Predictors of government subsidized pharmaceutical use in patients with diabetes or cardiovascular disease in a primary care setting: evidence from a prospective randomized trial

Abstract

Objectives:

This study uses data from a prospective randomized controlled trial to estimate predictors of pharmaceutical expenditure in diabetes (DM) or cardiovascular disease (CVD) patients. Identifying drivers of pharmaceutical use and the extent to which they are modifiable may inform cost-effective policy-making.

Methods:

The trial followed 260 patients aged >18 years (mean 68) from three general practices for 12 months. Patients had type 2 diabetes (90 patients) or cardiovascular disease (170 patients). Costs for pharmaceuticals prescribed on the Pharmaceutical Benefits Scheme (PBS) were obtained retrospectively at 12 months. Sociodemographic data and health-related quality-of-life (QoL) were recorded from questionnaires. Clinical measures (including body mass index (BMI), blood pressure, high and low density lipoprotein (LDL), and HbA1c) were also collected.

Results:

Mean pharmaceutical costs for DM patients (AU$4119) was greater than CVD patients (AU$2424). The largest contributor to costs in both groups was pharmaceuticals used for management of conditions other than CVD or DM. QoL (EQ5D) and BMI were significant predictors of costs in both groups. A history of cardiac events, HbA1c, age, and unemployment were significant predictors of costs in the DM group. A diagnosis of heart failure, frequency of hospital admissions, and LDL levels were significant predictors of costs in the CVD group. Roughly one third of total variation of costs can be explained by the regressors in both models.

Limitations:

Generalizability will be limited as data was derived from a trial and the study was not powered for this post-hoc analysis. Missing data imputation and self-reporting bias may also impact on results.

Conclusions:

Factors such as QoL BMI, HbA1c levels, and a history of cardiac events are significant predictors of costs. The results suggest there may be a place for interventions that improve quality-of-life and concurrently reduce pharmaceutical costs in patients with CVD or DM.

Key words::

• Drug costs
• Drug utilization
• Diabetes mellitus
• Cardiovascular diseases

Introduction

Cardiovascular disease and diabetes mellitus are two inter-related chronic conditions which are leading causes of global burden of disease, particularly in middle and high income countries1. Expenditure on pharmaceuticals related to these conditions accounts for a substantial proportion of total pharmaceutical expenditure. Australian Government expenditure on pharmaceuticals via the national Pharmaceutical Benefits Scheme (PBS) totalled $7.7 billion for the year ending 30 June 20092. This represented a 9.2% increase in expenditure from the previous year, the 4th year in the past 7 where growth has exceeded 9%3. Key drivers of this growth include the aging of the Australian population and the increasing proportion of Australians categorized as overweight or obese (Body Mass Index >25)4. In common with other developed countries, an increasing economic burden from chronic diseases such as cardiovascular diseases and type 2 diabetes mellitus are a result of this changing risk profile. Consequently, the largest contributors to pharmaceutical expenditure by the Australian Government are the Anatomical Therapeutic Chemical (ATC) group of cardiovascular system drugs ($2.3 billion in the financial year 2008–9), the ATC group of nervous system drugs ($1.4 billion), and the ATC group of alimentary tract and metabolism drugs which include insulin and oral glucose lowering agents ($1.1 billion)2.

International studies have found diabetes mellitus (DM) and cardiovascular disease (CVD) diagnoses to consistently predict increased use of pharmaceuticals5–7. While public health programs exist that directly target the risk factors for these chronic conditions, it is not clear to what extent other factors are associated with increased use of government subsidized pharmaceuticals in Australia. Factors found to be associated with pharmaceutical use include insurance status, being female, obesity, multiple comorbidities, and poor self-reported health5–7. If predictors of pharmaceutical expenditure can be identified in an Australian context and are associated with risk factors that are considered manageable through lifestyle interventions, then this could have implications for pharmaceutical and other health resource cost control.

This paper uses data from a prospective randomized trial of individuals with diabetes and/or cardiovascular disease in a primary care setting8 to describe the usage and cost of PBS pharmaceuticals in a community-dwelling people with a chronic disease, and to estimate drivers behind pharmaceutical expenditure among patients with these chronic diseases. The contribution of specific ATC drug categories9 to the total cost of medicines prescribed for these patients on the PBS will also be considered.

