Worked to death? A census-based longitudinal study of the relationship between the numbers of hours spent working and mortality risk

Abstract

Background There has been an increasing interest in the health effects of long working hours, but little empirical evidence to substantiate early case series suggesting an increased mortality risk. The aim of the current study is to quantify the mortality risk associated with long working hours and to see if this varies by employment relations and conditions of occupation.

Methods A census-based longitudinal study of 414 949 people aged 20-59/64 years, working at least 35 h/week, subdivided into four occupational classes (managerial/professional, intermediate, own account workers, workers in routine occupations) with linkage to deaths records over the following 8.7 years. Cox proportional hazards models were used to examine all-cause and cause-specific mortality risk.

Results Overall 9.4% of the cohort worked 55 or more h/week, but this proportion was greater in the senior management and professional occupations and in those who were self-employed. Analysis of 4447 male and 1143 female deaths showed that hours worked were associated with an increased risk of all-cause mortality only for men working for more than 55 or more h/week in routine/semi-routine occupations [adjusted hazard ratios (adjHR) 1.31: 95% confidence interval (CI) 1.11, 1.55] compared with their peers working 35–40 h/week. Their equivalent risk of death from cardiovascular disease was (adjHR 1.49: 95% CI 1.10, 2.00).

Conclusions These findings substantiate and add to the earlier studies indicating the deleterious impact of long working hours but also suggest that the effects are moderated by employment relations or conditions of occupation. The policy implications of these findings are discussed.

Hours worked, mortality, longitudinal study

Introduction

Although there is now a wealth of research demonstrating the deleterious effects to health of unemployment and low job security,^1–3 it is also recognised that not all forms of employment are salutogenic.⁴ The potential effects on mortality risk of one of the most contentious aspects of employment—long working hours—have not as yet been widely studied in Western societies.⁵ In contrast, both the Japanese and the Chinese have a word (‘karōshi’, ‘guolaosi’, respectively) for death due to overwork, responsible for a small number of deaths annually in these societies, and recognized as an official cause of death for which there may be government compensation.⁶

A broad body of evidence suggests that long working hours are associated with a range of adverse health consequences. There are clear associations with fatigue⁷ which can lead to an increased risk of work-related and other accidents.^8,9 Other associations include work-related stress, especially if combined with other factors such as poor job control;¹⁰ their recognition as a risk factor for anxiety and depression;^11,12 and a link with declining mental functioning in middle-aged men.¹³ Long working hours during pregnancy, especially if associated with heavy lifting and prolonged standing, may also increase the risk of preterm delivery and low birthweight.¹⁴ A meta-analysis has found small but significant correlations between hours worked and physiological and psychological health symptoms,¹⁵ though the underlying mechanisms are unclear and may be due to other factors such as lack of sleep, stress, disruption to work-life balance or exposure to work-related hazards. Long working hours may also damage health by increasing maladaptive lifestyles, such as drinking, which are detrimental to physical health.^16,17 On the other hand, it is argued that longer working patterns might be beneficial for some as they offer a structured environment, companionship and social support.¹⁸ However, it is the association with cardiovascular disease—first recognized in Japan¹⁹—that has attracted most interest, and in some texts karōshi is synonymous with cardiovascular mortality and its antecedents. This association has been subsequently confirmed in a variety of international studies,^20–23 and a recent systematic review and meta-analysis of prospective observational studies indicated an approximately 40% increased risk of coronary heart disease amongst employees working long hours.²⁴ However, the association between long working hours and mortality, the initial cause of concern, has not been studied in detail.

There is also evidence that the cardiovascular and other adverse health effects of work vary according to the characteristics of the job and that outcomes are worse when there are high demands and low levels of control (decision latitude)²⁵ or a perceived imbalance between effort and reward^26,27 and, as these are likely to correspond to occupation and job position, it may therefore be anticipated that any deleterious effects of long working hours may not be the same for all groups of workers.

The aims of this study are to quantify the effect of long working hours (as recorded in the 2001 UK Census) on subsequent mortality risk in both men and women, to determine how this is modified by labour market position and, particularly, to examine these effects for cardiovascular disease, accidents or alcohol-related causes, as is suggested by earlier studies.

