Skip to main content
eScholarship
Open Access Publications from the University of California

Scholarly Works (8 results)

  • Thesis
  • Peer Reviewed
UC Irvine Electronic Theses and Dissertations (2023)

Health text ranging from patient-generated online forum posts to clinician-authored unstructured notes contain valuable information that can potentially improve healthcare service quality, patient experiences, and patient and population health outcomes. Health text data are also highly heterogeneous, produced in different contexts and serve different purposes, which require careful study design and methodological innovations to ensure study validity. However, the current practices of computational analysis on health text are often inconsistent and lack considerations of the contexts in which health text is produced.

My dissertation includes three major studies that analyzed different types of health text including public-generated social media data and clinical notes of patients with rare diseases. In the first study, I conducted a systematic literature review that revealed multiple issues in the current practices of how computational sentiment analysis is applied on health-related social media data. I also comprehensively evaluated the commonly used sentiment analysis tools on several social media datasets and found that they failed to accurately label the sentiments conveyed in health-related social media data. In the second study, I developed and applied computer-assisted qualitative analysis pipelines to analyze health-related social media data including tweets and online physician reviews. The results identified public attitudes and concerns toward mask wearing during the COVID-19 pandemic and patient concerns around healthcare service quality. These insights contribute to better public health communication strategies and ways of enhancing patients’ experiences when interacting with healthcare systems. In the third study, I switched gears to develop a pipeline that extracts various clinical entities including diagnosis, environmental exposures, substance use, performance status, and staging from unstructured notes of patients with lymphoid malignancies. The pipeline achieved satisfying performance and an error analysis identified issues with current documentation practices of key clinical information and provided recommendations for future improvement of the pipeline. The extracted clinical entities will be further used to facilitate clinical research to understand the association between environmental exposures and cancer outcomes.

Collectively, these studies contribute a set of methodological and empirical insights into how to design and choose an appropriate computational method to analyze different types of health text data. Moving forward, my future work will integrate and adapt the emerging Large Language Models into health text analysis, assess their performances, and identify potential biases when analyzing different types of health texts from various patient populations.

    Cover page: Computational Analysis of Health TextCreative Commons 'BY-SA' version 4.0 license
    • Article
    • Peer Reviewed
    UC Irvine Previously Published Works (2019)
    Sentiment analysis has been increasingly used to analyze online social media data such as tweets and health forum posts. However, previous studies often adopted existing, general-purpose sentiment analyzers developed in non-healthcare domains, without assessing their validity and without customizing them for the specific study context. In this work, we empirically evaluated three general-purpose sentiment analyzers popularly used in previous studies (Stanford Core NLP Sentiment Analysis, TextBlob, and VADER), based on two online health datasets and a general-purpose dataset as the baseline. We illustrate that none of these general-purpose sentiment analyzers were able to produce satisfactory classifications of sentiment polarity. Further, these sentiment analyzers generated inconsistent results when applied to the same dataset, and their performance varies to a great extent across the two health datasets. Significant future work is therefore needed to develop context-specific sentiment analysis tools for analyzing online health data.
      Cover page: How Do General-Purpose Sentiment Analyzers Perform when Applied to Health-Related Online Social Media Data?
      • Article
      • Peer Reviewed
      UC Irvine Previously Published Works (2022)
      While it has been scientifically proven that COVID-19 vaccine is a safe and effective measure to reduce the severity of infection and curbing the spread of the SARS-CoV-2 virus, skepticism remains widespread, and in many countries vaccine mandates have been met with strong opposition. In this study, we applied machine learning-based analyses of the U.S.-based tweets covering the periods leading toward and after the Biden Administration's announcement of federal vaccine mandates, supplemented by a qualitative content analysis of a random sample of relevant tweets. The objective was to examine the beliefs held among twitter users toward vaccine mandates, as well as the evidence that they used to support their positions. The results show that while approximately 30% of the twitter users included in the dataset supported the measure, more users expressed differing opinions. Concerns raised included questioning on the political motive, infringement of personal liberties, and ineffectiveness in preventing infection.
        Cover page: Public Opinions toward COVID-19 Vaccine Mandates: A Machine Learning-based Analysis of U.S. Tweets.
        • Article
        • Peer Reviewed
        UC Irvine Previously Published Works (2021)

        Objective

        Sentiment analysis is a popular tool for analyzing health-related social media content. However, existing studies exhibit numerous methodological issues and inconsistencies with respect to research design and results reporting, which could lead to biased data, imprecise or incorrect conclusions, or incomparable results across studies. This article reports a systematic analysis of the literature with respect to such issues. The objective was to develop a standardized protocol for improving the research validity and comparability of results in future relevant studies.

