Application of Statistical Cross-Extrapolation Techniques to Derive Surrogate Acute Exposure Guideline Levels (AEGLs) Open Access

Chu, MyDzung Thi (2012)

Permanent URL:


Application of Statistical Cross-Extrapolation Techniques to Derive Surrogate Acute Exposure Guideline Levels (AEGLs)

AEGLs are comprehensively peer-reviewed health guidance values (HGVs) for assessing the risk of acute once-in-a-lifetime or rare exposures to hazardous inhalation chemicals. For each inhalation compound, up to fifteen AEGL values may be developed that address three health effects severity thresholds (AEGL-1: discomfort/reversible, AEGL-2: disabling/irreversible, AEGL-3: life threatening) at five exposure durations (1/6, 1/2, 1, 4, and 8 hours). Currently, only 74 compounds have Finalized AEGLs, while 187 are Interim and 12 are Proposed. Among these, 42% have unassigned AEGLs due to insufficient data or biological implausibility of estimates. Also as of November 2011, the AEGL Program no longer reviews new compounds. Therefore, a need for a rapid and cost-effective substitute for AEGL development is imminent. The aim of the present work was to develop an efficient method for the derivation of provisional AEGLs for inhalable hazardous compounds with unassigned AEGLs. Such method is plausible due to uniformity of procedures by which the AEGLs have been developed, and due to similarities in the physical-chemical characteristics of inhalable compounds.Qualitative and quantitative data for AEGLs were derived from the US Environmental Protection Agency's published technical support documents. Pearson correlation and Deming linear regression (DLR) analyses of the AEGL database were employed to develop a total of 105 unique univariate cross-extrapolation models for duration-and-threshold-specific AEGLs. 95% confidence and prediction intervals (CIs and PIs) of each model were constructed using bootstrap resampling. The most predictive DLR models were applied to compounds with unassigned AEGLs. Obtained estimates were externally validated using other available health guidance data, including occupational exposure limits (OELs). Model performance was also internally validated by comparing estimated and actual AEGLs for compounds with the full set of data. All Pearson correlation coefficients ( r) were greater than 0.88. Higher coefficients generally corresponded to cross-extrapolation models with narrower 95% PIs. The narrowest PIs, i.e. the most confident cross-extrapolation, were observed for pairs of AEGLs that were most similar in exposure duration and severity of health effects. Conversely, the widest PIs were obtained for functionally most distant AEGL pairs; however, even the worst estimates were within two orders of magnitude of the actual values. Comparison of estimated AEGLs to occupational HGVs suggested that numerically STELs and TWAs were more correlated with AEGL-1 and -2s at 4 h and 8 h. External validation of cross-extrapolated numbers against these occupational HGVs for a test set of 14 chemicals showed statistical identity at the 95% level for 8 of the 14 compounds. Our findings suggest that the DLR models are statistically valid and predictive of unassigned AEGL values for compounds in the database. Model performance is dependent on the severity threshold and exposure duration of the cross-extrapolated quantities. External validation using occupational HGVs shows that our cross-extrapolation estimates are sound. Yet, the uncovered relationships are not fully vetted. In the future,structure-activity, time-scaling, and the biological plausibility of AEGL predictions will be investigated.

Application of Statistical Cross-Extrapolation Techniques to Derive Surrogate Acute
Exposure Guideline Levels (AEGLs)

Smith College

Thesis Committee Chair:
P. Barry Ryan, PhD
Eugene Demchuk, PhD

A thesis submitted to the Faculty of the Rollins School of Public Health of Emory University
in partial fulfillment of the requirements for the degree of Master of Science in Public Health in Environmental Health-Epidemiology 2012

Table of Contents


i. Significance/Rationale
ii. Specific Aims
iii. Background and Literature review
iv. AEGL application in public health

i. Hypothesis
ii. Methods of data collection
iii. Methods of analysis and rationale

i. Descriptive statistics of AEGL database
ii. Model building
iii. Model analysis
iv Model selection
v. Model application
vi. Cross-validation with existing HGVs





i. Software
ii. Literature



About this Master's Thesis

Rights statement
  • Permission granted by the author to include this thesis or dissertation in this repository. All rights reserved by the author. Please contact the author for information regarding the reproduction and use of this thesis or dissertation.
Subfield / Discipline
  • English
Research field
Committee Chair / Thesis Advisor
Committee Members
Partnering Agencies
Last modified

Primary PDF

Supplemental Files