Circadian activity rhythms and mortality: the Study of Osteoporotic Fractures

Participants

SOF is a longitudinal epidemiologic study of 10,366 women age 65 or older, recruited from four study centers located in Baltimore, MD; Minneapolis, MN; Portland, OR; and the Monongahela Valley near Pittsburgh, PA. Women were excluded if they had a bilateral hip replacement or were unable to walk without assistance. The baseline SOF exams were conducted from 1986-88, where 9,704 Caucasian women were recruited (30). SOF was originally designed to investigate risk factors for osteoporosis and osteoporotic fractures, and initially African-American women were excluded from the study due to their low incidence of hip fractures, but from February 1997 to February 1998 662 African-American women were enrolled (31). At all subsequent visits, no exclusion criteria were used. All participants were community-dwelling at baseline. Since then follow-up exams have taken place approximately every two years.

The focus of this analysis was data gathered at SOF Exam 8, which took place between January 2002 and February 2004. Of the 4727 women at this visit, 3676 (78%) participants with clinic or home visits were eligible for collection of wrist actigraphy data. Eligibility was based on having a home visit, participant willingness and overall ability to perform the study. Of the 3,676 participants with a home or clinic visit, 12% refused, had advanced frailty or cognitive problems and were deemed ineligible by the study staff. The success rate for those given actigraphs was 94%. The patient flow diagram gives more information on the reasons for missing actigraphy data (Figure 1). Of those who were eligible to participate in the actigraphy study: <1% had an actigraph malfunction, <1% had a software or initialization problem, 1.3% removed the actigraph and did not replace it, 2.4% did not have adequate proportional integration mode (PIM) mode data (which computes movement as counts per minute based on an area under the curve analysis that takes into account both intensity and frequency of movement), and <1% did not have full 24-hour PIM data to calculate the rest-activity parameters. The institutional review boards on human research approved the study at each institution, and all participating women provided written informed consent.

Actigraphy

Activity data were collected with the Sleep-watch-O (SleepWatch-O ®, Ambulatory Monitoring, Inc) actigraph, a small device worn on the wrist. Movement is measured by a piezoelectric linear accelerometer (sensitive to 0.003 g and above), which generates a voltage each time the actigraph is moved. These voltages are gathered continuously and summarized over 1-minute epochs. The actigraph was initialized in the clinic prior to the visit, and placed on the participant’s nondominant wrist by the examiner during the visit. Women wore the actigraphs continuously for a minimum of three 24-hour periods (i.e., 72 hours).

An extension to the traditional cosine model was used to map the circadian activity rhythm to the activity data (32). This extended cosine model applies a non-linear transformation to the cosine curve, the anti-logistic function, and is sometimes referred to as a 5-parameter extension of the 24 hour cosine curve. Activity data often assumes a shape more similar to a squared wave than a cosine curve and this extension to the traditional cosine curve fits each individual’s data and allows for this shape. Circadian activity parameters of the extended cosine model were calculated using nonlinear least squares. The following activity rhythm parameters were calculated from the extended cosine curve: Amplitude, an indicator of the strength of the rhythm, is the peak to nadir difference in activity (measured in arbitrary units of activity [counts/min]); Mesor, mean level of activity (measured in arbitrary units of activity [counts/min]); Robustness of the circadian actvity rhythm (pseudo-F statistic for goodness of extended cosine fit; higher pseudo-F values indicate stronger rhythms); and Acrophase, timing of peak activity measured in portions of hours (time of day). Circadian amplitude, mesor and robustness were examined based on quartile distributions. Acrophase was examined in terms of the deviation from the population mean. We identified three categories based on having a peak time of more than 1.5 standard deviations (SDs) above and below the population mean for the study population. Phase advanced participants were defined as having an acrophase of <12:50PM (−1.5 SD from the mean) and phase delayed participants were defined as having an acrophase of >4:33PM (+1.5 SD from the mean).

Mortality

Vital status after visit 8, with cause of death verified through death certificates, was ascertained during an average of 4.1+1.1 years of follow-up. Participants were contacted by postcard or telephone every 4 months to ascertain vital status. Information from designated proxy sources (e.g., family member or a close friend) was used if the participant has died. Follow-up since the baseline visit has remained over 95% complete. Deaths were confirmed by death certificates. Causes of death were confirmed by death certificates, and when available, hospital discharge summaries. The median follow-up period was 4.3 years (range of 5 days to 5.2 years). We used International Classification of Diseases (ICD)-9 codes to classify causes of death as coronary heart disease (codes 410–414), stroke (codes 430–438), atherosclerosis (codes 401–404, 410–414, 425, 427.5, 428, 429.2, 430–438, 440–444, and 798), cancer (codes 140–239), or all other causes (non-cancer and non-atherosclerotic deaths). There were insufficient numbers of specific other causes of death to analyze these as separate outcomes. The most common causes of death in the ‘other’ category were pulmonary (n=38; codes 415–417.9, 460–529.9, 786, 796, 799.1) and cognitive-related (n=17; codes 290–290.9, 331–331.9, 332–332.1).

