Workers are selected into and out of physically demanding jobs with regard to their health. The study of occupational mortality and morbidity is hampered by this selection. Furthermore, social selection and rapid turnover are involved in health selection. Because different stages of disease form only one continuum (dissatisfaction - death), the correct interpretation of the results also requires measures softer than mortality. Earlier studies have concluded that soft and hard measures of health follow one another as explanations for the termination-of-employment rate. The aim of this study was to determine which age and exposure categories are the most prone to health selection. Mortality and morbidity were studied on three different exposure levels defined primarily according to the physical demands of the work: heavy level (iron foundries); medium level (manufacture of metal products); and light level (manufacture of electrical devices). The population comprised 15 714 men hired in 1950-1976 to work in the three branches of the metal industry. Another cohort, a cross-sectional one, of 1292 workers (who had been hired earlier and were still working in 1950) in the three industrial branches was used to clarify selection due to disability and mortality. Data for the mortality and disability analyses were obtained from national death and disability registers. The periods of follow-up were 1950-78 and 1950-87. A questionnaire on occupational history, morbidity, and the causes of turnover was sent to 400 current and 600 former workers from each industrial branch. A questionnaire concerning occupational history was also sent to the nearest relatives of a total of 450 decidents. The occupational histories of the current and former workers were compared for changes in the exposure level throughout their complete occupational histories. The occupations during the workers' lifetimes were also classified into three exposure levels on the basis of physical demands (heavy, medium, and light). The three exposure levels showed different patterns of change according to age throughout the workers' complete occupational histories. Selection into and out of jobs within and between different exposure levels appeared to be a continuous process, a chain of selection. This conclusion was ascertained when the complete occupational histories were analyzed according to the exposure levels (heavy, medium, light) of the occupations from which the workers came and to which they transferred. The foundry workers entered the industry from either heavy or medium-level occupations, and most of them sought lighter work in medium-level occupations. The metal product workers either began their work lives within the metal product industry or they transferred to it from work that entailed the same exposure level. After leaving a job, the metal product workers generally moved to medium (ie, the same level) or light occupations. The electrical workers switched from medium-level work, or they began their worklives within that industrial branch. When they left a job, they chose medium or light work in the metal industry. Due to the chain of selection from one exposure level to another, the mortality and morbidity rates for a certain exposure level can be underestimated or overestimated if complete occupational histories are not available. The selection process was different for different diseases and was manifested as different stages of disease. Ache or pain in the musculoskeletal system within the last 12 months was more frequent among the foundry and metal product workers than among the electrical workers. Both the currently employed and former foundry workers had significantly more earlier-diagnosed musculoskeletal diseases than the electrical workers. The younger age classes of foundry workers (<45 years) had a higher occurrence of musculoskeletal diseases than the metal product workers; in the older age classes the opposite was true. The former foundry workers had more diagnosed musculoskeletal diseases than the current workers; there was no such difference among the metal product workers. The results indicate that musculoskeletal diseases clearly lead to selection away from heavy work. This conclusion was also supported by the rate of disability due to musculoskeletal diseases, which showed a greater excess in the cross-sectional cohort than in the longitudinal one. Both the current and the former foundry workers (in different smoker categories) had higher rates of chronic bronchitis than the corresponding groups of metal product and electrical workers. Selection away from heavy exposure was indicated by the finding that former foundry workers had higher occurrences of chronic bronchitis than current foundry workers from the age class of 45--54 years on. A similar trend was found for the metal product workers, but the rates were lower. Diagnosed lung diseases were also more frequent in heavy than medium and light work. Health selection during employment was also suggested by the disability rates for respiratory diseases. Respiratory diseases were more frequent than expected in both the cross-sectional and the longitudinal cohorts, but the excess was greater in the cross-sectional one. Negative health selection was shown as higher occurrences of diagnosed emphysema and bronchitis in the youngest age class of electrical workers than among the foundry and metal product workers of the same age. Angina pectoris (irrespective of smoking habits) was more prevalent among both the current and former foundry workers than among the electrical