Mortality of workers in the British cotton industry in 1968-1984.

The mortality of 3458 cotton industry workers originally enrolled in a study of respiratory symptoms in the period 1968-1970 was followed to the end of 1984. Both the total mortality and the mortality from respiratory disease were less than expected, and they both decreased as length of service increased. However, for the subjects who initially reported byssinotic symptoms, the mortality from respiratory disease was slightly raised overall, and it increased with length of service. These patterns of mortality indicate a survivor effect (ie, a tendency for those with respiratory weakness to leave the industry), together with a long-term effect reflected in respiratory mortality on the health of those workers susceptible to the effects of cotton dust. The mortality from lung cancer was lower than expected, and it decreased with length of service. This finding is consistent with other observations that exposure to cotton dust may reduce the risk of lung cancer.

The survey was carried out in two stages. In the first ,35 cotton mills were visited, 26 spinning coarse and 9 medium cotton. All were situated in the Rochdale, Oldham, Ashton area. In the second (1968-70) 29 mills, 21 coarse and 8 medium, were re-visited two years after the initial visit and in the same week in the year. In addition, 13 coarse mills, 2 medium mills, and 2 fine mills were visited in other areas of the country, namely, Lancashire, West Riding of Yorkshire, and Scotland. Five of the medium mills included between 30 % and 40 0J0 rayon with their cotton and two others did so occasionally. At all the mills operatives in the cardrooms and blow-rooms were examined in the middle of the working week. In I I mills (6 coarse, 3 medium and 2 fine) the operatives in ring and winding rooms were also examined.
Of the total population available, approximately 10 0J0 refused to co-operate. A further 13 0J0 had to be excluded. Some of these were absent at the time when the mill was visited and the rest were seen but had to be excluded either because of ethnic difficulties (Cotes and Malhotra, 1965; Cotes, 1968) [7,8] which made prediction of normal values for ventilatory function impossible, or because of failure to collect all the information necessary for the analysis. Of the 77 0J0 included in the analysis, 2,316 were seen on the first occasion and 2,556 on the second . As 886 were seen on both occasions the total population for analysis was 3,986. In order to be sure that the population excluded did not include an exceptionally high proportion of disabled persons the frequency of persons drawing disablement benefit was compared in the included and excluded groups. No difference was found [p 43].
In 1971-1974 a third round of survey examinations was undertaken. No new subjects were included, but an attempt was made to trace all those seen in either of the previous rounds, whether or not they were stilI working in the cotton industry, A total of 1497 subjects was seen .
At the end of this exercise all available survey records were entered onto a computer file, and identification details were sent on cards to the National Health Service Central Register (NHSCR) for tracing and flagging for the follow-up of the population's mortality. The total number of records sent to the NHSCR was 3885, ie, 101 (2.5 070) less than the original study total. Of the total sent for tracing, 351 (9.0 %) had no record of their survey examination results. Figure 1 summarizes these figures and gives the numbers not traced in each category. The overall success rate for tracing was 97.8 %.
The analysis described in this report is predominantly based on the 3458 traced subjects with examination data. The mortality of the 340 traced subjects without examination data is also presented. All the death certificates were coded by the Office of Population Censuses and Surveys according to the eighth revision of the International Classification of Diseases (ICD) for deaths occurring before 1979 and the ninth revision of the ICD for deaths occurring from 1979 onwards.
The Medical Research Council's standardized questionnaire on respiratory symptoms (9), as modified by Roach & Schilling (10), was administered to the survey subjects. Byssinosis was graded on the basis of the subjects' responses to the questionnaire as follows: grade 0: no byssinotic symptoms; grade 1/2: occasional chest tightness on the first workday of the week; grade 1: chest tightness on the first workday of every week; and grade 2: chest tightness on the first workday and other days of the week.
The self-reported symptom pattern taken to define "byssinosis" can be found in people not exposed to cotton. It has been estimated that the prevalence of "byssinosis" as measured by the questionnaire among persons without occupational exposure to cotton dust can be as high as 5 % (11). The questionnaire thus recorded as "byssinotic" some individuals who may have had a wider susceptibility to respiratory disease than a strictly specific sensitivity to cotton dust. However, for simplicity of presentation we have referred to individuals "with byssinosis" in place of the more strictly accurate description "reporting byssinotic-type symptoms.' , The smoking habits of the subjects were coded to the four categories defined in the footnote to table 8 in the Results section.
At each subject's first survey examination, an employment history was taken. From this history, each subject's total years of employment in the cotton industry prior to the survey was calculated. At subsequent visits to the mills -or at follow-up survey examinations of ex-workers -the dates of leaving cotton work were recorded. About half the population had such a date recorded. For the purposes of this analysis, subjects without a recorded date of leaving were assumed to have left one year after the date of their last examination. The justification for this assumption was that subjects not appearing in the third round of examinations had very probably left the industry. Workers who did appear in the third round, and who were at the time (1971)(1972)(1973)(1974) stilI working with cotton, will have had a high probability of leaving in the next few years since the industry was contracting very rapidly through the 1970s.

