The Southern Highlands Division of General Practice (SHDGP) in New South Wales (NSW) implemented a diabetes program in 1995. All GPs in the SHDGP participate in the program. The GPs remain the case managers for their patients, and the program facilitates case management by provision of information and education to the GPs and by direct service provision to patients referred by the GPs. The Division funds and arranges diabetes education programs, dietitian services and an exercise program, and arranges access to podiatry services.
The core of the program is the centralized database of diabetic patients which is regularly updated with clinical information from the general practices. This database, which includes information on care provision as well as clinical indicators such as HbA1c measures, is used to send recall reminders to GPs, to provide regular audit reports to GPs on their adherence to guidelines, and to provide regular and ad hoc clinical alerts. In particular, the Division identifies patients who may be at risk of developing complications, and reports on them to their GPs.
The centralized database has been adapted for the diabetes program from a patient information database designed to manage information on cardiac and diabetic patients (the CarDiab database [22]). While there are gaps in the database where clinicians have failed to report the results of consultations, or patients have failed to attend for consultations or tests, the quality of data used is high. All pathology data are provided electronically from the pathology laboratories, and demographic and other clinical data are provided directly by the GPs to the Division, either paper-based or electronically. The analysis undertaken in this study draws on information from this centralized database and a range of other sources as outlined below.
The SHDGP is a rural division with the majority of the population living in three small towns. Its catchment population at the 2001 Census was 48,400 with 16% aged over 65 and 33% under 25. The population is above the median for socio-economic status as measured by the Socio-Economic Index for Australia (SEIFA) prepared by the Australian Bureau of Statistics [23]. The Division has 50 GP members in 16 practices and a relatively low level of GP services compared to the national average (3.7 GP services per head of population compared with 4.7 for Australia as a whole) [24].
A cost-effectiveness analysis of the Southern Highlands Division Diabetes Program is undertaken using a decision-analytic approach. As with many recent studies [13, 6, 25] a simulation model is used to predict longer term patient outcomes and complications for patients in the program. To our knowledge there is no diabetes simulation model available based on Australian data. The United Kingdom Prospective Diabetes Study (UKPDS) Outcome Model [19, 26] used in this study is based on ten years data from the UKPDS. This model predicts the costs and the outcomes for patients from their time of diagnosis with type 2 diabetes until death. Simulations are based on a probabilistic discrete-time illness-death model which has been validated within the UKPDS populations. The UKPDS model is designed to capture the association between different types of complications at an individual patient level[27]. It includes the major complications of diabetes, but not other problems such as ulceration and hyper/hypoglycemia [26].
Projections from the UKPDS outcomes model are determined by observed clinical measures (including HbA1c, blood pressure, BMI and cholesterol measures) both at diagnosis and where available over a post-diagnosis period. The model was used in this study without modification to simulate the progression of diabetes and its complications from time of diagnosis to death, and from a time point which is 5 years after initial participation in the program to death. The model simulates the progression of diabetes under “conventional treatment” within the UKPDS study. “Conventional treatment” is prescribed in the UKPDS as clinic attendance every three months and dietary advice, aimed to maintain fasting plasma glucose below 15 mmol/L without symptoms of hyperglycemia [2]. Conventional treatment in Australian practice will be much more varied. The implications of this potential difference are likely to be small, as the issue of interest is the difference between projections from diagnosis and projections after 5 years in the program. Where the estimated impact of the program on the costs of complications as derived from the model can be compared with estimates derived from other sources, results are of a similar order.
It is common practice for models to draw on data from countries other than those which are the subject of their study. The CORE model for example [28], used data from the UKPDS, data from the Framingham Study and other data sources in United States, data from Sweden and information from the international DIGAMI trial.
A sample of patients enrolled in the program was selected as outlined below and the average annual incremental cost of managing them in the diabetes program estimated. These costs encompassed both the direct costs of managing the program, and any incremental costs or savings flowing from changes in care for program participants. Four categories of costs were included: the cost of the SHDGP diabetes program; the primary care costs costs arising from adherence to the guidelines for management of diabetes; the cost of pharmaceuticals; and the cost of inpatient hospital services.
Differences in costs in three of these categories (primary care services provided in compliance with guidelines, pharmaceutical treatment, and hospital treatment) cannot be directly measured from the Division database. However, it is possible to estimate these costs by comparing costs in the SHDGP with costs for NSW or for Australia as a whole. Cost estimates for managing complications from the UKPDS Outcomes Model are used to provide estimates of the costs of hospital treatment for comparison with the direct estimates available from Australian information.
In summary, data from SHDGP patients are used to estimate program impacts on the life years and quality adjusted life expectancy of these patients, and to estimate the costs of diabetes caused complications. These cost estimates contribute to the estimated impact of the program on hospital costs, while the program impacts on life years and QALE are used to estimate cost effectiveness.
