An artificial intelligence-based chatbot for Helicobacter Pylori Health Education: A Potential tool for Community Awareness

doi:10.21203/rs.3.rs-8070077/v1

Download PDF

Research Article

An artificial intelligence-based chatbot for Helicobacter Pylori Health Education: A Potential tool for Community Awareness

https://doi.org/10.21203/rs.3.rs-8070077/v1

This work is licensed under a CC BY 4.0 License

You are reading this latest preprint version

Background

Helicobacter pylori (H. pylori) is a gram-negative bacterium that infects approximately 50% of the global population. It is a major risk factor for serious gastrointestinal diseases, including peptic ulcers and gastric cancer. The prevalence is highest in low- and middle-income countries due to factors such as poor sanitation and limited access to healthcare. However, in these high-risk, underserved communities, public awareness of H. pylori remains critically low. Artificial intelligence (AI)-powered chatbots represent a novel approach to health education by delivering instant, evidence-based, and personalised information. This study examines the feasibility of an AI-driven chatbot to enhance Helicobacter pylori awareness and support early diagnosis and prevention.

Methodology

A cross-sectional study was conducted to evaluate the usability, perceived usefulness, and overall effectiveness of an AI-powered H. pylori chatbot over three months (June-August 2024). The chatbot was developed using the Botpress Cloud platform, incorporating natural language understanding (NLU) and multilingual capabilities, with a knowledge base curated from peer-reviewed literature, CHADF Website Contents and global health guidelines. The study recruited adults aged 18 years and above through convenience sampling, with electronic informed consent obtained before chatbot access. Data collection comprised automated metrics captured via the Botpress analytics dashboard, tracking user engagement, message frequency, and session duration. Additionally, user perception was automatically assessed through sentiment analysis, classifying feedback as positive, neutral, or negative. These key metrics were evaluated to determine the chatbot's effectiveness in enhancing awareness and supporting prevention efforts related to H. pylori.

Results

A total of 102 users engaged with the chatbot across 324 sessions, generating 246 user messages. Queries predominantly focused on symptoms (21%), treatment (19%), dietary recommendations (17%), and transmission (16%). User engagement patterns revealed peak activity during evenings and weekends, indicating accessibility outside traditional healthcare hours. Feedback highlighted the Chatbot’s ease of use, reliability, and role in bridging health literacy gaps.

Conclusion

AI-powered chatbots show promise in enhancing H. pylori health education by providing accessible, evidence-based, and personalised information. This pilot study assessed effectiveness through user engagement, message frequency, and session duration. With 102 users generating 324 sessions and 246 messages over three months without external promotion, the engagement level was satisfactory for a feasibility study. The chatbot supported meaningful interactions, especially outside clinical hours, and addressed key topics like symptoms, treatment, and prevention. These findings highlight its potential to bridge health literacy gaps and contribute to reducing H. pylori-related disease burdens.

H. pylori

AI chatbot

health education

community awareness

gastric infection

Helicobacter pylori (H. pylori) is a gram-negative bacterium that colonizes the human gastric mucosa and is predominantly transmitted via the fecal–oral route. With an estimated global infection rate of 50%, prevalence is significantly higher in developing countries, largely attributable to factors such as poor sanitation and dietary habits [1]. Long-term H. pylori infection is a major etiological agent for chronic gastritis, peptic ulcers, and gastric mucosa-associated lymphoid tissue (MALT) lymphoma. Its role in gastric carcinogenesis led the World Health Organization (WHO) to classify it as a Group I carcinogen [2]. Despite its significant global health burden, public awareness of H. pylori, its associated risks, and prevention strategies remains critically low, particularly in underserved regions [3].

The prevalence of H. pylori is strongly correlated with socioeconomic status, with the highest infection rates observed in low- and middle-income countries where access to clean water and healthcare infrastructure is limited [1, 4]. In East Africa, including Tanzania, the infection burden is substantial. This region faces dual challenges: a healthcare system strained in providing widespread preventive care and diagnostic services, and a significant gap in public health literacy regarding gastrointestinal health [5, 6]. Consequently, opportunities for early detection and intervention are frequently missed, allowing infections to persist and progress to severe clinical outcomes, such as gastric cancer [4].

