Send letter to journal:
Re: Publication by Kraus et al., December 2002

Dear Editor

The paper by Kraus and colleagues[1] in the December 2002 issue of Occupational and Environmental Medicine addresses the important issue of the relationship of personal exposure to possible effect in an industrial setting. After reading this paper I would like to make some comments with regard to the design, conduct and statistical analysis of the study.

In an area where exposure assessment and confounding effects are so important, it is interesting to see that seemingly important associations are being drawn with so little information.

Kraus and colleagues state the goal of the paper as “To correlate the prevalence of respiratory tract symptoms and diseases with dust and fibre exposure in the soft tissue industry in Germany” rather than to investigate whether such a relationship actually exists.

The authors state that: “In every company pure cellulose was the basis of the production process. Recycled paper was not a relevant component of the products.”, however, in the introduction they site many references of toxicological studies that used insulation cellulose (Isofloc, which is ground up used newspapers sprayed with boric acid) and other not pure cellulose products. The contaminants in used newspapers with boric acid would be expected to produce a radically different toxicological response than inhaled pure cellulose. Only studies that used pure cellulose of similar source and composition as used in the plants in this study group should be sited and for such studies the route of exposure and doses should be provided.

Sampling: From the text, it appears that in this study, data was obtained on dust measurements in various places in the 7 chosen German soft tissue companies. These were then compared to symptoms of persons working in these companies.

The authors state that “According to the employers’ information, we interviewed all persons from the different workshops employed during the time period 1996–98.” They then explain that “The data concerning dust measurements were gathered from 1991 to 1997.” Thus, it appears that the dust measurements were not directly related to the individuals with the symptoms nor were they even necessarily coinciding with the presence of the symptoms. There is no mention of how and where within each plant the exposures were determined, of whether the dust levels were similar from 1991 to 1997 and whether there was any personnel monitoring to determine exposures. However, the authors do state that “Because of differences in the exposure intensity between the seven companies a cumulative exposure index was generated for describing the individual exposure.” And as such mean exposure levels were used for comparison.

The confidence that can be placed in conclusions drawn from samples depends in part on sample size. Small samples can be unrepresentative just by chance, however, the scope for chance errors can be quantified statistically. Given this importance, why haven’t the authors stated the sample sizes in each of the groups used in the statistical analyses?

More problematic are the errors that arise from the method by which the sample is chosen. In situations with high dust levels there is more likelihood of sampling which would be atypical of the dust levels more generally. Such systematic errors cannot usually be measured, and assessment therefore becomes a matter for subjective judgment. Systematic sampling errors can be avoided by use of a random selection process in which each working condition has a known (non-zero) probability of being included in the study sample. However, this requires an enumeration of all working conditions in each plant over the entire sampling period (1991 – 1998).

Not only was there no mention of such a sampling procedure being performed in this study but any sensitivity to specific plant conditions has been lost through the use of mean values.

Group composition: The authors then gloss over the choice of controls which were persons from the “management department”. Certainly the working condition and environment of office workers is not comparable to that of maintenance workers, electricians and mechanics. Also of concern is the disproportionality in the controls compared to exposed in terms of number, sex and smoking history. No mention is made of age and how long the persons in each group worked in the plant. This information would permit the reader to assess whether the symptoms were associated with either age or length of employment. Also, the smoking histories are not comparable between the groups. From the means in Table 1, there is a positive association between exposure group and smoking which is never discussed in the paper. If smoking habits were asked “in a detailed way” as stated, why haven’t the data have been presented? As there appears to be a dose relationship with smoking and exposure group, the authors should have discussed in the introduction the possibility that smoking can also produce similar effects as reported in this paper and provide appropriate references.

The authors mention using an attenuation factor (“divided by four and multiplied by individual time of exposure”) under recommendation of technical experts. This implies that it was not possible to obtain any kind of relationships using the company individual monitoring data in relationship to the individuals working in each company. What is the rational and scientific basis for this? Without such information, the choice of cut points for the attenuation factor appears arbitrary and leads the reader to wonder whether the choice of cut points was determined iteratively in order to find an attenuation factor which produced correlations?

Statistical analysis: The authors use the Cochran-Armitage trend test without adjustment for other variables and even say “that the results are prone to being confounded”. Yet, they discuss at length the results from this test in Tables 3 and 4. Further, it is not at all clear that even this analysis was performed correctly. Hothorn and Bretz (2000),[2] explain that the Cochran-Armitage trend test is a test based upon a linear dose response relationship and that the test lacks power for other shapes. Most of the parameters shown in Tables 3 and 4 of the Kraus paper do not appear to have a linear dose-response.

