Application of Topic Modeling on Artificial Intelligence Studies as a Foundation to Develop Ethical Guidelines in African American Dementia Caregiving.

We applied natural language processing and topic modeling to publicly available abstracts and titles of 263 papers in the scientific literature mentioning AI and demographics (corpus 1 before Covid-19, corpus 2 after Covid-19) extracted from the MEDLINE database. We found exponential growth of AI studies mentioning demographics since the pandemic (Before Covid-19: N= 40 vs. After Covid-19: N= 223) [forecast model equation: ln(Number of Records) = 250.543*ln(Year) + -1904.38, p = 0.0005229]. Topics related to diagnostic imaging, quality of life, Covid, psychology, and smartphone increased during the pandemic, while cancer-related topics decreased. The application of topic modeling to the scientific literature on AI and demographics provides a foundation for the next steps regarding developing guidelines for the ethical use of AI for African American dementia caregivers.

Analyzing Twitter-Based Social Networks of Support Communities for Hispanic and African American Family Caregivers of Persons with Dementia.

We applied social network analysis to compare Hispanic and Black dementia caregiving networks on Twitter that were established as part of a clinical trial from January 12, 2022, to October 31, 2022. We extracted Twitter data from our caregiver support communities (N=1980 followers, 811 enrollees) via the Twitter API and used social network analysis software to compare friend/follower interactions within each Hispanic and Black caregiving network. Analysis of the social networks revealed that enrolled family caregivers without prior social media competency had overall low connectedness compared to both enrolled and non-enrolled caregivers with social media competency, who were more integrated into the communities that developed through the clinical trial, partly due to their ties to external dementia caregiving groups. These observed dynamics will help to guide further social media-based interventions and also support the observation that our recruitment strategies effectively enrolled family caregivers with various levels of social media competency.

Comparing Emotional Valence Scores of Twitter Messages from Human Coding and Machine Learning Algorithms Among Hispanic and African American Family Caregivers of Persons with Dementia.

We compared emotional valence scores as determined via machine learning approaches to human-coded scores of direct messages on Twitter from our 2,301 followers during a Twitter-based clinical trial screening for Hispanic and African American family caregivers of persons with dementia. We manually assigned emotional valence scores to 249 randomly selected direct Twitter messages from our followers (N=2,301), then we applied three machine learning sentiment analysis algorithms to extract emotional valence scores for each message and compared their mean scores to the human coding results. The aggregated mean emotional scores from the natural language processing were slightly positive, while the mean score from human coding as a gold standard was negative. Clusters of strongly negative sentiments were observed in followers' responses to being found non-eligible for the study, indicating a significant need for alternative strategies to provide similar research opportunities to non-eligible family caregivers.

Using Artificial Intelligence to Develop a Lexicon-Based African American Tweet Detection Algorithm to Inform Culturally Sensitive Twitter-Based Social Support Interventions for African American Dementia Caregivers.

We extracted 3,291,101 Tweets using hashtags associated with African American-related discourse (#BlackTwitter, #BlackLivesMatter, #StayWoke) and 1,382,441 Tweets from a control set (general or no hashtags) from September 1, 2019 to December 31, 2019 using the Twitter API. We also extracted a literary historical corpus of 14,692 poems and prose writings by African American authors and 66,083 items authored by others as a control, including poems, plays, short stories, novels and essays, using a cloud-based machine learning platform (Amazon SageMaker) via ProQuest TDM Studio. Lastly, we combined statistics from log likelihood and Fisher's exact tests as well as feature analysis of a batch-trained Naive Bayes classifier to select lexicons of terms most strongly associated with the target or control texts. The resulting Tweet-derived African American lexicon contains 1,734 unigrams, while the control contains 2,266 unigrams. This initial version of a lexicon-based African American Tweet detection algorithm developed using Tweet texts will be useful to inform culturally sensitive Twitter-based social support interventions for African American dementia caregivers.

Comparing Themes Extracted via Topic Modeling and Manual Content Analysis: Korean-Language Discussions of Dementia on Twitter.

We randomly examined Korean-language Tweets mentioning dementia/Alzheimer's disease (n= 12,413) posted from November 28 to December 9, 2020, without limiting geographical locations. We independently applied Latent Dirichlet Allocation (LDA) topic modeling and qualitative content analysis to the texts of the Tweets. We compared the themes extracted by LDA topic modeling to those identified via manual coding methods. A total of 16 themes were detected from manual coding, with inter-rater reliability (Cohen's kappa) of 0.842. The proportions of the most prominent themes were: burdens of family caregiving (48.50%), reports of wandering/missing family members with dementia (18.12%), stigma (13.64%), prevention strategies (5.07%), risk factors (4.91%), healthcare policy (3.26%), and elder abuse/safety issues (1.75%). Seven themes whose contents were similar to themes derived from manual coding were extracted from the LDA topic modeling results (perplexity: -6.39, coherence score: 0.45). Our findings suggest that applying LDA topic modeling can be fairly effective at extracting themes from Korean Twitter discussions, in a manner analogous to qualitative coding, to gain insights regarding caregiving for family members with dementia, and our approach can be applied to other languages.

