Skip to main content

Table 4 Characteristics and results of studies that examined validity and accuracy of ActiGraphs for measuring sedentary behaviour in older community-dwelling, healthy adults (mean age ≥ 60 years), ordered from largest to smallest sample size

From: Validity of objective methods for measuring sedentary behaviour in older adults: a systematic review

Study Participants Monitor and epochs analysed Methods Results for sedentary behaviour
Keadle et al. (2014) [28] n: 7650
Only females
Mean age: 71.4 ± 5.8 y
USA: Data collected for an ancillary study of participants from the Women’s Health Study, a randomized trial of aspirin and vitamin E to reduce risk of cardiovascular disease and cancer. Data collected after completion of the trial.
ActiGraph GT3X+
Worn on hip
60-s epochs
Free-living
Activities: activities with VA < 100 cpm and VM < 200 cpm
Criterion: paper logs
Observation period: 7 consecutive days during waking hours
Valid hours and days: ≥10 h, ≥1 day and ≥ 4 days
Non-wear algorithm: Troiano et al. [62] and Choi et al. [61]
Analysis: computed min/day of sedentary activities and used Wilcoxon signed rank sum test to compare output between VA and VM
VA: min/day (95% CI)
Log+Troiano et al. algorithm: 530.1 (480.1, 578.6)
Log+Choi et al. algorithm: 581.6 (521.1, 639.8)
VM: min/day (95% CI)
Log+Troiano et al. algorithm: 474.6 (417.0, 529.6)
Log+Choi et al. algorithm: 506.0 (439.2, 570.9)
Differences between VA and VM were significant (p < 0.001)
Using dates from logs combined with Choi algorithm minimalised missing data and researcher burden. Using algorithm only resulted in misclassification of days when accelerometers were being posted to participants as accelerometer wear days.
Evenson et al. (2015) [33] n: 200
Females only
Mean age = 75.5 ± 7.7 y
USA: Data collected for a calibration sub-study of participants from the Women’s Health Initiative Long Life Study. Data collected after completion of main study.
ActiGraph GT3X+
Worn on hip
15-s epochs
Laboratory-based
Activities: sitting, watching DVD; sitting assembling a puzzle
Criterion measure: portable calorimeter
Observation period: 7 min per activity
Analysis: computed Spearman correlation using normal and low frequency extension for ActiGraph; used ROC analysis to determine optimal cut-points; computed AUC, sensitivity and specificity for normal and low-frequency extension filter
MAXIMISING SUM OF SENSITIVITY+SPECIFICITY
VA: Normal filter
Optimal: 0 counts/15 s
AUC: 0.73; Sensitivity: 91%; Specificity: 62%
VA: Low-frequency extension filter
Optimal: 0 counts/15 s
AUC: 0.79; Sensitivity: 79%; Specificity: 81%
VM: Normal filter
Optimal: ≤42 counts/15 s
AUC: 0.88; Sensitivity: 87%; Specificity: 80%
VM: Low-frequency extension filter
Optimal: ≤65 counts/15 s
AUC: 0.90; Sensitivity: 87%; Specificity: 81%
BALANCING NUMBER OF FALSE POSITIVES AND FALSE NEGATIVES
VA: Normal filter
Optimal: 0 counts/15 s
AUC: 0.73; Sensitivity: 91%; Specificity: 62%
VA: Low-frequency extension filter
Optimal: 0 counts/15 s
AUC: 0.79; Sensitivity: 79%; Specificity: 81%
VM: Normal filter
Optimal: ≤12 counts/15 s
AUC: 0.88; Sensitivity: 76%; Specificity: 88%
VM: Low-frequency extension filter
Optimal: ≤31 counts/15 s
AUC: 0.90; Sensitivity: 71%; Specificity: 88%
Bai et al. (2016) [34] n: 194
Females only
Mean age = 75.4 ± 7.7 y
USA: Data collected for a calibration sub-study of participants from the Women’s Health Initiative Long Life Study. Data collected after completion of main study.
