Fujishiro et al.1 recently published data on the association of job
demands and control with carotid artery intima-media thickness (IMT). The
joined effect of demands and control (strain) was analyzed by five
different strain definitions:
1. a quadrant term (median splits of demands and control),
2. combinations of tertiles of demands and control,
3. an additive term (demands minus control) ,
4. a quotient term (the ra...
Fujishiro et al.1 recently published data on the association of job
demands and control with carotid artery intima-media thickness (IMT). The
joined effect of demands and control (strain) was analyzed by five
different strain definitions:
1. a quadrant term (median splits of demands and control),
2. combinations of tertiles of demands and control,
3. an additive term (demands minus control) ,
4. a quotient term (the ratio) and
5. a multiplicative term (the product).
The first three terms are linear combinations of demands and control,
which are less informative than the corresponding linear combination based
on regression analyses of the mutually adjusted effects of demands and
control. The quotient term implies interaction between demands and control
but does not examine if there is one, its size, form or statistical
significance. An effect of any of the first four strain terms may be due
to an effect of only one of the two factors. Why introduce a strain
measure of the joined effect of demands and control, if it may only
reflect the effect of one of these variables?
A parsimonious and informative way to examine the joined effect of demands
and control is regression analyses with demands, control and their
multiplicative term included as covariates. The authors published the
effects of the multiplicative term but not the main effects. These are
needed to evaluate the form of any interaction. The authors only
illustrate the form of the interaction by dichotomous combinations of
demands and control.
The authors interpret the interaction as confirmation of the job strain
theory because high job control protected against thick IMT, especially
among persons with high job demands. However, they overlook that high job
demands also protected against thick IMT. The interaction effect as a
whole was not in accordance with the job strain model.
References
1. Fujishiro K, Diez Roux AV, Landsbergis P, et al. Associations of
occupation, job control and job demands with intima-media thickness: The
Multi-Ethnic Study of Atherosclerosis (MESA). Occup Environ Med
2011;68:319-326.
We read with interest the article by Sorahan et al., “Cancer risks in a historical UK cohort of
benzene exposed workers” [1]. We note that the authors showed an increased SMR and SRR for
lung cancer among this group. They comment that “there was evidence of increased mortality
for lung and lip cancers and for ANLL, and increased morbidity for lung and pleural cancers.
There is no reason to suspect that benze...
We read with interest the article by Sorahan et al., “Cancer risks in a historical UK cohort of
benzene exposed workers” [1]. We note that the authors showed an increased SMR and SRR for
lung cancer among this group. They comment that “there was evidence of increased mortality
for lung and lip cancers and for ANLL, and increased morbidity for lung and pleural cancers.
There is no reason to suspect that benzene is responsible for the increased lung and pleural
cancer risks in this study.” The authors then go on to point out that it is likely that some
members of the cohort were exposed to other lung carcinogens. However, we feel that if this
was a likely explanation for the observed excess, then the SMR for lung cancer would probably
have been heterogeneously distributed among the industries studied, the risk being elevated only
in those industries with known exposures to lung carcinogens. This does not seem to have been
the case. Further , confounding by cigarette smoking seems unlikely because the SMR for some
other tobacco-related causes are not significantly elevated (e.g. non-maliganant diseases of the
respiratory system).
There have been two papers based on a Chinese cohort of benzene exposed workers [2,3], which
are relevant to this issue. The first of these, showed an RR for lung cancer of 2.31 among
exposed non-smokers, and while 95% confidence intervals were not reported the authors did
state this was statistically significant [2]. The second reported a relative risk for lung cancer of
1.4 (95% CI 1.0-2.0) among benzene exposed males [3]. Therefore, we think it is important to
recognize that while the excess described in this paper may be due to exposure to other
carcinogens, it is also possible that we are seeing accumulating evidence of an association
between benzene exposure and lung cancer.
References
1 Sorahan T, Kinlen LJ, Doll R. Cancer risks in a historical UK cohort of benzene exposed
workers. Occupational and Environmental Medicine 2005; 62:231-236.
2 Yin SN, Li GL, Tain FD, et al. A retrospective cohort study of leukemia and other
cancers in benzene workers. Environmental Health Perspectives 1989; 82:207-13.
3 Yin SN, Haze RB, Linet MS, et al. A cohort study of cancer among benzene exposed
workers in China: Overall results. American Journal of Industrial Medicine 1996;
29:227-35.