Comparing Emotional Valence Scores of Twitter Posts from Manual Coding and Machine Learning Algorithms to Gain Insights to Refine Interventions for Family Caregivers of Persons with Dementia.

We randomly extracted Korean-language Tweets mentioning dementia/Alzheimer's disease (n= 12,413) from November 28 to December 9, 2020. We independently applied three machine learning algorithms (Afinn, Syuzhet, and Bing) using natural language processing (NLP) techniques and qualitative manual scoring to assign emotional valence scores to Tweets. We then compared the means and distributions of the four emotional valence scores. Visual examination of the graphs produced indicated that each method exhibited unique patterns. The aggregated mean emotional valence scores from the NLP methods were mostly neutral, vs. slightly negative for manual coding (Afinn 0.029, 95% CI [-0.019, 0.077]; Syuzhet 0.266, [0.236, 0.295]; Bing -0.271, [-0.289, -0.252]; manual coding -1.601, [-1.632, -1.569]). One-way analysis of variance (ANOVA) showed no statistically significant differences among the four means after normalization. These findings suggest that the application of NLP can be fairly effective in extracting emotional valence scores from Korean-language Twitter content to gain insights regarding family caregiving for a person with dementia.

Application of Natural Language Processing to Learn Insights on the Clinician's Lived Experience of Electronic Health Records.

We interviewed six clinicians to learn about their lived experience using electronic health records (EHR, Allscripts users) using a semi-structured interview guide in an academic medical center in New York City from October to November 2016. Each participant interview lasted approximately one to two hours. We applied a clustering algorithm to the interview transcript to detect topics, applying natural language processing (NLP). We visualized eight themes using network diagrams (Louvain modularity 0.70). Novel findings include the need for a concise and organized display and data entry page, the user controlling functions for orders, medications, radiology reports, and missing signals of indentation or filtering functions in the order page and lab results. Application of topic modeling to qualitative interview data provides far-reaching research insights into the clinicians' lived experience of EHR and future optimal EHR design to address human-computer interaction issues in an acute care setting.

Applying Participatory Design Mixed-Methods Approaches to Refine Twitter-Based Social Support Interventions for African Americans and Hispanic Family Caregivers of Persons with Dementia.

We applied mixed-methods to refine our first version of the Twitter message library (English 400, translated into Spanish 400) for African Americans and Hispanic family caregivers for a person with dementia. We conducted a series of expert panels to collect quantitative and qualitative data using surveys and in-depth interviews. Using mixed methods to ensure unbiased results, the panelists first independently scored them (1 message/5 panelist) on a scale of 1 to 4 (1: lowest, 4: highest), followed by in-depth interviews and group discussions. Survey results showed that the average score was 3.47, indicating good to excellent (SD 0.35, ranges from 1.8 to 4). Quantitative surveys and qualitative interviews showed different results in emotional support messages.

Analyzing Topics and Sentiments from Twitter to Gain Insights to Refine Interventions for Family Caregivers of Persons with Alzheimer's Disease and Related Dementias (ADRD) During COVID-19 Pandemic.

We randomly extracted Tweets mentioning dementia/Alzheimer's caregiving-related terms (n= 58,094) from Aug 23, 2019, to Sep 14, 2020, via an API. We applied a clustering algorithm and natural language processing (NLP) to publicly available English Tweets to detect topics and sentiment. We compared emotional valence scores of Tweets from before (through the end of 2019) and after the beginning of the COVID-19 pandemic (2020-). Prevalence of topics related to caregiver emotional distress (e.g., depression, helplessness, stigma, loneliness, elder abuse) and caregiver coping (e.g., resilience, love, reading books) increased, and topics related to late-stage dementia caregiving (e.g., nursing home placement, hospice, palliative care) decreased during the pandemic. The mean emotional valence score significantly decreased from 1.18 (SD 1.57; range -7.1 to 7.9) to 0.86 (SD 1.57; range -5.5 to 6.85) after the advent of COVID-19 (difference -0.32 CI: -0.35, -0.29). The application of topic modeling and sentiment analysis to streaming social media provides a foundation for research insights regarding mental health needs for family caregivers of a person with ADRD during COVID-19 pandemic.

