Cardiometabolic thresholds for peak 30-min cadence and steps/day
Cardiometabolic thresholds for peak 30-min cadence and steps/day
Bryan Adams 0 1
Katie Fidler 0 1
Noah Demoes 0 1
Elroy J. Aguiar 1
Scott W. DucharmeID 1
Aston K. McCulloughID 1
Christopher C. MooreID 1
Catrine Tudor-Locke 1
Diana ThomasID 0 1
0 Department of Mathematical Sciences, United States Military Academy, West Point, New York, United States of America, 2 Department of Kinesiology, School of Public Health and Health Sciences, University of Massachusetts Amherst , Amherst, Massachusetts , United States of America
1 Editor: Adewale L. Oyeyemi, University of Maiduguri College of Medical Sciences , NIGERIA
Funding: The authors received no specific funding
for this work.
To provide empirically-supported thresholds for step-based intensity (i.e., peak 30-min
cadence; average of the top 30 steps/min in a day) and steps/day in relation to
cardiometabolic health outcomes.
Receiver operating characteristic curve analysis was applied to the National Health and
Nutrition Examination Survey (NHANES) 2005?2006 accelerometer-derived step data to
determine steps/day and peak 30-min cadence as risk screening values (i.e., thresholds) for
fasting glucose, body mass index, waist circumference, high blood pressure, triglycerides,
and HDL cholesterol. Thresholds for peak 30-min cadence and steps/day were derived that,
when exceeded, classify the absence of each cardiometabolic risk factor. Additionally,
logistic regression models that included the influence of age and smoking were developed using
the sample weights, primary sampling units (PSUs), and stratification variables provided by
the NHANES survey. Finally, a decision tree analysis was performed to delineate criteria for
at-risk versus healthy populations using cadence bands.
Peak 30-min cadence thresholds across cardiometabolic outcomes ranged from 66?72
steps/min. Steps/day thresholds ranged from 4325?6192 steps/day. Higher thresholds
were observed in men compared to women. In men, higher steps/day thresholds were
observed in age ranges of 30?39, while in women, higher thresholds were observed in the
age-range 50?59 years. Decision trees for classifying being at low risk for metabolic
syndrome contained one risk-free leaf at higher cadence bands, specifically for any time
accumulated at 120 steps/min.
Minimum thresholds representing absence of cardiometabolic risk range from 4325?6192
steps/day and 66?72 steps/min for peak 30-min cadence. Any time accumulated at 120
Competing interests: The authors have declared
that no competing interests exist.
Abbreviations: AUC, area under the curve; BMI,
body mass index; PAM, physical activity monitor;
ROC, receiver operating characteristic.
steps/min was associated with an absence of cardiometabolic risk. Although based on
cross-sectional data, these thresholds represent potentially important and clinically
interpretable daily physical activity goals.
The number of cases of metabolic syndrome in U.S. adults are on the rise, with a current
estimated prevalence of 35% [
]. Since the presence of metabolic syndrome is known to increase
the risk of cardiovascular events and all-cause mortality [
], there is a growing need for the
prevention, diagnosis, and treatment of this chronic disease.
Physical inactivity is an established modifiable risk factor for metabolic syndrome [
response, national physical activity guidelines have been developed communicating the dose
(e.g. volume, frequency, and intensity) associated with preventing and treating metabolic
]. The 2018 Physical Activity Guidelines Advisory report [
] specifically targeted
daily step counts (steps/day) as a publicly consumable metric for measuring and prescribing
physical activity volume. Additionally, a strong relationship between cadence (steps/min) and
absolutely-defined physical activity intensity (e.g., metabolic equivalents; METs), has been
demonstrated in laboratory-based studies (r = 0.93) [
]. The application of cadence combined
with daily step counts is particularly attractive since it is easily tractable from the same
timestamped accelerometry devices used to report daily step counts [
]. Cadence patterns can be
further distilled by averaging the cadence values of the 30 highest (but not necessarily
consecutive) minutes in a day, and averaging these values over one week, providing an index metric
known as ?peak 30-min cadence? [
]. We note that the NHANES step data are reported as step
counts accumulated in 1-min epochs (i.e., steps per minute) and represents an average cadence
over a one-minute interval and thus is only an approximation of instantaneous cadence [
The peak 30-min cadence metric represents not only the best effort intensity for a given
day, but also the persistence of highest-intensity behavior across a week. Together, steps/day
and peak 30-min cadence can provide simple and understandable translations of physical
activity volume and intensity measurements, which may then be associated with risk of
developing chronic diseases, including metabolic syndrome.
Reliance on step data to deliver prescriptions requires rigorous data driven analysis that
provide steps/day and peak 30-min cadence thresholds associated with specific health
]. Some dose-response relationships have been identified between steps/day and
metabolic syndrome [
], but to date analysis strategies have solely focused on daily steps counts.
Here, using the National Health and Nutrition Examination Survey (NHANES) 2005?2006
dataset, we for the first time use a receiver operating characteristic and decision tree analysis to
derive clinically interpretable thresholds for steps/day and peak 30-min cadence for numerous
cardiometabolic risk factors. The derived thresholds provide minimum bounds on steps/day and
peak 30-min cadence associate with the absence of each risk factor. We also identify time spent
in sedentary to higher intensity activities [
] that are associated with complete absence of any
risk factor which can also be thought of as a threshold goal associated with ideal health status.