Patients and methods

Data was sourced from a prospective, randomized trial assessing the feasibility, acceptability, and cost-effectiveness of practice nurse (PN)-led collaborative, chronic disease care against general practitioner (GP) care (http://www.anzctr.org.au/trial_view.aspx?ID=82100). The trial was conducted in three Australian general practices (urban and metropolitan in Queensland, and rural in Victoria), purposively selected by willingness to participate, geographical location to aid in generalization, distance from each other to reduce patient’s visiting another study practice, their level of computerization, and their employment of a PN. Inclusion criteria comprised all patients in each practice during the period of the study aged over 18 years who could provide informed consent with at least one of the targeted chronic diseases. Patients were excluded if their condition was unstable, or if their GP considered there were other major health problems that made trial participation unsafe. Each practice generated a list of eligible patients using its electronic medical records. Eligible patients were posted an invitation letter from the GP containing the patient information sheet and consent form, until the target sample size was obtained. A GP management plan was developed for those randomized to the PN-led arm during an initial consultation, when they were informed that they should see their GP for any concerns unrelated to their chronic disease. The initial population included 285 patients aged 18 years or older with chronic disease type 2 diabetes (DM) and/or cardiovascular disease (defined as hypertension and/or ischemic heart disease) (CVD), without unstable, or major, health problems. Disease status was determined by the GP. Sociodemographic data and health-related quality-of-life (via the EQ-5D10) were recorded from questionnaires collected at baseline, 6 and 12 months. Clinical measures (including body mass index (BMI), blood pressure (BP), high and low density lipoprotein (HDL, LDL), and glycosylated hemoglobin (HbA1c) were collected at baseline and during the 12 month intervention phase, and retrospectively for the 12-month period prior to the intervention. (Clinical measures were not routinely collected at baseline for the GP group, since this group continued ‘usual practice’, i.e., they did not have a planned additional baseline appointment as a result of the trial. Ethical approval for the trial was granted from the University of Queensland, Griffith, and Bond Universities.

Data related to the use of pharmaceuticals prescribed on the PBS during the 1-year intervention period were obtained retrospectively by a research assistant at the end of the 12 month follow-up period. Records of dosage and duration of use for each medication and medication changes were made during the 1-year intervention period, and a micro-costing approach used to estimate pharmaceutical costs for each individual. Only pharmaceuticals eligible for supply under the PBS were costed. For medications where an exact dosage and/or duration were uncertain (e.g., for pain or anti-diarrheal medication on an as required basis, or topical creams) it was assumed that one pack was dispensed. Costs (in 2010 Australian dollars; AU$1 ≈ US$1) were assigned to the pharmaceuticals using the listed maximum price for dispensed quantity in the PBS11.

All data analysis was conducted using the IBM SPSS Statistics 19 program. Trial participants were excluded from the analysis if they were not on a care plan and therefore had no drug data (nine excluded), died during the trial (two excluded), or withdrew during the trial (11 excluded). A further three cases were excluded because they contained no baseline data. Many patients suffered from both DM and CVD to different degrees, and a GP classified patients into each group. However, patients categorized as CVD patients who had a concurrent diabetes diagnosis and were purchasing diabetes-related pharmaceuticals were recoded as DM patients (11 recoded). Thus, a total of 260 patients (90 DM patients and 170 CVD patients) were included in the analysis. Where available, baseline values were used for clinical variables; where not available, missing values for HbA1c were imputed for DM patients by taking their average HbA1C reading during or before the intervention period (18 values imputed). The same method was used to impute missing LDL cholesterol (121 values imputed), BMI (91 values imputed), and blood pressure (80 values imputed) for both DM and CVD groups. Test results at the end of the intervention period were missing for some patients because GPs continued ‘usual care’, and ordering tests was at their discretion. Therefore, the imputed missing data was considered missing at random (MAR).

There were no significant differences between the randomized trial groups (PN vs GP) for mean cost of pharmaceuticals on the PBS (herein termed ‘pharmaceutical costs’). Therefore, the data for the two trial arms were pooled then stratified into chronic disease sub-groups. Mean pharmaceutical costs for each condition (DM and CVD) were categorized by selected ATC codes. Differences in the means of specific variables across the groups (DM vs CVD) were tested using independent sample t-tests.

