Skip to main content

Table 4 Convergent validity of accelerometer-based methods for toddlers, sorted by methodological study quality, quality of evidence (level of evidence), and study result rating

From: Systematic review of accelerometer-based methods for 24-h physical behavior assessment in young children (0–5 years old)

  Study Study populationa Outcome(s) & Setting Comparison measure(s) Method
ological study qualityb
Placement Device-based methodc Results Study result ratingd Quality of evidencee
(level of evidence)
Convergent validity Kwon et al. (2019) [49] n = 21
Age = 25 ± 2.5 (13 to 35) months
Sex = 50% girls
Activity type: running (forward from one place to another), walking (forward from one place to another), climbing up/down (the stairs/foam climber), crawling (moving forward on two hands and two knees), riding a ride on toy (sitting on toy, moving forward using two feet), standing (without lifting a foot), sitting (on the ground), stroller (sit on stroller/wagon pushed by adult), and being carried (by adult while adult is walking)
Free-living
(free play in commercial indoor playroom)
DVO using a developed coding scheme to score activity type by two independent raters VG Hip AG-RF-VA/HA/DA/VM-5 Best differentiating features: DA FFT SD (d' = 0.64), DA FFT max (d' = 0.61), HA FFT SD (d' = 0.47). Feature importance all 78 features < 0.1 (highest: SD of VM 0.039). Top 10 ranked features: basic quantiles (e.g., min, median, max) of single axis direct values and FFT values
RF: accuracy = 89%, precision = 88%, recall = 89%, F-score = .88
58% of carried labels were correctly (SB) classified, whereas 89% of ambulation was correctly classified
+  Low
(level 1)
Costa et al. (2014) [46] n = 20
Age = 2.99 ± 0.48 (2 to 3) years
Sex = 60% girls
Activity intensity: Total time in SB, LPA, and MVPA
Laboratory (semi-structured activity sessions)
DVO using CARS (modified) to score activity type and determine activity intensity (SB: stationary with no movement, and with movement of the limbs; LPA: slow/easy translocation; MVPA: translocation fast, and with moderate effort) by one rater (repeated after one month for random minute of each child) VG Right hip AG-C-VM-5 Total PA: CCC = .80, MD = 48.0 s*, LoA (-217.9 to 121.9)
SB: CCC = .74, MD = 48.0 s*, LoA (-121.9 to 217.9)
LPA: CCC = .34, MD = -163.5 s***, LoA (-343.0 to 16.0)
MVPA: CCC = .40, MD = 117.8 s***, LoA (-51.4 to 286.9)
± 
(SB + ; PA -)
Low
(level 1)
AG-C-VA-5 Total PA: CCC = .84, MD = 5.3 s, LoA (-171.1 to 181.6)
SB: CCC = .77, MD = -5.3 s, LoA (-181.6 to 171.1)
LPA: CCC = .45***, MD = -112.5 s, LoA (-320.5 to 95.5)
MVPA: CCC = .42, MD = 115.5 s***, LoA (-49.3 to 280.3)
± 
(SB + ; PA -)
AG-Pa-VA-15 Total PA: CCC = .71, MD = 50.5 s, LoA (-183.0 to 284.0)
SB: CCC = .75, MD = -22.0 s, LoA (-209.7 to 165.7)
LPA: CCC = .38, MD = -115.3 s**, LoA (-397.27 to 166.8)
MVPA: CCC = .30, MD = 165.8 s***, LoA (-71.2 to 402.73)
± 
(SB + ; PA -)
AG-T12-VA-15 Total PA: CCC = .85, MD = 15.3 s, LoA (-152.4 to 182.9)
SB: CCC = .79, MD = -15.3 s, LoA (-182.9 to 152.4)
LPA: CCC = .36, MD = -150.5 s***, LoA (-391.1 to 90.1)
MVPA: CCC = .30, MD = 165.8 s***, LoA (-71.2 to 402.7)
± 
(SB + ; PA -)
van Cauwenberghe et al. (2011) [45] n = 31
Age = 20 ± 4 (12 to 30) months
Sex = 45.2% girls
Activity intensity: SB, LPA and MVPA
Free-living
(out- & indoor free play at childcare)
DVO using OSRAC-P to score activity type and derive activity intensity (SB: stationary and motionless, and stationary with limb/trunk movement; LPA: slow, easy movement; MVPA: moderate, and fast movement) by two independent raters VG Right hip AG-cts-VA-15 mean rsp = .66***
epoch-by-epoch intensity level rsp = .52***
- Low
(level 1)
AG-Pa-VA-15 accuracy = 58.3%
SB: AUC-ROC = .71, Se = 67.0%, Sp = 75.4%
