There are an increasing number of studies using self-reported sickness absence as an outcome. The value of such retrospective self-reported information depends on the quality of the data. Poor quality may lead to biased results of the studied associations due to differential or non-differential misclassification. Studies of the quality of different types of sickness absence data are, however, scarce and there is a need for more knowledge on this matter.1

Sickness absence has a multifactorial background that comprises risk factors at different structural levels (for example, societal, demographic, work-related and individual factors),2 and it has been described as a global measure of health in a working population.3 4 Both self-reported and employer-registered sickness absence have been found to be associated with self-reported general health5^–7 and mortality,4 5 8 and poor self-reported health is also a predictor of subsequent mortality.4 9 10

Earlier studies that have validated self-reported information of sickness absence with register data have mostly been based on small populations, between 150 and 600 employees within different occupations (coal miners, office and production work).11^–15 Several of these studies have focused on sickness absence due to back pain.11^–13 One recently published larger study with focus on sickness absence irrespective of diagnoses was based on the Whitehall cohort of about 8000 civil servants in London.7

In some investigations, the sensitivity of applying self-reported sick-leave data to detect the occurrence of sickness absence, using register information as the standard, has varied between 55% and 88% during follow-up periods of one month to four years.11–13 15 These studies reported specificities ranging from 83% to 98%. In a Dutch study of five recall periods (2 and 4 weeks, and 2, 6 and 12 months), the cumulative proportions displaying a maximum difference of two days between self-reported and recorded sickness absence were 95%, 95%, 87%, 57% and 51%, respectively—that is, a decrease with increased length of the recall periods.14 Moreover, the larger Whitehall study compared number of self-reported sick-leave days over the last 12 months with data from the employers’ registers, and a maximum discrepancy of two sick-leave days (adjusted for employment grade and number of sickness absence days) was found for 60% of the women and 73% of the men.7 Analyses also revealed that both self-reported and recorded sickness absence showed a similar association with self-reported health.

The target group of the present study is employees in the public sector in Sweden, about one million people, of whom about 80% are women. Almost half of the female work force is employed in this female-dominated public sector and about 50% are over 45 years of age. This heterogeneous group of various types of occupations involves both physically and mentally demanding work mainly within the areas of health care, social welfare, and education—so-called “human service” occupations. Swedish national data on sickness absence show that public sector employees, specifically those working in the municipalities, are over-represented among those on sick leave for 60 days or more per annum.16 The intention with the current study was partly to replicate the Whitehall study,7 but on a different population—mainly women in human service occupations, with a different age distribution, and with higher levels of sickness absence.

The aim of the present study was to compare self-reported information on sick-leave days in the last 12 months with sickness absence data from employers’ registers, irrespective of diagnoses. Furthermore, as in the Whitehall II study, another goal was to ascertain whether the self-reported sickness absences and the register data from the employers show the same degree of association with self-reported general health.

METHODS

Study population

The investigation was conducted within the framework of the Swedish study entitled “Work and Health in the Public Sector” (denoted HAKuL).17 The data used covered a cohort of employees at local authorities in five municipalities and three county councils in different parts of Sweden. A questionnaire was administered in 2001–2 to 7229 employees who were deemed eligible because they had permanent employment and were not on long-term sick leave (>90 days) at the time the information was collected. The response rate was 75% (n = 5411). For 90% (n = 4869) of those employees, we had access to complete register data on employment status and absence from work during the 12-month study period. Self-reported sickness absences were missing for 122 individuals, and the final study population consisted of 4747 employees (84% women) aged 19–65 years, with mean ages of 45 and 46 for women and men, respectively.

Considering occupations, most of the women were teachers, preschool teachers, child care centre staff, nurses, assistant nurses, home care workers in elderly care, cleaners and kitchen staff, and carers of mentally retarded persons. Most of the men were teachers or various types of production and maintenance workers, including operating personnel, construction workers, and park wardens. Information on occupation and age was derived from employers’ registers.

Measures of sickness absence

The number of self-reported sick-leave days in the last 12 months was derived from the following question: “How many days, approximately, have you been off work for health reasons the last 12 months?”

Register data on all spells of sickness absence were obtained from the employers’ computerised personnel records for the 12 months prior to the questionnaire survey. The date of the first and the last day of each spell of absence was available for individual employees.

Overlapping, consecutive, or duplicate sick-leave spells were merged into a single spell. All calendar days registered as a sick-leave day were counted, regardless of whether they were work days, and sickness absence part of the day was defined as a sick-leave day. For each employee, the numbers of sick-leave days in the 12 months before the questionnaire were computed by considering the date on which the questionnaire was returned as the starting date for calculating sick-leave days.

