Wot's Uh... The Deal With How We Pick Antidepressants and Antipsychotics Pt. 2

Depression is Temporary, Schizophrenia is Forever™

This is the second part of an essay about the selection of antidopaminergics and antidepressants. In Part I, I outlined (what I see as) the prevailing attitude of psychiatrists towards their selection of antidepressants and antidopaminergics, and my uncertainty about the extent to which that was supported by the data. I went through two big meta-analyses (one for each drug class) and talked about the size of individual drug effects and their efficacy relative to one another. By the end I had come to the conclusion that we’re probably too dismissive about antidepressant selection and excessively particular about antidopaminergic selection.

Part II is my attempt to work through why this narrative might exist, add a little more nuance to the data, and see where we end up.

Before I get into this, I just want to remind everyone reading that I am still in training and my personal sample size is relatively small. Whenever you see me talk about my “clinical experience,” add a few handfuls of salt.

Maybe about this much. If you have hypertension, you can use sodium free salt. Alternatively, just don’t worry about it. It’s a metaphor.

Baseline Severity and Treatment Response

I admit I’ve been hiding the ball on this one.

There is some evidence to suggest that baseline symptom severity in schizophrenic individuals predicts treatment response.

You can see this illustrated for the treatment of psychosis with the antidopaminergics in this Furukawa et al. paper, which shows how effect-size increases with a higher baseline PANSS score.

The specific numbers here aren’t super important¹, but the trends as you go from a low to high baseline PANSS certainly are!

I looked around at some other papers and this generally seems to be a real effect. Some papers note that, in particular, a higher baseline positive symptom subscore on the PANSS also increases effect size.

The Huhn et al. supplement includes a sub-analysis which controls for baseline severity by forcing a baseline PANSS of 94:

Some minor shuffling occurs - clozapine gets a bit worse, risperidone looks a bit better - but overall it seems like baseline severity does not change how effective these drugs are relative to one another for a given PANSS. In other words, if this severity-response relationship really does exist, it doesn’t seem like it is unique to a particular antidopaminergic².

Does this apply to the antidepressants?

It seems unclear.

A 2024 review by Tröger et al., analyzes 65 head-to-head RCTs that exclusively compared pharmacotherapy vs. psychotherapy. They found that baseline severity does not moderate treatment efficacy of either modality.

The Cipriani et al. meta-analysis from Part I suggests that there might be an effect here, but:

Studies with patients with more severe depression showed larger effectiveness and less dropout than studies with moderate/low depression for most drugs and for most active interventions… However the heterogeneity remains unchanged once accounting for differences in baseline severity and given the risk of ecological bias we did not consider this evidence conclusive.

Some other studies I found — where pharmacotherapy or psychotherapy are compared to control conditions (i.e. placebo or no therapy) — do show a moderating effect of baseline severity like the antidopaminergics³.

I find this to be totally perplexing, and can’t think of any good reasons why a study being placebo controlled vs. not should modulate whether or not baseline severity matters.

Who Shows Up

I think this baseline severity effect matters a lot when you think about who is showing up for psychiatric treatment.

First, consider the frequency of these illnesses. Prevalence rates for schizophrenia in the US range between 0.25-0.64%. Compare that to the 1 year prevalence for MDD which, according to Kessler et al., is approximately 6.6%.

Second, consider how likely a person with a given CGI-S score for each of our two diagnostic groups is to show up for treatment. (NB: This is based on my experience. Remember the salt.)

The psychotic patients that we come into contact with are typically seriously ill individuals — I would guess rarely lower than a CGI of 5 — and are usually not showing up because they just can’t wait for us to put them in a locked unit and start them on risperidone. Those with less severe symptoms usually aren’t showing up for treatment, either because they don’t think they need it (and their behavior isn’t dangerous enough that anyone can force them to get it), -or- because they’re people who used to be at a CGI ≥5 but have improved with medication and have enough insight to keep going to their appointments and taking their medication.