Method

The Northern Ireland Mortality Study (NIMS) is a prospective record-linkage study, derived from the 2001Census returns for the whole enumerated population, to which subsequent registered deaths to 2009 have been linked in 8.7 years of follow-up. Details of NIMS and linkage processes are described elsewhere.²⁸ These data were anonymized, held in a safe setting by the Northern Ireland Statistics and Research Agency (NISRA) and made available to the research team for this study. The use of the NIMS for research was approved by the Office for Research Ethics Committees Northern Ireland (ORECNI).

The research reported here is based on analyses of a subset of this database comprising a cohort of men and women in full-time employment and aged between 20 and 59/64 years (the official retirement ages) at the 2001 Census. People living in institutional settings were not included. All personal characteristics were drawn from the census returns: these include age (in 10-year bands), sex and marital status (grouped as married, never married, and widowed/separated or divorced). Two indicators of self-reported morbidity were included—presence of a long-term illness limiting usual activities (LLTI), and another on general health (GH) in the year preceding the Census: the LLTI question had a yes/no response and the GH question allowed three responses—good, fairly good or not good. Socioeconomic status was determined using a number of census indicators, including housing tenure (grouped as owner occupier, privately renting or social renting), household car availability (two or more, one only no access), educational attainment (university degree or equivalent, A-level, GCSE, no qualifications recorded), and occupational social class. This latter used the National Statistics-Socioeconomic Classification (NS-SEC) which categorizes in terms of both position in the labour market and work situation (or authority, reward and control over the work environment²⁹): this was summarized as four groups on the basis that the effects of long working hours may vary across the categories. These groups are classed as follows: the first most advantaged group (managerial/professional) comprised large-scale employers, managers and professionals where, according to the NS-SEC, they enjoy a salaried ‘service relationship’ with longer-term benefits and possibility of advancement; the second (routine occupations) included technical, semi-routine or routine occupations and were the least advantaged—all characterized by a ‘labour contract’ or exchange relationship based on payment received for work done; and a small intermediate group lay between these extremes. The final group comprised ‘own account’—or self-employed—workers, a group which included farmers.

The UK Census in 2001 asked ‘How many hours a week do you usually work in your main job?’, with expected responses stating, to the nearest whole hour, the average number of hours per week worked in the 4 weeks prior to census. There is no legal definition of a normal working week and no agreed definition of what constitutes a long working week, so researchers often use the 48-h limit introduced by the European Directive on Working Time. However, the responses to the census question showed marked ‘heaping’ around numbers of hours ending in either zero or five, and the modal value of 40 h per week was stated by 23.1% of the cohort and 37 h (the next most frequent) by 10.8%. The working week was therefore categorized as 35–40 h, 41–48 h, 49–54 h and ≥55 h. The first category defines the usual working week for the majority of the population; the second represents more than usual but less than the limit recommended by the Working Time Directive; and ≥55 h was chosen as the upper category that has been cited most frequently in previous studies to define long working hours.

Finally, information on the number of dependent children in the household and also on attendant informal caregiving responsibilities was included, as these may impact on both the number of hours worked and subsequent mortality risk.³⁰ These also were derived from the census returns—the latter from responses to the question asking about the number of hours spent on informal (unpaid) care given to family members, friends, neighbours or others. In keeping with previous work, this was categorized as hours providing no care, light caregiving responsibilities (1–19 h/week) and heavy caregiving (≥20 h/week).³¹

Analysis strategy

Of the 808 301 non-institutionalized enumerated men and women aged 20–59/64 years at the Census with full information on hours worked, 576 587 were economically active and 231 714 inactive (defined as people without a job at the time of the Census who had not actively sought work in the preceding 4 weeks and/or were not available to start work in the next 2 weeks). This latter group comprised 36.6% unable to work because of permanent sickness, 33.5% homemakers, 6.2% students, 7.3% retired and 16.4% others (mostly long-term unemployed and those who had never worked). The final cohort for analysis comprised 414 949 people (270 011 men and 144 938 women) who were employed for at least 35 h/week: 69.3% of this group worked 35–40 h, 13.9% worked 41–48 h, 7.3% 49–54 h and 9.4% ≥55 h/week.

