Big Data Being Used To Predict Patient Outcomes In Healthcare

Image Credit: By DARPA (Defense Advanced Research Projects Agency (DARPA)) [Public domain], via Wikimedia Commons

A recent and growing trend in the healthcare media coverage is about application of algorithms, machine learning, big data and other such technologies in healthcare. While such technologies have large scale applications in healthcare, one of the important application is in strengthening the accuracy of predictions in patient care and thereby helping in improving patient outcomes.

Uncertainty in Healthcare and Need of Big Data Models

Currently, there is a lot of uncertainty in health sciences and it achieves its peak when one is hospitalized. Some of the uncertainties that come to mind are, what are the chances that the patient would be successfully discharged from the hospital in a given number of days? What are the chances that one would need a re-hospitalization for the same disease? What are the chances of cure? etc.

While such questions are easy to ask, they can only rarely be answered scientifically. The existing medical knowledge, as well as our current understanding and techniques, does not allow us to exactly quantify these risks and calculate the right probabilities. It is hoped that with the help of advanced statistical and computing techniques more robust data models can be created which can derive better answers to these questions.

In the first instance it can be argued that why should one be bothered about predicting hospital outcomes? What if a tool can predict that Mr. X has a 95 % chance of dying or of a re-hospitalization? The same reason which holds true for predicting bad weather, wars, famines, economic cycles, etc. holds good in this case also, which is "to reduce the magnitude of their adverse effect". If one can accurately predict a patient's outcomes based on a given condition, one can work towards eliminating suffering and provide the best possible care to patients.

Another thought that may come to mind is that what is the need for big data models and why can we not have such answers derived by simple existing data models? The current simple data models pose significant challenges in predicting accurate outcomes in health services. Existing prediction models generally give results that are around 80 % accurate as they only take into account only some important variables and miss out on others which are needed in model to take the accuracy above 95% levels.

For example, an existing model may only include age, gender, and core disease variables like stage and time since disease onset, etc, whereas a big data model may include a large number of additional variables like social status, coexisting diseases, individual genetic makeup, etc.

Experiments With Big Data Models

The existing computing techniques and software are generally compatible with the old statistical methods and cannot accommodate such a huge number of variables. Specialized coding languages like R, Python, and MATLAB, etc. have been used to address this issue.

Have there been any real successes off late or are such things still in the trial? Researchers have been using big data techniques like machine learning and neural networks in healthcare for quite a long time now. The real success has been met in the area of cardiology where various studies have shown better predicting models for heart failure patients. These are equivalent or better than the existing models at predicting hospital re-admissions, predicting prognosis in acute coronary syndrome, and for stroke prediction in atrial fibrillation

Is it a very rosy path ahead for big data in healthcare modeling or it has its own pitfalls? While the techniques may be better, the data sets on which these work sometimes contain erroneous entries. These like the simpler models also suffer from false negatives and false positives and classification issues. The gap between demonstrating improvements in model performance and impact through actual implementation is seen in cardiology practice. There are some exciting things in the picture also like linking big data to areas beyond cardiology, wearable patient devices, and clinical decision support integration.

About the author: Dr. Naval Asija is a licensed MBBS Physician from India. MBBS is the equivalent of the MD degree offered by international medical schools. He is based in Delhi, India, and works as a medical writer, editor, and consultant. He supports medical researches as an author's editor, medical communication companies involved in medico-marketing activities, and medical technology companies in improving their products. He can be contacted via his LinkedIn Profile: https://www.linkedin.com/in/navalasija/

Disclaimer: The content provided here is NOT a substitute for professional medical advice. If you consider taking any action based on the above information, we recommend that it should be first confirmed with your doctor. Our detailed disclaimer statement can be read at our homepage