Prediction of early retirement on the basis of a health examination. An 11-year follow-up of 264 male employees in a

ASTRAND N-E, ISACSSON s-o, OLHAGEN GO. Prediction of early retirement on the basis of a health examination: An II -year follow-up of 264 male employees in a Swedish pulp and paper company. Scand J Work Environ Health 14 (1988) 110-117. In a Swedish pulp and paper company 264 men in the age groups 36, 46, and 56 years underwent a health examination in 1972. An l l-year follow-up in respect of early retirement was performed . On the basis of collected data, five scales to assess work ability were established. The scales were (i) self-assessment of health and work capacity, (ii) the doctor's evaluation of general medical work capacity, (iii) the doctor's assessment of medical adaptation to work, (iv) aerobic work capacity, predicted from a submaximal bicycle test, and (v) work performance, determined from an interview with the subject's nearest superior. Multivariate analysis of the five scalesand about 30 other variables showed that general medical work capacity was one of the strongest predictors of early retire ment. It was also shown that the medical variables were better predictors of early retirement than the psychological and sociological ones. It was concluded that it is possible to predict early retirement on the basis of a health examination within an occupational health organization.

Preemployment and periodic health examinations performed by occupational health services usually aim at recognizing risk individuals, risk groups, and risk factors and not at detecting cond itions respond ing to conventional medical treatment. Appropriate scales for the evaluation of an individual' s general work capacity and ability to adapt to a specific type of work ar e a prerequisite for the systematic identification of tho se persons who are not suitable for a specific job or who require special precaut ions in order to do their job without risk to their health .
There are several reports regardin g functional ability and disability scales, but they are mor e useful for rating disabiliti es to perform certain movements and muscular activities such as climbing stairs, carr ying heavy loads, and oth er such activities. An overview of these scales has been written by Kaufert (20).
There are also scales to evaluate a number of defined work -related abilities and disabilities. Some of them are structured as work -capacity profiles with corresponding scales to evaluate physical demands at the work site (16,18). Scales for evaluatin g general medical work capacity are used in some Swedish occupational health services, but no prospective evaluations can be found in published reports.
In an article fro m 1986 Charl son et al (10)  110 but that a systematic appraisal of such ratings has not yet been reported. They showed that physicians' severit y ratings were a significant predicto r of inhospital mortality.
Prospective studies of self-rated health in genuine working populations have not been reported. Palmore showed that self-assessed health among 60-to 79-yearold men was the best pred ictor of longevity among several tested variables, bette r than the medical assessment of physical function , and the best predi ctor of earl y retirement (26,27).
Aerobic work capacity, more often referr ed to as physical work capacity, was earlier characterized as an indicator of ability to produce, but the need for physical energ y expenditure has dimini shed with the diminishing numb er of heavy occupations in industrial countries (21). Physical work capacity is today, therefore, investigated mor e as a health factor (5, 11 , 17, 28,29, 31).
Several studies report an association between physical wor k capacity and risk factors for co ronary heart disease (8, 11 ,28,31 ). We found only one prospective stud y reporting physical work capa city as a predi ctor for a defined disease, myocardial inf arction, in a homogeneou s working population (29).
An assumed asso ciation of physical work capacity with sick leave and productivity measures has been investigated in several studies, but the result s were contrad ictor y (5,6,7,12,22).
The objecti ve of the present stud y was to investigate (i) the cross-sectional relations of work performance with aerobic work capacity, self-ratings, and doctor's rat ings of general work cap acity and medical ada ptation to work and (ii) the pred ictive power of the afore-mentioned factors and oth er risk indicato rs for early retiremen t.

