Billing computer medical software

To succeed in healthcare, new technology must prove itself not only to users, but also to those who pay for it.

"A new technology cannot displace an established technology--with its installed base of plants, equipment, training, personnel, and satisfied customers--unless the innovation is about ten times more cost-effective than its predecessor."

--Peter Drucker

One new technology that shows potential for cost effectiveness in healthcare billing is natural language processing (NLP).

NLP software extracts facts, such as ICD-9 codes, from narrative text that is typically created by transcriptionists working from physician dictation. The results of several formative studies suggest that NLP can improve medical record coding productivity and consistency without sacrificing quality. In fact, commercial NLP products for radiology and emergency medicine are now being sold. But whether the technology works as promised is only part of what is required for commercial success in a mainstream market.

Other factors that will play a role in the rate of adoption and the staying power of NLP include:

* How does NLP affect workflow?

* What mode of clinical data capture (unstructured text or structured/coded) will emerge as most desired by clinicians?

* Are other supporting technologies required for mainstream adoption?

* Can NLP fit into existing healthcare information technology legacy environments without requiring significant human re-engineering and/or costs?

* Is there a natural buyer?

* Will NLP increase revenue or reduce cost? By how much?

* Does NLP address healthcare executives' high priorities?

Three vendors--A-Life Medical, CodeRyte and Paradigm--have made significant investments in NLP software designed to automate various aspects of medical record coding from narrative text. A-Life began in 1996 with an ICD-9 and CPT-4 coding product for emergency medicine and in 2000 expanded its scope to radiology. CodeRyte started in mid-1999, aggressively built an autocoding ICD-9 and CPT-4 product for radiology and has already expanded into cardiology. Paradigm's NLP software, built over a number of years, identifies ICD-9 diagnosis and procedure codes from dictated/transcribed inpatient charts.

NLP and workflow

Since NLP products for radiology and emergency medicine are now emerging in the early adopter market, NLP's end-to-end impact on workflow and workload will soon be apparent. 3M sponsored studies of products from three NLP companies conducted in "laboratory" settings that have demonstrated 30 percent to 50 percent improvement in coder-productivity, reduction in workload reflected in the number of charts that can be coded without human intervention (40 percent to 65 percent), and improved inter-coder consistency with no reduction in coding accuracy. However, what these "laboratory" studies do not tell us is what the environmental, workflow and integration affects of NLP will be in a variety of healthcare settings, factors that will influence the rate of the adoption.

NLP radiology clients of A-Life Medical and CodeRyte are beginning to realize significant workload reductions through NLP batch processing. NLP customers electronically ship their dictated/transcribed ASCII records to the remote service centers where their charts are coded overnight and returned ready to be shipped to third-party payors. Charts that can't be "autocoded" by NLP software are flagged by the system for human review.

Clinical Data Capture

Among those who study the potential of NLP as an autocoding or coding-assist tool for billing, debate exists over the size of the market opportunity, in light of the variation in prerequisite use of dictation/ transcription.

In acute care markets, transcription services may actually be on the rise with transcribed documents representing a small but important percent of the total patient chart. Physician dictation and transcription services are heavily used in many hospitals for pathology, operative, history and physical exam reports as well as discharge summaries. Radiology and emergency medicine are also heavy transcription users.

But in other areas of medicine such as physician offices and clinic settings, handwritten notes are still the norm. In an October 1999 Harris poll, physicians were found to document in the following ways (frequencies):

* Handwritten notes (54.2 percent)

* Dictation/transcription (31.1 percent)

* Speech systems (4.4 percent)

* Computer keyboard (4.3 percent)

* Handheld devices, like PalmPilot (0.2 percent)

* Mixture of other (5.8 percent)

A major assumption made by those investing in NLP is that, in spite of the high cost of transcription services, "free text" or ASCII is not going away and may be on the rise.

Supporting Technologies

Automated transcription of continuous speech has yet to deliver on its promise, but may eventually be a big market driver for NLP by eliminating transcription costs. Without automated speech recognition, NLP vendors will be pinning their hopes on physician preference to talk versus type and that the demand for that preference will outweigh the cost of transcription services (about $10,000 per year per physician at many hospitals). For institutions already absorbing the cost of transcription, NLP software costs may seem relatively small compared to the benefit. Yet the sales process for NLP vendors to institutions accustomed to either handwritten or template-based clinical documentation will likely be more difficult.

Legacy Environments

To optimize their software for a production environment and to enhance its viability for a mainstream coding market, vendors are working to integrate NLP with coding reference tools, coding clinic guidelines, drug databases and legacy hospital, clinical, transcription and billing systems through interfaces. This integration is key for the success of almost any new medical software application.

Technology vendors in the emerging phase of a new product are typically caught in a situation where they need early adopter buyers who will insist that a third-party vendor provide an interface to their new system so that the value of the new technology can be fully realized. For products with substantial and obvious return on investment, this early adopter phase is short lived. It remains to be seen whether NLP vendors, now in this phase, will emerge quickly as a clear choice for a next-generation coding solution.

Natural Buyers

Increased regulatory requirements for coding, coupled with increased fines for inaccurate coding, are creating market pressures to improve existing coding methods and technologies.

A persistent shortage of competent coding professionals also causes a need for increased coder productivity. This should make medical records directors and CFOs motivated buyers. They are also natural buyers because they already have a budget for coding and classification software, at least in the hospital market. NLP vendors will be responsible to demonstrate that the benefits of their software do result in revenue increases by optimizing billing codes for classification (e.g., DRGs, APCs) or through staffing reductions.

Cost Effectiveness

An NLP buyer must consider how the new technology will influence overall workflow and consider what the ripple effect of change will be on people and budgets. The goals of optimizing workflow must include adequacy of documentation, accuracy of coding, minimizing physician time in the documentation process, minimizing cost and optimizing revenue.

The cost of transcription services, the transcription turnaround time, software costs, hardware costs (new computer workstations may be needed to run NLP technology), and interface costs (how will the new technology interface with existing systems) all have to be considered. Testimonials of paying clients will ultimately reveal whether there is a significant return on investment.

High Priorities

The Health Care Advisory Board recently reported that the three biggest concerns for hospital CEOs in recent months were:

1. Staffing shortage (e.g., nursing, clinicians, coders)

2. Shortage of space (e.g., need more operating rooms)

3. Service quality (e.g., patient wait times in the emergency room)

If NLP vendors can demonstrate their technology's ability to relieve medical record coding staffing pressures and provide a cost effective end-to-end billing solution for healthcare providers, mainstream adoption may be just around the corner, provided NLP can scale to other domains and that there will be "acres of ASCII" awaiting them in those domains.

Homer R. Warner, Jr., Ph.D., is business development manager for 3M Health Information Systems, Salt Lake City, UT. Contact him at hrwarner@mmm.com.