Identifying and Analyzing Bot-Generated Responses in Healthcare Research (Preprint)
Bibliographic record
Abstract
<sec> <title>BACKGROUND</title> The increasing reliance on online surveys for collecting patient-reported outcome measures (PROMs) and patient-reported experience measures (PREMs) has led to growing concerns over fraudulent responses generated by bots. These automated responses threaten data integrity by fabricating survey results, distorting statistical analyses, and potentially misguiding policy decisions. Addressing this issue is critical for maintaining the validity of research findings that inform healthcare practice and policy. </sec> <sec> <title>OBJECTIVE</title> This study aimed to develop a robust set of criteria for identifying bot-generated responses in online healthcare surveys and to examine how these responses impact data quality. We then explored differences in survey results between human and bot respondents in a survey assessing PROMs and PREMs within a geographic region in Ontario, Canada. </sec> <sec> <title>METHODS</title> A survey was conducted from July to October 2023 using REDCap, distributed with a generic link via email and later shared on social media. The survey collected data on healthcare use, patient experiences, health outcomes, digital healthcare engagement, and demographics. A three-tier classification system was developed to detect bot responses based on predefined “red flags,” including duplicate open-ended responses, inconsistent demographic reporting, identical timestamps, and location discrepancies. Quantitative analysis included chi-square tests to assess differences between human and bot responses and Spearman’s correlation tests to examine relationships amongst healthcare indicators. </sec> <sec> <title>RESULTS</title> Analysis included 1,154 responses, with 58% (n=668) classified as bot-generated. The most frequent bot-identification criterion was duplicated open-ended responses (n=293). Chi-square tests revealed statistically significant differences (P<.001) between bots and humans across nearly all survey items. Bots demonstrated response patterns concentrated in the middle of Likert scales, whereas humans were more likely to select extreme values. Correlation analyses showed that expected relationships between key health indicators (e.g., depression symptoms) were present in human responses but reversed in bot-generated data, highlighting the potential for compromised validity in unfiltered survey datasets. </sec> <sec> <title>CONCLUSIONS</title> The findings underscore the necessity of implementing bot prevention and detection methods in online healthcare surveys to preserve data integrity. Failure to do so risks distorting research conclusions, particularly in health equity studies where demographic misclassification may bias results. The study highlights effective bot detection strategies, including open-text analysis, timestamp evaluation, and geographic validation, and recommends integrating these techniques into survey design. As bots continue to evolve, ongoing advancements in bot prevention and detection will be crucial to maintaining the reliability of digital health research. </sec>
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Direct model labels (unvalidated)
Per-model category and study-design labels from the labeling rounds. They are machine output, unvalidated, and the disagreement between models ships as data. No study design here is MEDLINE-validated yet.
| Model arm | Categories | Study design | Confidence |
|---|---|---|---|
| gemma | MetaresearchResearch integrity Domain: Methods · Genre: Empirical About the Canadian research system: no · About a Canadian topic: no | Observational | low |
| gpt | Metaresearch Domain: Methods · Genre: Empirical About the Canadian research system: no · About a Canadian topic: no | Observational | low |
Full frame distilled prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. Learned from the 10,348 direct Codex labels and 10,348 direct Gemma labels. Candidate is the union of thresholded teacher heads; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels or direct frontier model labels.
Codex and Gemma teacher scores by category
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.005 | 0.001 |
| Meta-epidemiology (narrow) | 0.000 | 0.000 |
| Meta-epidemiology (broad) | 0.000 | 0.000 |
| Bibliometrics | 0.002 | 0.001 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.001 | 0.001 |
| Open science | 0.000 | 0.002 |
| Research integrity | 0.000 | 0.002 |
| Insufficient payload (model declined to judge) | 0.000 | 0.000 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedLabeled directly by 2 models reading the full record.
The models disagree on parts of this classification; every voice is preserved in the section at the end of the page.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".