Participants

Our sample consisted of 96 adults (42 men, 54 females; no one self-identified as neither or other), ranging in age from 18–67 years (mean = 30 years). Persons who were feeling unwell at the time or recently, were pregnant, had a chronic illness, or were taking any medication that could depress one’s temperature were excluded. Ethnicities of participants were 55% European origin, 16% East Asian, 16% South Asian, 8% Middle Eastern, and 5% mixed. No participant smoked. Participants were recruited via posters and social media, gave their written informed consent, and were paid $10 to thank them for their participation. No participant refused to participate since potential participants contacted us to volunteer. The study received human subjects approval from the University of British Columbia and Vancouver Coastal Health.

Materials and procedure

Life Brand^® Fast Read Digital Oral Thermometers were used. They are accurate to 0.1°iover the range of 34.0° to 42.0°. To take their temperature, participants turned on the thermometer, placed it under their tongue towards the back of their mouth, closed their lips, and waited until the thermometer beeped.

Participants were to take seven pairs of temperature readings in the evening right before going to sleep and first thing the following morning over a two-week period (yielding a total of 1,333 temperature readings). Thus all temperature readings were taken in participants’ own homes in the greater Vancouver, BC area. Participants were instructed not to eat, drink, or engage in strenuous physical exercise during the 30 minutes prior to taking their temperature and not to take their temperature for this study when feeling unwell. We also collected data on factors that might affect body temperature including weight, height, age, ethnicity, and menstrual phase (when a woman’s last period began and ended and when she expected her next period). about the length of their menstrual cycle and dates of their last and next (estimated) periods.

We calculated BMI from height and weight (weight divided by height squared [kilograms/meters²). We estimated midluteal phase menstrual phase by counting back 6–8 days from the anticipated onset of the next menstruation, and early follicular phase by counting forward 3–5 days after the onset of menstruation. We did not ask about use of oral contraceptives unfortunately. All data were collected between January and May in Vancouver, BC, where the mean outdoor temperature during this period was 7°C (44°F) and ranged from -8°C (18°F) to 16°C (61°F).

Couples participating were asked to be in the same room and wear similarly heavy or light clothing at the time of the temperature readings and for ≥30 minutes beforehand to try expose their bodies to the same ambient environment with similar protection against heat or cold. Our sample included 27 couples and 42 singles. All couples were heterosexual.

We provided each participant with a thermometer and a temperature collection form for recording temperature readings and their dates and times. Importantly, participants were unaware of our hypotheses. Participants gave a $10 deposit for the thermometer, which was returned when they handed in their collection form and thermometer.

Wide range of normal temperatures

Are there marked differences in the mean or “normal” temperature? Yes. The range of mean temperatures in our study spanned 2.2°C (4.0°F). Across a representative sample of reviews and other studies, the range in normal oral temperature has varied from 0.5°C (0.9°F) to 2.6°C (4.7°F; see Table 2). Other researchers who found variation in the normal temperature of healthy adults to be ≥2° include Geneva et al. [4], Sund-Levander et al. [13], Mackowiak et al. [5], Gomolin et al. [14, 15], and Keilson et al. [16] See Table 2 [17]. Individual temperature readings in our study ranged from 33.8°C (92.8°F) to 37.7°C (99.9°F); a difference of 3.9°C (7.1°F); see Fig 1b. Such large ranges mean that body temperature varies enough from person to person that using the same mean for everyone will cause marked errors.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Results from a representative sample of studies and reviews of oral temperature in healthy adults.

https://doi.org/10.1371/journal.pone.0245257.t002

Marked stability of temperature over 14 readings spanning 2 weeks

The stability of temperature readings taken at roughly the same times across several days had not previously been reported. We found it to be quite stable. Thermal stability was equally true for men and women, and older (i.e., middle-aged) and younger adults. This is consistent with a snapshot of someone’s temperature taken at each yearly check-up being sufficient to establish the average temperature for that person, realizing that slight dips very early morning and slight increases just before and after dusk are typical [2, 5, 7–9, 14].

We did not find the diurnal changes that are so well established. That was probably because of the times of day when we assessed temperature (07:00, after the morning nadir; and 22:30, after the peak around dusk; [5, 8, 18]). While studies find that temperatures tend to rise in late afternoon, they also find they dip just before sleep onset [19–21]. Also, many of the temperature measurements for our study may have been taken while participants were in bed (after having just awakened or before going to sleep). Prone or recumbent posture has associated with lower temperature [20, 22, 23]. This may also be why the temperatures we observed were toward the lower end of the range of temperatures found in other studies. Also our study is more recent and Protsiv et al. [24] report that temperatures have been declining.

37.0°C (98.6°F) is not an accurate approximation of the normal temperature of most adults

The mean oral temperature of our healthy volunteers was 36.1°C (97.0°F). The mean temperatures from reviews and studies listed in Table 2 vary from 36.0°C (96.8°F) to 36.8 °C (98.2°F), except for Wunderlich. Wunderlich’s estimate is especially too high when oral versus rectal thermometers are used because oral and axillary temperatures tend to be 0.3–0.6°C (0.5–1.0°F) lower than rectal temperature, though the range is similar [7, 13, 25]. Experts have been arguing to abandon 37.0°C (98.6°F) as not having any “special significance vis-à-vis the oral temperature of healthy adults” since at least 1992 [5 p. 1580]. A just-published study based on massive datasets [24] reports that at least in the U.S., temperatures have been declining at the rate of 0.03°C per decade of birth.

