A Simulation Study to Advance Human-Centred Artificial Intelligence via Digital Citizen Science: Can Large Language Models Transform Current Approaches to Missing Data Imputation?
Why this work is in the frame
A frame that forgets how it found something cannot be audited. These are the routes that admitted this work.
Bibliographic record
Abstract
Abstract Background Missing data is a persistent challenge in digital health research, and traditional approaches like Multiple Imputation by Chained Equations (MICE) may not capture complex patterns. While large language models (LLMs) could offer a viable alternative, their use in this context remains understudied. Moreover, a critical gap remains in embedding human-centred artificial intelligence (AI) approaches that integrate equity, transparency, and stakeholder participation. Digital citizen science, which leverages citizen-owned devices for ethical, participatory big data collection, offers a foundation to advance such approaches in digital health. Objective To evaluate and compare the imputation accuracy of MICE with the OpenAI o3 model for categorical variables in a simulated digital health dataset under different missingness mechanisms and levels, while situating this evaluation within the broader vision of human-centred AI enabled by digital citizen science. Methods A complete digital health dataset collected through a digital citizen science platform was used to simulate missingness under Missing at Random (MAR) and Missing Completely at Random (MCAR) at 10%, 25%, and 50%. MICE used logistic regression with five imputations and ten iterations per chain. For the o3 model, structured prompts were generated for each missing entry using all available non-missing variables from the same record. Both methods were evaluated on each simulated dataset using classification accuracy and a closeness metric representing similarity to the original data. Statistical differences were tested with a two-sample Z-test, and misclassification patterns were examined by variable type and category frequency. Results Under MAR conditions, MICE and o3 performed similarly with an average accuracy of 0.60 and 0.59, and closeness metrics of 0.83 and 0.85, respectively. Under MCAR, both methods achieved 0.59 accuracy, with closeness metrics of 0.84 and 0.85. No statistically significant differences were found across conditions (all p > 0.05). Conclusion While MICE remains preferred for continuous data, the o3 model shows promise as a complementary tool for categorical imputation in smaller datasets. Beyond methodological comparability, this study demonstrates how digital citizen science can serve as an ethical foundation for embedding human-centred AI into digital health research, positioning large language models not only as technical tools but also as vehicles for advancing equity, transparency, and participatory innovation in healthcare.
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.
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 | 0.003 | 0.001 |
| Meta-epidemiology (narrow) | 0.000 | 0.000 |
| Meta-epidemiology (broad) | 0.000 | 0.000 |
| Bibliometrics | 0.001 | 0.002 |
| Science and technology studies | 0.001 | 0.000 |
| Scholarly communication | 0.001 | 0.001 |
| Open science | 0.002 | 0.001 |
| Research integrity | 0.000 | 0.000 |
| 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 it