← All articles
An adult seen from behind walking away on a wide concrete city sidewalk at bright midday, ordinary dark coat, hard noon light, no signs in the foreground

The busiest minute of the day lined up with fewer deaths

For a reader who already moves on ordinary days and still cannot tell which part of that day is worth a conversation with a doctor. Inside is a comparison of people as they already were, not a plan for the next week.

In the German National Cohort, 65,028 adults wore a sensor on the right hip for seven days. They were 19 to 74 years old when the week was recorded, between 2014 and 2019. The mean age was 50.0 years, with a standard deviation of 12.6, and 52.1% were women. Death records then ran for a median of 8.0 years: 1,634 deaths, 514,297.5 person-years, and 3.2 deaths per 1,000 person-years. Patricia Fromherz and Michael J. Stein contributed equally. The paper was published online in BMC Medicine on 10 November 2026.

This material is informational and does not replace a conversation with a doctor.

What the hip sensor counted

The devices were ActiGraph models GT3X+, wGT3X+ and wGT3X-BT, worn on the right hip and sampling at 100 Hz. The team turned the raw trace into ENMO with GGIR version 3.2-9. ENMO is the Euclidean norm of acceleration minus one, reported in milli-gravity. A milli-gravity is a unit of acceleration. It is not a milligram of any substance.

Volume is the average acceleration across the 24-hour period, averaged over the whole measurement. Duration is the minutes per day of movement that was neither sedentary nor sleep. The intensity gradient is the distribution of time across light, moderate and vigorous intensity. A less negative value means a greater share of high-intensity activity. Mx is the average acceleration during the most active 1, 5, 15, 30, 60 or 120 minutes, whether those minutes sat together or were scattered through the day. The continuous version uses the most active uninterrupted bout of the same length, and it records the acceleration that at least three quarters of that bout stayed above.

A day counted only with at least 16 hours of recording. A person stayed in the analysis with at least two weekdays and one weekend day. Of 72,430 people who wore a sensor, 7,397 had a trace that failed that bar and 5 had no mortality record, which left 65,028. The parent study enrolled 205,415 adults at 18 centres. The baseline response was 15.6% of those invited.

Three pictures of the same week

The models are Cox models with age as the time scale. Each hazard ratio below compares people at the 50th or the 75th percentile of a measure with people at the 25th percentile. It is not the result of adding a minute to someone's day, and it is not a figure a reader can claim for themselves.

Average acceleration lined up with death along a straight slope. The test for a bend gave p = 0.127. Against 9.2 milli-gravity at the 25th percentile, 11.2 milli-gravity had a hazard ratio of 0.87 (95% confidence interval 0.83 to 0.90), and 13.7 milli-gravity had 0.73 (0.67 to 0.80). Each extra 1 milli-gravity of average acceleration had a hazard ratio of 0.93 (0.91 to 0.95). The authors describe 1 milli-gravity as roughly several extra minutes of brisk walking, and they say the translation depends on the movement and on how the device is calibrated. That is their bridge to everyday language. It is not a step target measured in this dataset.

Duration bent, with p for non-linearity below 0.001. Against 265.4 minutes a day, 321.6 minutes had a hazard ratio of 0.85 (0.81 to 0.88), and 383.3 minutes had 0.77 (0.72 to 0.82). Past the middle of the distribution the curve flattened.

The intensity gradient was a shallow U, again with p below 0.001. Against −2.91, the value −2.74 had a hazard ratio of 0.93 (0.90 to 0.96). Against the same reference, −2.56 had 0.98 (0.91 to 1.06), an interval that includes no difference. A less negative gradient was not uniformly lined up with death, and the upper end of the curve rests on few people.