Self-report of duration of computer use is usually overestimated. The
search for a valid measure of exposure to keyboard/mouse use resulted in
the development of a computer registration software. The use of this new
software generated unexpected results when IJmker et al.1 found software-
recorded computer use was not significantly associated with upper
extremity/neck symptom onset while self-reported computer use was
sig...
Self-report of duration of computer use is usually overestimated. The
search for a valid measure of exposure to keyboard/mouse use resulted in
the development of a computer registration software. The use of this new
software generated unexpected results when IJmker et al.1 found software-
recorded computer use was not significantly associated with upper
extremity/neck symptom onset while self-reported computer use was
signficantly associated with symptoms in the neck/shoulder and arm/hand.
What is captured in the self-report that is missing in the software-
recorded duration of computer use? In the editorial by Gerr and Fethke2
reference is made to work by Homan and Armstrong3 that noted the potential
negative effect of time spent with hands held over the keyboard but
without keying. In our Medical-Ergonomic Program4 we refer to this
position as the 'action ready' posture when the forearm(s) is in full
pronation over the keyboard or mouse causing muscle activation of the
forearm extensor muscles. This may lead to the development of painful
trigger points in the forearm extensor muscles, a common area of
complaints by computer users.4 Activities such as reading, talking,
thinking etc. while using the computer are frequently accompanied by this
'action ready' posture. Time spent in these activities is included when
self-reporting duration of computer use but would not be captured in
computer registration software.
Other posture issues without keystrokes or mouse clicks involve the
neck/shoulder area. Computer users have a habit of not sitting up
straight against the back of the chair and carry their shoulders forward.
This posture activates the muscles involved with scapulae stabilization
and shortens the pectoralis minor4 resulting in painful trigger points in
the overused muscles . Neck/shoulder muscles are also activated when mouse
use is with the arm extended away from the body, when the monitor is too
far away and the chin juts forward or when the keyboard is too high and
the shoulders remain hiked to compensate. Maintenance of these postures
with or without keystrokes and mouse clicks are an etiology for upper
extremity symptoms that needs to be added to the exposure equation for
computer use.
1. IJmker S, Huysmans MA, van der Beek AJ, et al. Software-recorder
and self-reported duration of computer use in relation to the onset of
severe arm-wrist-hand pain and neck-shoulder pain. Occup Environ Med
2011;68:502-209.
2. Gerr F and Fethke N. Ascertaining computer use in studies of
musculoskeletal outcomes among computer workers: differences between self-
report and computer registration software. Occup Environ Med 2011;68:465-
466.
3. Homan MM and Armstrong TJ. Evaluation of three methodologies for
assessing work activity during computer use. AIHA J (Fairfax, VA)
2003;64:48-55.
4. Bleecker ML, Celio MA, Barnes SK. A medical-ergonomic program for
symptomatic keyboard/mouse users. JOEM 2011;53:561-567.
We have read the study on respiratory disease and cardiovascular
morbidity by Koskela and coworkers with great interest.[1] They found no
obvious effect of direct dust exposure on ischaemic heart disease (IHD)
among granite workers and workers in metal industry such as foundry
workers and iron foundry workers in Finland. Furthermore, there was a weak
association between dust exposure and chronic bronc...
We have read the study on respiratory disease and cardiovascular
morbidity by Koskela and coworkers with great interest.[1] They found no
obvious effect of direct dust exposure on ischaemic heart disease (IHD)
among granite workers and workers in metal industry such as foundry
workers and iron foundry workers in Finland. Furthermore, there was a weak
association between dust exposure and chronic bronchitis and
pneumoconiosis, respectively, among granite workers. However, there was an
association between chronic bronchitis and IHD among granite workers and
iron foundry workers.
We want to make one comment regarding a possible bias towards unity-
effect by the inclusion of non-exposed workers.
The study by Koskela et al. comprises six cohorts. Three of these
cohorts consist of 1000 iron foundry workers, 1000 metal product workers
and 1000 electrical workers, respectively. The subjects selected into the
cohorts were 400 current and 400 former male workers with the longest
duration of employment and a further 200 with the shortest duration of
employment.
A possible association between air pollutants and IHD may have a
similar mechanism as the association between smoking and IHD. Smoking more
than 25 cigarettes per day is estimated to double the risk of IHD compared
with non-smokers. This risk decreases after quitting smoking and becomes
closer to the risk of non-smokers.[2] Thus, former dust exposed workers
followed for 16 years may ultimately have a risk similar to non-exposed
workers.
Consequently, the findings of the study by Koskela and coworkers may
underestimate the relation between dust exposure and the occurrence of
IHD. We certainly agree with our Finnish colleagues that efforts to
prevent IHD should both include the prevention of respiratory diseases and
the control of dust exposure.
