Pan African Clinical Trials Registry
South African Medical Research Council, South African Cochrane Centre
PO Box 19070, Tygerberg, 7505, South Africa
Telephone: +27 21 938 0506 / +27 21 938 0834 Fax: +27 21 938 0836
Email: pactradmin@mrc.ac.za Website: pactr.samrc.ac.za
Trial no.: PACTR202506786881322 Date of Approval: 20/06/2025
Trial Status: Registered in accordance with WHO and ICMJE standards
TRIAL DESCRIPTION
Public title CHEWA: Evaluating the Impact of an LLM-Enabled Virtual Voice Assistant on Community Health Extension Workers’ Decision-Making and Patient Safety: A Multi-Site, Prospective, Pre-Post Comparative Study
Official scientific title CHEWA: Evaluating the Impact of an LLM-Enabled Virtual Voice Assistant on Community Health Extension Workers’ Decision-Making and Patient Safety: A Multi-Site, Prospective, Pre-Post Comparative Study
Brief summary describing the background and objectives of the trial Community health extension workers (CHEWs) play a crucial role in Nigeria’s health care system, particularly in rural and underserved areas where they often provide the first point of contact for health care services. Their responsibilities include providing preventive care, health education, basic health screenings, and referral of more complex cases requiring higher levels of care. The Nigerian national health policy recognizes CHEWs as primary health care providers, acknowledging their role in delivering essential health care in the absence of a sufficient number of physicians (Kress et al. 2016) (O. Abah 2023). Advancements in large language models (LLMs) present a new opportunity to support CHEWs in delivering accurate, timely, and comprehensive health care. By providing real-time, contextually appropriate information, LLMs could help CHEWs overcome existing barriers in accessing medical knowledge (Nazi and Peng 2024). This study aims to assess the potential of LLMs in frontline health care settings. By addressing gaps in real-time clinical guidance, the integration of LLMs into the CHEW workflow could reduce unnecessary referrals, improve patient outcomes, and strengthen Nigeria’s primary health care system.
Type of trial Observational
Acronym (If the trial has an acronym then please provide) CHEWA
Disease(s) or condition(s) being studied
Sub-Disease(s) or condition(s) being studied
Purpose of the trial Supportive care
Anticipated trial start date 11/06/2025
Actual trial start date
Anticipated date of last follow up 10/07/2025
Actual Last follow-up date 31/08/2025
Anticipated target sample size (number of participants) 3000
Actual target sample size (number of participants)
Recruitment status Not yet recruiting
Publication URL
Secondary Ids Issuing authority/Trial register
82324756 Nigeria Clinical Trials Registry
STUDY DESIGN
Intervention assignment Allocation to intervention If randomised, describe how the allocation sequence was generated Describe how the allocation sequence/code was concealed from the person allocating the participants to the intervention arms Masking If masking / blinding was used
Single Group Non-randomised Allocation Sequence/Code was not concealed Open-label(Masking Not Used)
INTERVENTIONS
Intervention type Intervention name Dose Duration Intervention description Group size Nature of control
Experimental Group CHEWA 3 weeks In Nigeria, we are trialling a voice-first LLM-enabled ‘expert-in-your-pocket’, designed to support community health extension workers (CHEWs) via feature/basic mobile phones. After obtaining informed consent from a patient, CHEWs dial a toll-free phone number to ask clinical questions and receive real-time guidance generated by an LLM. By using feature phones, we seek to address another major challenge, that of accessible and affordable connectivity and devices, by evaluating tools that can be utilised without immediate reliance on internet access. This study seeks to not only assess proximal outcomes, i.e., the quality of LLM responses based on adherence to medical science, relevance, clarity, and contextual appropriateness, but also uses a pre/post design at the individual level to assess the tool’s impact on decision making. Crucially, a "safety net" mechanism ensures that a senior clinician reviews all LLM advice within one hour. If potentially harmful advice is identified, the CHEW is immediately contacted by the senior clinician, and corrective actions are initiated. This real-time oversight seeks to enable the generation of high-quality interventional study data for AI-supported decision making, while minimizing risks to patients. 3000
ELIGIBILITY CRITERIA
