Can AI Help With Medication Management for Treatment-Resistant Depression?

The word "treatment-resistant" carries more clinical weight than the term lets on.

For the person carrying that label, it often means years of trying medications that did not work, or stopped working, or worked enough to function but not enough to feel well. It means appointments where the chart says "treatment failure" before the prescriber has asked whether the dose was ever therapeutic or whether the trial actually ran long enough to be a trial.

The monitoring structure around psychiatric medication has not historically been designed to catch those distinctions. It is designed around scheduled appointments, which means the information available to make better treatment decisions is limited to what a patient can reconstruct and report in a 15-minute follow-up visit. AI-assisted medication management changes what is visible between those appointments, and that is where it matters most for patients with treatment-resistant depression.

What Treatment-Resistant Depression Actually Means

The clinical definition of treatment-resistant depression is more precise than the term suggests. According to Cleveland Clinic and multiple clinical consensus frameworks, TRD is defined as major depressive disorder that has not responded to at least two different antidepressants at adequate doses and for adequate duration, which means a minimum of six to eight weeks each. Both conditions must be met. A trial that ended at week four due to unmanaged side effects, or one that stayed at the starting dose without escalation, does not meet the clinical threshold for an adequate trial.

The STAR*D trial, the largest real-world study of antidepressant treatment sequencing, documented that remission rates decline significantly with each additional treatment step: approximately 37% in the first step, with substantially lower rates at each subsequent step. This means that every unnecessary cycle through a medication that was not adequately dosed or not given adequate time increases the clinical difficulty of the next attempt.

A significant proportion of patients who arrive at a new clinical platform carrying a TRD label have not actually completed adequate trials. The American Academy of Family Physicians clinical guidance on TRD specifically notes that before determining a patient is nonresponsive, the prescriber should confirm the accuracy of the diagnosis, medication adherence, and whether the depression is being worsened by coexisting conditions. In standard care, this confirmation often does not happen because the information required to make it is not available.

The Monitoring Gap That Creates Apparent Resistance

Most antidepressant treatment failure happens in the gaps between appointments.

Side effects emerge in the first two weeks, before therapeutic benefit appears. Without someone monitoring in real time, patients stop their medication and never tell their prescriber because the next appointment is eight weeks away. The prescriber records the trial as complete. The medication gets labeled as ineffective. The clinical picture moves one step toward treatment resistance.

Research reviewed in the Primary Care Companion to the Journal of Clinical Psychiatry documented that approximately 50% of psychiatric patients discontinue antidepressant therapy prematurely within six months of initiation, driven by side effects, misperceptions about medication, and inadequate prescriber follow-up. Many of these discontinuations happen before an adequate trial is completed.

Dose adequacy is a related problem. Starting doses are calibrated for tolerability, not for therapeutic effect. The clinical expectation is that doses will be titrated upward as tolerated. In practice, without between-visit monitoring data to flag incomplete response, doses often stay at the starting level through multiple appointments. A patient on sertraline 50 mg for six months who has partial response is not in a failed trial. They are in an undertreated one.

AI-assisted medication management addresses both of these failure modes by making what happens between appointments visible to the prescriber. Not as a patient's reconstructed verbal account, but as structured, time-stamped data on adherence, symptom trajectory, side effects, and functional indicators that can be reviewed continuously.

What AI Can Actually Do in Medication Management

The clinical applications of AI in psychiatric medication management are more specific and more modest than popular coverage suggests. They operate at the intersection of data structure and clinical oversight.

Structured intake. A well-designed AI intake systematically captures symptom history, prior medication trials, adherence patterns, side effect experiences, comorbid conditions, and treatment goals in a format that a prescriber can review efficiently before making any clinical decision. This is different from a patient verbal summary in every clinically relevant dimension: it is standardized, complete, and reviewable at depth.

Between-visit monitoring. Daily or frequent check-ins capture mood trajectory, sleep quality, side effect emergence, and medication adherence on a timeline that matches the actual pace of antidepressant pharmacology. A prescriber reviewing this data can see whether a partial response is developing, whether side effects are persisting beyond the typical window, or whether adherence gaps are occurring that would undermine the trial before it reaches adequate duration.

The American Psychiatric Association's clinical guidance on AI applications in psychiatry identifies medication adherence monitoring, pattern recognition for adverse effects, and clinical decision support for treatment selection as the highest-evidence areas of current AI application in psychiatric care.

Pattern recognition. AI can identify signals in longitudinal data that are difficult to detect in isolated appointments: the sleep disruption that typically precedes mood decline by several days, the partial response plateau that suggests dose optimization before switching, the adherence pattern that explains apparent non-response.