The participant data included in this analysis were sourced from the NHANES 2005?2006
Physical Activity Monitor (PAM) dataset. The purpose of NHANES is to assess the health,
2 / 14
nutritional and physical activity status of noninstitutionalized adults and children in nationally
representative sample from the United States. A detailed description of the NHANES PAM
protocols are available online (http://www.cdc.gov/nchs/data/nhanes/nhanes_05_06/BM.pdf),
and a catalog of variable definitions and data treatment rules has been assembled and reported
]. All NHANES protocols were approved by The National Center for Health
Statistics ethics review board. Participants were required to provide informed consent. In the
2005?2006 survey cycle, physical activity was objectively measured using the hip worn
ActiGraph 7164 accelerometer (ActiGraph, Ft Walton Beach, Florida). In addition to the more
conventional activity count output, the ActiGraph 7164 also provides an enumeration of steps
taken (stored in 1-min epochs), allowing for the calculation of steps/day, peak 30-min cadence
and time spent in different cadence bands (defined below).
Accelerometry. The NHANES 2005?2006 database was prepared for analysis using the
software package R (R Core Team (2013)). We first restricted the database to adults 18 years
or older. Non-wear time was defined as 60 consecutive minutes of zero accelerometer counts/
min, with a valid day defined as 10 hours [
]. Any participant without 4 valid wear days
] was removed from the dataset. The NHANES protocol requested participants wear the
accelerometer for a 7-day period and in this way ensure that a weekend is included. Restricting
to a minimum of 4 valid wear days may lose weekend wear time in some of the subjects. Using
the standard definition of non-wear time (i.e. 60 consecutive minutes of zero counts/min [
The R package ?dplyr? was used to remove all non-wear time from the database. The
consecutive 60 minutes of non-wear time was determined by the accelerometer count equaling zero
with a rolling sum of the difference in accelerometer counts between minutes for each subject.
Data were removed when the difference did not change for 60 minutes and the step count
equaled zero. Next, only recorded minutes that were deemed reliable by the NHANES team
under the NHANES identifier PAXSTAT and when the physical activity monitor (PAM) was
calibrated (PAXCAL) were retained [
]. We also removed (i.e., censored) steps associated
with intensity levels less than 500 activity counts/min as per recommended convention to
make the data more scalable to pedometer-based output [
]. Without removing the steps
associated with intensity levels less than 500 activity counts/min, the total steps/day are high
and implausible as indicated in [
]. Finally, any data over 180 steps/min were also removed.
These data were removed after comparison of the PAM?s output of intensity with the output
of cadence (S1 Fig?a scatter plot comparing steps/min with the PAM?s activity-count based
intensity output). Of note is that, when cadence exceeds 180 steps/min, the step count
increases as the activity-count based intensity decreases, which is implausible. This observation
was also supported by Rowlands et al. [
], who demonstrated that the ActiGraph GT1M
underestimated the step count at higher speeds.
Cardiometabolic risk factor thresholds. We applied published thresholds to classify risk
cut-offs for BMI, waist circumference, high-density lipoprotein (HDL) cholesterol,
triglycerides, systolic blood pressure (SBP), diastolic blood pressure (DBP), high blood pressure, and
fasting glucose (Table 1). In some cases (e.g., waist circumference or blood pressure) there are
low-risk and high-risk thresholds. For these variables, the low-risk and high-risk thresholds
were independently analyzed, thereby generating a low-risk and a high-risk thresholds for
peak 30-min cadence and steps/day. Additionally, each participant was assessed for metabolic
syndrome, defined as a diagnosis of exceeding at least three of the five metabolic risk
] (Table 1).
Subject characteristics after data processing appear in Table 2.
3 / 14
Design. Three main questions drove this study analysis.
1. What are the minimum steps/day and peak 30-min cadence (steps/min) that one should
achieve to avoid exhibition of each specific cardiometabolic risk factor? These thresholds
theoretically represent a minimum target value below which an individual should attempt
to not fall.
2. How does age or gender affect the thresholds in Question 1? Specifically, do the thresholds
remain uniform or are they age or gender dependent?
3. At what cadence and how much time spent at this cadence can individuals be classified as
healthy, that is, enjoying an absence of all cardiometabolic risk factors tested. These
thresholds theoretically represent an upper target value for more optimal health.
To address these three questions, we used the publicly available NHANES 2005?2006
database that contains minute-by-minute step accumulation data along with measured waist
circumference, high-density lipoprotein (HDL) cholesterol, triglycerides, systolic blood pressure
(SBP), diastolic blood pressure (DBP), high blood pressure, and fasting glucose.
In reference to Question 1, we employed a receiver operating characteristic (ROC) analysis
to determine the capacity of peak 30-min cadence and steps/day to classify individuals in the
NHANES dataset for the different cardiometabolic risk factors.