We do not have any competing interest regarding this letter.
References:
1. Koskela R-S, Mutanen P, Sorsa J-A, Klockars M. Respiratory disease
and cardiovascular morbidity. Occup Environ Med 2005; 62: 650-655.
2. Doll R, Peto R, Wheatley K, Gray R, Sutherland I. Mortality in
relation to smoking: 40 years’ observations on male British doctors. BMJ
1994; 309: 901-911.
Bengt Sjögren, MD, PhD
Work Environment Toxicology
Institute of Environmental Medicine
Karolinska Institutet
P.O. Box 210
SE-171 77 Stockholm
Sweden
Tel: 46 8 524 822 29
Fax: 46 8 31 41 24
E-mail: Bengt.Sjogren@ki.se
Carl-Göran Ohlson, MD, PhD
Department of Occupational and Environmental Medicine
Örebro University Hospital
SE-701 85 Örebro
Sweden
Tel.: +46 19 6022468
Fax.: +46 19 120404
E-mail: carl-goran.ohlson@orebroll.se
We thank Dr. Mikkelsen and colleagues for their constructive comments
on our paper. Our responses to their three major questions are listed
below.
1) Why did we present various formulations of job strain?
The five formulations of job strain have been commonly reported in
the literature. Often, authors chose one or two formulations and thus
would not know if their results were consistent across differe...
We thank Dr. Mikkelsen and colleagues for their constructive comments
on our paper. Our responses to their three major questions are listed
below.
1) Why did we present various formulations of job strain?
The five formulations of job strain have been commonly reported in
the literature. Often, authors chose one or two formulations and thus
would not know if their results were consistent across different
formulations. In our study, with the encouragement of an OEM peer
reviewer, we presented results for all five formulations so that readers
can compare the results both across different formulations and with
previous job strain studies. We agree with Mikkelsen et al. that four of
the five formulations of job strain we examined could be the result of the
effect of job control only. As shown in Table 3, job demands was not
significantly associated with IMT whereas job control was.
2) Why did we not show the main effect coefficients for job demands
and control as we showed the coefficient for the multiplicative term in
Table 4?
When a multiplicative term is included in a regression model, the
main effects of the interacted variables have a more complex
interpretation;1 that is, the coefficients represent the magnitude of
effect for each variable when the other is 0. We showed the main effects
of the two variables in Table 3, which represent the magnitude of each
variable's effect when the other is controlled for. For the sake of
brevity, we did not show the coefficients for demands and control after
the multiplicative term was included in the model. This was explained in
the footnote for Table 4.
3) Why did we illustrate the interaction in dichotomous terms (Figure
1) and ignored that high job demands were protective?
Figure 1 is an illustration of the interaction, which could have been
constructed using the mean+1SD as "high" and the mean-1SD as "low" or
other ways. We used the median split again for the sake of simplicity and
also because it is a commonly used approach in the job strain literature.
We do recognize that the demand-control model was only partially supported
in our study; that is, job control was protective only for those who
reported high job demands. Overall, job demands did not have a
significant association with IMT, as shown in Table 3. The following
paragraph is our discussion on this finding from an earlier draft.
Unfortunately, this paragraph was excluded from the final version because
of the word limit.
Contrary to the demand-control model's prediction,2 we did not find
significant associations between IMT and job demands. The Cardiovascular
Risk in Young Finns Study also failed to find the same significant
association.3 The majority of CVD studies have found positive
associations between job demands and CVD,4 but Belki? et al.4 identified
five studies that reported an inverse association between psychological
job demands and CVD.5-9 In the MESA cohort, of which about 30% were
immigrants, the job demands scale had acceptable scale reliability among
U.S.-born participants but not among immigrants (Cronbach's alpha = 0.75
for U.S.-born, ranged from 0.45 to 0.65 for immigrants, depending on the
language used in data collection10). Limitations in the measure of job
demands used in heterogeneous samples like ours may have limited our
ability to detect associations of job demands with IMT.
References:
1. Cohen J, Cohen P. Applied Multiple Regression: Correlation
Analysis for the Behavioral Sciences. 2nd ed. Hillsdale, NJ: Lawrence
Erlbaum, 1983.
2. Karasek RA. Job demands, job decision latitude, and mental strain:
Implications for job redesign. Administrative Science Quarterly
1979;24:285-308.
3. Hinsta T, Kivim?ki M, Elovainio M, Vahtera J, Hintsanen M, Viikari
JSA, et al. Is the association between job strain and carotid intima-media
thickness attributable to pre-employment environmental and dispositional
factors? The Cardiovascular Risk in Young Finns Study. Occupational and
Environmental Medicine 2008;65:676-82.