What the Clinical Evidence Shows

The research base for AI-assisted clinical decision support in depression treatment has strengthened meaningfully.

A 2025 cluster randomized trial published in the Journal of Clinical Psychiatry by Benrimoh and colleagues evaluated the AID-ME (Artificial Intelligence in Depression: Medication Enhancement) system, a deep learning-enabled clinical decision support platform for personalized depression treatment selection and management. The trial found that AI-assisted treatment selection improved outcomes compared to usual care in adult patients with major depressive disorder.

A companion paper in npj Mental Health Research described the treatment prediction model underlying the system, which predicted probabilities of remission across multiple antidepressants using a dataset of over 9,000 participants from clinical trials. The model was explicitly designed to surface differential treatment probabilities for clinicians to consider, not to replace prescriber judgment.

Earlier work on the PETRUSHKA AI tool, evaluated in UK NHS settings, demonstrated that patients whose antidepressant was selected using AI-assisted personalization were significantly more likely to continue treatment and experience lasting symptom improvement than those in usual care. The mechanism was treatment fit: matching medication to patient profile from the start reduced early discontinuation driven by poor tolerability or inadequate efficacy.

The common thread across this evidence: AI improves outcomes not by replacing clinical judgment but by giving clinical judgment better data to work with. Remission probabilities across medications, adherence patterns, side effect trajectories, and symptom response timelines that would not be visible in standard quarterly appointments.

What AI Cannot Do, and Why That Matters

The limits of AI in psychiatric medication management are not details. They are the framework within which any responsible clinical use of these tools must operate.

The APA's position on AI prescribing is explicit: AI should be understood as augmented intelligence that supports, rather than replaces, clinician judgment. The APA emphasizes that psychiatric prescribing depends on nuanced assessment of mood, cognition, and social context that is difficult for automated systems to replicate, and that medication decisions require the longitudinal clinical relationship and professional accountability that only a licensed prescriber provides.

For patients with treatment-resistant depression specifically, this matters in additional ways. TRD presentations often involve diagnostic complexity: comorbid conditions that were not identified, bipolar spectrum presentations that respond differently to antidepressants than unipolar depression, psychosocial contributors that require clinical evaluation, and medical contributors including thyroid dysfunction, vitamin deficiencies, and sleep disorders that can produce depressive symptoms without responding to antidepressant monotherapy.

AI can flag patterns that warrant clinical attention. It cannot evaluate what those patterns mean with the depth a clinician brings to the question. Advanced TRD treatments, including esketamine nasal spray, transcranial magnetic stimulation, and electroconvulsive therapy, require in-person specialized clinical assessment. These interventions are outside the scope of any AI-assisted monitoring platform and require direct prescriber involvement that goes beyond what digital care models currently provide.

If you are in a mental health crisis or experiencing thoughts of self-harm or suicide, call 988 or go to your nearest emergency room. AI-assisted monitoring is not a crisis intervention tool.

How SiggyMD Approaches Medication Management for Complex Cases

For patients arriving with a history of prior antidepressant trials, including those carrying a treatment-resistant label, the first clinical question is whether those prior trials were adequate. The structured AI intake captures medication history, doses, duration, side effect experience, and reasons for discontinuation in a format that makes this assessment possible before the first prescriber review.

After intake, daily check-ins provide the longitudinal data that closes the monitoring gap. When a patient's check-in trajectory shows declining symptom scores despite reported adherence, the prescriber sees this in real time, not at a quarterly appointment after the decline has compounded. When partial response appears, the data supports a dose optimization discussion before switching medications becomes the default option.

"A significant fraction of patients who have been told they are treatment-resistant actually have not had adequate trials," says Daniel Montville, MD, Psychiatrist at SiggyMD. "The label gets applied when the monitoring structure fails, not necessarily when the biology fails. Better between-visit data changes what I can see and what I can do. Most of the time, the first question is not which medication to try next. It is whether the current medication was actually given a real chance."

What Members Are Saying

Member stories reflect real experiences. Names and identifying details have been changed to protect privacy. Results vary. SiggyMD is currently invite-only.

Frequently Asked Questions

What Is Treatment-Resistant Depression?

Treatment-resistant depression (TRD) is clinically defined as major depressive disorder that has not responded to at least two different antidepressants at adequate doses and for adequate duration, meaning a minimum of six to eight weeks each for both medications. Approximately 30% of people with major depressive disorder develop TRD by this definition. However, many patients who carry the label have not completed adequate trials: some stopped too early due to side effects, some were on starting doses that were never escalated, and some had no monitoring structure to catch these gaps.

