Study sites, design and questionnaires
Ten Aboriginal communities were selected for a study of housing improvement and child health in 2004–5. Communities were spread across the NT, and included coastal, inland tropical and desert communities ranging in size from 250 to 1,450 people, with an average population of 730. Communities were selected because they were known to be receiving significant housing upgrades and the primary research question of the HICH study was whether improved housing was associated with an improvement in child health [15, 17, 18]. Therefore, the study represents prospective cohort design in which children were followed up; however, for the current analyses only baseline data are used and is therefore cross-sectional in nature. The scope of the study included all children less than 7 years of age. Consent for children in the study was obtained from their primary carer. Consent was also obtained separately from all carers' and householders' participating in the study.
The HICH study was designed to collect detailed information on housing conditions, and importantly on carers, householders, children and community contexts to allow for the complex relationships that exist to be accounted for in statistical analyses. The use of the HICH data to explore the complex relationships between carer characteristics, housing conditions, community factors, child health and reported gambling problems is consistent with the overall goals of the study in gaining a better understanding of how household and community environments affect child health in Indigenous communities in the NT.
Six survey instruments were used to collect data on the: 1) community, 2) householder (head of house), 3) dwelling, 4) primary carer of child, 5) child health and demographics (carer report), and 6) child health (audit of health clinic data). Carer, householder and child information were all collected through face-to-face interviews and in all communities local interpreters were employed to assist in translation of the survey questions when interviews were being carried out. The survey instruments were piloted in two communities prior to commencing the survey and modification and deletion of questions made as required .
Ethics approval (Application: 02/66) for the project was obtained from the Human Research Ethics Committee of the Northern Territory Department of Health and Menzies School of Health Research (Registration: EC0153) and the Central Australian (no reference number supplied) Ethics Committee, and formal agreements to participate were signed by peak organisations in each of the ten Indigenous communities.
Primary outcome variables
Child health was measured by asking the primary carer of the child about the presence or absence of common childhood illnesses for the two weeks preceding the survey. From this, six child health variables were constructed and are: 1) skin infections with no scabies, 2) scabies (with or without skin infections), 3) respiratory infections, 4) diarrhoea and/or vomiting, 5) ear infections, and 6) a composite child health variable indicating that the carer reported two or more of the five illnesses for the child. The study also collected child health data through an audit of health centre data; however, exploratory data analysis revealed this data to be less reliable than carer report, due to differential recording of illness between community health centres (see [15, 18]).
Reported gambling problems were measured using the Negative Life Events Scale (NLES), which was included in the carer and householder surveys. The NLES measures a person’s exposure to ‘life stressors’ and was designed for use in surveys of both the Indigenous and non-Indigenous populations of Australia. The scale was developed by the ABS in consultation with a special advisory group that guided items for inclusion. This group comprised experts in Indigenous information, research and cultural issues, who were nominated “from Indigenous organisations, peak Indigenous information bodies, Commonwealth and state/territory government agencies with Indigenous program responsibilities, and relevant academic research institutions” , p. 1. The NLES, as used in ABS surveys asks respondents since last [insert current month] last year, have any of these [list of stressors] been a problem for you, or your family or friends? However, for the HICH study this question was modified to include only the respondent and any one in the house in which they reside. Specifically respondents were asked if any of these things [list of stressors] has been a problem for yourself or someone in this house in the last 12 months. Respondents were then shown/read out a list of ‘negative life events’, or ‘life stressors’ for which a yes or no response was elicited. The stressors included: gambling problem; alcohol or drug related problems; being a witness to violence; being abused or in a violent crime; trouble with the police; divorce or separation; not able to get a job; lost a job or sacked; death of family member or close friend; serious illness or disability in a family member; serious accident by family member; overcrowding; racism or treated different; and vandalism.
The NLES was administered to both the carer and householder, so the reported gambling problems variable was available for the carer and householder. A community level gambling problems variable was also generated indicating that carer’s report of gambling problems in households for the community were: (i) less than 20%, (ii) 20–49%, and (iii) 50% or more. Both the carer and householder report of gambling problems yielded very similar estimates for communities, so only the carer reported gambling problem variable was aggregated for the community level variable.
The measure of gambling problems used in this study is not a measure of problem gambling prevalence, but is broadly consistent with the Australian national definition which states that problem gambling is characterised by difficulties in limiting money and/or time spent on gambling which leads to adverse consequences for the gambler, others, or for the community, p. 126. A more detailed discussion of the NLES, including its strength and weaknesses has been reported previously [10, 12] and the NLES, as used in the HICH study, bas been found to be a reliable measure .
