Re: Lung Cancer Risk Among Female Textile Workers Exposed to Endotoxin

We previously reported ( 1 ) an inverse exposure–response relation between cumulative exposure to endotoxin and lung cancer risk that was based on findings from a nested case–cohort study of female textile workers in Shanghai. The inverse risk trends became increasingly pronounced as recent exposures were discounted (lagged) by 5, 10, 15, and 20 years [see table 2 ( 1 )]. With a 20-year lag of cumulative exposure, increased endotoxin exposure was associated with decreased risk of lung cancer (when the highest quintile was compared with the unexposed worker reference group, relative risk = 0.60, 95% confidence interval = 0.43 to 0.83, P for trend = .002). We interpreted this pattern of results as suggestive of a possible protective effect of endotoxin exposure that acted preferentially at early stages of lung carcinogenesis.

Recently, during reanalyses of these data, we discovered an inadvertent error in the calculations of our lagged exposure estimates. For case patients, cumulative exposure estimates were correctly lagged from diagnosis date; however, for noncase subcohort workers, exposures were incorrectly lagged from final follow-up (the earlier of study end on December 31, 1998, or date of death). Estimates should have been calculated for subcohort members still at risk of incident lung cancer as of the diagnosis dates for case patients within individual risk sets. We have reanalyzed the data reported in the original study by use of methods and Statistical Analysis Software macros that were developed by Langholz and Jiao ( 2 ) after publication of that study ( 1 ). As noted by Langholz and Jiao, practical data analysis methods for handling computational challenges inherent to the stratified case–cohort design with time-dependent covariates had not previously been available. For our reanalysis, we organized the data into risk sets, each composed of a case patient and all subcohort women still at risk at the time the case patient was diagnosed. We then computed exposure estimates for each risk set up to and lagged from the diagnosis time of the case patient, which defined the risk set. We computed confidence intervals for relative risk estimates (as hazard ratios) by incorporating birth-year stratum-specific sampling weights for the subcohort into the variance estimation. The corrected results are shown in Table 1 .

Table 1.

Lung cancer incidence by cumulative exposure to endotoxin ^*

	Cumulative endotoxin exposure †
Exposure lag time	Unexposed	≤1283	>1283 to ≤2056	>2056 to ≤2801	>2801 to ≤4075	>4075	P trend ‡
None
HR (95% CI)	1.00 (ref)	1.06 (0.79 to 1.41)	0.98 (0.74 to 1.31)	0.79 (0.59 to 1.07)	0.88 (0.66 to 1.17)	0.71 (0.52 to 0.95)	.004
No. of case patients/No. of control subjects	186/913	88/425	89/424	75/425	90/424	74/424
5-y lag
HR (95% CI)	1.00 (ref)	1.08 (0.82 to 1.44)	0.95 (0.72 to 1.28)	0.79 (0.59 to 1.07)	0.91 (0.68 to 1.20)	0.69 (0.51 to 0.93)	.003
No. of case patients/No. of control subjects	186/916	94/456	86/424	74/407	91/416	71/416
10-y lag
HR (95% CI)	1.00 (ref)	1.07 (0.81 to 1.41)	0.93 (0.70 to 1.25)	0.78 (0.58 to 1.06)	0.91 (0.68 to 1.21)	0.70 (0.51 to 0.95)	.005
No. of case patients/No. of control subjects	188/923	99/507	84/422	72/387	89/400	70/396
15-y lag
HR (95% CI)	1.00 (ref)	1.07 (0.82 to 1.40)	0.88 (0.66 to 1.18)	0.85 (0.63 to 1.14)	0.87 (0.64 to 1.17)	0.72 (0.52 to 0.98)	.008
No. of case patients/No. of control subjects	193/965	106/555	83/402	77/366	76/372	67/375
20-y lag
HR (95% CI)	1.00 (ref)	1.00 (0.78 to 1.29)	0.75 (0.56 to 1.00)	0.87 (0.64 to 1.17)	0.88 (0.63 to 1.23)	0.74 (0.53 to 1.02)	.03
No. of case patients/No. of control subjects	208/1090	122/541	81/410	76/343	56/312	59/339

^*Data were adjusted for age at baseline (continuous) and smoking status (never [ref], former, or current). Analysis was based on cumulative exposure to endotoxin at each risk set time, excluding subjects who ever worked as machinists (n = 114), worked with wool (n = 17), or worked in sanitation jobs in production (n = 44). CI = confidence interval; HR = hazard ratio; ref = referent.

^†Exposure categories represent quintiles of exposure among noncase patients with no exposure lag. Cumulative endotoxin exposure is expressed as (endotoxin units/cubic meter) × years.

^‡Two-sided Wald test for the hypothesis that the slope equals zero of hazard trend (treated as a continuous variable), including unexposed category. Category scores were based on the median of category exposure range.

Additionally, we have identified an error with the original trend test results. In the original study ( 1 ), the tests were based on the geometric mean of the range of exposures in each category rather than on the median exposure values, as reported. This error has been corrected so that trend tests are now based on the exposure category median exposure values that are specific for the various lag intervals. The test for trend at baseline (time 0 − lag time) still yielded a statistically significant result [ P for trend = .004, rather than P for trend = .02, as reported previously ( 1 )].

The new findings continued to indicate a dose-related inverse lung cancer risk that was associated with cumulative endotoxin exposure, which is consistent with the previous hypothesis ( 1 ) and supported by other epidemiological studies of endotoxin-exposed occupational groups ( 3 ). However, a possible anticarcinogenic effect at early stages of lung cancer pathogenesis was not evident. Future studies of cancer incidence in the Shanghai female textile worker cohort should help to clarify the relation between endotoxin and lung cancer.