        Materials and methods

        We developed the Protocol of Analysis of senTiment in Health (PATH) based on a systematic review that analyzed common research design choices and how such choices were made, or reported, among eligible studies published 2010-2019.

        Results

        Of 409 articles screened, 89 met the inclusion criteria. A total of 16 distinctive research design choices were identified, 9 of which have significant methodological or reporting inconsistencies among the articles reviewed, ranging from how relevance of study data was determined to how the sentiment analysis tool selected was validated. Based on this result, we developed the PATH protocol that encompasses all these distinctive design choices and highlights the ones for which careful consideration and detailed reporting are particularly warranted.

        Conclusions

        A substantial degree of methodological and reporting inconsistencies exist in the extant literature that applied sentiment analysis to analyzing health-related social media data. The PATH protocol developed through this research may contribute to mitigating such issues in future relevant studies.
          Cover page: Developing a standardized protocol for computational sentiment analysis research using health-related social media data.Creative Commons 'BY' version 4.0 license
          • Article
          • Peer Reviewed
          UC Irvine Previously Published Works (2020)
          Online physician review (OPR) websites have been increasingly used by healthcare consumers to make informed decisions in selecting healthcare providers. However, consumer-generated online reviews are often unstructured and contain plural topics with varying degrees of granularity, making it challenging to analyze using conventional topic modeling techniques. In this paper, we designed a novel natural language processing pipeline incorporating qualitative coding and supervised and unsupervised machine learning. Using this method, we were able to identify not only coarse-grained topics (e.g., relationship, clinic management), but also fine-grained details such as diagnosis, timing and access, and financial concerns. We discuss how healthcare providers could improve their ratings based on consumer feedback. We also reflect on the inherent challenges of analyzing user-generated online data, and how our novel pipeline may inform future work on mining consumer-generated online data.
            Cover page: What Do Patients Care About? Mining Fine-grained Patient Concerns from Online Physician Reviews Through Computer-Assisted Multi-level Qualitative Analysis.Creative Commons 'BY' version 4.0 license
            • Article
            • Peer Reviewed
            UC Irvine Previously Published Works (2021)

            Objective

            Facial masks are an essential personal protective measure to fight the COVID-19 (coronavirus disease) pandemic. However, the mask adoption rate in the United States is still less than optimal. This study aims to understand the beliefs held by individuals who oppose the use of facial masks, and the evidence that they use to support these beliefs, to inform the development of targeted public health communication strategies.

            Materials and methods

            We analyzed a total of 771 268 U.S.-based tweets between January to October 2020. We developed machine learning classifiers to identify and categorize relevant tweets, followed by a qualitative content analysis of a subset of the tweets to understand the rationale of those opposed mask wearing.

            Results

            We identified 267 152 tweets that contained personal opinions about wearing facial masks to prevent the spread of COVID-19. While the majority of the tweets supported mask wearing, the proportion of anti-mask tweets stayed constant at about a 10% level throughout the study period. Common reasons for opposition included physical discomfort and negative effects, lack of effectiveness, and being unnecessary or inappropriate for certain people or under certain circumstances. The opposing tweets were significantly less likely to cite external sources of information such as public health agencies' websites to support the arguments.

            Conclusions

            Combining machine learning and qualitative content analysis is an effective strategy for identifying public attitudes toward mask wearing and the reasons for opposition. The results may inform better communication strategies to improve the public perception of wearing masks and, in particular, to specifically address common anti-mask beliefs.
              Cover page: Why do people oppose mask wearing? A comprehensive analysis of US tweets during the COVID-19 pandemic
              • Article
              • Peer Reviewed
              UC San Diego Previously Published Works (2020)

              Background

              In Alzheimer's Disease (AD), about one-third of the risk genes identified by GWAS encode proteins that function predominantly in the endocytic pathways. Among them, the Ras and Rab Interactor 3(RIN3) is a guanine nucleotide exchange factor (GEF) for the Rab5 small GTPase family and has been implicated to be a risk factor for both late onset AD (LOAD) and sporadic early onset AD (sEOAD). However, how RIN3 is linked to AD pathogenesis is currently undefined.

