[Content warning: panic, suffocation]


I recently presented this case at a conference and I figured you guys might want to hear it too. Various details have been obfuscated or changed around to protect confidentiality of the people involved.

A 20-something year old woman comes into the emergency room complaining that she can’t breathe. The emergency doctors note that she’s breathing perfectly normally. She says okay, fine, she’s breathing normally now, but she’s certain she’s about to suffocate. She’s having constant panic attacks, gasping for breath, feels like she can’t get any air into her lungs, been awake 96 hours straight because she’s afraid she’ll stop breathing in her sleep. She accepts voluntary admission to the psychiatric unit with a diagnosis of panic disorder.

We take a full history in the psych ward and there’s not much of interest. She’s never had any psychiatric conditions in the past. She’s never used any psychiatric medication. She’s never had any serious diseases. One month ago, she gave birth to a healthy baby girl, and she’s been very busy with all the new baby-related issues, but she doesn’t think it’s stressed her out unreasonably much.

We start her on an SSRI with (as usual) little immediate effect. On the ward, she continues to have panic attacks, which look like her gasping for breath and being utterly convinced that she is about to die; these last from a few minutes to a few hours. In between these she’s reasonable and cooperative but still very worried about her breathing. There are no other psychiatric symptoms. She isn’t delusional – when we tell her that our tests show her breathing is fine, she’s willing to admit we’re probably right – she just feels on a gut level like she can’t breathe. I’m still not really sure what’s going on.

So at this point, I do what any good psychiatrist would: I Google “how do you treat a patient who thinks she’s suffocating?” And I stumble onto one of the first convincing explanations I’ve ever seen of the pathophysiology of a psychiatric disorder.


Panic disorder is a DSM-approved psychiatric condition affecting about 3% of the population. It’s marked by “panic attacks”, short (minutes to hours) episodes where patients experience extreme terror, increased heart rate, gasping for breath, feeling of impending doom, choking, chest pain, faintness, et cetera. These episodes can happen either after a particular stressor (for example, a claustrophobic patient getting stuck in a small room) or randomly for no reason at all when everything is fine. In a few cases, they even happen when patients are asleep and they wake up halfway through. The attacks rise to the level of a full disorder when they interfere with daily life – for example, a patient can’t do her job because she’s afraid of having panic attacks while engaged in sensitive activities like driving.

The standard model of panic disorder involves somatosensory feedback loops. Your body is always monitoring itself to make sure that nothing’s wrong. Any major organ dysfunction is going to produce a variety of abnormalities – pain, blockage of normal activities like digestion and circulation, change in chemical composition of the blood, etc. If your body notices enough of these things, it’ll go into alarm mode and activate the stress response – increased heart rate, sweating, etc – to make sure you’re sufficiently concerned.

In the feedback model of panic disorder, this response begins too early and recurses too heavily. So maybe you have an itch on your back. Your body notices this unusual sensation and falsely interprets it as the sort of abnormality that might indicate major dysfunction. It increases heart rate, starts sweating, et cetera. Then, because it’s stupid, it notices the increased heart rate and the sweating that it just caused, and decides this is definitely the sort of abnormality that indicates major dysfunction, and there’s nothing to do except activate even more stress response, which of course it interprets as even more organ dysfunction, and so on. At some point your body just maxes out on its stress response, your heart is beating as fast as it can possibly go and your brain is full of as many terror-related chemicals as you can produce on short notice, and then after a while of that it plateaus and returns to normal. So panic disorder sufferers are people who are overly prone to have the stress response, and overly prone to interpret their own stress response as further evidence of dysfunction.

This is probably part genetic and part learned – I have a panic disorder patient who has a bunch of really bad allergies, whose body would shut down in horrifying ways every time he accidentally ate a crumb of the wrong thing, and this seems to have “sensitized” him into having panic attacks; that is, his body has learned that worrying sensations often foretell a health crisis, and lowered its threshold accordingly to the point where random noise can easily set it off. I’ve done a lot of work with this guy, but none of it has been “just ignore your panic attacks, you’ll be fine”. His body knows what it’s doing, and we’ve got to work from a position of respecting it while also teaching it not to be quite so overzealous.