Traditional health education campaigns in these regions have often been insufficient, struggling with issues of scalability, cost, and reach. There is a clear and urgent need for innovative, accessible tools to disseminate crucial health information. Recently, artificial intelligence (AI) has emerged as a transformative force in digital health. AI-powered chatbots, in particular, have demonstrated considerable potential in delivering scalable, personalized, and evidence-based health education directly to users [7, 8]. These tools can operate 24/7, overcome geographical barriers, and can be tailored to local languages and contexts, making them uniquely suited to address health literacy gaps in remote and resource-limited settings [7].

While AI chatbots have been successfully deployed in other health domains, such as cancer awareness and patient education [8, 9], their application specifically for H. pylori education remains largely unexplored. An AI-driven chatbot dedicated to H. pylori could provide immediate, accurate information on transmission, symptoms, prevention, and the importance of seeking treatment, thereby empowering individuals to make informed health decisions [1, 10].

This study designed and evaluated an AI-driven chatbot for H. pylori health education, with the objectives to: (i) develop a curated, guideline-aligned knowledge base and conversational flows covering symptoms, transmission, testing, treatment, and dietary guidance; (ii) assess real-world use and engagement (sessions, message patterns, topic distribution); and (iii) explore the feasibility of this approach to strengthen community awareness in East Africa, including Tanzania. By addressing persistent informational gaps with a scalable digital tool, this work seeks to inform strategies that support earlier recognition, testing, and treatment of H. pylori in resource-constrained settings.

Study Aim, Design and Setting

The purpose of this study was to design and evaluate an artificial intelligence (AI) chatbot intended to improve public understanding of Helicobacter pylori. The study used a descriptive, development-oriented approach. Work began with building the chatbot and then making it available for public use, after which user interactions were reviewed to understand how people engaged with the tool. The development phase and testing period ran from October 2024 to January 2025. Once the system was stable, it was posted online and could be accessed through both mobile phones and web browsers by users in Tanzania and anyone with access to the link.

Materials and AI Platform

The chatbot was developed on the Botpress platform, an open-source system that provides tools for natural language understanding (NLU), intent recognition, and multilingual communication. These features helped the chatbot interpret user questions in a conversational manner. Botpress was selected mainly because of its flexibility and the straightforward setup process reported in earlier digital health projects [2, 3]. Offering access on both mobile and web interfaces ensured that people with different devices could still reach the tool. No special hardware or clinical equipment was used during the project.

Knowledge Base and Training Data

The information used to train the chatbot was drawn from several sources. Scientific articles describing the epidemiology, diagnosis, and treatment of H. pylori formed the core of the knowledge base [4–7]. In addition, recognised clinical guidelines from regional and international bodies were consulted to ensure accuracy and alignment with recommended practices [8]. To make explanations easier for the general public, health education materials written for non-specialist audiences were reviewed as well. All information was arranged into themes including symptoms, transmission, prevention, available tests, treatment options, diet, and possible complications. This form of organisation, which has been shown in other studies to strengthen the reliability of chatbot responses, helped guide the structure of the conversations [4, 9].

Chatbot Development Procedures

Prompt and Completion Training

The chatbot was trained using a Prompt-and-Completion (P&C) method. In this approach, prompts were created to resemble the types of questions that members of the public typically ask about H. pylori. Each prompt was then paired with an answer crafted from the literature and guidelines mentioned above. This technique has been used in other chatbot projects and tends to produce replies that remain steady and predictable across different conversations [2, 3].

Conversation Flow and Interface Design

The flow of conversations was arranged into clear sections so that users could easily locate the information they needed. Decision tree-style structures were used to guide users through questions related to symptoms, testing procedures, treatment, diet, or preventive measures. When a question fell outside the chatbot’s knowledge area, a fallback response directed the user to trustworthy external sources. The interface itself was intentionally kept plain and easy to navigate. Designs that minimise unnecessary complexity are known to help users with different levels of digital experience engage more comfortably with health tools [2, 10].

Deployment and User Testing

After development, the chatbot was made publicly available. During the testing period, all interactions were recorded automatically through the Botpress analytics dashboard. Data collected included the total number of sessions, the number of messages exchanged between users and the chatbot, the time each interaction occurred, and which topics were accessed most often. These indicators are commonly used when assessing the performance of health-related chatbots and give an idea of how the tool was used in practice [3, 11]. No personal information or demographic details were collected at any point.

