Australian arm of the International Spinal Cord Injury (Aus-InSCI) community survey: 1. population-based design, methodology and cohort profile

The Aus-InSCI survey forms part of a global cross-sectional study to describe the lived experience of people with SCI, within and across countries and corresponding health and social support systems, policies, services, and care. Details of the InSCI survey are described elsewhere [8].

Study design and participation of data custodians

The Aus-InSCI study combined 11 databases from nine data custodians across four Australian states (New South Wales, Queensland, South Australia and Victoria), creating a representative, population-based, anonymised master database that serves as the sampling frame for individuals with SCI. Data custodians included the specialist SCI clinical services/units in each state, a government insurance agency and three not-for-profit SCI consumer associations. Two other consumer associations were invited but did not participate.

Prior to data collection, the anticipated composition of the target population was considered based on expert opinion and reports from the Australian Spinal Cord Injury Register, a national register of SCI incidences treated in the seven SCI units in Australia [9]. These anticipated characteristics included proportions living in metropolitan and regional/rural settings (70%, 30%); with paraplegia and tetraplegia (50%, 50%); complete and incomplete impairments (40%, 60%); aged <40 years, 40–60 years & >60 years (40%, 30%, 30%), and time post-injury <10 years, 10–20 years, >20 years (33% each).

Participants

Adults aged 18 years or over, who were residing in the community and at least 12 months post-injury, were able to fill in the questionnaire in English and had either a traumatic injury (e.g., due to motor vehicle crash, fall) or non-traumatic, non-progressive SCI disease or disorder (e.g., from spinal stenosis, infection, vascular accident or primary neurological tumour) were eligible.

Adults with a congenital SCI (such as spina bifida) or neurodegenerative disorders (including multiple sclerosis and amyotrophic lateral sclerosis, or peripheral nerve damage, such as Guillain-Barré Syndrome), those currently receiving acute or subacute care in hospital or unable to complete the survey due to severe cognitive impairments (i.e., severe traumatic brain injury, major mental health condition or dementia) or inability to speak English, were excluded.

Data linkage and creation of master database

Each data custodian prepared a dataset containing records of all eligible individuals with identifiable information (such as name and date of birth) and sociodemographic and injury-related information. The nine participating custodians prepared a total of 11 datasets, and securely transferred them to a third-party data linkage facility, the Population Health Research Network - Centre for Data Linkage (PHRN-CDL) based at the Curtin University in Western Australia. The PHRN-CDL cleaned, merged and de-duplicated these datasets to create a single master database, which served as the sampling frame for recruitment. The data cleaning phase included the standardisation of data, such as the same codes for gender and the same formats for dates of birth. Missing values or placeholders for missing data were also identified and standardised. The merging and de-duplication of data included a deterministic pass where exact matches were identified. Probabilistic data linkage was then used to determine matches where there were variations in records (e.g., differences due to typographic errors or even changes in addresses). The probabilistic method compares two records and assigns weights based on how closely each field matches. Weights are summed across each field comparison to produce a total weight for the record pair. Only those record pairs with a weight above a certain threshold are accepted as a match. Multiple matching passes ensure that all possible record pairs are assessed. The linkage strategies used in this project were adapted from those used in other multi-jurisdictional data linkage studies, which have been shown to return high-quality linkage results [10, 11]. The master database was then forwarded to the Australian Institute of Health and Welfare (AIHW) in Canberra for linkage with the National Death Index (NDI), identifying individuals who were deceased [12]. AIHW returned the NDI-linked dataset to the PHRN-CDL. A final cleaned and linked master dataset was prepared, assigning a master key identifier with unique national and international IDs and passwords. Eleven re-identifiable datasets containing unique records were then returned to the respective nine data custodians for recruitment. Additionally, a de-identified, population-based master dataset, including basic injury characteristics and National and International IDs and passwords was sent to the national co-ordinating study centre, John Walsh Centre for Rehabilitation Research (JWCRR), Kolling Institute, Sydney.