List inclusion criteria List exclusion criteria Age Category Minimum age Maximum age Gender
CHEW based in Kano or Oyo State Non-CHEW or CHEW outside Kano or Oyo states Adult: 19 Year-44 Year 19 Year(s) 44 Year(s) Both
ETHICS APPROVAL
Has the study received appropriate ethics committee approval Date the study will be submitted for approval Date of approval Name of the ethics committee
Yes 12/05/2025 Institute of Advanced Medical Research UIUCH Research Ethics Committee University of Ibadan
Ethics Committee Address
Street address City Postal code Country
University of Ibadan Ibadan 200111 Nigeria
Has the study received appropriate ethics committee approval Date the study will be submitted for approval Date of approval Name of the ethics committee
Yes 31/12/2024 Kano State Ministry of Health
Ethics Committee Address
Street address City Postal code Country
Post Office Road, PMB 3066 Kano Kano 700101 Nigeria
OUTCOMES
Type of outcome Outcome Timepoint(s) at which outcome measured
Primary Outcome The primary outcome of this study is the quality assessment of CHEWA responses based on a predefined, study-specific evaluation rubric. An expert panel of experienced clinicians will assess the CHEWA responses by reviewing interaction transcripts, including the questions posed by CHEWs and the corresponding answers generated by the LLM-based virtual assistant. Two expert panel members will independently evaluate each response based on four key domains. Responses in Arm 1 (patient-specific advice) will be evaluated across all four domains: (A) adherence to accepted medical science and consensus, (B) relevance to the CHEW’s query, (C) clarity of communication, and (D) contextual appropriateness for a resource-constrained setting. In contrast, responses in Arm 2 (general, non-patient-specific clinical queries) will be assessed only on domains A, B, and C, as contextual relevance and suitability are less applicable to broader clinical inquiries. Pre-Post intercation between CHEW and CHEW
Secondary Outcome This secondary outcome will assess whether the CHEW’s management plan changes after interacting with the LLM and whether these changes improve or worsen patient care. To establish a baseline, the CHEW will be asked: ‘Can you summarise the patient’s background, your planned next steps, and what is your first question’? After completing their interaction with the LLM component of CHEWA, the CHEW will answer the question: ‘Did you follow CHEWA’s advice’? and the CHEW will have to provide a response. If they did not follow or only followed some of the advice, the CHEW will be prompted to answer: ‘What did you do differently to CHEWA’s advice, and why’? The evaluation panel will be given all relevant documented information in order to determine the ground truth for each case and to compare both the pre-CHEWA and post-CHEWA management plans against this reference. pre-CHEWA and post-CHEWA interaction
RECRUITMENT CENTRES
Name of recruitment centre Street address City Postal code Country
Several health facilities in the state Several Primary Health facilities Oyo Nigeria
Several health facilities Several health facilities Kano Nigeria
FUNDING SOURCES
Name of source Street address City Postal code Country
The Gates Foundation 500 5th Ave N Seattle WA 98109 United States of America
SPONSORS
Sponsor level Name Street address City Postal code Country Nature of sponsor
Primary Sponsor PATH 2201 Westlake Avenue Seattle WA 98121 United States of America Charities/Societies/Foundation
COLLABORATORS
Name Street address City Postal code Country
CONTACT PEOPLE
Role Name Email Phone Street address
Principal Investigator Elizabeth Omoluabi elizomoluabi@gmail.com +2348136174697 F Road 20, Citec Mbora
City Postal code Country Position/Affiliation
Abuja 900001 Nigeria CRERD
Role Name Email Phone Street address
Public Enquiries Charles Ohikhuai charles.ohikhuai@viamo.io +2348025954513 Viamo Abuja
City Postal code Country Position/Affiliation
Abuja 900001 Nigeria Head of Programs
Role Name Email Phone Street address
Scientific Enquiries Bilal Mateen bmateen@path.org +12028220033 455 Massachusetts Ave NW
City Postal code Country Position/Affiliation
Washington DC DC 20001 United States of America Chief AI Officer
REPORTING
Share IPD Description Additional Document Types Sharing Time Frame Key Access Criteria
