Lung Cancer Mortality and Fiber Exposures among North Carolina Asbestos Textile Workers
Dana Loomis, John M. Dement, Susanne H. Wolf, David B. Richardson
SUPPLEMENTARY MATERIAL
METHODS
Data Analysis
Computation of SMRs when race is unknown
Analysis of SMRs requires complete information on race. One approach to handling missing race information is to calculate SMRs afterexcluding workers with unknown race. This can lead to bias, however, because death certificates can be used to verify race, so workers of unknown race are likely to be alive or lost to follow-up. Another approach, used in this study, is to impute the race of those with missing information.
The reference death rates used to compute SMRs required race to be classified as “white” or“non-white” for these analyses. However, all of the workers of known racein the study cohort were reported to be either white or black; other races were rare in the area near the plants when they were in operation, so it is likely that virtually all of the workers classified as nonwhite were black.
Race was missing for 1096 members of the cohort (19%). Based on the observation that historical hiring patterns in Southern textile mills were related to race and sex and changed over time, we used logistic regression to estimate the probability of being nonwhite as a function of year, plant, department and age at hire for men and of year, department and age at hire for women. Estimated values from fitting from these sex-specific models were used to classify individual race: workers with >50% estimated probability of being nonwhite were classified as nonwhite for the analyses reported here. Approximately 80% of workers whose race was known were correctly classified by this method, and the distribution of known and imputed race was similar. Alternative classifications using estimated probabilities of >25% and >75% to code workers as nonwhite were also investigated.
Sensitivity analyses were conducted to examine the impact on the SMR of imputing race and of excluding workers with unknown race by repeating analyses for all causes, all cancer and lung cancer.
Exposure-response models
Poisson regression was used to assess exposure-response within the cohort. The Poisson regression model has the general form: ln(λ)= α + βX + ΣγZ, where λ is the rate of mortality, X is the level of exposure, β is a regression coefficient describing the relationship of X to the log mortality rate, γZ is a vector of other explanatory variables and their corresponding regression coefficients, and α is the baseline mortality rate. The estimated rate ratio (RR) for the effect of any increment in exposure X1-X0 is obtained as exp[β(X1-X0)] and 95% (CI) are estimated from the standard error of β using a normal approximation.
Models for the relationship of each outcome to known demographic and occupational predictors of mortality were developed initially without considering fiber exposures. At this stage, model fit measured by Akaike’s Information Criterion (AIC) (1) and linearity of log RRs were considered in selecting variables and model form. Evidence of confounding was assessed by adding unlagged fiber exposure to the preceding models and examining the change in the coefficient for fibers associated with deleting each covariate. Age was assessed first, in both continuous and categorical form. Plots of log RRs obtained by modeling age both in categorical form and as a spline function suggested sublinear effects of younger and older ages, so a continuous linear age variable was not justified. Five and 10 year age categories were evaluated, but due to small numbers of deaths in the youngest and oldest categories, ages below 60 and above 80 were collapsed. An age variable with 4 categories (<60, 60-69, 70-79, 80+ years) provided the best compromise of fit and model form. Race, sex and calendar decade were included along with age in subsequent models. Race was coded as white, and other or unknown for internal analyses, obviating the need for imputation when race was missing. Other covariates evaluated for inclusion included length of follow-up, birth cohort, active work status, and years since termination. Models including birth cohort in addition to age, race, sex and calendar time were judged to provide the best fit for lung cancerafter considering AIC and the number of parameters. The preferred model for asbestosis was similar, but did not include birth cohort.
Confounding was assessed by adding unlagged fiber exposure to the previous models and examining the change in the coefficient for fibers associated with deleting each covariate. For lung cancer, race, calendar time and birth cohort changed the coefficient for exposureby approximately 10% or more and were selected for inclusion. To assess the effectiveness of the 4-category age variable in controlling age-related confounding, additional models were fit in which age was entered as a spline function, allowing the effect of age to be nonlinear (2). The coefficients for fiber exposure in these models and in models with categorical age were nearly identical. For asbestosis, raceand calendar time had weak confounding effects (<10% change in the coefficient for fibers), but were retained for their contribution to model fit and for consistency with lung cancer.
RESULTS
Additional results are presented in Tables S1-S4.
