Methods to Achieve High Interrater Reliability in Data Collection From Primary Care Medical Records
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
PURPOSE: We assessed interrater reliability (IRR) of chart abstractors within a randomized trial of cardiovascular care in primary care. We report our findings, and outline issues and provide recommendations related to determining sample size, frequency of verification, and minimum thresholds for 2 measures of IRR: the κ statistic and percent agreement. METHODS: We designed a data quality monitoring procedure having 4 parts: use of standardized protocols and forms, extensive training, continuous monitoring of IRR, and a quality improvement feedback mechanism. Four abstractors checked a 5% sample of charts at 3 time points for a predefined set of indicators of the quality of care. We set our quality threshold for IRR at a κ of 0.75, a percent agreement of 95%, or both. RESULTS: Abstractors reabstracted a sample of charts in 16 of 27 primary care practices, checking a total of 132 charts with 38 indicators per chart. The overall κ across all items was 0.91 (95% confidence interval, 0.90-0.92) and the overall percent agreement was 94.3%, signifying excellent agreement between abstractors. We gave feedback to the abstractors to highlight items that had a κ of less than 0.70 or a percent agreement less than 95%. No practice had to have its charts abstracted again because of poor quality. CONCLUSIONS: A 5% sampling of charts for quality control using IRR analysis yielded κ and agreement levels that met or exceeded our quality thresholds. Using 3 time points during the chart audit phase allows for early quality control as well as ongoing quality monitoring. Our results can be used as a guide and benchmark for other medical chart review studies in primary care.
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 | 0.034 | 0.014 |
| Meta-epidemiology (narrow) | 0.000 | 0.000 |
| Meta-epidemiology (broad) | 0.001 | 0.000 |
| Bibliometrics | 0.000 | 0.001 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.000 | 0.000 |
| Open science | 0.003 | 0.001 |
| Research integrity | 0.000 | 0.000 |
| Insufficient payload (model declined to judge) | 0.001 | 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