Premium
Automated movement recognition to predict motor impairment in high‐risk infants: a systematic review of diagnostic test accuracy and meta‐analysis
Author(s) -
Raghuram Kamini,
Orlandi Silvia,
Church Paige,
Chau Tom,
Uleryk Elizabeth,
Pechlivanoglou Petros,
Shah Vibhuti
Publication year - 2021
Publication title -
developmental medicine and child neurology
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.658
H-Index - 143
eISSN - 1469-8749
pISSN - 0012-1622
DOI - 10.1111/dmcn.14800
Subject(s) - psycinfo , cinahl , meta analysis , medline , cerebral palsy , motor impairment , generalizability theory , confidence interval , medicine , physical medicine and rehabilitation , receiver operating characteristic , artificial intelligence , psychology , computer science , developmental psychology , political science , law
Aim To assess the sensitivity and specificity of automated movement recognition in predicting motor impairment in high‐risk infants. Method We searched MEDLINE, Embase, PsycINFO, CINAHL, Web of Science, and Scopus databases and identified additional studies from the references of relevant studies. We included studies that evaluated automated movement recognition in high‐risk infants to predict motor impairment, including cerebral palsy (CP) and non‐CP motor impairments. Two authors independently assessed studies for inclusion, extracted data, and assessed methodological quality using the Quality Assessment of Diagnostic Accuracy Studies‐2. Meta‐analyses were performed using hierarchical summary receiver operating characteristic models. Results Of 6536 articles, 13 articles assessing 59 movement variables in 1248 infants under 5 months corrected age were included. Of these, 143 infants had CP. The overall sensitivity and specificity for motor impairment were 0.73 (95% confidence interval [CI] 0.68–0.77) and 0.70 (95% CI 0.65–0.75) respectively. Comparatively, clinical General Movements Assessment (GMA) was found to have sensitivity and specificity of 98% (95% CI 74–100) and 91% (95% CI 83–93) respectively. Sensor‐based technologies had higher specificity (0.88, 95% CI 0.80–0.93). Interpretation Automated movement recognition technology remains inferior to clinical GMA. The strength of this study is its meta‐analysis to summarize performance, although generalizability of these results is limited by study heterogeneity. What this paper adds Automated movement recognition is sensitive and specific and warrants further investigation. Sensor‐based technologies have higher specificity but are less portable. The performance of automated movement recognition is inferior to clinical General Movements Assessment. Emerging technologies such as 3D video analysis may improve its performance.