Applying Social Network Analysis to Compare Dementia Caregiving Networks on Twitter in Hispanic and Black Communities.

We applied social network analysis (SNA) on Tweets to compare Hispanic and Black dementia caregiving networks. We randomly extracted Tweets mentioning dementia caregiving and related terms from corpora collected daily via the Twitter API from September 1 to December 31, 2019 (initial corpus: n = 2,742,539 Tweets, random sample n = 549,380 English Tweets, n= 185,684 Spanish Tweets). After removing bot-generated Tweets, we first applied a lexicon-based demographic inference algorithm to automatically identify Tweets likely authored by Black and Hispanic individuals using Python (n = 114,511 English, n = 1,185 Spanish). Then, using ORA, we computed network measures at macro, meso, and micro levels and applied the Louvain clustering algorithm to detect groups within each Hispanic and Black caregiving network. Both networks contained a similar proportion of dyads and triads (Hispanic 88.2%, Black 88.9%), while the Black caregiving network included a slightly larger proportion of isolates (Hispanic 0.8%, Black 4.0%). This study provides useful baseline information on the composition of existing large groups and small groups. In addition, this work provides useful guidance for future recruitment strategies and the design of social support interventions regarding emotional needs for Hispanic and Black dementia caregivers.

Application of Social Network Analysis of COVID-19 Related Tweets Mentioning Cannabis and Opioids to Gain Insights for Drug Abuse Research.

We applied social network analysis (SNA) to Tweets mentioning cannabis or opioid-related terms to publicly available COVID-19 related Tweets collected from Jan 21st to May 3rd, 2020 (n= 2,558,474 Tweets). We randomly extracted 16,154 Tweets mentioning cannabis and 4,670 Tweets mentioning opioids from the COVID-19 Tweet corpora for our analysis. The cannabis related Tweets created by 6,144 users were disseminated to 280,042,783 users and retweeted 11 times the number of original messages while opioid-related Tweets created by 3,412 users were disseminated to smaller number of users. The opioids Twitter network showed more cohesive online group activities and a cleaner online environment with less disinformation. The cannabis Twitter network showed a less desirable online environment with more disinformation (false information to mislead the public) and stakeholders lacking strong science knowledge. Application of SNA to Tweets provides insights for future online-based drug abuse research during the outbreak.

Application of Topic Modeling to Tweets as the Foundation for Health Disparity Research for COVID-19.

We randomly extracted publicly available Tweets mentioning COVID-19 related terms (n=2,558,474 Tweets) from Tweet corpora collected daily using an API from Jan 21st to May 3rd, 2020. We applied a clustering algorithm to publicly available Tweets authored by African Americans (n=1,763) to detect topics and sentiment applying natural language processing (NLP). We visualized fifteen topics (four themes) using network diagrams (Newman modularity 0.74). Compared to the COVID-19 related Tweets authored by others, positive sentiments, cohesively encouraging online discussions (e.g., Black strong 27.1%, growing up Blacks 22.8%, support Black business 17.0%, how to build resilience 7.8%), and COVID-19 prevention behaviors (e.g., masks 4.7%, encouraging social distancing 9.4%) were uniquely observed in African American Twitter communities. Application of topic modeling techniques to streaming social media Twitter provides the foundation for research team insights regarding information and future virtual based intervention and social media based health disparity research for COVID-19.

Applying Artificial Intelligence to Predict Self-Reported Poor Health Among Black and Hispanic Caregivers with Mild Cognitive Impairment.

We applied artificial intelligence techniques to build correlate models that predict general poor health in a national sample of caregivers with mild cognitive impairment (MCI). Our application of deep learning identified age, duration of caregiving, amount of alcohol intake, weight, myocardial infarction (MI) and frequency of MCI symptoms for Blacks and Hispanics whereas frequency of MCI symptoms, income, weight, coronary heart disease (CHD), age, and use of e-cigarette for the others as the strongest correlates of poor health among 81 variables entered. The application of artificial intelligence efficiently provided intervention strategies for Black and Hispanic caregivers with MCI.

Trends in Poor Health Indicators Among Black and Hispanic Middle-aged and Older Adults in the United States, 1999-2018.

