Skip to main content

Table 3 Criterion and convergent validity of accelerometer-based methods for infants, 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) Methodological study qualityb Placement Device-based methodc Results Study result ratingd Quality of evidencee (level of evidence)
Criterion validity Galland et al. (2012) [30] n = 31
Age = 13.0 ± 3.1 (10.0 to 22.3) weeks
Sex = 27.3% girls
Sleep quality: sleep latency, TST, sleep efficiency, and WASO (duration and number)
Laboratory
(daytime nap)
Polysomnography to score sleep stages as sleep (active, quiet, intermediate) and wake VG Shin Ac-CS-omni-15 accuracy = 86.3%, Se = 85.7%, Sp = 84.3%, κ = .66, PABAK = .72; Sleep latency rp = .79*** (underestimated by 3 min*), TST rp = .83***, Sleep efficiency rp = .87***, WASO (duration) rp = .48** (overestimated by 6.5 min***), WASO (number) rp = .35 (overestimated by 3***) +  Low
Ac-CS-omni-30 accuracy = 86.1%, Se = 84.7%, Sp = 87.4%, κ = .66, PABAK = .72 + 
Ac-CS-omni-60 accuracy = 84.5%, Se = 79.8%, Sp = 90.2%, κ = .63, PABAK = .69 -
Ac-S0-omni-15 accuracy = 82.3%, Se = 94.8%, Sp = 57.8%, κ = .55, PABAK = .65; Sleep latency rp = .80*** (underestimated by 5 min***), TST rp = .76*** (overestimated by 11 min*), Sleep efficiency rp = .76*** (overestimated by 16.5%***), WASO (duration) rp = .48** (overestimated by 7 min***), WASO (number) rp = .38* (overestimated by 3***) ± 
Ac-S0-omni-30 accuracy = 86.3%, Se = 91.0%, Sp = 77.7%, κ = .65, PABAK = .72 ± 
Ac-S0-omni-60 accuracy = 84.5%, Se = 80.0%, Sp = 89.9%, κ = .63, PABAK = .69 ± 
Ac-PS-omni-15 accuracy = 84.3%, Se = 95.1%, Sp = 63.1%, κ = .59, PABAK = .69; Sleep latency rp = .93***, TST rp = .80*** (overestimated by 11 min*), Sleep efficiency rp = .81*** (overestimated by 13.6%*), WASO (duration) rp = .50** (overestimated by 3.5 min***), WASO (number) rp = .13 (overestimated by 3***) ± 
Ac-PS-omni-30 accuracy = 86.8%, Se = 90.8%, Sp = 79.0%, κ = .66, PABAK = .74 ± 
Ac-PS-omni-60 accuracy = 84.6%, Se = 79.9%, Sp = 90.3%, κ = .63, PABAK = .69 ± 
Ac-WSC-omni-15 accuracy = 83.2%, Se = 94.6%, Sp = 60.2%, κ = .57, PABAK = .66; Sleep latency rp = .78*** (underestimated by 4 min*), TST rp = .78*** (overestimated by 10 min*), Sleep efficiency rp = .79*** (overestimated by 15.2%*), WASO (duration) rp = .41* (overestimated by 7 min***), WASO (number) rp = .29 (overestimated by 3***) ± 
Ac-WSC-omni-30 accuracy = 86.5%, Se = 90.9%, Sp = 78.2%, κ = .66, PABAK = .73 ± 
Ac-WSC-omni-60 accuracy = 84.5%, Se = 80.0%, Sp = 89.9%, κ = .63, PABAK = .69 ± 
Insana, Gozal, & Montgomery-Downs (2010) [31] n = 22
Age = 14.1 ± 0.6 (13.0 to 15.0) months
Sex = 45.5% girls
Sleep quality: TST and WASO (duration)
Laboratory (in the hospital)
Polysomnography to score sleep stages as sleep and wake VG Ankle Aw-ACT40-uni-15 accuracy = 89.6%, Se (89.0 to 96.3%), Sp = 58.9%
TST: rp = .83***, ICC = .80, MD = -72.3 (SD = 61.5) min***, d = .70
WASO: rp = .52*, ICC = .65, MD = 13.9 (SD = 30.9) min*, d = .44
±  Low
Aw-ACT80-uni-15 TST: rp = .84***, ICC = .86, MD = 52.1 (SD = 60.4) min**, d = .50
WASO: rp = .52*, ICC = .65, MD = 6.4 (SD = 26.8) min
± 
Rioualen et al. (2015) [32] n = 24
Age = 2.5 ± 0.7 days
Sex = 50.0% girls
Sleep stage: wake, active sleep, and quiet sleep
Laboratory (in the hospital)
Polysomnography to score sleep stages as sleep (active, quiet) and wake VG Wrist Aw-ASA40-uni-60 Se = 93%, 95% CI (89 to 96)
Sp = 20%, 95% CI (14 to 27)
±  Low
Ankle Aw-ASA40-uni-60 accuracy = 58.5%, Se = 87%, 95% CI (81 to 93), Sp = 31%, 95% CI (24 to 39) -
Lewicke, Sazonov, & Schuckers (2004) [35] n = 25
Age = n.r
Sex = n.r
