{"status":"ok","message-type":"work","message-version":"1.0.0","message":{"indexed":{"date-parts":[[2026,9,17]],"date-time":"2026-09-17T09:57:52Z","timestamp":1789639072946,"version":"4.0.0"},"reference-count":35,"publisher":"Public Library of Science (PLoS)","issue":"2","license":[{"start":{"date-parts":[[2021,2,26]],"date-time":"2021-02-26T00:00:00Z","timestamp":1614297600000},"content-version":"vor","delay-in-days":0,"URL":"http:\/\/creativecommons.org\/licenses\/by\/4.0\/"}],"content-domain":{"domain":["www.plosone.org"],"crossmark-restriction":false},"short-container-title":["PLoS ONE"],"abstract":"<jats:p>Child physical abuse is a leading cause of traumatic injury and death in children. In 2017, child abuse was responsible for 1688 fatalities in the United States, of 3.5 million children referred to Child Protection Services and 674,000 substantiated victims. While large referral hospitals maintain teams trained in Child Abuse Pediatrics, smaller community hospitals often do not have such dedicated resources to evaluate patients for potential abuse. Moreover, identification of abuse has a low margin of error, as false positive identifications lead to unwarranted separations, while false negatives allow dangerous situations to continue. This context makes the consistent detection of and response to abuse difficult, particularly given subtle signs in young, non-verbal patients. Here, we describe the development of artificial intelligence algorithms that use unstructured free-text in the electronic medical record\u2014including notes from physicians, nurses, and social workers\u2014to identify children who are suspected victims of physical abuse. Importantly, only the notes from time of first encounter (e.g.: birth, routine visit, sickness) to the last record before child protection team involvement were used. This allowed us to develop an algorithm using only information available prior to referral to the specialized child protection team. The study was performed in a multi-center referral pediatric hospital on patients screened for abuse within five different locations between 2015 and 2019. Of 1123 patients, 867 records were available after data cleaning and processing, and 55% were abuse-positive as determined by a multi-disciplinary team of clinical professionals. These electronic medical records were encoded with three natural language processing (NLP) algorithms\u2014Bag of Words (BOW), Word Embeddings (WE), and Rules-Based (RB)\u2014and used to train multiple neural network architectures. The BOW and WE encodings utilize the full free-text, while RB selects crucial phrases as identified by physicians. The best architecture was selected by average classification accuracy for the best performing model from each train-test split of a cross-validation experiment. Natural language processing coupled with neural networks detected cases of likely child abuse using only information available to clinicians prior to child protection team referral with average accuracy of 0.90\u00b10.02 and average area under the receiver operator characteristic curve (ROC-AUC) 0.93\u00b10.02 for the best performing Bag of Words models. The best performing rules-based models achieved average accuracy of 0.77\u00b10.04 and average ROC-AUC 0.81\u00b10.05, while a Word Embeddings strategy was severely limited by lack of representative embeddings. Importantly, the best performing model had a false positive rate of 8%, as compared to rates of 20% or higher in previously reported studies. This artificial intelligence approach can help screen patients for whom an abuse concern exists and streamline the identification of patients who may benefit from referral to a child protection team. Furthermore, this approach could be applied to develop computer-aided-diagnosis platforms for the challenging and often intractable problem of reliably identifying pediatric patients suffering from physical abuse.