Descriptive statistics illustrated the socio-demographic characteristics and health status by levels of economic activity. Logistic regression quantified the risks associated with those working the longest hours (55+ per week) when compared with those working fewer hours. Cox proportional hazards models were used to examine mortality risk with censoring at the end of the follow-up period to allow any health impact of long working hours for those close to retirement age to become manifest. Models with person-years censored at age of retirement were also calculated and are available on request. The time-dependent interaction term between working hours and the logarithm of the follow-up period for mortality was non-significant (P = 0.263) confirming that the proportional hazards assumption was not violated. The preliminary models compared mortality by economic activity and subsequent models quantified risk by economic activity and hours worked within the subset working >35 h/week (with hours worked tested as both a continuous and categorical variable). Models adjusted for other possible confounders including baseline health status were used to examine for health selection effects. Tests for interaction were used to determine if the health effects associated with long working hours differed by age, health status or occupational social class. Differences in cause-specific mortality were explored using the following broad ICD10 classifications: circulatory disease (I00-I99); ischaemic heart disease (I20-I25); cerebrovascular disease (I60-I69); respiratory disease (J00-J99); external causes (V01-Y98); accidents (V01-V99 and W00-X59); suicides (X60-X84, Y10-Y34, Y87.0) and alcohol-related deaths (F10,G31.2, G62.1, I42.6, K29.2, K70, K73, K74 (excluding K74.3-K74), K86.0, X45, X65, and Y15.³² Although there were few a priori reasons for expecting an increase in deaths due to malignant neoplasms (C00-C97), they were included as a comparator for any changes in risk from other causes of death.

Results

Table 1 shows that over half (51.3%) of men and women aged 20–59/64 years in Northern Ireland were employed and working full-time (≥35 h/week), though this varied by sex, with almost two-thirds (63.9%) of women either economically inactive or working for <35 h/week compared with 33.7% of men. Overall, 9.4% of those working full-time, 12.4% of men and 3.8% of women, were working for ≥55 h/week. This table also compares the characteristics of the analysis cohort (working full-time) against both those working fewer hours and those classified as economically inactive. The cohort group recorded smaller proportions at older ages and among those not currently married. They also recorded higher proportions in the more advantaged categories associated with educational attainment, housing tenure and car availability. However, the largest differences related to health status—less than 6% of full-time workers reported a limiting long-term illness and only 2% said their general health was not good, compared with 9% and 4%, respectively, for those working fewer hours and 50% and 33%, respectively, for the economically inactive. There is evidence that a higher proportion of women are not in full-time employment because of family or caregiving commitments: 41% of employed women working <35 h/week have two or more dependent children and 20% are providing unpaid caregiving, compared with 25% and 13%, respectively, for men.

Table 1

Characteristics in percentages of fully employed men and women in Northern Ireland aged 20–54/64 years by average hours worked per week, compared with those economically inactive and those working less than 35 h/week

			Economically active: Average working h/week
	Economically inactive	<35	35–40	41–48	50–54	55+
Cohort totals	231 714	161 638	287 697	57 794	30 389	39 069
Men	37.6	30.9	58.1	75.8	83.6	86.0
Women	62.4	69.1	41.9	24.2	16.4	14.0
Age (years)
20–34	27.9	36.3	44.3	42.9	33.3	30.5
35–49	33.7	42.1	38.1	38.9	44.2	44.5
50–59/64	38.4	21.6	17.6	18.2	22.5	25.0
Marital status
Married	56.2	60.1	56.6	61.4	69.3	70.5
Never married	26.3	27.6	35.7	31.6	24.2	23.0
Widow/separated/divorced	17.6	12.3	7.8	7.0	6.5	6.5
Dependent children
None	52.2	42.6	55.0	54.3	49.6	48.6
One	17.6	21.2	19.2	18.1	17.7	17.0
Two or more	30.2	36.2	25.8	27.6	32.8	34.3
Caregiver
None	81.0	82.1	86.4	86.0	84.9	84.9
1–19 h/week	6.9	12.3	10.3	10.9	11.5	11.3
20 + h/week	12.0	5.7	3.3	3.1	3.6	3.8
Occupational group
Managerial/professional	14.8	19.7	38.4	42.5	46.7	35.5
Intermediate	9.4	12.4	16.4	8.2	3.2	4.1
Own account	5.7	6.6	7.9	11.1	25.2	41.4
Routine occupations	51.8	43.6	37.3	38.1	24.9	19.1
Other	18.3	17.7	N/A	N/A	N/A	N/A
Housing tenure
Owner	59.8	75.4	87.7	88.5	90.1	88.7
Private rent	9.8	8.1	5.5	6.2	6.4	7.9
Social rent	30.4	16.6	6.8	5.2	3.5	3.4
Education
Degree	7.9	14.4	25.7	27.7	32.2	25.1
A-level	6.8	11.4	11.0	10.5	8.3	7.2
Intermediate	26.3	38.7	38.1	36.9	31.9	32.7
None	59.0	35.6	25.2	25.0	27.7	35.0
Car access
Two or more	28.7	42.7	51.5	55.3	64.6	65.4
One car	44.4	40.9	41.1	39.0	32.2	31.2
No car	26.9	16.5	7.4	5.7	3.2	3.3
Limiting long-term illness
No	50.2	91.0	94.9	95.0	94.7	94.2
Yes	49.8	9.0	5.1	5.0	5.3	5.8
General health
Good	40.6	74.4	81.8	82.5	83.2	81.9
Fair	26.4	21.6	15.9	15.6	14.7	15.8
Not good	33.0	4.0	2.3	2.0	2.0	2.3