Subjects
Th e subjects consisted of 264 men who were all wor king at the time of the base-line examinat ion in 1972. They were split into three groups according to age (36-, 46-, and 56-year -olds). The y were all emplo yees of Svenska Cellulo saaktiebolaget (SCA), a pulp and paper company, and worked at one of four work sites situated in or close to Sund svall, Sweden.
The selection procedure was designed to fit pur poses oth er than those of the present study. The 36-year-olds were selected from the sta ff register as tho se 100 men who worked at anyone of the four work sites and who were aged 36 years, or as close to that age as possible, at the time o f the examina tion in 1972. The selection was performed in 1971. The 46-and 56-year-old s constituted the currently employed fraction of two group s of about 100 men who were selected in 1961 for an investigat ion into the relat ion between age and health among working men. The selection procedure was the same as described for the aforementioned 36-year-olds. Twenty-four percent of the 46-and 56-year-olds included in this study had left the comp any for other emplo yment dur ing the period 1961-1972.
The nonparticipation rate was 15 % for the 36-yearolds, 14 o for the 46-year-olds , and 5 % for the 56-year-o lds .
Th e mortality rates for the participant s and nonparticipant s du ring the follow-up were 1 and II % for the 36-year-olds, 3 and 23 % for the 46-year-olds, and 13 and 20 0,10 for the 56-year -olds, respectively. The corre sponding early retirement rates were 2 0,10 and unkn own for the 36-year-old s, 10 and 33 0,10 for the 46-year-olds, and 39 and 25 070 for the 56-year-olds, respectively.
The to ta l number of subjects in the age groups 36, 46, and 56 years who were examined in 1972 was 270. Five subj ects were granted earl y retirement with a disability pension before 1972 and were excluded from thi s study. One subje ct was missing becau se of technica l reaso ns.
Abo ut 25 (I lo of the men worked as cler ks and 75 0,10 as manu al labore rs. The assessments were based on affiliation in trade unions. Supervisors, engineers, and man agers were included in the group o f clerks. The manual work ers were occupied in bar king, grind ing, washing and straining, bleaching, operat ing paper and pulp machine s, workin g in the sod a boiler room, producing chlorine, packing, wor king in the stea m power sta tion, or repairing. The production units had the character of a proc ess indu str y. Changing positions within the company was common. The repa irers usually work ed a bout half o f the time in the produ ction unit s and half of the time in the workshop. Exposure to heat, moisture, noise, and, at some places, cold and draft was common but mostly of limited dur a-tion. Ph ysically stressful work loads occurred only to a limited extent and were usually not of a continuous character.

Methods
Th e medical examination was carried out during the spri ng of 1972 and was performed by one doctor and two experienced nurses. It started with urine and blood tests on sample s taken from the subjects while they were in a fastin g state. After the tests a light morning meal was served. The examination was continued with a questionnaire which was a Swedish mod ification of the Cornell Medical Ind ex (CMI) (9,14). A structur ed inter view and a non struc tured interview complementing the questionnaire were included in the medical examination. The former comprised question s about angina pectoris, which was defined as chest pain on exertion, relieved within 10 min by the subject' s resting (30). Questions on drug s prescribed for pain relief, sedation, and sleep and on smoking hab its and alcohol consumption were included in the stru ctured interview. Blood pressure was recorded indi rectly by a mercur y manometer from the right arm accord ing to norms of the World Health Organ ization with the subj ect in the supine position (34). It was read at the nearest 5 mm Hg (0.67 kP a). Phase five of the diastolic pressure was recorded.
Aerob ic work ca pacity was expre ssed as maximum oxygen uptake (" V0 2 max) levels and was based upon submaximal ergom etric bicycle procedures described by Astrand (1). The exclusion crit eria were in accordance with a memorandum from the Medical Department of the National Board of Occupational Safety and Health in Stockholm (personal communication ). A bicycle ergometer was used, and a tr ained nurse carried out the tests, which were performed with work load s of 50, 100, and 150 W, 6 min at each load. All the 46-and 56-year -olds had und ergone a similar fitness test ea rlier. The test experience of the 36-yearolds was unknown.
Neuroticism was reco rded on the basis of a scale of neuroticism constructed by Ma rke & Nyman (23) for 32 questions from the Swedish modi ficat ion of the CM I (23). The reliab ility o f thi s scale is reported to be 0.88 for men. The discriminating power was at least 0.60 for 27 of the 32 question s of the scale.
The psycholo gical exam ination was performed by one psychologist. Three tests measuring mental abilities were performed. One of them, the most age-independ ent , was used in this study -" Synonyms," in Sweden called "SRB: I," a test measuring verbal ability with a reported reliabil ity of 0.95 (13).
Th e sociological examin ation was carried out by two sociologists instructed by one coordinato r. The sociological variables used in this study were education , recorded from a stru ctured interview, and seven scales on self-perception of work and the work en-III vironment, recorded from an inquiry supervised by one of the sociologists.
The seven scales included ratings regarding general sati sfaction with present work and employer, work invol vement, job-related mental strain , influence on one's own work situation, feeling of affinity with co-workers, possibility of communication with coworkers , and perception of physical work environment factors as strenuo us.
The scales used, subsequently referred to as Gardell's scales, were constructed on the basis of factor analysi s. The first six scales ha ve been described in a publication by Gardell (15) and the sevent h in a personal communication from him in 1971. Determination of the reliability of the int ernal consistency by Cronbach's alpha showed the majority of the scores to be greater than 0.80, and the minimum score to be 0.75 (14,15).
The follow-up was terminated on 31 December 1983. The details of the early retirements with disability pensions granted in the period 1972-1983 were obtained from the Public Social In surance Offices in 1984. The records included all diagnoses obtained from the medical certificates required for earl y retirement. Up to five diagnoses per case were reported . The principal diagnoses at earl y retirement were recorded from the National Social Insurance Board in 1986.
The outcome variable in this study was the first event of early retirement with a granted disability pension, partial or complete, temporary or permanent.
In spite of the differences in regard to the degree and permanency of work disability , there is a minimum requirement common to all forms of early retirement with disability pen sion s in Sweden, ie, a min imum of 50 070 reduction in work capacity over a long period of time .
By Swedish law (since 1976), one can also be granted early retirement because of rea son s referable to the labormarket situation. Th is type of retirement is restricted to persons aged 60 years or more . In this stud y no subject was granted a disability pension for labormarket reasons.
Based on the aforementioned examinations and interv iews, th e following five central variables were constructed: 1. Self-assessed health and work capacity: a trichotomic scale, in this study dichotomized into "good" and " reduced ." 2. General medical work capacit y: doctor's assessment , a quadrotomic scale, in this stud y trichotomized into "good," "fair," and "reduced." 3. Medical adaptation to work: doctor's assessment, a quadrotomic scale , in thi s study d ichotomized into " good" and " reduced." 4. Aerobic work capacity: maximal oxygen uptake in liters per minute. The variable was normally distributed 112 and divided into quartiles befor e the analysis. For use in the life-t able ana lysis, it was tricho tomized into the low first quartile, a middle interval compr ising the second and third qu ar tiles, and th e high fourth quartile.

