How To: Mitigate increasing risk in healthcare with advanced analytics

Sharp Healthcare: Predicting patient decline with machine learning and EMR data

Based in San Diego, California, Sharp HealthCare runs a specialized rapid response team for medical emergencies. Using technologies from Intel, Cloudera and ProKarma, it developed a predictive model that can leverage the hospital’s EMR data to identify patients at risk for requiring a rapid response team intervention within a defined period of time. The solution analyzes a range of data including blood pressure, temperature, and pulse rate, and uses machine learning to train the algorithm over time.

When Sharp tested its proof of concept model against historical data, it found it to be 80 percent accurate³ in predicting the likelihood of a rapid response team event within the next hour, demonstrating the potential to drive real-time clinical interventions, improve outcomes and enhance resource utilization.

Penn Medicine: Predicting and preventing sepsis and heart failure

Penn Medicine operates a network of healthcare facilities in Pennsylvania and southern New Jersey. Its dedicated data science team aims to harness the power of data to assist with identifying patients at risk of critical illnesses that may have been missed by current diagnostic techniques.

To this end, Penn Medicine has worked with Intel to create a collaborative open source data science platform called Penn Signals* that can help clinicians make faster, smarter decisions based on large-scale clinical data and big data.

The first trials of the platform covered two of the most common and costly issues for hospitals: sepsis and heart failure. In the case of sepsis, the pilot was able to improve its identification of cases from 50 percent to 85 percent, and improve identification time before the onset of septic shock from two hours to as much as 30 hours.⁴

For heart failure patients meanwhile, the algorithm provided Penn Medicine’s clinical team with its first accurate measure of the quantity and distribution of sufferers within or between its hospitals. Using standard diagnostic tools, 20–30 percent of heart failure patients had not been properly identified. With the predictive model, these patients could be identified and given education to self-manage their condition, leading to a marked decrease in readmission rates.⁵

Montefiore Health System: Advancing patient care through holistic data analysis

Montefiore Health System, which runs a number of healthcare facilities in the New York, has deployed an advanced analytics solution, Semantic Data Lake, to automate and accelerate the identification of patient risk.

The solution provides Montefiore’s facilities with a holistic and realistic profile of patients by leveraging multiple data sources and formats. This can range from clinician data, through to demographic, environmental, behavioral and wellness research findings, population demographics, and medical imaging. Advanced analytics algorithms are then used to analyze the data, and machine learning is applied to optimize the insights over time.

The first implementation of Semantic Data Lake was able to identify patients at risk of death or in need of intubation within the following 48 hours (the window of opportunity to complete a successful intervention) with accurate prediction at a rate of more than 70 percent⁶. This allowed physicians to deliver more timely, pertinent and accurate treatment able to help prevent fatal episodes or respiratory failure.

Find out more about how to get started with predictive analytics at your healthcare organization in this new white paper from Intel.