ActiGraph GT3X+
Worn on hip
1-s epochs
Laboratory-based
Activities: sitting, watching DVD; sitting assembling a puzzle
Criterion measure: portable calorimeter
Observation period: 7 min per activity
Analysis: used ROC analysis to compare an activity index created for this study with activity count using the normal filter and LFE, and another method of summarise raw data, the Euclidean Norm Minus One (ENMO); computed AUC also to compare these measures on predicting energy expenditure greater than SB
Compare watching DVD vs washing dishes or doing laundry, respectively
Activity index: AUC: 0.98, 0.98
Activity count (normal filter): AUC: 0.39, 0.74
Activity count (LTF): AUC: 0.27, 0.87
ENMO: AUC: 0.40, 0.69
Predicting whether MET is < or ≥ 1.5 MET
Activity index: AUC: 0.96
Activity count (normal filter): AUC: 0.86
Activity count (LFE): AUC: 0.91
ENMO: AUC: 0.85
Chudyk et al., 2017 [26] n: 106
39 men and 76 women
Mean age = 74.1 ± 6.4 y
Canada: Data collected for Walk the Talk: Transforming the Built Environment to Enhance Mobility in Seniors, a cross-sectional study of older adults living on low incomes. A random stratified design based on neighbourhood walkability was used to recruit older adults who received a provincial government rental subsidy.
ActiGraph GT3X+
Worn on right hip
60-s epochs
Free-living
Activities: VA < 100 cpm
Criterion: paper log
Observation period: 7 days during waking hours
Valid hours and days: ≥8 h; ≥4 days
Non-wear algorithms:
1. ≥60 min of continuous zeroes; allow for up to 2 min of counts ≤100 counts as non-wear time [62]
2. ≥90 min of consecutive zeroes; allow for up to 2 min of non-zero counts if the interruption was accompanied by 30 consecutive min of 0 counts upstream or downstream [61].
3. ≥90 min of continuous zeroes; no allow for interruptions, as non-wear time
4. ≥90 min of continuous zeroes; allow for up to 2 min of counts≤50 counts as non-wear time
5. ≥90 min of continuous zeroes, while allowing for up to 2 min of counts ≤100 counts as non-wear time
Analysis: used Bland Altman methods to compare logs to non-wear time algorithms
COMPARISON OF EACH ALGORITHM TO LOGS
Mean differences in SB min/day between log and ActiGraph (95% CI)
1. 37.5 (25.7, 49.3)
2. 5.8 (− 4.4, 16.0)
3. − 4.4 (− 14.6, 5.8)
4. 5.5 (− 4.9, 15.9)
5. 8.1 (− 2.3, 18.5)
95% limits of agreement between log and ActiGraph for wear-time
1. − 84.6, 159.6
2. − 100.2, 111.8
3. − 110.5, 101.8
4. − 103.0, 114.0
5. − 100.2, 116.4
Koster et al., 2016 [29] n: 62
26 males, 36 females
(58.1% females)
Mean age = 78.4 ± 5.7 y
USA: Data collected for a methodological sub-study of
the Developmental Epidemiologic
Cohort Study
ActiGraph GT3X+ Worn on hip, right wrist and left wrist concurrently 15-s and 60-s epochs Free-living
Activities: sitting, lying
Concurrent measure: ActivPAL worn on right thigh
Observation period: 7 full days, with permission given for removal at night
Valid hours and days: ≥10 h; ≥1 day
Non-wear algorithm: Choi et al. [61]
Analysis: for each monitor, used ROC to determine optimal cut-points; computed AUC, sensitivity, specificity, and kappa statistic
STANDARD CUT-OFF POINTS FOR 60-SEC EPOCHS
VA < 100 cpm:
Sensitivity: 94%; Specificity: 58%; kappa: 0.55
Mean difference: − 114.3 min/day (95%CI -140.5, − 88.1)
VM < 200 cpm:
Sensitivity: 88%; Specificity: 79%; kappa: 0.68
Mean difference: − 9.9 min/day (95%CI -32.8, 13.0)
OPTIMAL 60-SEC EPOCHS
Hip-worn ActiGraph
VA < 22 cpm:
AUC: 0.85; Sensitivity: 85%; Specificity: 74%; kappa: 0.59