On the other hand, we see a lot of depressed individuals with CGIs of 3-5. The cultural tide has shifted to the point where it is much more socially acceptable to seek psychiatric help for depression/anxiety, there’s (usually) not a psychotic thought process that would dissuade them from seeking care, and (at least in Boston) clinicians are pretty willing to hospitalize people even for fleeting thoughts of suicide/thoughts that life would be better if they were dead.

If we put these together we can see how they really could distort the perception of how effective these medications are relative to one another, right?

With schizophrenics, psychiatrists are seeing the sickest chunk of a low prevalence illness where baseline severity seems to have a large effect on medication efficacy — so they see big, memorable effects in these patients.

On the flip side, psychiatrists see many more depressed individuals with symptoms of low/moderate severity. These symptoms are treated with antidepressants and their symptoms get 50% better… but that 50% improvement is much harder to notice.

---

Diagnostic Uncertainty

RCTs are not real life. Studies typically go out of their way to recruit patients who are relatively uncomplicated from a diagnostic point of view⁴ and who are most likely to respond to treatment (e.g. the Cipriani paper excluded studies where >20% of the patients were considered “treatment resistant”).

Real life is much messier, and in a way that makes it much harder for the antidepressants to demonstrate efficacy.

Schizophrenia usually doesn’t look much like anything else. Sure, you have drug-induced psychosis, psychotic mania, psychotic depression, etc. but if you’re able to get any sort of long-term history you can usually tease them apart.

Additionally, the role of antidopaminergics in treatment are more clearly defined in psychiatry. We use them to treat psychosis. Yes, there are some more niche applications, like quetiapine for anxiety/depression, adjunctive aripiprazole for MDD, and so on. The point is, though, that most of the time we don’t give aripiprazole to a floridly psychotic individual, see it work (or not!), and then wonder if the patient became less psychotic because we just happened to treat a comorbid depressive disorder.

Treating with the antidepressants is a whole different mess, because the rate of psychiatric comorbidity in depression is very high. Some think it is so high that it means that the entire diagnosis is garbage and meaningless, but that’s a topic for another essay.

This 2020 paper from Steffen et al. used ambulatory claims data from Germany and found that “64% of mild depression cases (72% of moderate, 78% of severe) had a comorbid mental disorder.” I’ve seen how sloppy American physicians are with diagnostic coding, and I don’t want to get myself in trouble by subscribing to cultural stereotypes about German meticulousness and efficiency, so I think this is probably a gross over-estimate⁵.

Thaipisuttikul et al. did a smaller, but more careful analysis of 250 individuals that used a structured interview to diagnose MDD and psychiatric comorbidities. They ended up excluding 60 of them because they turned out to have bipolar disorder (keep this in mind), and of the 190 remaining individuals:

There were 67 patients (35.3%) with one or more psychiatric comorbidities. Comorbidities included dysthymia (19.5%), any anxiety disorders (21.1%) (panic disorder [6.8%], agoraphobia [5.8%], social phobia [3.7%], obsessive–compulsive disorder [OCD] [4.7%], generalized anxiety disorder [5.3%], and post-traumatic stress disorder [4.2%]), alcohol dependence (0.5%), psychotic disorder (1.6%), antisocial personality (1.1%), and eating disorders (0%).

“Hey!” you might say, “don’t we use the antidepressants to treat most of those comorbidities too? And couldn’t someone’s depressive symptoms just be secondary to their PTSD or horrible anxiety? How would anyone be able to tell what exactly they were treating, and whether or not they would expect the patient to get better?”

Exactly.

Here is a list of psychiatric illnesses that the antidepressants are FDA approved for:

• Major depressive disorder

• Dysthymia

• Generalized anxiety disorder

• Social anxiety disorder

• Panic disorder

• Bipolar depression

• PTSD

• OCD

• Fibromyalgia

• Neuropathic pain

• Premenstrual dysphoric disorder

• Bulimia nervosa

There is also a whole list of off-label uses for the antidepressants that include:

• Borderline personality disorder

• Being very angry sometimes

• Cocaine use disorder

• Preventing your psychiatrist from having to deal with the discomfort of explaining to you that no medication will solve the fact that you are in a loveless marriage of 30 years, your children don’t speak to you because you are a bit of a jerk, and you have no friends.