The mean currently in use (~37.0°C [~98.6°F]) is too high, but this has persisted despite 35 years of research consistently showing that it’s too high. Using this mean causes fevers in those with normally lower temperatures to be missed.

Medical historians trace the wide acceptance that normal body temperature of healthy adults is ~37.0°C (98.6°F) primarily to Wunderlich [1]. Wunderlich was the first to systematically use a thermometer to measure human body temperature. His data are based on ~25,000 patients, whose care he supervised. There are a number of limitations with this classic work, however.^a

Whereas most researchers and clinicians before us have argued for a new, lower estimate of normal adult temperature, we are urging that normal temperature be personalized because what is normal for one person can be quite different from what is normal for another. As Edelsberg [27] wrote: “[N]o single value of mean body temperature has much clinical importance because of individual and population variability.”

We are particularly worried about fevers not being detected in persons with low normal temperatures (e.g., ≤36.5°C [97.6°F]), which in our study was 77% of participants, 82% of just the women, and 89% of adults ≥43 years age. Indeed, 30% of our sample had mean temperatures <35.9°C (96.6°F). Similarly, Waalen et al. [3 pp. 490–491] wrote: “[I]n every age cohort up to age 50, there is a small discrete subset of individuals with body temperatures less than 96°F. The representation of that population increases with increasing age.” Since studies consistently report lower mean temperatures in seniors than in younger adults, it is particularly likely that troubling temperature elevations in elderly adults might be missed if one “normal” temperature continues to be assumed for everyone at all ages.

Thus, all of our principal hypotheses were confirmed. We turn now to our secondary hypotheses; only one of which were confirmed.

Our secondary hypotheses

More than half our participants (57%) experienced a temperature rise overnight, but overall there was no consistent trend in either direction. Others have shown that temperature is lower during sleep and its onset [22, 28, 29]. Two studies had previously examined overnight temperature fluctuations, though both looked at only one night per participant, did not align temperature changes to when a participant actually fell asleep or awoke, and had small samples (Baker et al. [21]; Cagnacci et al. [30]).^{a, b}

We found that women had slightly higher temperatures than men, consistent with Wunderlich et al. [6], Mackowiak et al. [5], and Sund-Levander et al. [13], but the opposite of that reported in McGann et al.’s study [31] or Geneva et al.’s review [4]. As far as we know, no one had previously compared the temperatures of spouses. We asked couples to be in the same room before and during taking their temperatures and to wear clothing of similar warmth. Even within couples we found that women had slightly higher temperatures than men. However, if there is a reliable sex difference in adults’ temperatures it seems tiny and unlikely to be of clinical significance.

Women tend to have higher temperatures in the midluteal phase [2, 21, 30]. Our sex-difference results do not seem to be menstrual-phase driven, however, since only 12 women measured their temperature even once during their estimated midluteal phase. We found a trend for temperatures to be higher during the midluteal phase, but that was not significant. Another reason we might have found higher temperatures in women than men is that many of our female participants were probably taking oral contraceptives. Women’s temperatures tend to be higher when they are on oral contraceptives [21, 32].

Although it has been consistently reported that older adults have lower temperatures than younger adults [2–6, 9], we did not find that. Almost certainly that is because our older adults were middle-aged rather than elderly. The mean age of our older adults was 47 years and only two participants were ≥60 years. The participants in Mackowiak et al. [5] were 18–40 and no age-related difference in temperature was found. The participants in Adhi et al. [33] were 9–70 (mean age = 34 years) and they, too, found no age-related temperature difference.

We did find a larger range of temperatures among younger adults, but no greater diurnal variability. Perhaps we found no age difference in diurnal temperature fluctuations because of when we assessed temperatures or because our participants were more similar than different in their degree of physical activity. Thermal variability is greater in those more physically active [34].

Limitations

Limitations of our study include a small sample size and our not having asked about use of oral contraceptives or level of physical activity. We should have asked participants to note the room temperature when taking each temperature reading and note their posture. We had only two participants ≥60 years old, which limited our ability to look at age differences throughout the adult lifespan. We did not confirm menstrual phases with hormone assays. Not having assessed temperatures between 16:00–21:00, we didn’t include the window of peak temperatures. We should have measured participants’ height and weight ourselves rather than accepting participants’ self-reports. It might have been more accurate, though less practical, if we had taken the temperature readings ourselves, but given the small variability between readings and our means and ranges being comparable to those found by others, participants seem to have accurately measured their temperatures.

Conclusions

Our take-home message is that it is time to personalize body temperature. It is needed and easily doable. The range in normal temperature is sufficiently large that using a standard “normal” temperature value will lead to errors for many individuals. For the many people whose normal temperature is much lower than 37.0°C (98.6°F), healthcare professionals may fail to identify a pathologically high fever or may obtain false negatives when using oral temperature to screen for COVID-19, mistaking what is actually an elevated temperature for normal.

We propose that a person’s temperature be taken at each doctor visit, just as we do now with blood pressure. This would be simple to implement and would permit an accurate estimate of each person’s normal temperature because, as we have shown, temperature varies little across days. Such records over time would also provide a more accurate understanding of how temperature changes over the lifespan.