Other Measurements

All participants completed questionnaire data, which included questions about medical history, self-reported health, smoking status, alcohol use, caffeine intake, married status, and whether or not the participant walked for exercise. The Geriatric Depression Scale (GDS) was used to assess depressive symptoms, with the standard cutoff of ≥6 symptoms used to define depression (33). Medication use was ascertained by asking participants to bring all current prescription and nonprescription medications used in the past 30 days to their clinic visits. For women who completed a home visit, the interviewer gathered medication use information at the home. A computerized medication coding dictionary was used to categorize all medications (34). The Mini-Mental State Examination (MMSE) was administered to assess cognitive function, with higher scores on a scale of 0 to 30 representing better cognition (35). Functional status was assessed by collecting information on 6 instrumental activities of daily living (IADL), which included walking 2 to 3 blocks on level ground, climbing up to 10 steps, walking down 10 steps, preparing meals, doing heavy housework, and shopping for groceries or clothing. Participants were asked if pain was making sleeping difficult. Body weight and height were measured, and body mass index (BMI) was calculated as weight in kilograms divided by the square of height in meters. A history of cardiovascular disease was defined as a prior diagnosis of myocardial infarction, angina, congestive heart failure, or other heart disease. A history of medical conditions was defined as a prior diagnosis of stroke, diabetes, Parkinson’s disease, Alzheimer’s disease, COPD, cancer or cardiovascular disease. All measurements were collected at visit 8.

Statistical analysis

Characteristics known to be related to activity rhythms or mortality (36) were summarized using means and SDs for continuous data and percentages for categorical data. We compared characteristics among categories of amplitude and acrophase using ANOVA for continuous covariates that were normally distributed, Kruskal-Wallis tests for skewed continuous data, and chi-square tests for categorical data. To determine the relationship between circadian activity rhythms and mortality, we used Cox proportional hazards models to estimate hazard ratios (HRs) and 95% confidence intervals (CIs). To identify potential confounders, we considered a list of predictors thought to be associated with circadian activity rhythms and mortality, based on biological plausibility or previous studies, including age, BMI, cognitive function, self-reported health, number of IADLs, physical activity, depression, pain, race, benzodiazepines and antidepressant use, sleep medication use, alcohol use, smoking status, caffeine intake, married status, and prior medical conditions. Variables that were significantly related (p<0.10) to at least one activity rhythm predictor measure and all-cause mortality were included in the final multivariate analyses (age, clinic site, race, BMI, cognitive function, walking for exercise, IADL impairments, depression, current use of benzodiazepines or antidepressants, alcohol use, smoking status, self-reported health status, married status, and co-morbidities). Multivariate-adjusted Kaplan-Meier curves were generated to assess the cumulative incidence of all-cause mortality. Additional analyses were performed to determine if any association found between peak and mean activity levels and all-cause mortality were independent of physical activity level. Models were stratified by whether or not walking for exercise was reported, and formal interactions between walking for exercise and mesor and amplitude were performed. Statistical analysis was performed using the statistical software program SAS version 9.1 (SAS Institute, Inc., Cary, NC).

Characteristics of the study population

The analysis cohort was composed of 3027 women (mean age 84 ± 4 years, range 77–99 years). We previously compared characteristics between this analytical data set of women with actigraphy data and the remaining women who did not have actigraphy measured 21. The group without actigraphy was on average 1.4 years older, had a slightly lower percentage of African Americans, had a slightly higher prevalence of several health conditions and experienced an almost 2-fold increased mortality rate compared to women with actigraphy measurements. We compared the 192 women who were given the actigraph but had unusable activity rhythm data to those in the analysis subset. Those with unusable data had slightly lower MMSE scores, 27.3 vs. 27.9 (p=0.0003), and shorter follow-up times, 3.85 years vs. 4.05 years (p=0.009).

We compared characteristics among quartiles of amplitude (Table 1). In general, women in the lowest quartile of amplitude were more likely to be older and have a high BMI, more IADL impairments, more medical conditions, poor health and more likely to be taking antidepressants and to smoke. Characteristics of the lower mesor and less robust groups (Appendix) were similar to those for amplitude; the delayed acrophase group also was largely similar to amplitude with younger age being the major exception (Table 2).

