What an IME actually proves about effort, and where grip testing falls short

When a workers comp or disability claim turns on effort, the examiner gets a hard question: did this person give full effort, or hold back? The tools we lean on are weaker than most reports admit. Here is what grip strength coefficient of variation actually proves, and where it stops.

When a workers comp or disability claim turns on effort, the examiner gets handed a hard question: did this person give a full effort, or hold back? A lot rides on the answer. Benefits, settlements, and a person's word all sit on it. But the tools we lean on to answer that question are weaker than most reports admit. Here is what one of the most common ones, grip strength coefficient of variation, actually proves, and where it stops.

The grip CV test, and what the numbers really say

The idea is simple. Ask someone to squeeze a dynamometer several times. A person giving full effort tends to produce a tight cluster of readings. Someone holding back tends to scatter. The coefficient of variation, or CV, puts a single number on that scatter. Higher CV, more scatter, more suspicion of submaximal effort.

The best evidence on how well this works comes from Shechtman's 2001 study in the Journal of Hand Therapy PMID 11511013. It reports the area under the ROC curve at 0.79 for three trials and 0.87 for five. At the common 15 percent cut-off, three trials give a sensitivity of 0.55 and a specificity of 0.94.

Read those two numbers slowly, because they point in opposite directions.

Specificity 0.94 means that when the test flags someone as inconsistent, it is usually right. Very few full-effort people scatter that much by chance. So a positive result is a fair rule-in. It is real evidence, not proof, that effort was submaximal.

Sensitivity 0.55 is the other half, and it is the half that gets left out of reports. The test misses almost half of the people who are actually holding back. A clean, consistent result clears nobody. It is a weak rule-out, because plenty of submaximal efforts look perfectly steady on a grip dynamometer.

One more thing, since I chased it down myself. A figure that floats around the reviews, specificity 0.74 at the 15 percent cut-off, is wrong. It comes from mixing two different operating points in the original paper. The real 15 percent specificity is 0.94, which is a harder and more honest bar. A biostatistician flagged the error to me, I bought the full text to settle it, and the tables settle it.

One test is not a verdict

Grip CV earns a place in the file. It does not earn the last word. Used honestly it is a rule-in with teeth and a rule-out with none. An examiner who writes "effort was consistent, therefore the claim is valid" has overread the tool. An examiner who writes "grip CV exceeded 15 percent, consistent with submaximal effort, weighed with the rest of the exam" has it right.

Where video scoring comes in, and where it breaks

Lately a lot of tools, mine included, have tried to read effort straight from video. Film the movement, track the joints, score the consistency. It is a good idea and I am building toward it. It is also easy to oversell, so I want to be plain about the limits.

A single phone camera has a geometry problem that no software fixes. From the side it under-reads how far a joint actually travels. From the front it is worse. Put the same clip against 3D motion-capture ground truth and the range of motion a single camera reports can be off by tens of degrees, with the rep count drifting along with it.

That matters for one blunt reason: the moment you drop a movement number into a report that a defense expert can pull up and check against a lab, that number has to survive the check, and a lot of them will not.

There is a quieter trap underneath it. On some movements a single camera reports a consistency gap the body never produced. Sloppy form projects into a flat 2D image more erratically than clean form, so the camera shows scatter that the true 3D motion does not. I tested whether a per-setup calibration removes that artifact. It does not. Calibration can fix the level of a number, not the scatter a camera invents.

I am saying this out loud because the alternative, a tool that quietly prints a consistency score it cannot defend, is how a whole field loses its credibility in one deposition. A single-camera effort score is not defensible today. A second camera, a depth signal, or a capture gate that throws out the angles that read badly, those are the honest routes, and that is where the work goes next.

What to take from this if you order or rely on effort findings