Faculty Bibliography

Social media have created a new environmental context for the study of social and human behavior and services. Although social work researchers have become increasingly interested in the use of social media to address social problems, they have been slow to adapt tools that are flexible and convenient for analyzing social media data. They have also given inadequate attention to bias and representation inherent in many multimedia data sets. This article introduces the Visual and Textual Analysis of Social Media (VATAS) system, an open-source Web-based platform for labeling or annotating social media data. We use a case study approach, applying VATAS to a study of Chicago, IL, gang-involved youth communication on Twitter to highlight VATAS"™ features and opportunities for interdisciplinary collaboration. VATAS is highly customizable, can be privately held on a secure server, and allows for export directly into a CSV file for qualitative, quantitative, and machine-learning analysis. Implications for research using social media sources are noted.

Mining social media data for studying the human condition has created new and unique challenges. When analyzing social media data from marginalized communities, algorithms lack the ability to accurately interpret off-line context, which may lead to dangerous assumptions about and implications for marginalized communities. To combat this challenge, we hired formerly gang-involved young people as domain experts for contextualizing social media data in order to create inclusive, community-informed algorithms. Utilizing data from the Gang Intervention and Computer Science Project-a comprehensive analysis of Twitter data from gang-involved youth in Chicago-we describe the process of involving formerly gang-involved young people in developing a new part-of-speech tagger and content classifier for a prototype natural language processing system that detects aggression and loss in Twitter data. We argue that involving young people as domain experts leads to more robust understandings of context, including localized language, culture, and events. These insights could change how data scientists approach the development of corpora and algorithms that affect people in marginalized communities and who to involve in that process. We offer a contextually driven interdisciplinary approach between social work and data science that integrates domain insights into the training of qualitative annotators and the production of algorithms for positive social impact.

Objectives: We will: 1) describe the WonderLab, a digital health initiative within the New York University Langone Health Department of Child and Adolescent Psychiatry; 2) introduce When to Wonder: Picky Eating, which is the WonderLab's first early childhood mental health digital study; and 3) present preliminary data from this study. Our first objective is to demonstrate how smartphone-based tools developed to assess children in their homes and the use of advanced data analytics can transform how, when, and where we assess young children's development and mental health. Our second objective is to share how our multidisciplinary team and agile development methodology enable us to build and launch a consumer-facing pediatric health app within an academic medical center.
Method(s): The WonderLab creates scalable mobile digital health tools to collect multimodal data in children's homes at the individual, family, and population levels. In December 2018, we released When to Wonder: Picky Eating, a national study with consent, enrollment, study activities, and feedback fully integrated in iOS and Android apps that parents download from the app stores. When to Wonder: Picky Eating focuses on the emotions and behaviors related to picky eating in children under the age of 7 years. Data sources include parent-report, video, audio, and an active task that children and parents play independently to quantify children's food preferences.
Result(s): We will present preliminary data from When to Wonder: Picky Eating to characterize normative and clinically significant emotions and behaviors related to picky eating. We will also share data on recruitment and engagement using social media, app performance, and "lessons learned" about digital pediatric health.
Conclusion(s): We create clinically and scientifically valid digital tools that parents and children want to use. We integrate clinical, scientific, engineering, design, data science, and bioethics expertise with collaborative user engagement and a "build, measure, learn" agile development culture. Our app-based study demonstrates how to build digital health tools that collect and analyze population-level and individual-level, multimodal data about children and families in the home. These new tools and approaches have the potential to transform our engagement with families and our delivery of care. EA, EC, MED
Copyright

Objectives: Digital innovation has the potential to transform both the science and practice of child mental health. Creation of pediatric digital health tools requires that bioethics, human-centered design, and clinical and scientific expertise are integrated with digital tool development, digital data collection, and data analytics. In this talk, we will describe the opportunities for innovations in pediatric digital mental health and the concurrent ethical and security risks. We will then present a framework and design methodology for creating ethical, human-centered, clinically informed, and evidence-based digital tools for children's mental health.
Method(s): The data presented will come from our experience founding and leading the New York University Langone Department of Child and Adolescent Psychiatry's WonderLab, which creates pediatric digital mental health tools that are evidence based, scalable, and ethical, as well as beautiful and fun so that parents and children would want to use them. The WonderLab brings clinical, scientific, digital engineering, digital design, data science, and bioethics expertise together with user engagement and a "build, measure, learn" agile development culture and methodology. We will use the WonderLab team's development and launch of our first app-based study, "When to Wonder: Picky Eating," to illustrate our framework and methodology.
Result(s): We will describe the innovation opportunities in pediatric digital mental health, including innovation in measurement, engagement, access, and collaborative methodologies. We will then present the ethical, privacy, security, and safety risks related to digital health applications and app-based data collection with children and their families. Finally, we will describe how the WonderLab team, methodology, and products innovate across multiple domains within an explicit ethical and clinically informed framework.
Conclusion(s): Digital innovation and data science have great potential to address the challenges facing our patients and our field. To build ethical and useful digital health tools for children's mental health requires multidisciplinary teams, user engagement, collaborative agile methodology, and a framework that ensures that innovations are integrated with and reflect our ethics and commitment to children. R, COMP, DAM
Copyright