The second question was evaluated by grouping participants into 10-year age ranges and
performing the ROC analysis to determine age-specific classification thresholds. This analysis
was performed to facilitate clinical interpretation and application of the thresholds. We also
developed logistic regression models with steps/day and peak 30-min cadence, both adjusted
by age and smoking. Each logistic regression model used the sample weights, PSUs, and
stratification variables provided by the NHANES survey team [
]. While the quality of the model
can be assessed, the thresholds are not clinically interpretable and so we report instead the
odds ratio (OR).
Finally, Question 3 was evaluated using a decision tree analysis that classified participants
into healthy versus at risk groups based on time spent in cadence bands [
]. This analysis
provides information regarding the required time spent in derived cadence bands associated with
the absence of risk.
ApproachReceiver operating characteristic curve analysis. A ROC curve analysis was
conducted using the statistical program R (R Core Team (2013)). The R package ?dplyr? was
used to group and filter the NHANES data by age and to assign the binary outputs of 0 if the
designated metabolic health risk threshold (Table 1) was not met and 1 if the standard was
met. The R package, pROC was then used to classify true positives, false positives, true
negatives, and false negatives for each disease state described in Table 1. The pROC package
outputs the threshold, which simultaneously maximizes true positives and minimizes false
negatives and the resulting area under the curve (AUC) of the ROC curve. Ninety-five percent
confidence intervals for the AUC and thresholds were also provided by the pROC package.
Logistic regression model
The R package ?survey? was used to perform logistic regression. The survey package takes into
account the sample weights, PSUs, and stratification variables provided by the NHANES
]. We tested the necessary assumption that the logit was linear in each continuous
covariate (peak 30-min cadence and age).
Since peak 30-min cadence is a continuous covariate, reporting the Odds Ratio (OR) for
the increase of 1 step/min in a peak 30-min cadence would be of little interest. Instead, each
OR was calculated using an increase of 20 steps/min to match the 20 steps/min discretization
of cadence bands, OR ? e20bb, subsequently the 95% confidence intervals are reported using
?e20bb 1:96 20 ScE?bb??. The choice of 20 steps/min was selected because this is equivalent to
moving the peak 30-min cadence up one cadence band [
]. Accordingly, ORs for peak 30-min
cadence can be interpreted as ?for every increase in 20 steps/min, an individual is [OR] times
less likely to be at risk for the respective negative metabolic health outcome.? Similar to the
ORs for peak 30-min cadence, steps/day ORs were calculated using a scalar of 1000 steps/day.
Thresholds by age ranges. The data were grouped by age ranges [
], [30, 39], [40,49],
[50,59], [60,69] and [70,85]. Sex-specific ROC curve analyses were performed using the data in
each age strata. The resulting AUC and l thresholds were calculated for each sex and age strata.
Decision tree analysis of cadence bands. A decision tree analysis was performed using
the Classification and Regression Tree (CART) algorithm through the R package, ?rpart? [
5 / 14
The CART algorithm uses input variables to classify an outcome; in our case a binary outcome
of either having or not having the cardiometabolic risk factor. The algorithm is sometimes
referred to as recursive partitioning because the goal of the process is to partition the data into
groups iteratively until no predictive improvement is achieved when further partitioning the
group. More specifically, the original dataset is first split into two groups using the optimal
variable, i.e., the variable that decreases the risk the most. One group is generated with the lowest
risk and the other group represents the remainder of the dataset. The same process is then
applied again separately to these two new groups, breaking them into further subgroups until
either the algorithm has reached a minimum group size or if no more improvements can be
made in predicting whether the group can be further delineated into at-risk versus risk-free
The NHANES step data were first separated into cadence bands representing the total
number of minutes spent at: 0 steps/min (zero cadence; non-movement during wear time), 1?19
steps/min (incidental movement), 20?39 steps/min (sporadic movement), 40?59 steps/min
(purposeful stepping), 60?79 steps/min (slow walking), 80?99 (medium walking), 100?119
(brisk walking) and 120 steps/min (all faster ambulation) [
]. The amount of time spent in
each cadence band was averaged over valid days of wear. The time spent in cadence bands
were then used as model inputs to classify individuals as above or below the low risk metabolic
syndrome threshold. The decision tree classification was repeated 1000 times by assigning 80%
of the data as training dataset and 20% of the data reserved to evaluate the cross-validated
classification accuracy. The decision tree analysis was performed in the statistical package R (R
Core Team (2013)) using the ?rpart? (Recursive Partitioning and Regression Trees) package.
The entire dataset prior to data cleaning consisted of 10348 individuals, of whom 5080 were
men and 5268 were women. From the full dataset, 3377 participants had valid accelerometer
data. To identify the existence of metabolic syndrome concomitant metabolic measurements
of waist circumference, triglycerides, HDL cholesterol, systolic and diastolic blood pressure
and fasting glucose are required. Of the 3377 participants, there were a total of 1065
participants who had all five complete metabolic measurements available to determine whether or
not they had metabolic syndrome. Out of the participants with valid steps/day and peak
30-min cadences, 178 were classified with ?high-risk? for metabolic syndrome, 1742 were
classified without ?high-risk? metabolic syndrome, 1143 were classified with ?low-risk? metabolic
syndrome, and 706 were classified without ?low-risk? metabolic syndrome. The remainder of
the participants could not be classified as ?high-risk? (1457 participants) or ?low-risk? (1528
participants) because of missing measurements. Average BMI in the final analytical sample
was 28.47 ? 6.60 kg/m2 with mean age 48.15 ? 19.51 years.