4. Belki? KL, Landsbergis PA, Schnall P, Baker D. Is job strain a
major source of cardiovascular disease risk? Scand. J. Work Environ.
Health 2004;30(2):81-128.
5. Alterman T, Shekelle RB, Vernon SW, Burau KD. Decision latitude,
psychologic demand, job strain and coronary heart disease in the Western
Electric Study. American Journal of Epidemiology 1994;139:620-27.
6. Bobak M, Hertzman C, Skovoda Z, Marmot MG. Association between
psychosocial factors at work and non-fatal myocardial infarction in a
population based case-control study in Czech men. Epidemiology 1998;9:43-
47.
7. Hall EM, Johnson JV, Tsou TS. Women, occupation, and risk of
cardiovascular morbidity and mortality. Occupational Medicine 1993;8:709-
19.
8. Johnson JV, Stewart W, Hall EM, Fredlund P, Theorell T. Long-term
psychosocial work environment and cardiovascular mortality among Swedish
men. American Journal of Public Health 1996;86:324-31.
9. Steenland K, Johnson JV, Nowlin S. A follow-up study of job strain
and heart disease among males in the NHANES1 population. American Journal
of Industrial Medicine 1997;31:256-59.
10. Fujishiro K, Landsbergis P, Diez Roux AV, Hinckley Stukovsky K,
Shrager S, Baron S. Factorial invariance, scale reliability, and validity
of the decision latitude and psychological demands scales for immigrant
workers: The Multi-Ethnic Study of Atherosclerosis (MESA). Journal of
Immigrant and Minority Health 2010;13:533-40.
Sir, the recent paper by Lin et al.[1] in the November issue of the
journal was a thought provoking piece of work. In their paper the authors
try to prove the theoretically derived hypothesis that biomarkers of
exposure have smaller variance ratios and would typically provide less
biased surrogates of exposure compared to air measurements. Although I
entirely agree with the theoretical part of this s...
Sir, the recent paper by Lin et al.[1] in the November issue of the
journal was a thought provoking piece of work. In their paper the authors
try to prove the theoretically derived hypothesis that biomarkers of
exposure have smaller variance ratios and would typically provide less
biased surrogates of exposure compared to air measurements. Although I
entirely agree with the theoretical part of this scientific discussion, I
do not think the data presented in their paper actually support the
notion.
I would like to raise three issues to support my case. First, in
order to proof the hypothesis one would have to compare concurrently
collected environmental and biological data gathered from the same
individuals during the same time period. The main analysis presented in
the paper by Lin et al.1 does not fulfil this criterion, because as the
authors pointed out datasets were added with biomarker data only. These
added datasets might not have been representative for the other situations
considered and as the authors point out might be from exposure situations,
which favours biomonitoring a priori (exposure with a long residential
half-life). When we restrict ourselves to the data presented in Figure 5
we see no statistically significant difference in variance ratios between
the two measures of exposure except for pesticides, but these exposure
situations remarkably show variance ratios that favour air measurements.
There is also a puzzling issue when one compares table 3, figure 4 and
figure 5. From figure 4 one learns that 64 lambda values could be
estimated for biomarker measurement datasets and 43 for air measurement
datasets. One would expect that when the authors fall back to exposure
settings where both measurements were performed, they would present
results for 43 datasets. However, in figure 5 results are presented for 54
datasets. It becomes even more puzzling when one considers table 3 where
33 air measurement datasets and 46 biomarker measurement datasets are
mentioned. In addition, comparing figure 4c with 5c it is remarkable that
the number of datasets with biomarkers with intermediate residence time
increases from 17 to 18. Unless the residence time of the biomarkers was
recoded this would have been impossible.
The second issue that might have biased the presented results is the
unequal number of both monitored workers and occasions. From table 1 and
Appendices A and B it is clear that in most situations more air samples
were collected than biomarker samples. As was shown earlier[2], when the
number of observations increases (more individuals measured on more
occasions covering a larger area of possible conditions) the estimated
between- and within-worker variance components have a tendency to increase
as well. So if one does not restrict the datasets to the same individuals
measured over the same period, comparing variance ratios of air and
biomarker data becomes problematic. From the Appendices it looks like that
in most cases the data collection covered the same period, but not
necessarily the same individuals (see for instance reference 24 in
Appendix B, where 249 airborne styrene measurements were performed on 48
individuals during a period of 2-3 months, while 146 blood styrene
measurements were taken from 29 individuals in a period of 3-4 months).