              Methods

              Quantitative PCR and immunoblotting were used to measure the RIN3 expression level in mouse brain tissues and cultured basal forebrain cholinergic neuron (BFCNs). Immunostaining was used to define subcellular localization of RIN3 and to visualize endosomal changes in cultured primary BFCNs and PC12 cells. Recombinant flag-tagged RIN3 protein was purified from HEK293T cells and was used to define RIN3-interactomes by mass spectrometry. RIN3-interacting partners were validated by co-immunoprecipitation, immunofluorescence and yeast two hybrid assays. Live imaging of primary neurons was used to examine axonal transport of amyloid precursor protein (APP) and β-secretase 1 (BACE1). Immunoblotting was used to detect protein expression, processing of APP and phosphorylated forms of Tau.

              Results

              We have shown that RIN3 mRNA level was significantly increased in the hippocampus and cortex of APP/PS1 mouse brain. Basal forebrain cholinergic neurons (BFCNs) cultured from E18 APP/PS1 mouse embryos also showed increased RIN3 expression accompanied by early endosome enlargement. In addition, via its proline rich domain, RIN3 recruited BIN1(bridging integrator 1) and CD2AP (CD2 associated protein), two other AD risk factors, to early endosomes. Interestingly, overexpression of RIN3 or CD2AP promoted APP cleavage to increase its carboxyl terminal fragments (CTFs) in PC12 cells. Upregulation of RIN3 or the neuronal isoform of BIN1 increased phosphorylated Tau level. Therefore, upregulation of RIN3 expression promoted accumulation of APP CTFs and increased phosphorylated Tau. These effects by RIN3 was rescued by the expression of a dominant negative Rab5 (Rab5S34N) construct. Our study has thus pointed to that RIN3 acts through Rab5 to impact endosomal trafficking and signaling.

              Conclusion

              RIN3 is significantly upregulated and correlated with endosomal dysfunction in APP/PS1 mouse. Through interacting with BIN1 and CD2AP, increased RIN3 expression alters axonal trafficking and procession of APP. Together with our previous studies, our current work has thus provided important insights into the role of RIN3 in regulating endosomal signaling and trafficking.
                Cover page: Upregulation of RIN3 induces endosomal dysfunction in Alzheimer’s disease
                • Article
                • Peer Reviewed
                UC Irvine Previously Published Works (2023)

                Introduction

                Adverse drug events (ADEs) are associated with poor outcomes and increased costs but may be prevented with prediction tools. With the National Institute of Health All of Us (AoU) database, we employed machine learning (ML) to predict selective serotonin reuptake inhibitor (SSRI)-associated bleeding.

                Methods

                The AoU program, beginning in 05/2018, continues to recruit ≥ 18 years old individuals across the United States. Participants completed surveys and consented to contribute electronic health record (EHR) for research. Using the EHR, we determined participants who were exposed to SSRIs (citalopram, escitalopram, fluoxetine, fluvoxamine, paroxetine, sertraline, vortioxetine). Features (n = 88) were selected with clinicians' input and comprised sociodemographic, lifestyle, comorbidities, and medication use information. We identified bleeding events with validated EHR algorithms and applied logistic regression, decision tree, random forest, and extreme gradient boost to predict bleeding during SSRI exposure. We assessed model performance with area under the receiver operating characteristic curve statistic (AUC) and defined clinically significant features as resulting in > 0.01 decline in AUC after removal from the model, in three of four ML models.

                Results

                There were 10,362 participants exposed to SSRIs, with 9.6% experiencing a bleeding event during SSRI exposure. For each SSRI, performance across all four ML models was relatively consistent. AUCs from the best models ranged 0.632-0.698. Clinically significant features included health literacy for escitalopram, and bleeding history and socioeconomic status for all SSRIs.

                Conclusions

                We demonstrated feasibility of predicting ADEs using ML. Incorporating genomic features and drug interactions with deep learning models may improve ADE prediction.
                  Cover page: Using machine learning to develop a clinical prediction model for SSRI-associated bleeding: a feasibility study.
                  Top