N/A, not appropriate for this group.

Table 2 records the socio-demographic characteristics associated with those working long hours (taken as 55+ h/week). A higher proportion of men than women work long hours—the proportions increasing with age from 9.6% and 2.8% for men and women, respectively, at age 20–34 years, to 14.7% and 6.3%, respectively, at age 50–59/64 years. Those married record higher likelihoods of working longer hours than those not currently married. Within the cohort there is no evidence of substitution of work for family or caring responsibilities, as men and women with two or more dependent children or who described themselves as caregivers were more likely to be working long hours than those with no children or caregiving duties. Finally, occupational social class shows marked variation in the likelihood of working long hours. Compared with men in managerial and professional classes, those in intermediate or routine occupations were about half as likely to work 55+ hh/week whereas self-employed men recorded a 3-fold excess [odds ratio (OR) 3.42: 95% CIs 3.33, 3.52]. Although this group comprised only 17% of the cohort of men in full-time employment, they accounted for almost half of those working for 55+ h/week. The relative difference between managerial and professional classes and own-account workers was more marked for women (OR 6.38: 95% CIs 5.95, 6.85).

Table 2

Socio-demographic characteristics of men and women in Northern Ireland working long hours (LWH = 55 or more h/week) when compared with those working 35–40 h/week. Data represent the percentages within the group of fully employed men and women working long hours and odds ratios (and 95% confidence intervals) from logistic regression

	Men (n = 270 011)			Women (n = 144 938)
	% LWH	Age-adjusted	Fully adjusteda	% LWH	Age-adjusted	Fully adjusteda
20–34 years old	9.6	1.00	1.00	2.8	1.00	1.00
35–49 years old	13.9	1.51 (1.47, 1.55)	1.11 (1.08, 1.15)	4.2	1.52 (1.43, 1.62)	1.23 (1.15, 1.32)
50–59/64 years old	14.7	1.62 (1.57, 1.67)	1.13 (1.09, 1.18)	6.3	2.38 (2.21, 2.56)	1.80 (1.65, 1.96)
Marital status
Married	14.1	1.00	1.00	4.4	1.00	1.00
Never married	9.0	0.70 (0.68, 0.73)	0.84 (0.81, 0.87)	3.0	0.88 (0.82, 0.94)	1.10 (1.02, 1.18)
Widow/separated/divorced	12.2	0.83 (0.79, 0.87)	0.99 (0.94, 1.04)	3.6	0.76 (0.70, 0.84)	0.91 (0.82, 1.00)
Dependent children
0	11.4	1.00	1.00	3.7	1.00	1.00
1	11.7	1.05 (1.01, 1.08)	0.96 (0.92, 0.99)	3.2	0.90 (0.84, 0.97)	0.87 (0.81, 0.95)
2+	14.6	1.31 (1.28, 1.35)	1.07 (1.03, 1.10)	4.5	1.26 (1.18, 1.35)	1.09 (1.01, 1.17)
Caregiving
None	12.2	1.00	1.00	3.5	1.00	1.00
Light	14.1	1.09 (1.05, 1.13)	1.10 (1.06, 1.15)	4.4	1.09 (1.01, 1.18)	1.08 (0.99, 1.17)
Heavy	13.3	1.01 (0.95, 1.08)	1.08 (1.01, 1.16)	6.9	1.70 (1.52, 1.89)	1.66 (1.49, 1.86)
Occupational class
Managerial/prof	11.9	1.00	1.00	3.8	1.00	1.00
Intermediate	5.6	0.45 (0.43, 0.48)	0.46 (0.43, 0.49)	1.2	0.32 (0.28, 0.35)	0.32 (0.29, 0.35)
Own-account workers	32.0	3.42 (3.33, 3.52)	3.42 (3.33, 3.52)	21.4	6.38 (5.95, 6.85)	6.40 (5.96, 6.88)
Routine occupations	6.0	0.48 (0.46, 0.49)	0.49 (0.47, 0.50)	2.8	0.73 (0.68, 0.79)	0.73 (0.68, 0.79)

