Predictive analytics is becoming more and more popular in the healthcare industry.

Based on a survey of 223 health payer and provider executives conducted in 2017 by the Society of Actuaries, 57% of executives forecast that predictive analytics will save their organization 15% or more over the next five years.

47% of providers currently use predictive analytics, with 89% of providers currently using predictive analytics or planning to begin in the next 5 years.  Outcomes where providers anticipate improvement are: patient satisfaction, hospital readmissions, and staffing/workforce needs.

What is predictive analytics, exactly?  We can answer that question by examining an established market pipeline in analytics: Descriptive, predictive, and prescriptive.  Descriptive analytics provide insight into the past, through aggregation and presentation of historical data. Predictive analytics provide insight into the future, by using statistical and machine learning techniques to ascertain what will happen in the future based on retrospective patterns.   Finally, prescriptive analytics are a newer branch of analytics, advising client on what they should do.  This pipeline is presented in sequential order as well as from least to most complex.  Descriptive analytics in a world where massive amounts of medical data are harvested already provide great value to providers and payers alike.  This is a testament, then, to how powerful predictive analytics can be when assumptions are reasonable, appropriate methods are used, and the solution is properly integrated into clinical workflow.

Obstetrics Department Birth Prediction

Let’s look at some cases of exploring such solutions in real clinical environments.  IPS designed a forecast for a hospital obstetrics department, predicting how many births would occur in subsequent months.  The forecast was created by aggregating individual predictions, using the fact that patients with recent appointments or who are nearing delivery of their baby are more likely to deliver at that particular hospital.  This was of tremendous utility to administration, who would have to decide how many staff members are required to meet patient demand, as well as make decisions on whether to defer patients to network hospitals due to capacity constraints.

Feasibility of Harms Prediction

In another example, IPS participated in the evaluation of the feasibility of predicting a harm in a hospital environment.  Inclusion-exclusion criteria were defined for determining the use-case of the work; a harm was eventually decided upon based on data availability, impact of a potential predictive solution, and whether predictive information would be actionable.   The team developed a prototype of the model and performed a clinical usability study to understand how to best integrate it into a physician’s daily workflow.  The model was found to have moderate performance, predicting at the point of admission whether a patient would acquire the harm during their episode of care.  Next steps in improving the effort include integration of patients’ outpatient histories, as well as providing continuous real-time predictions.  Ultimately, clinicians will be able to risk stratify and provide appropriate interventions – such as antibiotic administration and resuscitative fluid control — for patients who need critical care most urgently

Predictive analytics has tremendous implications towards operational efficiency, harm prevention and, by extension, overall patient satisfaction.  IPS has learned from experience that there are several important factors for a successful predictive analytics initiative.  Namely, clinicians have a reluctance to consider or adopt solutions that fit outside of their established clinical workflow.  They also have resistance to information overload – who wants to make sense of the fact that one of their patients has a 1.7% chance and another has an 8.2% chance of acquiring a Healthcare-Associated Infection (HAI)?   However, given appropriate feasibility studies guided by human-centered design principles, solutions can be configured and designed based on the needs and workflow of the client.  The resulting impact on patient care is very powerful: clinicians are able to more successfully reduce preventable clinical deterioration, harms, avoid readmissions, and more.

About the Author

Richard Bryant is a data scientist with experience using data analysis, predictive modeling, and discrete event simulations to drive innovative technology solutions.  Utilizing an M.S. in Applied Statistics from the University of Michigan and a Lean Six Sigma Green Belt certification, Richard has provided analytical support to healthcare process improvement projects in various hospital departments, such as radiology, obstetrics, and pharmacy.  Prior to his role at Improvement Path Systems, he provided statistical consulting services to the Michigan Surgical Quality Collaborative and to the educational technology provider Echo360.