So this is where my understanding of panic disorder stood until I Googled “how do you treat a patient who thinks she’s suffocating?” and came across Donald Klein’s theory of panic as false suffocation alarm. You might want to read the full paper, as it’s got far too many fascinating things to list here, including a theory of sighing. But I’ll try to go over the basics.

Klein is a professor of psychiatry who studies the delightful field of “experimental panicogens”, ie chemicals that cause panic attacks if you inject them in someone. These include lactate, bicarbonate, and carbon dioxide, all of which naturally occur in the body under conditions of decreased respiration.

But this is actually confusing. All of these chemicals naturally occur in the body under conditions of decreased respiration. But they don’t cause panic attacks then. During exercise, for example, your body has much higher oxygen demand but (no matter how much you pant while running) only a little bit higher oxygen supply, so at the muscle level you don’t have enough oxygen and start forming lactate. But exercise doesn’t make people panic. Even deliberately holding your breath doesn’t make you panic, although it’s about the fastest way possible to increase levels of those chemicals. So it looks like your body is actively predicting how much lactate/bicarbonate/CO2 you should have, and only getting concerned if there’s more than it expects.

So Klein theorized that the brain has a “suffocation alarm”, which does some pretty complicated calculations to determine whether you’re suffocating or not. Its inputs are anything from blood CO2 level to very high-level cognitions like noticing that you’re in space and your spacesuit just ruptured. If, after considering all of this, and taking into account confounding factors like whether you’re exercising or voluntarily holding your breath, it decides that you’re suffocating, it activates your body’s natural suffocation response.

And the body’s natural suffocation response seems a lot like panic attacks. Increased heart rate? Check. Gasping for breath? Check. Feeling of impending doom? Check. Choking? Check. Chest pain? Check. Faintness? Check. Some of this makes more sense if you remember that the brain works on Bayesian process combining top-down and bottom-up information, so that your brain can predict that “suffocation implies choking” just as easily as “choking implies suffocation”.

A quick digression into medieval French mythology. Once upon a time there was a nymph named Ondine whose lover was unfaithful to her, as so often happens in mythology and in France. She placed a very creative curse on him: she cursed him not to be able to breathe automatically. He freaked out and kept trying to remember to breathe in, now breathe out, now breathe in, now breathe out, but at some point he had to fall asleep, at which point he stopped breathing and died.

So when people discovered a condition that limits the ability to breathe automatically, some very imaginative doctor named the condition Ondine’s Curse (some much less imaginative doctors provided its alternate name, central hypoventilation syndrome). People with Ondine’s curse don’t exactly not breathe automatically. But if for some reason they stop breathing, they don’t notice. Needless to say, this condition is very, very fatal. The usual method of death is that somebody stops breathing at night (ie sleep apnea, very common among the ordinary population, but not immediately dangerous since your body notices the problem and makes you start breathing again) and just never starts again.

Klein says that this proves the existence of the suffocation alarm: Ondine’s Curse is an underactive suffocation alarm – and thus the opposite of panic disorder, which is an overactive suffocation alarm. In Ondine’s Curse, patients don’t feel like they’re suffocating even when they are; in panic disorder, patients feel like they’re suffocating even when they’re not.

This picture has since gotten some pretty powerful confirmation, like the discovery that panic disorder is associated with ACCN2, a gene involved in carbon dioxide detection in the amygdala. If you’re looking for something that causes you to panic when you’re suffocating, a carbon dioxide detector in the amygdala is a pretty impressive fit.

I don’t think this is necessarily a replacement for the somatosensory feedback loop theory. I think it ties into it pretty nicely. The suffocation alarm is one of the many monitors watching the body and seeing whether something is dysfunctional, maybe the most important such monitor. It goes through some kind of Bayesian learning process to constantly have a prior probability of suffocation and update with incoming evidence. Let me give two examples.

First, my patient with the bad allergies. Every time he eats the wrong thing, he goes into anaphylactic shock, which prevents respiration and brings him to the edge of suffocating. His suffocation alarm becomes sensitized to this condition, increases its prior probability of suffocation, and so drops its threshold so low that it can be set off by random noise.

