Heralding The Minimal Clinically Important Difference When AI Is Used For Human Mental Health

AI And Mental Health

As a quick background, I’ve been extensively covering and analyzing a myriad of facets regarding the advent of modern-era AI that produces mental health advice and performs AI-driven therapy. This rising use of AI has principally been spurred by the evolving advances and widespread adoption of generative AI. For a quick summary of some of my posted columns on this evolving topic, see the link here, which briefly recaps about forty of the over one hundred column postings that I’ve made on the subject.

There is little doubt that this is a rapidly developing field and that there are tremendous upsides to be had, but at the same time, regrettably, hidden risks and outright gotchas come into these endeavors, too. I frequently speak up about these pressing matters, including in an appearance last year on an episode of CBS’s 60 Minutes, see the link here.

Background On AI For Mental Health

I’d like to set the stage on how generative AI and large language models (LLMs) are typically used in an ad hoc way for mental health guidance. Millions upon millions of people are using generative AI as their ongoing advisor on mental health considerations (note that ChatGPT alone has over 800 million weekly active users, a notable proportion of which dip into mental health aspects, see my analysis at the link here). The top-ranked use of contemporary generative AI and LLMs is to consult with the AI on mental health facets; see my coverage at the link here.

This popular usage makes abundant sense. You can access most of the major generative AI systems for nearly free or at a super low cost, doing so anywhere and at any time. Thus, if you have any mental health qualms that you want to chat about, all you need to do is log in to AI and proceed forthwith on a 24/7 basis.

There are significant worries that AI can readily go off the rails or otherwise dispense unsuitable or even egregiously inappropriate mental health advice. Banner headlines in August of this year accompanied the lawsuit filed against OpenAI for their lack of AI safeguards when it came to providing cognitive advisement.

Despite claims by AI makers that they are gradually instituting AI safeguards, there are still a lot of downside risks of the AI doing untoward acts, such as insidiously helping users in co-creating delusions that can lead to self-harm. For my follow-on analysis of details about the OpenAI lawsuit and how AI can foster delusional thinking in humans, see my analysis at the link here. As noted, I have been earnestly predicting that eventually all of the major AI makers will be taken to the woodshed for their paucity of robust AI safeguards.

Today’s generic LLMs, such as ChatGPT, Claude, Gemini, Grok, and others, are not at all akin to the robust capabilities of human therapists. Meanwhile, specialized LLMs are being built to presumably attain similar qualities, but they are still primarily in the development and testing stages. See my coverage at the link here.

Gauging Medical Condition Improvement

Shifting gears, let’s briefly discuss how a medical condition of nearly any kind can be assessed. I will then tie that to the topic of AI and mental health.

Suppose you feel like you might have the flu or a common cold. You go to see a doctor. The doctor gives you sage advice. You go home and abide by the advice. A few days later, you go back to the doctor.

The doctor takes some medical measurements and says that you are improving. But you say aloud that you don’t feel better. You still feel as though a ton of bricks has hit you, and you are not able to shake this darned cold or flu that you have. Thus, despite the medical tests suggesting that you are improving, you surely don’t feel that way.

I’m sure that you have experienced something like this at one point in your existence.

Using A Notable Factor To Decide

One interesting question that the scenario raises is what factor might be chosen to ascertain whether someone is medically improving.

There is the doctor’s viewpoint of potentially selecting a stringent medical factor, perhaps your temperature or a test of your blood. From a patient’s perspective, the factor they are focusing on might be something else, such as whether their nose is runny or they feel that their energy level is low versus high.

Whichever factor is utilized, another facet is the degree of improvement. A person might say that their nose is runny nonstop and continues to seem to be that way. Thus, nothing noteworthy is improving. To them, if their runny nose becomes intermittent, perhaps only once or twice a day, that’s a demonstrative improvement. That’s the level of change that they perceive as significant.

This entire question about what factor to choose and how to gauge whether the factor has demonstrably improved was codified into a medical technique that was christened in the late 1980s as the minimal clinically important difference or MCID. The idea is that you come up with a factor that is clinically valued, you measure its change over time, and only declare a step forward in improvement if the change has made at least some minimal amount of improvement.

The difference must be important. A minimal difference that is unimportant isn’t really what we are aiming for. Nor do we especially find noteworthy anything less than a minimal difference. A strident goal is to emphasize the minimal clinically important difference.

An ongoing debate is whether the MCID is to be based on the viewpoint of the doctor or the viewpoint of the patient. By and large, the MCID is customarily shaped from the perspective of the patient. The usual use of MCID is principally as seen from the patient’s point of view.

