How (Not) to Validate Race-Neutral Lung Function Reference Equations
Bibliographic record
Abstract
Abstract RATIONALE: Prediction models and reference equations should be extensively validated in the target population before implementation. Lung function reference equations are designed to estimate healthy lung function using data from healthy volunteers. However, in patients with respiratory diseases, “healthy lung function” is an unobservable counterfactual, presenting a validation challenge: how to assess validity of lung function reference equations without a definitive ground truth label? Several studies have attempted to assess the construct validity of reference equations by measuring their discriminative accuracy for clinical outcomes such as mortality. However, when reference equations exclude some variables due to fairness considerations (e.g., race), their association with outcomes might remain spuriously strong due to the confounding effect of such variables. As such, prediction accuracy per se might become a misleading indicator of validity. METHODS: We analyzed data from the National Health and Nutrition Examination Survey (NHANES) III and NHANES 2007-2012, with mortality follow-up through December 31, 2019. We included participants with a history of smoking, diagnosis of a respiratory condition, or respiratory symptoms. We used area-under-the-curve (AUC) to compare discriminative accuracy in predicting multiple clinical outcomes using FEV1 Z-scores from three alternative reference equations: race-averaged GLI-Global, race-specific GLI-2012, and a naïve equation that assigned the same predicted FEV1 to all individuals. Evaluated clinical outcomes included 10-, 20-, and 30-year all-cause mortality, medical visits for wheezing, overnight hospital admissions within the past year, and concurrent respiratory symptoms. RESULTS: FEV1 Z-scores from race-averaged GLI-Global were non-inferior to those from race-specific GLI-2012 in predicting clinical outcomes (Table 1). Similar results in the literature have been used to argue that race-averaged GLI-Global performs as well as GLI-2012 even after removing race as an explicit predictor. However, FEV1 Z-scores based on the clearly flawed naïve reference equation showed significantly better accuracy in predicting death and comparable accuracy for healthcare utilization and symptoms among respiratory patients in NHANES, compared to GLI-2012 and GLI-Global (Table 1). Based on statistical assessment of prediction accuracy, one should advocate for using the naïve equation, which is clearly flawed. CONCLUSIONS: Using prediction accuracy for clinical outcomes as a proxy for validating lung function reference equations can be misleading. We urge a careful reassessment of the evaluation methods for lung-function reference equations that objectively reconcile fairness criteria and prediction accuracy. Table 1 Discriminative Accuracy of Reference-Adjusted FEV1 Z-scores for Predicting Clinical Outcomes in Participants with or at Risk of Respiratory Diseases
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 | 0.325 | 0.616 |
| Meta-epidemiology (narrow) | 0.002 | 0.001 |
| Meta-epidemiology (broad) | 0.002 | 0.003 |
| Bibliometrics | 0.001 | 0.002 |
| Science and technology studies | 0.002 | 0.003 |
| Scholarly communication | 0.005 | 0.005 |
| Open science | 0.005 | 0.003 |
| Research integrity | 0.004 | 0.005 |
| Insufficient payload (model declined to judge) | 0.003 | 0.003 |
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; the direct Gemma label and the distilled Codex classifier agree on what is shown here.
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".