Linear regression was employed to estimate predictors of pharmaceutical costs for each of the two chronic disease groups. The distribution of total pharmaceutical costs failed tests for normality (Kolmogorov-Smirnov test: DM p = 0.027, CVD p = 0.039). Therefore, the variable was transformed to the natural log scale (herein termed LnCosts). LnCosts passed tests for normality (Kolmogorov-Smirnov test: DM p = 0.58, CVD p = 0.119). An association with LnCosts was tested for a range of variables including physiologic characteristics (BMI, blood pressure, HDL, LDL, HbA1C), age, gender, treatment location, history of hospital admissions, self-reported health-related quality-of-life, education levels, employment type, smoking status, alcohol consumption, and co-morbidities (including self-reported diagnosis of anxiety or depression).

For each group bivariate tests of association between all variables were performed, and variables were selected for the multivariate model if they had significant correlations (p < 0.10) with the natural log of total costs, and were not highly correlated with other variables. Spearman’s rho was used for continuous variables and Mann-Whitney U-tests for categorical variables. Where two variables were correlated with a coefficient greater than 0.30 the variable with the weaker association with LnCosts was excluded from further analysis.

As the model used a log transformed dependent variable with untransformed independent variables (known as a log-lin model), the coefficients are interpreted as the percentage change in the geometric mean of total patient pharmaceutical costs per unit increase in independent variable12. The effect of binomial independent variables is also interpreted as the percentage change in the geometric mean of the dependent variable and is calculated by taking the anti-log of the coefficient for the binomial variable and subtracting one12.

Results

The participant characteristics, the composition of costs by drug category, are presented and compared across the DM and CVD groups in Table 1.

Table 1.  Participant characteristics and assessment of differences between DM and CVD groups.

	DM (n = 90)	CVD (n = 170)	p-value
Mean age (SD)	65 (10)	69 (11)	0.006
Sex – Male (%)	53	45	0.220
BMI	31.5	29.3	0.005
LDL mmol/L	2.3	2.6	0.031
Mean (SD) EQ-5D domain score	80 (20.3)	79.5 (20.9)	0.853
Site (%):
Metropolitan (QLD)	9	26
Regional (QLD)	34	32
Rural (VIC)	57	42	0.001
Mean (SD) total PBS drug costs, $	4119 (3371)	2424 (1844)	<0.001
PBS Pharmaceutical cost ($) and % of participants prescribed:
A10A – Insulins and analogs	653 (12.2)	0 (0)	<0.001
A10B – Blood glucose lowering drugs (ex. Insulin)	269 (74.4)	0 (0)	<0.001
B01A – Antithrombotic	195 (54.4)	127 (58.2)	0.249
C01 – Cardiac therapy	41 (21.1)	62 (33.5)	0.256
C02 – Antihypertensive	5 (5.6)	6 (5.9)	0.864
C03 – Diuretic	9 (14.4)	9 (16.5)	0.984
C05 – Vasoprotective	10 (3.3)	4 (1.2)	0.421
C07 – Beta blocking agent	73 (32.2)	63 (38.2)	0.627
C08 – Calcium channel blocker	71 (27.8)	95 (31.8)	0.215
C09 – Renin-angiotensin agents	251 (78.9)	238 (77.6)	0.547
C10 – Lipid modifying agent	758 (77.8)	482 (57.1)	<0.001
V04C – Diagnostic tests (non-urine)	74 (23.3)	0 (0)	<0.001
Other drugs	1711 (93.3)	1340 (93.1)	0.103
Number of other drugs (%):
0 other drugs	7	6
1–4 other drugs	40	38
5–9 other drugs	32	39
10+ other drugs	21	18	0.868

Two-sided t-tests for differences between groups.

DM, diabetes mellitus; CVD, cardiovascular disease; n, number of participants; BMI, body mass index; LDL, low density lipoprotein.

