Smartwatch sweat-loss estimates were off by a quarter to a third

Most current running watches will tell you how much you sweated. The number appears alongside pace and heart rate with the same air of measurement, though sweat loss is not measured by any sensor on the wrist — it is estimated from workload, duration and environmental inputs. A validation study published this month put two commercially available smartwatches against the reference method and found the estimates are useful for tracking trends and not much else [s1].

How it was tested

The study enrolled 111 participants (mean age 40 ± 14 years; height 171.0 ± 8.9 cm; body mass 70.8 ± 14.7 kg; VO2max 46.8 ± 8.7 mL/kg/min) who completed up to four running trials each under varying conditions: indoors and outdoors, distances from 2.5 km to 20 km, and interval runs [s1].

The criterion was the standard laboratory approach — change in nude body mass before and after exercise [s1]. Validity was assessed with mean absolute error and mean absolute percentage error, Lin's concordance correlation coefficient, Pearson's r, Deming regression, equivalence testing on the confidence interval for the difference in means, and Bland-Altman plots for bias [s1]. Results were also stratified by sex, weight and fitness category [s1].

That is a more demanding validation package than consumer device studies usually apply, and it is worth noting because the headline result depends on which statistic you look at.

The results

On correlation, both devices did reasonably: Lin's concordance correlation coefficients ranged from 0.71 to 0.90 [s1]. On error, they did not: mean absolute percentage error ranged from 25.38% to 33.21% [s1].

Those two findings are not in conflict. A high concordance coefficient means the devices ranked and tracked sweat loss in roughly the right order — bigger, hotter, longer runs produced bigger estimates. A mean absolute percentage error near 30% means that for a run in which you actually lost 1.5 litres, an estimate somewhere between roughly 1 litre and 2 litres is typical.

Stratifying by sex, weight and fitness category did not rescue the numbers; the study reports similar results across strata, with no additional analyses meeting the combined validity thresholds [s1].

The authors' conclusion is that wearable devices are a promising and convenient tool for general sweat-loss tracking but lack the precision to replace laboratory methods for hydration management, and that traditional methods remain essential where accuracy matters [s1].

Why the precision question is not academic

The obvious response is that a 30% error is fine for a recreational runner deciding whether to carry a bottle. That is probably true. The reason the accuracy question has weight is that the same class of estimate is increasingly invoked in occupational heat settings, where the stakes are kidney injury rather than a dry mouth — and where the evidence that hydration alone is protective is weaker than assumed.

A study published earlier in November in the American Journal of Physiology - Renal Physiology tested whether drinking to thirst during occupational heat stress reduces markers of kidney injury [s2]. Thirteen healthy adults, five of them women, completed two two-hour simulations of occupational heat stress — eight circuits of treadmill walking and rowing at a wet bulb globe temperature of 33.1 ± 0.2 °C [s2]. In one trial they were offered 237 mL of a non-caloric sports drink every 15 minutes and drank ad libitum; in the other, no fluid was provided [s2].

The drinking trial worked as intended in the sense that mattered least: ad libitum intake averaged 1,394 ± 316 mL and body mass loss was smaller than in the no-drink trial (1.3 ± 0.8% versus 2.8 ± 0.9%, P < 0.001) [s2]. Peak core temperature did not differ between trials (38.5 ± 0.4 °C versus 38.6 ± 0.4 °C, P = 0.346) [s2].

The biomarkers did not differ either. Urinary thioredoxin-1, a marker of oxidative stress, was elevated after exercise and in recovery (P < 0.001) but not different between trials (P = 0.743) [s2]. Monocyte chemotactic protein-1, a marker of inflammation, rose in both serum and urine and again did not differ between trials [s2]. The acute kidney injury risk product [IGFBP7·TIMP-2] was elevated post-exercise and in recovery in both trials, with no difference between them [s2].

The authors state that these findings challenge the assumption that current hydration recommendations are protective, and that additional strategies are needed to mitigate heat-induced kidney injury in occupational settings [s2].

What to take from the pair

Thirteen participants in a laboratory simulation is a small study and a proxy for real occupational exposure, and the authors present it as a challenge to an assumption rather than a refutation of hydration guidance. Nothing in it suggests fluid replacement is unnecessary.

What the two papers jointly undercut is a tidy chain of reasoning that consumer devices encourage: that a watch can tell you how much you lost, that replacing that amount is the goal, and that hitting the number protects you. The first link carries about 30% error [s1], and the third did not hold in a controlled heat-stress trial [s2].

Sources

• Evaluation of Exertional Sweat Loss Estimates in Wearable Technology — Sports Health, 2025-11-30
• Ad libitum drinking does not mitigate acute kidney injury risk nor elevations in markers of oxidative stress and inflammation during simulated occupational heat stress — American Journal of Physiology - Renal Physiology, 2025-11-14