PWPhilip Wang

Philip Wang

*This is an algorithmically generated author record based on publicly available information.

Current Organizations / Institutes

The organizations/institutes from the scholar's latest article


Philip Wang has a total of 61 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Open Access

    Large Language Models Generate Stigmatizing Language During Reasoning Over Real-World Clinical Data

    by Yutong Yang, Bowen Gu, David B. Hathaway, Richard Wyss, Laura Marengo, Jason B. Gibbons, Stanley Lyndon, Jiageng Wu, Qingyu Chen, Nan Liu, et al.

    Stigmatizing language in clinical documentation, which conveys negative stereotypes, attitudes, or judgments toward patients, is a recognized source of documentation bias and is associated with poorer care and adverse health outcomes. Although prior stigma-related research has focused on clinician-written EHR notes, the increasing use of large language model (LLM)-generated documentation in clinical workflows raises new concerns about its potential to reproduce or amplify bias and affect patient safety. In this study, we conducted a large-scale assessment of stigmatizing language in LLM-generated reasoning text on 35 real-world clinical tasks across 107 LLMs. We applied a psychiatrist-validated, natural language processing (NLP) system to detect stigma terms in LLM reasoning text and quantified stigma rates of LLM-generated reasoning texts across 3,745 model-task pairs. Results showed that stigma rates ranged from 0% to 33.33%, with 84.06% of pairs containing stigma terms. Open-source models and reasoning models showed statistically higher stigma rates than proprietary (1.97% vs. 1.60%; p < 0.01) and non-reasoning models (2.35% vs. 1.70%; p < 0.0001), while the stigma rate difference between the general and medical models is not statistically significant (2.00% vs. 1.80%; p = 0.26). Stigma rates of LLM outputs correlated negatively with task accuracy (r = -0.304; p < 0.001) and positively with input clinical-text stigma (r = 0.569; p < 0.001), with 19.76% of model-task pairs amplifying stigma in the original input notes. Applying prompt engineering as a destigmatizing approach helped reduce model stigma rates by as much as 91.91% without affecting the model performance. This study shows that stigmatizing language generation is common but reducible during LLMs’ reasoning traces, suggesting that well-implemented approaches for LLM monitoring and destigmatizing will be essential for healthcare systems to implement.


  • Open Access

    A Randomized Clinical Trial of the Acceptability and Preliminary Effectiveness of an Online CBT Platform in Diverse Youth

    by Margaret Danielle Weiss, Rajendra Aldis, Danta D. R. Bien-Aime, Taylor Witkowski, Peyton Williams, Katie E. Holmes, Dharma E. Cortés, Philip S. Wang, Benjamin Lê Cook, Eleanor Castine Richards

    Background: The objective of this study was to evaluate the preliminary effectiveness of a self-administered digital mental health intervention, COPE2Thrive, in improving symptoms and functional impairment in youth in a diverse community setting. Methods: Outcome measures included self-reported Kiddie-Computerized Adaptive Testing (K-CAT) scores on seven diagnostic domains, and self-reported scores of six domains and the total score of the Weiss Functional Impairment Rating Scale (WFIRS)—Self Report. All youth were eligible to participate unless they were currently in another therapy. Primary analyses followed the stepped-wedge randomized design, comparing participants assigned to one of three randomized intervention sequences. Secondary analyses treated the study as an open-label intervention study and evaluated pre- to post-treatment change among participants who engaged in the intervention, using the last observation carried forward for non-completers. Results: We successfully recruited a majority non-White sample that was 31.5% Black, 31.5% Hispanic/Latino, and 19.6% Asian. Of the eligible participants, 30.5% enrolled in the study, 23.4% completed the intervention, and 12.5% participated in the follow-up. Racial and ethnic minority subjects were less likely to have received outside behavioral health therapies, both in the screening sample and among those who participated in the intervention. There was no statistically significant improvement in the total score or any of the domain scores on the K-CAT or the WFIRS-S, either in the RCT or in the open-label analysis. Conclusions: Given the low acceptability of the program and the absence of clinically significant improvement in this pilot study, the findings suggest that the intervention, as implemented, may have limited utility for addressing youth distress in diverse communities.