• Ennui

You may have noticed that there is a whole lot of symptom overlap between these different diagnoses.

Unfortunately, the efficacy of the antidepressants for each one varies quite considerably. Relative to MDD, they appear to be even more efficacious in primary anxiety disorders. On the other hand, they appear to perform very poorly in PTSD and bipolar depression (NNT = ~30). The latter one is particularly unfortunate because — unless someone has had a episode of mania/hypomania prior to their first depressive episode — bipolar depression is virtually indistinguishable from MDD.

So, when treatment with antidepressants doesn’t work (or only modestly works), we’re are left uncertain as to why. Is it because our diagnosis was incorrect and we expected too much, or does this patient’s depression just not respond to this particular medication? When it succeeds we’re not much better off. Maybe they were depressed because they were so anxious all the time, and we just treated the anxiety really well? Or maybe it’s…

Spontaneous Remission

If all of this uncertainty wasn’t enough, the manufacturers of depressive disorders decided to include a feature called ‘spontaneous remission’ that is standard to all depressive disorders. It appears uniquely designed to vex regular clinicians desperately trying to understand the efficacy of their interventions. I can’t say that I’m unhappy about the fact that people spontaneously get better, but it would make things a whole lot easier if they didn’t.

Though the manufacturers of MDD have not publicly disclosed the built-in rates of spontaneous remission, dubbing them “trade secrets,” a 2013 consumer report from Whiteford et al. reviewed a collection of 21 wait-list controlled trials and observational cohort studies and found the following rates of remission in untreated depression:

• 3 months: 23%

• 6 months: 32%

• 12 months: 53%

Simon et al. looked at a sample of 545 untreated individuals who met ICD-10 criteria for depressive disorder and found that higher baseline severity predicted a lower remission rate at 12 months, as you might predict. However, you would probably not predict that even in the patients with severe baseline symptoms, 57.8% of them experienced total remission of their symptoms! I’ve included the table below:

Simon et al. 1999 Table 2. - Recog refers to patients who were recognized as symptomatic and treated by their PCPs.

A short-term meta-analysis by Mekonen et al., of 16 waitlist controlled studies with a medication duration of 8 weeks (range: 4-12 weeks) published in 2022 found that:

The overall pooled remission from untreated depression was 12.5%, 95% confidence interval (7.8, 18.0%). Due to lack of published data, we were unable to determine if the severity of depression symptoms was associated with remission rates.

This just doesn’t happen with schizophrenics⁶, at least not in the short-term. I was going to say that I don’t think this happens at all outside of first-episode psychosis —which are a relatively unique population with really high rates of remission. But as I was writing this, Awais Aftab published an excellent piece that does a really nice review of the remission literature in schizophrenia. It looks like remission without antidopaminergics does happen, but is still quite uncommon.

Time to Onset

The antidopaminergics seem to work relatively quickly. This review by Agid et al. suggests that >50% of all the improvement you will get from an antidopaminergic in the first year is seen in the first 4 weeks of treatment, and that the improvement in the first 2 weeks is the largest of any other 2 week period afterwards (i.e. at least 25% of your improvement is seen in the first 14 days).

Previously it was thought that the time to onset was much longer, which was primarily supported by studies (say it with me) in rats, that showed a “depolarization block” in dopaminergic neurons only after 3 weeks of continuous treatment. Apparently this was just a cool, but irrelevant finding.

My clinical experience⁷ has matched up with the Agid paper’s timeline - I’ve seen many psychotic patients respond quite substantially within the first few days of treatment.

The antidepressants, however, continue to be plagued by the ghost of “4-6 weeks”; essentially this idea that patients on antidepressants shouldn’t expect to see any real effects until 4-6 weeks after starting treatment. This is oft-repeated by PCPs and psychiatrists alike, but doesn’t appear to be true.