The third issue I would like to raise is the comparison of
environmental versus occupational exposures. Based on the presented fold
ranges and lambda values the authors claim “biomonitoring may be more
advantageous in environmental settings than in occupational settings”.
Here the authors appear to forget that fold ranges and lambda’s are by
definition of a relative nature. A fold range of 100 at nanogram level
might be biologically totally irrelevant, while a fold range of 2 around a
level where biological effects can be expected, might be relevant to study
in an epidemiological setting. When we studied a group of pig farmers in
The Netherlands3 their between-farmer fold range in average air exposure
to endotoxins was very low (bwR95=4.1) and the variance ratio such that
we could expect heavily biased exposure response relations (λ= 4.7).
Nevertheless even within a group so uniformly exposed meaningful exposure-
response relations could be discerned by applying an innovative exposure
assessment method where measurement results were modelled and exposure
predicted based on diary information.
Finally, I would like to comment and I am convinced that the authors
will fully agree, that one should not choose between air sampling and
biomonitoring, but one should perform both whenever possible. Not only
because more data is needed to prove the case for biomonitoring (for which
currently data is insufficient), but also air and dermal measurements are
a necessity in order to link (internal) exposure to possibilities for
hazard control. Prevention of occupational and environmental exposures
before they reach harmful concentrations will continue to be our first
priority in the field of occupational and environmental health.
References:
1. Lin YS, Kupper LL, Rappaport SM. Air samples versus biomarkers for
epidemiology. Occupational and Environmental Medicine 2005; 62:750-60.
2. Kromhout H, Symanski E, Rappaport SM. A comprehensive evaluation of
within- and between-worker components of occupational exposure to chemical
agents. the Annals of Occupational Hygiene 1993; 37:253-270.
3. Preller L, Kromhout H, Heederik D, Tielen MJM. Modeling long-term
average exposure in occupational exposure-response analysis. Scandinavian
Journal of Work Environment and Health 1995; 21:504-12.
In a recent editorial Gerr et al.[1] discuss computer work and
musculoskeletal outcomes based on self-reported exposure versus objective
recordings using computer software. They state that only one small study
(n=27) using objective recordings was published before a large study by
Ijmker et al.[2], published in the same issue as the editorial. They
failed to consider the results of two NUDATA papers based on more than
2...
In a recent editorial Gerr et al.[1] discuss computer work and
musculoskeletal outcomes based on self-reported exposure versus objective
recordings using computer software. They state that only one small study
(n=27) using objective recordings was published before a large study by
Ijmker et al.[2], published in the same issue as the editorial. They
failed to consider the results of two NUDATA papers based on more than
2000 study participants, one of them published in the OEM[3].
The results of the study of Ijmker et al. and the NUDATA studies
consistently indicate that sustained or severe pain outcomes were not
related to objective computer work recordings.
The editorial argues that the results of the study of Ijmker et al does
not invalidate the much larger literature in which self-reported computer
use was associated with musculoskeletal symptoms. The main argument seems
to be that objective recordings do not capture the relevant exposures,
e.g. holding the hands over the keyboard without keying and that different
cut points for such non-activity periods may invalidate the objective
recordings. However, objective software-based computer work recordings are
in very good accordance with other objective measures like video-
recordings, and much better than self-reported exposure. Furthermore,
within reasonable limits, different cut-off values for non-activity
periods do not change these relations or computer times very much. This is
consistent evidence from several studies and not from "preliminary
investigations", as stated in the editorial. Finally, exposure times based
on different cut off's are highly correlated, and their relation to
musculoskeletal outcomes will not vary much with different cut-offs [4].
Contrary to the editorial, we find it very unlikely that retrospective
self-reports about computer use during several months should capture
biologically important aspects of computer work which are not captured by
a validated objective method, which prospectively collects exact computer
use data on a daily basis.
References
1. Gerr F, Fethke N. Ascertaining computer use in studies of
musculoskeletal outcomes among computer workers: differences between self-
report and computer registration software. Occup Environ Med 2011; 68: 465
-66
2. IJmker S, Huysmans MA, van der Beek AJ, et al. Software-recorded
and self-reported duration of computer use in relation to the onset of
severe arm-wrist-hand pain and neck-shoulder pain.
Occup Environ Med 2011; 68: 502-9
3. Andersen JH, Harhoff M, Grimstrup S, et al. Computer mouse use
predicts acute pain but not prolonged or chronic pain in the neck and
shoulder. Occup Environ Med 2008;65 :126-31.