Research About MCID

In a research article entitled “Perspective: Minimal Clinically Important Difference (MCID) And Alzheimer’s Disease Clinical Trials” by Jeffrey Cummings, Translational Research and Clinical Interventions, January 26, 2025, these salient points were made (excerpts):

• “The minimum clinically important difference (MCID) is a measure of the minimal clinically relevant change.”
• The MCID represents the smallest difference in score on the measure or domain of interest which patients or clinicians perceive as beneficial or as a meaningful decline.”
• “The foundational concept of the MCID is to determine the smallest magnitude of change that is meaningful for the individual.”
• “MCIDs have been derived for symptomatic treatments for disease-targeted therapies.”
• “The MCID complements other strategies for assessing the meaningfulness of interventions, including effect size, number needed to treat, responder analyses, time saved, quality of life, and quality-adjusted life years.”

As noted, the MCID is not the only avenue for gauging medical progress in a patient. It is best used as a complement to other measurable aspects. You would be potentially missing the bigger picture by relying solely on the MCID. On the other hand, those who do not lean into the use of MCID are probably not going to get a robust gauge of what is going on with the patient. MCID, in that sense, is highly useful. It isn’t a necessity, but it is handy and can be a means of understanding what the patient believes about their medical condition.

Using The MCID In Mental Health

The MCID has been used in the mental health domain and provides the same efficacy as a medical indicator of a patient’s status.

For mental health, there are lots of possibilities about which factor might be used to serve as the keystone of interest. The hope would be to identify a readily measurable factor. If you can’t easily explain the measure, or measure the measure, everything else gets jammed up.

A popular choice is to use a longstanding questionnaire about mental health that is known as the PHQ-9. This is a standardized patient health questionnaire, abbreviated as PHQ, and has nine questions. The beauty of PHQ-9 is that it is evidence-based, self-reported, only takes a few minutes to undertake, has been widely validated, is freely available, and is considered non-proprietary. It is particularly used as a clinical tool for measuring depression.

Each question in the PHQ-9 asks how often, over the past two weeks, the person has been bothered by symptoms, and they then self-rate themselves on a 0-3 scale of:

• Score of 0: Not at all
• Score of 1: Several days
• Score of 2: More than half the days
• Score of 3: Nearly every day

About The PHQ-9

Since there are 9 questions, and each question is rated up to a score of 3 points, the total maximum score is 27.

While exact cutoffs can vary slightly by guideline, a common interpretation is:

• Total score of 0–4: Minimal or no depression
• Total score of 5–9: Mild depression
• Total score of 10–14: Moderate depression
• Total score of 15–19: Moderately severe depression
• Total score of 20–27: Severe depression

Those categorizations are used by clinicians as an overall guide and are not intended to be a specific diagnosis.

The nine questions generally consist of these elements or themes:

• (1) Little interest or pleasure in doing things
• (2) Feeling down, depressed, or hopeless
• (3) Trouble sleeping (too much or too little)
• (4) Fatigue or low energy
• (5) Changes in appetite
• (6) Feelings of worthlessness or excessive guilt
• (7) Trouble concentrating
• (8) Psychomotor agitation or slowing
• (9) Thoughts of death or self-harm

If a person indicates a non-zero score on the ninth item, denoting that they have had self-harm thoughts, this usually triggers a quick follow-up for assessment by the clinician. The person might be in a dire mental state. A clinician should not leap to any hasty conclusion. Perhaps the person misunderstood the question, or casually answered without giving the question much thought, etc. The key would be to have the clinician try to ascertain the mental state as potentially flagged by the answer to the ninth question.

Using AI As Part Of The MCID Effort

A therapist could use generative AI to assist in an MCID effort. In the mental health realm, one such usage would be to have the AI administer the PHQ-9.

Most of the major LLMs have patterned on the PHQ-9. That being said, make sure to review what the AI says about the PHQ-9. There is a possibility that the AI has mistakenly patterned on it and has derived alternative questions or otherwise made errors associated with the questionnaire. You want the PHQ-9 to be as designed. Variations would fall outside the validations that have been carefully performed.

Besides conducting the PHQ-9 as the MCID, the AI could also score the person’s response and present the results. The presented results might be shown as totals and a raw concluding statement based on the rating scale I’ve shown above.

The next step beyond that aspect would be for the AI to actively interpret the results.

