Cumulative Retrospective Exposure Assessment (REA) as a predictor of amphibole asbestos lung burden: validation procedures and results for industrial hygiene and pathology estimates

Figures & data

Table 1. Variable definitions.

Variable (s)	Description	Type	Units
A, B, C, D	The A, B, C, and D variables refer to sets of individual cumulative exposure estimates for each of the four raters that were logarithmically-transformed for the 328 cases.	Continuous	Log10 (f/cc-years)
EXP	EXP refers to the population of 328 arithmetically averaged, log-transformed exposure estimates from each rater. EXP = (A + B + C + D)/4 for each of the 328 cases for which all raters submitted an estimate.	Continuous	Log10 (f/cc-years)
AB/g	Log10 – transformed values of the set of linearly averaged light microscopy laboratory determinations of the number of asbestos bodies per wet gram of lung tissue for each case for each of 327 cases.	Continuous	Log10 (ab/g)
SEM AB/g	Log10 – transformed values of the set of linearly averaged SEM laboratory determinations of the number of asbestos bodies per wet gram of lung tissue. Analytical data was available for 277 out of the 327 cases.	Continuous	Log10 (SEM ab/g)
F/g	Log10 – transformed values of the set of linearly averaged SEM laboratory determinations of the number of asbestos fibers longer than 5 µ per wet gram of lung tissue for each case for 308 overlapping cases with the 327 cases.	Continuous	Log10 (f/g)
LAB	A combination variable of AB/g, SEM AB/g, and F/g that was transformed into the same units as AB/g. More on this variable is found in the discussion, below, and in Table 2.	Continuous	Log10 (ab/g)
ET	The data set was further reduced from the original categorization into 11 exposure types (or categories) for analysis of covariance (ANCOVA) (Rosner, 2006). The main asbestos exposure types were based on occupation, but other categories such as household exposure and the type of asbestos product from which exposures occurred were also utilized. These exposure type (ET) categories are defined in Figure 4, along with numbers of associated cases.	Class	Single Integers (Dummy Variable)
FT	For 308 of the 327 cases, asbestos fiber-type information associated with lung fiber burden analysis was available. Thus, 308 cases were classified into one of four categories. These four categories are defined in Figure 4, along with numbers of associated cases for each category, which allowed correlation by ANCOVA.	Class	Single Integers (Dummy Variable)

Table 2. Correlation of AB/g as a function of (SEM AB/g) and F/g. Coefficients were utilized to calculate LAB, a combined function of AB/g, SEM AB/g and F/g, all in the same units of AB/g.

Independent Variable	N (all available lab analysis) from original data set	Correlation Coefficient (r)	Slope (Standard Error)	Intercept (Standard Error)
SEM AB/g	277	0.91	1.17 (0.031)	−1.21 (0.13)
F/g	341	0.87	1.29 (0.040)	−2.72 (0.18)

Figure 1. Correlation of a Rater’s Exposure Estimates with EXP.

Figure 2. Residual error histogram with probit plot and quantile data for regression of a sample rater’s exposure estimates with EXP.

Table 3. Intraclass ICC for asbestos bodies and fiber burden analysis.

Data Set	Corresponding variable	Number of cases	Cases with replicates	Replicates per Case	ICC
Asbestos Bodies in lung Tissue: Light Microscopy	AB/g	363	166	≥2	0.91
Asbestos Bodies in lung Tissue: SEM	SEM AB/g	1249*	54	≥2	0.90
Free Asbestos Fibers in Lung Tissue >5 µ by SEM	F/g	1249*	67	≥2	0.88

*Additional data were obtained from the pathologist for the purpose of calculating the laboratory ICCs.

Table 4. Correlation of the logarithm of various lung asbestos body and fiber burden determinations with average log of rater estimates (EXP).

Regression Dependent Variable	Overall Significance	r	Slope
AB/g	p < 0.0001	0.65	1.22 (Significantly different from 1, p = 0.004)
SEM AB/g	p < 0.0001	0.65	0.87 (Barely significantly different from 1, p = 0.040)
F/g	p < 0.0001	0.66	0.81 (Significantly different from 1, p = 0.0004)
LAB (normalized to AB/g units)	p < 0.0001	0.67	1.13 (Barely significantly different from 1, p = 0.045)

Figure 3. Correlation of LAB with EXP.

