Trajectories of disability throughout early life and labor force status as a young adult: Results from the Longitudinal Study of Australian Children

Objectives Young people with disabilities have poorer labor force outcomes than their peers without disabilities. These understandings, however, are largely based on research assessing disability at one time point only, an approach that potentially obscures variation in disability over time. We aimed to identify trajectories of disability during childhood/adolescence and assess associations between trajectory membership and labor force status in young adulthood. Methods We conducted group-based trajectory modeling of disability status information from six waves [waves 2–7 (age 4/5 to 16/17 years)] of the Longitudinal Study of Australian Children. The trajectories were used to predict labor force participation (employed, unemployed, not in the labor force) at wave 8 (18/19 years), adjusted for confounders. Results We identified four trajectory groups of the prevalence of disability: low (75.5% of cohort), low increasing (9.7%), high decreasing (10.9%), and consistently high (3.9%). Individuals in the low increasing trajectory were nearly three times as likely to be unemployed at age 18/19 years compared to individuals in the low trajectory [risk ratio (RR) 2.96, 95% confidence interval (CI) 1.94–4.53]. Individuals in the consistently high trajectory had a greater RR of not being in the labor force at age 18/19 years compared to individuals in the low group (reference) (RR 3.65, 95% CI 2.21–6.02). Conclusions Results suggest that prolonged and increasing experiences of disability among young Australians may be differentially associated with future labor force outcomes. Additional support to prepare young people for the labor force should focus on individuals who consistently or increasingly report a disability.


Supplementary
3c. Is a description provided of how missing data in the analyses were dealt with?
Yes; methods 4. Is information about the distribution of the observed variables included?
Yes, Supplementary Table S3 5. Is the software mentioned?
Yes; methods 6a. Are alternative specifications of withinclass heterogeneity considered (e.g., LGCA vs. LGMM) and clearly documented? If not, was sufficient justification provided as to eliminate certain specifications from consideration?
Yes; we had considered using GMM but decided GBTM was appropriate for this exploratory analysis. See Appendix 2.
6b. Are alternative specifications of the between-class differences in variancecovariance matrix structure considered and clearly documented? If not, was sufficient justification provided as to eliminate certain specifications from consideration?
No; not an option in the available software. 7. Are alternative shape/functional forms of the trajectories described?
Yes, Supplementary Table S4 8. If covariates have been used, can analyses still be replicated? Yes; methods 11. Are the total number of fitted models reported, including a one-class solution?
Yes; Supplementary Table S4 12. Are the number of cases per class reported for each model (absolute sample size, or proportion)?
Yes; Supplementary Table S4 13. If classification of cases in a trajectory is the goal, is entropy reported?
Yes; Supplementary Table S4 14a. Is a plot included with the estimated mean trajectories of the final model?
Yes; Figure 1 14b. Are plots included with the estimated mean trajectories for each model? The Bayesian Information Criterion (BIC) is calculated as: BIC=log(L)-0.6k log(N), where L is the model's maximized likelihood, N is the sample size, and k is the number of parameters in the model (1).
The highest order polynomial was dropped if it was not significant (p<0.05). We repeated this process until all higher order terms were significant.
Entropy reflects the quality of the classification of the model, with values closer to 1.0 indicating better classification. Entropy is calculated by averaging the posterior probabilities after individuals have been assigned to their most likely trajectory. Values may range from 0 to 1 (2).

5
Supplementary  Does the study child have a medical condition or disability that has lasted for 6 months or more? If yes: Which medical conditions or disabilities does the study child have? Sight problems not corrected by glasses or contact lenses Hearing problems Speech problems Blackouts, fits, or loss of consciousness Difficulty learning or understanding things Limited use of arms or fingers Difficulty gripping things Limited use of legs or feet Any condition that restricts physical activity or physical work (other physical condition) Any disfigurement or deformity None of the above conditions (i.e. some other condition) To make wave 2 comparable with waves 5-7 we also included the item: Which restrictions does the study child have? Any mental illness for which help or supervision is required long-term

Waves 3 & 4
Does the study child have a medical condition or disability that has lasted, or is likely to last, for 6 months or more? GBTM assumes that the population is composed of distinct groups, each with a different underlying trajectory. These trajectory groupings represent individuals who follow similar developmental courses on the outcome of interest, (3) in this case disability. The groups can then be further explored to identify differences in baseline characteristics or subsequent outcomes (2).
To apply GBTM, the modeler must specify the number of trajectory groupings, and the trajectories themselves are then estimated parametrically directly from the data (4). Unlike methods such as growth mixture modeling (GMM) which incorporate random effects in each group's trajectory model, GBTM does not estimate within-group variation (2). However, GBTM approximates a more complex distribution of trajectories and is appropriate for exploratory analysis as it is less computationally demanding, involves fewer assumptions, and the results are more straightforward to interpret (2).