  • Open Access

    Mental Health Care Provider Experiences of Remote Measurement-Based Care Rollout in an Urban Safety-Net Psychiatry Department: Three-Site Mixed Methods Hypothesis-Generating Implementation Study

    by Ana M Progovac, Carl Fulwiler, Margaret Lanca, Daisy Wang, Kate Zona, Norah Mulvaney-Day, Rajendra Aldis, H Stephen Leff, Philip Wang, Lisa C Rosenfeld

    Background: Measurement-based care (MBC), including remote MBC, is increasingly being considered or implemented for mental health treatment and outcomes monitoring in routine clinical care. However, little is known about the health equity implications in real-world practice or the impact on patient-provider relationships in lower-resource systems that offer mental health treatment for diverse patients. Objective: This hypothesis-generating study examined the drivers of MBC implementation outcomes, the implications for health equity, and the impact of MBC on therapeutic alliance (TA). The study was conducted 1 year after the implementation of remote MBC at 3 outpatient adult clinics in a diverse, safety-net health system. Methods: This explanatory sequential mixed methods study used quantitative surveys and qualitative focus groups with mental health care providers. Repeated surveys were first used to understand mental health care provider experiences over a 6-month period, at least 1 year after MBC implementation. Surveys were analyzed to refine focus group prompts. Six mental health providers participated in repeated surveys over 6 months, after which the same 6 providers and 1 additional mental health provider took part in focus groups. Results: Surveys revealed stable acceptability and utility ratings, concerns that MBC was not equally benefiting patients, little endorsement that MBC improved TA, and slightly decreasing feasibility scores. In focus groups, mental health care providers shared concerns about the acceptability, appropriateness, feasibility, and equity of processes for collecting MBC data. These providers had less first-hand experience with sharing and acting upon the data but still voiced concerns about the processes for doing so. TA both impacted and was impacted by MBC in positive and negative ways. The potential drivers of the findings are discussed using qualitative data. Conclusions: More than 1 year after the implementation of remote MBC for mental health, mental health care providers had enduring concerns about its implications for health equity as well as its bidirectional relationship with TA. These findings suggest that further study is needed to identify system-level strategies to mitigate potential negative effects of real-world MBC implementations on health equity, particularly in low-resource settings with diverse populations.


  • Open Access

    Mental Health Care Provider Experiences of Remote Measurement-Based Care Rollout in an Urban Safety-Net Psychiatry Department: Three-Site Mixed Methods Hypothesis-Generating Implementation Study (Preprint)

    by Ana M Progovac, Carl Fulwiler, Margaret Lanca, Daisy Wang, Kate Zona, Norah Mulvaney-Day, Rajendra Aldis, H Stephen Leff, Philip Wang, Lisa C Rosenfeld

    BACKGROUND Measurement-based care (MBC), including remote MBC, is increasingly being considered or implemented for mental health treatment and outcomes monitoring in routine clinical care. However, little is known about the health equity implications in real-world practice or the impact on patient-provider relationships in lower-resource systems that offer mental health treatment for diverse patients. OBJECTIVE This hypothesis-generating study examined the drivers of MBC implementation outcomes, the implications for health equity, and the impact of MBC on therapeutic alliance (TA). The study was conducted 1 year after the implementation of remote MBC at 3 outpatient adult clinics in a diverse, safety-net health system. METHODS This explanatory sequential mixed methods study used quantitative surveys and qualitative focus groups with mental health care providers. Repeated surveys were first used to understand mental health care provider experiences over a 6-month period, at least 1 year after MBC implementation. Surveys were analyzed to refine focus group prompts. Six mental health providers participated in repeated surveys over 6 months, after which the same 6 providers and 1 additional mental health provider took part in focus groups. RESULTS Surveys revealed stable acceptability and utility ratings, concerns that MBC was not equally benefiting patients, little endorsement that MBC improved TA, and slightly decreasing feasibility scores. In focus groups, mental health care providers shared concerns about the acceptability, appropriateness, feasibility, and equity of processes for collecting MBC data. These providers had less first-hand experience with sharing and acting upon the data but still voiced concerns about the processes for doing so. TA both impacted and was impacted by MBC in positive and negative ways. The potential drivers of the findings are discussed using qualitative data. CONCLUSIONS More than 1 year after the implementation of remote MBC for mental health, mental health care providers had enduring concerns about its implications for health equity as well as its bidirectional relationship with TA. These findings suggest that further study is needed to identify system-level strategies to mitigate potential negative effects of real-world MBC implementations on health equity, particularly in low-resource settings with diverse populations.