4. Mikkelsen S, Lassen CF, Vilstrup I, et al. Does computer use
affect the incidence of distal arm
pain? A one-year prospective study using objective measures of computer
use. Int Arch Occup Environ Health 2011 May 24 [Epub ahead of print]
We appreciate Dr. Kromhout’s comments regarding our article “Air
samples versus biomarkers for epidemiology”[1] and are pleased that he
supports our recommendation that both air samples and biomarkers be
collected whenever possible. Kromhout raises three points in his letter.
First, he suggests that our conclusion that biomarkers tend to be better
surrogates for exposures than air samples might have b...
We appreciate Dr. Kromhout’s comments regarding our article “Air
samples versus biomarkers for epidemiology”[1] and are pleased that he
supports our recommendation that both air samples and biomarkers be
collected whenever possible. Kromhout raises three points in his letter.
First, he suggests that our conclusion that biomarkers tend to be better
surrogates for exposures than air samples might have been biased by our
inclusion of some data from studies that contained only biomarker
measurements. Second, he questions our comparisons of air measurements
and biomarkers in situations where sample sizes differ between the two
types of data. And finally, Kromhout argues that because variance ratios
(values of lambda) are relative measures, they should not be the sole
criteria for choosing between air and biological measurements in an
epidemiology study. We will respond to each of these points in turn.
We agree that it is best to compare environmental and biomarker data
that were collected concurrently from the same subjects. Indeed, as we
indicated, we made every effort to do so and were successful in 36 of the
47 biomarker datasets (77%) listed in table 1. We have no reason to
believe, as Kromhout suggests, that the remaining 11 biomarker datasets
were less representative of biomarker measurements than those with
concurrent air measurements. But, even if we restrict our analyses to
datasets with concurrent air and biological measurements, 62% of these
datasets had lambda ratios less than one (median=0.46, p.755, par. 2),
indicating that a typical biomarker should be a less biasing surrogate for
exposure in an epidemiology study than a typical air measurement. This
was the main conclusion of our analyses. Although sample sizes were small
for stratified comparisons, we also observed that lambda ratios for metals
were significantly smaller than one (Figure 5B), while those for pesticide
exposures were significantly greater than one. Although Kromhout
concluded from the latter result that air measurements would be favoured
for pesticides, we cautioned that a lambda ratio greater than one could
also result from multiple routes of pesticide exposure including
inhalation, dermal contact, and ingestion; if this were the case, then
biomarkers would be preferred for pesticide exposures as well.
Regarding the issue of sample sizes, Kromhout suggests that the variance
components estimated in our study might have been influenced by the
relative numbers of air and biomarker measurements in our database. His
argument is based upon previous work, which indicated that the within-
subject variance components, estimated from air measurements of workplace
exposures, tended to be greater when sample sizes were greater than 10
measurements than when they were less than 10 measurements’.[2] In fact,
the median number of repeated observations from a given subject in our
database was three for both biomarkers (range: 2-30) and air measurements
(range: 2-37). So, different sample sizes were unlikely to have
influenced our results. We also note that the variance components
estimated in our study were obtained after adjustment for fixed time
effects, whereas those from Kromhout’s study were not.[2] Because sample
sizes tend to be positively associated with the duration of a study,
larger numbers of measurements tend to suggest longer periods of
observation which can introduce trends and seasonal effects into the data.
Indeed, in our study we identified significant fixed effects of time in
about 1⁄2 of the sets of biomarker measurements and about 1/3 of the sets of
air measurements. We reported in our paper that ignoring such fixed time
effects tended to increase the estimated within-person variance component
and to decrease the estimated between-person variance component,
consistent with the earlier work of Symanski et al.[3]
Concerning the relative nature of the variance ratio (lambda), we agree
that it should not be the sole criterion used to choose between air and
biological measurements. As we indicated on p.758 of our paper, “…the
optimal measure of exposure for an epidemiology study depends not only on
variance ratios of the air and biomarker measurements (smaller is better),
but also on projected sample sizes (larger is better), based on practical
considerations and costs, and knowledge of the dominant route of exposure
(if multiple routes, biomarkers are preferred).”