Yes The study is fully compliant with the General Data Protection Regulation. To ensure confidentiality, informed consent will be obtained from all participants (CHEWs and patients) providing clear information about the study and their right to privacy. The data in Viamo’s database is secured and encrypted with restricted access. All members of the research team understand the importance of confidentiality and the imperative to use the data received only for the purpose of this research. All data to be shared publicly will be de-identified with no direct link to participants. On the dashboard, results will not include CHEW or patient names or phone numbers. All data stored on Viamo’s database will be kept for at least five years or may be deleted from the platform upon request from PATH. The data can also be downloaded and shared with PATH for further use or storage if necessary. Informed Consent Form,Statistical Analysis Plan,Study Protocol 12 months Legitimate request
URL Results Available Results Summary Result Posting Date First Journal Publication Date
No
Result Upload 1: Result Upload 2: Result Upload 3: Result Upload 4: Result Upload 5:
Result URL Hyperlinks Link To Protocol
Result URL Hyperlinks
Changes to trial information
Section Name Field Name Date Reason Old Value Updated Value
Intervention Intervention List 09/06/2025 We do not have an experimental and control group. This is an observational study, so the total sample is 3000. However it is taking place in two states: Oyo and Kano. Ecah state is enrolling 1500 participants each. Experimental Group, CHEWA, , 3 weeks, Interaction between LLM and Community Health Workers, 1500, Experimental Group, CHEWA, , 3 weeks, Interaction between LLM and Community Health Workers, 3000,
Section Name Field Name Date Reason Old Value Updated Value
Intervention Intervention List 17/06/2025 We do not have an experimental and control group. This is an observational study, so the total sample is 3000. However it is taking place in two states: Oyo and Kano. Ecah state is enrolling 1500 participants each. Experimental Group, CHEWA, , 3 weeks, Interaction between LLM and Community Health Workers, 3000, Experimental Group, CHEWA, , 3 weeks, In Nigeria, we are trialling a voice-first LLM-enabled ‘expert-in-your-pocket’, designed to support community health extension workers (CHEWs) via feature/basic mobile phones. CHEWs dial a toll-free phone number to ask clinical questions and receive real-time guidance generated by an LLM. By using feature phones, we seek to address another major challenge, that of accessible and affordable connectivity and devices, by evaluating tools that can be utilised without immediate reliance on internet access. This study seeks to not only assess proximal outcomes, i.e., the quality of LLM responses based on adherence to medical science, relevance, clarity, and contextual appropriateness, but also uses a pre/post design at the individual level to assess the tool’s impact on decision making. Crucially, a "safety net" mechanism ensures that a senior clinician reviews all LLM advice within one hour. If potentially harmful advice is identified, the CHEW is immediately contacted, and corrective actions are initiated. This real-time oversight seeks to enable the generation of high-quality interventional study data for AI-supported decision making, while minimizing risks to patients. , 3000,
Section Name Field Name Date Reason Old Value Updated Value
Intervention Intervention List 17/06/2025 We do not have an experimental and control group. This is an observational study, so the total sample is 3000. However it is taking place in two states: Oyo and Kano. Ecah state is enrolling 1500 participants each. Experimental Group, CHEWA, , 3 weeks, In Nigeria, we are trialling a voice-first LLM-enabled ‘expert-in-your-pocket’, designed to support community health extension workers (CHEWs) via feature/basic mobile phones. CHEWs dial a toll-free phone number to ask clinical questions and receive real-time guidance generated by an LLM. By using feature phones, we seek to address another major challenge, that of accessible and affordable connectivity and devices, by evaluating tools that can be utilised without immediate reliance on internet access. This study seeks to not only assess proximal outcomes, i.e., the quality of LLM responses based on adherence to medical science, relevance, clarity, and contextual appropriateness, but also uses a pre/post design at the individual level to assess the tool’s impact on decision making. Crucially, a "safety net" mechanism ensures that a senior clinician reviews all LLM advice within one hour. If potentially harmful advice is identified, the CHEW is immediately contacted, and corrective actions are initiated. This real-time oversight seeks to enable the generation of