A new interesting paper was published 2 weeks ago on depression and temperature. I saw a news article about the paper today that I didn’t think captured the core findings particularly well — and I wasn’t happy with some of the authors’ conclusions either — so I briefly wanted to summarize these issues here.

Study results

The researchers followed about 20,000 people over 7 months. Participants were queried once a month about their self-reported depressive symptoms (things like sad mood, insomnia, fatigue, concentration problems) and temperature, both self-reported and using a wearable device. The authors find that depression and temperature are correlated. This is the main figure (pink additions by me).

A few things stand out. First, this is a really interesting finding that I believe will replicate. The authors identified what we call a robust phenomenon or explanandum: a finding that requires explanation. What do you think is the explanation for the finding? We’ll get back to this later.

Second, the main finding to me is not that depression severity and temperature are related, but that around 0.5% of all participants — those with severe depression symptoms (dark red line) — have elevated temperature. Interesting, but not the same.

Third the difference in temperature is very small. This is very important for reasons of clinical utility, or what we call predictive accuracy — let me explain this. Men are taller than women, on average, but they aren’t twice as tall. This means that if you tell me someone’s height, I may provide an educated guess if they are male or female, but in many cases it’s just a coinflip: I’m totally guessing. The same goes for temperature: the differences are so small that the temperature here isn’t useful to help us guess whether someone is depressed or not.

Explaining the results

An news article covering the piece now asks the right question: what explains this correlation?

News: “While it’s not clear if poor body temperature regulation is a symptom of depression or vice versa”

This is mistaking correlation with causality: just because temperature and depression symptoms are correlated does not mean one causes the other. There is a third explanation. Take the correlation between ice cream consumption and pool drownings: one could make the same mistake and conclude that ice cream consumption causes pool drownings, or pool drowning cause ice cream consumption, but what actually happens is that a third variable explains both: when the sun shines, people eat more ice cream and also go to pools more. We call this a ‘common cause’, where one variable causes two other variables, explaining the correlation among them. You have seen many examples of common causes. For example, when you observe a kid that has both Koplik’s spots (red dots in the mouth) and fever, you would be a poor doctor to conclude that one causes the other, when it fact the kid has measles, which causes both symptoms.

What common cause could be part of a causal model for depressive symptoms and temperature? I’m not an expert on temperature and asked around, and my best guesses would be medication (people with very severe depression may take different medication that could slightly alter metabolism/temperature); other mental or physical disorders (people with severe depression have, in over 50% of cases, other health problems, which may impact thermoregulation); BMI, weight, fitness, or inflammation could also play a role (depressed people exercise less, may have higher weight, and there is some evidence for higher levels of inflammation — all of which may impact thermoregulation). There are many other possibilities (e.g., parts of the menstrual cycle can be related to both mood-related symptoms and thermoregulation differences), and the authors actually mention e.g. inflammation in their paper. Based on this, I don’t think the treatment implications they talk about next make a lot of sense.

Treatment implications

News: “It could be a way in for non-pharmaceutical treatment … If thermoregulation is a symptom of depression, as the Oura Ring study suggests, temperature treatment could be an effective way to treat depression.”

The news article turns an observed correlation into a causal interpretation: depression causes temperature alterations, hence, temperature is a symptom of depression (symptoms are consequences and indicated an underlying issue — Koplik’s spots is a consequence of measles and indicates measles). In this case, temperature treatment would be nonsense: you don’t treat Koplik’s spots to treat measles or treat cough to cure the flu or treat pain to cure a broken bone. You cure the underlying disease, and then the symptoms will disappear as a consequence.

Authors: “What is we can track the body temperature of people with depression to time heat-based treatments well? .. Given the climbing rates of depression in the United States, we’re excited by the possibilities of a new avenue of treatment .. [results] may hold implications for development of novel approaches to the treatment of major depressive disorder.”

This doesn’t make sense to me.

1. If the causal model is ‘depression -> temperature’, as the news article suggests, temperature treatment would not affect depression.
2. If the causal model is ‘common cause -> depression & temperature’, temperature treatment would not affect depression.
3. And if the causal model is ‘temperature regulation -> depression’, as the authors suggest in the abstract when they say that temperature alterations are “potentially relevant factors in depression etiology”, this appears inconsistent with their data. Smoking is part of the etiology of lung cancer, and that means that first comes the smoking, and later comes the lung cancer. Some types of unhealthy diets are part of the etiology of cardiovascular diseases, and that means that heart attacks follow after people have eaten unhealthy for years. Here, thermoregulation difficulties should be seen before people develop depressive episodes, but this doesn’t appear to be the case.

Conclusion

This is an interesting study, with results that I believe are genuine (i.e., will replicate), demonstrating that a few folks with severe depression have slightly elevated temperature. I mean, just look at it — how cool is that? (red line is again a few dozen folks with severe depression)?

But that is all the study shows: an interesting correlation we now need to find explanations for. The authors don’t have any explanations, given their study design. That is fine and makes for an interesting paper, and the authors should be proud. But in cases like this, lets just stay away from talking about treatment implications please. This goes for authors, university press releases, but also journalists. Correlation ≠ causality is not a high bar.