Guest Article: Top 10 tips for successfully using speech recognition in EHRs and healthcare apps

Home > Guest Article: Top 10 tips for successfully using speech recognition in EHRs and healthcare apps

Moving away from paper is an automation journey that is both challenging as well as rewarding and choosing the right data entry technology is certainly one of the biggest challenges. Choose the right data entry mechanism and your journey is smooth; choose unwisely, and you’re in for a great deal of pain. There is no right mechanism for everyone so you’ll need to go through the various options with great care. One of my favorite data entry techniques is speech recognition (something I use for dictating blogs and writing articles) so I invited Nick van Terheyden, MD, Chief Medical Information Officer (Clinical Language Understanding) from Nuance Healthcare to offer tips and tricks that can help simplify and ease the move towards electronic records using speech recognition. In addition to spending several years as a medical practitioner, Dr. van Terheyden was also behind the development of an electronic health record in the early 1990’s and later became a business leader in one of the first speech recognition Internet companies so he knows his stuff. He’s a very interesting physician and you can follow him on Twitter @drnic1 and on his blog, Voice of the Doctor. Here’s what Dr. van Terheyden had to say about how to maximize speech recognition in EHRs as well as some pitfalls to avoid:

  1. Have the right hardware installed – while technology has improved and Moore’s law remains in effect doubling the number transistors (and hence power) of the chip. But doubling the power in the laboratory and store doesn’t translate into doubling of power of the desktops in a clinical setting that may have been installed several years ago and are now underpowered for current applications. In fact for many the hardware is woefully inadequate and nothing will kill the value proposition than the dreaded hourglass cursor or worse repetitive delays and even crashes. Speech technology requires processing power and memory – make them available. Buying the low cost home computer from CostCo or Best Buy that features an older processor such as the older Intel Celeron. This is especially true for the Netbooks that just don’t have enough processing power or memory to meet the need. Much better to buy a higher end business machine with an Intel Pentium4 or AMD Athlon 64 chip with at least 1 Gb of RAM and preferable more.
  2. Intelligent application Coexistence – putting a speech recognition application on the same desktop as any graphic intensive application that creates 3-D models out of 2-D input (such as Vital Images Vitrea, Tera-Recon Aquarius or Barco’s Voxar 3D) is asking for trouble. Both these application need powerful processors and lots of memory. If both are installed they’ll be competing and likely neither will work well and frustrate users. Same applies to other processor and memory intensive applications such as video editing and processing applications or heavy data query/reporting tools.
  3. Use good quality microphones and sound recording equipment. Buying the cheapest microphone and expecting to get good results from speech recognition is similar to using the cheapest quality meats and food ingredients and expecting a 5 star dish worthy of placement in an Iron chef competition. If you want good results you need good quality audio. Most vendors have a recommended list of recording devices – find a range of choices and offer them to your users. Most of the modern Plantronic headsets are firm favorites with Speech recognition users including myself. The DSP 300 offers wireless connection, and the Audio 310 offers a flexible wired connection with straight audio jacks and a handy USB adaptor for computers that have no audio jacks built in. But for those looking for low profile connections the RevoLabs RF:xTag offers a device that clips on to your lapel for almost invisible recording capability. Bonus Tip: Assemble the full collection of styles and types of devices and keep them on hand available for users to view and test to determine personal suitability. One colleague has a suitcase filled with the various types of devices that is he takes to show colleagues the choices and help them settle on the most appropriate choice for them in their clinical setting
  4. Environmental Considerations. Think about the clinical environment – a busy and noisy setting may need a different approach for capturing audio. But there are basic actions – a radio playing loudly in the background is unlikely to help a speech recognition solution return accurate results. While the staff may want to have the radio playing this may not be the best choice if it interferes with the effectiveness of the technology. Bonus Tip: There has been some work that suggests that better than a quiet environment is an environment populated with White Noise – in troublesome areas introducing some white noise generators (assuming hey are acceptable to patients and staff) may help improve accuracy and the overall acceptance and usefulness of a speech solution.
