Fever is one of my favorite topics. Our body’s physiologic response to (usually) infection by raising its internal temperature is fascinating on many levels, in particular the fact that it is so misunderstood by so many people. This has resulted in a significant amount of “fever phobia” and countless unnecessary blood draws, antibiotic prescriptions, and doses of antipyretics. I discussed this in detail in one of my earliest posts for Science-Based Medicine back in 2012:

The occasional abnormal elevation in body temperature associated with infection is as much a part of the human condition as abstract thought or the desire to lose weight without exercise or cutting calories. Commonly known as fever, this powerful yet misunderstood physiologic response has been documented in a variety of animal species including fish, reptiles and of course humans. We have all had fever at least once in our lives, and probably several times. And many of us have undoubtedly spent a few anxious nights cradling febrile little ones, afraid more of the repercussions of the fever itself than the potential sequelae of the underlying cause.

Along those lines, fever is one of the most common reasons for parents to seek medical care for their children, with roughly a third of pediatric acute care visits related to it, as well as a frequent impetus for late night nursing calls to sleepy hospitalists. Actually only about half of after-hours calls are about fever but who’s counting. Unfortunately most medical professionals, including many pediatricians, have a poor understanding of the pathophysiology of fever, and their panicked approach to its management in many children involves unnecessary laboratory tests, imaging studies, and doses of broad spectrum antibiotics. It also adds to parental anxiety and helps to establish a vicious cycle as patients of over worried caregivers tend to undergo more aggressive evaluation and treatment.

I’ve been thinking about fever, or more accurately what is a “normal” body temperature, a lot over the past week. In order to determine what a fever is, after all, you have to have a firm grasp of what a fever isn’t…right? Well, to be honest, firm isn’t exactly how I would describe our grasp of normal human body temperature. Not that we are wholly ignorant. We have it down to a decent range that is good enough for clinical purposes, but there are numerous genetic, hormonal, metabolic, and environmental variables that make determining what temperature any one healthy individual should be at any given time virtually an impossibility. And the early data upon which most people still base their understanding of “normal” body temperature, particularly the myth of 98.6°F, is over 150-years-old.

On January 7th, a paper on this topic was published in eLife that caught my attention thanks to several news articles and a Stanford Medicine press release claiming that the average temperature of humans has been steadily decreasing over the past 150 years or so. Not only do the authors claim that humans (in the United States) have been getting cooler by about .03°C on average per birth decade, they suggest several possible explanations. Though they discuss the possibility that changes in resting metabolic rate (RMR) may have played a role, they believe that the most plausible factor is a population-level decrease in inflammation:

Economic development, improved standards of living and sanitation, decreased chronic infections from war injuries, improved dental hygiene, the waning of tuberculosis and malaria infections, and the dawn of the antibiotic age together are likely to have decreased chronic inflammation since the 19th century.

To explain temperature data from more recent cohorts, when presumably the above factors would not be an issue, they raise the possibility that widespread use of anti-inflammatory medications is causing the slow decline in body temperature over time. Also mentioned by the authors was the potential role of the advent of climate controlling technology such as air conditioning:

Thus, the amount of time the population has spent at thermoneutral zones has markedly increased, potentially causing a decrease in RMR, and, by analogy, body temperature.

In my opinion, it is premature for the study authors to float possible explanations for a decrease in average human body temperatures (in the United States). We should prove that there truly is a decrease, right? The authors believe they have done this but I’m not convinced.

In order to determine that human temperature, at least in a resource rich developed region, is decreasing, the researchers looked at three cohort populations obtained over the past 157 years which involved a total of over 600,000 individual measurements. Data from each has potential issues that would make it difficult to draw firm conclusions regarding temperature trends in my opinion.