Table S1. Standardized mortality ratios for all workers employed in 4 North Carolina asbestos textile plants, 1950-2003. Expected deaths estimated from United States mortality rates.
White Males / All Other Males / White Females / All Other FemalesCause / O* / SMR* / 95% CI* / O / SMR / 95% CI / O / SMR / 95% CI / O / SMR / 95% CI
All causes / 1208 / 1.44 / 1.36-1.52 / 767 / 1.52 / 1.41-1.63 / 543 / 1.49 / 1.37-1.62 / 65 / 1.41 / 1.09-1.80
Tuberculosis / 2 / 1.05 / 0.13-2.84 / 1 / 0.21 / 0.01-1.20 / 0 / 0.00 / 0.00-7.26 / 1 / 4.91 / 0.12-27.4
All cancers / 303 / 1.42 / 1.27-1.59 / 179 / 1.60 / 1.37-1.85 / 148 / 1.26 / 1.06-1.48 / 12 / 1.04 / 0.54-1.82
Buccal cavity & pharynx / 4 / 0.77 / 0.21-1.97 / 10 / 2.39 / 1.15-4.40 / 2 / 1.47 / 0.18-5.30 / 0 / 0.00 / 0.00-21.8
Digestive & peritoneum / 44 / 0.84 / 0.61-1.13 / 35 / 1.10 / 0.76-1.52 / 28 / 1.16 / 0.77-1.68 / 2 / 0.73 / 0.09-2.65
Peritoneum, other & unspec. / 0 / 0.00 / 0.00-4.67 / 1 / 2.74 / 0.07-15.3 / 0 / 0.00 / 0.00-7.45 / 0 / 0.00 / 0.00-87.1
Respiratory / 164 / 2.09 / 1.78-2.43 / 74 / 1.83 / 1.43-2.29 / 48 / 1.78 / 1.31-2.35 / 3 / 1.48 / 0.30-4.32
Larynx / 3 / 1.12 / 0.23-3.29 / 3 / 1.44 / 0.30-4.22 / 0 / 0.00 / 0.00-9.18 / 0 / 0.00 / 0.00-63.4
Trachea, bronchus and lung / 157 / 2.09 / 1.77-2.44 / 71 / 1.86 / 1.46-2.35 / 46 / 1.74 / 1.28-2.32 / 3 / 1.54 / 0.32-4.50
Pleura / 3 / 13.5 / 2.78-39.4 / 0 / 0.00 / 0.00-68.3 / 1 / 23.3 / 0.59-130 / 0 / 0.00 / 0.00-1540
Breast / 0 / 0.00 / 0.00-13.8 / 1 / 6.12 / 0.15-34.1 / 20 / 0.85 / 0.52-1.31 / 2 / 0.78 / 0.09-2.81
Female genital organs / - / - / 17 / 1.19 / 0.69-1.91 / 2 / 1.21 / 0.15-4.37
Male genital organs / 22 / 1.40 / 0.87-2.11 / 18 / 1.51 / 0.89-2.39 / - / -
Urinary organs / 13 / 1.19 / 0.63-2.03 / 7 / 1.94 / 0.78-3.99 / 1 / 0.32 / 0.01-1.77 / 0 / 0.00 / 0.00-13.7
Other and unspecified sites / 36 / 1.26 / 0.88-1.74 / 15 / 1.30 / 0.73-2.14 / 23 / 1.65 / 1.04-2.47 / 1 / 0.82 / 0.02-4.55
Mesothelioma / 4 / 13.90 / 3.79-35.6 / 0 / 0.00 / 0.00-114 / 0 / 0.00 / 0.00-83.8 / 0 / 0.00 / 0.00-1830
Lymphatic and hematopoietic / 20 / 0.95 / 0.58-1.46 / 19 / 2.37 / 1.43-3.70 / 9 / 0.88 / 0.40-1.67 / 2 / 2.30 / 0.28-8.32
Benign & unspecified / 3 / 1.14 / 0.24-3.34 / 1 / 0.75 / 0.02-4.18 / 2 / 1.18 / 0.14-4.13 / 0 / 0.00 / 0.00-17.6