The mean age of participants in the trial was 68 years (SD = 11), approximately half of participants were female (52%), approximately one third (35%) were classified as diabetic, with the remaining two thirds (65%) were classified as having CVD without diabetes. Participants with both CVD and diabetes were analysed as part of the DM group. The rural Victorian site had the highest proportion of DM patients, whereas the metropolitan Queensland site had the highest proportion of CVD patients. The total annual cost of pharmaceuticals in the DM group (mean AU$4119, median $3314, IQR $3592) was greater than for those in the CVD group (mean AU$2424, median $2050, IQR $2148) (p < 0.0001). The composition of total drug costs in terms of ATC drug category highlights which classes of pharmaceuticals contribute most to overall expenditure. The cost of insulins and analogs (A10A) and blood glucose lowering drugs (A10B) used in treating diabetes are the greatest contributor to cost differences between the two groups. Of the cardiovascular drugs, lipid modifying agents were the greatest contributor to costs in both groups (with an average cost of $758 and $482 per person for the DM and CV groups, respectively), but had a significantly higher cost for the DM group (p = 0.001). However, the largest contributor to overall costs in both groups are pharmaceuticals used for the management of conditions other than cardiovascular disease or diabetes, with ∼20% of patients taking 10 or more other government subsidized prescription medications. Other pharmaceuticals most commonly prescribed to CVD and DM patients included drugs for obstructive airway diseases, analgesics, and drugs for acid-related disorders (Table 2).

Table 2.  Average cost per patient of top seven ‘other’ medications contributing to total costs.

ATC Drug classification		DM ($)	CVD ($)
R03	Drugs for obstructive airway diseases	315.08	266.35
N02	Analgesics	295.24	284.09
A02	Drugs for acid related disorders	192.80	145.59
G04	Urologicals	158.26	35.96
N06	Psychoanaleptics	93.56	50.54
M01	Anti-inflammatory and anti-rheumatic products	76.28	53.92
N05	Psycholeptics	69.03	35.97

ATC, anatomical therapeutic chemical; CVD, cardiovascular disease; DM, diabetes mellitus.

The proportions of patients in each chronic disease group being prescribed drugs from each ATC drug category were reflected in the cost differences between sub-groups. For example, a large percentage of DM patients (74.4%) received blood glucose lowering drugs compared to none of the CVD patients, and a greater proportion of patients in the DM group received lipid modifying agents than patients in the CVD group. Costs for other drugs were higher in the DM group (although not statistically significantly so), despite the proportion of patients and the total number of other drugs being no different to the CVD group. This suggests that the drug costs for co-morbid conditions in the DM group are more costly than in the CVD group. Seventeen variables were found to be significantly correlated with LnCosts in at least one of the disease groups (Table 3).

Table 3.  Summary of bivariate tests of association with LnCosts.

Variable	DM	CVD
Start age	0.043	0.018
Diastolic blood pressure	0.058	0.032
BMI	0.186	0.045
Diabetes diagnosis	—	—
HbA1c	0.060	—
LDL	0.162	<0.001
EQ5D Utility	<0.001	<0.001
History of cardiac events	0.059	<0.001
Visited an emergency department	0.777	0.008
Number of emergency department visits	0.887	0.005
Admitted to hospital	0.063	0.020
Number of hospital admissions	0.066	0.007
Diagnosed with heart attack	0.095	0.006
Diagnosed with heart failure	0.316	0.005
Diagnosed with Angina	0.076	<0.001
Diagnosed with Depression	0.290	0.002
Unemployed	0.021	0.583

Pearson correlation and Spearman’s rho for continuous variables.

Mann-Whitney U-test and Kruskal-Wallis for categorical variables.

BMI, body mass index; LDL, low density lipoprotein.

Three variables (self-reported diagnosis of angina or heart attack, and number of visits to an emergency department) were excluded from further consideration for the models as they were highly correlated with other variables. The regression model began with the remaining 14 variables. The models were developed by alternating between backward and forward stepped approaches until stable results were produced with adequate explanatory power and normally distributed residuals. The results of the analyses for the DM and CVD groups are shown in Tables 4 and 5.

Table 4.  Factors associated with total PBS costs in the regression models.