^aAdjusted for age, marital status, number of dependent children, caregiving duties and occupational class. N/A, not applicable for this group.

Over the 8.7 years of follow-up, the total group of economically active and inactive men and women experienced 21 034 deaths. In models adjusted for age, marital status and socio-economic status (results not shown in Table 3), women were half as likely to die as men (HR 0.46: 95% CIs 0.45, 0.48), and those not currently married had a 27% excess compared with those who were married. In the fully adjusted model, each of the indicators of socio-economic status, educational attainment, housing tenure and car ownership remained as an independent predictor of mortality risk. Mortality risk was lowest for economically active men and women working ≥35 h/week and highest for those who were economically inactive. In models adjusted for age, marital status and socio-economic status, the hazard ratio for economically inactive men, compared with men working ≥35 h/week was 2.71 (95% CIs 2.60, 2.82)—the equivalent result for women was 2.14 (95% CIs 1.99, 2.30). Economically active men working for less than ≥35 h/week also had a higher mortality risk than their full-time working peers (HR 1.40: 95% CIs 1.32, 1.50), but no mortality differences were recorded for women (HR 0.98: 95% CIs 0.91, 1.07). Table 3 shows the mortality associated with hours worked, with 4447 male and 1143 female deaths occurring over the 8.7 years of follow-up. In models adjusted for age and marital status, men and women working 41+ h/week tended to have a 10% lower mortality risk than those working 35–40 h/week. However, these differences were either eliminated or greatly attenuated by adjustment for variation in socio-economic status. Further adjustment for dependent children, caregiving responsibilities and baseline levels of LLTI and general health resulted in a further minor attenuation in risk across work patterns. The final model records no difference in mortality risk between men working 55+ h/week and those working fewer hours; women working 49–54 h/week had a 17% higher risk than those working a normal week, and those working 55+ h/week had a 14% lower risk, but both of these are based on small numbers of events and the wide confidence intervals for both estimates encompass unity. In the fully adjusted model, the indicators of socio-economic circumstance and health status make independent contributions to risk. In relation to occupational social class, it is notable that the mortality associated with own-account workers, who work the longest hours, is only a little higher than that of the managerial/professional workers (HR 1.08: 95% CIs 0.95, 1.25) and in models where hours worked are included as the final element of the model, this small excess is not altered by the inclusion of hours worked. The equivalent relative risk for own account female workers is lower (HR 0.95: 95% CIs 0.72, 1.25).

Table 3

All-cause mortality risk associated with number of hours worked for 270 011 men and 144 938 women in Northern Ireland aged 20–59/64 years. Data represent hazard ratios (95% confidence intervals) from Cox proportional hazards modelling

	Deaths	Adjusted for age & marital status	+Socioeconomic status	+Dependent children and caregiving	+Limiting long-term illness and general health
Men
Hours/week
35–40	2800	1.00	1.00	1.00	1.00
41–48	663	0.92 (0.85, 1.00)	0.96 (0.88, 1.04)	0.96 (0.88, 1.04)	0.96 (0.88, 1.05)
49–54	418	0.91 (0.82, 1.01)	1.01 (0.91, 1.12)	1.01 (0.91, 1.12)	1.02 (0.92, 1.13)
55 and over	566	0.88 (0.80, 0.96)	0.96 (0.87, 1.05)	0.97 (0.88, 1.06)	0.97 (0.88, 1.07)
Women
Hours/week
35–40	950	1.00	1.00	1.00	1.00
41–48	99	0.93 (0.75, 1.14)	0.99 (0.80, 1.22)	0.98 (0.79, 1.21)	0.98 (0.80, 1.21)
49–54	49	1.03 (0.78, 1.38)	1.18 (0.88, 1.58)	1.18 (0.88, 1.58)	1.17 (0.87, 1.57)
55 and over	45	0.80 (0.59, 1.08)	0.86 (0.63, 1.18)	0.87 (0.64, 1.19)	0.86 (0.63, 1.17)