<\/jats:p>","DOI":"10.1371\/journal.pone.0247404","type":"journal-article","created":{"date-parts":[[2021,2,26]],"date-time":"2021-02-26T13:32:19Z","timestamp":1614346339000},"page":"e0247404","update-policy":"https:\/\/doi.org\/10.1371\/journal.pone.corrections_policy","source":"Crossref","is-referenced-by-count":61,"title":["A natural language processing and deep learning approach to identify child abuse from pediatric electronic medical records"],"prefix":"10.1371","volume":"16","author":[{"ORCID":"https:\/\/orcid.org\/0000-0002-9097-6511","authenticated-orcid":true,"given":"Akshaya V.","family":"Annapragada","sequence":"first","affiliation":[],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Marcella M.","family":"Donaruma-Kwoh","sequence":"additional","affiliation":[],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Ananth V.","family":"Annapragada","sequence":"additional","affiliation":[],"role":[{"vocabulary":"crossref","role":"author"}]},{"ORCID":"https:\/\/orcid.org\/0000-0003-4295-4227","authenticated-orcid":true,"given":"Zbigniew A.","family":"Starosolski","sequence":"additional","affiliation":[],"role":[{"vocabulary":"crossref","role":"author"}]}],"member":"340","published-online":{"date-parts":[[2021,2,26]]},"reference":[{"issue":"3","key":"pone.0247404.ref001","first-page":"195","article-title":"Non-accidental trauma in pediatric patients: a review of epidemiology, pathophysiology, diagnosis and treatment.","volume":"3","author":"AR Paul","year":"2014","journal-title":"Transl Pediatr"},{"key":"pone.0247404.ref002","unstructured":"Child Maltreatment 2017; (2019). https:\/\/www.acf.hhs.gov\/cb\/resource\/child-maltreatment-2017"},{"issue":"6","key":"pone.0247404.ref003","doi-asserted-by":"crossref","first-page":"e1537","DOI":"10.1542\/peds.2014-1329","article-title":"Validation of a Clinical Prediction Rule for Pediatric Abusive Head Trauma","volume":"134","author":"KP Hymel","year":"2014","journal-title":"Pediatrics"},{"key":"pone.0247404.ref004","first-page":"1","article-title":"A case study of algorithm-assisted decision making in child maltreatment hotline screening decisions","volume":"81","author":"A Chouldechova","year":"2018","journal-title":"Proceedings of Machine Learning Research"},{"issue":"2","key":"pone.0247404.ref005","doi-asserted-by":"crossref","first-page":"290","DOI":"10.1542\/peds.2014-3993","article-title":"Validation of a prediction tool for abusive head trauma","volume":"136","author":"L Cowley","year":"2015","journal-title":"Pediatrics"},{"key":"pone.0247404.ref006","doi-asserted-by":"crossref","first-page":"24","DOI":"10.1038\/s41591-018-0316-z","article-title":"A guide to deep learning in healthcare","volume":"25","author":"A Esteva","year":"2019","journal-title":"Nature Medicine"},{"key":"pone.0247404.ref007","article-title":"Deep learning approach for microarray cancer data classification","author":"H Basavegowda","year":"2020","journal-title":"CAAI Transactions on Intelligence Technology"},{"key":"pone.0247404.ref008","doi-asserted-by":"crossref","first-page":"1950022","DOI":"10.1142\/S0217984919500222","article-title":"Multi-objective differential evolution based random forest for e-health applications","volume":"33","author":"M Kaur","year":"2020","journal-title":"Modern Physics Letters B"},{"key":"pone.0247404.ref009","article-title":"Efficient Prediction of Drug-drug interaction using Deep Learning Models","volume":"14","author":"K Pr Shukla","year":"2020","journal-title":"IET Systems Biology"},{"issue":"3","key":"pone.0247404.ref010","doi-asserted-by":"crossref","first-page":"e28","DOI":"10.2196\/medinform.5909","article-title":"Prediction of Sepsis in the Intensive Care Unit With Minimal Electronic Health Record Data: A Machine Learning Approach","volume":"4","author":"T