high-quality interventional study data for AI-supported decision making, while minimizing risks to patients. , 3000, Experimental Group, CHEWA, , 3 weeks, In Nigeria, we are trialling a voice-first LLM-enabled ‘expert-in-your-pocket’, designed to support community health extension workers (CHEWs) via feature/basic mobile phones. After obtaining informed consent from a patient, CHEWs dial a toll-free phone number to ask clinical questions and receive real-time guidance generated by an LLM. By using feature phones, we seek to address another major challenge, that of accessible and affordable connectivity and devices, by evaluating tools that can be utilised without immediate reliance on internet access. This study seeks to not only assess proximal outcomes, i.e., the quality of LLM responses based on adherence to medical science, relevance, clarity, and contextual appropriateness, but also uses a pre/post design at the individual level to assess the tool’s impact on decision making. Crucially, a "safety net" mechanism ensures that a senior clinician reviews all LLM advice within one hour. If potentially harmful advice is identified, the CHEW is immediately contacted by the senior clinician, and corrective actions are initiated. This real-time oversight seeks to enable the generation of high-quality interventional study data for AI-supported decision making, while minimizing risks to patients. , 3000,
Section Name Field Name Date Reason Old Value Updated Value
Recruitment Centre RecruitmentCentre List 09/06/2025 Putting each recruitment centre in its line Several health facilities in the state, Several Primary Health facilities, Oyo and Kano, , Nigeria Several health facilities in the state, Several Primary Health facilities, Oyo , , Nigeria
Section Name Field Name Date Reason Old Value Updated Value
Recruitment Centre RecruitmentCentre List 09/06/2025 Putting each recruitment centre in its line Several health facilities , Several health facilities , Kano, , Nigeria
Section Name Field Name Date Reason Old Value Updated Value
Ethics Ethics List 09/06/2025 There are two locations and we received ethics approval from each site: Kano and Oyo. I am now trying to submit both TRUE, Kano State Ministry of Health, Post Office Road, PMB 3066 Kano, Kano, 700101, Nigeria, , 31 Dec 2024, +2348088566747, smoh@smoh.org.ng, 34815_31976_4737.pdf
Section Name Field Name Date Reason Old Value Updated Value
Ethics Ethics List 09/06/2025 There are two locations and we received ethics approval from each site: Kano and Oyo. However, I am unable to load the two documents on the platform, so I submitted one. It would also not accept the second one, even if it is a PDF. Please advise How I can seubmit both approval letters. TRUE, Institute of Advanced Medical Research UIUCH Research Ethics Committee University of Ibadan, University of Ibadan, Ibadan, 200111, Nigeria, , 12 May 2025, +234803268431, ikeajayi2003@yahoo.com, 34815_31971_4737.pdf TRUE, Institute of Advanced Medical Research UIUCH Research Ethics Committee University of Ibadan, University of Ibadan, Ibadan, 200111, Nigeria, , 12 May 2025, +234803268431, ikeajayi2003@yahoo.com, 34815_31971_4737.pdf
Section Name Field Name Date Reason Old Value Updated Value
Reporting IPD description 09/06/2025 Added more information on IPD description Cleaned and de-identified data. The study is fully compliant with the General Data Protection Regulation. To ensure confidentiality, informed consent will be obtained from all participants (CHEWs and patients) providing clear information about the study and their right to privacy. The data in Viamo’s database is secured and encrypted with restricted access. All members of the research team understand the importance of confidentiality and the imperative to use the data received only for the purpose of this research. All data to be shared publicly will be de-identified with no direct link to participants. On the dashboard, results will not include CHEW or patient names or phone numbers. All data stored on Viamo’s database will be kept for at least five years or may be deleted from the platform upon request from PATH. The data can also be downloaded and shared with PATH for further use or storage if necessary.
Section Name Field Name Date Reason Old Value Updated Value
Eligibility Minimum age 09/06/2025 Correcting for age of community health workers in the study 18 Year(s) 19 Year(s)
Section Name Field Name Date Reason Old Value Updated Value
Eligibility Maximum age 09/06/2025 Correcting for age of community health workers in the study 55 Year(s) 44 Year(s)