Participants

Any person who accessed the chatbot during the testing period was counted as a participant. Because the tool did not require users to log in or identify themselves, all interactions were anonymous. This meant that the study drew on a broad and informal group of people seeking information on H. pylori.

Data Collection and Management

Overall, 324 sessions and 246 user messages were downloaded for analysis once the testing window closed. The exported data contained only the text of the messages and a small amount of metadata such as timestamps and session identifiers. No identifying information was present, and there was no direct communication between the research team and the users.

Data Analysis

Descriptive Analysis

The quantitative information was summarised using basic descriptive techniques. This included total session counts, the number of messages sent by users and the chatbot, the share of returning users compared with new ones, and patterns of use across different days and times. Because the study relied on naturally occurring interaction data, a power calculation was not required.

Thematic Analysis

To understand the content of the questions being asked, a thematic analysis was carried out. Messages were sorted into six categories: symptoms, treatment, transmission and prevention, diet, diagnostic tests, and complications or risk factors. Two reviewers completed the coding independently and then compared and resolved any differences. This process follows established approaches used in qualitative digital health research [12].

Evaluation Metrics

The chatbot was evaluated using three main considerations. First, the accuracy of its responses was examined by comparing the information it provided with established clinical guidelines to ensure that the advice aligned with accepted standards of care [7, 8]. The second aspect concerned accessibility, which was assessed by looking at when and how often people used the system, as well as the variety of times during the day or week when interactions occurred. These patterns offered a sense of how easily users could reach the tool. The final element involved user satisfaction. Although the chatbot did not formally collect feedback, some users left short remarks during their conversations, and these comments were reviewed in the same way that informal feedback has been used to judge the usefulness of other conversational agents in previous studies [3, 11].

User Interaction

During the observation period, the chatbot logged 324 sessions in total. Most of the users who tried the system were new users. Of all the users interacting with the chatbot, 70 were new users, accounting for 69%, while 32 came back for the second or later visit, accounting for 31% of all users. Figure 1 shows this on a graph. The trend is that the tool attracted a relatively consistent flow of new users, while a smaller but significant portion returned for either follow-up questions or further clarification. The mixture of new and returning users is an early indicator that this chatbot was both findable and interesting enough to bring some visitors back for more.

Engagement Metrics

The depth of the user engagements with the chatbot varied, but the patterns of engagement did not change much throughout the testing period. Within these 324 sessions, the chatbot generated a total of 303 replies, while users contributed 246 messages. These are represented in Fig. 2. This slight difference in the number of messages from the chatbot is because several user questions required more than one response for the sake of clarity. The volume of user messages also illustrates that users did not merely open the platform and leave; they asked direct questions and created multiple turns. This reflects an actual interest in the information being provided.

User Interface Experience

Informal feedback from users and observations during the testing support the view that, in general, the interface was easy to use. Figure 3 shows the layout is relatively simple, enabling participants to easily read the responses and find their way around between topics without any confusion. There was also a lack of superfluous features to distract them from their questions. Although the interface was simple, this actually worked in its favour by lessening the barriers for a section of people who are not as accustomed to digital health tools.

Thematic Distribution of User Queries

The 324 recorded questions were categorized by topic area to gain an understanding of what the users were trying to learn. Table 1 summarizes the themes that emerged from the inquiries. The most common subject discussed was symptoms at 21% of all messages. A number of users sought to understand if given stomach discomforts or related symptoms could be indicative of H. pylori infection. Questions on treatment options comprised 19% and showed great interest in how the infection is treated, treatment duration, and potential medication side effects. The queries on diet and lifestyle comprised 17% of the dataset. Many of the questions asked included inquiries about what to eat and avoid that would exacerbate or alleviate the symptoms.

16% of the topics involved transmission and prevention, usually regarding whether food, utensils, or close contact spread the infection. Testing and diagnosis accounted for 13%, with users asking which tests are the most accurate and when testing should be repeated. Risk factors and complications were also 13% of the queries, often concerning ulcers, long-term infection, or cancer. Overall, the thematic analysis shows that users were not focused on a single issue but, instead, asked about a wide range of topics. The interest in symptoms, treatment, and daily management does suggest that many individuals were dealing with stomach-related concerns and wanted guidance on whether these could be linked to H. pylori.