All subjects in this study were covered by the Swedish national health insurance, which provides sickness benefits to those who fulfil the requirement of being unable to work due to disease or injury. At the time of this study, compensation for the first 14 days of a sick-leave spell was paid by the employer, with the exception of the first qualifying (“waiting”) day, for which no benefit was paid. The first seven days of absence in a sick-leave spell could be self-certified; thereafter a medical certificate was required. All sickness absence, both self-certified and certified by a physician, was included in the present study.

Self-reported general health

Self-reported general health was measured by a single item on the questionnaire: “In general, how would you describe your health?” The response options were excellent, very good, good, fair, or poor, and the answers fair and poor were considered to indicate “poor general health”.

Statistical analyses

Mean annual self-reported and recorded sick-leave days stratified by sex, age and occupational skill level were calculated, and the differences between the two absence measures are presented with 95% confidence intervals. Analysis of variance was carried out to determine whether the differences between the mean number of self-reported and recorded sick-leave days per year for women and men were associated with age, occupational skill level, and number of recorded sick-leave days. In addition, the proportions of women and men with a maximum discrepancy of two days between the two absence measures were calculated.

The total numbers of self-reported and employer-registered sick-leave days were stratified into the five categories: 0, 1–7, 8–14, 15–27 and ⩾28 days, which are presented in cross tables showing the proportion of complete agreement for each category (tables 2 and 3). Estimates of the sensitivity and specificity of the self-reported sick-leave days over the last 12 months were calculated on the assumption that the register information from the employers was correct.

Logistic regression analyses were performed to study the associations between self-reported general health and the five categories of self-reported and recorded sick-leave days. The odds ratios were adjusted for age, sex and occupational skill level. Occupational skill level was classified according to the International Standard Classification of Occupations (ISCO-88), and the occupations were stratified into four skill levels: managers and university or postgraduate university degree, occupations that require 1–3 years of education after high school, skilled workers with vocational training and unskilled workers.

RESULTS

The numbers of self-reported sick-leave days was lower than the corresponding employer recorded numbers for both women and men, among different age groups, and in the four groups of occupational skill level (table 1). No discrepancies in the difference between recorded and self-reported mean annual number of sick-leave days were found in the four groups of occupational skill level among women. However, among men in occupations requiring 1–3 years of education after high school the difference between recorded and self-reported mean annual number of sick-leave days was lower (one day) compared with the difference for skilled workers with vocational training (5.8 days). The sex differences in mean annual number of sick-leave days according to the two different methods of recording remained significant after adjustment for age and level of occupational skills (results not shown). Additional adjustment for recorded number of sick-leave days reduced this difference somewhat (p = 0.099).

Table 2 presents the total numbers of self-reported and recorded sick-leave days for women divided into five categories. The greatest overall agreement between the two sickness absence measures for women was found for no sick-leave days, 1–7 days and ⩾28 days, as indicated by 74%, 72% and 67% agreement, respectively. In all, 91% (2335) of the 2566 female employees who had any sickness absence according to the register data also reported at least one sick-leave day on the questionnaire (sensitivity). Among women with no recorded sick-leave days (n = 1402), 74% reported that they had not been on sick leave (specificity). The sensitivity of ⩾28 recorded sick-leave days was 67% with a specificity of 98%.

The results for men were similar to those for women, showing substantial total agreement of 78%, 71% and 74% with the categories no sick-leave days, 1–7 days and ⩾28 days, respectively (table 3). Moreover, the sensitivity for any number of sick-leave days was the same as for women, 91%, with a specificity of 78%. For ⩾28 recorded sick-leave days, the sensitivity was 73% and the specificity 99%.

Crude and adjusted analyses of the associations between poor self-rated health and self-reported and recorded number of sick-leave days are presented in table 4. The point estimates of the associations with poor self-rated health were slightly higher for self-reported days than for recorded days but there was no real difference in the strength of the associations. The strongest associations were found among those with ⩾28 self-reported days (adjusted OR 8.25, 95% CI 6.62 to 10.29) and ⩾28 recorded days (adjusted OR 7.27, 95% CI 5.93 to 8.91).

DISCUSSION

In this study self-reported sickness absence collected using a questionnaire was compared with sickness absence data compiled by employers. The results show that the number of self-reported sick-leave days was lower than the recorded number of sick-leave days over a period of 12 months. Analyses of five categories of sick-leave days using sickness absence data from employers’ registers as the standard indicated good agreement between self-reported and recorded information for any self-reported sick-leave day, 1–7 sick-leave days and sickness absence of 28 days or more. Both measures of sickness absence were associated with self-rated health, and this association was strengthened with increasing number of sick-leave days.

Methodological considerations

Since the study population originated from the longitudinal HAKuL study, which included interventions at the workplaces, the selection of participating county councils and municipalities was not random. However, the cohort represented all kinds of activities performed by county councils and municipalities in different parts of Sweden, including both rural and urban areas, although there was some predominance of home care and healthcare workers.17 Excluding employees on long-term sick leave when the study started implies a relatively healthy study population and a lower proportion of employees in the category ⩾28 sick-leave days. However, we do not think that this has affected the agreement between self-reported and registered number of sick-leave days or the association between sickness absence and self-reported general health.

