All cause and cause specific mortality associated with transition to daylight saving time in US: nationwide, time series, observational study

Mortality data

We obtained state level mortality data from the National Center for Health Statistics, which had aggregated death counts on a weekly basis from 50 US states and the District of Columbia, from the start of 2015 to the end of 2019.29 Causes of death were identified according to ICD-10 codes (international classification of diseases, 10th revision), including Alzheimer's disease (G30), dementia (F00-F03), circulatory diseases (I00-I09, I11, I13, I20-I51, I60-I69), malignant neoplasms (C00-C97), and respiratory diseases (J00-J06, J10-J18, J30-J39, J40-J47, J67, J70-J98). These conditions were the top five causes of death in the original dataset of the National Center for Health Statistics. The mortality data were grouped by age, and by race and ethnic group. This observational study used secondary aggregated data that were publicly available and thus the study was exempt from ethical approval.

To calculate mortality rates, we obtained data from the US Census Bureau for the mid-year annual population size by age, and by race and ethnic group, for each state, from 2014 to 2020.30 We converted the annual data to weekly data by linear interpolation, where we included data from 2014 and 2020 as the boundary conditions for implementing linear interpolation. Mortality and population data were grouped into six age groups (<25, 25-44, 45-64, 65-74, 75-84, and ≥85 years) and five race and ethnic groups (Hispanic, non-Hispanic Asian, non-Hispanic black, non-Hispanic white, and other race and ethnic groups).

Daylight saving time

Since the official implementation in the US in 1966, two daylight saving time shifting dates were applied to 48 states and the District of Columbia (excluding the states of Arizona and Hawaii), on the second Sunday of March (in the 10th or 11th epidemiological week) and the first Sunday of November (in the 44th or 45th epidemiological week). We considered the periods of interest as eight weeks after the shifting dates for transitions to both spring and autumn daylight saving time (0-7 weeks after either spring or autumn daylight saving time). Apart from these 16 weeks, the rest of the year was considered the control period. The period of interest (0-7 weeks after either spring or autumn daylight saving time) was selected among different periods, ranging from 0-2 weeks to 0-10 weeks, and 0-7 weeks was the longest duration among those with a significant (adjusted P value <0.05) effect size (online supplemental table S32).

Meteorological variables and air pollution assessment

Daily ambient air pollution and meteorological data were obtained from the US Environmental Protection Agency's Air Quality System, including fine particulate matter (PM_2.5), ozone, ambient temperature, relative humidity, and wind speed.31 Based on these daily data, we calculated the state level weekly average data for these air pollutants and meteorological variables by taking an average of all monitoring stations within each state and by each week. Online supplemental appendix section S1 describes the data and variables used in this study.

Statistical analyses

To account for the overdispersion feature, which occurs when count data contain a relatively large heterogeneity, weekly death counts were analysed with fixed effects negative binomial log linear regression models with the natural logarithm of population size as an offset (online supplemental appendix section S2).32 The associations between the weeks after transition to daylight saving time and mortality risks were examined while accounting for long term trends, seasonality, federal holidays, ambient temperature, relative humidity, wind speed, PM_2.5, and ozone. The potential non-linear associations between meteorological factors and mortality were considered by using spline functions with the degree of freedom selected according to the Akaike information criterion score.33 34

We expressed the results of the study in terms of both relative risk and absolute risk. Relative risk was measured by the mortality rate ratio. Absolute risk was measured by the absolute excess mortality rate associated with the weeks after transition to daylight saving time, and was defined as α×(1–1/rate ratio),35 where α=weekly mortality rate for a particular epidemiological week, cause of death, and population subgroup. Under the likelihood based statistical inference framework, the statistical uncertainty of model estimates was assessed by constructing the 95% confidence interval and Benjamini-Hochberg adjusted P values from Wald's tests. We reported both week specific and aggregated estimates of relative risk and absolute excess mortality rate for the period 0-7 weeks after transition to daylight saving time. The aggregated estimates were calculated as the pooled effect size with all week specific estimates (16 weeks, including 0-7 weeks after transition to spring or autumn daylight saving time), which could be interpreted as an average association between daylight saving time and mortality during the period 0-7 weeks after spring or autumn daylight saving time.

