The Prescription for Effective AI in Healthcare is Quality Data

April 02, 2019

Nowhere is Artificial Intelligence (AI) more promising than in the healthcare arena. The potential of using AI to help find cures to diseases or more effectively bring treatment into underserved, hard-to-reach areas is huge. Although some developments may be many years away, AI is currently being used in many areas of healthcare, from prevention and diagnosis, to treatment and monitoring, and more.

AI is enabling this by crunching huge quantities of data, detecting patterns in images, behaviors and symptoms and becoming smarter by teaching itself in real time in order to address some of the most pressing healthcare issues today. But to get the most value from AI, healthcare organizations – and the industry as a whole – must overcome some key challenges.

Capturing the right data

There’s no doubt that more data is being collected today and that trend is expected to continue. Some sources estimate that healthcare is currently responsible for 30 percent or more of all data and an IDC study predicts that this will increase by 36 percent compounded annually by 2025. This will outpace the growth rate of the manufacturing, financial services and media and entertainment industries. Part of this growth will come from data culled from “smart” diagnostic equipment and devices, including wearables, that can be uploaded for analysis.

While there’s a lot of healthcare data and more to come, it’s important that we’re collecting the right type of data. There are several key principles data scientists follow to develop effective AI algorithms, and one of the most important ones is that the more data, the better. When a data scientist develops algorithms to try to test a hypothesis or solve a business problem, he/she uses data as the fuel for the AI engine. The more data that is available, the more accurate a picture the program can develop of what’s going on.

At the same time, the quality of the data is essential. That requires data to be relevant, accurate and properly prepared, e.g., coded, cleaned, etc. In addition, it’s important that a variety of data is collected. If you only provide information from one source, it’s too limited. It may not include the accurate answer to the problem you are trying to solve or to accurately predict outcomes. Without the right data, a clinician could be missing medical conditions or misdiagnosing them.

Data challenges in healthcare

There are several data challenges inherent in the healthcare industry that must be overcome in order to develop sound AI programs that provide value, including:

  • The focus has been elsewhere. For the past several years, the major tech investment and focus in healthcare has been on converting paper-based medical records to electronic health records (EHRs) to comply with government regulations, gain incentives and avoid penalties. While that’s clearly important, the potential for investment in AI to develop better diagnostic, research and treatment options shouldn’t be overlooked.
  • We’re not aggregating as many different data types as we should. While the industry is furiously working to collect data, a lot of it is coming from medical records and other text-based data sources. But to get a more complete picture, consider all the sources, including audio, imaging and video. For example, why can’t a device be placed on a stethoscope to capture the audio of a heart murmur, arrhythmia or other heart-related ailment and then upload it to a centralized database? It can not only be used to monitor the progression of the condition for that particular patient, but also by AI programs for training data to help diagnose the condition in others. Or based on how that patient responds to a particular treatment (how the heart rhythm has changed), it can be used to provide data for predictive analytics programs. In a similar way, images can be captured of skin conditions or other problems, and videos can be taken of a person’s gait or other conditions. Since AI programs are able to easily work with videos, images and other types of unstructured data, it makes sense to make that data available for analysis.
  • Healthcare data is siloed. Many departments, from pharmacy to radiology, still have data trapped in siloed systems that they are not sharing. In addition to investing in integration, healthcare providers need to offer incentives to get these different departments to share the information that they have. Organizations need to determine who the owners are and communicate clear financial rewards and benefits for tearing down the siloes and sharing data.

While AI in healthcare is still in the early stages, the possibilities it offers in prevention, diagnosis, treatment and monitoring – just to name a few – are immense and exciting. The first step is to get our data houses in order so we can take advantage of the greenfield opportunities that are available to help clinicians make better, more informed decisions, get better outcomes, and – even get further along the path of curing cancer and other diseases.

 

Leave a Reply

  • (will not be published)