Community health workers improve diabetes care in remote Australian Indigenous communities: results of a pragmatic cluster randomized controlled trial

The study setting was 12 small remote communities (Indigenous population range 260–3,000) in far north Queensland where the majority of the population was Aboriginal or Torres Strait Islander, served by a single provider and where the health service had agreed to participate in the trial. Primary health care is provided by either a community-controlled service (n = 4) or the Queensland Government (n = 8). The distance to the nearest tertiary hospital is between one and 12 hours by road or air.

The unit of randomisation was the community health service which was allocated to either the health-worker led case management intervention or to a waitlist control group (where the intervention was provided after 18 months). Following patient recruitment and baseline data collection, the 12 services were randomly allocated (names out of a hat) to either the intervention (n = 6) or waitlist group (n = 6). The study was not blinded as the allocation arm was known following recruitment and baseline data collection and the study was designed as a pragmatic trial reflecting effectiveness in real world practice [8]. The data reported here were collected at two timepoints. Baseline data at the time of recruitment (2011) and follow-up data in 2013.

The study was approved by the Cairns and Hinterland Institutional Ethics Committee with support from the peak Aboriginal and Torres Strait Islander Health Councils. The trial is registered as a clinical trial, ANZCTR number 12610000812099.

Patient eligibility criteria included having type 2 diabetes and at least one major comorbidity, age 18 or more, poor glycemic control (HbA1c > =8.5% or 69 mmol/mol), mentally competent and able to provide informed consent, and obtaining regular care from the identified health service. Exclusion criteria were major mental ill health (major psychosis or depression requiring inpatient treatment) or substance misuse, renal dialysis or end-stage renal disease, a cancer diagnosis or current pregnancy. Eligible patients were identified from their records by the health service staff, who then approached them to be in the trial. Enrolment occurred between December 2011 and July 2012. Baseline (2011) and follow-up (2013) interviews were conducted by Indigenous researchers in either plain English or Creole, depending on the preference of the patients.

The trial was powered to demonstrate a reduction in median HbA1c by 1.0% over 18 months compared to baseline, as the primary outcome measure. This estimate was based on a mean drop of 1.3% HbA1c over one year (from 9.9% to 8.6%, following initiation of intensive drug treatment in T2DM) reported by a large US Health Maintenance Organization [9]. A sample size of 49 in each group would have 90% power to detect a difference in mean HbA1c between the intervention and control group after 18 months of 1.0%, assuming that the common standard deviation was 1.5% using a two group t-test with a 0.05 two-sided significance level. With 12 communities (6 intervention, 6 control), this would require 9 subjects per community. With adjustment for clustering, assuming a design effect of 1.2 derived from a similar study [10], the required number of subjects per community is 11 for the primary outcome, or 132 in total [11]. However, due to potential difficulty of maintaining subjects in these communities in the trial, high rates of mobility, and the potential for a more modest effect size in this group, we aimed to recruit 300 subjects. In the event, 327 patients were assessed by the health services as eligible and 213 (65%) agreed to participate, providing written informed consent. Over the study period, 22 patients (10%) were lost to follow-up: 6 died, 15 moved away from the community permanently and one withdrew. More patients in the intervention group than the waitlist group were lost to follow up (Figure 1).

Each site allocated to the intervention arm recruited an Indigenous health worker resident in the community (selected by the health service) to work as part of the primary care team, and allocated a caseload of between 9 and 26 clients. The health workers with low caseloads worked part-time. All health workers at the commencement of the study received an intensive 3-week training in clinical aspects of diabetes and other chronic condition care, including how to support patients in self-management skills, advice on medications, routine foot care, nutrition, smoking cessation, follow up referrals to other providers, and scheduled tests. The roles of the health workers included helping patients make and keep appointments, understand their medications and nutrition and the effects of smoking and where appropriate, work with the family to help support the patient in self-management. Home visits and out-of-clinic care were features of the trial, however visits were conducted according to the patients’ preferences.

The curriculum included specific training and practice in:

During the 18 month intervention period, the health workers attended two workshops where they underwent refresher training, including in Good Clinical Practice and reflective practice. During these sessions, they reported on their patients’ progress and shared approaches to problem solving with the clinical support team and peers.

The primary outcome measure for this trial, glycemic control (HbA1c) was measured by Queensland Medical Laboratories using standard high-pressure liquid chromatography methods. Blood pressure, height, weight, serum fasting lipids (Cholesterol fractions and triglycerides) were abstracted from clinic files and electronic records. Taking insulin was defined as having any of long-acting, medium- or short-acting insulin. Albuminuria was defined as urinary ACR > =3.4 mg/mmol. Estimated glomerular filtration rate (eGFR) was calculated from serum creatinine using the CKD-EPI formula, where GFR = 141 × min (Scr/κ,1) ^α × max (Scr/κ,1)^-1.209 × 0.993^Age × 1.018 [if female] × 1.159 [if black], where Scr is serum creatinine (mg/dL), κ is 0.7 for females and 0.9 for males, α is −0.329 for females and −0.411 for males, min indicates the minimum of Scr/κ or 1, and max indicates the maximum of Scr/κ or 1 [12].

Test of Functional Health Literacy for Adults (TOFHLA) [11] was administered at study enrolment to all participants to gauge the patients’ general understanding of health messages and procedures. In general, TOFHLA was scored highly in both groups, and it was concluded that they would not have major difficulty working with the diabetes care team.

Quality of Life was estimated using the Assessment of Quality of Life (AQoL) instrument, a multi-attribute utility instrument developed using Australian importance weights [13]. Socio-demographic data was by self-report, including years of formal education, household income, employment, food insecurity (“Do you frequently not have enough money to buy food?”), current smoking and medication adherence.

Guideline recommended clinical checks, including General Practitioner Management Plans (GPMP) and specialist referrals [14], were abstracted from primary care records for the 18 months prior to the baseline and follow-up assessments. For patients with clinical indications, appropriate medication use (as indicated in the clinical guidelines) was recorded, including insulin, statins, ACEi or ARB drugs, vaccination.

Implementation Fidelity and “dose” was estimated through diaries kept by the IHWs which recorded time spent on specific intervention-related tasks versus other clinical work, as we anticipated that due to the small size of the communities and the limited number of health staff, that the trial workers would be called upon to perform regular acute clinical care to the general patient population.

Baseline demographic and clinical data were analysed for differences between intervention and waitlist patients using t-tests, Wilcoxon rank sum tests and chi-square tests as appropriate. At follow-up all analyses were on an intention to treat basis. The 22 people who had either died or were lost to follow up were included in the dataset; however, their 18 month follow-up results were left blank. Analysis was by Generalized Estimating Equations (GEE), using a Gaussian family and identity link function. The GEE regression models used Group, Time, and a Group × Time interaction term as predictor variables. The formal test of an intervention effect was whether or not the coefficient for the interaction term was statistically significant at the 5% level. An initial analysis using a mixed effects model showed that adjustment for clustering by community made little difference, based on a likelihood ratio test. That being the case, further models were not adjusted for clustering. There were a few cases where data was available for only one time point. Where there was missing data, a second analysis was performed used multiple imputation with 20 copies of the dataset and using a regression approach for imputation. Nonparametric tests (ranksum test) were used to examine HbA1c change by GPMP exposure between groups. A further analysis was done looking at the impact of the service model (Community controlled versus Government provided) on the likelihood of clinical change, independently of group allocation.

All analyses were done using Stata version 13.