Table 1
A table that organizes the user questions by topic area (N = 324)
Topic Area	Questions	Frequency	Percentage
Symptoms of H. pylori Infection	- "What are the symptoms of an H. pylori infection?" - "How can I tell if my stomach issues are caused by H. pylori?" - "Does H. pylori cause pain and nausea?" - "Are there any early signs of H. pylori?" - "Can H. pylori cause acid reflux or bloating?"	68	21%
Treatment Options	- "What is the best treatment for H. pylori?" - "How long do I need to take antibiotics?" - "Are there natural remedies for H. pylori?" - "Side effects of H. pylori medications?" - "Is there a non-antibiotic cure for H. pylori?"	62	19%
Transmission and Prevention	- "How is H. pylori transmitted?" - "Can I catch H. pylori from sharing food?" - "How to prevent H. pylori infection?" - "Is H. pylori contagious?" - "Hygiene practices to avoid H. pylori?"	52	16%
Dietary Recommendations	- "What foods should I avoid with H. pylori?" - "Diet for managing H. pylori?" - "Foods that worsen H. pylori symptoms?" - "Foods that help fight H. pylori?" - "Should I avoid spicy foods?"	56	17%
Testing and Diagnosis	- "What test confirms H. pylori?" - "Is there a blood test for H. pylori?" - "How accurate is the breath test?" - "When should I retest after treatment?" - "Can I test for H. pylori at home?"	43	13%
Risk Factors and Complications	- "What happens if H. pylori is untreated?" - "Does H. pylori increase cancer risk?" - "Can H. pylori lead to ulcers?" - "Long-term effects of H. pylori?" - "Is there a link between H. pylori and digestive issues?"	43	13%

Overview of Key Findings Taken together, the engagement metrics and thematic patterns demonstrate several key points: people were willing to use the chatbot and engage in substantive exchanges rather than browsing; second, the questions they asked reflect a clear need for accessible information on symptoms, treatment procedures, preventive practices, and dietary guidance. Finally, the fact that some users returned suggests that the tool may have offered value beyond a single interaction.

The findings from this study have established that a simple AI-based chatbot can be a feasible delivery mode for the dissemination of health information on H. pylori to the public. Admittedly, the tool was still in an early phase, but the level of activity it generated suggests willingness to use a digital platform when questions about stomach symptoms, infection risks, or treatment options arise. The types of questions asked by users, mainly symptoms, treatments, diets, and the mode of spread of the infection, reflect those commonly raised during clinical consultations. This suggests that the chatbot was able to address genuine gaps in community knowledge.

When compared to related work on health-focused chatbots, such as those on cancer awareness or general patient education, the patterns presented here were generally consistent with prior results [3, 10]. Users interacted during non-working hours-especially evenings and weekends, similar to other studies. The main difference observed is that most of the questions asked were introductory in nature, reflecting low public awareness about H. pylori in many low-resource settings. This observation is also consistent with prior reports of very limited knowledge of the infection in areas with a high disease burden [5]. Overall, the findings support the role of chatbots in increasing access to health information among populations that would otherwise have limited sources of reliable advice.

These results also bear important implications for public health practice. The times at which users engaged indicate that people seek information at moments when hospitals and clinics are not open, underlining the value of having an always-available educational tool. The range of topics raised-from basic symptoms to questions about cancer risk-reveals that many people lack reliable sources of guidance on H. pylori. While a chatbot cannot substitute for medical care, it can provide timely explanations that may assist users in making more informed choices, including when to seek professional evaluation. This support could contribute to earlier recognition and better outcomes in communities with poor health literacy.

This study has several strengths: it shows that an inexpensive, easily deployable tool can be attractive even without active promotion, and it directly shows what people want to know about H. pylori, through real messages rather than hypothetical survey responses. This natural interaction yields insights that might be very difficult to capture in a traditional questionnaire format.

