An investigation of the longitudinal trajectory patterns of health-related quality of life among Australians with disabilities: explaining disability types and properties

The Household, Income and Labour Dynamics in Australia (HILDA) is a nationally representative broad socioeconomic longitudinal survey for Australian residents. Other comparable surveys in the world like HILDA include the US Panel Study of Income Dynamics (PSID), the UK Household Longitudinal Study (UKHLS), and the German Socio-Economic Panel (SOEP). Readers interested in learning more about the HILDA survey might check the HILDA user manual or Nicol Watson's review of the HILDA survey [16, 17].

The HILDA survey incorporated a top-up sample of 2000 households in 2011 to alleviate the impact of sample attrition and uphold population representativeness amidst shifting demographics in Australia. Therefore, wave 11 (2011) was selected as the baseline year for our analysis. This study analyzed individuals who self-reported a disability in wave 11 of HILDA survey data and followed up on those individuals up to wave 21 (Individuals who responded fewer than three times were excluded, as at least three data points are needed to understand a trajectory). This resulted in 3724 individuals and 34,539 observations for our study (i.e., 3724 individuals reported having a disability in 2011 and took part in the subsequent surveys at least three times. Individuals who did not report a disability in 2011 but later developed one, or those who had a disability previously but did not report it in 2011, are excluded from our analytical sample). A sequential imputation strategy addresses missing data in individual responses (item-wise). We prioritize forward imputation, where the missing value is estimated based on the respondent’s most recent preceding response. In cases where forward imputation is not applicable (i.e., the missing value is at the beginning of the sequence), we employ backward imputation, utilizing the respondent’s closest following response. The participant flow into the analytic sample is shown in Fig. 1.

The existing literature employs a variety of Health-Related Quality of Life (HRQoL) instruments to assess health and well-being for both individuals with and without disabilities (e.g., EQ-5D, ASCOT, AQoL, SF-6D etc.)[18, 19]. Since the HILDA survey provides only one measure of preferences-based HRQoL: SF-6D, we used it as our outcome measure. The SF-6D utility score, a widely used preference-based HRQoL measure was developed by Brazier and colleagues [20]. Utilizing the six domains (six physical functioning items, four role limitations items, five social functioning items, six pain items, five mental health items, and five vitality items) of the SF-36 instrument [21], the SF-6D is capable of predicting utility scores for 18,000 health states (6 × 4 × 5 × 6 × 5 × 5). In our study, the SF-6D is the primary outcome variable. Our study sample includes 641 distinct health states, with scoring ranges from 0.301 to 1.

The HILDA survey collects participants’ disability status through a face-to-face interview. Participants who reported disability conditions at wave 11 were selected for this study and followed longitudinally. Using the guidelines provided by earlier studies [22‐24], this study classified the 17 different types of disability into four major categories: (I) Sensory disability, (II) Physical disability, (III) Psychosocial disability, and (IV) Other disability or long term conditions. The study also considers whether the individuals have single or multiple disabilities and whether disability status is work-limiting or not. Details of the disability types are provided in Table A1 of the supplementary material (Appendix A).

In addition to the above-mentioned clinical factors, the study also includes demographic covariates (age and gender) and socioeconomic covariates (equivalized household income quintile, education and current labour force status) to comprehend the relationship between the various trajectory groups identified from our HRQoL measure.

To begin, we described the study participants’ demographic characteristics. We used group-based trajectory modeling (GBTM) to identify subgroups of PwD who share similar quality-of-life trajectories based on their SF-6D health state utility scores at each point [25‐27]. Following group identification, we examined the distribution of disability types over time and estimated the transition probabilities of disability conditions for each group (STATA command “xttrans” is used to estimate transition probabilities). Finally, we assessed demographic, clinical and socioeconomic factors’ association with quality-of-life trajectory groups through a multinomial logistic regression technique.

Table 2 tabulates the BIC, Bayes Factor (B₁₀), entropy, group size, average posterior probability (APP) and Odds of Correct Classification (OCC) for model fits to the HRQoL data of PwD in HILDA survey. We know that, if the Bayes factor is greater than the value of 150 (or 2log_eB₁₀ > 10), the evidence against the null model is very strong [25]. Table 2 shows that the number of classes with Bayes factor threshold value greater than 150 is seven classes. However, from seven class trajectories, the entropy is also lower than 0.8, and the smallest group size fell below 5%, violating our threshold criteria. For this reason, we determined that, there are six trajectory classes of HRQoL for people with disabilities.