  • Comparisons of mental health treatment frequency and risk of suicidal thoughts and behaviors among youth

    by Gareth J. Parry, Lindsay Overhage, Peyton Williams, Katie Holmes, Akhil Reddy, Gabriel E. Rios Perez, Albert Y.H. Lo, Abigail Thomas, Brian Mullin, Sanam Bhakta, et al.


  • Open Access

    The Development of a Brief but Comprehensive Therapeutic Assessment Protocol for the Screening and Support of Youth in the Community to Address the Youth Mental Health Crisis

    by Margaret Danielle Weiss, Eleanor Castine Richards, Danta Bien-Aime, Taylor Witkowski, Peyton Williams, Katie E. Holmes, Dharma E. Cortes, Miriam C. Tepper, Philip S. Wang, Rajendra Aldis, et al.

    Objective: The objective of this study was to explore the acceptability and feasibility of a therapeutic assessment protocol for the Screening and Support of Youth (SASY). SASY provides brief but comprehensive community-based screening and support for diverse youth in the community. Methods: SASY screening evaluates symptoms, functioning and clinical risk. The Kiddie Computerized Adaptive Test was used to evaluate seven different diagnoses and symptom severity. The Weiss Functional Impairment Rating Scale-Self was used to measure functional impairment. Measures were scored according to nationally developed norms. An algorithm was developed to aggregate symptom and function ratings into an overall score for clinical risk. The results are discussed with participants in a motivational interview designed to promote insight, followed by the opportunity for the participant to engage in an online intervention. Protocol changes necessitated by social distancing during the pandemic led to innovative methods including the use of a QR code for recruitment, integration of both online and offline participation, and expansion from in-person recruitment within the schools to virtual engagement with youth throughout the community. The final sample included disproportionately more Black or African American and Hispanic youth as compared to school and community statistics, suggesting that optimization of online and offline methods in research may facilitate the recruitment of diverse populations. Qualitative interviews indicated that the screening and feedback raised youth awareness of their wellbeing and/or distress, its impact on their functioning, and engagement with options for improved wellbeing. Conclusions: The emergence of innovative methods optimizing the advantages of both online and offline methods, developed as a necessity during the pandemic, proved advantageous to the feasibility and acceptability of community-based recruitment of at-risk, minoritized youth.


  • Open Access

    Acceptability and Feasibility of Hybrid Research Methods for Adaptive Assessment of Youth in Community Settings

    by Margaret Danielle Weiss, Eleanor Castine Richards, Danta Bien-Aime, Taylor Witkowski, Peyton Williams, Katie E Holmes, Dharma E Cortes, Miriam C Tepper, Phillip S Wang, Rajendra Aldis, et al.