Finally, Kromhout questioned the numbers of datasets we used for
various analyses and noted a discrepancy regarding the numbers of datasets
referred to in Figures 4C and 5C. His confusion may have resulted from
the fact that some studies reported multiple biomarkers for a given air
exposure (e.g., reference 24, appendices B and D) and, occasionally, two
types of air exposure for a given biomarker (e.g., ref 26, appendices B
and D). Thus, 43 datasets generated 64 estimates of lambda for biomarkers
(Figure 4) and generated 54 lambda ratios (Figure 5). While rechecking
our files, we found that one set of blood lead measurements had been coded
as an intermediate-term biomarker for the estimate of lambda (Figure 4C)
and as a long-term biomarker for the estimate of the lambda ratio (Figure
5C). We also discovered that an asterisk (indicating statistical
significance) had been inadvertently deleted from Figure 5B under the bar
‘Metal’. We apologize for any confusion this may have caused.
Sincerely,
S.M. Rappaport, Ph.D.
Yu-Sheng Lin, Ph.D.
Lawrence L. Kupper, Ph.D.
School of Public Health
University of North Carolina
Chapel Hill, NC 27599-7431
U.S.A.
References
1. Lin YS, Kupper LL, Rappaport SM. Air samples versus biomarkers for
epidemiology. Occup Environ Med. 2005 Nov;62(11):750-60.
2. Kromhout H, Symanski E, Rappaport SM. A comprehensive evaluation of
within- and between-worker components of occupational exposure to chemical
agents. Ann Occup Hyg. 1993;37(3):253-70.
3. Symanski E, Kupper LL, Kromhout H, Rappaport SM. An investigation of
systematic changes in occupational exposure. Am Ind Hyg Assoc J.
1996;57(8):724-35.
We appreciate the careful reading of our editorial [1] by Drs.
Mikkelsen and Andersen. We regret our omission of the one published NUDATA
study available at the time our editorial was submitted [2]. That study
reported significant associations between mouse usage time collected with
memory resident software and both, acute neck pain and acute shoulder
pain, among 2146 technical assistants. However, because i) median mouse...
We appreciate the careful reading of our editorial [1] by Drs.
Mikkelsen and Andersen. We regret our omission of the one published NUDATA
study available at the time our editorial was submitted [2]. That study
reported significant associations between mouse usage time collected with
memory resident software and both, acute neck pain and acute shoulder
pain, among 2146 technical assistants. However, because i) median mouse
usage time was 5.2 hours/week and median keyboard usage time was 0.9
hours/week, and ii) rates of moderate or greater musculoskeletal pain were
very low among the participating computer users, we are concerned about
the generalizability of the observed associations to workers with greater
mouse and keyboard use.
Regarding differences in associations with musculoskeletal disorders
(MSDs) observed across studies using self-reported estimates of computer
use versus memory resident software documentation of computer use, we made
no argument that one was correct and the other was incorrect. Rather, we
raised the concern that these two exposure assessment methods capture
different (but not totally unrelated) aspects of computer use relevant to
MSD risk. The absence of perfect correlation between self-reported
estimates of computer use and memory resident software documentation of
computer use may be due to error in self report, differences in the kind
of exposure information captured, or both. The claims of methodological
objectivity and validity presented by Mikkelsen and Andersen do not
address this fundamental question. We continue to believe, as noted in our
editorial, that a better understanding of the attributes of work captured
by self report and by computer registration software will clarify what
appear to be inconsistent results reported by studies using them.
1. Gerr F, Fethke N. Ascertaining computer use in studies of
musculoskeletal outcomes among computer workers: differences between self-
report and computer registration software. Occup Environ Med 2011; 68: 465
-66.
2. Andersen JH, Harhoff M, Grimstrup S, et al. Computer mouse use
predicts acute pain but not prolonged or chronic pain in the neck and
shoulder. Occup Environ Med 2008;65 :126-31.
This article is presenting both the applied question and the
statistical methods used in a very well organised and excellent way.
Though, I have some comments on the use of the words risk, effect and
predictor as these imply a casual relationship and make us believe that we
model the development of neck pain.
The data in the article is longitudinal and in the analysis the
authors are using this...
This article is presenting both the applied question and the
statistical methods used in a very well organised and excellent way.
Though, I have some comments on the use of the words risk, effect and
predictor as these imply a casual relationship and make us believe that we
model the development of neck pain.
The data in the article is longitudinal and in the analysis the
authors are using this. In the marginal model all repeated measures of
“current neck pain” are included and time is in the model. In the
transition model “current neck pain” is modelled and “neck pain previous
year” is included in the model. These are interesting and valuable models
that gives more information on the association of interest than a cross-
sectional study would, but it is not the same as modelling risk or
incidence.
The two models used and the analysis of them is correct as long as
one sees that they are not models of the risk of developing neck pain.