LPA: AUC-ROC = .62, Se = 60.0%, Sp = 63.2%
MVPA: AUC-ROC = .57, Se = 21.5%, Sp = 91.3%
-
(SB -; PA -)
AG-Si3-VA-15 accuracy = 52.7%
SB: AUC-ROC = .58, Se = 91.8%, Sp = 23.9%
LPA: AUC-ROC = .52, Se = 14.6%, Sp = 89.0%
MVPA: AUC-ROC = .53, Se = 8.9%, Sp = 97.1%
-
(SB -; PA -)
AG-vC-VA-15 accuracy = 52.2%
SB: AUC-ROC = .56, Se = 94.4%, Sp = 17.2%
LPA: AUC-ROC = .51, Se = 9.0%, Sp = 93.7%
MVPA: AUC-ROC = .53, Se = 10.0%, Sp = 96.9%
-
(SB -; PA -)
Trost et al. (2012) [44] n = 18
Age = 2.3 ± 0.4 years
Sex = 55.6% girls
Activity intensity: SB, LPA and MVPA
Free-living (regularly scheduled play)
DVO using CARS (modified) to score activity type and determine activity intensity (SB: lying down or sitting; LPA: standing; MVPA: walking and running) by two raters VG Right hip AG-T12-VA-15 SB: MD = -7.6***, LoA (-17.6to 2.3)
LPA: MD = 7.2***, LoA (-2.0 to 16.3)
MVPA: MD = 0.5, LoA (-2.6 to 3.5)
? Low
(level 1)
Pre-school cut-points: AG-Si/vC/Re/Pa/N-VA-15 VG Preschool and toddler cut-points overestimated SB*** and underestimated LPA***; Si** and vC ** underestimated MVPA; Pa (≥ 420) same trend as the toddler cut-point T12: MD = 0.5, 95% LoA (-2.5 to 3.5) ? Low
(level 2)
Albert et al. (2020) [48] n = 22
Age = 1.5 ± 0.5 (1.1 to 2) years
Sex = 54.6% girls
Activity type: Run/walk, crawl, climb, stand, sit, lie down, carried, and stroller as defined by Kwon et al. (2019)[49]
Free-living (guided play)
DVO using a developed coding scheme to score activity type by three independent raters A Waist AG-RF-HA/VA/DA-2 accuracy = 63.8%
RF + HMM: accuracy = 64.8%
Run/walk: recall = 80.0%, precision = 88.2%
Crawl: recall = 81.2%, precision = 68.3%
Climb: recall = 56.0%, precision = 29.0%
Stand: recall = 45.4%, precision = 49.2%
Sit: recall = 66.6%, precision = 66.9%
Lie down: recall = 61.7%, precision = 76.0%
Carried: recall = 58.8%, precision = 43.5%
Stroller: recall = 28.2%, precision = 41.0%
Sit, stand and stroller collapsed: accuracy = 79.3%
- Very low
(level 1)
AG-SVM-HA/VA/DA-2 accuracy = 58.6%
SVM + HMM: accuracy = 60.1%
-
AG-LR-HA/VA/DA-2 accuracy = 57.0%
LR + HMM: accuracy = 59.1%
-
AG-J48-HA/VA/DA-2 accuracy = 57.3%
J48 + HMM: accuracy = 57.5%
-
AG-kNN-HA/VA/DA-2 accuracy = 52.6%
kNN + HMM: accuracy = 54.1%
-
Pulakka et al. (2013) [43] Validation
n = 40
Age = 16.9 ± 5.8 (16.0 to 18.3) months
Sex = 60% girls
Cross-validation
n = 30
Age = 17.0 ± 0.6 (16 to 18.5) months
Sex = 60% girls
Activity intensity: SB, LPA, MPA, and VPA
Free-living (free play sessions
DVO using CPAF to score activity type and derive activity intensity (SB: stationary, no movement; LPA: stationary, limb movement, MPA: slow trunk movement; VPA: rapid trunk movement) by one rater (and 19.5% of the videos by a second rater) D Right hip AG-Pul-VM-15 Validation
SB vs. LPA: AUC-ROC = .73, CI (.67 to .80)
VPA vs. MPA: AUC-ROC = .67, CI (.56 to .78)
SB vs LPA and MVPA: AUC-ROC = .98, CI (.97 to .99)
LPA vs. MPA: AUC-ROC = .94, CI (.91 to .97)
Cross-validation
SB and LPA vs. MVPA: accuracy = 92%, Se = 94.2%, Sp = 90.9%, κ = .83
+ 
(SB + ; PA ±)
Very low
(level 1)
AG-Pul-VA-15 Validation
SB vs. LPA: AUC-ROC = .62, CI (.56 to .67)
VPA vs. MPA: AUC-ROC = .59, CI (.47 to .72)
SB vs. LPA and MVPA: AUC-ROC = .95, CI (.93 to .96)
LPA vs. MPA: AUC-ROC = . 90, CI (.87 to .94)
Cross-validation
SB and LPA vs. MVPA: accuracy = 84%, Se = 84.1%, Sp = 84.6%, κ = .67
± 
(SB + ; PA ±)
Hager et al. (2016) [42] n = 24
Age = 24.5 ± 5.2 (14.7 to 35.5) months
Sex = 41.7% girls
Activity intensity: SB, LPA, and MVPA
Laboratory (structured activities)
DO using CARS to score activity type and derive activity intensity (SB: stationary with no movement; LPA: stationary with movement of the limbs, and slow/easy translocation; MVPA: translocation fast, and with moderate effort) by one rater D Left ankle Ac-Ha-omni-30 rsp = .75