An advantage of our study was the high questionnaire response rate of 75%. High validity of the employers’ registers of sickness absence could be assumed, because these records are the basis for calculating salaries. The mean number of recorded sick-leave days was greater for employees who had information on self-reported sick-leave days (16.9 days) than for those for whom such information was missing (12.6 days). The starting point for calculating the number of recorded sick-leave days was the date when the completed questionnaire was returned. This dislocation of the follow-up periods might have led to a few days of discrepancy between the two sickness absence measures, but that would have had only a minor effect on the results.

Several different measures of sickness absence can be used, and they may provide dissimilar results.1 Therefore, the definition of a measure of sickness absence is of importance. In Sweden it is possible to be on part-time sick leave for 25%, 50% or 75% of working hours. The proportion of part-time sick leave was 26% of all sick-leave days in 2001 and this was more common among women than among men.18 However, part-time sick leave is rare during the first 14 days of sick leave—about 5% of these days are part-time sick-leave days compared to more than 35% of the sick-leave days in spells of more than 180 days.18

In the present study all sick-leave days recorded by employers were regarded as full days, and, if part-time sick-leave days were added together by the respondent to yield full days, this may have resulted in an under-reporting of sick-leave days in the questionnaire compared to the days entered in the register. We do not know to what extent this actually occurred in our study, but it may be one explanation for the lower numbers of self-reported sick-leave days in the questionnaire versus the register. This emphasises the need for clear instructions in a questionnaire/interview regarding how to report part-time sick-leave days. Consequently, we agree with Johns,19 who recommended that pre-coded responses be avoided and that the responders should be requested to report the actual number of sick-leave days.

The mean number of sick-leave days per person is not a common epidemiological measure, regardless of whether the data are self-reported or obtained from a register. Notwithstanding, this is often used as a complementary measure in studies of sickness absence. In most populations, the distribution of sickness absence data is skewed such that the majority of the people in a population have only a few or no sick-leave days, whereas a minority generates the greater part of the sickness absence.1 In the present study, we categorised the number of self-reported and the number of employer recorded sick-leave days into five different categories of sick-leave days in addition to calculating the mean number of sick-leave days. Either way, our results indicate that self-reported data tend to underestimate the actual number of sick-leave days.

Levels of sickness absence

A common problem with epidemiological studies on sickness absence is the lack of comparable studies to verify results from earlier studies. Six earlier studies have compared self-reported and register-based information on sickness absence,7 11^–15 and in five of those investigations the recorded data originated from employers, whereas in the sixth study information from social insurance offices was used as the standard.13 The largest of the studies focused on the British Whitehall cohort, and involved about 8000 civil service employees in London.7 The Whitehall cohort differed from our cohort in that Britain has a different social security system and the study population was restricted to white-collar workers aged 35–55, mostly men and participants with lower overall levels of sickness absence.

The present study based on public employees, mainly women in human service occupations, aged 19–65, had a higher mean number of sick-leave days indicated by recorded data compared with the Whitehall investigation.7 This finding was expected because Sweden, together with Norway and the Netherlands, has one of the highest levels of sickness absence in Europe.20 In accordance with other studies, we also observed a higher mean number of sick-leave days for women than for men, and for those with lower occupational skills compared to those with a higher education.2

Discrepancy between self-reported and registered sickness absence

The sex differences between the mean annual number of recorded and self-reported sick-leave days can be explained by the fact that women have more extensive sickness absence than men in Sweden. This agrees with the findings of the Whitehall study, which show higher mean numbers of sick-leave days for women as well as for employees in lower occupational grades.7 It was not possible to compare our results with the other five of the six studies mentioned above with respect to the employees occupational grades, because those investigations were based on small and relatively homogeneous samples of employees.11^–15 Longer spells of sickness absence in our study might also explain why a maximum discrepancy of two days between mean number of self-reported and recorded sick-leave days was seen among smaller proportions of women and men in our investigation (56% and 67%, respectively) than in the Whitehall study (60% and 73%).7

The best overall agreement between self-reported and recorded sick-leave days (67–78%) was found in the groups that had no sick-leave days, 1–7 sick-leave days, and ⩾28 sick-leave days. Substantial overall concurrence was also observed in the Whitehall study for those with no sick-leave days (about 60%), except for men in the lower employment grades, almost 80% of whom showed overall agreement.7 One of the six investigations mentioned above also found a high level of overall agreement (73%) for the category no sick-leave days, although the greatest concurrence (87%) was found for >14 sick-leave days.12 However, the recall period for self-reported sickness absence in that study was only six months.