Sleep stage: wake, active sleep, and quiet sleep
Laboratory
Polysomnography to score sleep stages as sleep (active, quiet, indeterminate) and wake I Left hip mSCA-LVQ-uni-30 Training: accuracy = 80.7%, Se = 94.6%, Sp = 48.2%
Validation: accuracy = 75.3%, Se = 92.3%, Sp = 42.4%
±  Very low
Convergent validity Camerota et al. (2018) [33] n = 82
Age = 3.6 ± 0.6 (2.7 to 5.2) months
Sex = 43% girls
Sleep quality: sleep onset time, rise time, sleep period, WASO (number and duration), TST, and longest sleep period
Free-living
(at home)
Videosomnography to score sleep stages as sleep and wake VG Left ankle Aw-ACTdef-uni-15 κ = .47, Se = 52%, Sp = 95%
Sleep onset time: r = .79*, underestimated by 29 min***, d = .43
Rise time: r = .76*
Sleep period: r = .78*, overestimated by 23.6 min*, d = .28
WASO: (number) r = .47*, underestimated by 1.1***, d = .54; (duration): r = .59*, overestimated by 58.4 min***, d = 72
TST: r = .54*
Longest sleep period: r = .38*, overestimated by 35.5 min*, d = .28
- Moderate
(level 2)
Greenspan et al. (2021) [29] n = 16
Age = 3.1 ± 1.1 (1.8 to 5) months
Sex = 51% girls
Body position: supine (parallel to floor stomach facing up), reclined (trunk tilted about 45° posteriorly from upright), upright (90°), inclined (trunk tilted about 45° anteriorly from upright), and prone (parallel to floor stomach facing down)
Free-living (structured and free play)
DVO to score body position by two independent raters using outlined definitions VG Trunk GG-Gr-HA/VA/DA-1 κw = .84, 95% CI (.83 to .84)
Supine: accuracy = 90.7%, rsp = .97***
Reclined: accuracy = 73.5%, rsp = .89***
Upright: accuracy = 83.1%, rsp = .97***
Inclined: accuracy = 41.4%, rsp = .67**
Prone: accuracy = 78.4%, rsp = .96***
±  Low
(level 1)
Jun & Choi (2020) [40] n = 9
Age = 144 ± 284.5 (2 to 720) days
Sex = 59.1% girls
Activity type: sleeping (without substantial movement), strong movement (struggling with crying or in agony), weak movement (moving in comfy state), and movement by external force (from nurse or caregiver)
Laboratory (at the hospital)
DVO to score activity type by two independent raters using a 3-stage classification scheme A Chest ICM-DNN-HA/VA/DA-4 accuracy = .96, F-score = .95, precision = .98, recall = .93, Sp = .98
Sleep: accuracy = .97, F-score = .99, precision = .99, recall = 1.0, Sp = .94
Strong movement: accuracy = .97, F-score = .95, precision = .96, recall = .94, Sp = .99
Weak movement: accuracy = .95, F-score = .93, precision = .97, recall = .90, Sp = .99
External force: accuracy = .95, F-score = .95, precision = 1.0, recall = .89, Sp = 1.0
+  Very low
(level 1)
Airaksinen et al. (2020) [39] n = 22
Age = 6.7 ± 0.8 (4.5 to 7.7) months
Sex = 59.1% girls
Body position: prone (navel on floor), supine (lower back on floor), side left/right, crawl posture (supported by hands, and knees or feet);
Movement: macro still (no movement), turn left/right (change of posture along prone-side-supine axis), pivot left/right (change of facing whole body direction, without movement), crawl proto (practice crawl without moving during prone/side posture and move multiple limbs during supine posture), crawl commando (crawl with forward movement)
Laboratory
(semi-structured)
DVO to score body position and movement by three independent raters using a developed annotation scheme A Arms + legs; proximally SM-CNN-VM-2 Movement track: F-score = 80%
Full agreement: posture: accuracy = 99.1%, movement: accuracy = 90.7%; All frames: posture: accuracy = 98.2%, movement: accuracy = 81.7%
Most frequent confusions in prone-side-supine axis as well as between crawl posture and prone