During the above database handling process, rigorous data management protocols were applied by the PHRN-CDL to protect the privacy and confidentiality of individuals. These include strict data governance procedures covering people, processes and information technology; role separation and restricted data flows to mitigate risks to privacy by limiting access to certain information [13]. The ethically approved record linkage process in Australia was without the specific written consent of each person with SCI and on the basis that this data was believed to be in the public interest and low risk (under Section 95 A of Commonwealth Privacy Act 1988/2014).

Recruitment and data collection

Eligible individuals were invited to participate by their respective data custodians, with two reminders sent to individuals who had not responded at 3 and 6 months after the initial invitation. At each time point, participants were sent a package, including an invitation or reminder letter, participant information sheet, a blank Aus-InSCI survey (with a unique international ID and a password to access online completion) and a pre-paid self-addressed return envelope. Recruitment was by an opt-out approach. Participants in this study were not under any obligation to complete the questionnaire. Implied consent was used for participants who completed surveys.

The study commenced on 5 March 2018, and recruitment finished on 31 January 2019. Participants could complete the survey as a paper version returned via the pre-paid self-addressed envelope, online by logging into the InSCI website (using their unique Australian ID and password provided to them in the invitation/reminder package) or via telephone interview.

The Aus-InSCI questionnaire

The InSCI data model, based on the International Classification of Functioning, Disability and Health (ICF) Core Sets for SCI and Rehabilitation, has previously been described [7]. The Aus-InSCI questionnaire is compiled in English, comprising the InSCI module (with 125 questions) and an additional national module, including 68 questions. The InSCI questionnaire includes sociodemographic factors, SCI characteristics, body functions and structures, activities and participation, environmental and personal factors, and health and well-being, and it took between 45–60 min to complete. For more details, see Appendix A of paper 2 of this series [14].

Statistical analysis

Eligibility, response status and participation rates were described according to the standard definitions of the American Association for Public Opinion Research [15]. Participants’ questionnaire responses were used to describe cohort characteristics. A minimal dataset of core sociodemographic and injury-related information from data custodians on all eligible individuals was used to compare participant and non-participant characteristics. Participation status (participation vs. non-participation) was regressed on a set of sociodemographic and injury characteristics to identify potential predictors for participation using logistic regression analysis, both before and after adjustment for other factors. Odds ratios (OR) and 95% confidence intervals are reported, whereby OR above 1 indicate a higher probability for survey participation and OR below 1 indicate a lower probability of participation.

To correct for potential bias due to unit non-response, logistic regression models for propensity to participate were developed considering age, gender, socioeconomic status, geographical region, recruitment source, injury level and injury duration. Predicted propensities for participation derived from these models were used to generate inverse probability weights, which were then used in subsequent analyses to correct for the potential non-response bias. Reweighted estimates for the percentage of individuals in current paid work and for mean quality of life ratings, modified self-reported Spinal Cord Independence Measure (m-SCIM-SR) total scores [16], and Nottwil Environmental Factor Inventory Short Form (NEFI-S) scores [17] were compared with unweighted estimates, both overall and by gender and lesion level, using survey-weighted generalised linear models. The m-SCIM-SR score used in these analyses involved 12 questions derived from the standard SCIM-SR measure, covering self-care, sphincter management, use of the toilet, and three mobility questions (ability to perform bed-mobility activities unassisted, and degree of independence in transferring from bed to a wheelchair, and in moving moderate distances of 10-100 metres), rescaled to range from 0 (least independent) to 100 (most independent) [16]. The NEFI-S evaluated environmental barriers to participation in society over the past four weeks and was scored between 0 (fewest barriers) and 100 (most barriers) [17]. The self-rated quality of life ratings used were coded from 1 (very poor) to 5 (very good).

Differences in sociodemographic and injury characteristics of participants were examined by recruitment source, including the type of data custodian (consumer organisation, government agency, or SCI unit) and location of data custodian (New South Wales, Queensland, South Australia, or Victoria), and by response characteristics, including speed of response to survey invitations (first three months, next three months, or last four months of data collection period) and response mode (online, telephone, or paper-based).

留言 (0)

沒有登入
gif