To explore spatial heterogenicity and differences in personal characteristics for transition to daylight saving time on the risk of mortality, we conducted subgroup analyses with different fixed effect regression models but the same model complexity (as the main analysis), separately for different subgroups of data, according to cause of death, time zone, age, and race and ethnic group. To assess the robustness of our findings, we carried out a series of sensitivity analyses under various model settings of adjustments of covariables. These adjustments included changing the model flexibility for long-term trends, seasonality, and confounding adjustments for variables. For partial data, we excluded deaths in the weeks after daylight saving time for model fitting. Online supplemental appendix section S3 has the details of the subgroup and sensitivity analyses.

As a negative control, for comparison, we also repeated our main analysis for regions where transitions to daylight saving time were not implemented, including the state of Hawaii and most regions in the state of Arizona. All statistical analyses were performed with R statistical software (version 4.2.1).36 Results were considered significant when the Benjamini-Hochberg adjusted P value was <0.05.

Patient and public involvement

Because this study used deidentified mortality data, which cannot be used to identify individuals, no patient or member of the public was involved in implementing the study design. Public data (released by the US government) were used in this study, and thus patients and the public were not involved in the research. We have no plans to disseminate the results of the research directly to study participants.

Principal findings

We investigated the relation between transition to daylight saving time and all cause and cause specific mortality in the US, based on a nationwide dataset. The dataset used in this study had nearly 14 million reported deaths across different states, ages, and race and ethnic groups, over a five year period (2015-19). Our results suggested a slight but non-significant increase in mortality after spring daylight saving time, but a substantial decrease in mortality after autumn daylight saving time. Although the increase in mortality after spring daylight saving time was not significant, similar positive associations were reported in the literature for a range of severe medical conditions.8 13 The non-significant rate ratio estimates after spring daylight saving time were likely because of low time resolution (ie, weekly rather than daily data). In addition, we found a substantial decrease in mortality after autumn daylight saving time for different causes of death. The protective effect of the transition to autumn daylight saving time on the risk of mortality was more pronounced in elderly people aged ≥75 years, in the non-Hispanic white population, and in those residing in the eastern time zone.

Comparison with other studies

We found a substantial decrease in mortality after autumn daylight saving time for different causes of death. This finding seems to be consistent with the literature that assumed that switching back to standard time in autumn could have no effect,1 8 10 or a protective effect against diseases, unintentional injuries, and mortality.6 14 16 22 37 Moreover, we found that the combined effect of transitions to both spring and autumn daylight saving time was a decrease in the risk of mortality but the decrease was not significant. Similar to a previous study that investigated general mortality risks after daylight saving time in 16 European countries,28 our finding was also in contrast with the general scientific postulate in the daylight saving time literature that minimal sleep deprivation might lead to an increase in the risk of mortality.8 20 38

We investigated changes in mortality during a relatively long period, up to eight weeks after the transition to daylight saving time. Although not univocally stated, changes in the risk of adverse health conditions was suggested to be greatest on the transition day (Sunday) and the following days within one week. Few studies, however, have investigated the health consequences after a minimal shift in circadian rhythm (eg, daylight saving time) compared with a relatively long term shift. The effects of daylight saving time might be shortlasting for cardiovascular diseases,1 8 11 and we also reported a generally decreasing trend in the rate ratio for mortality from circulatory diseases as the weeks increased after autumn daylight saving time. For deaths caused by dementia, malignant neoplasms, and respiratory diseases, however, we detected a decrease in the risk of mortality until six weeks after autumn daylight saving time. We also found an increasing trend for mortality from dementia during the first five weeks after spring daylight saving time, including a rate ratio of 1.050 (95% confidence interval 1.007 to 1.095) in the fifth week, although the underlying mechanisms were unclear. These findings suggest that the effect of transition to daylight saving time on mortality could last longer than one week for a range of causes.