Figure 2 depicts the six HRQoL trajectories for PwD. From the graph, we can see that Trajectory 1 and 2 both exhibit a low HRQoL trajectory throughout the study. We classified the individuals exhibiting these two trajectories as the low HRQoL trajectory group. Similarly, individuals exhibiting Trajectories 4 and 6 exhibit a high HRQoL trajectory throughout the study period and we classified this group as the High HRQoL trajectory group. Trajectory 3 started with lower than the High trajectory group but gradually improved over the study period. We classified individuals following this trajectory as the Moderate-Improving HRQoL trajectory group. Finally, Trajectory 5 almost started like Trajectory 4, but gradually deteriorated and had lower HRQoL than Trajectory 3 (Moderate-Improving group) after six years. We defined individuals following this trajectory group as Moderate-Deteriorating HRQoL trajectory group. Hence, the four distinct groups form these six trajectories are: (1) Group 1: High HRQoL trajectory group (33% of the population) (2) Group 2: Moderate-Improving HRQoL trajectory group (18.9% of the population) (3) Group 3: Moderate-Deteriorating HRQoL trajectory group (11.6% of the population) and, (4) Group 4: Low HRQoL trajectory group (36.5% of the population).

Figure 3 illustrates the distribution of four types of disabilities (sensory, physical, psychosocial, and other) among the four HRQoL trajectory groups (Group 1: High, Group 2: Moderate-Improving, Group 3: Moderate-Deteriorating, and Group 4: Low). Group 4 has a high distribution trend for all four major disability categories (sensory: 24–29%, physical 67–78%, psychosocial: 23–29% and other: 70–78%). On the contrary, Group 1 has a low distribution trend for all four major disability categories (sensory: 14–22%, Physical: 21–39%, psychosocial: 3–8%, and other: 31–60%). Group 2 (sensory: 17–21%, physical 43–55%, psychosocial: 10–17% and other: 47–66%) usually has a higher distribution than Group 3 (sensory: 17–24%, physical 31–57%, psychosocial: 5–15% and other: 45–66%) at the beginning in physical, psychosocial, and other disabilities and the trend gradually dropped compared to Group 3. From wave 15 and onwards, Group 3 usually has a higher disability distribution than Group 2 and gradually increases. This makes sense, as, Group 3’s HRQoL dropped over time. Upon careful examination of the graphs, it becomes intriguing to observe that disability prevalence closely reflects the trajectory groups. Trajectories characterized by high disability prevalence are negatively related to low HRQoL scores.

The implication of distribution trends suggests that the absence, presence, or accumulation of disability conditions may have significantly affect an individual’s HRQoL. Table 3 provides the transition probabilities of disability status (no condition, single disability condition and multiple disability condition). Transition probabilities are the likelihood of transitioning from one state (e.g. single disability condition) to another (e.g., multiple disability condition). Transition counts between observations are used to estimate transition probabilities. Overall, there is a 65% chance that a person with no condition would have no more condition in the next period. Similarly, there is a 45% chance that an individual with a single disability condition would stay at the single disability condition in the next period and a 73% chance that a person with multiple disability conditions would stay at multiple disability conditions. A person with a single disability has a 30% probability of transitioning to multiple disabilities and a 25% probability of transitioning to a disability condition in the following period. A person with multiple disabilities has about a 20% probability of transitioning back to a single disability and approximately 7.5% of transitioning to no condition in the subsequent period. For the trajectory groups, Group 1 has a higher rate of transitioning to a better condition than Group 2, Group 2 has a higher transition rate to a better condition than Group 3, and so on.