Further analysis showed that the relationship between hours worked and mortality risk varied by social class for men (chi square 24.48; P = 0.004) but not for women (chi square 7.51; P = 0.584). The relationship was not influenced by age, baseline physical ill health, dependent children or caregiving status for either sex (results available on request). Figure 1 shows the results of fully adjusted analyses stratified by social class and sex. For men, the managerial/professional workers working >40 h/week recorded lower mortality risks than those working fewer hours, with the risk for those working 55+ h/week at 0.77 (95% CIs 0.63, 0.94). For own-account workers, no relationship between hours worked and mortality risk was evident. However, men in routine occupations and working 55+ h/week recorded a 31% excess risk (95% CIs 1.11, 1.55) compared with their occupational peers working <40 h/week. There was a suggestion of a dose-response between hours worked and mortality risk for this group, with point estimates for 55–59, 60–69, 70–79 and 80+ h/week at 1.17, 1.18, 2.04 and 1.52, respectively. A pattern of increasing mortality with longer working hours was evident for those in the intermediate class but here the numbers were smaller and all confidence intervals included unity. The wider confidence intervals associated with the hazard ratios for women are a consequence of the smaller number of deaths. However, the point estimates suggest that there is no evidence of an increased mortality risk associated with long working hours for the managerial/professional or intermediate class workers but perhaps some suggestion of an increased risk for own-account workers and those in routine occupations.

Figure 1

All-cause mortality risk by average hours worked and stratified by occupational group for 270 011 men (4447 deaths) and 144 938 women (1143 deaths) in Northern Ireland. Data represent hazard ratios (95% CIs) in models fully adjusted for the covariates listed in Table 1

Table 4 shows cause-specific mortality associated with long working hours, stratified by occupational group. For brevity, results shown compare those working 55+ h/week with those working 35–40 h/week (the reference group): full results are available on request. About 25% of deaths in the cohort were due to cardiovascular disease (of which 68% were due to ischaemic heart disease) and 47% due to cancer. The increased mortality associated with working 55+ h/week is evident for male workers in routine occupations who, compared with peers working least hours, record excess mortality from cardiovascular disease (HR 1.49: 95% CIs 1.10, 2.00), and a more than 2-fold excess from respiratory causes (HR 2.67: 95% CIs 1.42, 5.01). For this group there was no increased mortality risk associated with malignant neoplasms, external causes, accidents or suicides. For men working longer hours, where analysis was possible the estimates for alcohol-related causes of death indicated lower relative mortality risks across all occupation groups. The equivalent results for women are also presented, but the numbers of deaths included in over half the individual occupation categories were too small to analyse and the confidence intervals for the remaining categories are wide.

Table 4

Cause-specific mortality risk (hazard ratios and 95% confidence intervals) for men and women in Northern Ireland working 55+ hours/week, compared with those working 35–40 h/week, for each of the four occupational groups