Thresholds for peak 30-min cadence and steps/day. All AUC values for both peak
30-min cadence and steps/day were greater than or equal to 0.50 (Tables 3 and 4). The highest
AUC values were associated with high-risk fasting glucose (AUC = 0.701 for steps/day) and
high-risk metabolic syndrome (AUC = 0.68 for peak 30-min cadence). The lowest AUCs were
observed for high-risk HDL cholesterol (AUC = 0.50 for steps/day) and low-risk HDL
cholesterol (AUC = 0.51 for peak 30-min cadence) with the lower bound of the confidence interval
below 0.50. Peak 30-min cadence ranged from 66?72 steps/min across cardiometabolic risk
factors. Thresholds for steps/day ranged from 4325?6192 steps/day.
Logistic regression models. Four different models were created to assess the impact of
peak 30-min cadence and steps/day to estimate thresholds for waist circumference, blood
6 / 14
pressure, and metabolic syndrome. One model included peak 30-min cadence or steps/day
as univariate predictors, and the remaining models were adjusted for age and smoking. Each
model?s Akaike information criterion (AIC) was compared to select the best model (see S5
Table for AIC?s for each model). In all cases, the model that adjusted for age and smoking had
the lowest or nearly lowest AIC. As such, these are the reported OR?s.
Table 5 reports OR values for all cases where the results were statistically significant. OR
ranged from 1.18 to 1.60 with the highest OR occurring in peak 30-min cadence adjusted for
age and smoking classifying cardiometabolic risk. The 95% CI for the OR in adjusted peak
30-min cadence classifying high-risk metabolic syndrome was (1.13, 2.04).
Thresholds by age ranges. Fig 1 provides peak 30-min cadence and steps/day thresholds
for the sex-specific and combined full datasets, and steps/day thresholds for males and females
(Panel A and B respectively). Comprehensive statistical results (e.g., each age-respective AUC
and confidence intervals) are provided in S1, S2, S3 and S4 Tables. Higher thresholds indicate
that higher intensity (steps/min) and/or more steps/day are required to reach an absence of
metabolic health disease. Higher thresholds were observed in men compared to women. In
men, higher steps/day thresholds were observed in age ranges of 30?39, while in women,
higher thresholds were observed in the age-range 50?59 years. Peak 30-min cadence
7 / 14
thresholds were highest for men in the 18?39 year age range, while thresholds for women were
highest in the 50?59 year age range. Thresholds decline in older age bins, however, the sample
sizes also decline as age increases.
Decision tree analysis. Fig 2 represents the decision tree result for low-risk metabolic
syndrome. In Fig 2, spending virtually any time in the seventh cadence band ( 120 steps/min),
4.2 minutes in the fifth cadence band (80?99 steps/min), and 9.8 minutes in the first cadence
band (1?19 steps/min) would result in a person being classified as not at low-risk metabolic
syndrome. Additionally, a person who does not spend any time in the seventh cadence band
( 120 steps/min) would be classified as at-risk. When the observations were split into 1000
different training sets with 80% of the data reserved for training and 20% for testing, the tree
correctly predicted an average of 66.3% of the locations in the test data sets.
Waist Circumference (cm)
Blood Pressure (mm Hg)
8 / 14
Fig 1. Sex-specific and total population thresholds for metabolic syndrome determined for each age decade. Higher thresholds
indicate higher steps/min or more steps/day are required to achieve positive health outcomes. A. Thresholds for peak 30-min
cadence B. Thresholds for steps/day.
The tree had one leaf with this initial data split that categorized the initial split variable into
a high probability of absence of a low metabolic syndrome risk factor. The other branches of
the decision tree suggest a pathway to absence of risk by spending longer time in lower cadence
bands. For example, in Fig 2, spending virtually no time in cadence band 7 (120 steps/min)
Fig 2. Decision tree classifying low-risk metabolic syndrome by thresholds of time spent in minutes in each
cadence band. The value ?No? represents absence of risk and ?Yes? represents presence of risk. In each box, the values
on the left are the number of participants in the box that did not have the risk factor and the value on the right are the
number of participants who did have the risk factor. The percentage represents what percent of the total population
were contained in the box. Spending virtually no time in cadence band 7 (120 steps/min) would result in a person
being classified as at-risk. The only pathway that exists to be classified as not being at-risk is to spend time in cadence
band 7 (120 steps/min), 4.2 minutes in cadence band 5 (80?99 steps/min), and 9.8 minutes in cadence band 1 (1?
9 / 14
would result in a person being classified as at-risk. The only risk-free pathway is by spending
any time in cadence band 7 (120 steps/min), over 4.2 minutes in cadence band 5 (80?99 steps/
min), and over 9.8 minutes in cadence band 1 (1?19 steps/min).