workers, whereas the corresponding differences between workers and metal product workers were not significant. The former foundry workers had somewhat higher occurrences than the current workers. The same was true for the metal product workers. The current electrical workers had higher occurrences than the former electrical workers. This result may reflect negative health selection to light from heavy and medium work. Analysis according to complete occupational history showed that workers at the heavy level had more diagnosed cases of cardiovascular diseases in the younger age classes while workers at the medium level had more diagnosed cases of cardiovascular diseases in the older age classes; both the findings which can be interpreted as the result of selection. Disability due to all cardiovascular diseases and also to coronary heart disease alone were significantly lower for the foundry workers than for the active male population. This finding can be explained both by health selection into and out of heavy-level exposure and by the mortality from cardiovascular diseases in the older age groups. The use of complete occupational histories and the combination of similar exposure levels throughout the lifetime produced clearer differences in the occurrences of musculoskeletal, respiratory, and cardiovascular diseases between the three exposure levels. The heavy level had significantly higher occurrences of any previously diagnosed musculoskeletal disease which had prevented work for at least one month (45%) than the medium (34%) or the light (22%) exposure level. The medium level had higher occurrences than the light exposure level. The prevalences of all diagnosed lung diseases were the highest (42%) at the heavy level, the second highest (36%) at the medium level, and the lowest (32%) at the light exposure level. The differences between the branches were not statistically significant. This analysis of previously diagnosed heart diseases according to the exposure level did not reveal great differences between the exposure levels. The medium level had the highest prevalence (16%), the heavy level the second highest prevalence (14%), and the light level the lowest prevalence (10%). However, workers 45 years or older on the medium exposure level had significantly more previously diagnosed heart diseases than workers of the same age on the light level. Just as changes in the heaviness of the work formed a chain of selection during the workers' lifetimes, changes in the workers' health also formed a chain through their life cycles. The combined results (with the life-tables technique) for mortality, disability, and morbidity indicated that different stages of disease decreased the cohorts on the heavy, medium, and light exposure levels in different ways. The cohort of workers on the heavy level decreased because of decreased work capacity, disability and death. The cohort of workers on the medium exposure level was reduced mainly through disabling diseases and disability in the older age classes. Workers on the light exposure level had the smallest decrement attributable to different stages of disease; however, they had an increased decrement due to death in the older age classes, a finding which indicates negative health selection. The life-table technique was also used to show the points of inflection according to the age at which the measures of different stages of disease turned from soft measures into hard ones. The occurrences of excellent work capacity, as well as soft measures of disease (dissatisfaction, discomfort, disease) were calculated for the workers still active, whereas hard measures (disability, death) were considered complementary. The points of inflection in age varied according to the exposure level. Comparisons of the three industrial branches revealed that the proportion of foundry workers with excellent work capacity decreased the most rapidly with age. The highest point of dissatisfaction (low pay) was found before the age of 30 years and turned to discomfort (heavy work) at the age of 30--34 years. Discomfort again changed to disease at about the age of 40 years. Since heavy work and different symptoms led to selection out of heavy-level jobs, disease did not change to disability until near the age of 50 years. Thereafter hard measures of disease were the most prominent. Low pay was more important throughout the age classes among the metal product workers than among the foundry workers. Correspondingly, the proportion of workers who experienced the work as heavy was clearly smaller than among the foundry workers and turned to disease later (at about the age of 44 years). Nevertheless, disease changed to disability early in the next age class (45--49 years). The electrical workers retained "excellent" work capacity until older ages than the foundry workers and metal product workers. They also had the highest occurrences of dissatisfaction (low pay) throughout all the age classes. The proportions of heavy work and illness were low. Because light-level work is not physically demanding and because there was negative health selection, the occurrence of disease was slightly higher than that of discomfort. Discomfort and disease turned to disability in the age class of 50--54 years. Hard and soft measures thus formed a continuum (dissatisfaction - death) and followed each other as explanations for the termination of employment. Consequently, a chain of selection was not only formed during workers' lifetimes by changes in the heaviness of the work, but also by health changes during their life cycles.