Individual sensors lower performance (left arm posture: accuracy = 71%, movement: accuracy = 70%, left leg posture: accuracy = 90%, movement: accuracy = 68%) compared to four-sensor setup (posture: accuracy = 95%, movement: accuracy = 80%), and two-sensor setup (right leg + arm posture: accuracy = 94%, movement: accuracy = 78%; left leg + arm posture: accuracy = 95%, movement: accuracy = 79%; left + right arm posture: accuracy = 85%, movement: accuracy = 72%; left + right leg posture: accuracy = 93%, movement: accuracy = 72%)
+ 
Arm: -
Arms: ± 
Leg: ± 
Legs: ± 
Combined
2-sensor: ± ; 4-sensor: + 
Very low
(level 1)
SM-SVM-VM-2 Movement track: F-score consistently 5–10% lower performance in prone positions (crawl proto*, turn right***; pivot left*)
Posture classification comparable between CNN and SVM
± 
Smith et al. (2015) [41] n = 12
Age = 6.8 ± 2.9 (1 to 12) months
Sex = 66.7% girls
Activity type: leg movement (change of limb position)
Laboratory
DVO to score activity type by one rater D Legs Op-AAV-VM-n.r accuracy = 92.7%, Se = 92.0% +  Very low
(level 1)
Hewitt et al. (2019) [34] n = 32
Age = 15.2 ± 6.4 (4.7 to 24.9) weeks
Sex = 40.6% girls
Body position: prone on floor (tummy with both hips touching the floor), prone supported (held in prone, on parent’s chest), and non-prone (supine, left/right side lying, reclined in car seat/pram, upright, supported sitting, cradle hold)
Laboratory (structured positions)
DVO to score body position by one rater (one randomly chosen video was scored by four independent raters) using outlined definitions D Right hip AG-He-VA/HA-1 Prone: accuracy = 90%, MD = -18.3 s, LoA (-97.0 to 60.3)
Non-prone: accuracy = 99.9%, MD = -0.2 s, LoA (-1.2 to 0.9)
Prone supported: accuracy = 63.6%, MD = -127.3 s, LoA (-324.7 to 70.2)
±  Very low
(level 1)
GA-Ai-VA/HA-1 Prone: accuracy = 95.4%, MD = -8.4 s, LoA (-78.2—61.3)
Non-prone: accuracy = 98%, MD = 31.2 s, LoA (-154.9 to 92.4)
Prone supported: accuracy = 52.2%, MD = -166.9 s, LoA (-390.7 to 56.8)
± 
Ankle AG-He-VA/DA-1 Prone: accuracy = 87.9%, MD = -22.1 s, LoA (-124.0 to 79.7)
Non-prone: accuracy = 96.3%, MD = -56.6 s, LoA (-209.5 to 96.3)
Prone supported: accuracy = 53.3%, MD = -163.1 s, LoA (-431.7 to 105.6)
± 
Chest MB-He-VA/HA/DA-1 Prone: accuracy = 79.2%, MD = -38 s, LoA (-194.5 to 118.5)
Non-prone: accuracy = 99.9%, MD = 2.0 s, LoA (-14.0 to 10.8)
Prone supported: accuracy = 66.1%, MD = -113 s, LoA (-355.6 to 129.6)
-
Horger et al. (2021) [37] n = 9
Agestart = 7.96 ± 2.51 (0.89 to 8.19) months
Ageend = 10.89 ± 1.14 (9.5 to 13.08) months
Sex = 44.0% girls
Sleep quality: sleep onset, morning wake time, WASO, sleep efficiency, and sleep duration
Free-living (at home)
Videosomnography to score sleep quality D Left ankle MM-PSinf-uni-60 Sleep onset: rp= .94**, MD = 0.76 min, SD = 34.5 min
Morning wake: rp= .74, MD = -14.1 min, SD = 40.4 min
WASO: rp= .20, MD = 1.2, SD = 3.9
Sleep efficiency: rp= .30
Sleep duration: rp= .36
- Very low
(level 2)
MM-PS-uni-60 Sleep onset: rp= .86**
Morning wake: rp= .55
WASO: rp= .27
Sleep efficiency: rp= .51
Sleep duration: rp= .12
-
Sadeh et al. (1995) [38] n = 41 (10 newborns; 11 3-month-olds; 10 6-month-olds; 10 12-month-olds)
Age = 0 to 12 months
Sex = 46.4% girls
Sleep stage: sleep (active and quiet) and wake
Free-living
(newborns at hospital; others at home)
DO using Thoman’s observation scheme (1975, 1995) to score sleep stage as sleep (active and quiet), wake, sleep–wake transition and uncertain D Left ankle AMA-PS-uni-60 All infants: accuracy = 83.4% (sleep–wake = 95.6%; wake = 93.5%; active sleep = 74.9% and quiet sleep = 78.0%) (excl. 12-month-olds)