This fact is not clearly put forward either in this article or in texts
about analysis of longitudinal ordinal data. The fact that the data is
longitudinal ensures us that there is more information than data from a
cross-sectional study can give. Note though that the fact that the data is
longitudinal does not automatically imply that it is possible to analyse
cause and effect or to analyse the risk of developing neck pain.
The marginal model in the article models the prevalence of neck pain
using all the information in the repeated measures, time trend etc. The
transition model is modelling the prevalence of neck pain and all the
explanatory variables are adjusted for “previous neck pain”. That is the
OR, for example of “mental stress”, is the OR for having neck pain now,
averaged over all possible values of “previous neck pain”. Note though
that this is not the same as modelling the development of neck pain, in
other words the risk of getting neck pain. In the transition model one
interaction is included, “previous neck pain” and “hand above shoulder”.
This means that for the specific explanatory variable “hands above
shoulder” one can talk about the effect on moving from severe to mild neck
pain (developing neck pain) or the other way around due to “hands above
shoulder”.
The topic of analysing longitudinal ordinal data as an alternative to
dichotomising variables and analysing them traditionally is very
interesting and needs further discussions. There are statistical models
and methods available, but to realise what we model is of great
importance.
I want to thank the authors for a nice article and for introducing
these models in the area of occupational medicine. I look forward to
further discussions and developments in the methodological area of
longitudinal ordinal data.
Fujishiro et al.1 recently published data on the association of job demands and control with carotid artery intima-media thickness (IMT). The joined effect of demands and control (strain) was analyzed by five different strain definitions: 1. a quadrant term (median splits of demands and control), 2. combinations of tertiles of demands and control, 3. an additive term (demands minus control) , 4. a quotient term (the ra...
Dear Editor
We read with interest the article by Sorahan et al., “Cancer risks in a historical UK cohort of benzene exposed workers” [1]. We note that the authors showed an increased SMR and SRR for lung cancer among this group. They comment that “there was evidence of increased mortality for lung and lip cancers and for ANLL, and increased morbidity for lung and pleural cancers. There is no reason to suspect that benze...
Self-report of duration of computer use is usually overestimated. The search for a valid measure of exposure to keyboard/mouse use resulted in the development of a computer registration software. The use of this new software generated unexpected results when IJmker et al.1 found software- recorded computer use was not significantly associated with upper extremity/neck symptom onset while self-reported computer use was sig...
Dear Editor,
We have read the study on respiratory disease and cardiovascular morbidity by Koskela and coworkers with great interest.[1] They found no obvious effect of direct dust exposure on ischaemic heart disease (IHD) among granite workers and workers in metal industry such as foundry workers and iron foundry workers in Finland. Furthermore, there was a weak association between dust exposure and chronic bronc...
We thank Dr. Mikkelsen and colleagues for their constructive comments on our paper. Our responses to their three major questions are listed below.
1) Why did we present various formulations of job strain?
The five formulations of job strain have been commonly reported in the literature. Often, authors chose one or two formulations and thus would not know if their results were consistent across differe...
Dear Editor,
Sir, the recent paper by Lin et al.[1] in the November issue of the journal was a thought provoking piece of work. In their paper the authors try to prove the theoretically derived hypothesis that biomarkers of exposure have smaller variance ratios and would typically provide less biased surrogates of exposure compared to air measurements. Although I entirely agree with the theoretical part of this s...
In a recent editorial Gerr et al.[1] discuss computer work and musculoskeletal outcomes based on self-reported exposure versus objective recordings using computer software. They state that only one small study (n=27) using objective recordings was published before a large study by Ijmker et al.[2], published in the same issue as the editorial. They failed to consider the results of two NUDATA papers based on more than 2...
The Editor,
We appreciate Dr. Kromhout’s comments regarding our article “Air samples versus biomarkers for epidemiology”[1] and are pleased that he supports our recommendation that both air samples and biomarkers be collected whenever possible. Kromhout raises three points in his letter. First, he suggests that our conclusion that biomarkers tend to be better surrogates for exposures than air samples might have b...
We appreciate the careful reading of our editorial [1] by Drs. Mikkelsen and Andersen. We regret our omission of the one published NUDATA study available at the time our editorial was submitted [2]. That study reported significant associations between mouse usage time collected with memory resident software and both, acute neck pain and acute shoulder pain, among 2146 technical assistants. However, because i) median mouse...
Dear Editor,
This article is presenting both the applied question and the statistical methods used in a very well organised and excellent way. Though, I have some comments on the use of the words risk, effect and predictor as these imply a casual relationship and make us believe that we model the development of neck pain.
The data in the article is longitudinal and in the analysis the authors are using this...
Pages