Abstract
BackgroundThe data from longitudinal complex surveys based on multi-stage sampling designs contain cross-sectional dependencies among units due to clustered nature of the data and within-subject dependencies due to repeated measurements. Special statistical methods are required to analyze longitudinal complex survey data.MethodsStatistics Canada's longitudinal National Population Health Survey (NPHS) dataset from the first five cycles (1994/1995 to 2002/2003) was used to investigate the effects of demographic, social, life-style, and health-related factors on the longitudinal changes of mental distress scores among the NPHS participants who self-reported physician diagnosed respiratory diseases, specifically asthma and chronic bronchitis. The NPHS longitudinal sample includes 17,276 persons of all ages. In this report, participants 15 years and older (n = 14,713) were considered for statistical analysis. Mental distress, an ordinal outcome variable (categories: no/low, moderate, and high) was examined. Ordered logistic regression models based on the weighted generalized estimating equations approach were fitted to investigate the association between respiratory diseases and mental distress adjusting for other covariates of interest. Variance estimates of regression coefficients were computed by using bootstrap methods. The final model was used to predict the probabilities of prevalence of no/low, moderate or high mental distress scores.ResultsAccounting for design effects does not vary the significance of the coefficients of the model. Participants suffering with chronic bronchitis were significantly at a higher risk (ORadj = 1.37; 95% CI: 1.12-1.66) of reporting high levels of mental distress compared to those who did not self-report chronic bronchitis. There was no significant association between asthma and mental distress. There was a significant interaction between sex and self-perceived general health status indicating a dose-response relationship. Among females, the risk of mental distress increases with increasing deteriorating (from excellent to very poor) self-perceived general health.ConclusionsA positive association was observed between the physician diagnosed self-reported chronic bronchitis and an increased prevalence of mental distress when adjusted for important covariates. Variance estimates of regression coefficients obtained from the sandwich estimator (i.e. not accounting for design effects) were similar to bootstrap variance estimates (i.e. accounting for design effects). Even though these two sets of variance estimates are similar, it is more appropriate to use bootstrap variance estimates.
Highlights
The data from longitudinal complex surveys based on multi-stage sampling designs contain cross-sectional dependencies among units due to clustered nature of the data and within-subject dependencies due to repeated measurements
Higher proportions in the moderate or high level distress categories were observed for i) respondents who selfreported asthma or chronic bronchitis that have been diagnosed by a health professional; ii) younger respondents; iii) females; iv) non-white people; v) widowed/separated/divorced or single respondents; vi) immigrants; vii) respondents living in urban areas; viii) respondents from Atlantic and Quebec regions; ix) respondents in low and middle income categories; x) respondents with low education (= 12 years); xi) respondents with low social involvement score; xii) current smokers; xiii) respondents exposed to smoke within household; and xiv) respondents with 'poor' self-perceived health status
Unadjusted Odds ratios The strength of relationship between mental distress and each of the independent variables based on the Generalized Estimating Equations (GEE) approach is presented as an estimate of odds ratio (OR) and 95% confidence interval in Table 3 and are described below
Summary
The data from longitudinal complex surveys based on multi-stage sampling designs contain cross-sectional dependencies among units due to clustered nature of the data and within-subject dependencies due to repeated measurements. The complex multi-stage sampling designs used for these longitudinal surveys contain cross-sectional dependencies among units (caused by inherent hierarchies in the data) in addition to the within-subject dependencies due to repeated measurements. As previously reported [3,4] for longitudinal dichotomous outcome, modelbased analytical approach of complex survey data sets ignores stratification and clustering and may lead to biased results. The first set of models (known as model-based) was based on the assumption that the study design involved only subject-level clustering due to repeated measurements and was based on the Generalized Estimating Equations (GEEs) approach [7,8], ignoring the complexities of the survey design. The bootstrap method based on the replication approach was used in this article for the variance estimation
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a callDisclaimer: All third-party content on this website/platform is and will remain the property of their respective owners and is provided on "as is" basis without any warranties, express or implied. Use of third-party content does not indicate any affiliation, sponsorship with or endorsement by them. Any references to third-party content is to identify the corresponding services and shall be considered fair use under The CopyrightLaw.