Second, claustrophobics. There’s a clear analogy between being crammed into a tiny space, and suffocating – think of people who are buried alive. For claustrophobics, for some reason that link is especially strong, and just being in an elevator is enough to set off their suffocation alarm and start a panic attack. Now, why agoraphobics get panic attacks I’m not sure. Maybe fear makes them feel woozy and hyperventilate, and the suffocation alarm treats wooziness and hyperventilation as signs of suffocation and then gets stuck in a feedback loop? I don’t know.


Bandelow et al find that you’re about a hundred times more likely to develop a new case of panic disorder during the postpartum period than usual.

This can be contrasted with two equally marked trends. Panic attacks decrease markedly during pregnancy, and disappear entirely during childbirth. This last is really remarkable. People get panic attacks at any conceivable time. When they’re driving, when they’re walking, when they’re tired, when they’re asleep. Just not, apparently, when they’re giving birth. Childbirth is one of the scariest things you can imagine, your body’s getting all sorts of painful sensations it’s never felt before, and it’s a very dangerous period in terms of increased mortality risk. But in terms of panic attack, it’s one of the rare times when you are truly and completely protected.

Maternal And Fetal Acid-Base Chemistry: A Major Determinant Of Perinatal Outcomes notes that:

There is a substantial reduction in the partial pressure of carbon dioxide in pregnancy…this fall is found to reach a mean level of 30-32 mmHg and is associated with a 21% increase in oxygen uptake. The physiological hyperventilation of pregnancy is due to the hormonal effect of progesterone on the respiratory center.

In other words, you’re breathing more, you have more blood oxygen, you have less blood CO2, and you’re further away from suffocation. This nicely matches the observation that there’s fewer panic attacks.

According to Klein, “There is a period of extreme hyperventilation during delivery, which drops the blood carbon dioxide to the minimum recorded under nonpathological conditions”. This explains the extreme protective effect of labor against panic disorder, despite labor’s seeming panic-inducing properties. When your CO2 is that low, even an oversensitive suffocation alarm is very far from a position where it might be set off.


Then you give birth, and progesterone – the hormone that was increasing respiratory drive – falls off a cliff. Your body, which for nine months has been doing very nicely with far more oxygen than it could ever need, suddenly finds itself breathing much less than usual and having a normal CO2/oxygen balance. This explains the hundredfold increased risk of developing panic disorder! Somebody who’s previously never had any reason to think they’re suffocating finds themselves with much less air than they expect (though still the physiologically correct amount of air they need), and if they’ve got any sensitivity at all, their suffocation alarm interprets this as possible suffocation and freaks out.

This can go one of two directions: either it eventually fully readjusts to your new position and becomes comfortable with a merely normal level of oxygen. Or the constant panic and suffocation feelings sensitize it – the same way that my allergy patient’s constant anaphylaxis sensitized him – the alarm develops a higher prior on suffocation and a lower threshold, and the patient gets a chronic panic disorder.

The reason my patient was so interesting was that she was kind of in the middle of this process and had what must have been unusually good introspective ability. Instead of saying “I feel panic”, she said “I feel like I’m suffocating”. This is pretty interesting. It’s like a heart attack patient coming in, and instead of saying “I feel chest pain”, they say “I feel like I have a thrombus in my left coronary artery”. You’re like “Huh, good job”.

So I explained all of this to her, and since she didn’t know I used Google I probably looked very smart. I told her that she wasn’t suffocating, that this was a natural albeit unusual side effect of childbirth, and that with luck it would go away soon. I told her if it didn’t go away soon then she might develop panic disorder, which was unfortunate, but that there were lots of good therapies for panic disorder which she would be able to try. This calmed her down a lot and we were able to send her home with some benzodiazepines for acute exacerbation and some SSRIs which she would stay on for a while to see if they helped. She’s scheduled to see an outpatient psychiatrist for followup and hopefully he will monitor her panic attacks to see if they eventually get better.


I realize that case reports are usually supposed to include a part where the doctor does something interesting and heroic and tries an experimental new medication that saves the day. And I realize there wasn’t much of that here. But I think that in psychiatry, a good explanation can sometimes be half the battle.

