Background:
Self-reported disability pension (DP) and sickness absence are commonly used in epidemiological and other studies as a measure of exposure or even as an outcome. The aims were (1) to compare such self-reports with national register information in order to evaluate the validity of self-reported DP and sickness absence, and (2) to estimate the concordance of reporting behaviour in different twin zygosity groups, also by sex.
Methods:
All Swedish twins born 1933-1958 who participated in the Screening Across the Lifespan Twin study (SALT) 1998-2003, were included (31,122 individuals). The self-reported DP and long-term sickness absence (LTSA) at the time of interview was compared to the corresponding register information retrieved from the National Social Insurance Agency by calculating the proportions of agreements, kappa, sensitivity, specificity, concordance rates, and chi-square test, to evaluate construct validity.
Results:
The proportions of overall agreement were 96% and specificity 99% for both DP and LTSA, while the sensitivity was 70% for DP and 45% for LTSA. Kappa estimates were 0.76 for DP, and 0.58 for LTSA. The proportions of positive agreement were 64% for DP and 42% for LTSA. No difference in response style was found between zygosity groups among complete twin pairs for DP and LTSA. Results were similar for women and men and across age. Kappa estimates for DP differed somewhat depending on years of education, 0.68 (college/university) vs. 0.77 (less than 13 years in school) but not for LTSA.
Conclusions:
Self-reported DP data may be very useful in studies when register information is not available, however, register data is preferred especially for LTSA. The same degree of twin similarity was found for truthful self-report of DP and LTSA in both monozygotic and dizygotic twin pairs. Thus, the response style was not influenced by genetic factors. One consequence of this would be that when estimating the relative importance of genetic and environmental effects from twin models, heritability estimates would not be biased.