Previous work has been published on associations between cardiometabolic risk factors, steps/
day and peak 30-min cadence [
]. In the present analysis of the 2005?2006 NHANES
accelerometer data, we extend this foundational work by deriving minimum thresholds for
peak 30-min cadence and steps/day that classify cardiometabolic risk through a ROC analysis.
These thresholds are readily interpretable; steps/day or peak 30-min cadence values below
each respective threshold are associated with the presence of cardiometabolic risk. We also
derived thresholds for daily time accumulated in defined cadence bands associated with the
absence of cardiometabolic risk factors using a decision tree analysis. These thresholds can be
interpreted as goals associated with the absence of cardiometabolic risk factors.
Peak 30-min cadence and steps/day thresholds for cardiometabolic risk
Peak 30-min cadence thresholds across all cardiometabolic risk factors were between 66?72
steps/min. Average national peak 30-min cadence calculated from these same data is 71 steps/
min , which is within this range. Although these data are cross-sectional and therefore
prevent clear conclusions about causality, the results suggest that public health efforts made to
elevate peak 30-min cadence beyond these minimal thresholds may be a potent strategy to
minimize cardiometabolic risk. Though there are many ways to achieve a peak 30-min cadence
above 72 steps/min, the most straightforward method to achieve this target would be to
ambulate at a cadence 72 steps/min for 30 minutes each day.
Volume thresholds ranged between 4325?6192 steps/day. These thresholds appear lower
than the 7,100 to 11,000 steps/day range that has been mapped to assembled studies of
objectively-determined 30 minutes of daily moderate-to-vigorous intensity physical activity, which
represents the current physical activity standards [
]. There are two reasons for this
discrepancy. The first is that our thresholds represent a minimum bar. The threshold is the lowest
value that one can achieve before the presence of cardio-metabolic risk. Thus, the threshold
should not be considered as a physical activity goal. Recent research that identifies thresholds
associated with mortality found similar lower bounds [
]. Second, the benchmark
accumulation of 150 minutes per week of moderate-to-vigorous intensity physical activity itself is a
product of years of research primarily based on self-reported behavior [
] and a literal
translation to objectively monitored time has been questioned [
]. In contrast, the thresholds
generated herein are not shaped by any form of preconceived notion related to duration or
intensity. Instead they represent minimum objectively-monitored bounds emerging directly
from the data set that are associated with a variety of accepted cardiometabolic risk factors.
An AUC over 0.50 indicates the classifier model performs better than a random classifier.
The closer the AUC is to 1, the better the classification. While all of the AUCs for both peak
30-min cadence and steps/day were over the value of 0.50, the signals derived from each were
not overly strong. The highest AUC was 0.69. The weak signals may be due to several reasons.
First, energy expenditure and physical activity are both known to exhibit biological variability
and are dependent on many factors (e.g., sex, age, mass, etc.), some of which may not be fully
understood (for example, genetic variability) . Second, the step data derived from
accelerometer-based devices itself is known to exhibit measurement error when compared to actual
steps taken [
]. For example, Toth et al., demonstrated that device accuracy relative to the
criterion measure of stepping (direct observation; hand count) varied between ~5 and 120%
10 / 14
mean absolute percent error (MAPE) under free-living conditions, depending on the device,
wear location, and the step detection algorithm/data processing techniques employed [
Thus, studies employing different wearable technologies including different brands of
accelerometers and pedometers to determine the relationships between steps/day and
cardiometabolic risk may produce varying results. Even with these considerations, the results here
demonstrate that while peak 30-min cadence and steps/day only serve as estimated proxies
for quality and quantity of physical activity, they are still associated with components of
metabolic syndrome. There are also other proxies for habitual physical activity intensity, for
example steps accumulated over 10-minute bouts. However, when we plot the highest number of
steps over 10-minute bouts against peak 30-min cadence (S2 Fig), there is a strong correlation
(r = 0.90). Finally, examining time spent in cadence bands, as we have performed in the
decision tree analysis, provides analysis without collapsing the data first.
In addition to our initial evaluations using step-based metrics, future work could derive
thresholds for other markers physical activity (e.g., sedentary time, and light, moderate and
vigorous PA) and compare them to our findings with peak 30-min cadence[
Age dependent thresholds for cardiometabolic risk
We demonstrate here through a logistic regression model that the estimation of
cardiometabolic risk from peak 30-min cadence is age dependent. These findings were further confirmed
through the ROC analysis performed in 10-year age ranges. Interestingly, the peak 30-min
cadence and steps/day thresholds were highest for women between 40?59 years of age. The
interpretation of this finding is that the quality and quantity of daily activity needs to be higher
in this age range in order to avoid the presence of cardiometabolic risk. This surprising finding
is supported by a study that performed energy balance measurements in perimenopausal
women who were within the age range identified in our study [
]. The investigators of that
study reported decreased energy expenditure and fat oxidation in women at the onset of
menopause, concluding that during this life transition women need to increase their physical
activity and/or decrease their energy intake to maintain body weight.
The increased steps/day thresholds in males between the ages of 30?39 were surprising.
