Research setting
Tennant Creek (population 3,500), is located on the Stuart Highway 510 kilometres (km) north of Alice Springs and 670 km south of Katherine. It services ‘The Barkly’ tableland, an area of similar size to the UK or New Zealand (240,000 km2) located between the tropical 'Top End' and the arid 'Red Centre' of Australia’s Northern Territory. It consists largely of open grass plains with cattle stations, mines and Indigenous communities.
Health service infrastructure in Tennant Creek includes a twenty-bed hospital which provides services to assess, diagnose and treat short-term illnesses and injuries, as well as an eight-chair renal dialysis unit. The primary healthcare needs of the Aboriginal community of Tennant Creek have long been served by the local Aboriginal Medical Service.
Characteristics of the natural experiment
Accessibility to local primary healthcare for non-Aboriginal residents of Tennant Creek was, prior to November 2004, maintained continually by a solo male GP (1.0FTE) who ceased practicing at that time (Intervention A). Through to December 2006 the town lacked local access to a GP service, prompting the introduction of the Tennant Creek General Practice Service (TCGPS) (Intervention B) which was established by the RFDS (Central Operations) at the request of local authorities and supported by a Memorandum of Understanding between key Commonwealth, Territory and local stakeholders. Operating with a solo male GP (1.0FTE) the purpose of the TCGPS was to increase local access to comprehensive primary healthcare services to residents and visitors of Tennant Creek and surrounding Barkly region. Its services included medical examinations and health screening, preventative health, antenatal and counselling services as well as other RFDS-operated health and education programs such as the Rural Women’s General Practitioner Service. The TCGPS was primarily a full fee paying practice with concessions available to healthcare card holders (government health care assistance) and children under the age of 16. On average, since the second quarter of operation, the TCGPS consistently performed 930 patient consultations per quarter including 64 new patients per quarter. There was no data available to the study on the quantum of activity performed by the solos GP practice prior to November 2004.
Outcome measurement
De-identified unit records were retrieved from the RFDS Central Operations Flight database. All records with flights originating in Tennant Creek (International Civil Aviation Organisation airport code YTNK) for the life of the database (1999q2 to 2012q3) were extracted. Extraction was restricted to the non-Aboriginal resident population of Tennant Creek (i.e. the underlying population on which the interventions were design to primarily impact). Consistent with the expected health benefits of primary healthcare [
6,
7], outcomes were defined in terms of aeromedical service events for primary care sensitive conditions [
8] adjudged at the ICD-9 chapter heading level. For example, it was not deemed that the service interventions would logically or theoretically have an impact on ‘Injuries and Poisoning’ (ICD-9 800–999) thus aeromedical services with these condition codes were excluded.
Despite the remarkably stablenon-Aboriginal population in Tennant Creek between 1999 and 2012, quarterly aeromedical service use was analysed both as raw counts and rates (calculated as a proportion of the non-Aboriginal Tennant Creek population as at the 2001, 2006 and 2011 Australian Census’ for records occurring between 1999 to 2003; 2004 to 2008; and 2009 to 2012 respectively). Further, despite the sex distribution remaining stable, the population had aged and to account for such population shifts over time and any effect these might have on propensity for aeromedical service use, quarterly rates were standardised to the age and sex distribution of the non-Aboriginal population of Tennant Creek in 2011 using direct standardisation. The study protocol was approved by the Human Research Ethics Committee of the University of South Australia.
Intervention specification
Three intervention models of the form suggested by Box and Tiao [
14] were considered based on the possible impacts Intervention A (GP service withdrawal) and B (GP service replacement) could have on the aeromedical service rate: (1) abrupt and permanent; (2) gradual and permanent; and (3) abrupt and temporary. To test these potential effects, exposure of the population to the intervention was specified (as dummy variables) as follows:
(a)
For the ‘abrupt and permanent’ and ‘gradual and permanent’ shifts:
0 before the intervention quarters, 0.33 in the quarter of intervention implementation (as both Interventions A and B occurred one-third of the way through a quarter), and 1 thereafter
(b)
For the ‘abrupt and temporary’ shift:
0 before the intervention quarters, 1 during the intervention quarter and 0 thereafter.
Thus, the following Intervention A models were tested:
Model 1: Abrupt and permanent impact: Yt = μ + ω1I1t + at
Model 2: Gradual and permanent impact: Yt = μ + [ω1/(1 − δΒ)] I1t + at
Model 3: Abrupt and temporary impact: Yt = μ + [ω1/(1 − δΒ)] I2t + at
where: Yt is the aeromedical service rate at time t,
I1t is the intervention A dummy variable coded as (a) above
I2t is the intervention A dummy variable coded as (b) above
ω1 is the effect of intervention A: i.e. the net difference in the level of the time series before and after the introduction of the intervention A (for abrupt and permanent impact); the change in the level of time series at the moment of intervention A (for the other two models)
δ is a rate parameter describing how quickly (or slowly) the effect is realised (in the case of the gradual and permanent impact) or decays to zero (in the case of the abrupt and temporary impact)
Β is the “backshift operator” such that, for example, BYt = Yt-1
Following this, the best Intervention A model (now considered the pre-intervention B model) was expanded to estimate the effects of Intervention B as follows (e.g. using Intervention A, Model 1: Yt = μ + ω1 I1t + at).
Model 1: Abrupt and permanent impact: Yt = μ + ω1I1t + ω2I2t + at
Model 2: Gradual and permanent impact: Yt = μ + ω1I1t + [ω2/(1 − δΒ)] I2t + at
Model 3: Abrupt and temporary impact: Yt = μ + ω1I1t + [ω2/(1 − δΒ)] I3t + at
where: Yt is the aeromedical service rate at time t,
I1t is the intervention A dummy variable coded as (a) above
I2t is the intervention B dummy variable coded as (a) above
I3t is the intervention B dummy variable coded as (b) above
ω1 is the effect of intervention A: i.e. the net difference in the level of the time series before and after the introduction of the intervention A (for abrupt and permanent impact); the change in the level of time series at the moment of intervention A (for the other two models)
ω2 is the effect of intervention B: i.e. the net difference in the level of the time series before and after the introduction of the intervention B (for abrupt and permanent impact); the change in the level of time series at the moment of intervention B (for the other two models)
δ is a rate parameter describing how quickly (or slowly) the effect is realised (in the case of the gradual and permanent impact) or decays to zero (in the case of the abrupt and temporary impact)
Β is the “backshift operator” such that, for example, BYt = Yt-1