Insights into Auditory Verbal Hallucinations Through Mobile Technology

Hallucinations are experiences where a person perceives something that isn't actually there. One type of these is auditory hallucinations, where individuals might hear sounds or voices without any external source. Auditory Verbal Hallucination (AVH) specifically refers to the experience of hearing voices when no one is speaking. This phenomenon is commonly observed among people diagnosed with mental health issues, such as bipolar disorder and schizophrenia.

#Importance of Valence in Auditory Verbal Hallucinations

To understand AVH better, we look at the nature of the voices-are they negative or positive? This aspect is known as "valence." Assessing how people feel about the voices they hear can give insights into the severity of their mental health conditions. In this study, we worked with 435 individuals who reported experiencing AVH. These participants shared information about their experiences through a mobile application, rating how they felt about the voices they heard several times a day for a month.

#How We Collected Data

Participants received prompts on their phones four times daily, asking them to indicate the nature of the voices they heard. They could choose from options that ranged from "not at all" to "extremely." This method, known as Ecological Momentary Assessment (EMA), allowed us to gather real-time data. Along with these self-reports, participants recorded audio diaries where they verbally expressed what they were experiencing.

We also collected additional data through Mobile Sensing, which included information about participants' surroundings and behaviors without needing them to actively engage. By using various features collected through their mobile devices, we could analyze the relationship between these factors and the occurrence of AVH.

#Our Approach to Prediction

We wanted to understand how well the information gathered from audio diaries and mobile sensing could predict the occurrence and nature of auditory verbal hallucinations. Using advanced techniques in machine learning and data fusion, we built a model that could predict the valence of the voices based on the collected data. Our neural model achieved a top-1 score of 54% and a top-2 score of 72% in predicting the nature of the hallucinated voices.

#The Prevalence of Auditory Verbal Hallucinations

Auditory verbal hallucinations are not uncommon. Research shows that between 5% to 28% of the general population experiences AVH at some point. While these experiences are often associated with severe mental illnesses like schizophrenia, many people who hear voices may not fit the criteria for a psychotic disorder.

It’s critical to distinguish between individuals who may need clinical care and those who do not. Some individuals can manage their experiences without professional help, while others may need more intensive support. This variability in experiences makes it crucial to study AVH further.

#Mobile Sensing as a Tool

Mobile technology has become a key resource for monitoring mental health. These devices can passively gather data about individuals' surroundings and behaviors, enabling researchers to identify early signs of mental health issues like anxiety or depression. The variety of sensors in smartphones allows for a broad spectrum of Data Collection, from tracking location and revealing patterns to measuring physiological responses such as heart rate.

We also utilized mobile applications to facilitate the EMA process, prompting participants to respond to questions about their current experiences. The real-time nature of this approach helps capture more accurate data compared to traditional retrospective methods.

#Language and Mental Health

Language use has been an area of great interest in relation to mental health. Researchers have found that subtle aspects of everyday language can indicate underlying mental health issues. Certain linguistic markers may predict the likelihood of psychosis, and auditory hallucinations are among the primary symptoms of psychosis.

Recent advancements in natural language processing have opened up new possibilities for analyzing speech patterns and detecting the early signs of auditory hallucination. By examining how individuals talk about their experiences, we can gain valuable insights into their mental health.

#Our Research Objectives

Our primary goal was to assess the valence of auditory verbal hallucinations. We sought to understand whether the voices were primarily negative, which may indicate a higher risk for severe psychotic episodes. Our research aims to illustrate how mobile technology can be used effectively in assessing mental health, especially related to AVH.

#Study Methodology

We worked with individuals who reported hearing voices and collected data through EMAs and audio diaries. Participants received instructions on how to use the app and what data would be collected over the study. The data included both self-reports and passive sensing information, enabling us to examine how different variables might relate to AVH experiences.

#Data Collection Process

Participants accessed the study through an online advertisement. They underwent a screening process to ensure they met specific criteria related to experiencing auditory verbal hallucinations. Those eligible were guided through the installation of our mobile application, where they would provide their consent and be informed about the data being collected.

#Participant Demographics

In total, 435 individuals participated in our study. The majority of participants were female, and a significant portion fell between the ages of 25 and 50. We asked participants to report any mental health diagnoses to better understand the population we were studying.

#Ecological Momentary Assessment (EMA)

The EMA process was central to our study. Participants received prompts on their phones, asking if they were currently hearing voices. If they answered "yes," follow-up questions would assess the valence, loudness, control over the voices, and their perceived power. The data collected through these questions formed the backbone of our analysis.

#Active vs. Passive Data

Our study utilized both active and passive data collection methods. Active data came from audio diaries participants recorded when prompted by the EMA. Participants could describe their AVH experiences in detail. Passive data, on the other hand, included sensor information collected in the background while participants used their mobile devices.

#Audio Diaries and Their Role

When participants reported an AVH experience, they had the option to record an audio diary explaining what they heard. These recordings were transcribed and analyzed for deeper insights into the nature of the voices. By combining audio diary data with EMA responses, we aimed to establish a clearer picture of each participant's experiences.

#Passive Sensing Data

Our mobile application also gathered passive sensing data while participants used their devices. This data captured contextual information, such as location, phone usage, and conversations happening around them. The sensing data served as an additional layer of information to correlate with the self-reported experiences of AVH.

#Data Analysis and Model Training

After collecting data, we moved to the analysis phase, where we trained various models using the collected feature sets. Our approach involved generating auditory features from audio diaries, extracting textual features using speech-to-text technology, and transforming passive sensing data to make it suitable for machine learning.

We applied advanced techniques like transfer learning to enhance our model's performance. By utilizing existing pre-trained models, we were able to improve our neural net's capabilities in predicting the valence of auditory verbal hallucinations.

#Research Results

Our results revealed that a hybrid model that combined features from audio diaries, text transcripts, and passive sensing data performed best in predicting AVH events. The model achieved a 54% score for top-1 predictions and 72% for top-2 predictions, indicating the effectiveness of the integrated approach.

#Implications of the Study

Our findings suggest that mobile applications could be instrumental for healthcare professionals in monitoring patients experiencing AVH. Such tools could enable continuous assessments of auditory hallucination events, giving insight into their severity and occurrence over time. By providing a platform for regular updates, both patients and professionals can respond promptly to changes in mental health.

#Addressing Privacy Concerns

With the rise of mobile data collection comes the responsibility to prioritize participants' privacy and data security. One potential solution is to limit access to raw data, ensuring that sensitive information remains on participants' devices. By conducting necessary analyses locally through on-device models, we can minimize the risk of data being misused or improperly shared.

#Related Studies

Our research aligns with other studies emphasizing the benefits of mobile technologies in monitoring mental health. Various studies have demonstrated the value of passive data in recognizing signs of anxiety, depression, and other mental health conditions. Using EMA-based approaches, researchers have successfully captured real-time data, leading to more accurate assessments of mental states.

#Conclusion

In summary, our study showcases the potential of mobile technology in understanding and assessing auditory verbal hallucinations. By integrating self-reported data with passive sensing, we can create a comprehensive view of individuals experiencing AVH. We hope that this research will pave the way for more effective monitoring and intervention strategies in mental health care, ultimately improving the quality of life for individuals facing these challenges.