Calculation of expected deaths
Person-years were calculated from the date of the first examination to the date of death or emigration or to 31 December 1984, whichever came first. The period of observation was divided into the thr ee sub periods 1966-1972, 1973-1978, and 1979-1984; five-year age groups were used from 15 to 84 years with an open-ended upp er inter val for age 85 years on. National (En gland and Wales) average sex-, age-, and period-specific death rates were derived from national mo rtality statistics an d applied to the person -year s in the calculation of the expected numbers of deaths for a ra nge o f causes . Th e first two years o f ob servation (1 966-1 967) came pr ior to the adoption of the eighth revision of the ICD, and the 1968-I972 average rat es were used for this period . When the tabul at ion s were prepared, the 1983 mortality data for England and Wa les had not been pr ocessed , and therefore the 1979-I982 rates were ap plied to th e whole period 1979-1984. Table I shows the major cau ses of death (20 or mor e dea ths observed or expected) for the main study gro up (3458 subjects with dat a from at least on e survey medical examination). Th ere was an overall def icit of mortality (583 ob served, 684.5 expected). Th e deficits for all causes and all cancers reached statistical significance at the 1 ulo level, and that for diseases of th e respiratory system was statistically significant at the 5 Ufo level. The deficit in overall mortality was not onl y due to a lack of deaths from respiratory and circulatory diseases, but was also substantially con tributed to by reduced mor ta lity from lung and breast cancer. Table I also shows the sta nda rdized mort ality rati o (SMR) value s for th e first five years of follow-u p (counting from each individual's date of first examinat ion ) and for th e subsequent follo w-up. Th ere was a substantial di fference between the two periods. During the first five years of follow-up th e observed numbers of death s were abo ut one-hal f the expecte d numbers; whereas during later periods of follow-up, the numbers of observed dea ths were close to tho se expected , though deaths from respiratory disease remained somewhat low. Table 2 gives the o verall mortalit y of the 340 subjects with no avai lable exami natio n dat a for the same majo r cause gro ups as shown in table 1. The observed death s fro m all the major categories were much higher than th e number expected, and stat istically significant excesses were recorded for all cau ses and fo r circulator y disease. For th is gro up of subjects the pattern of mortality in relation to period of follo w-up was th e reverse of that seen for the main study group, mortality being sub stantially and significantly raised for the period immediately aft er entry into the study and close to no rma l thereafte r.

Respiratory and cardiac diseases
Disease categories of special interest were selected for further analyses of mortality in correlation with cotto n expo sur e and reported byssinotic symptoms. (These analyses were based on the main study gro up of the 3458 subjects with examination data.) To ensure, as far as possible, that all mortality from respirato ry conditions was identified, we included two categories of circulat ory disease (pulmonary heart disease and congestive heart failure) which might occur as a consequence of lung disease. Because of the recent suggestion (12) that the deficit of lung cancer in some other studies of cotton wor kers might be due to a protective effect connected with endotoxins, lung cancer was also included.  Table 3 shows the mortality for the selected cau ses for the men and women separately. Deficits were found for all th e categories but on e, ie, the men showed a nonstatistically significant excess of death s due to congestive cardiac failure (6 ob served, 3.2 expected) . Table 4 shows mortality by years of employment as cotton workers . The re were significant deficits of mortality among the workers with 30 or more years of service in the industry. For all causes the significant deficit occur red for bot h sexes, but for circulatory disease statistical significance was only reached for the men , although the women showed the same trend . The defi cit of lung cancer for these workers was statistically signifi cant when the men and women were combined [22 observed, 34.4 expected, SMR 64, 95 070 confidenc e interval (95 % CI) 40-97].