Instead, it turns out that there are probably different cohorts of patients that respond at different rates. Uher et al. 2011 did a study with 811 individuals on either escitalopram or nortriptyline over 12 weeks in which they found 9 distinct trajectories that can really be chunked into three categories:

1. People who get worse or don’t improve at all

2. Early responders: People who show improvement within the first 3 weeks

3. Late responders: People who don’t show improvement until 3-4 weeks

Patients seem to be split between these categories relatively evenly; it’s a 40/60 split between early/late responders. So, about 40-50% of patients who don’t show any response at 2 weeks will still go on to have a significant improvement in symptoms in weeks 3-6, and by week 12 the early and late responder groups look fairly similar to one another. You can check out the table for yourself:

Interestingly, this study also found that there was a difference in the trajectories of patients depending on if they received nortriptyline or escitalopram. There were about 3x as many individuals with “delayed complete remission” trajectories in the escitalopram group as there were in the nortriptyline group⁸.

You’ll see how this matters when we talk about…

How We Manage The Severe Cases

In my experience, the severely depressed patient is rarely⁹ just put on a SSRI/SNRI and watched in an inpatient unit for 4-6 weeks to see how they do. They might get 7-14 days, but then people get antsy. Sometimes the people getting antsy are us, sometimes they’re the case managers telling us that insurance isn’t going to cover any more inpatient days and can you please stop setting the hospital’s money on fire thank you.

When it’s case management, patients are started on something and discharged unless we think they will literally either die or kill themselves. If we can’t discharge them because we think that the patient might kill themselves or starve/dehydrate themselves to death, we usually add another medication and/or do ECT.

Either way, we don’t give ourselves the chance to see the big changes in depressive symptoms from these medications; forget thinking that we might get the chance to do it with enough different kinds of antidepressants to start differentiating them.

Severe cases of schizophrenia tend to be handled differently. ECT is not often used unless patients are catatonic or not eating/drinking. Usually we just rely on oral antidopaminergics (or IMs if they’re often assaultive/threatening/court mandated) and wait for them to improve. Often there is some adjunctive benzodiazepine usage, but there’s really only one intervention being used to treat the psychosis per se at any given time.

Depression is Temporary, Schizophrenia is Forever™ (or at least a pretty long time)

The difference between these two approaches are easily explained — I think — by how we think about the prognoses of these two illnesses.

Depression is episodic, and our goal of treatment is to make it go away… or at least make the symptoms less severe while we wait for it to go away. If you remember the rates of spontaneous remission above, this is actually an OK strategy for most patients!

Schizophrenia, unfortunately, is (mostly) forever. ECT might be effective, but everyone knows it’s a temporary measure. Finding the “right” antidopaminergic feels like the only thing we can do to improve/stabilize these patients to try and return them to some level of independence in the long-term.

Time + Management

Hopefully you can see how these differences come together yet again to lessen our perception and certainty of the impact of the antidepressants relative to the antidopaminergics.

The most symptomatic patients are in whom we’re most likely to see a dramatic difference. We get to see that in schizophrenics because we’re typically using a single type of intervention that works substantially within the duration of a hospitalization. We don’t do this in the severely depressed because we’re not keeping them around long enough to see major improvements, and/or we’re using multiple interventions to treat their symptoms that cloud the picture.

---

A Better Narrative?

When schizophrenic patients show up for treatment, there is not usually much question about whether or not they are schizophrenic. They tend to be quite symptomatic in a way that is very visible and unusual. Fortunately, the sicker they are, the bigger effect our medications seem to have. We treat them with medications that work quickly enough that the psychiatrists who are caring for them in the acute phase of their illness can see a substantial difference even during the relatively short time frame of an inpatient stay, and their outpatient providers are also likely to notice a significant change too. Spontaneous remission is not really a thing and usually there’s just a single agent treating the psychosis, so there’s not much uncertainty about why their symptoms improved.

When depressed patients show up for treatment, it is often unclear whether or not they have Major Depressive Disorder or something else that looks quite similar. They are more likely to have symptoms of mild/moderate severity, which may mean that antidepressants are less likely to work well, and definitely means that we are less likely to notice an effect clinically. In the event that they do have severe symptoms, the antidepressants work too slowly for inpatient units to keep them for the full 6 weeks to see full effect and/or they are treated with augmenting agents/alternative treatments that make it difficult for us to be confident in what led to improvement.