For different underlying causes of death, weak but significant decreases in mortality risk were detected during 0-7 weeks after autumn daylight saving time for dementia, circulatory diseases, malignant neoplasms, and respiratory diseases (table 1). Studies in recent decades have frequently reported changes in risk in the onset and severe outcomes of daylight saving time for mental or behavioural disorders,6 39 and cardiovascular diseases,5 7 9–12 including circulatory deaths.40 Our findings suggested that transition to daylight saving time might have an important role not only on mental health and cardiovascular conditions, but could also be associated with changes in the mortality risks of other non-accidental diseases, including respiratory and neoplastic diseases. Positive associations were found between a mild shift in circadian rhythm (not necessarily by daylight saving time) and the risk of various pulmonary diseases41 42 and cancers.43–46 A slight increase in mortality from cancer associated with daylight saving time was found in the first week after transition to spring daylight saving time, and thus extra care might be helpful for patients with cancer during spring daylight saving time.

To explore spatial heterogenicity, we compared mortality risks after daylight saving time for different time zones, an important research gap in investigations of daylight saving time recognised by the American Academy of Sleep Medicine in 2020.25 We used the Hawaii time zone as a negative control because daylight saving time was not implemented in Hawaii, and consistently, we detected no effects of spring or autumn daylight saving time. Similarly, daylight saving time was not implemented in most regions in Arizona, and we found no evidence of any difference between observed and expected mortality rates. For the six time zones in the US, we detected evidence indicating a decrease in all cause mortality only after autumn daylight saving time in the eastern time zone. The eastern time zone in the US is the closest to coordinated universal time (UTC), where sunset (ie, the start of daytime) occurs earlier than in other time zones. Although the exact mechanisms were not clear, our findings suggest that the changes in risk of mortality associated with daylight saving time might be influenced by the eastward time zone in the US. Considering that 47.6% of the continental US population live in the eastern time zone, however, which also had the highest general mortality rate among all six time zones, the only decreased mortality risks after autumn daylight save time in eastern time zone might also benefit from large population size and high mortality rate, leading to a relatively higher statistical power. Hence further study is needed to verify our results.

For differences in personal characteristics, we found that deaths associated with daylight saving time were disproportionate for different ages and different race and ethnic groups. The effects of shifts in daylight saving time on general mortality risks were strongly increased by age. We found that the protective effects on mortality associated with autumn daylight saving time increased with age, and these protective effects were more pronounced in people aged ≥75 years. Between 2015 and 2019, 3 287 963 (23.6%) and 4 331 998 (31.1%) of 13 912 837 deaths were reported among those aged 75-84 and ≥85 years, respectively. Although we are not aware of any study that has reported on the age modifying effects of mortality associated with daylight saving time, elderly people could be more susceptible to the effects of daylight saving time on adverse health outcomes.6 12

Also, despite the non-significant outcome of the trend test, we found a steadily increasing trend in rate ratio estimates for spring daylight saving time as age increased. The non-significant results for both rate ratio estimates and trends for rate ratio were likely because of the low time resolution of the time series data used in the statistical analysis. Healthcare services are likely to be used more frequently by elderly patients with critical conditions after the transition to daylight saving time in spring. Acknowledging the non-significant rate ratio estimates after transitions to either spring or autumn daylight saving time, the trend for the rate ratio suggested that the overall risk of mortality was more likely to be decreased, rather than increased, after combining the effects of transitions to spring and autumn daylight saving time.

Assessing mortality risks after daylight saving time at the general population level was an important contribution to the strategic decision making process about the future of the policy on daylight saving time. Compared with recent studies in Europe,28 38 we found that the mortality patterns associated with daylight saving time in the US were different from those in Europe, especially after the transition to autumn daylight saving time. Given the conflicting results in recent studies of mortality associated with daylight saving time based on population level death records, as well as many other studies with different study designs, a discussion of the topic based on different sources of datasets and from different angles might be useful and timely.

Our findings indicated a change in all cause mortality associated with transition to daylight saving time in the general US population. We found that changes in mortality risks associated with daylight saving time were more pronounced in elderly people, in the non-Hispanic white population, and in those living in the eastern time zone. For these subgroups of individuals with critical conditions, improvements in care and drug treatments for a two month period after the spring daylight saving time, and preventive measures to improve health status for a short period before the spring daylight saving time, might be helpful to prevent excess deaths. For different causes of death, ages, race and ethnic groups, and time zones, the decrease in mortality risks after autumn daylight saving time was more pronounced than the increase after spring daylight saving time.