Newborns: accuracy = 74.9% (wake = 82.8%; active sleep = 74.4.1%, quiet sleep = 54.9%)
3-month-olds: accuracy = 87.3% (wake = 92.5%, active sleep = 78.3%, quiet sleep = 87.2%; sleep–wake = 93.8%); 6-month-olds: accuracy = 83.2% (wake = 97.8%, active sleep = 66.4%, quiet sleep = 76.7%; sleep–wake = 97.9%); 12-month-olds: wake = 99.3%; sleep–wake = 97.2%
Wake time: r (.85 to .99), % active sleep time: r (.78 to .98), % quiet sleep time: r (.36 to .85) (lowest correlations in newborns)
Wake: MD (0.2 to 3.5%), sleep: MD (0.4 to 11.3%)
± 
 < 3 months: -
 > 3 months: + 
Very low
(level 3)
Gnidovec, Neubauer, & Zidar (2002) [36] n = 10
Age = 1 to 6 months
Sex = 40% girls
Sleep stage: Sleep (active and quiet) and wake (active and quiet)
Free-living
(at home)
DO using Thoman’s observation scheme (1990) to score sleep stage as sleep (active and quiet), wake (active and quiet), sleep–wake transition and uncertain I Left ankle Gw-ASW-uni-10 1-month: accuracy = 72.0% (sleep = 88.7%, wake = 57.5%); 3-months: accuracy = 90.1% (sleep = 97.0%, wake = 84.5%); 6-months: accuracy = 95.0% (sleep = 88.4%, wake = 98.0%)
After 3rd month 88.9% and 6th month 94.4% accuracy (calibration sample n = 5)
Invalid discrimination between active- and quiet sleep all age groups < 75%, best for 3-month-olds (accuracy = 74.3%; active sleep: accuracy = 49.3%, quiet sleep: accuracy = 88.3%), as compared to 1-month (accuracy = 58.5%; active sleep: accuracy = 35.9%, quiet sleep: accuracy = 90.5%) and 6-months (accuracy = 66.5%; active sleep: accuracy = 39.9%, quiet sleep: accuracy = 87.1%)
± 
1 month: -
 > 3 months: + 
Very low
(level 3)
  1. Abbreviations: AAV acceleration and angular velocity algorithm [41], Ac Actical, ACTdef Actiware software with default wake threshold value of .888 * mean acceleration, ACT40 Actiware software with wake threshold value of 40, ACT80 Actiware software with wake threshold value of 80, ASA40 Actiwatch activity and sleep analysis software with wake threshold value = 40, AG ActiGraph, Ai Activinsights software, AMA AMA-32, ASW automatic sleep wake scoring algorithm, Aw Actiwatch, CI confidence interval, CNN convolutional neural network, CS count-scaled algorithm [30], d Cohen’s d, DA diagonal axis (z-axis), DNN deep neural network, DO direct observation, DVO direct video observation, GA GENEActiv, GG get around garment with ADXL335, Gr Greenspan’s cut-points (2021) [29], Gw Gaewiler, HA horizontal axis (x-axis), He Hewitt’s cut-points (2019) [34], ICC intraclass correlation coefficient, ICM ICM20600 chip, κ Kappa, κw weighted Kappa, LoA limits of agreement, LVQ neural network learning vector quantization, MB MonBaby, MD mean difference,MM MicroMini sleep watch, mSCA miniature semiconductor chip accelerometer, n.r. not reported, omni omnidirectional, Op Opal APDM, PABAK prevalence- and bias-adjusted Kappa, PS probability scaled algorithm [38], PSinf probability scaled algorithm for infants [92], r correlation coefficient (unknown), rp correlation coefficient (Pearson), rsp correlation coefficient (Spearman rank), SD standard deviation, Se sensitivity, SM Suunto Movesense sensor, Sp specificity, SVM support vector machine, S0 zero-threshold computation [93], TST total sleep time, uni uniaxial (axis was not specified), VA vertical axis (y-axis), VM vector magnitude, WASO wake after sleep onset, WSC weighted sum activity algorithm [30], 1 1 s epoch, 2 2 s epoch, 4 4 s epoch, 10 10 s epoch, 15 15 s epoch, 30 30 s epoch,  60 60 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