The International Spinal Cord Injury (InSCI) community survey is a large-scale collaborative cross-sectional survey completed across 22 countries in 2017–18 [14, 15]. This study used data completed by Australian participants of the InSCI community survey (henceforth, Aus-InSCI) [16]. The full survey included 193 questions. A subset of these 193 questions related to use of healthcare practitioners, unmet needs, satisfaction with healthcare providers and details about participants’ primary health care provider. Further details of the Aus-InSCI survey are published elsewhere [16, 17].

Study setting

This cross-sectional study was informed from 11 datasets of people with SCI supplied from nine data custodians. Data custodians from four Australian states participated: New South Wales, Queensland, South Australia and Victoria. Custodians included specialist SCI unit(s) or service(s) in each of the four participating states, three not-for-profit SCI consumer associations and a government insurance agency [16].

Databases were securely provided by custodians to an external institution (Curtin University) for linkage to identify and remove duplicates, creating a single master dataset [16]. This dataset was securely sent to the Australian Institute of Health and Welfare (AIHW) for linkage to the National Death Index. This was returned to Curtin University and following cleaning, de-duplication and removal of deceased individuals data, the individual re-identifiable datasets were returned to their respective data custodian for recruitment [16]. Full details on the secure data linkage process are described in the Aus-InSCI protocol paper [16].

Ethics was approved by the Northern Sydney Local Health District HREC (HREC/16/HAWKE/495) and AIHW Ethics Committee (EO2017/1/341). Implied consent was used for participants who completed surveys.

Participants

Participants sourced from data custodian databases were eligible if they were aged 18 and over, living in the community, at least 12-months post SCI and were able to complete the questionnaire in English. Paper-based surveys were mailed to eligible participants in 2018 and could be completed as a hardcopy or online via enclosed unique participant login details. Follow-up reminders were mailed at 3- and 6-months after the initial contact for those participants who did not respond.

Data measures

Survey questions on health service use and unmet healthcare needs were derived from the Model Disability Survey [18]. The specific questions are described below.

Healthcare provider utilisation

Service use was assessed with the question ‘who were the health care providers you visited in the community or hospital, or who visited you in your home, in the last 12-months?’. Fifteen types of healthcare providers, including ‘other’, were listed. Participants also identified their main contact for SCI-specific problems (i.e., general practitioner, local specialist, spinal specialist, or other).

Experiences with healthcare providers

The survey collected information on participants’ last visit with a healthcare provider: “For your last visit to a health care provider, how would you rate the following: (1) ‘your experience being treated respectfully?’, (2) ‘how clearly health care providers explained things to you?’, and (3) ‘your experience being involved in making decisions for your treatment?’”. Responses for these questions were: ‘very good’, ‘good’, ‘neither’, ‘bad’, or ‘very bad’.

Satisfaction with healthcare services provided

This was measured using three questions: ‘how satisfied are you with the services provided by (1) your general practitioners? (2) your local general hospital/s? (3) the Spinal Cord Injury Unit/Services in your state?’. Participants provided a response on a 5-point Likert scale from ‘very satisfied’ to ‘very dissatisfied’.

Unmet healthcare needs

The following question was used as a proxy for unmet healthcare needs, ‘in the last 12 months, have you needed health care but did not get it?’. Participants who responded ‘yes’ were classified as having unmet healthcare needs. Reasons for not being able to access healthcare were also collected (e.g., the cost of the service, unavailable services, inadequate provider skill).

Hospitalisations

Hospitalisations were collected based on the question: ‘over the last 12-months how many times were you a patient in a hospital, rehabilitation facility or care facility for at least one night?’

General satisfaction was captured using the question: ‘in general, how satisfied are you with how the health care services are run in your area?’. Participants provided a response on a 5-point Likert scale from ‘very satisfied’ to ‘very dissatisfied’.

Health measures

Fourteen secondary conditions were assessed, with responses given on a 5-point Likert scale ranging from 1 ‘no problem’ to 5 ‘extreme problem’. The 14 secondary conditions captured were: bowel, bladder and sexual dysfunctions; urinary tract infections; contractures; spasticity; pressure ulcers; respiratory problems; injury caused by loss of sensation; autonomic dysreflexia; postural hypotension; sleep and circulatory problems; and pain. Participants were given a score (between 0 and 14) based on the number of severe secondary conditions (that is, a response 4 or 5 ‘extreme problem’). A higher score on this scale indicates a greater number of severe secondary conditions.

Sociodemographic and injury characteristics

Sociodemographic factors included gender, age, marital status, education, rurality, daily household assistance being received (no/yes); weekly household income; and disability pension statues (no/yes) (see Table 1 for variable levels). Injury characteristics included injury level and completeness, injury aetiology (i.e., traumatic or non-traumatic) and time since injury (years).

Table 1 Variables included the six Bayesian penalised regression models.Functional independence

The Spinal Cord Independence Measure (SCIM) is a 16-item measure assessing mobility, self-care, and respiration. A score is calculated ranging from 0 to 100 where higher scores indicate higher functional independence [19]. The InSCI survey used a modified, self-reported version of the SCIM (henceforth, modified-SCIM-SR). A subset of the SCIM with scores ranging from 0 to 66 was used. This included all self-care (feeding, bathing, grooming), bowel and bladder management, and some mobility measures (bed to wheelchair transfers, moving moderate distances, activities able to be performed without assistance or aids). This version was rescaled to the original measure (i.e., 0–100) for easy interpretation, where higher scores again indicate greater functional independence [19].

Data analysis

All analyses were undertaken in R (version 4.1.1) [20] using the RStudio environment (version 1.4.1717) [21]. Sociodemographic and injury characteristics, income and education, measures of health, health service use, unmet healthcare needs and satisfaction with healthcare services provided with services were descriptively summarised as count and percent or median and first and third quartile.

Bayesian penalised regression was used to model the six binary outcome variables. Penalised regression guards against overfitting [22], with the aim to shrink small non-important effects to zero while maintaining true, larger effects. Shrinkage (horseshoe) prior distributions with three degrees of freedom were specified for regression coefficients in all models. A logit link function was used to connect the Bernoulli distribution with the regression coefficients. Bayesian models were fit with Stan [23] using the brms interface [24].

The six outcomes were: (1) unmet healthcare needs; (2) general practitioner use; (3) allied health practitioner use; (4) medical specialist use; (5) rehabilitation specialist use (inclusive of SCI specialists); and (6) hospitalisations in the past 12-months. Five of the six models included the same 14 predictor variables: age, gender, geographical area, education, daily professional household assistance, receives disability pension, marital status, weekly household income, level and completeness, injury aetiology, time since injury, number of severe secondary conditions, modified-SCIM-SR and unmet health care needs (Table 1). The model for unmet healthcare needs also included provider experiences (Table 1).

There were missing values in 14 of the 16 variables included in models, with the percentage missing ranging from 0.1% to 19.5% (Suppl. 1) [25]. Values were assumed to be missing at random. Missing values were imputed using predicted mean matching, with the results averaged over five imputed datasets.

Posterior estimates were generated using Markov chain Monte Carlo procedures (50,000 iterations, 4 chains and thinned by a factor of 5) and are reported as the mean and 95% credible interval (CrI). Parameter estimates are shown on the logit scale in figures and are reported as odds ratios in supplementary tables, along with the posterior probability that the odds ratio was greater than the null value of 1 (Pr OR > 1) and less than the null value of 1 (Pr OR < 1). The results were interpreted using estimation methods [26].