models that are accurate approximately 20 per cent of the time. Our machine learning algorithms within QuartzClinical have an overall accuracy exceeding 80 per cent for most models. Some models, such as those for dialysis patients, have an accuracy exceeding 90 per cent.
Healthcare organisations that are moving toward a value-based framework with a focus on pay for performance will benefit from the business intelligence opportunities created by healthcare analytics platforms
The impact of this work is the ability to identify the patient, hospital and surgeon-specific factors that impact healthcare. These factors are individualised to the patient, thus delivering truly personalised medicine. When combined with a performance improvement program, it is possible to mitigate the risk factors that are most likely to lead to an adverse outcome, thus using a targeted approach to improve healthcare. When measuring key outcomes variables such as length of stay, mortality, readmission and cost of care, these models offer the ability to quantifiably measure finite end points.
Reducing costs
As an example, our group developed a predictive model to identify vascular surgery patients most likely to be readmitted to the hospital. With over 100 million data points in the big data model, our machine learning algorithm was able to narrow the number of meaningful predictors of readmission to just a few dozen. The impact was a 57.3 per cent decrease in 30-day readmissions, leading to a projected US$290,400 decrease in cost of care. If such a model were to be implemented nationwide, the overall savings would be in the hundreds of millions of dollars.
There are several issues associated with widespread adoption of these tools. The most significant issue is access to the billions of data points needed to understand a specific disease process. One of the key limitations is lack of insight into procedure-specific data points which are simply not recorded by an electronic medical
record. The second deals with poor interoperability between a hospital’s various data sources – the electronic medical record, financial system, supply chain system, etc. Another key limitation is that a single hospital system will simply not have the scale necessary to achieve the number of data points needed for a successful machine learning algorithm. For these reasons, contracting with a vendor who specialises in data analytics and utilises sophisticated machine learning algorithms to help analyse this information can provide significant strategic advantage.
Healthcare organisations that are moving toward a value-based framework with a focus on pay for performance will benefit from the business intelligence opportunities created by healthcare analytics platforms. Improving both the quality and cost of healthcare can lead to a significant improvement in the operating margin, particularly in today’s challenging financial climate. ■