5.
Work performance: foreman' s or the employee's superior ' s assessment, a scale of five categories, in th is study trichot omized into "good ," "fair ," a nd " reduced ." A detailed description of th e scales can be obtained upon request.
Th e interviews were per form ed in October and No vemb er 1972.
In the case of employees who left the company and went to oth er enterprises, the scale used, with explanations, was sent to the respective superior as a questionnaire.
The five centr al va riables and an additional 33 oth er variables were used in the mult ivari ate analysis. Th e var iables were chosen on principle in accordance with Palmore's theoretical mod el for the pr ediction of earl y retirement (27), which includes demographic, socioeconomic, health, and job factors and attitudes towards retirement. Attitud es to wards ret irement were not reco rd ed for the 36-and 46-yea r-old s and subsequently were excluded from this study. Variables known to contribute to the doctor's assessments were given priority. An amestic and physical examination data were used . Th ese included blood pressure, evaluation of electrocardiograms, histo ry of angina pecto ris, and reported sympto ms from th e musculo skeletal system . Other varia bles included were seru m cholesterol levels, use of toba cco, alcohol , and analgesics, tranq uilizing or sedative drugs , self-report s of ph ysical exertion during leisure time , result s of the Marke-Nyman sca le of neuroticism , result s of the psychological verbal test, result s of the seven Gardell scales, and variables appl icable to education and occupational status.

Statistical methods
Fischer's exac t test was used to calculate the significance of differences in mortality and earl y retirement between the participant s and nonparticipants.
Yates' corrected chi-square was used fo r other classes with one degree of freedom , and Pear son 's chisquare was used for classes with more than one degree of freedom.
An analysis of variance was used for the calcul ation of the statistical significance of the variation of aerobic work capacity with work performan ce.
The Kaplan-Meier meth od was used to a na lyze th e predictive po wer of each o f the five cent ral var iables with regard to earl y retirement.
The Cox regre ssion was used for the multi variate an alysis of earl y retirement.  with II % fo r subje cts with a fair ratin g , and 3 % for th ose with a good rating (P <O.OOI). Figure 3 shows the co rresponding analysis of med ical adaptation to wo rk . Subject s with a reduced medical adaptation to work showed 50 % ea rly retirem en t in co m pa riso n with 10 % for those with good ada ptation (P < O. OO I).