Model	Variables	Coefficients		p	Adjusted R^2
		B	Std. Error
DM	(Constant)	5.605	1.200	<0.001
	Age	0.021	0.009	0.027
	BMI	0.033	0.016	0.041
	HbA1c	0.169	0.077	0.030
	EQ5D Utility	−1.444	0.399	0.001
	History of cardiac events	0.540	0.226	0.019
	Unemployed	−1.122	0.345	0.002	0.313
CVD	(Constant)	8.088	0.513	<0.001
	EQ5D Utility	−0.638	0.317	0.047
	BMI	0.025	0.013	0.049
	LDL	−0.338	0.082	<0.001
	Diagnosed with Heart Failure	0.606	0.263	0.024
	Number of hospital admissions	0.406	0.158	0.012	0.273

BMI, body mass index; DM, diabetes mellitus; LDL, low density lipoprotein.

Table 5.  Interpretation of regression model coefficients.

Model	Variable	Change	Effect on costs
Diabetes	Age	1 year increase in age	2.1% increase in costs
	BMI	1 point increase	3.4% increase in costs
	HbA1c	0.1 mmol/L increase	1.8% increase in costs
	EQ5D Utility	0.1 decrease in utility	7.6% increase in costs
	History of cardiac events	Yes	71.6% increase in costs
	Unemployed	Yes	67.4% decrease in costs
CVD	EQ5D Utility	0.1 decrease in utility	4.7% increase in costs
	BMI	1 point increase	2.5% increase in costs
	LDL	1 mmol/L increase	28.7% decrease in costs
	Diagnosed with Heart Failure	Yes	83.3% increase in costs
	Number of hospital admissions	Per admission	50.1% increase in costs

BMI, body mass index; LDL, low density lipoprotein.

In both the CVD and the DM groups for every 0.1 decrease in quality-of-life, as measured by the EQ5D, total costs increased by ∼5–8%. Additionally, BMI was associated with around a 2.5–3.4% increase in costs for every additional point increase. A history of cardiac events was found to have a stronger effect in the diabetes group (72% increase in costs if a cardiac event history was reported), but was not a significant predictor of costs in the CVD model. Conversely, LDL levels in the CVD model were associated with a 29% decrease in costs for every 1 mmol/L increase in LDL. LDL was not a significant predictor of costs in the diabetes model. HbA1c, age, and being identified as unemployed were significantly associated with costs in the diabetes model only. A diagnosis of heart failure and the frequency of hospital admissions also contributed significantly to costs in the CVD group. Roughly one third of total variation of costs can be explained by the regressors in both models (Adjusted R²= 0.313 and 0.273 for DM and CVD model, respectively). This can be considered a reasonable level of explanatory power given the number of parameters in the model, variability in patients, nurses, and GPs, and small sample size.

Discussion

The findings of this study identified a number of factors that are associated with pharmaceutical costs in participants with chronic disease in a primary care setting. In individuals with cardiovascular disease a self-reported diagnosis of heart failure was observed to have the greatest impact on total pharmaceutical costs, followed by a previous hospital admission. A self-reported diagnosis of heart failure was associated with a statistically significant greater number and cost of several drug categories (B01A—antithrombotics, C01—cardiac therapy, C10—lipid modifying agents). An increase in LDL was associated with a substantial reduction in cost. This may seem counterintuitive, as dyslipidemia is known to predict cost of cardiovascular medicines13. However, it may be that LDL is confounded by lipid lowering therapy—i.e., those with a lower LDL are more likely to be taking a lipid-lowering agent, which in turn is contributing to higher pharmaceutical costs. An increase in BMI and a reduction in quality-of-life (as measured by an EQ-5D generic utility weight) had a smaller but significant cost impact. Our findings related to the impact of BMI are broadly consistent with recent research in Australia which reported that obesity predicts annual pharmaceutical costs in a population with existing cardiovascular disease and/or risk factors13. However, an increase in diastolic blood pressure was not observed to predict costs as might have been expected13. It is possible that the sample used in the current study may have been too small to identify an association.

Individuals with diabetes (with or without co-morbid CVD) were observed to have substantially (54%) higher pharmaceutical costs than those with cardiovascular disease alone, and this is consistent with the findings of others13. Having a comorbid cardiovascular condition was associated with a large (72%) increase in pharmaceutical costs in diabetes. In fact, costs for drugs used to treat patients with cardiovascular disease (ATC category C) are generally higher in the DM group ($1218) than the CVD group ($959, Table 2), confirming that patients in the diabetic group are likely to be suffering from cardiovascular disease in addition to diabetes. Being unemployed was significantly associated with lower pharmaceutical costs in the diabetes sub-group. This may be an issue of affordability for unemployed persons with chronic disease, or may be an artefact due to the low numbers of unemployed people in the sample. A reduction in quality-of-life or an increase in BMI, age, or HbA1C all had a smaller but significant cost impact.

