The Cost Consequences of the Gold Coast Integrated Care Programme

Introduction: The Australian Gold Coast Integrated Care programme trialled a model of care targeting those with chronic and complex conditions at highest risk of hospitalisation with the goal of producing the best patient outcomes at no additional cost to the healthcare system. This paper reports the economic findings of the trial. Methods: A pragmatic non-randomised controlled study assessed differences between patients enrolled in the programme (intervention group) and patients who received usual care (control group), in health service utilisation, including Medicare Benefits Schedule and Pharmaceutical Benefits Scheme claims, patient-reported outcome measures, including health-related quality of life, mortality risk, and cost. Results: A total of 1,549 intervention participants were enrolled and matched on the basis of patient level data to 3,042 controls. We found no difference in quality of life between groups, but a greater decrease in capability, social support and satisfaction with care scores and higher hospital service use for the intervention group, leading to a greater cost to the healthcare system of AUD$6,400 per person per year. In addition, the per person per year cost of being in the GCIC programme was AUD$8,700 equating to total healthcare expenditures of AUD$15,100 more for the intervention group than the control group. Conclusion: The GCIC programme did not show value for money, incurring additional costs to the health system and demonstrating no significant improvements in health-related quality of life. Because patient recruitment was gradual throughout the trial, we had only one year of complete data for analysis which may be too short a period to determine the true cost-consequences of the program.

Step 1. Intervention group: Raw data was received from GCHHS Health Analytics for 1,682 intervention participants. Those with no inpatient admissions (or missing data) during this period were ineligible for matching, therefore 133 participants were excluded. The remaining 1,549 were split into two groups: (a) 993 participants with at least one admission with primary or associative diagnoses of heart disease, diabetes mellitus (DM), renal failure or chronic lower respiratory disease and (b) the remaining 556 participants with at least one admission with other diseases. Matching was undertaken using data at baseline (recruitment) of the intervention group and then matched to hospital records for the control group.
Passive control group: De-identified raw data was received for 86,665 passive control candidates. One hundred percent of this group had at least one inpatient admission during this period and were therefore eligible for matching. The group was split into two: (a) 18,207 patients with at least one admission with primary or associative diagnoses of heart disease, DM, renal failure, or chronic lower respiratory disease, and (b) 68,458 who were admitted at least once with a diagnosis other than the above listed four primary or associative diagnoses. Matching of a control to one intervention participant was performed separately by these sub-groups: 1,205 passive controls were matched in the (a) sub-group, and 1,025 were matched in the (b) sub-group.
Active control group: The active control group was randomly selected from the control group to provide external validation to the HHS data and other measures such as quality of life. A sample of 20% was selected from the passive controls to recruit at least 600 participants for the active control group. Allowing for some deaths and losses to follow-up, over-sampling by 25% was undertaken (750). Raw data was received for 893 active control participants. Those with no inpatient admissions during this period were ineligible for matching, therefore 25 participants were excluded. The remaining 868 were split into two groups: (a) 781 participants with at least one admission with primary or associative diagnoses of heart disease, DM, renal failure, or chronic lower respiratory disease, and (b) the remaining 87 participants with at least one admission with other than the above listed diagnoses. Matching of a control to one intervention participant was performed separately by these subgroups. One hundred percent of the 868 active controls were matched to an intervention participant. Figure S1. Control group matching flow chart Step 2.
a. Direct radius matching was done, which was a multi-step process using the radmatch command in Stata. This user-written procedure shuffles the controls in a random order, then selects the required number of matches for each intervention participants based on a list of variables of interest and maximum tolerances (radius) for each. Radius of zero means exact match required. Matching in sub-group (a) used the following list of variables based on presentations for those with the four main conditions: 1) gender (radius=0), 2) age in years in 2015 (radius=5), 3) heart disease (I20-52) as primary or associative diagnosis (radius=0), 4) chronic lower respiratory disease (J40-J47) (radius=0) 5) diabetes mellitus (E10-E14) as primary or associative diagnosis (radius=0), 6) renal failure (N17-N19) as primary or associative diagnosis (radius=0), 7) number of discharges (radius=0), and 8) percentage of discharges linked to potentially avoidable hospitalisations (radius=20).
The matching process was repeated (with slight adjustments) until 100% of intervention participants in sub-group (a) were matched with exactly two control participants.
The matching process was repeated (with slight adjustments) until 100% of intervention participants in (b) were matched with exactly two control participants.
As recommended by Billot et al (2016), matching performance was investigated by comparing the baseline characteristics, healthcare utilisation and disease profile between the two study groups as shown in Table 1. Standardised difference was used as this is not influenced by sample size [43,44]. Table 2 presents the participant characteristics by subgroup (a) and sub-group (b).