Desautels","year":"2016","journal-title":"JMIR Med Inform"},{"key":"pone.0247404.ref011","doi-asserted-by":"crossref","first-page":"26094","DOI":"10.1038\/srep26094","article-title":"Deep Patient: An Unsupervised Representation to Predict the Future of Patients from the Electronic Health Records","volume":"6","author":"R Miotto","year":"2016","journal-title":"Scientific Reports"},{"key":"pone.0247404.ref012","doi-asserted-by":"crossref","first-page":"120","DOI":"10.1016\/j.ijmedinf.2016.09.014","article-title":"A machine learning-based framework to identify type 2 diabetes through electronic health records","volume":"97","author":"T Zheng","year":"2017","journal-title":"International Journal of Medical Informatics"},{"key":"pone.0247404.ref013","article-title":"Scalable and accurate deep learning with electronic health records","volume":"18","author":"A Rajkomar","year":"2018","journal-title":"npj Digital Medicine"},{"issue":"2","key":"pone.0247404.ref014","doi-asserted-by":"crossref","first-page":"153","DOI":"10.1007\/s10654-018-0470-0","article-title":"Use of natural language processing in electronic medical records to identify pregnant women with suicidal behavior: towards a solution to the complex classification problem","volume":"34","author":"QY Zhong","year":"2019","journal-title":"Eur J Epidemiol"},{"key":"pone.0247404.ref015","doi-asserted-by":"crossref","first-page":"364","DOI":"10.1093\/jamia\/ocy173","article-title":"Natural language processing of symptoms documented in free-text narratives of electronic health records: a systematic review","volume":"26","author":"TA Koleck","year":"2019","journal-title":"Journal of the American Medical Informatics Association"},{"issue":"5","key":"pone.0247404.ref016","first-page":"e180052","article-title":"Automated Organ-Level Classification of Free-Text Pathology Reports to Support a Radiology Follow-up Tracking Engine Radiology","volume":"1","author":"JM Steinkamp","year":"2019","journal-title":"Artificial Intelligence"},{"issue":"C","key":"pone.0247404.ref017","doi-asserted-by":"crossref","first-page":"402","DOI":"10.1016\/j.eswa.2017.06.035","article-title":"Identifying child abuse through text mining and machine learning","volume":"8","author":"C Amrit","year":"2017","journal-title":"Journal Expert Systems with Applications: An International Journal archive"},{"key":"pone.0247404.ref018","volume-title":"Natural Language Processing with Python","author":"S Bird","year":"2009"},{"key":"pone.0247404.ref019","first-page":"133","article-title":"Using tf-idf to determine word relevance in document queries","volume":"242","author":"Juan Ramos","year":"2003","journal-title":"Proceedings of the 1st Instructional 1044 Conference on Machine Learning"},{"key":"pone.0247404.ref020","first-page":"2418","volume-title":"Proceedings of COLING 2016, The 26th International Conference on Computational Linguistics","author":"A. Dahou"},{"issue":"22","key":"pone.0247404.ref021","doi-asserted-by":"crossref","first-page":"4360","DOI":"10.3390\/ijerph16224360","article-title":"Comparison of Word Embeddings for Extraction from Medical Records","volume":"16","author":"A. Dudchenko","year":"2019","journal-title":"International journal of environmental research and public health"},{"key":"pone.0247404.ref022","doi-asserted-by":"crossref","first-page":"1532","DOI":"10.3115\/v1\/D14-1162","article-title":"GloVe: Global Vectors for Word Representation.","author":"J Pennington","year":"2014","journal-title":"Conference on Empirical Methods in Natural Language Processing (EMNLP 2014), Doha"},{"key":"pone.0247404.ref023","doi-asserted-by":"crossref","first-page":"160035","DOI":"10.1038\/sdata.2016.35","article-title":"MIMIC-III, a freely accessible critical care database","volume":"3","author":"A