To our knowledge, there have been no studies examining changes in energy expenditure or
energy balance at key time points in males like the ones performed in females [
findings suggest more research on longitudinal energy expenditure/balance changes in males is
Decision tree classification
The decision tree analysis performed herein demonstrated that time accumulated at higher
cadence bands resulted in a lower probability of cardiometabolic risk factors. Steps taken in
the higher cadence bands are thought to represent increasingly more purposeful movement
]. Despite strong evidence for the cardiometabolic benefits associated with light
intensity physical activity [
], this finding reinforces the additional importance of time
spent accumulating these more purposefully higher intensity cadences. On the other hand,
there was also a leaf on the decision tree associated with sedentary behavior that led to paths
with high-risk, accentuating the relationship between less movement and high
cardiometabolic risk. Regardless, the decision tree analysis presented here has several limitations. The
first is that the variable (i.e., time spent in cadence band 7 at 120 steps/min) determined
by the algorithm in the R program for the initial split cannot directly be considered the most
important factor in determining risk. A deeper analysis that forces splits may be required
because the data herein may not allow for sufficient splits at the lower cadences. Despite
11 / 14
these concerns, this initial decision tree analysis does provide more rigorous and quantitative
step-based goals that effectively delineate healthy populations from those at risk in terms of
cardio metabolic health.
This study is the first to calculate volume and effort-specific step-based thresholds for
cardiometabolic health-risk stratification. Though there is ample evidence showing a clear
doseresponse relationship between walking behavior and health [
], no such reports have been
presented using step-based physical activity intensity or effort-related metrics. In light of the
widely-recognized ease of interpreting physical activity recommendations based upon steps/
day and the need for step-based intensity guidelines [
], this novel analysis offers a utilitarian
platform from which we may continue to build and advance empirical support for step-based
public health recommendations.
S1 Table. Peak 30-min cadence, AUC and thresholds to classify each of the known
highrisk metabolic syndrome. Peak 30-min cadence above the threshold classifies positive health
S2 Table. Peak 30-min cadence, AUC and thresholds to classify each of the known low-risk
metabolic syndrome. Peak 30-min cadence above the threshold classifies positive health
S3 Table. Total steps/day, AUC and thresholds to classify each of the known high-risk
metabolic syndrome. Total steps/day above the threshold classifies positive health outcomes.
S4 Table. Total steps/day, AUC and thresholds to classify each of the known low-risk
metabolic syndrome. Total steps/day above the threshold classifies positive health outcomes.
S5 Table. AIC for logistic regression models.
S1 Fig. Plot of steps/min with the PAM?s activity-count based intensity output. When
cadence exceeds 180 steps/min, the step count increases as the activity-count based intensity
decreases, which is implausible.
S2 Fig. Steps accumulated in 10-minute bouts versus peak 30-min cadence. To calculate
bouts, each participant?s 10-minute bout total steps were calculated by summing the top 3
non-overlapping total steps taken in a 10-minute consecutive time interval. Each valid wear
day was used in calculating the mean total steps taken in the top 3 x 10-minute bouts.
Conceptualization: Catrine Tudor-Locke, Diana Thomas.
Formal analysis: Bryan Adams, Katie Fidler, Noah Demoes, Diana Thomas.
Methodology: Bryan Adams, Katie Fidler, Noah Demoes, Elroy J. Aguiar, Scott W. Ducharme,
Aston K. McCullough, Christopher C. Moore, Diana Thomas.
12 / 14
Project administration: Bryan Adams, Diana Thomas.
Supervision: Catrine Tudor-Locke.
Writing ? original draft: Bryan Adams, Katie Fidler, Elroy J. Aguiar, Scott W. Ducharme,
Catrine Tudor-Locke, Diana Thomas.
Writing ? review & editing: Elroy J. Aguiar, Scott W. Ducharme, Aston K. McCullough,
Christopher C. Moore, Catrine Tudor-Locke, Diana Thomas.
13 / 14
Obesity. Circulation. 2009; 120(16):1640?5. Epub 2009/10/07. https://doi.org/10.1161/
CIRCULATIONAHA.109.192644 PMID: 19805654.
1. Moore JX , Chaudhary N , Akinyemiju T. Metabolic Syndrome Prevalence by Race/Ethnicity and Sex in the United States , National Health and Nutrition Examination Survey , 1988 - 2012 . Prev Chronic Dis. 2017 ; 14 : E24 . Epub 2017/03/17. https://doi.org/10.5888/pcd14.160287 PMID: 28301314 .
2. Gami AS , Witt BJ , Howard DE , Erwin PJ , Gami LA , Somers VK , et al. Metabolic syndrome and risk of incident cardiovascular events and death: a systematic review and meta-analysis of longitudinal studies . J Am Coll Cardiol . 2007 ; 49 ( 4 ): 403 - 14 . Epub 2007/01/30. https://doi.org/10.1016/j.jacc. 2006 . 09 . 032 PMID: 17258085 .
3. Mottillo S , Filion KB , Genest J , Joseph L , Pilote L , Poirier P , et al. The metabolic syndrome and cardiovascular risk a systematic review and meta-analysis . J Am Coll Cardiol . 2010 ; 56 ( 14 ): 1113 - 32 . Epub 2010/09/25. https://doi.org/10.1016/j.jacc. 2010 . 05 .034 PMID: 20863953 .