Table 4 examines the associations between the baseline (wave 11) demographic, clinical and socioeconomic factors and the trajectory groups. The Relative Risk Ratio (RRR) and the multinomial logistic regression statistics are utilized to profile individual characteristics. The RRR represents the exponentiated coefficient in a multinomial logistic regression model. It reflects the change in the relative risk of the outcome in the comparison group compared to the referent group. An RRR > 1 indicates the risk of outcome falling in the comparison group increases compared to the referent group, and vice versa. Group 1: High HRQoL trajectory group is the reference group to which other groups are compared. The significant factors of Group 2: Moderate-Improving HRQoL trajectory group vs the reference groups are as follows (RRR are in brackets): aged 25–44 years (2.054), 45–54 years (1.666), 65 + years (1.612), female (1.237), physical disability (1.820), psychosocial disability(2.868), other disability (1.572), work-limiting disability (2.005), income: middle (0.622), rich (0.635), richest (0.615), and not in the labour force (1.568). Significant factors for Group 3: Moderate-Deteriorating HRQoL trajectory group vs reference group are as follows: aged 65 + years (2.169), income: middle (0.666) and not in the labour force (1.402). This result is interesting because only three factors significantly differ from the High HRQoL reference group. Hence, it is reasonable to assume that Group 3 primarily comprises older individuals with disabilities who are presently not part of the workforce and are experiencing deteriorating health.

On the contrary, except for multiple disability status and certificate/diploma education, all factors are highly significant for Group 4: Low HRQoL group versus the reference group (multiple disability status is not significant in any of the groups). The significant factors and their RRR for Low HRQoL groups are as follows: age: 25–44 years (3.817), 45–54 years (3.121), 65 + years (3.959), female (1.532), sensory disability (1.474), physical disability (3.524), psychosocial disability (6.090), other disability (2.439), work-limiting disability (4.146), income: middle (0.487), income: rich (0.472), income: richest (0.363), bachelor or higher education (0.631), unemployed (2.428), and not in the labour force (2.140). In general, from the regression analysis, we can see that Group 4: Low HRQoL has higher RRR in demographic and clinical factors (risk of falling in Group 4 is higher than the Group 1) and lower RRR in socioeconomic factors for better SES (risk of falling in group 4 is lower than the group 1 for higher SES). Further, females have a higher RRR, indicating they are at risk of failing in Group 2 versus Group 1 or Group 4 vs Group 1, indicating vulnerability of disabled females as well as HRQoL improving capacity of females. Among clinical factors, psychosocial disability has the highest RRR for Group 4, followed by work-limiting disability and physical disability, implying having these factors greatly increases the risk of falling in Group 4. Furthermore, the RRRs in Group 4 (Low) exhibit a greater magnitude than those in Group 2 (Moderate-Improving), suggesting that all these factors pose a higher risk of being classified into Group 4 (Low) than Group 2 (Moderate-Improving). Lastly, as people’s health deteriorates with age, elderly people have a greater chance of falling into Group 3 (Moderate-Deteriorating) or Group 4 (Low), as they have a high RRR of falling into these groups. In summary, various types of disability, as well as socioeconomic and demographic status, have distinct effects on the HRQoL of people with disabilities.

Our results show that, High HRQoL trajectory group, fared better outcomes than other trajectory groups. A question arises whether this group genuinely fared better than other disability groups. Since our analysis showed that High HRQoL trajectory has a lower proportion of disability distribution of all types (Fig. 3), it is safe to assume that the High HRQoL group genuinely fare better than other trajectory groups and ‘disability paradox’ actually is not a paradox. Some PwD just fare better than others. This has significant policy implications for resource allocation for health interventions. NDIS and DSP could improve allocative efficiency if group trajectories are considered.

The onset and nature of a disability impact a person’s overall HRQoL. However, prior research has focused on the association between HRQoL and having disability [37‐39]. Other studies have examined how changes in a particular aspect of HRQoL (e.g., mental health) correlate with disability acquisition [40]. Therefore, existing studies cannot identify HRQoL trajectories of people with different types of disability. The main contribution of our study to the existing literature is showing HRQoL pathways and changes in HRQoL over time in people living with disabilities through group-based trajectory modelling. HRQoL was measured using a preference-based score (SF-6D), an additional strength of our study. Another major strength of the study is that we have studied 17 distinct forms of disability and categorize them into four broad groups to better understand how they each influence HRQoL pathways instead of focusing on disability in general. Other strengths of our study include using longitudinal data and a large sample size to precisely estimate the trajectories in HRQoL after disability onset. Our study has some limitations that need to be discussed. First, self-reported bias may arise since data on disability were collected through a self-completion questionnaire. Second, we did not consider the severity of the disability to explore its effect on HRQoL. Third, we cannot incorporate some important covariates that influence HRQoL. For example, we did not include chronic disease as a potential confounder in the model due to unavailability in all waves considered for the present study.