Of the 10
308 participants in the Whitehall II study, 3830 were excluded from our analysis (figure). Reasons for exclusion were missing data on mortality (n=11) or absence records (n=1124) and an incomplete three year exposure period (n=2695; 120 died and 2575 left the civil service during the exposure period). We compared mortality after 1993 in the remaining sample of 6478 participants with the sample excluded from analysis. After adjustment for age, sex, and employment grade, mortality was lower in the analysed sample than in the excluded sample (hazard ratio 0.81, 95% confidence interval 0.67 to 0.98).
Flow of participants through study
Overall sickness absence and mortality
Of the 6478 participants, 288 died during the follow-up period from the first day after the sickness absence exposure to September 2004. Mean follow-up was 13.2 (SD 2.3, range 0.1-16.1) years. Mortality in the 12 civil service departments where absence diagnoses were recorded from baseline was similar to that in the other eight departments where recording began in 1991 (hazard ratio 0.98, 0.74 to 1.31).
Table 1 shows the numbers of participants who had one or more spells of medically certified sickness absence during the three year exposure period. A total of 1906 employees (29%) had one or more such spells, 18% having one spell only and 11% having two or more. Altogether, there were 3214 such spells of absence, with a median length of 16 days.
Table 1 Distribution of spells of medically certified sickness absence during three year exposure period among civil service employees
Table 2 shows the associations of these spells of absence with subsequent mortality, (adjusted for age, sex, and employment grade). Among the participants who had had one or more medically certified absence in the three year exposure period mortality was 1.7 times higher than among those with no such absences, and further adjustment for self rated health, longstanding illness, and the composite physical illness indicator had little effect on this (hazard ratio 1.59, 1.16 to 2.17). We also found a dose-response association between rates of medically certified absence with subsequent mortality: compared with having no medically certified absences, the hazard ratio for having two or more medically certified absences was 1.97 (1.43 to 2.71) and for having one such absence was 1.48 (1.11 to 1.98).
Table 2 Hazard ratios for all cause mortality (adjusted for age, sex, and employment grade) among 6478 civil service employees by specific diagnoses for medically certified spells of sickness absence
Diagnosis-specific sickness absence and mortality
The commonest diagnostic categories were respiratory, surgery, musculoskeletal, psychiatric, infectious diseases, and injury—which together accounted for 73% of all medically certified absences in the three year exposure period. Among those participants who had two or more absences, 77% had absences attributed to two or more of the 25 different diagnostic categories.
As shown in table 2, medically certified absences attributed to infectious diseases, respiratory diseases, and injury had similar hazard ratios for mortality, ranging from 1.5 to 1.7. The increased mortality risk was greater, but not significantly so, for spells of absence with a non-psychotic psychiatric diagnosis (hazard ratio 1.9), and this increased risk was seen with both subcategories “neurosis” (1.9) and “neurosis ill-defined” (2.0). Hazard ratios were higher for sickness absences attributed to circulatory disease (4.7) and surgical operations (2.2). Fewer than 10 people had sickness absence attributed to cancer diagnoses; this group had a particularly high hazard ratio of 21.3 (6.7 to 67.4). Those having absence with a musculoskeletal diagnosis did not have increased mortality compared with participants who took no medically certified absences (hazard ratio 1.0).
Comparison of our models with and without sickness absence diagnoses indicated that inclusion of diagnoses significantly improved the prediction of mortality (P for improvement in χ2=0.03). This statistical test of improvement in model fit is equivalent to testing whether hazard ratios for mortality vary by diagnosis. The only diagnosis where the hazard ratio for all cause mortality significantly exceeded the hazard ratio of 1.66 for all absences was circulatory disease (P<0.001).
None of the interactions between diagnostic category and sex was significant. Having multiple absences with two or more different diagnoses during the three year exposure period was associated with slightly higher mortality, but this was not significant (hazard ratio 1.34, 0.75 to 2.39, for ≥2 absences with different diagnoses v ≥2 absences with same diagnosis).
We repeated analyses after excluding the 31 employees who died in the three years immediately after the three year sickness absence exposure period, to exclude the possibility that absence close to death might account for our findings. The hazard ratio for any medically certified absence compared with no such absence was 1.59 (1.23 to 2.07). Likewise, hazard ratios for diagnosis-specific absence were also hardly altered (results not shown).
Sickness absence and cause-specific mortality
Table 3 presents associations between medically certified spells of absence and cause-specific mortality. The two leading causes of death were cancer (144 deaths) and cardiovascular mortality (72 deaths), accounting for 50% and 25% of all deaths. Other causes (68 deaths) accounted for 24% of all deaths. Cause was unknown for four deaths. Among the cancer deaths, 30% were assessed as smoking related.
Table 3 Hazard ratios for cause-specific mortality (adjusted for age, sex, and employment grade) among 6478 civil service employees by specific diagnoses for medically certified spells of sickness absence
Medically certified spells of sickness absence were associated with both cardiovascular mortality (hazard ratio 2.0) and cancer mortality (hazard ratio 1.7). Absence attributable to psychiatric disorder showed a stronger association with cancer mortality (hazard ratio 2.5) than with cardiovascular mortality (hazard ratio 1.2), but the difference was not significant. The association with cancer mortality was seen for both psychiatric categories—“neurosis” (hazard ratio 2.2, 0.8 to 6.2) and “neurosis ill-defined” (3.0, 1.5 to 5.9)—and was also observed in the subgroup whose absences were all attributable to psychiatric disorder (hazard ratio 2.4). To exclude the possibility that this association might be explained by higher rates of sickness absence for psychiatric causes among people with cancer diagnosed, we repeated the analysis excluding the 166 participants with a self reported diagnosis of cancer either at baseline or follow-up in 1995-6. In the subgroup without cancer, participants who had absences with a psychiatric diagnosis had cancer mortality 2.4 times higher (95% CI 1.2 to 4.6) than did those with no absence.
Otherwise, rates of diagnosis-specific absence tended to show stronger associations with cardiovascular mortality than cancer mortality, although not significantly so. However, absences with a musculoskeletal diagnosis were not associated with either cardiovascular or cancer mortality. None of the interactions between sex and diagnosis-specific absence were significant either for cancer mortality or cardiovascular mortality. Medically certified absences were associated with mortality from suicide (ICD-9 code E95, ICD-10 codes X60-X84) or external cause of undetermined intent (ICD-9 E98, ICD-10 Y10-Y34) (six deaths, hazard ratio 6.9, 1.2 to 39.1).
Adjustment for smoking, alcohol consumption, body mass index, and hypertension did not remove the associations of medically certified sickness absence with cause-specific mortality (table 4). As might be expected, the association between absences with a circulatory diagnosis and cardiovascular mortality was somewhat reduced after adjustment for these risk factors, but it remained significant (hazard ratio 3.4). However, adjustment for these risk factors only slightly attenuated the increased risk of cardiovascular mortality associated with absences for other diagnoses. Absences for both psychiatric and non-psychiatric causes maintained their associations with cancer mortality after adjustment for risk factors.
Table 4 Hazard ratios for cause-specific mortality among civil service employees by specific diagnoses for medically certified spells of sickness absence
Exclusion of participants who died in the first three years after assessment of sickness absence had little effect on hazard ratios for cause-specific mortality (table 4). Similarly, when deaths in the first five years were excluded, there was still a twofold increase in cancer mortality among those having absences for psychiatric reasons.