Healthcare is one of the most challenging industries to be in today in the US. The costs of insurance, medications, equipment, employee salaries and complying with regulations lead to very thin margins for most hospital systems and private practices. Because of this, there is a lot of pressure on healthcare providers to minimize costs, improve service and reduce readmissions. Many providers have computerized systems and have implemented asset tracking solutions, which has led to better care and lower inventory costs, however, this is not enough. Systems that integrate machine learning into hospitals and provider offices, will improve care while reducing costs. In this first of two blogs (there was too much for just one!), we’ll talk about how machine learning can help improve care by increasing the accuracy of a diagnosis.

According to a study by the VA, 1 in 20 American adults– or 12 million adults — is misdiagnosed in outpatient clinics each year. This is a staggering number. Many times the misdiagnosis can be the result of not having a complete picture of the patient or the healthcare provider not being able to or have the time to interpret the test results and compare it to previous results from the same or different tests.  Also, pathologists can overlook certain factors that signal cancer. In a study conducted by the American Medical Association, 13% of the interpretations missed Stage 1 breast cancer cells and 48% of the interpretations failed to note atypia hyperplasia, a common precursor to breast cancer. Once again, detection before a condition occurs or when it’s in an early stage could mean shorter recovery times, less invasive treatment and a lower fatality rate.

So how could machine learning help with this? One of the ways to do this by creating a machine learning powered image processing software to analyze medical images. Often radiologists use their experience and their “eye” to identify abnormalities. However, no matter how experienced a radiologist is, there is always a chance that a human error will occur because an interpretation needs to be made quickly. Because of this time constraint, the radiologist doesn’t have the time  (nor access to) to compare the current image to hundreds of images. An automated solution could take in the current image and compare it, in real time, against hundreds or thousands of historical images. The solution could identify minuscule similarities that the human eye would have a very difficult time identifying. By identifying these small similarities, conditions could be detected early on and lead to less aggressive treatments and lower mortality rates. Ultimately, reducing costs and improving outcomes for patients. To see a solution similar to this, check out our imaging demo.

Also, another example where machine learning could predict a diagnosis is by forming a complete picture of the patient and other patients with similar conditions. Electronic health records have helped doctors track a patient’s care and be able to quickly see how a patient has changed over time. For example, with a few clicks a doctor could see that Patient X has reduced their blood pressure by 3% every year and her cholesterol has been in the normal range for the past 5 years. However, it is possible to be able to predict a diagnosis by using one patient’s data and then comparing it to other patients’ information. For example, Patient X comes in for her yearly check up and a wide range of vitals are taken and tests are performed. Currently, her doctor would have to analyze these results and look for anomalies. But, just like in the imaging example, her doctor probably does not have on hand the health records of hundreds of other patients with multiple diseases to compare Patient X’s results to and he definitely doesn’t have the time to do this by hand. A machine learning powered solution could take in Patient X’s vitals and test results and compare it to thousands of records of patients with both common and obscure conditions. Any significant similarities that Patient X has to other patients could be quickly identified and measures could be taken to treat the condition or take preventative care to reduce the risk of the condition advancing. Conditions such as seizures, diabetes, cancer and heart attacks may be able to be identified before they occur or when they are in very early stages.

These machine learning solutions can help doctors confirm a diagnosis is correct. This can be extremely helpful for doctors that are pressed for time and also doctors who don’t have access to a large auxiliary staff, like in small or rural hospitals. By having machine learning tools that can reconfirm their diagnosis patients will receive the right care at the right time – which can lead to lower readmissions, higher patient satisfaction scores and lower costs for hospitals. Truly a win-win-win.

Entrigna specializes in helping healthcare organizations and hospital systems create machine learning based solutions to that improve care,  minimize costs, and increase revenue. For more information on our product, please visit our website or e-mail us at