This extended duration of improvement is slow enough that we are less likely to notice the change, and when we do it may not feel as large as it is objectively. Even if patients do improve, there is somewhere between a 50 and 90% chance (depending on severity) that their symptoms would have remitted within 12 months even without treatment, so it’s hard to be confident that the medications did anything anyway. If they don’t improve, the diagnostic uncertainty inherent in depression means that it is hard to be confident that the patient is simply treatment resistant; it is often equally as plausible that our diagnosis is incorrect and that our expectations of efficacy should be lowered accordingly.

…

When you put it like that, is it any wonder that our perception is that the antidepressants are such mediocre medications that they aren’t worth spending the time agonizing about?

What’s the Remedy?

Futile Grousing about the DSM-5

You can say a lot of negative things about the construct validity of the diagnosis of schizophrenia, but when you put it next to something like major depressive disorder… it doesn’t look so bad does it?

This is the part where I say that the diagnostic construct of major depression is bad, the field of psychiatry should feel badly about it, and we should probably come up with something better so that we have more diagnostic clarity between patients and we stop doing all of these studies examining the effectiveness of a class of medications for a poorly defined illness. Maybe we just do these studies and track particular symptoms or something. Maybe we switch to a spectrum model like HiTOP. I dunno. Write your local DSM-6 section chair or whatever.

Practical Steps

This next bit is what I’m going to try and do from a practical point of view to make sure I’m paying the right amount of attention to antidepressant drug selection. At the outset, this requires you to have agreed with me based on the data that I presented in Part I that there are probably meaningful differences between the antidepressants in depressed (and anxious) individuals. If I haven’t convinced you of that, well, thanks for continuing to read I guess?

My approach here is to try and make treatment outcomes more legible to myself, in the way that the outcomes in schizophrenia seem to be.

1. Be detailed in my symptom collection and put my money down on a diagnosis

I find that there’s a temptation when I see patients with some mixture of anxious/depressive symptoms to not worry too much about the precise details¹⁰ because the treatment ends up being pretty much the same in the end. It’s not that I couldn’t tell you what I think the most likely diagnosis is in 90% of cases, but it hasn’t felt like the best use of my time to agonize over the details. I mostly just put them on a SSRI and do a sort of CGI at every visit. Sure, that might be what I do at the end of the day, but it harms my ability to calibrate my priors.

Instead, I’m going to make an effort to do more thorough symptom inventories and use them commit to a single primary diagnosis followed by informed guesses about the likelihood of alternative and/or comorbid diagnoses.

2. Systematically track symptom trajectories

Really this just means “use clinical scales consistently.” I already do this to some degree, but not as much as I should. I know that there are a lot of psychiatrists who scoff at clinical scales, but I don’t really understand why. They’re very time efficient and I think probably more thorough than most psychiatrists are in their questioning about particular groups of symptoms.

I’ve already found one instance in which doing this was helpful in adjusting my perception of antidepressant efficacy, so I’ll close this essay with a little vignette. (Note: I’ve changed details to maintain anonymity)

Just in case you forgot about the salt

One patient of mine had been off of antidepressants for a couple years. He had a history of mild/moderate episodes of depression and had tried just about every SSRI/SNRI we had. He was always equivocal about whether or not antidepressants did much for his depressive symptoms. From my point of view, he was a pretty functional individual with some mild mood symptoms that seemed to generally be within the realm of normal for his life situation, but not someone who presented as clinically depressed.

At some point he told me that his symptoms had started to worsen, but he wanted to see whether or not he could turn things around on his own with behavioral changes first. From my point of view, very little had changed about him. Beyond his rather nebulous reports about being less interested in socializing and feeling more down, he certainly didn’t look more depressed in our meetings and continued to function more-or-less the same.

Eventually, after a few months of what he reported was a steady decline (that I couldn’t really perceive), he decided that he would like to give medication another go. I was skeptical that medication would achieve much, but he had put in the effort to try and deal with these symptoms on his own first, which usually makes me more willing to prescribe for people with mild/moderate symptoms.