OBJECTIVES/OBJECTIVE:Access to firearms is an independent risk factor for completed suicide and homicide, and the American Academy of Pediatrics recommends that pediatricians screen and counsel about firearm access and safe storage. This study investigates how often pediatric residents screen for access to firearms or counsel about risk-reduction in patients with suicidal or homicidal ideation. METHODS:Retrospective chart review of visits by patients under the age of 19 years presenting to the pediatric emergency department (ED) of a tertiary academic medical center January-December 2016. Visits were eligible if there was an ultimate ED discharge diagnosis of "suicidal ideation," "suicide attempt," or "homicidal ideation" as identified by ICD-10 codes and the patient was seen by a pediatric resident prior to evaluation by psychiatry. Descriptive statistics were used to analyze results. RESULTS:Ninety-eight patients were evaluated by a pediatric resident for medical assessment before evaluation by a psychiatry team during the study period and were therefore eligible for inclusion. Screening for firearm access was documented by a pediatric resident in 5/98 (5.1%) patient encounters. Twenty-five patients (25.5%) had no documented screening for firearm access by any provider during the ED visit, including in five cases when patients were discharged home. CONCLUSIONS:Pediatric residents rarely document screening for firearm access in patients with known suicidal or homicidal ideation who present to the ED. Additional understanding of the barriers to screening and potential strategies for improving screening and counseling are critical to providing appropriate care for high-risk pediatric patients.

Adolescents are an important population to represent in biobanks. Inclusion of biospecimens from adolescents advances our understanding of the long-term consequences of pediatric disease and allows the discovery of methods to prevent adult diseases during childhood. Consent for biobanking is complex, especially when considering adolescent participation, as it brings up issues that are not present with general clinical research. The development and successful implementation of an adolescent capacity assessment tool applied specifically to biobanking can potentially provide researchers and clinicians with contextualized information on participants' understanding, appreciation, reasoning, and voluntary choice for biobanks. This tool would enhance current studies looking at the role of shared decision-making in biobanking, as well as provide a formal measurement when considering decisions around pediatric and adolescent biobanking participation. This study adapted the MacCAT-CR for use with a hypothetical adolescent biobank study and examines predictors of MacCAT-CR scores on healthy and chronically ill adolescents.

Purpose: Instagram has grown over the years to become one of the most popular social media platforms, and three quarters of teens who use social media use Instagram. In recent years, "Finstas", or "fake" Instagrams have grown in popularity among US teenagers. Finsta accounts are subsidiary Instagram accounts with highly selected audiences where owners can post material that is not associated with their main account. Public Twitter posts (tweets) can provide insight into communication about these clandestine accounts not available through Instagram due to the inherent private nature of these accounts. This exploratory study uses natural language processing (NLP) techniques on tweets about Finsta accounts to gain insight into this phenomenon. Method(s): An R-script was developed to pull data from the Twitter API to capture tweets longitudinally that were in English, from North America, and specifically mention some form of the stem and lemmatized word "Finsta." As there are no current studies on Finsta accounts, a comprehensive thematic analysis was then performed on the corpus of tweets to develop qualitative insights on this phenomenon. A quantitative process involved further cleaning and removing of stop-words to develop a Ngram frequency chart of the lemmatized words in the corpus of tweets to better understand the ways in which people were communicating about Finsta accounts. Result(s): 10,000 tweets containing the word "Finsta" were pulled from the Twitter API. After a comprehensive cleaning process, 5,159 tweets were then analyzed qualitatively to identify themes as a preliminary inquiry into this relatively new phenomenon. Themes identified within the corpus were: a desire for privacy compared to their main account, a place to share information that may be politically incorrect or would get users in trouble if shared on accounts with wider viewership, and a place to showcase real life. Ngram frequency words highlight similar words common to social media, "follow," "like," and "post" being amongst the most popular; however, within this corpus there are high frequencies of the words, "private," "sad," "nudes," "spam," "rant," "exposed," "emotional," and "outlet," tied to contextual themes indicating that Finstas may be an outlet for emotional catharsis in a "safe space." A Finsta user may have twenty followers (as opposed to 1000 on their main account) that include their closest friends. They may post blurry pictures without filters, with long captions detailing their negative emotional state. This sensitive content is posted with the underlying assumption that their friends will keep this information private. Conclusion(s): Preliminary analyses indicate that Finstas are a new way for teens to connect with peers in a controlled space online, where they can truly express themselves. Additionally, there is also a great deal of gossip, exhibitionism, risk-taking, and other attention-seeking behaviors typical of adolescence that manifest in ways not seen on users' primary accounts. Finsta accounts fulfill a vital role in the lives of adolescents looking for ways to authentically connect, share, and create community that is not offered through traditional uses of social media. Sources of Support: NYU CAMS Undergraduate Internship

