Affiliations: Section of Personal Injury Prevention, Department of
Clinical Neuroscience, Karolinska Institutet, Stockholm, Sweden | Department of Health and Society, Division of Social
Medicine and Public Health, Linköping University, Sweden
Note: [] Address for correspondence: Jan Hagberg, Section of Personal
Injury Prevention, Department of Clinical Neuroscience, Karolinska Institutet,
171 77 Stockholm, Sweden. Tel.: +46 8 524 832 11; Fax: +46 8 524 832 06;
E-mail: [email protected]
Abstract: Several methodological challenges arise when attempting to analyse
individual data on changes of sick-leave diagnoses over several years.
Sick-leave spells for a person can recur, have different sick-leave diagnoses,
and both these aspects are dependent of previous episodes, the numbers of
repeated periods vary across subjects, and standard statistical methods are not
valid for variables on nominal scales, e.g. sick-leave diagnoses. Objective: Our aim was to ascertain whether the number and pattern of changes
in sick-leave diagnoses are associated with future disability pension (DP) and
to test methods for analysis of repeated measurements on nominal data. Participants: Data from a 12-year prospective cohort study of the 8000
sick-leave periods of the 213 persons aged 25–34 who, in 1985, had a new
sick-leave spell ⩾ 28 days with back diagnoses were
used. Methods: We used entropies, uncertainty coefficients adjusted for
repeated measurements, and transition matrices to examine the changes in
sick-leave diagnoses that occurred during follow up. Results: In the
12 years 22% were granted DP and they had changed sick-leave diagnosis less
frequently and more often had new sick-leave periods with musculoskeletal
diagnoses than the others. The variation in diagnoses and the degree of
dependence between consecutive diagnoses were associated with DP. Conclusions: Many tools in statistics are based on linear methods that require
numerical variables, but such methods are not valid for repeated measurements
on discrete variables on nominal scales, as for sick-leave diagnosis. In such
cases, it can be beneficial to use tools that are applied in statistical
information theory.