Diabetes mellitus / 20 / 1.24 / 0.76-1.92 / 11 / 0.96 / 0.48-1.71 / 11 / 1.07 / 0.54-1.92 / 4 / 2.00 / 0.54-5.11
Diseases of blood & blood-forming organs / 2 / 0.75 / 0.09-2.71 / 1 / 0.54 / 0.01-3.00 / 2 / 1.30 / 0.16-4.68 / 0 / 0.00 / 0.00-12.3
Mental & psychiatric disorders / 18 / 2.33 / 1.38-3.69 / 20 / 2.32 / 1.41-3.58 / 8 / 2.16 / 0.93-4.25 / 2 / 3.36 / 0.41-12.1
Nervous system disorders / 19 / 1.54 / 0.93-2.40 / 11 / 2.00 / 1.00-3.59 / 11 / 1.33 / 0.66-2.38 / 1 / 1.52 / 0.04-8.45
Diseases of the heart / 379 / 1.27 / 1.15-1.40 / 194 / 1.35 / 1.17-1.55 / 143 / 1.43 / 1.21-1.69 / 14 / 1.16 / 0.63-1.95
Other circulatory diseases / 79 / 1.22 / 0.96-1.52 / 96 / 1.97 / 1.59-2.40 / 47 / 1.30 / 0.95-1.72 / 7 / 1.27 / 0.51-2.62
Respiratory diseases / 125 / 2.06 / 1.72-2.46 / 43 / 1.42 / 1.03-1.91 / 64 / 2.19 / 1.62-2.80 / 5 / 2.13 / 0.69-4.97
Chronic obstructive pulm. disease / 50 / 1.64 / 1.22-2.16 / 12 / 1.34 / 0.69-2.35 / 29 / 1.92 / 1.29-2.76 / 0 / 0.00 / 0.00-7.00
Pneumoconiosis / 41 / 3.91 / 2.80-5.30 / 12 / 2.13 / 1.10-3.72 / 18 / 4.13 / 2.44-6.52 / 2 / 4.13 / 0.50-14.9
Digestive diseases / 56 / 1.46 / 1.10-1.89 / 40 / 1.55 / 1.10-2.10 / 32 / 1.99 / 1.36-2.80 / 3 / 1.31 / 0.27-3.82
Genito-urinary diseases / 14 / 1.29 / 0.70-2.16 / 13 / 1.16 / 0.62-1.98 / 14 / 2.21 / 1.21-3.71 / 2 / 2.21 / 0.46-6.45
Diseases of the skin / 0 / 0.00 / 0.00-5.81 / 1 / 1.18 / 0.03-6.58 / 2 / 3.96 / 0.48-14.3 / 0 / 0.00 / 0.00-26.6
Musculoskeletal & connective tissues diseases / 2 / 1.11 / 0.13-4.03 / 0 / 0.00 / 0.00-3.17 / 2 / 0.91 / 0.11-3.28 / 2 / 4.80 / 0.58-17.3
Symptoms & ill-defined conditions / 11 / 1.30 / 0.65-2.32 / 4 / 0.36 / 0.10-0.91 / 1 / 0.31 / 0.01-1.71 / 0 / 0.00 / 0.00-4.50
Accidents / 80 / 1.44 / 1.14-1.79 / 47 / 1.23 / 0.90-1.64 / 22 / 1.79 / 1.12-2.71 / 1 / 0.51 / 0.01-2.83
Violence / 53 / 1.93 / 1.45-2.53 / 39 / 1.30 / 0.92-1.78 / 4 / 0.70 / 0.19-1.78 / 4 / 3.06 / 0.83-7.84
Other & unspecified causes / 42 / 2.30 / 1.66-3.11 / 66 / 3.54 / 2.73-4.50 / 30 / 3.17 / 2.14-4.52 / 6 / 2.44 / 0.90-5.32
* O, oserved number of deaths; SMR, standardized mortality ratio (national reference rates); CI, confidence interval.
Table S2. Standardized mortality ratios for selected causes of death for workers in 4 North Carolina asbestos textile plants with at least 20 years latency, 1950-2003.