Exposure to cotton
Mortality for the workers with fewer years of exposure was much less favorable, and a stati stically significant excess of circulatory disease was found in the 15-to 29-year group for the women alone and for both The overall pattern of associations with length of cotton expo sure sho wed trends both for deaths due to lung cancer and, though less marked, for deaths due to respiratory disease. In both cases (and for both sexes), the workers with longer exposure showed lower SMR values. Table 5 shows the mortality for persons who, on the basis of their replies in the examination, were given a grade of 1/2 or higher for the presence of byssinosis. For th is group overall mortality was higher than for the total population, and the deficit for circulatory disease was not as apparent in this subgroup, while the SMR values for respiratory disease were raised, but not to a statistically significant level. Table 6 shows the mortality from lung cance r and circulatory and respiratory diseases in relation to the sex and smoking status of the subjects in the main study group, and the presence of byssinotic symptoms amon g them . Lung can cer and circulatory disease showed no relat ion to byssinotic symptoms but were stro ngly affected, as expected, by smoking. Both the male and female nonsmokers ha d a statistically significa nt deficit of lung cancer (SMR 12 and 13, respectively). The male smokers had an SMR of 97, and that of the female smokers was 160. The pattern for circulatory disease mortality was similar. The ma le and female nonsmokers' SMR values were 64 and 66, respectively, the corresponding smokers' SMR values being 97 and 151, respectively. The more marked smoker/ nonsmo ker contrast for the women was a pred ictab le consequence of th e lower prevalence of smoking among women (ie, a female smo ker is less " typical" than a male smo ker). For resp iratory disease the general pattern was f o r mor-   Table 6. Observed (0) number of deaths and t he standardi zed mortality ratio (SMR) for lung cancer, circulatory disea se, and respiratory disease by smok ing and byssinosis statu s for the men and women in the main stud y group separately. Among the persons with byssinosis, the men showed a strong positive relation and the women a weaker negative one. For each sex the patterns were maintained across all the smoking categories.

Analysis of byssinosis levels
To explore the effects of age, smoking, and cotton exposure on symptoms of byssinosis, a logistic regression model was fitted predicting the prevalence of byssinosis (grade 1/2 or more) in terms of age, years of cotton exposure, and smoking. An additional factor was also defined to identify ex-smokers as a separate category. Table 8 shows the analysis of deviance for the chosen model. Age, years of cotton exposure, and smoking each made independent and highly significant contributions to the model. The adjustment for ex-smokers was not statistically significant. When the ex-smoker adjustment was included, the corresponding model parameter gave ex-smokers predicted byssinosis rates between those of light and medium smokers. Figure 2 displays the predicted values from this model in terms of years of cotton exposure for subjects aged 40-49 years. For clarity only the heavy, light, and nonsmoker curves are shown, the medium smoking category curve lies between the heavy and light smoker curves. XCOT (Exposure to cotton, six levels: < 10,10-,20-,30-, 40-, 2:50 years). XSMO (Ex/current smoker, two levels. All ex-smokers were assigned the 15-24 cigarettes/d level on SMOK).
Byssinosis prevalence was modeled as the logit of SMOK + AGE + XCOT Lung cancer and cotton exposure Table 9 shows lung cancer mortality in terms of cotton exposure. There was a consistent pattern of decreasing SMR values with increasing length of exposure in all the smoking categories (except for nonsmokers and female heavy smokers, for whom the numbers were too small to indicate a trend).