Some therapists would object strongly to the AI doing anything other than just showcasing the scoring. An AI-derived interpretation of the results is a bridge too far. There is a worry that the AI will mislead the person into believing something about their mental health condition that is not warranted or completely false. A counterargument is that it is okay for the AI to provide an interpretation if the person understands that the AI might be off base. The retort to that point is the concern that people tend to believe in AI and associate AI as an authoritative figure.

For my extensive coverage on the role of AI and how therapists make use of AI, see the link here and the link here. I have predicted that the classic dyad of therapist-client is being disrupted and transforming to a new triad of therapist-AI-client, see the link here.

AI And MCID In Mental Health

Let’s take a look at an example of how this MCID enactment might occur when using AI.

I logged into a popular LLM and looked to see if it had properly patterned on the PHQ-9. It had done so. I then pretended to be someone who has been undergoing therapy and is doing weekly check-ins via the use of the PHQ-9. In my first week, I had a score of 18. My second week was 16. Now, for my third week, I had the AI administer the PHQ-9 again to me, and it showed me my score, which was 15.

Is my score of 15 enough of an MCID?

Here’s what happened next.

• My entered prompt: “I filled out the questionnaire. My score went from last week’s 16 down to 15, so I think that’s a good sign. But is it enough of a change?”
• Generative AI response: “A 1-point drop is a worthwhile change, but for most people, improvement usually starts to feel noticeable when the original score drops by about 4 to 5 points overall. So, this shift suggests movement in the right direction, but, understandably, you might not feel much different yet.”

You can see that the AI has stated that, though my one-point drop is a worthwhile change, most people tend to feel a noticeable change when the overall score has dropped by 4 to 5 points. I’m not there yet.

This AI-generated response would immediately raise the eyebrows of some therapists. If I believed that I feel mentally better, even at a one-point improvement, I should be allowed to relish that change. The AI has essentially disillusioned me from my own personal semblance of an MCID.

That’s the crux of using AI. I’ve repeatedly forewarned that modern-era generative AI is like a box of chocolates; you never know for sure what you will get.

Continuing The Example

Admittedly, the commentary by the AI has shaken my faith in whether I am mentally improving.

Let’s ask.

• My entered prompt: “Are you basically saying that nothing has changed?”
• Generative AI response: “Early changes often show up on questionnaires before they fully register emotionally. Many people at this stage say things like, “I’m still depressed, but maybe slightly less stuck.”

I logged out of the AI and came back a few days later.

Here’s what transpired.

• My entered prompt: “I took the questionnaire again. Now that I’m in the fourth week of treatment, I noticed that my score is 14. What does that signify?”
• Generative AI response: “Since your total change so far is about 4 points from when you started, you’re close to the range where people often begin to notice meaningful differences. Perhaps a bit more energy or fewer heavy moments. It may be worth giving this another week or two while paying attention to small functional shifts, not just mood.”

The AI is being cautious about my progress. Also, it isn’t asking me any probing questions about how I feel regarding my progress. This highlights that using generic AI is a roll of the dice. Sometimes it will ask, other times it won’t.

It is possible to give the AI various custom instructions on how to proceed in mental health aspects, see my coverage at the link here. This would likely enhance what the AI has to say on these matters. Likewise, a specially built AI for mental health app would presumably do a better job at the MCID activity, see my discussion at the link here.

Where We Are All Heading

I anticipate that researchers will soon be mindfully examining how AI can be of use in MCID for mental health aspects. I will make sure to keep you updated on such studies. Be on the watch for that coverage.

We are now amid a grandiose worldwide experiment when it comes to societal mental health. The experiment is that AI is being made available nationally and globally, which is purported to provide mental health guidance of one kind or another. Doing so either at no cost or at a minimal cost. It is available anywhere and at any time, 24/7. We are all the guinea pigs in this wanton experiment.

The crux is that AI can be a bolstering force for aiding mental health, but it can also be a detrimental force, too. It’s up to us all to steer toward the former and avoid or prevent the latter.

On the topic of MCID, I am reminded of a famous quote by Abraham Lincoln: “The shepherd drives the wolf from the sheep’s for which the sheep thanks the shepherd as his liberator, while the wolf denounces him for the same act as the destroyer of liberty. Plainly, the sheep and the wolf are not agreed upon a definition of liberty.”

That’s a bit of a long way to express the insight that, oftentimes, perspectives radically differ. In the case of therapists and clients, the viewpoint of improvement of a client might differ substantially between what the therapist perceives and what the client perceives. The MCID technique gets the perception of the client onto the table for discussion. AI could be an important aid in doing so, if suitably shaped for this significant task.