Designing Effective Health Surveys: HBM, AI, and Kis Platform Guide

Researchers often find it difficult to make new information relatable and interesting for respondents. When survey items are too abstract, participants struggle to connect them with what they already know, which can reduce data quality. To illustrate how these challenges can be addressed, we will walk through a case‑study questionnaire on malaria vaccine awareness, perception, and willingness to accept.

Frameworks for Questionnaire Design

Two main approaches guide questionnaire construction. A theory‑based approach uses a pre‑defined framework to ensure completeness, while Grounded Theory lets concepts emerge from the data itself. For health‑promotion topics such as vaccines, a theory‑based approach is generally recommended because it helps capture all relevant constructs. The Health Belief Model (HBM) is highlighted as a suitable framework for vaccine‑related surveys.

Health Belief Model Components

The HBM predicts health behaviours based on individual perceptions. Its key components are:

• Modifying variables (demographics, knowledge)
• Perceived seriousness of the health problem
• Perceived susceptibility to the problem
• Perceived benefits of taking action
• Perceived barriers to taking action
• Perceived threats (a combination of seriousness and susceptibility)
• Self‑efficacy (confidence in one’s ability to act)
• Cues to action (triggers that prompt the behaviour)

All of these elements should be reflected in the questionnaire if the study claims to be grounded in the HBM.

Analysis of the Malaria Vaccine Questionnaire

Demographic Section Review

The original questionnaire collected age, marital status, education, occupation, number of children, and residence. Several issues emerged:

• Incomplete items – “age” lacked units or response options.
• Ambiguous response categories – overlapping age ranges and unclear marital‑status definitions.
• Unclear measurement – “children” did not specify whether it referred to biological children, those living at home, or those under 18.
• Inconsistent formatting – mixture of camel case and sentence case.
• Missing “other” and “not sure” options – needed for exhaustiveness.
• Lack of explicit instructions – such as “please select one”.

Awareness, Perception, and Willingness Sections

• Awareness: The yes/no format (“Have you ever heard about malaria vaccine?”) was too simplistic. Skip logic should direct respondents based on their answer.
• Perception: Many items were leading, e.g., “How serious do you think malaria is?” which presumes seriousness. Neutral phrasing like “How serious or not serious do you consider malaria?” is preferred.
• Willingness: Questions about vaccinating children were asked of all respondents, even those who were unwilling or unsure. Skip logic can prevent unnecessary burden, and technical terms such as “effectiveness” and “safety” should be simplified.

Using AI for Questionnaire Improvement

ChatGPT and Bard were employed to compare the questionnaire against the HBM. The AI tools identified missing elements—most notably the lack of perceived susceptibility items—and suggested draft questions. While AI can quickly highlight gaps and generate wording, human review remains essential to catch bias, ensure contextual relevance, and refine language. The recommended workflow is to draft the questionnaire first, then use AI to spot omissions and improve phrasing.

Detailed Revision Principles

Each survey item must contain three elements:

1. The question itself
2. Clear instructions (e.g., “please select one”)
3. Response options

Response options should be mutually exclusive (no overlap) and collectively exhaustive (cover all possibilities). Additional principles include:

• Avoid leading or ambiguous language.
• Use sentence case consistently.
• Provide explicit units and definitions (e.g., continuous age data).
• Include “other” and “I don’t know” choices where appropriate.
• Combine related items to reduce respondent burden (“questions are real estate”).
• Apply skip logic so participants only answer eligible questions.

Practical Application: Kis Platform

The Kis platform was demonstrated for end‑to‑end survey management:

• Project setup – title, aims, keywords, and collaborator invitations.
• Importing – questionnaire transferred from Word, with sections formatted as headers.
• Editing – age converted to a continuous field, exclusive source‑of‑information options added, and selectable answer limits defined.
• Logic configuration – inclusion criteria based on age and children, skip patterns for willingness to vaccinate.
• Preview and launch – logic and flow checked before publishing the survey link.
• Data visualization – immediate response monitoring and automatic methodology report generation.

Q&A Highlights

• Likert scales can be treated as structured data for quantitative analysis.
• Mixed‑methods designs are possible by adding open‑ended questions.
• Survey length matters: exceeding 30 minutes may reduce data quality; shuffling items can mitigate order bias.
• Sample size without prevalence data requires specialized calculation methods.
• Response options should be exhaustive, even if some appear extreme (e.g., “very unsafe”).
• Continuous vs. categorical data – collecting age as a continuous variable allows flexible later categorization.
• AI assistance is useful for grammar checks and simplification, but the initial draft should be created by the researcher.
• Maximum question count and logical branching depend on study goals and respondent fatigue considerations.