Given his difficulty with articulating both his symptoms and historical response to antidepressants in a useful manner, I figured that it would be good for both of us if he filled out a Beck Depression Inventory (BDI) prior to starting the medication. His score was a 23, which is right in the middle of the “moderate depression” range (score: 19-29). A bit higher than I would’ve predicted based just on his outward appearance and description of his symptoms. I start him on escitalopram with a plan to get him up to 10mg qDay within a week or so.

Six weeks later we meet. Things are significantly better, he says, though not all the way. He struggles to articulate exactly how so. To my eye he seems maybe a little brighter and perhaps sounds a bit happier in the day-to-day. I would not have guessed that his BDI had dropped to an 11, just one point away from the bottom end of the “mild depression” range (10-18). He tells me that improvement plateaued about 2 weeks ago, and wonders if we can try a higher dose - we go up to 15mg.

Another 6 weeks later, we meet. He feels like he is almost back to normal, though again has a hard time telling me what has shifted. Again, he seems maybe a little bit brighter to me. Again, his BDI has improved down to a 5, below the cutoff for mild depression.

Without the BDI scores, I think my perception of this treatment would’ve been that my patient maybe got mildly depressed and that maybe the escitalopram helped a bit according to the patient (though it didn’t really seem like it to me.)

Consider my actual perception with careful tracking of symptoms. My patient got moderately depressed, despite not looking it and being unable to articulate it very well, and responded pretty well to the escitalopram!

Of course, this might be placebo response or spontaneous remission. You might also accuse me of being a bad diagnostician and contend that a more experienced psychiatrist would’ve asked better questions and more accurately picked up on the non-verbal cues from the patient.

And sure, maybe I am just a bad psychiatrist, but I’d like to think that this will make me at least a little less bad in the long run.

1. This paper uses a much smaller number of studies than the Leucht analysis and they set their MIC at 10 for reasons that I don’t agree with, so I think that they’re overly optimistic

2. It is worth noting that this analysis does not suggest that the effect isn’t real. We would need to see how the model behaves at various different PANSS scores (e.g. 50, 80, 110, etc.) which I don’t think they did.

3. It’s worth mentioning that the Huhn paper also thought about this for the antidopaminergics, but did not find any differential effect of baseline severity on efficacy that was unique to placebo-controlled studies.

4. They also report a significant difference in the number of “early marked response” patients - slightly more in the nortriptyline group - but I didn’t mention it because it doesn’t look like they corrected for multiple comparisons and it would fall outside of even a simple Bonferroni correction (here, p < 0.0055)

5. I have literally never seen this happen, but I’m sure it does in hospital systems that are willing to set money on fire. Like, more than usual amount they set on fire by having an inpatient psychiatry unit in the first place.

6. I am careful to use “schizophrenic” here in its DSM-5 sense, that is >6 months of positive and negative symptoms.

7. Salt

8. For the non-physicians: When we want to get paid for our work by insurance companies, we generally have to provide a diagnosis for which we are providing care. There is very little incentive for physicians to be thoughtful and specific about what diagnoses we submit to billing companies, and many incentives for us to just put down a collection of plausible - but uncertain - diagnosis that we know we will be paid for. In fact (at least in the US) more diagnoses are better because of how our reimbursement is structured! In practical terms this means that a patient with a collection of depressive/anxious symptoms will usually get slapped with diagnoses of major depressive disorder, generalized anxiety disorder, and maybe even a couple others.

9. To be clear, I am not saying that this doesn’t happen with psychosis. This study from the British Journal of Psychiatry suggests that 58% of individuals will experience near complete remission following their first episode of psychosis.

10. In other words, I don’t go through a whole SIGECAPS checklist. When a patient tells me that they’re crying uncontrollably for no reason every day for the past month, barely getting out of bed in the morning, and having regular thoughts of suicide I usually don’t feel the need to ask them whether or not they are experiencing excessive guilt, anhedonia, etc.