    Objective: To explore the acceptability and feasibility of conducting comprehensive, community-based, hy-brid (in-person and virtual) mental health screening and feedback with diverse youth in the community. Methods: The Screening and Support for Youth intervention provides comprehensive assessment and thera-peutic feedback in community settings including symptoms, diagnoses, functional impairment, suicide, and clinical risk. Symptom severity, seven diagnoses and suicidality were assessed with the Kiddie Computer-ized Adaptive Test. Functional impairment was evaluated by youth self-report on the Weiss Functional Im-pairment Rating Scale. Measures were scored according to nationally developed norms and an algorithm was developed to aggregate symptoms and functioning to determine clinical risk. The results were discussed with participants in a motivational interview to promote insight and engagement in an online intervention. Results: Methodological innovations included using a QR code for recruitment, integrated in-person and virtual participation, and expansion from in-person recruitment within the schools to virtual engagement with youth throughout the community. The final sample included disproportionately more Black or African American and Hispanic youth as compared to school and community statistics, suggesting that hybrid re-search may facilitate recruitment of diverse populations. Qualitative interviews indicated that the screening and feedback raised youth awareness of their wellbeing and/or distress, its impact on their functioning, and engagement with options for improving wellbeing. Conclusion: Hybrid research, developed as a necessity during the pandemic, proved advantageous to the feasibility and acceptability of community-based recruit-ment of at risk, minoritized youth.

  • Open Access

    Association between depression severity, mental health recovery and dropout from behavioral health care treatment

    by Jason B. Gibbons, Shelbi A. Cox, Loreen Straub, Josephine S. Au, Philip S. Wang, Jun Liu, Alyson Albano, Rachel Wood, Matthew Ruble, John Peloquin, et al.


  • Open Access

    Participants’ Engagement With and Results From a Web-Based Integrative Population Mental Wellness Program (CHAMindWell) During the COVID-19 Pandemic: Program Evaluation Study (Preprint)

    by Joseph A Rosansky, Kayley Okst, Miriam C Tepper, Ana Baumgart Schreck, Carl Fulwiler, Philip S Wang, Zev Schuman-Olivier

    BACKGROUND The COVID-19 pandemic involved a prolonged period of collective trauma and stress during which substantial increases in mental health concerns, like depression and anxiety, were observed across the population. In this context, CHAMindWell was developed as a web-based intervention to improve resilience and reduce symptom severity among a public health care system’s patient population. OBJECTIVE This program evaluation was conducted to explore participants’ engagement with and outcomes from CHAMindWell by retrospectively examining demographic information and mental health symptom severity scores throughout program participation. METHODS We examined participants’ symptom severity scores from repeated, web-based symptom screenings through Computerized Adaptive Testing for Mental Health (CAT-MH) surveys, and categorized participants into symptom severity-based tiers (tier 1=asymptomatic to mild; tier 2=moderate; and tier 3=severe). Participants were provided tier-based mindfulness resources, treatment recommendations, and referrals. Logistic regressions were conducted to evaluate associations between demographic variables and survey completion. The McNemar exact test and paired sample t tests were performed to evaluate changes in the numbers of participants in tier 1 versus tier 2 or 3 and changes in depression, anxiety, and posttraumatic stress disorder severity scores between baseline and follow-up. RESULTS The program enrolled 903 participants (664/903, 73.5% female; 556/903, 61.6% White; 113/903, 12.5% Black; 84/903, 9.3% Asian; 7/903, 0.8% Native; 36/903, 4% other; and 227/903, 25.1% Hispanic) between December 16, 2020, and March 17, 2022. Of those, 623 (69%) completed a baseline CAT-MH survey, and 196 completed at least one follow-up survey 3 to 6 months after baseline. White racial identity was associated with completing baseline CAT-MH (odds ratio [OR] 1.80, 95% CI 1.14-2.84; P=.01). Participants’ odds of having symptom severity below the clinical threshold (ie, tier 1) were significantly greater at follow-up (OR 2.60, 95% CI 1.40-5.08; P=.001), and significant reductions were observed across symptom domains over time. CONCLUSIONS CHAMindWell is associated with reduced severity of mental health symptoms. Future work should aim to address program engagement inequities and attrition and compare the impacts of CHAMindWell to a control condition to better characterize its effects.