4. Gennuso KP , Gangnon RE , Thraen-Borowski KM , Colbert LH . Dose-response relationships between sedentary behaviour and the metabolic syndrome and its components . Diabetologia . 2015 ; 58 ( 3 ): 485 - 92 . Epub 2014/12/06. https://doi.org/10.1007/s00125-014-3453-z PMID: 25476524 .
5. Physical Activity Guidelines Advisory Committee report , 2008 . To the Secretary of Health and Human Services. Part A: executive summary . Nutr Rev . 2009 ; 67 ( 2 ): 114 - 20 . Epub 2009/01/31. PMID: 19178654 .
6. 2018 Physical Activity Guidelines Advisory Committee Scientific Report. Department of Health and Human Services , 2018 .
7. Tudor-Locke C , Rowe DA . Using cadence to study free-living ambulatory behaviour . Sports Med . 2012 ; 42 ( 5 ): 381 - 98 . Epub 2012/04/03. https://doi.org/10.2165/ 11599170 -000000000-00000 PMID: 22462794 .
8. Tudor-Locke C , Camhi SM , Leonardi C , Johnson WD , Katzmarzyk PT , Earnest CP , et al. Patterns of adult stepping cadence in the 2005- 2006 NHANES. Prev Med . 2011 ; 53 ( 3 ): 178 - 81 . Epub 2011/06/29. https://doi.org/10.1016/j.ypmed. 2011 . 06 .004 PMID: 21708187 .
9. Tudor-Locke C , Brashear MM , Katzmarzyk PT , Johnson WD . Peak stepping cadence in free-living adults: 2005-2006 NHANES . J Phys Act Health . 2012 ; 9 ( 8 ): 1125 - 9 . Epub 2011/12/31. PMID: 22207030 .
10. Dall PM , McCrorie PR , Granat MH , Stansfield BW . Step accumulation per minute epoch is not the same as cadence for free-living adults . Med Sci Sports Exerc . 2013 ; 45 ( 10 ): 1995 - 2001 . Epub 2013/04/ 10. https://doi.org/10.1249/MSS.0b013e3182955780 PMID: 23568091 .
11. Sisson SB , Camhi SM , Church TS , Tudor-Locke C , Johnson WD , Katzmarzyk PT . Accelerometerdetermined steps/day and metabolic syndrome . Am J Prev Med . 2010 ; 38 ( 6 ): 575 - 82 . Epub 2010/05/ 25. https://doi.org/10.1016/j.amepre. 2010 . 02 .015 PMID: 20494233 .
12. Tudor-Locke C , Camhi SM , Troiano RP . A catalog of rules, variables, and definitions applied to accelerometer data in the National Health and Nutrition Examination Survey, 2003 - 2006 . Prev Chronic Dis. 2012 ; 9:E113 . Epub 2012/06/16. https://doi.org/10.5888/pcd9.110332 PMID: 22698174 .
13. National Health and Nutrition Examination Survey: 2005 -2006 Data Documentation , Codebook, and Frequencies . Centers for Disease Control , 2008 June 2008. Report No.
14. Tudor-Locke C , Johnson WD , Katzmarzyk PT . Accelerometer-determined steps per day in US adults . Med Sci Sports Exerc . 2009 ; 41 ( 7 ): 1384 - 91 . Epub 2009/06/12. https://doi.org/10.1249/MSS. 0b013e318199885c PMID: 19516163 .
15. Rowlands AV , Stone MR , Eston RG . Influence of speed and step frequency during walking and running on motion sensor output . Med Sci Sports Exerc . 2007 ; 39 ( 4 ): 716 - 27 . Epub 2007/04/07. https://doi.org/ 10.1249/mss.0b013e318031126c PMID: 17414811 .
16. Alberti KG , Eckel RH , Grundy SM , Zimmet PZ , Cleeman JI , Donato KA , et al. Harmonizing the metabolic syndrome: a joint interim statement of the International Diabetes Federation Task Force on Epidemiology and Prevention; National Heart, Lung, and Blood Institute; American Heart Association; World Heart Federation; International Atherosclerosis Society; and International Association for the Study of
17. National Health and Nutrition Examination Survey: 2005 -2006 Data Documentation , Codebook, and Frequencies , Demographic Variables & Sample Weights (DEMO_D) Centers for Disease Control , 2007 .
18. Barreira TV , Harrington DM , Schuna JM Jr., Tudor-Locke C , Katzmarzyk PT . Pattern changes in step count accumulation and peak cadence due to a physical activity intervention . J Sci Med Sport . 2016 ; 19 ( 3 ): 227 - 31 . Epub 2015/02/18. https://doi.org/10.1016/j.jsams. 2015 . 01 .008 PMID: 25687483 .
19. Breiman L. Classification and regression trees . New York, N.Y.: Chapman & Hall; 1993 . x, 358 p. p.
20. Therneau TM , Atkinson EJ . An Introduction to Recursive Partitioning Using the RPART Routines . CRAN (The Comprehensive R Archive Network ). 2018 .
