Tuesday, March 24, 2009

Speech Recognition and MT Compensation

Speech Recognition and its relationship to compensation took on a life of its own over at the MTChat message board in this thread titled MT Exchange: MTs and "Speech Wreck". There were strong words and a concerted attack on Julie Weight....Yikes! The confusion that ensued linking and even blaming a technology with poor business practices and in particular poor compensation models that appeared to be unfair missed the point.

But it was the posting by Jay Vance of XY Files in an MT World who posted a thoughtful response to some of the criticism being leveled at Speech Recognition in this posting "Is Speech Rec Wrecked" that even featured actual data (thanks for sharing this!) from a survey he conducted in 2006 of Speech Recognition editors. In fact the data presented was helpful in assessing the actual benefits (back in 2006 - a long time ago in technology terms!) that even then showed:
a total of 51% of respondents - saw an average increase in productivity of between 25% and 50%. This confirms the anecdotal information I had collected via informal conversations with MTs working as SR editors in a variety of situations on a variety of SRT platforms.
I don't think it is a stretch to assume that this must have gotten better and productivity has improved beyond this and for a greater proportion of editors. The survey included some review of compensation changes (there was a reduction in rate but hard to determine if this was a real reduction or represented a reduction in rate that was offset by increased productivity) and a final question on satisfaction with the technology:
31% said they were somewhat satisfied
26% said they were very satisfied. These two categories totaled 57%
Not great but better than average. Overall
there is a wide spectrum in terms of the impact of SRT on productivity, compensation, and overall satisfaction among MTs working as SR editors. Consequently, I don't believe there is enough objective evidence to conclude that speech recognition has proven to be a widespread disaster for the MT working class. As with any scenario involving people, technology, and money, mileage is going to vary widely. In my experience, there are simply too many factors that can influence productivity, compensation, and overall satisfaction with speech recognition technology to draw hard and fast conclusions about the impact SRT is having on working MTs on the whole.
And this was in part the point that Julie Weight was trying to make on the MTChat board - there are many factors and there is no use trying to stall the implementation of Speech technology - that trains has left, like outsourcing.

Both Jay and Julie make the point that this technology is in use and although I probably am a stronger advocate and believer in the Speech technology I think the overriding point here is that this can and should be a good thing for the industry. Reducing the labor intensive element of producing a report has to be a good thing....freeing up the medical editor to add value to the clinical information as part of the process of review, editing and validation.

Recognizing this is old data this gives us a good reason to update this information and there is a survey currently ongoing from MTIA that can be taken here and I would encourage you to participate. This is an extensive survey and needs input but if you don't have the time I put a 4 question survey here that. If you can spare the time please take the full survey, but if not I'd welcome hearing your responses.


Bob Mahan said...

The fundamental problem with paying Medical Language Specialists (MLS) by produced line is that most of the risk is borne by the MLS. The pay-by-line compensation model assumes that each MLS has equivalent access to jobs in terms of difficulty and recurrence. When change is introduced MLS productivity (and therefore compensation) usually declines although it may recover or even increase as in the case of speech recognition. MLS reluctantly accept the burden of temporary changes such as new accounts, dictator churn, new software, etc. However, Speech Recognition (SR) introduces a far more pervasive change.

In the pay-by-line compensation model the SR line rate is determined with reference to the transcription (TR) line rate. The SR rate is generally a fraction of the TR rate determined by the group’s average productivity lift of SR relative to TR. If the position of each MLS on the bell curve of TR productivity persists to the SR productivity bell curve then there would be few disgruntled MLS performing SR editing. However, the change introduced by SR is more complex.

A significant portion of low performing MLS, who reside on the left side of the TR productivity bell curve, move further to the right on the SR bell curve. Likewise, a significant portion of MLS on the right side of the TR productivity bell curve will move to the left. The permanent nature of these migrations represent a challenge to the overly simplistic approach to benchmarking group SR line rates to group TR line rates.

Even though the MLS group taken as a whole may, in fact, be whole financially, there will be many MLS making more in SR than previously in TR even though they may not represent the highest SR producers. Conversely many MLS will lose ground in SR even though they may be producing at an above average level. The problem lies in the assumption that each MLS will maintain his or her position on the productivity curve after moving to SR.

Although this type of risk has historically been shouldered primarily by the MLS, the introduction of SR has a much greater impact than any previous changes since the introduction of the word processor.

Nick van Terheyden, MD said...

Thanks Bob - these are great points and important to consider as part of the process of moving to a new technology.
There must be ways to mitigate the risk while transitioning to SR and allow for the variation between MLS.