However, the study also has obvious limitations. Participation was totally dependent on access to the internet, which could have excluded people in rural or underserved areas. Because background information was not collected by the chatbot itself, it's impossible to say which demographic groups were represented and whether some populations were underrepresented. Long-term outcomes were not measured, so it remains unclear how the behavior, testing rates, and treatment adherence of the people using the chatbot were impacted. The tool was available only in English, perhaps restricting its usefulness for those who were more comfortable with Swahili or other local languages.

These findings outline several avenues for future work. Extension of this chatbot to various languages would increase accessibility, especially within communities where the use of English is minimal. The tool could also be linked with health facilities offering test-booking options or even follow-up care reminders. A larger and more diverse user group, combined with systematic methods in the evaluation process, would help determine whether such a chatbot can lead to measurable improvements in awareness, early testing, or prevention of H. pylori infection. This study hence provides early evidence that AI-supported tools can augment conventional methods of public health education. For conditions like H. pylori, which often remain unnoticed until complications develop, these might help overcome the longstanding gaps in knowledge and facilitate earlier engagement with healthcare services.

This pilot demonstrates that a small, low-cost chatbot can reliably deliver scalable H. pylori information. Over a period of three months, 102 individuals conducted 324 sessions and asked 246 questions most frequently about symptoms, treatment, diet, and transmission. Usage clustered in evenings and weekends, suggesting that an always-available tool can meet information needs outside routine clinic hours. Taken together, these patterns suggest that a simple conversational system can augment community health education in helping people recognize when testing or care is needed and reinforce basic prevention messages in settings with limited health-literacy support.

The analysis is exploratory and based on anonymous analytics; it does not measure clinical outcomes, and it depends on internet access. This said, the approach is practical, inexpensive, and easy to scale. Next steps should include localisation (e.g., Swahili), linkage to testing and referral pathways, and a larger evaluation that tracks behaviour change and care-seeking. In sum, an AI-supported chatbot is a feasible addition to public-health communication on H. pylori, with clear potential to strengthen early information, triage, and prevention in resource-constrained communities.

Artificial Intelligence

H.pylori

Helicobacter pylori

LMICs

Low and Middle income Countries

MALT

Mucosa Associated Lymphoid Tissue

NLU

Natural Language Understanding

NLP

Natural Language Processing

P&C

Prompt and Completion

User Interface

WHO

World Health Organisation

Ethics Approval Statement

This study was conducted in accordance with ethical guidelines for research involving human participants. As this study involved the use of a publicly accessible AI-powered chatbot with no collection of personally identifiable information, no formal ethics approval was required. All participants provided electronic informed consent prior to engaging with the chatbot. The study adhered to ethical principles, ensuring privacy and confidentiality of all user interactions.

Author Contribution

H.J. conceived the study idea, developed the chatbot framework, curated the clinical content, and led the drafting of the manuscript. R.S. contributed to the chatbot development process, performed data extraction and analysis, and assisted in interpreting user engagement findings. H.J. and R.S. jointly revised the manuscript for important intellectual content. Both authors approved the final version of the manuscript and agree to be accountable for all aspects of the work.

Acknowledgement

The authors thank all individuals who interacted with the chatbot during the testing period; their participation made this work possible. We also acknowledge the Botpress development community for providing an open-source platform that supported the technical implementation of the chatbot. Special appreciation goes to Dr. Leonard Katalambula for his guidance and constructive insights during the development and refinement of this study.

Data Availability

The study used anonymised chatbot interaction logs consisting only of message text and timestamps. No personal or identifying information was collected. The dataset supporting the findings of this study is available from the corresponding author on reasonable request.

