<img height="1" width="1" style="display:none;" alt="" src="https://dc.ads.linkedin.com/collect/?pid=486348&amp;fmt=gif">
BlogHeader.jpg

Keep it realFull of thought

USE PREDICTIVE ANALYTICS FOR BETTER HEALTHCARE

Predictive Analytics, Health Care | [fa icon="comment"] 0 Comments

healthcare_-1

Following on from our last analytics blog Use Predictive Analytics For Better Employee Retention, a keen 360˚ view can likewise be applied to health care. By drawing on diverse new data sources (Automated Electronic Health Records, Mobile phone sensors, Nutrition, bio-sensors, …), combining them with the tried and the true (Demographic, social and socio-economic factors, Clinical study findings, …), and finally by employing Predictive Analytics to make sense of it all, health care professionals can improve outcomes for their patients.

Predictive Modelling in health care is not new. Statistical studies serving as fuel for evidence driven practice, is a cornerstone of modern medicine. What is new, and what analytics specialists such as Mindfull can bring to the table, is the extraction of insight from massive, modern “Big Data” scale data, and at the other end of the spectrum: The careful identification of customised patterns from small scale, “n = 1” patient data.

Two proven and widely used analytics techniques to start off with are predicting ward demand, and optimizing surgical units.

Predicting the demand for care within a ward is important for staff scheduling. We need to ensure that patients receive the level of attention they require, while not wastefully leaving excess resources idle. Admission records can be used to segment arrival flows of patients, and to calculate expected lengths of stay. Simulation can then be used, to help you make smart decisions.

Patient throughput is high for surgical units, especially so in acute wards. Tracking tools, devices and even patients can be a challenge in these quick, high stakes environments. By taking a step back and carefully modelling the surgical wards (or the entire hospital ecosystem), we can identify the bottlenecks and probable breaking points. Mathematical models can be developed to help understand and optimize processes. We can use sensitivity analysis and simulation to weigh up the “What If’s”. For example, what would be the impact of hiring another surgeon? What would be the impact of diverting a less critical class of patients to an elective ward? What would be the impact of the surgical unit remaining open for two more hours?

Moving to a lower granularity, we can use data to make predictions for individual patients. A patient being discharged too early is costly, and can endanger the patient by distancing their access to emergency staff and facilities. Using the Electronic Health Record, demographic information, Doctor’s notes, and the vast historical data of past patient outcomes, we can predict which patients are at high risk for readmission. Patients flagged by these models can be asked to remain as an inpatient, or can be given a more thorough examination prior to dismissal.

Back inside the hospital, with the ever increasing richness of monitored and recorded data, and development of real-time bio-sensing devices, we can predict some of the most life-threatening and time sensitive events, before they happen. Recent research using Deep Neural Networks have successfully predicted epileptic seizures, using patterns derived from patient EEG signals.

Support Vector Machines, have been trained to predict hospital cardiac arrests, at a rate superior to incumbent heart attack prediction best-practice methods, already used by Doctors (eg. Medical Early Warning Score, MEWS).

Prediction ultimately opens the door for more acute prevention, precaution, or to new medication possibilities.

At the most cutting edge, IBM Watson, a cognitive computing platform, is increasingly being adopted for tasks such as assisting oncologists in making their diagnosis and decisions. A cheap, quick and unbiased “second opinion”. Mobile phone apps, making use of the many sensors contained within modern phones, can be used to detect deviations from normal behaviour (eg. An elderly person falling down and injuring themselves, a seizure). These same sensors will eventually be able to detect the anomalous patterns which the above case studies have shown do exist in the lead-up just before harmful events (heart attacks, seizure, cancer and so on).

Mindfull Limited is currently working on an exciting, experimental project in this same area: Predicting professional sporting injuries, before they happen and hospital readmissions.

To find out more about predictive analytics and how it’s changing the playing field across a whole range of industries, join us for ‘The Power of Predictive’ breakfast seminar on Tuesday 26 May at Eden Park. For more information and to register, click here.

Sources:

[1] Patients, queues and hospital beds: modelling and optimisation.

Ilze Ziedins.

https://www.math.auckland.ac.nz/CULMS/wp-content/uploads/2010/08/Ziedins-ppt.pdf

[2] Classification of Patterns of EEG Synchronization for Seizure Prediction.

Piotr Mirowski MSc, Deepak Madhavan MD, Yann LeCun PhD, Ruben Kuzniecky MD.

http://yann.lecun.com/exdb/publis/pdf/mirowski-cneuro-09.pdf

[3] Early Code Blue Prediction Using Patient Medical Records.

S. Somanchi, S. Adhikari, A. Lin, E. Eneva, R. Ghani.

https://d277f6d674b2cfd0d2436b2145030d5d731cac78.googledrive.com/host/0B0TBaU3UgQ0Da3A2S2VWNTRzc1E/25.pdf

 

Topics: Predictive Analytics, Health Care

All in a days work

#32
cups of coffee consumed
zestfull
#600
lines of code cracked
skillfull
#450
fantabulous happy customers
delightfull