Figure 4. ANCOVA correlation of LAB as function of the continuous variable EXP and the class variable ET, no interaction term. Order of parallel lines from top: 1. Shipyard Insulator (n = 10), 2. Power Plant Worker (n = 7), 3. Asbestos Insulator (n = 41), 4. Shipyard Worker (n = 89), 5. Asbestos Manufacturing Except Friction Products (n = 5), 6. Secondary Household Exposures (n = 18), 7. Miscellaneous and Mixed Trade (n = 72), 8. Construction, Piping, and Boiler Trades Except Shipyard (n = 63), 9. Building Occupant (n = 5), 10. Refinery Workers (n = 6), and 11. Automobile and Friction Product Related (n = 11). Colors of data points correspond to the colors of the parallel lines.

Table 5. Standard errors associated with the data presented in Figure 4.

Exposure Type (ET) or (Parameter)	Intercept or (Slope)	Standard error	n
(Overall Slope)	(0.957)	0.098
Shipyard Insulator	3.07	0.44	10
Power Plant Worker	2.40	0.46	7
Asbestos Insulator	2.21	0.36	41
Shipyard Worker	2.17	0.30	89
Asbestos Manufacturing,  Except Friction Products	1.98	0.51	5
Secondary Household Exposures	1.95	0.36	18
Miscellaneous and Mixed Trade	1.83	0.30	72
Construction, Piping, and Boiler Trades,  Except Shipyard	1.82	0.31	63
Building Occupant	1.64	0.51	5
Refinery Workers	1.19	0.48	6
Automobile and Friction Product Related	1.07	0.28	11

Figure 5. LAB as function of EXP and fiber for four fiber type classes (with interaction). 1. Commercial amphibole asbestos, majority fiber type, no chrysotile (+), n = 216 2. Commercial amphibole asbestos, majority fiber type, chrysotile detected (x), n = 30. 3. Amosite and/or crocidolite detected, but not majority asbestos fiber-type (o), n = 35. 4. No crocidolite or amosite detected (Δ), n = 40.

Figure 6. Comparison of asbestos body and asbestos fiber burden ANCOVA correlations.

Table 6. Comparison of ANCOVA correlation with EXP and EXP*FT for each of the dependent variables, LAB, AB/g, and F/g.

		Slope 1 (Std Error)	Slope 2 (Std Error)	Slope 3 (Std Error)	Slope 4 (Std Error)
Dependent		Diff from 1? (p)	Diff from 1? (p)	Diff from 1? (p)	Diff from 1? (p)
Variable	r	Diff from 0? (p)	Diff from 0? (p)	Diff from 0? (p)	Diff from 0? (p)
LAB	0.75	1.11 (0.085)	0.68 (0.20)	0.34 (0.25)	0.27 (0.21)
		No (0.21)	No (0.11)	Yes (0.009)	Yes (0.0006)
		Yes (<0.0001)	Yes (0.0007)	No (0.18)	No (0.20)
AB/g	0.75	1.06 (0.094)	0.66 (0.22)	0.42 (0.28)	0.54 (0.23)
		No (0.49)	No (0.12)	Yes (0.036)	Yes (0.046)
		Yes (<0.0001)	Yes (0.0027)	No (0.13)	Yes (0.022)
F/g	0.70	0.88 (0.069)	0.58 (0.16)	0.20 (0.20)	0.20 (0.20)
		No (0.095)	Yes (0.0086)	Yes (<0.0001)	Yes (<0.0001)
		Yes (<0.0001)	Yes (0.0003)	No (0.32)	No (0.31)

Figure 7. Residual error histograms for F/g and AB/g as a function of EXP, FT and EXP*FT.

Rosner B. (2006). Fundamentals of biostatistics. 6th ed. Duxbury: Thomson

 Google Scholar