Sensitivity and specificity

The calculations of sensitivity measure to what extent the self-reported data identify the actual number of sick-leave days with the employers’ registers as the reference. These analyses showed rather high validity for any sick-leave day, with a sensitivity of 91% for both women and men. However, the sensitivity was lower—67% for women and 73% for men—when considering sickness absence of ⩾28 days. In earlier investigations, the sensitivities ranged from 55% to 88% for follow-up periods varying between one month and four years.11–13 15 One aim of those relatively small studies was to evaluate sick leave related to back pain, and different measures of sickness absence were used. In the present work we analysed five different categories of sick-leave days, whereas Fredriksson et al13 studied spells comprising >7 consecutive days, Van Poppel et al15 used the measure of short (⩽7 days) and long (>7 days) episodes of sick leave, and Agius et al restricted their analyses to spells lasting two days or more.11 The lower sensitivity for self-reported sick leave of 8–27 days in our study indicates that caution should be exercised when using very detailed measures of self-reported sick-leave days in epidemiological studies.

The specificity was the probability of being able to correctly classify employees as having no sick-leave days or not having ⩾28 sick-leave days based on self-reported information. The results of our study are similar to the findings of the above-mentioned studies, with specificity being higher for ⩾28 sick-leave days (almost 100%) than for not having any sick-leave days (74% for women and 78% for men).

Sickness absence and self-rated health

The associations between self-rated health and self-reported and recorded numbers of sick-leave days clearly demonstrated a decrease in self-reported general health with increasing numbers of sick-leave days. These results are well in accordance with the Whitehall study in which similar relations were found.7 The Whitehall study also discovered that annual numbers of self-reported and recorded sick-leave days were related to prevalent coronary heart disease, which was used as a more objective measure of health, and hence reporting bias should not be a significant problem. A possible explanation for the tendency towards a stronger association between self-reported health and self-reported sick-leave days could be that employees who report poor health also indicate more sick-leave days or vice versa. However, we do not think that was the case in our study, since the recorded and self-reported sick-leave days showed similar associations with self-reported general health.

There are some plausible reasons for the differences we observed between the mean annual days of self-reported and recorded sickness absence. It is probably more difficult to recall the exact number of many and repeated sick-leave days than to remember fewer sick-leave days, which might partly explain why the mean annual difference between self-reported and recorded sick-leave days was greater for women than for men in our study. Furthermore, it might be that sick-leave days during vacation and parental leave are not reported and it might be hard to remember correctly whether the sick-leave days occurred within the specific period of interest, particularly when using longer recall periods (for example, over 12 months).

Differences between the two measures of sickness absence might also be due to factors other than imperfect reporting of sick-leave days on the questionnaire. A likely example of this that was pointed out in two earlier studies is inaccurate coding in the records of sickness absence, seen as missing, fewer or too many sick-leave days in the registers.13 15 We assume that this was not a problem in our investigation, as the employers in question use the register data to calculate the employees’ salaries, and also because sickness benefit for the first 14 days of a sick-leave spell is paid by the employers. For most employees, sickness benefit is equivalent to 90% of the ordinary salary, but the first day of a sick leave spell is a qualifying day without any compensation.

In summary, it seems that differences in mean annual self-reported and recorded numbers of sick-leave days are due to misclassification, arising mainly as a result of recall bias associated with the total number of sick-leave days. The findings of our study and of the Whitehall investigation show that such misclassification does not depend on health status, nor does it seem to be dependent on the difference in the social security system in the two countries. In all likelihood, information on episodes of other types of absence from work, as well as sickness presenteeism, would greatly facilitate clarification of the discrepancy between self-reported and recorded data on sickness absence.

Another important aspect to consider is that computer-based absence data from employers’ registers are seldom compiled for research purposes, and they often require advanced and expensive preparation to allow analysis. When it is difficult to achieve recorded data on sickness absence, an alternative strategy to achieve information on sickness absence that is also useful in epidemiological studies or in the context of occupational health practice, is to use self-reported data on sickness absence. Furthermore, in light of the current results, we do not agree with the more restrictive proposal that only follow-up periods of 2–6 months can provide acceptable validity in epidemiological studies, as suggested in earlier studies.14 19 Indeed, shorter periods of retrospective follow-up of self-reported sickness absence might introduce other problems, such as seasonal variation in the number of sick-leave days.21

CONCLUSIONS

In this study we found good agreement between self-reported and recorded data on number of all-cause sickness absence days over a period of 12 months, and this was true for both women and men. The use of retrospectively collected self-reported sick-leave days can be very useful in epidemiological studies—for example, when describing the level of sickness absence over time, and also in evaluations of preventions and interventions at workplaces. However, in situations requiring knowledge of the exact number of sick-leave days, for instance when calculating health economic figures, it is preferable to use employers’ registers of sick-leave days as a source of more accurate information.