To convert the estimated annual program costs and the related costs and savings to a 40-year time horizon, the average annual estimates are assumed to be the same each year over the 40-year time frame of the analysis. The cost streams are discounted at five per cent per annum. Although the modelling incorporates both deaths and increased costs per capita due to complications as patients age, it does not address per capita cost changes for pharmaceutical costs or costs of GP compliance with guidelines. If the single year overall treatment savings followed the pattern of the costs of complications, the 40 year total estimate would be only 5 per cent different to the value estimated by assuming constant costs. The dominant factor in the overall costs is the direct cost of the program, which will not increase with patient ageing except at the margin (e.g. dietician costs), but will decline with the deaths of program participants. Allowance for the effect of these deaths on program costs reduces the long term costs by substantially more than any plausible increase in treatment cost effects, but a fixed value approach was taken to ensure that results were conservative.
The five per cent discount rate follows the standards used in Australia in assessing pharmaceuticals and medical services for public funding [29, 30]. This level of discounting is consistent with German, Swiss and French discounting rates [18], although for the United Kingdom rates of 3.5% are used [25] and rates of 1.5% for clinical outcomes and 6% for costs are quoted [18], and for the Netherlands 3% is used [12]. In the sensitivity analyses, a rate of 0% is tested, but as 5% is at the higher end of international discounting rates no higher value is used.
The health outcomes for the selected patients are estimated over a 40-year period from diagnosis, using the model of the long-term sequelae of patients with type 2 diabetes in the UKPDS Outcomes model. The measures used are the changes in years of life expectancy and in QALE. The impact on utility of different diabetes related complications used in the model to estimate QALE is based on a study of a UKPDS participants in 1997 [31].
Using National Health Survey data [32], the number of identified type 2 diabetics in the SHDGP in 2005 was expected to be 1,525. Overall the program had contact with around 85% of the expected total number of identified diabetics in the region in 2004, although for 220 of the known patients there were no clinical data.
While there were 1,087 type 2 diabetic patients included on the database at some time over the program, the number included in the analysis was considerably less than this as:
As a result of these exclusions, the final sample comprised 74 patients who registered at diagnosis and had complete information both at diagnosis and approximately 5 years after diagnosis.
As this group includes only five-year survivors, six additional patients who were randomly selected from among those deceased within five years of joining the program were added to the initial sample. Six patients were added as eight per cent of the total patient group who registered at or near diagnosis died during the following five years.
Gaps in the SHDGP data combined with the requirements of the UKPDS model resulted in a sample which could have introduced bias. However, as shown in Table 1, the characteristics of all patients on enrolment were similar to those of the study sample. Significant differences which exist are in the expected directions. Table 1 also shows the program has been effective in improving the status of the sample over 5 years. Most clinical measures (e.g. cholesterol) have improved or are stable, and the increase in average HbA1c is less than would have been expected over 5 years under normal treatment [27].
Life years and QALE were estimated over 40 years from diagnosis using the UKPDS Outcome model. Outcomes were first estimated from diagnosis with no constraint, and secondly estimated with the constraints of actual clinical measures after five years of program participation. These two sets of projections gave estimates of life years and QALE saved by program participants.
Division administrative costs, including the costs of employing the program manager/diabetes educator, data entry, general administration and IT costs, and the costs of patient access to dietetic services and the exercise program were provided by the Division. The details of these costs are shown at Additional file 1.
Administrative costs to practices included maintenance of records and transfer of data to the Division, and the management of patient recall. These costs were estimated by the Division based on consultation with the participating practices and are also shown at Additional file 1.
The cost of GP compliance with guidelines, and the flow on costs to pathology testing, ophthalmology etc, were based on data from the Service Incentives Payments (SIP) scheme [33]. SIP payments are made to registered GPs on completion of a 12 month cycle of guidelines based care for a diabetic patient. Claims were made for 31% of known diabetic patients in the SHDGP compared to 20% across Australia as a whole [33]. These percentages were combined with estimates of average national costs per diabetic patient-year for out of hospital services [34] (excluding pharmaceutical costs), and assumptions on relative costs of compliant and non-compliant treatment, to provide estimates of costs of guideline compliance (see Additional file 1).
A comparison of prescribing rates for oral antidiabetic agents in the SHDGP with national rates, using publicly available data from the Pharmaceutical Benefits Scheme (PBS), was used to estimate differences in pharmaceutical costs attributable to the program[35]. Details again are shown in Additional file 1. As it is not possible to divide insulin prescribing between type 1 and type 2 diabetic patients, the cost of insulin has not been included. On the other hand, these data do not include patient co-payments, which would lead to an under-estimate of any savings.
Three estimates of cost savings were obtained for hospital services. One estimate was based on the difference between rates and costs of hospitalization for patients with primary ICD-10 diagnosis codes 10–14 (diabetes) residing in the Southern Highlands and in New South Wales overall. These estimates were based on data provided by NSW Department of Health. As these are diabetes-specific codes, they will not include all complications of diabetes. The cost per patient-year for these two groups ($A255 and $A341 respectively) are well below the Australian estimate of around $A600 per patient year [34] for diabetes-related hospitalization. However, the actual hospitalization data provide an indication of the level of savings in hospital costs for program participants.
The second estimate applied the percentage reduction in the costs of treating complications attributed to the program from the UKPDS model to an estimate of the average annual cost of hospitalizations for type 2 diabetic patients in Australia [34]. The third approach used actual dollar values of savings estimated by the UKPDS Outcomes model. The three approaches provided estimates of a similar order of magnitude, as shown at Additional file 1.
