A bayesian way to correct for measurement error in drug risk estimates from EHR data.
Bibliographic record
Abstract
IntroductionData from electronic medical records is now readily available and records information needed in pharmacoepidemiological studies not usually found in administrative data such as risk factors and biometrics. Yet, EMR data leads to measurement error due to primary non-adherence. Bayesian bias correction could provide corrected estimates from administrative data. Objectives and ApproachWe present a method for correcting risk estimates from EMR data using linked data. In our example, we estimate the risk of cardiovascular events from oral-hypoglycemics in patients with type-2 diabetes in Boston, Quebec, and the UK between 2009 and 2012. Using linked EMR and administrative data in Quebec, we compute a positive and negative predicting value of prescription on dispensation for each class of oral-hypoglycemics. The cardiovascular risk is then analysed using a bayesian Weibull survival model adjusted for potential confounders. A similar model is then computed that accounts for exposure measurement error using the PPV and NPV. ResultsThe Quebec and Boston cohorts have similar sizes with 1197 and 2346 patients, but the UK was bigger at 41370 patients. In Quebec's data, there were important differences in PPV and NPV by class of oral-hypoglycemics with PPVs for Biguanides at 0.81, Sulphonylureas at 0.65, and others at 0.50. The pattern for NPV differed with the same classes having respectively values of 0.56, 0.97, and 0.99. Estimates from the naïve model are typical of similar analysis but compared to their correction, they were generally overprecise and biased towards the null. The adjusted estimated were adequately representing the increased uncertainty with hazard ratios for Sulphonylureas going from 1.72 (1.22, 2.41) to 3.19 (1.36, 5.93), and from 1.09 (0.86, 1.39) to 1.05 (0.45, 2.16) for no drugs Conclusion/ImplicationsBayesian adjustment for measurement error allowed us to use linked data to regenerate uncertainty and to correct the bias in our risk estimates. Our approach was impacted by the observed low predictive value of prescribing, by reduced transportability of our PPV and NPV estimates, and other sources of bias.
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 machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.042 | 0.142 |
| Meta-epidemiology (narrow) | 0.001 | 0.002 |
| Meta-epidemiology (broad) | 0.001 | 0.003 |
| Bibliometrics | 0.003 | 0.004 |
| Science and technology studies | 0.001 | 0.002 |
| Scholarly communication | 0.003 | 0.003 |
| Open science | 0.003 | 0.003 |
| Research integrity | 0.003 | 0.004 |
| Insufficient payload (model declined to judge) | 0.009 | 0.002 |
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 source (direct Gemma or distilled Codex), 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".