Purpose: There are an ever increasing number of social media platforms available for people to connect and build online communities. Pro-eating disorder communities, notably anorexia nervosa (AN), have developed a steady presence on Twitter. While these communities can be beneficial for individuals who are not yet ready or able to seek professional help, multiple studies have revealed the detrimental side effects these communities can have on users, such as normalization of maladaptive behaviors, encouragement of behaviors and sharing new ways to perpetuate behaviors. To date, few studies have investigated the ways in which available social media data from self-identified anorexic individuals could be used to better inform screening, treatment, and follow-up with these individuals. This study evaluated the ways in which natural language processing (NLP) and machine learning algorithms, coupled with qualitative methods, could collect, categorize, and inform clinician insights about pro-eating disorder communities. Method(s): Twitter crawling algorithms were developed and deployed through the Twitter API to find tweets based on key words such as: "ana" "proana" "thinspo" and "meanspo." This initial corpus of tweets containing the identified keywords was then qualitatively assessed to further refine the algorithmic process of identifying appropriate tweets and removing irrelevant tweets. This process resulted in a cleaned corpus of 970 unique tweets over a ten-day period. This cleaned dataset was then utilized for NLP to identify common words, phrases, and topics. Concurrently, data was hand coded in a thematic analysis process to identify deeper themes within the dataset. These themes could inform qualitative lines of inquiry as well as machine learning systems. Qualitative insights were utilized to improve sentiment analysis as well as classification of unstructured data though a semi-supervised machine learning process. Result(s): Analysis revealed that emotional restraint was not present and judgement of one's self on external standards was present within these Twitter communities. Additionally, users' frequently requested "meanspo," an extension of thinspo that serves as an inspiration for thinness by using aggressive and abrasive rhetoric to encourage users to aspire for thinness. Additionally, posts asking for an "ana buddy," a partner to help users hold each other accountable to their AN behaviors, were extremely common. Additional information about caloric restrictions, weekly weight loss goals, and a large number of individuals tweeting from residential treatment for ED about lying about wanting to get better as a means to be released, as well as genuine statements about wanting to change. Conclusion(s): AN is a complicated disease with multiple causes, side effects and comorbid illnesses. This pilot study offers a promising 'first-step' approach towards understanding the mindset, experiences, and potential gaps within current ED treatment approaches from a patient perspective. NLP and ML processes are now developed to scan, collect, and analyze this data in an ongoing way to develop new AI processes with the ultimate goal of identifying individuals with a higher likelihood of wanting to enter treatment. Overall, the present study highlights the benefits of using new available data streams to develop patient-informed comprehensive care models. Sources of Support: NYU CAMS Undergaduate Internship

Purpose: In 2016 there were roughly 77,000 PrEP users in the United States, while over 1.2 million Americans were identified as "high-risk" for HIV infection. The reasons for this discrepancy are vast; however, potential reasons that have been identified are stigma, ineffective or poorly targeted marketing, access, and cost, amongst other factors. This pilot project seeks to understand the ways in which people and companies talk about PrEP on social media to glean deeper insights on methods to increase PrEP use. The increased use of social media gives researchers, clinicians, policymakers, and health organizations the opportunity to have access to real time data and potentially influence awareness of PrEP. This inductive exploratory study uses natural language processing (NLP) and content analysis to examine the ways in which people are using social media to talk about PrEP. Method(s): An R script was utilized to crawl Twitter the Twitter API based on keywords related to PrEP and HIV, plus all lemmatized variations related to the word pair. Data cleaning was then performed to remove tweets that were not in English, tweets that had been retweeted, as well as removing any identifying information. The resulting data frame was then used both qualitatively and quantitatively for analysis. Qualitative analysis involved a comprehensive reading of tweets, development of a category dictionary, and identification of themes that would help to train an algorithm to automatically process and count tweets based on its category. The quantitative process involved further cleaning and removing of stop-words to develop a Ngram frequency cloud as well as development of a process to automatically categorize the different types of tweets based on the type of tweet (advertisement, question about PrEP, comment on cost or availability, criticism of manufacturer, etc.). Result(s): This processes resulted in identification of 587 unique HIV related PrEP tweets. Qualitative insights from this reduced dataset indicated that there are preventative concerns related to access and cost which may be preventing high-risk individuals from getting PrEP. Algorithmic sorting and categorization processes also identified concerns about targeted marketing, specifically the lack of campaigns focusing on transgender, female, and minority communities. Our bootstrap method of training and testing resulted in a process that had an 80% likelihood of identifying, analyzing, and classifying HIV related PrEP tweets. Once classified, 40% of tweets were advertising and messaging, the rest were concerns about cost (31%), requests for info/ways to pay for PrEP (20%), as well as other non-classified comments. Conclusion(s): There are a number of different conversations about HIV/PrEP awareness happening on Twitter. However, access and cost were consistently the most common themes being discussed. Currently, a 30-day supply of PrEP costs between 0-$1600, in the US, which may be creating a substantial barrier to further reducing HIV rates. Additionally, Improving online marketing strategies of PrEP could increase awareness and use by offering targeted information as well as identification of local resources to those interested or in need. Sources of Support: NYU CAMS Undergradaute Internship