A population based validation study of self-reported pensions and benefits: the Nord-Trøndelag health study (HUNT)
Myrtveit, Solbjørg Makalani; Steinsland Ariansen, Anja Maria; Wilhelmsen, Ingvard; Krokstad, Steinar; Mykletun, Arnstein
Abstract
Measures of disability pensions, sickness certification and long-term health related benefits are often
self-reported in epidemiological studies. Few studies have examined these measures, and the validity is yet to be
established.
We aimed to estimate the validity of self-reported disability pension, rehabilitation benefit and retirement pension
and to explore the benefit status and basic characteristics of those not responding to these items.
A large health survey (HUNT2) containing self-reported questionnaire data on sickness benefits and pensions was
linked to a national registry of pensions and benefits, used as “gold standard” for the analysis. We investigated two
main sources of bias in self-reported data; misclassification - due to participants answering questions incorrectly,
and systematic missing/selection bias - when participants do not respond to the questions.
Sensitivity, specificity, positive (PPV) and negative (NPV) predicative value, agreement and Cohen’s Kappa were
calculated for each benefit. Co-variables were compared between non-responders and responders.
Results: In the study-population of 40,633, 9.2% reported receiving disability pension, 1.4% rehabilitation benefits
and 6.1% retirement pension. According to the registry, the corresponding numbers were 9.0%, 1.7% and 5.4%.
Excluding non-responders, specificity, NPV and agreement were above 98% for all benefits. Sensitivity and PPV were
lower. When including non-responders as non-receivers, specificity got higher, sensitivity dropped while the other
measures changed less.
Between 17.7% and 24.1% did not answer the questions on benefits. Non-responders were older and more likely to
be female. They reported more anxiety, more depression, a higher number of somatic diagnoses, less physical
activity and lower consumption of alcohol (p < 0.001 for all variables). For disability pension and retirement pension,
non-responders were less likely to receive benefits than responders (p < 0.001). For each benefit 2.1% or less of
non-responders were receivers. False positive responses were more prevalent than false negative responses.
Conclusions: The validity of self-reported data on disability pension, rehabilitation benefits and retirement pension
is high – it seems that participants’ responses can be trusted. Compared to responders, non-responders are less
likely to be receivers. If necessary, power and validity can be kept high by imputing non-responders as
non-receivers.