Mean difference: − 5.0 min/day (95%CI -29.5, 19.5)
VM < 174 cpm:
AUC: 0.89; Sensitivity: 87%; Specificity: 80%; kappa: 0.67
Mean difference: 2.5 min/day (95%CI -20.4, 25.5)
Wrist-worn ActiGraph
VM < 2303 cpm (dominant wrist):
AUC: 0.86; Sensitivity: 81%; Specificity: 78%; kappa: 0.58
Mean difference: 30.2 min/day (95% CI 10.7, 49.6)
VM < 1853 cpm (non-dominant wrist):
AUC: 0.86; Sensitivity: 82%; Specificity: 77%; kappa: 0.57
Mean difference: 22.6 min/day (95%CI 0.5, 44.6)
OPTIMAL 15-S EPOCHS
Hip-worn ActiGraph
VA < 1 count/15 s:
AUC: 0.75; Sensitivity: 87%; Specificity: 61%; kappa: 0.50
Mean difference: − 53.4 min/day (95%CI -76.4, − 30.4)
VM < 20 counts/15 s:
AUC: 0.83; Sensitivity: 83%; Specificity: 73%; kappa: 0.56
Mean difference: 12.8 min/day (95%CI -8.8, 34.4)
Wrist-worn ActiGraph
VM < 517 counts/15 s (dominant wrist):
AUC: 0.81; Sensitivity: 75%; Specificity: 75%; kappa: 0.48
Mean difference: 64.7 min/day (95%CI 45.7, 83.7)
VM < 376 counts/15 s (non-dominant wrist):
AUC: 0.81; Sensitivity: 75%; Specificity: 74%; kappa: 0.47
Mean difference: 64.7 min/day (95%CI 44.3, 85.0)
Rosenberg et al. (2017) [30] n: 39
Only females
Mean age = 69.4 (range: 56–94)
USA: A convenience sample
ActiGraph GT3X+
Worn on right hip
60-s epochs
Free-living
Activities: sitting and riding in a vehicle
Criterion: DO (images produced via SenseCam camera worn on lanyard around neck)
Observation period: 7 days during waking hours
Valid hours and days: not reported
Non-wear algorithms: Choi et al. [61]
Analysis: used machine learning algorithm to classify activities (sitting and riding in vehicle analysed separately), computed sensitivity and specificity
Sitting
Sensitivity: 89%; Specificity: 91%
Median counts = 0 (IQR: 0, 17), indicating that sitting occurred at a lower intensity than would be detected by existing threshold of < 100 cpm
Riding in a vehicle
Sensitivity: 84%; Specificity: 99%
Median counts = 72 (IQR: 21, 177), indicating that riding occurred at a higher intensity than would be detected by threshold of < 100 cpm
Aguilar-Farias et al. (2014) [17] n: 37
13 males, 24 females (64.9% female)
Mean age: 73.5 ± 7.3 y
Australia: A convenience sample recruited mainly via flyers displayed at senior centres and exercise centres, and emails to university staff
ActiGraph GT3X+
Worn on right hip
1-s, 15-s, 60-s epochs
Free-living
Activities: sedentary activities defined as VM and VT counts being below cut-points set separately for 1-s, 15-s and 1-m epochs as follows: 1-s (< 1 to < 10 in increments of 1 counts/s), 15-s (< 1 to < 100 in increments of 5 counts/15 s) and 60-s epochs (< 1 to < 400 in increments of 25 cpm)
Concurrent measure: ActivPAL3™ worn on right thigh
Observation period: 7 consecutive days during waking hours
Valid hours and days: ≥10 h; ≥5 days
Non-wear algorithms: 90-min of consecutive zeros with no interruptions allowed
Analysis: used ROC to determine optimal cut-points, calculated AUC and computed sensitivity, specificity, percent correctly classified, mean bias (min/day)
OPTIMAL CUT-POINTS FOR VA
< 1 count/s
AUC:0.67; Sensitivity:92%; Specificity:43%
Correctly classified: 74%
Mean bias:156.61 (95%LoA: − 34.5, 347.7)
< 10 counts/15 s
AUC:0.70; Sensitivity:84%; Specificity:65%
Correctly classified: 79%
Mean bias: − 4.29 (95%LoA: − 141.3, 132.8)
< 25 cpm
AUC:0.79; Sensitivity:83%; Specificity:75%
Correctly classified: 80%
Mean bias:4.81 (95%LoA: − 157.2, 166.8)
OPTIMAL CUT-POINTS FOR VM
< 1 count/s
AUC:0.73; Sensitivity:85%; Specificity:62%