SB: Se = 81.8%, Sp = 77.5%; LPA: Se = 61.7%, Sp = 84.7%; MVPA: Se = 85.7%, Sp = 88.4%
± 
(SB ± ;
PA +)
Very low
(level 2)
Oftedal et al. (2014) [47] n = 10
Age = 29 ± 6 months
Sex = 50% girls
Activity intensity: SB and non-SB
Laboratory
(semi-structured activity sessions)
DVO using DO software Behavioral Evaluation Strategy and Taxonomy to score activity type and determine activity intensity (SB: lying/sitting with(out) limb movement, and standing still) by one rater I Waist AG–O-VM-5 Se = 82%, Sp = 83%
bias = -5.1%, LoA (-27.5 to 16.1%)
SB: +  Very low
(level 1)
AG–O-VA-5 Se = 76%, Sp = 93%, bias = -17.3%, LoA (-44.3 to 8.3%)* SB: ± 
Nam & Park (2013) [50] n = 10
Age = 22.4 ± 3.3 (16 to 29) months
Sex = 50% girls
Activity type: Wiggling, rolling, standing still, standing up, sitting down, walking, toddling, crawling, climbing up/down, and stopping (precise definitions not indicated)
Laboratory
(simulated real home environment)
DVO to score activity type by one rater I Hip mSCA-DT- VA/HA/DA/VM-n.r accuracy = 74.0% (39.5 to 97.2%) ±  Very low
(level 1)
mSCA-NB- VA/HA/DA/VM-n.r accuracy = 73.0% (28.0 to 90.9%) -
mSCA-BN-VA/HA/DA/VM-n.r accuracy = 84.8% (57.8 to 98.9%)  + 
mSCA-SVM-VA/HA/DA/VM-n.r accuracy = 86.2% (39.8 to 99.9%) + 
mSCA-kNN-VA/HA/DA/VM-n.r accuracy = 84.1% (67.8 to 94.6%) + 
mSCA-J48-VA/HA/DA/VM-n.r accuracy = 88.3% (71.7 to 98.7%) + 
mSCA-MLP-VA/HA/DA/VM-n.r accuracy = 84.8% (52.5 to 99.5%) ± 
mSCA-MLR-VA/HA/DA/VM-n.r accuracy = 86.9% (72.1 to 98.7%) + 
  1. Abbreviations: Ac Actical, AUC-ROC area under the receiver operating curve, AG ActiGraph, cts counts, BN Bayes net, C Costa’s cut-points (2014) [46], CARS children’s activity rating system, CCC concordance correlation coefficient, CI confidence interval, CPAF children’s physical activity form, d’ discriminability index, DA diagonal axis (z-axis), DO direct observation, DVO direct video observation, FFT fast Fourier transform, Ha Hager’s cut-points (2014) [42], HMM hidden Markov model, J48 decision tree (pruned), κ Kappa, kNN k-nearest neighbors, LoA limits of agreement, LPA light physical activity, LR logistic regression, MD mean difference, MLP multi-layer perceptron network, MLR multinomial logistic regression, MPA moderate physical activity, mSCA miniature semiconductor chip accelerometer, MVPA moderate-to-vigorous physical activity, N NHANES cut-points [88], NB naïve Bayes, n.r. not reported, O Oftedal’s cut-points (2014) [47], omni omnidirectional, OSRAC-P observational system for recording physical activity in children preschool, Pa Pate’s cut-points (2006) [52], PA physical activity, Pul Pulakka’s cut-points (2013) [43], Re Reilly’s cut-points (2003)[80], RF random forests, rsp correlation coefficient (Spearman rank), SB sedentary behavior, SD standard deviation, Se sensitivity, Si Sirard’s age-specific cut-points (2005) [81], Si3 Sirard’s cut-points (2005) for 3-year-olds [81], Sp specificity, SVM support vector machine, T12 Trost’s cut-points (2012) [44], VA vertical axis (y-axis), vC van Cauwenberghe’s cut-points (2011) [45], VM vector magnitude, VPA vigorous physical activity, 5 5 s epoch, 15 15 s epoch, 30 30 s epoch
  2. aAge presented as mean ± SD (range)
  3. bMethodological study quality based on newly developed checklist: VG very good, A adequate, D doubtful, I inadequate
  4. cDevice-based method described using code combinations of four elements resulting in the following format: brand-axis-approach-epoch length
  5. dStudy result rating based on COSMIN guideline: + sufficient, ± inconsistent, - insufficient, ? intermediate
  6. eQuality of evidence based on GRADE approach
  7. * p < .05
  8. ** p < .01
  9. *** p < .001