Cause / O* / SMR* / 95% CI*All causes / 2015 / 1.58 / 1.51-1.65
All cancers / 548 / 1.52 / 1.40-1.66
Stomach / 6 / 0.53 / 0.20-1.16
Intestine / 30 / 0.98 / 0.66-1.40
Rectum / 10 / 1.63 / 0.78-3.00
Peritoneum / 1 / 0.79 / 0.02-4.38
Larynx / 5 / 1.24 / 0.40-2.89
Trachea, bronchus and lung / 249 / 2.14 / 1.88-2.42
Pleura / 4 / 15.42 / 4.20-39.48
Mesothelioma / 4 / 10.92 / 2.98-27.96
Pneumoconiosis / 63 / 3.84 / 2.95-4.92
*O, observed number of deaths, SMR, standardized mortality ratio, CI, confidence interval.
Table S3. Rate ratios for lung cancer by category of employment duration among production workers at 3 North Carolina asbestos-textile plants, 1950-2003.
Duration (years) / Deaths / Person-years / RR*† / 95% CI* / LR (4 df)* / AIC*<1 / 85 / 71,976 / 1.00 / 8.32 / 2350.4
1-<5 / 49 / 33,038 / 1.12 / 0.79 / 1.60
5-<10 / 17 / 10,445 / 1.07 / 0.63 / 1.80
10-<20 / 17 / 6,040 / 1.67 / 0.97 / 2.85
≥20 / 13 / 2,530 / 2.35 / 1.27 / 4.33
*RR, rate ratio; CI, confidence interval; LR (4), likelihood ratio, 4 d.f.; AIC, Akaike information criterion.
†Adjusted via Poisson regression for age (<60, 60-69, 70-79 or ≥80 years), sex, race (white or other/unknown), decade of follow-up, and birth cohort (<1920, 1920-1939, and ≥1940).
Table S4a. Rate ratios per f-y/ml for lung cancer by average fiber exposure and cumulative + average fiber exposure by lag interval among production workers at 3 North Carolina asbestos-textile plants, 1950-2003.
Lag (years) / RR*† / 95% CI* / LR (1 df)* / AIC*0 / 1.0063 / 0.9982 / 1.0145 / 2.09 / 2350.6
10 / 1.0067 / 0.9988 / 1.0147 / 2.44 / 2350.3
20 / 1.0071 / 0.9996 / 1.0147 / 3.01 / 2349.7
30 / 1.0078 / 1.0010 / 1.0147 / 4.4 / 2348.3
*RR, rate ratio; CI confidence interval; LR (1 df), likelihood ratio for addition of exposure term with 1 degree of freedom; AIC, Akaike information criterion.
†Adjusted via Poisson regression for age (<60, 60-69, 70-79 or ≥80 years), sex, race (white or other/unknown), decade of follow-up, and birth cohort (<1920, 1920-1939, and ≥1940).
Table S4b. Rate ratios per f-y/ml for lung cancer by cumulative + average fiber exposure by lag interval among production workers at 3 North Carolina asbestos-textile plants, 1950-2003.
Lag (years) / RR*† cumulative / 95% CI* / RR† average / 95% CI / LR* / AIC*0 / 1.000934 / 1.000324 / 1.001544 / 1.001298 / 1.000324 / 1.001544 / 0.07 / 2345.5
10 / 1.000966 / 1.000338 / 1.001593 / 1.001386 / 0.992193 / 1.010665 / 0.08 / 2345.0
20 / 1.001013 / 1.000309 / 1.001718 / 1.001467 / 0.992346 / 1.010672 / 0.10 / 2345.2
30 / 1.001120 / 0.999347 / 1.002496 / 1.002257 / 0.993397 / 1.011196 / 0.24 / 2345.3
*RR, rate ratio; CI confidence interval; LR (1 df), likelihood ratio for addition of average exposure term with 1 degree of freedom; AIC, Akaike information criterion.
†Adjusted via Poisson regression for age (<60, 60-69, 70-79 or ≥80 years), sex, race (white or other/unknown), decade of follow-up, and birth cohort (<1920, 1920-1939, and ≥1940).
SUPPLEMENTAL REFERENCES
1. Akaike H. Information theory and an extension of the maximum likelihood principle. In: Petrov BN, Czaki F, eds. Second International Symposium on Information Theory. Budapest: Akademiai Kiado 1973:267-281.
2. Eilers PHC, Marx BD. Flexible smoothing with B-splines and penalities. Statistical Science 1996;11:89-121.