More details of the results of the logistic regression analyses should have been presented to support the conclusions stated by the authors. At least an estimate of the percent variation in the data explained by each of the logistic regressions should be provided in the tables in order to assess the importance of each relationship. Grouped and individual R2 values are well documented in Statistical texts and can be calculated for each logistic regression in order to assess importance.

Results: The authors should have clearly state whether they used inhalable dust or respirable dust in each analysis. Respirable is probably more relevant, however, Table 2 indicates that there were only 24 respirable dust measurements. With 4 exposure categories (Controls, <25, 25–100, >100) and 7 companies this would amount to less than 1 respirable dust sample per category per company over an 8 year of monitoring period! It is more likely that they used the inhalable dust measurements for which there were 105 measurements. Even with the inhalable dust measurements, this would amount to less than one inhalable dust measurement every two years per category per company!

The authors state that “The mean sampling time was 1.97 hours (range 0.16–3.8).” This amounts to each category in each plant being monitored for mean of 7.4 hours over a working time of 14,400 hours. Even if all 105 inhalable dust measurements were performed in one work year (which they were not), each category in each plant would have been monitored for a mean of 7.4 hours over a working time of approximately 1800 hours/year.

Yet, the authors are suggesting that somehow these few samples are representative of what 441 workers were exposed to over an 8 year sampling period. Not only are the samples not chosen correctly but the small sample size assures that these measurements are not representative.

The need for statistical significance also seems to have escaped the authors when discussing the disease parameters. The data presented indicate that none of the disease parameters are significantly different from the controls, yet the authors state that “However, sinusitis, laryngitis, and chronic bronchitis (a disease parameter) showed increasing odds ratios with increasing cumulative dust exposure (table 5).” What is the validity of stating this when there is no statistical significance in the relationship?

Again, in the discussion, the authors state that “After adjustment for smoking habits, chronic bronchitis was no longer significantly associated with dust or fibre exposure. However, the highest odds ratio (1.57) in the subgroup with longest and highest exposure suggests some deleterious potential.“ This is very worrisome that the authors suggest that there is a ‘deleterious potential’ when in fact there is no statistical basis from the data (95% CI of 0.7 to 4.0) for making such a statement.

In Tables 7 and 8 the authors group the ‘Diseases’ as part of the ‘Symptoms’ no longer differentiating between Diseases and Symptoms.

They then go on to state that “For example, intensity and duration of exposure have an almost equal influence on the symptom “blocked nose”. The calculated ORs rise with increasing exposure intensity from 6.4 to 10.8 and with increasing exposure duration from 6.4 to 12.5.” Fortunately, in Table 7 they present the 95% confidence limits which are (shown in parenthesis) for these values 6.4 (1.2 to 49.7) and 10.8 (2.9 to 70.5) which suggest that 6.4 is not different from 10.8 statistically. Not surprising given the statements by Kraus and colleagues, Tables 7 and 8 present no measure of statistical significance for any of the parameters yet they state that “For symptoms of the upper airways, clear dose- response relations could be found with relation to cumulative exposure indices”. This is certainly not supported by the data presented in these tables.

Discussion: Again the authors cite references to articles not dealing with the same pure cellulose as used in the plants. If there are studies using the same pure cellulose, then a discussion of the route of exposure and the exposed dose compared to what is seen here should have been included.

The Ericsson reference and others are not quoted correctly from the publications. As an example, Ericsson states that “There was a dose- dependent increase of symptoms from the upper respiratory tract. However, coughing and coughing with phlegm were not found to be more common among persons with heavy exposure compared to those with low exposure to the dust“, indicating that the Ericsson results are not comparable to those in the current manuscript.

The statement by the authors that “As expected, because of the distribution of inhalable and respirable dust fraction, symptoms of the lower respiratory tract have a weaker association with exposure after adjustment for confounders (for example, cough, phlegm, dyspnoea, and exercise induced dyspnoea)” should from the data presented in the paper be changed to state that “As expected, due to the distribution of inhalable and respirable dust fraction, symptoms of the lower respiratory tract have no statistically significant association with exposure after adjustment for confounders (e.g. cough, phlegm, dyspnea, exercise induced dyspnea).”

The question of respirator use in the plants should have been investigated. If respirators were used even by some employees, then any association to the measured dust levels for those persons is no longer valid.

I hope that the critical readers of your journal will decide based upon the statistical significance of the data whether there is a basis for stating that a relationship exists between exposure and effect in the soft tissue industry.

References

(1) T Kraus, A Pfahlberg, O Gefeller, H J Raithel, Respiratory symptoms and diseases among workers in the soft tissue producing industry. Occup Environ Med 2002;59:830–835.

(2) Hothorn LA and Bretz F. Evaluation of Animal Carcinogenicity Studies: Cochran-Armitage Trend Test vs. Multiple Contrast Tests. Biometrics Journal 2000;42(5):553-567.