Correctly classified: 76%
Mean bias: 0.98 (95%LoA: − 113.4, 15.4)
< 70 counts/15 s
AUC:0.79; Sensitivity:87%; Specificity:70%
Correctly classified: 82%
Mean bias:0.80 (95%LoA: − 188.5, 120.1)
< 200 cpm
AUC:0.84; Sensitivity:89%; Specificity:79%
Correctly classified: 85%
Mean bias:18.05 (95%LoA: − 107.2, 143.3)
Sasaki, 2016
[31]
n = 35
14 males; 21 females
Mean age: 70.8 ± 4.9 y
USA: A convenience sample
ActiGraph GT3X+
Worn on dominant hip, wrist and ankle
20-s epochs
LABORATORY-BASED
Activities and observation period: performed sitting and lying down postures (30 s each); sat doing crossword puzzles or playing cards (5 min)
Criterion measure: DO
FREE-LIVING (N = 15)
Activity: SB
Criterion measure: DO (trained observers coded activities with continuous focal sampling software in a personal digital assistant)
Observation period: 2–3 h
Analysis: output used to train random forest (RF) and support vector machine (SVM) algorithms to classify activities; different algorithms developed for different body placement of monitor; computed percent correct classification
% CORRECT CLASSIFICATION AS SB
Lab-based algorithms applied to lab and free-living conditions: using 20-s epochs
SVM hip: Lab: 92%; Free: 68%
SVM wrist: Lab: 97%; Free: 73%
SVM ankle: Lab: 92%; Free: 79%
RF hip: Lab: 92%; Free: 62%
RF wrist: Lab: 93%; Free: 71%
RF ankle: Lab: 89%; Free: 76%
Free-living-based algorithms applied to free-living conditions: using 20-s epochs
SVM hip: Free: 82%
SVM wrist: Free: 75%
SVM ankle: Free: 87%
RF hip: Free: 81%
RF wrist: Free: 81%
RF ankle: Free: 84%
Free-living-based algorithms applied to free-living conditions: using other epochs
5-s epochs:
hip: 79%; wrist: 70%; ankle: 78%
10-s epochs:
hip: 82%; wrist: 74%; ankle: 82%
15-s epochs:
hip: 82%; wrist: 75%; ankle: 87%
30-s epochs:
hip: 82%; wrist: 78%; ankle: 87%
RF ALGORITHMS
5-s epochs:
hip: 72%; wrist:73%; ankle: 75%
10-s epochs:
hip: 77%; wrist: 77%; ankle: 81%
15-s epochs:
hip: 81%; wrist: 81%; ankle: 84%
30-s epochs:
hip: 83%; wrist: 84%; ankle: 86%
FREE-LIVING FOR HIGHEST OVERALL CLASSIFICATION RATES ACROSS ANKLE, HIP and WRIST ALGORITHMS, AT 30-S EPOCHS
Sensitivity and specificity:
Ankle (SVM): 82%; 94%
Hip (RF): 79%; 93%
Wrist (RF): 69%; 92%
Bourke et al., 2016 [25] n: 20% females not provided
Mean age = 76.4 ± 5.6 y
Norway: A convenience sample
ActiGraph GT3X+
Worn on right hip
5-s epochs
LABORATORY-BASED
Activities: semi-structured protocol that included sitting and lying
Criterion measure: DO (video camera)
Observation period: in one session
FREE-LIVING
Activities: sitting and lying as part of tasks requested by researchers + normal routine
Criterion measure: DO (camera on head)
Observation period: partial day
Analysis: used laboratory-based and free-living data together to assess % correctly classified using researcher-developed algorithm
Across both conditions % correct classification:
Sitting: 75%
Lying: 51%
  1. Abbreviations: AUC Area under the ROC curve that is used to evaluate classification accuracy, cpm Counts per minute, IQR Inter-quartile range, ICC Intraclass correlation coefficient, LoA Low-frequency extension filter, LTE Limit of Agreement, PPV Positive predictive value, ROC Receiver operator characteristic analysis; valid hours and days: for free-living studies lasting at least 7 days, number of hours per day and days during observation period that were required for data to be included in analysis;VA Vertical axis, VM Vector magnitude, m Minutes, s Seconds, h Hours, y Years