Cut-off Scores vs Norms
Two test scoring and interpretation methods are particularly common: One uses cut-off scores based on the sensitivity/selectivity of test to determine those cut-off points.
Cut-Off Points
For example, if a patient’s total MMSE score is below the cut score 27/30 MMSE items passed, that indicates a deficit, vs. a total score of 27 or better is considered within normal limits (WNL). The cut-off point is determined by finding the score which yields the best sensitivity to a criterion (successfully picking out who has dementia, true positive yield) vs. selectivity (successfully selecting who doesn’t have it, true negative yield). No test is perfect, and setting a test’s cut-off score to correctly capture the most people with dementia (true positives) means also capturing many people without it (false negatives), and when one goes up the other goes down. A whole section, here, is dedicated to describing this test interpretation technique which is most common in medical settings. An alternative method does not use cut-off points at all, but instead compares the test to the scores of other people who are as similar as possible to the testee; i.e., to a normal sampling of people. This normative methods is often considered better, but there is debate about that, challenging this belief. What’s important is they both work successfully to differentiate people with and without disease conditions.
Norms
Regarding the normative method, we’ll continue to use the MMSE as an example. We use the MMSE, but ignore the cut-off score which applies to all people, young and old. It would be more useful if it had different cut-off scores at different ages as 18 year olds are expected to get different results than 90 year olds. Luckily, an intensive study was conducted by Crum and her colleagues in the 1970s of thousands of people of all ages, and ages groups were formed according to who scored the most similarly, like 20-30 years olds, 31 to 50 year olds, etc. The results were published in JAMA and are available to all on PubMed.
Our computer program avoids cut-offs and compares the patient’s MMSE total score to tens of thousands in Crum’s normative sample. She and her colleagues then statistically studied and found that two factors accounted for deficit vs. WNL patients; the patient’s age and their educational level. Amazingly, criteria like sex or socio-economic status showed little influence on MMSE scores, whereas the older a patient was the lower the total score should be to be WNL which was expected. However, consistent with other research, a patient’s educational level was found predictive of cognitive impairment. Eighty year olds with college degrees needed a much higher MMSE score than grammar school graduates to fall WNL. So our use of the MMSE is arguable better that the cut-off score model making it a real test, as opposed to the screening measure it is when used quickly with only a cut-off score in a PCP’s office. This is also why we continue to use the MMSE rather than the new Montreal Test of Cognitive Ability (MoCA) which requires formal test administration training and certification for non-neuropsychologists by its author.
