Automated Translation of Chronic Disease Diagnosis Codes using the ChatGPT Large Language Model
Bibliographic record
Abstract
BackgroundThe International Classification of Diseases (ICD) is revised over time and there are region-specific versions, including ICD-10-CA (Canada) and ICD-9-CM (USA). Studies spanning multiple ICD versions require crosswalks to translate diagnosis codes across versions, but manual crosswalk development is costly and requires clinical expertise. ObjectiveTo evaluate the accuracy of a pre-trained large language model (LLM) to automatically translate chronic disease diagnosis codes from ICD-10-CA to ICD-9-CM. ApproachEight prompts were developed to instruct the OpenAI Generative Pre-trained Transformer 4 (GPT-4) LLM to translate 1,272 ICD-10-CA codes for the Elixhauser Comorbidity Index to ICD-9-CM. Prompt accuracy (%) was measured against a crosswalk developed by the Canadian Institute of Health Information. Variability was assessed by replicating each prompt three times. Mean accuracy ± standard deviation was reported for each prompt across replications, for both five-digit and truncated three-digit codes. ResultsThe highest prompt performance was observed when assigning a persona of a medical coding specialist (40.8% ± 0.9%), requesting justification for the selected code (41.4% ± 1.1%), and providing diagnosis code labels (47.5% ± 0.7%). For truncated three-digit codes, these prompts achieved accuracy of 82.0% ± 0.5%, 80.8% ± 0.9%, and 82.7% ± 0.1%, respectively. Combining these three prompting techniques marginally improved accuracy to 48.6% ± 0.7% for five-digit codes and 84.3% ± 0.2% for truncated three-digit codes. ConclusionGeneral-purpose LLMs are currently not sufficiently accurate at automating ICD code translation for chronic diseases. ImplicationsAdditional experiments with fine-tuning, task-specific training, and prompt engineering are needed to improve accuracy and reduce variability.
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
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 | 0.000 | 0.000 |
| Science and technology studies | 0.001 | 0.000 |
| Scholarly communication | 0.000 | 0.002 |
| Open science | 0.001 | 0.000 |
| 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 itClassification
machine, unvalidatedMachine predicted; a candidate call from one teacher head, not a consensus.
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".