		Occupational group
Cause of death	Deaths (no.)	Managerial / professional	Intermediate	Own account	Routine occupations
Men (total)		94 848	21 171	45 993	107 999
All cardiovascular	1240	0.95 (0.69, 1.34)	0.55 (0.13, 12.26)	0.96 (0.72, 1.29)	1.49 (1.10, 2.00)
IHD	897	0.82 (0.55, 1.23)	0.36 (0.05, 2.61)	1.24 (0.88, 1.74)	1.53 (1.08, 2.17)
Stroke	153	1.43 (0.51, 3.97)	n/a	0.52 (0.23, 1.20)	2.65 (1.28, 5.50)
Respiratory disease	169	0.42, 0.13, 1.38)	2.54 (0.30, 21.3)	1.39 (0.56, 3.46)	2.67 (1.42, 5.01)
All cancers	1897	0.78 (0.58, 1.04)	1.30 (0.60, 2.84)	0.93 (0.73, 1.18)	1.24 (0.96, 1.61)
Alcohol-related	180	0.47 (0.14, 1.53)	n/a	0.52 (0.18, 1.53)	0.59 (0.21, 1.60)
All external causes	630	0.88 (0.50, 1.56)	1.16 (0.41, 3.31)	0.74 (0.48, 1.15)	1.14 (0.72, 1.80)
All accidents	304	0.44 (0.13, 1.44)	2.51 (0.71, 8.94)	0.61 (0.33, 1.14)	1.10 (0.56, 2.18)
Suicide	308	1.23 (0.63, 2.39)	0.45 (0.06, 3.37)	0.88 (0.46, 1.66)	1.24 (0.67, 2.31)
Women (total)		68 291	33 367	6919	36 361
All cardiovascular	170	1.42 (0.44, 4.60)	n/a	3.41 (0.83, 13.9)	1.13 (0.27, 4.71)
Ischaemic heart disease	60	1.25 (0.16, 9.43)	n/a	1.35 (0.77, 23.9)	1.93 (0.44, 8.47)
Stroke	62	1.07 (0.14, 8.00)	n/a	2.98 (0.48, 18.5)	n/a
Respiratory disease	34	n/a	n/a	n/a	3.07 (0.38, 24.9)
All cancers	708	1.10 (0.64, 1.90)	0.92 (0.23, 3.73)	0.86 (0.39, 1.90)	0.49 (0.18, 1.32)
Alcohol-related	54	n/a	n/a	2.07 (0.33, 12.9)	n/a
All external causes	86	n/a	n/a	n/a	1.43 (0.19, 10.8)
All accidents	44	n/a	n/a	n/a	n/a
Suicide	41	n/a	n/a	n/a	2.41 (0.30, 19.2)

IHD, Ischaemic heart disease; n/a, numbers of deaths too small to provide meaningful results.

Discussion

This study finds some evidence that working longer hours increases mortality risk, which was concentrated mainly in men in technical, semi-routine or routine occupations and only for those working for 55+ h/week. These findings require some consideration, though the case for a causal relationship is strengthened by an indication of a dose-response relationship above this threshold.³³ The findings also appear plausible, as 55+ h/week entails either an arduous 5-day week and/or a longer working week, with less time away from work impacting significantly on work-life balance. The findings also mirror the initial Japanese studies where the average hours worked were at least at this level.^6,34 The authors suggest that it is not possible to draw any safe conclusions from the current study about a possible mortality risk associated with long working hours for women, and most of the remaining discussion related therefore to the effects on men.

The variation in mortality risk by occupational class suggests that employment relations or conditions of occupation are significant moderators of the relationship between long working hours and mortality risk and these may relate to the degree of individual preference or choice and extent of control the worker can exercise. There is some evidence that the stresses and physical health symptoms associated with long working hours are most apparent where there is little flexibility in the choice of overtime working or autonomy in terms of how many hours are worked and when they are worked,^35–37 or in situations where overtime is necessitated by personal economic circumstances.¹⁶ The cardiovascular effects of long working hours are also more likely in those with lower levels of decision authority.³⁸ It is likely, though not proven, that these conditions are more prevalent in the technical, semi-routine or routine occupations. In the current study, working long hours was not associated with increased mortality risks for men in self-employed, small-employer or managerial and professional occupations—the groups that collectively are responsible for the majority of the overtime worked. However, these findings are also in keeping with the prevailing theories of work-related psychosocial stressors,^39,40 as it is presumed that own-account workers experience less effort-reward imbalance and social surveys indicate that, although the majority of overtime in managerial or professional occupations is unpaid, much of it is perceived as ‘part of the job’ or is undertaken in expectation of future advancement.⁴¹

The study also demonstrated that men in routine occupations working longer hours experienced a 49% excess in mortality from cardiovascular disease, but no increased mortality risks associated with cancer or accidents. One meta-analysis of observational studies examining the relationship between long working hours and coronary heart disease found an adjusted relative risk of approximately 80% for all studies and 40% for prospective studies, though there was evidence that the associations were stronger when the analysis was limited to men and where higher cut-offs were used to define excessive working hours.²⁴ The mechanisms underpinning this are unclear but the authors provide a comprehensive list of potential causes, including disruption of the hypothalamic-pituitary-adrenal and sympatho-adrenomedullary axes, increased cardiovascular risk factors associated with adverse lifestyle and poor sleep patterns. Risk of death due to respiratory causes was also increased and might be due to the recognized association between long working hours and adverse lifestyles.^16,17 The lower mortality risks associated with alcohol is not unexpected, as many men with a significant alcohol problem may have left the labour market.⁴³

This study has significant strengths and limitations. Its strengths are that it relates to the entire working population rather than selected subgroups and that cause of death is derived from validated records. The inclusion of NS-SEC classification, which reflects position in the labour market and work conditions, enabled an examination of variation in mortality risk across these dimensions. An additional strength is the inclusion of measures of homemaking and caregiving responsibilities which, though unpaid, can justifiably be included as work-related activities and implicitly intensify any effects of long working hours.