Can AI Prescribe Antidepressants or Diagnose Depression?

No. AI does not prescribe medications or diagnose conditions. The American Psychiatric Association classifies AI as augmented intelligence: a tool that supports clinician decision-making, not a replacement for it. In a clinically supervised platform, AI provides structured intake data, continuous monitoring data, and pattern recognition that a licensed prescriber reviews before making any clinical decision. Nothing moves forward without prescriber review and approval. The AI informs the prescriber's judgment. The prescriber makes the clinical decision.

How Does AI Help With Medication Adherence for Depression?

AI-assisted monitoring captures daily medication adherence through check-ins, identifies patterns of inconsistent dosing that may explain apparent non-response, and provides this data to the prescriber in a reviewable format. When adherence gaps appear, the prescriber can address them directly rather than making treatment decisions based on incomplete trial data. For patients with treatment-resistant depression, distinguishing between medication failure and adherence-related under-treatment is clinically critical because it determines whether the next step is switching medications or completing the current trial properly.

What Are the Limits of AI in Treating Treatment-Resistant Depression?

AI-assisted monitoring is appropriate for standard SSRI and SNRI management with continuous between-visit data. It is not appropriate for the advanced interventional treatments used in genuine treatment-resistant depression, including esketamine nasal spray, transcranial magnetic stimulation, or electroconvulsive therapy. These require in-person specialized clinical assessment beyond what digital platforms currently provide. For patients with complex comorbidities, significant diagnostic uncertainty, or severe symptom burden, the clinical complexity may require specialist in-person evaluation that goes beyond AI-assisted monitoring.

Does AI-Assisted Medication Management Improve Outcomes?

Clinical evidence is strengthening. A 2025 cluster randomized trial published in the Journal of Clinical Psychiatry found that AI-assisted clinical decision support for depression treatment selection improved outcomes compared to usual care. Research on AI treatment personalization tools has also shown that matching medication to patient profile from the start reduces early discontinuation. The mechanism in both cases is better information at the point of clinical decision-making, not AI acting independently of the clinician.

What Should I Do If I Think I Have Treatment-Resistant Depression?

Start with a clinical review of your prior medication history. A thorough prescriber who reviews the adequacy of your prior trials, whether doses were therapeutic, whether trials ran long enough, and whether adherence was consistent may find that your depression has not been treatment-resistant but treatment-incomplete. If you are experiencing a mental health emergency or thoughts of self-harm, call 988 or go to the nearest emergency room immediately.

Bottom Line

Treatment-resistant depression is a real clinical condition, but the label is applied more broadly than the definition warrants. Many patients who carry it have not completed adequate antidepressant trials, largely because the monitoring structure around psychiatric medication was not designed to catch the gaps where those trials fail.

AI-assisted medication management does not cure TRD. What it does is close the monitoring gap that creates apparent resistance where adequate treatment would otherwise succeed. Structured intake that captures prior trial adequacy, continuous between-visit data that makes side effects and adherence visible in real time, and pattern recognition that surfaces partial response before switching becomes the default answer: these are the specific mechanisms through which AI changes outcomes for patients who have been labeled as difficult to treat.

The prescriber is still the decision-maker. The AI gives the prescriber better data to make that decision with. That distinction is the entire framework for responsible use of these tools in psychiatric care.

Sources

1. Rush AJ, et al. Acute and longer-term outcomes in depressed outpatients requiring one or several treatment steps: a STAR*D report. American Journal of Psychiatry. 2006;163(11):1905-1917.
2. Benrimoh D, et al. Artificial Intelligence in Depression: Medication Enhancement (AID-ME): a cluster randomized trial of a deep learning-enabled clinical decision support system for personalized depression treatment selection and management. Journal of Clinical Psychiatry. 2025;86(3):24m15634.
3. Cleveland Clinic. Treatment-Resistant Depression. Cleveland Clinic. Accessed May 2026.
4. Preskorn SH. Antidepressant Adherence: Are Patients Taking Their Medications? Primary Care Companion to the Journal of Clinical Psychiatry. 2010;12(5).
5. American Psychiatric Association. Applications of Artificial Intelligence in Mental Health Care. APA. Accessed May 2026.
6. American Psychiatric Association. AI Prescribing in Psychiatry. APA. Accessed May 2026.
7. Little A. Treatment-Resistant Depression. American Family Physician. 2009;80(2):167-172.
8. NIMH. Questions and Answers About the STAR*D Study. National Institute of Mental Health. Accessed May 2026.