In the US, a bill entitled the Sunshine Protection Act (https://www.congress.gov/bill/118th-congress/senate-bill/582) was passed unanimously by the Senate in 2022 (but not by the House as yet), which proposed permanent daylight saving time in the US. From the general public perspective, large differences in attitudes towards eliminating changing the clocks were highlighted in two recent US public opinion polls in 2021 (https://today.yougov.com/politics/articles/39209-daylight-saving-time-americans-want-stay-permanent) and 2022 (https://apnorc.org/projects/dislike-for-changing-the-clocks-persists/). Regardless of the health related effects of transition to daylight saving time, increasing concerns exist about energy savings because electricity consumption shifts from lighting to transportation, commercial, and entertaining events, and hence daylight saving time contributes little to saving energy in the modern society. Therefore, discussion of the potential (or possible) effects of changing the policy of daylight saving time is timely, necessary, and of general interest. Although our findings suggested that abolishing daylight saving time might not be supported by the evidence of overall change in mortality associated with daylight saving time, future policies on daylight saving time require more solid evidence at the population level, and extensive understanding of the underlying mechanisms.

Limitations

Our study had several limitations. Because daylight saving time might affect health by affecting circadian rhythm,22 the absence of sleep parameters (eg, sleep quality, duration, and sleep-wake cycle) and chronotypes restricted our analysis without controlling for different types of biological responses to transitions to daylight saving time.47 Lack of adjustment for these factors might cause less precise estimates, but not bias.

The low time resolution of the time series data is a limitation because of loss of information on changes in the risk of mortality and the relatively low statistical power of the study. Because shifting of daylight saving time is only a minimal shift in circadian rhythm, time series data aggregated on a weekly basis could not reflect or adjust for a daily or even hourly distribution of mortality risk before versus after transition to daylight saving time, which was also noted in another time series study investigating daylight saving time.28 Mortality might be highest on the transition day to daylight saving time as well as in the following few days, with a sharp reduction in the first week, especially for cardiovascular mortality and road traffic incidents. Because this weekday-weekend pattern was not included in our study, mean weekly mortality was unlikely to be sufficiently sensitive (ie, with a relatively low statistical power) to identify associations between mortality and the shift in daylight saving time. Thus we speculate that significant increases in mortality might have been detected with the statistical methods in this study if daily or hourly time series data of deaths had been available. The low time resolution could also reduce the detection of differences in mortality risks between the east and west regions within one time zone, where this east-to-west difference might only exist for a relatively short time after the transition to daylight saving time.16 For the US, future studies might investigate changes in cause specific hospital admissions and mortality data associated with daylight saving time in Medicare and Medicaid databases with a daily or hourly based time resolution.

Although we assessed only five different causes of death in this study, circulatory diseases, malignant neoplasms, and respiratory diseases were the leading causes of death in the US, as well as in many other countries across the world, and were broadly reported among the general population. Identifying more specific causes of death that might be associated with shifts in daylight saving time are needed for further investigation.

The original dataset released by the National Center for Health Statistics covered the time period up to 2021, but we excluded data from 2020 onwards so that the effects of the covid-19 pandemic could be removed from the study. The information was collected from the whole of the US population, but the dataset used for analysis in this study was from a five year observation period, which was relatively short for a time series study. Hence further investigation with a longer study period might be needed to verify our findings.

This observational study was based on aggregated time series data, and thus causation cannot be inferred from our findings because the results might be confounded by unmeasured factors. Further studies are needed to explore the detailed relation between shifts in daylight saving time and the risk of mortality, as well as the underlying mechanisms, when data are available.

Conclusions

Transitions in daylight saving time were associated with changes in mortality in the US for different underlying causes. We found that the risk of mortality was increased, but not significantly, after the transition to spring daylight saving time and substantially decreased after transition to autumn daylight saving time, resulting overall in no effect on deaths associated with combined transitions to daylight saving time. Our findings might inform the ongoing debate on the policy of shifting daylight saving time.