21. Schuna JM Jr., Brouillette RM , Foil HC , Fontenot SL , Keller JN , Tudor-Locke C . Steps per day, peak cadence, body mass index, and age in community-dwelling older adults . Med Sci Sports Exerc . 2013 ; 45 ( 5 ): 914 - 9 . Epub 2012/12/19. https://doi.org/10.1249/MSS.0b013e31827e47ac PMID: 23247705 .
22. Tudor-Locke C , Bassett DR , Rutherford WJ , Ainsworth BE , Chan CB , Croteau K , et al. BMI-referenced cut points for pedometer-determined steps per day in adults . J Phys Act Health . 2008 ; 5 Suppl 1 : S126 - 39 . Epub 2008/04/12. PMID: 18364517 .
23. Tudor-Locke C , Schuna JM Jr., Han HO , Aguiar EJ , Green MA , Busa MA , et al. Step-Based Physical Activity Metrics and Cardiometabolic Risk: NHANES 2005-2006. Med Sci Sports Exerc . 2017 ; 49 ( 2 ): 283 - 91 . Epub 2016/09/27. https://doi.org/10.1249/MSS.0000000000001100 PMID: 27669450 .
24. Tudor-Locke C , Craig CL , Brown WJ , Clemes SA , De Cocker K , Giles-Corti B , et al. How many steps/ day are enough? For adults . Int J Behav Nutr Phys Act . 2011 ; 8 : 79 . Epub 2011/07/30. https://doi.org/ 10.1186/ 1479 -5868-8-79 PMID: 21798015 .
25. Lee IM , Shiroma EJ , Kamada M , Bassett DR , Matthews CE , Buring JE . Association of Step Volume and Intensity With All-Cause Mortality in Older Women . JAMA Intern Med . 2019 . Epub 2019/05/30. https://doi.org/10.1001/jamainternmed. 2019 .0899 PMID: 31141585 .
26. Haskell WL , Lee IM , Pate RR , Powell KE , Blair SN , Franklin BA , et al. Physical activity and public health: updated recommendation for adults from the American College of Sports Medicine and the American Heart Association . Med Sci Sports Exerc . 2007 ; 39 ( 8 ): 1423 - 34 . Epub 2007/09/01. https:// doi.org/10.1249/mss.0b013e3180616b27 PMID: 17762377 .
27. Troiano RP , Berrigan D , Dodd KW , Masse LC , Tilert T , McDowell M . Physical activity in the United States measured by accelerometer . Med Sci Sports Exerc . 2008 ; 40 ( 1 ): 181 - 8 . Epub 2007/12/20. https://doi.org/10.1249/mss.0b013e31815a51b3 PMID: 18091006 .
Westerterp KR . Physical activity and physical activity induced energy expenditure in humans: measurement, determinants, and effects . Front Physiol . 2013 ; 4 : 90 . Epub 2013/05/03. https://doi.org/10.3389/ fphys. 2013 .00090 PMID: 23637685 .
29. Toth LP , Park S , Springer CM , Feyerabend MD , Steeves JA , Bassett DR . Video-Recorded Validation of Wearable Step Counters under Free-living Conditions . Med Sci Sports Exerc . 2018 . Epub 1. 19 . 2018 . https://doi.org/10.1249/MSS.0000000000001569 PMID: 29381649
30. Lovejoy JC , Champagne CM , de Jonge L , Xie H , Smith SR . Increased visceral fat and decreased energy expenditure during the menopausal transition . Int J Obes (Lond) . 2008 ; 32 ( 6 ): 949 - 58 . Epub 2008/03/12. https://doi.org/10.1038/ijo. 2008 .25 PMID: 18332882 .
31. Fuzeki E , Engeroff T , Banzer W. Health Benefits of Light-Intensity Physical Activity: A Systematic Review of Accelerometer Data of the National Health and Nutrition Examination Survey (NHANES) . Sports Med . 2017 ; 47 ( 9 ): 1769 - 93 . Epub 2017/04/11. https://doi.org/10.1007/s40279-017-0724-0 PMID: 28393328 .
Williams PT . Dose-response relationship of physical activity to premature and total all-cause and cardiovascular disease mortality in walkers . PLoS One . 2013 ; 8 ( 11 ): e78777 . Epub 2013 /12/07. https://doi.
org/10 .1371/journal.pone.0078777 PMID: 24312170 .
33. Obesity : preventing and managing the global epidemic . 2000 Contract No.: 895.
34. Understanding Blood Pressure Readings [updated February 18 , 2018 ; cited 2018 April 14 , 2018 ]. http:// www.heart.org/HEARTORG/Conditions/HighBloodPressure/KnowYourNumbers/UnderstandingBlood-Pressure-Readings_UCM_301764_Article.jsp#. Ws_DhYdaFjo.
35. American Diabetes A. 2. Classification and Diagnosis of Diabetes: Standards of Medical Care in Diabetes-2018 . Diabetes Care . 2018 ; 41 ( Suppl 1 ): S13 - S27 . Epub 2017/12/10. https://doi.org/10.2337/dc18- S002 PMID: 29222373 .
36. Cholesterol test 1998-2018 [cited 2018 April 14 , 2018 ]. https://www.mayoclinic.org/tests-procedures/ cholesterol-test/about/pac-20384601.