  1. The Union Army Veterans of the Civil War, 1860-1940 (UAVCW): 23,710 individual adults, 83,900 total measurements. Unknown measurement technique (oral versus axillary) with unknown precision. Unknown time of day. Unknown ambient temperature. All male. Average age around 60.
  2. The National Health and Nutrition Examination Survey, 1971-1975 (NHANES I): 15,301 individual adults, 15,301 total measurements. Oral temperatures, unknown precision. Time of day not available with all measurements. Unknown ambient temperature. 61% women with unknown ovulation status. Average age around 40.
  3. The Stanford Translational Research Integrated Database Environment, 2007-2017 (STRIDE): 150,280 individual adults, 578,522 total measurements. Digital oral temperatures with known precision. Time of day available but not ambient temperature. 60% women with unknown ovulation status. Average age around 50.

Using the power of math, the authors claim that they have accounted for numerous potentially confounding variables and, most importantly, the possibility that differences may be the result of improved measurement accuracy over the decades rather than changes in human physiology. When all was said and done, they determined that both men and women, regardless of ethnicity, experienced a similar steady decline in average body temperature:

We observed a steady decrease in body temperature by birth cohort for both men (−0.59°C between birth decades from 1800 to 1997; −0.030°C per decade) and women (−0.32°C between 1890 and 1997; −0.029°C per decade).

Okay, not so fast. When I first read the paper and the accompanying press release, as well as the media coverage, there were some red flags. First off, a quote from the lead author demonstrated a perhaps not-so-solid understanding of human evolution:

“Physiologically, we’re just different from what we were in the past,” Parsonnet said. “The environment that we’re living in has changed, including the temperature in our homes, our contact with microorganisms and the food that we have access to. All these things mean that although we think of human beings as if we’re monomorphic and have been the same for all of human evolution, we’re not the same. We’re actually changing physiologically.”

To be clear, even if the study findings are true, it absolutely does not represent an evolutionary change in average human body temperature in response to environmental pressure. We aren’t cooling down because of a genetic mutation that resulted in a survival advantage in…checks notes…a world with less disease and a more comfortable life. Americans are also taller, heavier, and longer lived than 150 years ago, again not because of evolutionary changes in our physiology in response to environmental survival pressure but because of better health and easier access to large quantities of calories.

In the introduction, the authors describe how the normal range of human body temperature was largely determined by the work of German physician Wunderlich who, starting in 1851, began recording the axillary temperatures of his patients. He eventually collected data from 25,000 subjects over 15 years, amounting to millions of individual measurements, and came up with the infamous and thoroughly unhelpful 98.6°F average temperature. That useless number became chiseled into the collective consciousness of humanity, though in reality Wunderlich was a pioneer in establishing a range of temperatures in healthy and ill patients, and his work helped to establish temperature as an important factor in many conditions. Though he wasn’t the first to measure human body temperature, he was one of the first physicians to fiercely promote the idea that it was a clue to certain ailments, particularly infection.

In Wunderlich’s day, we had a very limited understanding of human physiology and have since learned a great deal about the normal variations in temperature related to hormonal cycles (higher in the late afternoon, lower in the early morning) as well as to numerous other internal and external factors. The concept of one temperature being used as a guideline for what is normal was long ago rendered meaningless and clinically unhelpful, even if it has persisted as modern day medical folklore because of cultural inertia. I do use specific values for what is a fever, 100.4F/38C in a young infant for example, but even then I recognize that this is somewhat arbitrary, purposefully oversensitive, and only one component of my assessment of a child’s risk for a life-threatening infection.

A problem when comparing modern temperature data to those measurements taken in 1851 is the fact that they were almost certainly inaccurate, although they had come a long way from the days of Galileo and Santorio. While Wunderlich and his contemporaries were well past the days of documenting human temperatures of 118F, the thermometers at his disposal could hardly be expected to have precision to tenths of a degree such as we have today. The first practical medical thermometer, which was still pretty large by today’s standards, wasn’t even invented until 1867, which means that in 1851 Wunderlich would have been using something extremely cumbersome, along the lines of a foot long instrument that took 20 minutes to yield a result.

More recent data mentioned in the eLife paper, such as a 2002 compilation of 27 modern studies and a 2017 study involving roughly 250,000 individual measurements, have found a lower average temperature of 97.9F/36.6C. The consensus conclusion from these papers is not that we are cooler but that we have better thermometers and use more accurate, though still imperfect, techniques. Taking temperatures orally with digital thermometers, for example. The authors of the new study beg to differ.