The caveats concern the information limitations of the Census. Respondents were asked only about the hours worked in their main job, so additional hours worked in subsidiary occupations are not included. Arguably though, the wording of the census question means that both paid and unpaid overtime, known to vary by occupational group,⁴⁴ are likely to be included. Also, for this study, it was not possible to include time spent travelling to and from work though, given the size of Northern Ireland, this omission may not be crucial for the majority of workers. A major concern, and one that cannot be dismissed, is that a proportion of working men in the reference group may curtail their hours because of poorer health⁴⁵ and that comparative mortality differentials represent a healthy worker effect. However, unlike other studies, the analysis population here is restricted to men working ≥35 h/week²⁴ (a ‘normal’ working week) and this can act to reduce the likelihood of health selection effects. Additionally, baseline self-reported health levels and mortality risks indicate that any health selection effects occur mainly between the economically active and the economically inactive groups; the higher mortality risk among part-time workers is in line with an earlier study of Sokejima and Kagamimori.⁴⁶ Among those working a full week, there was no relationship between levels of LLTI or general health and the number of hours worked. Furthermore, adjusting for health status at baseline did not attenuate the relationship between length of working week and mortality risk, and stratification according to health status did not show any variation in mortality risk according to health status at baseline.

A further caveat is that employment status and hours worked were collected at only one point in time (the Census) and the study therefore assumes their persistence. The Census did not collect data relating to shift work, which has been associated with higher risk for cardiovascular disease⁴⁷ and for a range of cancers,⁴⁸ and it is possible that some of the excess mortality recorded here is related to the irregularity rather than the number of hours per se.

These findings confirm and extend the finding of earlier studies showing associations between long working hours and different facets of mental and physical ill health, on work-life balance and on other family members.^5,12,24,33 In Northern Ireland the public sector is a major employer⁴⁹ which, with its more defined work patterns, means there is a smaller proportion of the workforce working excessive hours. However, the issue of long working hours may be of particular importance in the rest of the UK as the average number of hours worked by those in full-time employment are traditionally higher than most other EU countries⁴⁴ and, although in Europe the Working Time Directive limits the working week to 48 h, the UK has an opt-out whereby employees may work longer hours but cannot be forced to do so. The effects of long working hours are likely to be more apparent in other countries such as Japan where more extreme working weeks are not uncommon,⁶ or where the facility of leaving the labour market due to ill health afforded by the availability of statutory benefits is not available.

Although research suggests that the effects of longer working hours may be ameliorated by introducing flexibility in working patterns that give workers more choice or control over their working arrangements,³⁴ or by ensuring that the rewards are commensurate with the effort²⁵ or by maintaining a healthy lifestyle,⁵⁰ we suggest the requirement for such long working hours should be always be examined, given the poor correlation between productivity and hours worked.

Funding

The NIMS is funded by the Health and Social Care Research and Development Division of the Public Health Agency (HSC R&D Division) and NISRA. The NILS-RSU is funded by the ESRC and the Northern Ireland Government.

Acknowledgements

The help provided by the staff of the Northern Ireland Longitudinal Study/Northern Ireland Mortality Study (NIMS) and the NILS Research Support Unit is acknowledged. The authors would also like to thank the four anonymous reviewers for their constructive comments on this paper.

The authors alone are responsible for the interpretation of the data, and any views or opinions presented are solely those of the author and do not necessarily represent those of NISRA/NILS.

Conflict of interest: None declared.

KEY MESSAGES

• Long working hours are associated with a range of adverse health consequences but there is little empirical evidence to substantiate earlier case series suggesting an increased mortality risk.

• There is also evidence that occupationally-related cardiovascular and other health effects are more apparent when the occupation is characterized by higher demands and low levels of control or a perceived imbalance between effort and reward.

• Using a census-based longitudinal study, we demonstrate an increased mortality risk in the UK but only amongs men working >55 h/week and employed in routine or semi-routine occupations. The mortality risk for cardiovascular disease was particularly marked for this group.