Johnson","year":"2016","journal-title":"Sci. Data"},{"key":"pone.0247404.ref024","doi-asserted-by":"crossref","first-page":"71","DOI":"10.1186\/s12911-019-0781-4","article-title":"Clinical text classification with rule-based features and knowledge-guided convolutional neural networks","volume":"19","author":"L. Yao","year":"2019","journal-title":"BMC Med Inform Decis Mak"},{"key":"pone.0247404.ref025","unstructured":"Chollet F et al, Keras (2015) GitHub repositoryhttps:\/\/github.com\/fchollet\/keras"},{"issue":"11","key":"pone.0247404.ref026","doi-asserted-by":"crossref","first-page":"e0224365","DOI":"10.1371\/journal.pone.0224365","article-title":"Machine learning algorithm validation with a limited sample size","volume":"14","author":"A Vabalas","year":"2019","journal-title":"PLoS ONE"},{"key":"pone.0247404.ref027","article-title":"Visualizing and Understanding Neural Models in NLP","author":"J Li","year":"2016","journal-title":"arXiv:1506.01066v2"},{"issue":"4","key":"pone.0247404.ref028","doi-asserted-by":"crossref","first-page":"305","DOI":"10.1093\/brief-treatment\/mhm021","article-title":"Intimate partner violence and child maltreatment: Overlapping risk","volume":"7","author":"A. J. Zolotor","year":"2007","journal-title":"Brief Treatment and Crisis Intervention"},{"key":"pone.0247404.ref029","article-title":"Prevalence and correlates of the co-occurrence of family violence: a meta-analysis on family polyvictimization","author":"K. L. Chan","year":"2019","journal-title":"Trauma, Violence, & Abuse"},{"key":"pone.0247404.ref030","author":"U.S. Department of Health","year":"2017"},{"key":"pone.0247404.ref031","article-title":"Novel Efficient RNN and LSTM-Like Architectures: Recurrent and Gated Broad Learning Systems and Their Applications for Text Classification","author":"J. Du","journal-title":"IEEE Transactions on Cybernetics"},{"key":"pone.0247404.ref032","first-page":"106486","volume":"210","author":"L Jiang","year":"2020","journal-title":"DECAB-LSTM: Deep Contextualized Attentional Bidirectional LSTM for cancer hallmark classification, Knowledge-Based Systems"},{"issue":"1","key":"pone.0247404.ref033","doi-asserted-by":"crossref","first-page":"67","DOI":"10.1542\/peds.2008-3632","article-title":"Bruising characteristics discriminating physical child abuse from accidental trauma","volume":"125","author":"MC Pierce","year":"2010","journal-title":"Pediatrics"},{"issue":"2","key":"pone.0247404.ref034","doi-asserted-by":"crossref","first-page":"335","DOI":"10.1016\/j.burns.2017.08.018","article-title":"Raising suspicion of maltreatment from burns: Derivation and validation of the BuRN-Tool","volume":"44","author":"AM Kemp","year":"2018","journal-title":"Burns"},{"issue":"1","key":"pone.0247404.ref035","doi-asserted-by":"crossref","first-page":"e20153756","DOI":"10.1542\/peds.2015-3756","article-title":"Validation of the Pittsburgh infant brain injury score for abusive head trauma","volume":"138","author":"RP Berger","year":"2016","journal-title":"Pediatrics"}],"container-title":["PLOS ONE"],"original-title":[],"language":"en","link":[{"URL":"https:\/\/dx.plos.org\/10.1371\/journal.pone.0247404","content-type":"unspecified","content-version":"vor","intended-application":"similarity-checking"}],"deposited":{"date-parts":[[2021,2,26]],"date-time":"2021-02-26T13:32:59Z","timestamp":1614346379000},"score":1,"resource":{"primary":{"URL":"https:\/\/dx.plos.org\/10.1371\/journal.pone.0247404"}},"subtitle":[],"editor":[{"given":"Khanh N.Q.","family":"Le","sequence":"first","affiliation":[],"role":[{"vocabulary":"crossref","role":"editor"}]}],"short-title":[],"issued":{"date-parts":[[2021,2,26]]},"references-count":35,"journal-issue":{"issue":"2","published-online":{"date-parts":[[2021,2,26]]}},"URL":"https:\/\/doi.org\/10.1371\/journal.pone.0247404","relation":{},"ISSN":["1932-6203"],"issn-type":[{"value":"1932-6203","type":"electronic"}],"subject":[],"published":{"date-parts":[[2021,2,26]]}}}