Someone brings this up at a party. They always do. The cold plunge person, three sips in, tells you that a randomized trial of three thousand people found cold showers cut sick days by twenty-nine percent. And here is the part that makes it fun. They are right. The trial exists. The number is in the abstract. Twenty-nine percent, p equals 0.003.

So the interesting question is not whether the study is real. It is what happened to the denominator on the way to the party.

The paper is Buijze and colleagues, PLoS One, 2016, out of the Academic Medical Center in Amsterdam. They called it the Cool Challenge. Between January and March of 2015 they randomized 3,018 Dutch adults, all employed, all without severe heart or lung disease, none of them habitual cold showerers, into four arms. Three arms finished every shower with 30, 60 or 90 seconds of the coldest water their tap produced (about 10 to 12 degrees Celsius in a Dutch winter) for 30 consecutive days, then showered cold as often as they liked for another 60. The fourth arm showered normally for all 90 days.

Two primary outcomes. Illness days, meaning days you felt sick, and sickness absence, meaning days you did not go to work. Both self reported through a web form filled in three times over four months. Nobody was blinded, because you cannot blind a person to the temperature of their own shower.

Illness days did not move. Sickness absence did, by 29 percent.

Here is what the paper actually publishes for its primary outcome, and it is worth sitting with for a moment.

The median participant in every arm, including the control arm, missed zero days of work. The 29 percent is an incidence rate ratio of 0.71 from a negative binomial regression, which is the correct tool for count data this lopsided. The statistic is correct. It is also not a number of days.

The paper never prints the group means. But you can recover them from the model, and when you do, the whole thing deflates gently.

Four tenths of a day. Call it three working hours per person per quarter, on a self reported outcome, in an unblinded trial.

And the sample size calculation gives the game away entirely. The authors powered the study to detect a difference of 0.5 days of sickness absence. They designed a trial to find half a day and they found half a day. Twenty-nine percent is that same half day wearing a suit.

The crude proportions in the table tell the same story from a different angle. About 34.8 percent of controls missed at least one day, against about 32.1 percent pooled across the cold arms. Roughly 37 people would need to shower cold for three months for one additional person to avoid missing any work at all. That is a descriptive comparison rather than an adjusted one, and the trial never counted absence episodes, only days. But it is the number a person would actually want before deciding how to spend the next ninety mornings.

Credit where it is due. This is an honest paper about a modest finding, and most of what is wrong with it is disclosed in its own limitations section. A partial inventory.

Every outcome was self reported through a web form. Compliance could not be verified, by design. The trial ran during the longest Dutch influenza epidemic in more than forty years, with a partial vaccine mismatch, which raises the event rate but also means the result is anchored to one unusual season. Registration with the Netherlands Trial Register happened in June 2015, after the data were collected and before they were analyzed, which the authors state plainly. Twenty serious adverse events were reported and adjudicated unrelated, including one death from an occult pulmonary embolism in a participant whose condition was undiagnosed at enrollment.

Two more things worth noticing, neither of them a scandal. The author list includes the risk management arm of a Dutch insurer, which is exactly the kind of organization for which sickness absence is the outcome that matters and it is disclosed in the competing interests statement. And the data underlying Table 2 were omitted from the original supporting information and only released by a correction in August 2018, five months after Harvard Business Review had already told its readers the answer.

My favorite finding in the paper is the one nobody quotes. In the same regression model, regular physical activity carried a rate ratio of 0.65, a 35 percent reduction in sickness absence. Larger than the shower. It was sitting right there in Table 2 the whole time.

Cold water research is a field where the physiology is well characterized and the health claims are not. Roughly where things stand.

Immune modulation exists but is small and fast. The famous Kox PNAS 2014 study trained 12 volunteers in meditation, cyclic hyperventilation and ice immersion, then injected them with endotoxin. The trained group produced more IL-10, less TNF alpha, IL-6 and IL-8, and had fewer flu like symptoms. Real, and the follow-up work from the same group points at the breathing rather than the cold as the driver. It is also an acute experiment in a laboratory, not a claim about winters.

Metabolic adaptation is the best documented effect. Søberg and colleagues in Cell Reports Medicine found experienced Scandinavian winter swimmers running lower core temperatures with greater cold induced thermogenesis and no detectable brown fat activity at rest, which reads as genuine acclimation rather than a health benefit. The 2022 circumpolar health review of 104 studies lands on adipose tissue and insulin sensitivity as the most plausible durable effects, and concludes, after reviewing all of it, that the topic remains a subject of debate.

The athletic literature is the most mature and the least flattering. Cold water immersion reliably reduces muscle soreness and creatine kinase and improves perceived recovery (Moore 2022, Moore 2023, and a 2026 soccer meta-analysis whose prediction intervals all cross zero). It also blunts muscle fiber hypertrophy after resistance training. A 2025 trial in national level soccer players found it no better than a sham laser. And a 2025 Journal of Physiology study found hot water immersion beat cold for muscle regeneration after simulated injury, which will not be appearing on any plunge tub website.

Put together, the shelf says this. Cold water does measurable things to human physiology within minutes. Whether any of it adds up to being healthier a year later is a question nobody has yet answered with an objective endpoint.