  5. Create a Standard and Replicate. Once you have found the ideal configuration(s) using both technical specifications, local circumstances and requirements and some inevitable trial and error get several users to try the configuration to make sure this meets the broadest range of needs. Then rinse, lather and repeat. Take this configuration and replicate it faithfully throughout the facility, clinic or enterprise. So if you have found that Computer A, with a specific motherboard, memory size and type, hard drive, graphics and audio add options buy the identical set up and replicate the exact configuration down to the operating system version, software and even patches. This might sound daunting but this is exactly what corporate and enterprise IT system support centers do. They create one standard and then create an exact image of the drive and copy this to all the other machines (purchasing the necessary licenses for the number of installations created). This can be done with tools designed for this purpose for large numbers of machines or simply by following an identical setup procedure for each new machine. Once you have a working environment don’t be in a rush to update software components with the latest versions or patches before you test to make sure updates don’t break your working configuration.
  6. Anticipate Resistance. Expect resistance – resistance is part of life and best faced directly with a clear understanding of the reasons. Some of this arises from the prejudice based on historical bad experiences with technology that was implemented either before it was ready or without the right environment and support. Set up opportunities for everyone to see the technology in use and try it for themselves. Clinicians successfully using the solution in use in a there own clinical setting is most helpful in overcoming the resistance
  7. Quick Portable Guides. Develop a Short Guide/Cheat Sheet of Commands and tools and Techniques – taking a leaf out of the Apple playbook and simplify the experience. Overwhelming already frazzled clinicians with hours of training is unlikely to lead to success. Identify the key pain points of clinical documentation that can be helped using speech recognition and dictation, accepting that for some clinicians using speech may not be suitable for them at this time or in their particular circumstance and focus on delivering success here. This may not include all the features and short cuts available when using a speech recognition solution but early success is more important than using every element available in a solution.
  8. Preparation. Prepare, prepare and prepare some more – Once you have identified good targets offer a simple path to adoption that does not entail long user training sessions. Capitalize on the built in tools available in speech recognition products that jump-start the building of a customized profile. Pre load with previous reports for that user, carry out minimal audio enrollment and test the profile for effectiveness. At the elbow training is most helpful here to get a critical mass of users up to speed and using the solution effectively. Once you have the critical mass or early adopter super-users this group will help the late adopters when they elect to transition to using he technology (and this move will be driven in large part when they see their colleagues being successful using the technology)
  9. Horses for Courses. Some users will use the most basic features for simple typing replacement. Dictating into a box and seeing the results appear and being able to complete a document immediately may be enough. Mixed with some use of the keyboard and other documentation tools may be enough to keep some clinicians happy. Others may make use of templates, an the ability to customize standard content for different patient interactions. And there will be some who make extensive use of the features including all of the above as well as command and control to navigate applications, sign documents, display results and automate complex tasks. Accept the variation and be prepared to support all levels of usage –success is measured by each doctor not by a standard set by a group or department. If a doctor feels successful then he is
  10. Identify champion(s). More often than not facilities are full of existing users who have speech recognition in use in a range of circumstances. These individuals are often well connected to the industry and other power users through online forums and conferences. Identify them early and make them a apart of the process and implementation baring in mind that if you are asking them to give up clinical time they need to offset this lost income in some way – many contribute excess hours and time to projects to of goodwill and passion but finding ways to compensate them is likely to create a stronger more long lasting supporter

Digitizing medical records is not so much a destination but a journey and one that we must all take. There are challenges and achieving success is not entirely dependent on speech recognition technology but where it does make sense getting the best from the technology will go a long way to easing the strain everyone feels with change. The benefits are clear the question you must ask yourself as a physician is can you afford not to go digital and more importantly can your patients.

Also, if you’re looking for EHRs and other software don’t forget that you can get [free medical software advice on][4].


Shahid N. Shah

Shahid Shah is an internationally recognized enterprise software guru that specializes in digital health with an emphasis on e-health, EHR/EMR, big data, iOT, data interoperability, med device connectivity, and bioinformatics.