RANDOMLY SELECTED FROM THE POPULATION ON SICKLEAVE FOR PSYCHOLOG...

3605) randomly selected from the population on sickleave for psychological reasons, to the 3605 healthy work-expected, in the case of the representative study popula-ing employees, thus adhering better to the conditions intion (N = 3677), the cutoff score equals 11, with the samepractice. This study population is called the 'representa-restriction for specificity. Table 2 shows that the cutofftive study population'.score of 11 is located between the 75

^th

and 95

^th

percentileof the distribution of distress scores in the population ofFor use in occupational health practice, it is not onlyhealthy working employees: most healthy employeeshave less distress. In the sample of employees on sicknessimportant to find nearly all employees-at-risk, but also toexclude false-positive employees. Therefore, during theabsence, this cutoff score is located close to the 5

^th

percen-tile, which means that the overall majority exceeds thisevaluation, the establishment of an optimal cutoff pointis based on an optimal trade-off between sensitivity anddistress level in this population.specificity. In a screening situation, however, where theIn the representative study population (N = 3677), a cut-prevalence of absence due to distress is low (2%), specifi-off point equal to or higher than 11 has as a consequencecity is more crucial than sensitivity. Increasing the specifi-city at the expense of sensitivity will lead to a substantialthat 69 of 72 absent employees are correctly classified as

Table 1: Characteristics of the samples 'healthy working employees' and 'employees on sick leave due to psychological complaints'

Age

Gender

Marital status

Level of education

Sample

n

Mean years

SD

% Female

% Married

% Low

% Medium

% High

Healthy working employees

3605

43.9

8.1

9

78

26

49

25

Employees on sick leave due to psychological complaints

280

41.9

8.1

34

66

34

41

25

Table 2: Levels of distress (mean and SD score on 4DSQ distress scale) in the samples of 'healthy working employees' and 'employees

on sick leave due to psychological complaints'

Distress (range 0–32)

Percentile

Sample

n

Mean

SD

5

25

50

75

95

Healthy working employees

3605

4.0

5.0

0.0

0.0

2.0

6.0

14.0

Employees on sickness absence due to psychological complaints

280

22.3

6.7

9.0

18.0

24.0

28.0

31.0

Two issues require some discussion here. One issue is thatbeing absent due to psychological complaints, corre-sponding with a sensitivity of 95%. Within the popula-we used as our study population employees working for atelecom company, which in potential restricts the general-tion of 3605 employees without sickness absence, 3261employees are classified as not distressed, correspondingizability of the cutoff point to other working samples.with a specificity of 90%. The positive predictive value isTherefore, we recommend that more studies be under-.17, whereas the negative predictive value is .998. In addi-taken with a clear reference to the populations studied.tion, Table 3 shows the sensitivity and specificity of alter-native cutoff points.The second issue that should be kept in mind when imple-menting the results of this paper is that psychologicalThe Area Under the ROC Curve (AUC) statistic (Fig 1) hascomplaints range from zero to many, therefore distressbeen obtained by comparing the full range of possible cut-can be best viewed as a continuum as opposed to a dichot-omy. Applying a cutoff point to this continuum poten-off scores. The area under the curve was 0.98, which istially reduces information [51]. If the purpose of a studyexcellent, because in that case the positive likelihood ratio(LR+: the probability to find a positive test result inis to explore the etiology of distress, it is more informativestressed employees compared with employees who areto use a range of distress scores. A dichotomy, however, isnot stressed) is 10 or more and the negative likelihooduseful when the prevalence of distress has to be comparedratio (LR-: the probability to find a negative test result inin different subgroups or when employees have to beselected for stress management or treatment.not stressed) is 0.1 or less. This means that employeesUnfortunately, there is no other study to compare with,who score above the chosen cutoff score are far morelikely to report sick compared with employees who scorewhich reported a cutoff point based on the AUC statisticunder the cutoff score.for identifying cases of sickness absence related to distressin a working population. It is noteworthy that the use of acutoff point for inclusion in preventive stress manage-DiscussionIn the present study, a cutoff point ≥ 11 was chosen for thement programs has not often been reported until now.Moreover, in the meta-analysis of van der Klink et al [34],distress scale of the 4DSQ to measure distress in a workingpopulation. This cutoff point corresponds with a sensitiv-only four studies out of forty-eight involved participantselection with regard to high baseline stress levels.ity of 95% and a challenging specificity of 90% and nega-tive predictive value of .998, and indicates a distress levelthat puts an employee "at risk" for subsequent sick leaveThe choice of a cutoff point of 11 results in a measure thaton psychological grounds.can be used as a cutoff point in future studies on distress

Table 3: Sensitivity and specificity of alternative cutoff points in the 'equal sample population' and the 'representative sample

population'

Equal sample population (n = 560): 280 healthy working employees plus

Representative sample population (n = 3677): 3605 healthy working

280 employees on sick leave for psychological complaints

employees plus 72 employees on sick leave for psychological complaints

Cutoff

Sensitivity

Specificity

Cutoff

Sensitivity

Specificity