Consider Schachter and Singer (1962). They injected patients with adrenaline (a drug which among other things makes people physiologically agitated) or a placebo. Half the patients were told that the drug would make them agitated. The other half were told it was just some test drug to improve their eyesight. Then a confederate came and did some annoying stuff, and they monitored how angry the patients got. The patients who knew that the drug was supposed to make them angry got less angry than the ones who didn’t. The researchers theorized that both groups experienced physiological changes related to anger, but the patients who knew it was because of the drug sort of mentally adjusted for them, and the ones who didn’t took them seriously and interpreted them as their own emotion.

We can think of this as the brain making a statistical calculation to try to figure out its own level of anger. It has a certain prior. It gets certain evidence, like the body’s physiological state and how annoying the confederate is being. And it controls for certain confounders, like being injected with an arousal-inducing drug. Eventually it makes its best guess, and that’s how angry you feel.

In the same way, the suffocation monitor is taking all of its evidence about suffocation – from very low-level stuff like how much CO2 is in the blood to very high-level stuff like what situation you seem to be in – and then adjusting for confounders like whether you’re exercising. And I wonder whether telling a patient “You’re not actually suffocating, your panic comes from a known physiologic process and here are the hormones that control it” is the equivalent of telling them “You’re not really angry, your agitation comes from us giving you a drug that’s known to produce agitation”. It tells the suffocation alarm computer that this is a confounder to be controlled for rather than evidence on which to update.

I can’t claim to really understand this at a level where it makes sense to me. There are a lot of things that very directly increase CO2 but don’t increase panic, or vice versa. Hyperventilation can either cause or prevent panic depending on the situation. There seems to be something going on where the suffocation monitor controls for some things but not others, but this is an obvious cop-out that allows me to avoid making real predictions or narrowing hypothesis-space.

For example, this theory would seem to predict that waterboarding shouldn’t work. After all, its whole deal is artificially inducing the feeling of suffocation in a situation where the victim presumably knows that the interrogators aren’t going to let him suffocate. You would think that eventually the alarm realizes that “is being waterboarded” is a confounder to control for, but this doesn’t seem to be true.

(on the other hand, the inability to condition yourself seems relevant here. It seems like the brain might be not be controlling for whether something is reasonable, but only for whether something is produced by yourself. So maybe exercise counts because it’s under your control, but waterboarding doesn’t count because it isn’t. I wonder if anyone has ever tried letting someone waterboard themselves and giving them the on-off switch for the waterboarding device. Was Hitchens’ experience close enough to this to count? Why would this be different from letting someone hold their breath, which doesn’t produce the same level of panic?)

But overall I find Klein’s evidence pretty convincing and feel like this must be at least part of the story. And I think that giving this kind of explanation to somebody can comfort them, reassure them, and (maybe) even improve their condition.

New Comment
1 comment, sorted by Click to highlight new comments since:

(on the other hand, the inability to condition yourself seems relevant here. It seems like the brain might be not be controlling for whether something is reasonable, but only for whether something is produced by yourself. So maybe exercise counts because it’s under your control, but waterboarding doesn’t count because it isn’t. I wonder if anyone has ever tried letting someone waterboard themselves and giving them the on-off switch for the waterboarding device. Was Hitchens’ experience close enough to this to count? Why would this be different from letting someone hold their breath, which doesn’t produce the same level of panic?)


Hypothesis: the difference is in the failure mode.

If you hold your breath, you can always choose (assuming you are not underwater or in some other environment that prevents proper breathing) to stop holding your breath and save yourself from suffocating. If you are being waterboarded by friendly demonstrators, you can say the safe word and save yourself from "drowning". These may seem the same, but are not. In both cases the longer you hold the weaker you get - the way that weakening affects your ability to stop the ordeal is very different.

The longer you hold your breath the harder it gets to keep holding your breath - until at some point you are no longer able to hold your breath and are forced to breath. Even if you can keep holding your breath past that point - you are just going to pass out, and then you'll just switch to autopilot and breath automatically. Unless you suffer from Ondine's curse, failure will not kill you.

With waterboarding, if you become too weak to properly signal the "torturers" to stop - they won't stop. Sure, in Hitchens' case they noticed that he passed out and stopped it. Because they are professionals. But this is probably too high level for your subconsciousness and your body to rely on - as far as they care, failure can mean death. Holding as long as you can is no longer a safe option - so your body will try to scream at you to stop it as soon as possible.