Can Medical Record Abstraction Be Automated?

Ahhh, the age old question.  Can (or should) tasks that humans can do very well, but perhaps not very efficiently, be automated with technology?  While the views on this topic would vary greatly from an abstraction service provider to a low-budget medical research project manager, there are perhaps a few things that could be defined to help one decide what is best for their particular medical record abstraction situation.  Consider these ideas:

ONE. Longevity of project

Effective automation is best put in place if time is taken to truly understand the most common scenarios that will truly allow you to maximize the design of the automated processes.  Understanding the most common document types, what information per document type needs to be found and abstracted and where the information ultimately needs to end up are just some of the items that need to be defined and programmed in an automated system.  So, if you have a short-term or one-time project that is not likely to repeat itself, it does not make sense to put all that work into defining the automation for it.  However, if medical record abstraction is required on an on-going basis or for a longer-term, high-volume project, adding automation to the processes will likely increase efficiency and consistency in the long run.

TWO. Quality of the Documents

While technology has improved greatly in the last decade around translating poor quality documents into electronic data that can be automatically abstracted, there is a point in which technology cannot work if it cannot read the words.  So, if you are abstracting medical records from the 1880’s or if their quality is consistently poor, there is little chance that automation can be applied. I should mention handwriting at this point as well as the technology, from our perspective, has not yet reached the levels of accuracy to enable confident automation in all areas.  You don’t have to assume the answer here, however, nor do you need to assume that no automation can be applied.  A vendor will typically give you an assessment for free as to whether their automation technology will be successful or not.  We don’t typically like to pursue projects that won’t be successful, so reach out an ask for an assessment and you will likely get a very honest answer.

THREE. Variety of Documents and Variety of Sources

You will be surprised to know that this does not really matter. Automated document classification is powerful these days and can easily be used as a front end to medical data abstraction to either make your human abstractors more efficient or to assist an automated solution in being more accurate.  Robust algorithms and machine learning allow a system to identify documents pretty easily and then automate the sorting and abstraction of data based on the rules defined for that document type.

FOUR. Defining What You Need From Each Document

This needs to be done either way.  Either you need to inform abstractors of what you want from each document type or you need to train software.  While defining it for software involves more thought and definition (as software is only as smart as how it is programmed), it can be done and is already being done by software in many areas of healthcare.

So, back to the age old question: Can medical record abstraction be automated?  Yes.

Should it be? From experience we know that automating the collection, separation, classification, and abstraction of data can be done. In our opinion, the best practice we have seen is software beginning the process by collecting and organizing the documents and then suggesting to a human abstractor what should be abstracted and the human abstractor approving it.  This allows the power of technology to organize and search through documents at the speed of computer processing power, but then allows the intelligence of the human to act as a quality assurance step, by assuring it is correct or providing the necessary corrections to make it so.