Circadian activity rhythms and mortality

After a mean of 4.1 years of follow-up, 444 (15%) of the analytic sample had died. Those women with lower peak (amplitude) and mean (mesor) levels of activity and less robust rhythms (lower pseudo-F values indicate weaker rhythms) had the shortest overall survival. Amplitude was highly correlated with both mesor (r=0.78) and robustness (r=0.72) and the results for these measures were very similar. Compared to the highest quartiles, an approximately 2-fold adjusted higher risk of mortality was observed for those in the lowest quartiles of amplitude, mesor, and robustness value after multivariate adjustment (Table 3). The risks of all-cause mortality increased from highest to lowest quartile of amplitude (p-trend, p<0.0001).

An increased risk of atherosclerotic mortality was also observed for the lowest quartiles of amplitude and mesor activity levels as well as the lowest quartile of robustness (Table 3). The relationships with atherosclerotic mortality were largely driven by associations between circadian activity rhythms and either CHD or stroke mortality, which comprised 34% and 32% of atherosclerotic mortality cases, respectively. Increased risk of CHD mortality was observed in the lowest quartile of mesor. Robustness was associated with stroke mortality with an increased risk for the lowest quartile compared with the highest quartile.

Inverse associations between amplitude, mesor, robustness value and higher “other-cause” (non-atherosclerotic / non-cancer) mortality rates were also observed. Increased risk of “other-cause” mortality was observed in the lowest quartiles of amplitude, mesor, and robustness, compared with the highest quartiles. Further analysis excluding the two most common causes of “other cause” mortality (pulmonary and cognitive causes) produced similar results, suggesting that these two causes of death do not explain this association.

Acrophase deviation was not associated with all-cause or “other cause” mortality. A delayed acrophase (timing of peak activity >4:33PM) was associated with increased cancer and stroke mortality when compared to the mean peak range of 12:50PM–4:33PM. An increased but non-significant association was observed between a delayed acrophase and atherosclerotic mortality when compared to the population mean. An advanced acrophase (timing of peak activity <12:50PM) was associated with an elevated but non-significant association risk of stroke mortality.

We performed additional analyses to determine if the associations between amplitude, mesor and all-cause mortality were independent of physical activity level. The associations between amplitude, mesor and all-cause mortality were consistent among women who did and did not walk for exercise and the interactions (amplitude*exercise, p=0.51; and mesor*exercise, p=0.67) were not significant. Interactions between mesor and three ambulatory IADLs (difficulty walking 2–3 blocks, climbing up 10 steps, and climbing down 10 steps) were not significant for all-cause mortality (mesor*walking, p=0.15; mesor*up 10 steps, p=0.14; and mesor*down 10 steps, p=0.16). The only statistically significant interaction detected was between “difficulty walking 2–3 blocks” and amplitude for all-cause mortality (p=0.04). Participants in the lowest quartile of amplitude were still at increased risk of all-cause mortality whether they reported having difficulty (HR=3.07, 95% CI, 1.80–5.22) or not having difficulty (HR=1.62, 95% CI, 1.09–2.42) walking 2–3 blocks. The interactions between difficulty climbing steps and amplitude were not significant for all-cause mortality (amplitude*up 10 steps, p=0.09; and amplitude*down 10 steps p=0.25). These results suggest that these circadian parameters are not simply markers of low physical activity which increases the risk for death.

Circadian rhythms and sleep are influenced by circadian and homeostatic processes. Our study measured activity levels over several 24-hour periods which allowed us to specifically examine circadian activity rhythms and timing of activity independent of sleep. Indeed, sleep duration was not significantly correlated (P<0.05) with amplitude (r=0.05), mesor (r=−0.23), robustness (r=0.26) or acrophase timing (r=−0.03). Adding total sleep time or sleep efficiency to the models did not change the associations between circadian activity parameters and mortality (data not shown).

Kaplan-Meier survival curves show the cumulative incidence of all-cause mortality (Figure 2). There was evidence for a time trend of mortality in the lowest quartiles of amplitude, mesor, and robustness (Figure 2). The Kaplan-Meier curves diverged at approximately the first year and continued through 4 years with more events in the lowest quartile compared to the top three quartiles (Figure 2). For acrophase, the Kaplan-Meier curves diverged at approximately the first year and continued through 4 years with more events in the delayed acrophase group compared to those in the mean or advanced acrophase groups (Figure 2).