Discussion
Recent debate about the nature of byssinosis centers on the question of the permanence of its impact on affected workers' health. There is no doubt that exposure to cotton dust leads to acute respiratory changes in some workers. Numerous studies have demonstrated that this acute effect can be observed both in terms of subjective reports of the typical "Monday morning chest tightness," and the objective pre-to postshift Table 9. Observed (0) number of deaths from lung cancer and the standardized mortality ratio (SMR) by smoking habits and length of cotton exposure for the men and women in the main study group separately. changes in respiratory performance. Some studies (including the original morbidity analyses of this population) have also shown that the rat e o f lung function loss, and base-line (preshift) lung fun ction, is lower among long-term cott on workers than amon g the genera l population . Neverthele ss, it has been argued (1,(13)(14)(15)) that the existence of these effects does not prove that byssinosis is, in fact , perm anentl y and significantl y disablin g. The evidence of publi shed anal yses of mortality among cotto n workers (15)(16)(17)(18)(19)(20) is equ ivocal as to whether cotto n worker s' mortality is detectably worse than that of a ppro priate reference groups. None of these studies were able to allo w for the effects of smo king.
Th e mortality data presented in th is report was derived from a follow-up of a cross-sectiona l or census popul ation of cotton workers. Individual subjects were followed for a period ranging from 13 to 18 years. The data on cotton exposure relates predominantly to time prior to recruitment of an individual into the study.
The low overall mortality of the main study population reached statis tical significance at the 1 'Jlo level, but this overall deficit was due to the very low mortalit y in the first five years.
Because of its cross-sectional nature, the population may well have experienced selection bias, and this bias, alon g with the health y worker effec t enhanced by the relatively short period of follo w-up (maximum 18 years), may well have been respon sible for much of the deficit in mortality that was observed. This deficit was still present when the results for subjects with no exam ination data were included , and the pattern in relation to follow-up period was maintained, thou gh the difference between the first five yea rs and subsequent years was redu ced. It is clear that -by chance, or otherwise -the subjects without examination data included a disproportionate numb er of the unhe alth y members of the initial population . H owever their relatively sma ll number, and the fact th at their mortality for respiratory disease-the category of most interest -was close to expectation (6 observed , 5.3 expected, SMR 117), gives some assurance that the analyses based on the main study group were not seriously distort ed.
The interpret ation of the relationship of respirat ory mortality to years of exposu re to cotton is complicated by the fact that cotto n has an acute -and subjectively apparent -effect that will produce a tendency for workers so affected to leave cotton work (if possible). When this process operates, the work force is continuously pur ged of indi vidual s susceptible to the effects of cotto n dust, and thus the damage to their respirator y health is curtailed.
At first sight, the relationships between byssinosis, respirato ry mortality, and years of exposure to cott on in this population are contradictory . The presence of byssinot ic symptoms was strongly and positively relate d to cott on exposure after smoking and age were controlled for (table 8, figure 2). Byssinosis was po sitively related -though the numb ers were small -to respir ator y mo rtalit y (table 6) and remain ed so after control for smoking and sex (table 7). But cotton exposure was strongly and negatively related to respirator y mort ality (table 4). The key to this puzzle is the selection effect that has already been outlined -or, more accurately, its implicat ions for the " survivors." Indi vidual s who have spent , say, 40 years in expo sure to cotton and who have no byssinotic sympto ms ar e, evidentl y, not very sensitive to the effect s of cott on du st and , to this extent at least , have more robu st respiratory systems than persons with symptoms. (Sensitivity to cott on dus t may be an indicator of a mor e general respir atory sensitivity.) According to this assumption one would expect the respiratory mor tality of non symptomatic indi viduals in a cross -sectiona l survey to fall with increasing cotton exposure and for normal respiratory death rates to apply only to tho se persons with the shorter lengths of exposure. This is exactly the pattern shown for the nonbyssinotic population in table 7, for both sexes, and with some consistency over all smoking categories (although the "normal" SMR values for respiratory disease are, o f cour se, greater than 100 for smokers). For individuals with byssinoti c symptoms the picture is less clear. For the men, and for both sexes combined , the SMR for respiratory disease increased with increasing cott on exposure; but for the women the pattern was reversed (though based on only eight deaths in compari son with the 13 among the men) .
In considering these dat a, the effects of selection must be recalled . If those who ar e affected leave the indu str y before their exposure to cotton ha s done all the damage it might, the full potential effects will (happily) rarel y be seen, and the anal ysis of mortality amon g symptomatic ind ividual s will show a weak er relationship than otherwise. In addit ion it is likely that part-time or interrupted working patterns are more common among women than men so that, for women, the length of employment in the industry is a less satisfactory measure of exposure than is the case for men.
The effect s of smoking on mortality are demonstrated in tabl e 6. They were broadly as expected with raised levels of circulatory disease and lung cancer for smokers as compared with nonsmokers . However the male smokers' SMR values were lower for both causes of death and less dif ferent from tho se o f the nonsmokers than would normally be seen. Th is observation is prob ably explained by the fact that, while the prevalence of smoki ng in this population was close to average, the level of consumption amon g the men was lower. The proportion of male smokers consuming < 15 cigarettes per da y was 52 % for thi s population and was 35 010 for the general population at the time of the original survey (21). This low tobac co consumption is probably due to restri ction s on smoking in cotton mills. The fema le smokers in this populat ion did not have a lower tobacco consumption than the general (female) population, presumably because at avera ge levels of female cigarette consumption (much lower than men 's) the restri ction s on smo king in the mills ha ve less impact. The percentage of women smoki ng < 15 cigarettes per day was 62 % for the cotton workers , ie, the same per centage as for th e general femal e population.
The se considerations of restrictions on smoking are ob viously relevant to the question of whether exposure to endotoxins in cotton du st have a protective ef fect against lung cancer , as suggested by Enterline et al (12). Th e strong trends of decreasing lun g cancer mo rtality with incr easing length of cott on exposur e shown in tabl e 9 suggest that there may indeed be such an effect. Since this is a compa rison between smoking cotton workers , the ir level of consumption relative to th at of the general population is not relevant except to the extent that ex-cotton workers' smoking consumption may incre ase after they cease cott on work. There is some evidence that this may be the case, for women at least. Among the women last seen as ex-work ers, 16 % had increased their consumption from th at recorded in the first examination, and 8 % had redu ced their consumption . For the men there was very little difference (21 % increase, 19 % decrease). Th e median gap between the first exam ination and the last for these workers was 5.8 years. Thes e cha nges do no t seem sufficiently large to explain the strong trends observed, particularly since early smo king behavior is a much mo re powerful determinant of lung cancer mortality than more recent smoking.

Concluding remarks
Expo sure to cotton du st do es affect the risk of death fro m respiratory disease, bu t on ly in susceptible individuals. Smoking exacerba tes thi s effect. Th e acute byssinoti c response is (weakly) predictive of thos e at incr eased risk.
The study provides support for the hypothesis th at expo sur e to cotton dust reduces the risk of lung cancer , perh ap s through th e actio n o f th e associated endot oxins.