Results
T he criteri on of statistica l sign ificance was set at P < 0 .05 .
The number o f early retireme nts in the fo llo w-up per iod wa s 44 . The distribution was 34 ea rly reti rements a mo ng the 56-yea r-olds, eight a mo ng th e 46yea r-olds, a nd two among the 36-yea r-ol ds. Thirtysix percen t of the prin cipal diagn oses revealed by the ea rly ret irem ent certif icates wer e catego rized as mu sculoskelet al disorders, 34 fJlo as circ ulatory diso rders, a nd 9 lI /o as mental disorder s. T he average yea rly ea rly reti rem ent ra te was 3.4, 0.9, a nd 0.2 fo r the resp ective 56-, 46-, and 36-year-ol ds . Table I sho ws the five cent ral mea surement s of wor k a bilit y in relatio n to age. All showed an inverse relationship to ag e . Table 2 shows the correlation betw een work performance and the four other central va riables. Work performance was associated with self-assessed health and wo rk ca pac ity (P = 0.001), and with med ical adaptati on to wo rk (P = 0.002). It was not significantly associated with general medical wo rk capacity (P = 0. 063) or aerobic wo rk ca pa city (P =0.7).     Seventeen subjects were not tested for their aerobic work capa city. Five of the missing subjects were excluded from the test because of blood pressure (> 250/ 120), three because of heart disease, and four becau se of disorders of the musculoskeletal system. The test was stopped in two cases due to a high heart rate and in one case each due to intermittent claudication , angina pectoris, and back pain. a The relative risk represents the eff ect of a 10-year increase in age. b Self-perception of physical work environment factors as strenuous was associated with a risk of early retirement . The relative risk represents the effect of a one-degree increase on a five-degree scale . e The relat ive risk represents the effect of a to-rnm increase in systolic blood pressure. Figure 5 shows the life-table analysis of work performance as a predictor of early retirement. Subjects with a reduced work performance showed 34 lrfo early retir ement , subjects with a fair work performance 20 010, and subjects with a good work performance 10 010 (P = 0.001). Ten assessments regarding work performance were missing. Six of the subjects had left the company and started their own enterprises and had no superior . One person was undergoing rehabilitation and was not eligible for assessment. Three subjects were employed elsewhere, and answers were not obtainable. Table 3 shows the result s of the Cox regression analysis for the prediction of early retirement. Age was the strongest independent predictor, showing a relative risk of 7.2 for a lO-year increase in age. General med ical work capacity was the second strongest independent predictor, showing a relative risk of 4.2 for reduced general medical work capacit y. The con sumption o f prescribed analgesic dru gs showed a relative risk of 3.9 for the consumption of analgesic drug s for more than 10 d every year. The self-perception of ph ysical work environment factors as strenuous sho wed an association with early retirement, ie, an impli cation of a risk of 2.0 for an increase of on e degree on the five-degree scale.
During the Cox regression analysis of the work performance variable two dichotomized variables were created. The first was the genuine variable dichotomized by the limit of fair to reduced , ca lled Wp var I. The second was created correspondingly by the limit of good to fair and was called Wp var 2. The first variable (Wp var I) showed a relative risk of 4.6 for reduced work performance when compared with good -t-and fair performance. The second (Wp var 2) showed a relative risk of 2.5 for reduced and fair work performance in comparison with good performance.
Systolic blood pressure showed a relative risk of 1.2 for the effect of a lO Hg mm (1.33 kPa) increase in systolic blood pressure.
The consumption of sedative or tranquilizing drugs for more than 10 d a year showed a relative risk of 3.3.

