Heart Failure Education of African American and Hispanic/Latino Patients: Data Collection and Analysis

Itika Gupta, Barbara Di Eugenio, Devika Salunke, Andrew Boyd, Paula Allen-Meares, Carolyn Dickens, Olga GarciaACL 2020

TL;DR

This paper tackles the critical issue of health disparities in heart failure (HF) self-care among African American (AA) and Hispanic/Latino (H/L) patients. The authors are building a culturally sensitive dialogue agent to improve patient education and adherence, and this paper details their initial steps: collecting and analyzing real-world patient-educator interactions.

What Problem Does This Paper Address?

Heart failure (HF) is a serious global health issue, and effective self-care is crucial for patients to manage their condition. However, minority communities, particularly African Americans and Hispanic/Latinos in the US, experience higher mortality rates from HF. This is often attributed to a complex interplay of factors including genetic variations, access to quality healthcare, socioeconomic conditions, and health behaviors. A significant contributing factor is the lack of culturally sensitive and linguistically appropriate educational materials. Most existing self-care resources are designed for a White, educated population, failing to account for the unique cultural nuances and communication styles of minority groups. This gap leads to poorer self-care adherence and ultimately, worse health outcomes. The authors aim to bridge this gap by developing an interactive dialogue agent that can provide medically reliable and culturally sensitive HF self-care information.

Methods

The core of this paper lies in the data collection and analysis of real-world interactions. The authors recruited three health educators to conduct HF education sessions with AA and H/L patients. They recorded and transcribed 20 of these sessions, with a focus on capturing the nuances of patient-educator dialogue.

For analysis, they employed two main approaches:

  • Thematic Analysis: This involved identifying the most common topics discussed during the HF education sessions. They used Latent Dirichlet Allocation (LDA) for topic modeling, specifically focusing on the educator's turns due to the high proportion of filler sentences in patient turns. They also used Empath, a text analysis tool, to identify topical and emotional categories and highlight differences between patient and educator discourse.
  • Initiative Analysis: This method focused on understanding who controlled the conversation. They defined "initiative" as contributing to the conversation beyond answering questions or using fillers (e.g., by asking questions). They classified turns into questions, prompts, or assertions/commands and analyzed how control of the conversation transferred between the educator and the patient.

Datasets

The paper introduces a newly collected dataset of 20 transcribed heart failure education conversations between health educators and African American (18 sessions) and Hispanic/Latino (2 sessions) patients. This dataset is unique as it specifically targets minority populations and aims to capture culturally relevant dialogue, which is not readily available in existing public datasets.

Key Insights and Ideas

  • Cultural Nuances Matter: The authors emphasize that existing HF self-care materials often lack cultural relevance, leading to poorer outcomes for minority communities. Their work directly addresses this by aiming to build a culturally sensitive agent.
  • Educators Dominate Conversations: The analysis revealed that educators spoke significantly more words and turns than patients, indicating they held most of the conversational initiative. This highlights a potential imbalance in the current educational approach.
  • Patient-Initiated Turns are Crucial: Despite educators' dominance, the authors hypothesize that the turns where patients take control are vital for developing an effective dialogue agent. They envision the agent as a tool that patients will interact with on their own initiative, rather than a passive recipient of information.
  • Food is a Central Topic: Both thematic and word cloud analyses consistently pointed to "food" and "salt" as the most common topics of discussion, underscoring the importance of dietary advice in HF management.
  • Linguistic Practices are Key: The authors recognized that standard transcription services often smoothed over vernacular speech, losing important linguistic practices. They revised transcripts to ensure accurate representation of patient language, recognizing its significance for cultural sensitivity.

Key Results

  • Data Collection: 20 HF education sessions were recorded and transcribed, with 18 involving African American patients and 2 involving Hispanic/Latino patients.
  • Conversational Dynamics: Educators spoke approximately 2.5 times more words than patients. Educators also asked significantly more questions (26 per conversation on average) compared to patients (3 per conversation).
  • Initiative Control: Educators held the initiative for about 95 turns per conversation, while patients held it for 51 turns. Educator control lasted for about 5 turns on average, compared to 2 turns for patients.
  • Topical Analysis: "Food" was identified as the most common topic of discussion. Other significant patient topics included "family" and "children."

Limitations and Future Work

  • Limited Hispanic/Latino Representation: The dataset has a significant imbalance, with only 2 H/L patient sessions compared to 18 AA sessions. This is attributed to challenges in identifying H/L patients and the exclusion of Spanish-language interactions.
  • Focus on English: The current analysis and data collection primarily focus on English interactions, excluding H/L patients who communicate in Spanish.
  • Future Work:
    • Extracting topics from patient-controlled turns to inform the dialogue agent's capabilities.
  • Conducting focus groups with 10 self-identified AA and 10 H/L patients to gain deeper insights into their lives post-HF diagnosis and evaluate the acceptability of a dialogue agent.
  • Using insights from these interactions to inform the questions for the focus groups.
  • Manually identifying culturally relevant topics to model the dialogue agent accordingly.

Summary updated

BibTeX
@inproceedings{gupta2020heart,
  author = {Itika Gupta and Barbara Di Eugenio and Devika Salunke and Andrew Boyd and Paula Allen-Meares and Carolyn Dickens and Olga Garcia},
  title = {{Heart Failure Education of African American and Hispanic/Latino Patients: Data Collection and Analysis}},
  booktitle = {Annual Meeting of the Association for Computational Linguistics},
  year = {2020},
}
Figure