Alsulimani A, Akhter N, Jameela F, Ashgar RI, Jawed A, Hassani MA, et al. The impact of artificial intelligence on microbial diagnosis. Microorganisms. 2024;12(6):1–20. 10.3390/microorganisms12061051.
Benson T. Exploring the potential of customized AI chatbots in food policy research: capabilities and constraints in comparative perspective. 2023:1–30.
Benson T. Exploring the potential of customized AI chatbots in food policy research: capabilities and constraints in comparative perspective. 2023:1–30.
Burucoa C, Axon A. Epidemiology of Helicobacter pylori infection. Helicobacter. 2017;22(S1):e12403. 10.1111/hel.12403.
Car LT, Dhinagaran DA, Kyaw BM, Kowatsch T, Joty S, Theng YL, et al. Conversational agents in health care: scoping review and conceptual analysis. J Med Internet Res. 2020;22(8):e17158. 10.2196/17158.
Gao Z, Ge J, Xu R, Chen X, Cai Z. Potential application of ChatGPT in Helicobacter pylori disease-relevant queries. Front Med. 2024;11:1489117. 10.3389/fmed.2024.1489117.
Görtz M, Baumgärtner K, Schmid T, Muschko M, Woessner P, Gerlach A et al. An artificial intelligence-based chatbot for prostate cancer education: design and patient evaluation study. J Med Internet Res. 2023;&#8212.
Görtz M, Baumgärtner K, Schmid T, Muschko M, Woessner P, Gerlach A, et al. An artificial intelligence-based chatbot for prostate cancer education. Digit Health. 2023;9:1–10. 10.1177/20552076231173304.
Graham DY, Rugge M, Genta RM. Diagnosis and treatment of Helicobacter pylori: a clinical perspective. J Gastroenterol Hepatol. 2021;36(3):472–9. 10.1111/jgh.15229.
Jaroń K, Pietrzak A, Daniluk J, Adrych K, Gąsiorowska A, Skrzydło-Radomańska B, et al. Diagnostic and therapeutic recommendations on Helicobacter pylori infection: recommendations of the Working Group of the Polish Society of Gastroenterology. Prz Gastroenterol. 2023;18(3):225–48. 10.5114/pg.2023.131998.
Khairatun Hisan U, Miftahul Amri M. ChatGPT and medical education: a double-edged sword. J Pedagogy Educ Sci. 2023;2(01):71–89. 10.56741/jpes.v2i01.302.
Maida M, Ramai D, Mori Y, Dinis-Ribeiro M, Facciorusso A, Hassan C. The role of generative language systems in increasing patient awareness of colon cancer screening. Endoscopy. 2024;—. 10.1055/a-2388-6084
Medical P. Artificial intelligence in medicine: a new frontier. 1986;54–60.
Nijevitch A, Starostina O, Demidov A. Helicobacter pylori infection: epidemiological challenges in the 21st century. World J Gastroenterol. 2022;28(5):467–81. 10.3748/wjg.v28.i5.467.
Proposals RFOR. Development of a custom AI chatbot for the—. 1–9.
Yang H, Hu B. Diagnosis of Helicobacter pylori infection and recent advances. Diagnostics. 2021;11(8):1–15. 10.3390/diagnostics11081305.
Zhang W, Song LN, You YF, Qi FN, Cui XH, Yi MX, et al. Application of artificial intelligence in the prediction of immunotherapy efficacy in hepatocellular carcinoma: current status and prospects. Artif Intell Gastroenterol. 2024;5(1):–. 10.35712/aig.v5.i1.90096.
Zhou J, Li T, Fong SJ, Dey N, Crespo RG. Exploring ChatGPT’s potential for consultation, recommendations, and diagnosis: gastric cancer and gastroscopy case reports. Int J Interact Multimed Artif Intell. 2023;8(2):7–13. 10.9781/ijimai.2023.04.007.

No competing interests reported.

Download PDF

Editorial decision: Revision requested
19 Nov, 2025
Editor assigned by journal
19 Nov, 2025
Submission checks completed at journal
19 Nov, 2025
First submitted to journal
09 Nov, 2025

You are reading this latest preprint version

An artificial intelligence-based chatbot for Helicobacter Pylori Health Education: A Potential tool for Community Awareness

Status:

Version 1

Abstract

Figures

Background

Methods

Study Aim, Design and Setting

Materials and AI Platform

Chatbot Development Procedures

Prompt and Completion Training

Conversation Flow and Interface Design

Deployment and User Testing

Participants

Data Collection and Management

Data Analysis

Descriptive Analysis

Thematic Analysis

Evaluation Metrics

Results

User Interaction

Engagement Metrics

User Interface Experience

Thematic Distribution of User Queries

Discussion

Conclusions

Abbreviations

Declarations

Author Contribution

Acknowledgement

Data Availability

References

Additional Declarations

Status:

Version 1