Discussion
The nonparticipants were shown .to have somewhat worse health than the participants in the analysis of mortality. Since the nonparticipant rate was low and the strengths and significances in this study were high, the nonparticipants' rate could not be supposed to bias the predictive results to an y considerable degree.
The 36-year-olds were not experienced with the bicycle ergometer test. This inexperience could possibly have caused an underestimation of the V0 2 max. However the relation between the V0 2 max levels of the different age groups was similar to that of the Scandinavian standard (I).
Seventeen subjects who were supposed to have low V0 2 max values were excluded from the VO z max tests. Since similar exclusion criteria are widely used, similar subjects must be expected to be excluded in comparable populations. Therefore a discussion of the ir effect on V0 2 max as a predictor for early retirement is meaningless from a clinical point of view. Howe ver we have given the 17 missing subjects artificial values low enough to give them positions within the low quartile of the V0 2 max distribution. The life-table analysis then showed significant differences between the low, the two middle , and the high quartiles (P =0.001) .
Th e aerobic work capacity of all the subjects as a group was lower than the average in the Scandinavian standard, which reports values of 2.8-3.4 I . min-I as an average for men 30-39 years of age, 2.5-3 .1 I . min-I for men 40-49 years of age, and 2.2-2.8 I . min -I for men 50-59 years of age. The comparatively low values sho uld be considered in the light of the fact that the pulp and paper industry, pre viousl y known to be a comparatively heavy industry from a physiological standpoint, had become a proces sing indu str y with a reduction of physically heav y occupations as earl y as 1972. This evolution is illustrated by the fact that 66 070 of the subjects reported a heavy ph ysical work load in previous jobs and only 38 010 of the subjects reported a heavy physical work load in their present jobs. The Scandinavian standard was established from results from selected populations in the late 1950s, and it is probably too high to represent the Scandinavian population of the early I 970s (I). The results in thi s study agree very well with the values reported for drivers, repairers, post office personnel, and commercial employees in the I960s (2).
A rough comparison of the early retirement rate in this material with that of the whole of Sweden shows a somewhat lower rate for the former. This finding is an indication of the healthy worker effect (25). Information about nonparticipants, who showed a higher rate of early retirement only for the 46-year-olds, could aid in explaining the difference.
Of the three central medical assessment variables, general medical work capacity was the only one shown to be an independent predictor of early retirement by the Cox regression analysis. This observation is interesting since this variable can be considered to be the one of the three that contains the most purely medical health components. The variable medical adaptation to work includes work environment as a component by definition, and self-assessment of health is known to be influenced by different perceptions and aspiration levels (32,33). Therefore poor medical health leads to early retirement in the long run, and this statement applies even to those subjects who showed good medical adaptation to work in the base-line examination.
Work performance was the second of the central variables shown to be an independent predictor of early retirement. No earlier report of work performance as a predictor of early retirement has been found by us. The predictive power of work performance could reflect occupations in which good work performance is possible in spite of considerable disabilities and /or some mental or other health defects not detected by the doctor.
The work-related aspects of early retirement were reflected by the Gardell indices . Only one of these indices, self-perception of the physical work environment factors, proved to be an independent predictor of early retirement. This variable is in fact composed of two components, the actual work environment and the attitudes towards it (15). The used variable does not allow for a separation of the two factors, but the appearance of this variable as an independent predictor must be interpreted as an influence of high job demands or the perception of high job demands on health . This result can be compared to Karasek 's theory that a combination of high job demands and low decision latitude gives rise to mental strain and poor health (19). Despite our expectations, the influence on one's own work , as measured on one of Gardell's scales, was not an independent predictor of early retirement.
The use of analgesics, tranquilizers, and sedatives prescribed by a physician proved to be independent predictors of early retirement. Associations of the use of analgesic drugs with poor health have been reported earlier (24). The confinement to drugs prescribed by a doctor presumably increases the value of the use of analgesics as an indicator of poor health since it postulates repeated contacts with a physician and the physician's verification of a disorder that legitimates the use of drugs .

116
The psychological test regarding verbal ability was not an independent predictor of early retirement. This test has earlier been shown to be independently associated with back abnormalities but was not an independent predictor of early retirement (3,4).
The appearance of the central variable general medical work capacity as a predictor of early retirement is interesting . The predicti ve power of this variable was, except for age, the most important of the independent predictors. If this variable were to be excluded from the equation, some of the relative risks provided by other medical independent predictors would possibly increase. However the doctor integrated many health indicators into his assessment. Some of these factors are comparatively rare, and therefore no independent power was shown in this study . Other factors gave small, separate contributions to the reductions of general medical work capacity but interacted with other factors that could not be supposed to show separate independent predictive powers in the computed analyse s.
Moreover, our experience shows that complicated routines and analyses cannot be integrated into everyday occupational health services. This fact has also been pointed out by Greenwood (16). The use of a scale for the assessment of general medical work capacity founded on data from a health examination can therefore be recommended as central in the evaluation of the risk of individuals to be chosen for special observation or secondary preventive measures. This simple measure must be a cost-effective method. The addition of the other independent predictors reported in the present study increases the accuracy.
Of course the findings in this study cannot be applied to every working male population without objections. However, the present material was composed of subjects of diffe rent ages and with a variety of occupations and job demands . Therefore there is reason to believe that the method, as a whole, can be appli cable to diverse working populations.
The medical, psychological, and sociological measurements were used in this study independently of each other. The medical measurement s predominated the list of independent predictors. When the rea sons for this outcome are evaluated, the selection of variables, the qualit y of the used variables, and their mechanisms of action on the end point have to be considered . The selection of variables was in accordance with Palmore's model for the prediction of early retirement except for attitudes towards retirement , which were lacking in this study (27). The quality of the psychological and sociological variables used was documented. The medical dominance among the independent predictors of early retirement can be explained by different positions in the chain of action. It is well known that psychosocial factors influence medical health . In the chain of causes medical health factors are situated closest to the end point, and there-fore gi ve the b est in d ep en d en t prediction of early ret irement.