In order to support a claim that adults have gotten cooler, the study authors compared results of the above described cohorts to each other, but also looked within each cohort to compare subjects over time based on the years that they were born. They argue that each cohort would have used similar measurement techniques, thus avoiding the potential for bias of improved measurement techniques. In their analysis, they are forced to make numerous assumptions that I believe are problematic.

To give one example, the UAVCW data did not include measurement technique at all, but it is safe to assume that it consisted of a mix of both oral and axillary temperatures. These techniques yield different ranges and have different risks of measurement error. As an argument in support of ignoring the possibility that this might have made the data unreliable for their purposes of making comparisons over time, the study authors claim that axillary temperatures in the earliest cohort would have, if anything, resulted in an underestimation of the cooling.

The authors cite two modern studies finding that axillary temperatures are on average one degree Celsius lower than oral temperatures, but this is cherry picked and based on modern thermometers rather than the crude oral thermometers that would have been available at the turn of the 20th century. Axillary temperatures are more consistently found to be lower than the gold standard of rectal temperatures, but even in that case the data is varied and there is no accepted conversion. Because of wide normal ranges and the possibility for measurement bias, such as improper technique, ambient temperature differences, etc, an axillary temperature can easily be higher than an oral temperature. Ambient temperature was also unknown in the UAVCW cohort.

In addition to not knowing the ambient temperature in any of the cohorts, another glaring problem is that the time of day is not known in either of the first two sets of data. The difference in temperature from morning to late afternoon can be as much as .5C. Without the ability to take this variable into account, it is hard to have much confidence in any comparisons. Time of day was available in the third cohort, which is also the most recent and involved modern digital thermometers, so it has the most reliable data. But there are still many variables that couldn’t possibly be controlled for, and it can’t really be used to make comparisons to the older cohorts.

But if we assume that the most recent cohort, which involved measurements taken between 2007 and 2017, has no measurement errors and we can account for all the confounding variables, I still don’t see how the proposed explanations for a cooling add up. If it’s related to a decrease in inflammation, cooling should have hit a steady state by now. Regardless, it’s one set of data and would need replication.

Conclusion: Skeptical but my mind is open

As I said above, I don’t buy the claim that humans, at least in the United States, are cooling over time. My first instinct when I saw the headlines was that it’s a pretty ridiculous concept and reading the paper didn’t change my mind all that much. But the data from the most recent cohort is interesting, and I’m open to more research looking into this claim even if I honestly can’t imagine what utility it will have. Certainly there is no clinical usefulness and the proposed explanations are rather implausible.

The bottom line is that measuring temperature in humans in fraught with error and bias. Because obtaining measurements from the pulmonary artery or distal esophagus is risky, and most adults would frown upon rectal temperatures, we are forced to rely on more peripheral measurement techniques that give us an idea of what a subject or patient’s core temperature is. No technique, and there are more than just axillary or oral approaches, is particularly reliable, and what is normal is a range of temperatures throughout the day rather than one “normal” temperature.


Posted by Clay Jones

Clay Jones, M.D. is a pediatrician and a regular contributor to the Science-Based Medicine blog. He primarily cares for healthy newborns and hospitalized children, and devotes his full time to educating pediatric residents and medical students. Dr. Jones first became aware of and interested in the incursion of pseudoscience into his chosen profession while completing his pediatric residency at Vanderbilt Children’s Hospital a decade ago. He has since focused his efforts on teaching the application of critical thinking and scientific skepticism to the practice of pediatric medicine. Dr. Jones has no conflicts of interest to disclose and no ties to the pharmaceutical industry. He can be found on Twitter as @SBMPediatrics and is the co-host of The Prism Podcast with fellow SBM contributor Grant Ritchey. The comments expressed by Dr. Jones are his own and do not represent the views or opinions of Newton-Wellesley Hospital or its administration.