The single most active minute is a different cut. People at 247.2 milli-gravity in that minute, versus 171.3, had a hazard ratio of 0.58 (0.53 to 0.63). The same comparison across the most active 120 minutes, 44.3 versus 26.8 milli-gravity, was 0.69 (0.65 to 0.74). When that one minute had to be continuous, the hazard ratio was 0.71 (0.66 to 0.77), at 187.2 versus 117.2 milli-gravity. The methods say these intensity and accumulation models were also adjusted for total volume. The footnote of Table 2 lists the other covariates and does not restate volume.

Associations for continuous bouts differed by total activity volume

The authors split the cohort at the median volume, 11.2 milli-gravity, and looked at continuous bouts. At 1, 5 and 15 continuous minutes the interaction with volume was detectable, with p = 0.001, 0.014 and 0.022. The inverse association was stronger among people below that median. At 30, 60 and 120 continuous minutes the interaction was not detectable, with p = 0.097, 0.357 and 0.367. In the lower-volume group the curves fell and then rose slightly at the highest intensities.

They also split the most active minutes by how broken up they were. An Mx-ratio below 0.5 means those minutes were mostly scattered. In that intermittent group the inverse association ran from 1 minute through 120. At a ratio of 0.5 or higher, a more continuous block, the association was no longer clear at about 60 minutes.

After a correction for many tests, the activity measures did not interact with the covariates. Volume did not interact with the intensity gradient (p = 0.105). Leaving out the first two years of follow-up removed 123 cases. Leaving out 4,162 participants who already had cancer at the start, apart from non-melanoma skin cancer, or taking cardiovascular disease, diabetes and body mass index out of the adjustment, left the direction unchanged. Those checks are in the supplement. The proportional-hazards assumption held for the activity exposure.

What one week on the hip cannot say

The sensor ran once, at the start. Retirement, illness and a change of habit after that week are invisible. A hip device may underestimate cycling, swimming and work done with the arms. It also does not distinguish occupational, household, transport and leisure-time activity. The authors note that occupational activity has been linked with worse outcomes in other studies. Adjustment for employment and socioeconomic status is not the same as knowing which domain the movement came from.

In age-standardised baseline figures, the highest volume quartile averaged 16.5 milli-gravity and 406.6 minutes a day, against 7.8 milli-gravity, 242.7 minutes and 7.8 sedentary hours in the lowest. Higher volume went with less smoking, a lower mean body mass index, less reported cardiovascular disease and diabetes, and more employment, and also with more people over the study's alcohol threshold. Health and fitness can still sit behind the association. The response was 15.6% of those invited. The authors argue from other cohorts that this participation rate is unlikely to have changed the associations much. That argument is not a measurement in this dataset. Cause of death was not split.

The authors state that this analysis received no funding. The cohort itself is funded by the Federal Ministry of Research, Technology and Space (BMFTR), the federal states, the Helmholtz Association, the participating universities and Leibniz institutes, under application NAKO-955. Michael J. Stein is a member of the BMC Medicine editorial board. Fabian Bamberg discloses consultancy, a speakers bureau and an unrestricted research grant from Bayer Healthcare, and an unrestricted research grant from Siemens Healthineers. The paper says the remaining authors declare no competing interests. The design is observational. The authors write that the associations should not be read as causal.

What to take into a conversation with a doctor

The useful question is not a minute target. It is whether a week a sensor would see is mostly a long quiet day, a day with one intense burst, or a day broken into short pieces, and whether that pattern is a job, a commute or a choice. A hip trace from a single week is a description of people who already differed. It is not a prescription. Training plans belong in a conversation with a doctor, not in a percentile from this cohort.

Sources

Fromherz P, Stein MJ, Lendt C, et al. Daily physical activity volume, intensity, and accumulation patterns in relation to all-cause mortality in the German National Cohort (NAKO). BMC Medicine, online 10 November 2026. PMID 43015814. DOI 10.1186/s12916-026-05385-8.

Articles in this section are educational and are not medical advice, a diagnosis, or a prescription. Consult a qualified professional before acting on anything you read here.

Read next