Importance: Adults who belong to racial/ethnic minority groups are more likely than White adults to receive a diagnosis of chronic disease in the United States. Objective: To evaluate which health indicators have improved or become worse among Black and Hispanic middle-aged and older adults since the Minority Health and Health Disparities Research and Education Act of 2000. Design, Setting, and Participants: In this repeated cross-sectional study, a total of 4 856 326 records were extracted from the Behavioral Risk Factor Surveillance System from January 1999 through December 2018 of persons who self-identified as Black (non-Hispanic), Hispanic (non-White), or White and who were 45 years or older. Exposure: The 1999 legislation to reduce racial/ethnic health disparities. Main Outcomes and Measures: Poor health indicators and disparities including major chronic diseases, physical inactivity, uninsured status, and overall poor health. Results: Among the 4 856 326 participants (2 958 041 [60.9%] women; mean [SD] age, 60.4 [11.8] years), Black adults showed an overall decrease indicating improvement in uninsured status (ß = -0.40%; P < .001) and physical inactivity (ß = -0.29%; P < .001), while they showed an overall increase indicating deterioration in hypertension (ß = 0.88%; P < .001), diabetes (ß = 0.52%; P < .001), asthma (ß = 0.25%; P < .001), and stroke (ß = 0.15%; P < .001) during the last 20 years. The Black-White gap (ie, the change in ß between groups) showed improvement (2 trend lines converging) in uninsured status (-0.20%; P < .001) and physical inactivity (-0.29%; P < .001), while the Black-White gap worsened (2 trend lines diverging) in diabetes (0.14%; P < .001), hypertension (0.15%; P < .001), coronary heart disease (0.07%; P < .001), stroke (0.07%; P < .001), and asthma (0.11%; P < .001). Hispanic adults showed improvement in physical inactivity (ß = -0.28%; P = .02) and perceived poor health (ß = -0.22%; P = .001), while they showed overall deterioration in hypertension (ß = 0.79%; P < .001) and diabetes (ß = 0.50%; P < .001). The Hispanic-White gap showed improvement in coronary heart disease (-0.15%; P < .001), stroke (-0.04%; P < .001), kidney disease (-0.06%; P < .001), asthma (-0.06%; P = .02), arthritis (-0.26%; P < .001), depression (-0.23%; P < .001), and physical inactivity (-0.10%; P = .001), while the Hispanic-White gap worsened in diabetes (0.15%; P < .001), hypertension (0.05%; P = .03), and uninsured status (0.09%; P < .001). Conclusions and Relevance: This study suggests that Black-White disparities increased in diabetes, hypertension, and asthma, while Hispanic-White disparities remained in diabetes, hypertension, and uninsured status.

How Twitter Can Support the HIV/AIDS Response to Achieve the 2030 Eradication Goal: In-Depth Thematic Analysis of World AIDS Day Tweets.

BACKGROUND: HIV/AIDS is a tremendous public health crisis, with a call for its eradication by 2030. A human rights response through civil society engagement is critical to support and sustain HIV eradication efforts. However, ongoing civil engagement is a challenge. OBJECTIVE: This study aimed to demonstrate the use of Twitter data to assess public sentiment in support of civil society engagement. METHODS: Tweets were collected during World AIDS Days 2014 and 2015. A total of 39,940 unique tweets (>10 billion users) in 2014 and 78,215 unique tweets (>33 billion users) in 2015 were analyzed. Response frequencies were aggregated using natural language processing. Hierarchical rank-2 nonnegative matrix factorization algorithm generated a hierarchy of tweets into binary trees. Tweet hierarchy clusters were thematically organized by the Joint United Nations Programme on HIV/AIDS core action principles and categorized under HIV/AIDS Prevention, Treatment or Care, or Support. RESULTS: Topics tweeted 35 times or more were visualized. Results show a decrease in 2015 in the frequency of tweets associated with the fight to end HIV/AIDS, the recognition of women, and to achieve an AIDS-free generation. Moreover, an increase in tweets was associated with an integrative approach to the HIV/AIDS response. Hierarchical thematic differences in 2015 included no prevention discussion and the recognition of the pandemic's impact and discrimination. In addition, a decrease was observed in motivation to fast track the pandemic's end and combat HIV/AIDS. CONCLUSIONS: The human rights-based response to HIV/AIDS eradication is critical. Findings demonstrate the usefulness of Twitter as a low-cost method to assess public sentiment for enhanced knowledge, increased hope, and revitalized expectations for HIV/AIDS eradication.

Semi-supervised morphosyntactic classification of Old Icelandic.

We present IceMorph, a semi-supervised morphosyntactic analyzer of Old Icelandic. In addition to machine-read corpora and dictionaries, it applies a small set of declension prototypes to map corpus words to dictionary entries. A web-based GUI allows expert users to modify and augment data through an online process. A machine learning module incorporates prototype data, edit-distance metrics, and expert feedback to continuously update part-of-speech and morphosyntactic classification. An advantage of the analyzer is its ability to achieve competitive classification accuracy with minimum training data.