Data on socio-demographic, socioeconomic, psychosocial, housing condition and community characteristics were collected. Child level variables used included age, sex, mobility, relationship to the carer, relationship to householder, breastfeeding history and day care attendance. Householder data was limited in its use in the analyses due to a number of variables having large amounts of missing data. Carer variables included age, sex, education, financial stress and psychosocial status.
Two psychosocial measures were collected for carers, in addition to the information collected as part of the NLES previously described. The Brief Screen for Depression (BSD) contains a set of questions scored on a Likert scale and when summed produce a maximum score of 50, with scores of 25 or greater indicating depression . Positive and negative affect were measured using the two scales of the Affect Balance Scale , which gives scores of 0 to 4 (with 4 being most positive or most negative). Both scales were dichotomised with scores of 0 to 3 compared with scores of 4 (i.e. most negative or most positive). The BSD, Positive Affect Balance (PosAB), and Negative Affect Balance (NegAB) had Cronbach alpha scores (α = 0.57, α = 0.66, α = 0.58 respectively).
Data assessing both the hygienic state and physical functioning of various aspects of the house were collected according to two methods of rating houses that corresponded to twelve Healthy Living Practices (HLPs) as outlined in the National Indigenous Housing Guide , and an overall measure for the house. Briefly, the first method, known as the Failed HLP (FHLP), assessed the condition of individual items of infrastructure, which were then grouped according to the HLPs. Each HLP was then scored as a pass only if all items of infrastructure for the particular HLP passed. Eight out of the possible 13 HLPs were assessed using the FHLP method. The second method, known as the Surveyor Function Score (SFS), assessed the functional state of each HLP using a 7-point Likert scale that scored the infrastructure based on the perception of trained surveyors. A 7-point Likert scale was also used to assess the hygienic state of the house for each HLP and given by the Surveyor Condition Score (SCS). Full details of both HLP scoring methods are provided in more detail in previous publications [15, 17, 31].
Community level variables included in the analyses were: community ID (not identifiable), crowding, community housing function aggregated from eight of the HLPs, environmental health condition, whether the community has outstations (smaller communities on traditional homelands) attached, community location (desert, inland tropical, coastal tropical), frequency of access to a medical doctor, and community facilities (e.g. meeting hall, canteen (alcohol outlet), takeaway store, women’s centre and aged care facility).
Two distinct analyses were undertaken to address the research questions. The first analysis used carer report of gambling problems for themselves or someone in their house as the outcome variable, which was a binary variable making it suitable for logistic regression modeling. Unadjusted associations were first calculated for community, carer, householder and dwelling level variables. Carer and householder variables showing moderately significant (p < 0.10) associations with reported gambling problems were then used to develop multivariable adjusted model for carer report of gambling problems. This was done by entering all moderately significant variables into the model and applying backward elimination with removal set at p > 0.05. Once the multivariable adjusted model for the carer report of gambling problems was determined, community level variables showing a significant bivariate association with reported gambling problems were entered separately into the model to determine if any community contexts retained significance after adjustment for carer and householder level characteristics. Plausible first order interactions were explored and pair-wise tests carried out if significant.
The second analysis was at the child level, with child illness (report of skin infection, scabies, respiratory infection, ear infection, diarrhoea and vomiting and two or more reports in the two weeks preceding the survey) used as the dependent variables’. Again, these outcome variables were binary and logistic regression modeling was used to assess unadjusted associations between child illness and community, carer, householder and dwelling level variables. The reported gambling problems variable was excluded from this part of the analyses, as this was a primary variable of interest and was added later to models. The following process was carried out for each of the child health outcomes. Carer, householder and dwelling level variables showing a moderately significant (p < 0.10) bivariate associations with child illness were retained for the next stage in the analysis. Because of the large number of variables, a blocked step approach was used to develop final multivariable adjusted models. First, demographic variables retained from the previous stage were entered simultaneously into a model and backward elimination carried out with removal set at p < 0.05. This procedure was then repeated separately for socioeconomic, psychosocial and dwelling variables. This procedure was also carried out for community level variables, though many variables exhibited co linearity and were excluded. Next, all significant variables from each of the domains were entered into a model simultaneously and backward elimination carried out with removal at p > 0.05. After arriving at this model, the carer and householder reported gambling problems were entered separately to determine if they retained a multivariable adjusted association with child health. Lastly, community level variables were added one at a time and included in final models if significant. Plausible first-order interactions were then tested and presented if significant.
Confidence intervals for all analyses were adjusted for clustering of children and carers in dwellings and communities using the Huber-White sandwich variance estimator. All analyses were carried out using Stata 9.2 IC.