The results of this study highlight the similarities and differences in the drivers of costs for patients with a diagnosis of either diabetes or cardiovascular disease. The most notable consistency is the effect of lower quality-of-life on increased pharmaceutical use, with a 0.1 unit increase in utility weight as measured by the EQ-5D (on a scale from 0 representing death to 1 representing full health) consistently associated with a 5–8% decrease in pharmaceutical costs. Previous research on predictors of pharmaceutical cost in Australians with cardiovascular disease or risk factors has not explored the impact of quality-of-life on costs13,14. Quality-of-life in this context could be considered a proxy for comorbidities, which is consistent with the observation in the current study that the use of other drugs not necessarily related to diabetes or cardiovascular disease is the greatest contributor to total pharmaceutical costs (Table 1). An analysis of the composition of these other drugs indicated that, as might be expected based on national pharmaceutical use data, drugs for acid-related disorders (including proton-pump inhibitors (PPIs)) comprised a substantial proportion of these costs2. This is particularly pertinent given the association between BMI and acid-reflux15 (which PPIs treat) and an average BMI of the observed population of 30.1 (>30 is considered obese). Indeed BMI itself is a significant predictor of costs for both the diabetic and cardiovascular sub-groups. The most notable difference is perhaps that LDL was observed to impact pharmaceutical costs for the cardiovascular but not diabetes sub-group, despite the groups having similar mean LDL levels. This might be expected as a greater proportion of DM patients are on a lipid-lowering agent.

The implications of these findings for pharmaceutical cost containment are perhaps predictable. Quality-of-life is not a factor that is modifiable per se; however, policies targeted at reducing body weight and increasing LDL cholesterol may improve quality-of-life, delay the onset of diabetes, and lessen the severity of comorbidities and cardiovascular disease. These should also be priorities amongst patients already diagnosed with diabetes, as well as improving blood glucose management as measured by HbA1c.

Although this study has explored associations with pharmaceutical costs in chronic disease, it is important to highlight that high or rising pharmaceutical costs can be beneficial, if they are associated with an appropriate and cost-effective use of medicines. The 57% of patients in the CVD group that were on lipid-lowering agents had significantly lower LDL levels than those not receiving these drugs (2.2 vs 3.2 mmol/L, p < 0.001), with 16% of those not on lipid-lowering agents having LDL levels >4.0 mmol/L. This result suggests that some patients not receiving lipid-lowering agents probably should be. There were no differences in the proportion of patients taking rennin-angiotensin agents between the two sub-groups (78%). This is higher than a previous study of patients at risk of CVD where roughly 23% of participants received angiotensin II type 1 receptor antagonists13. The difference may be a result of the patients in the earlier study being at risk of CVD, whereas this study considers patients suffering from CVD. The findings from this study suggest that there may be a place for interventions that can reduce a reliance on pharmaceuticals, and policies that improve coverage of necessary medicines.

The analysis was subject to some limitations. This is a post-hoc analysis, and the trial was not powered for the purposes of this analysis. The size of the population observed was small enough to allow random variation to have an impact on results and may limit the generalizability of the findings to the general population. In addition to this, the lack of observations for some important variables further reduced the data available for the regression analysis. While the imputation methods increased the number of data points to include in the models, this additional data manipulation can reduce the reliability of results. Much of the data was self-reported and, therefore, the findings may be subject to reporting bias.

Despite these potential limitations, the results present one of the few published indications of individual level pharmaceutical costs associated with chronic disease in Australia. As such, these results may be utilized in economic evaluations of interventions for the management of chronic disease. Further research into the predictors of pharmaceutical costs in chronic disease, and potential methods for disentangling pharmaceutical costs associated with the appropriate use of medicines from usage that might be sub-optimal, would be beneficial to inform healthcare evaluation in chronic disease. Future work could also be targeted at assessing the impact of interventions that improve blood glucose management and reduce BMI in patients with or at risk of DM or CVD. The relationship between quality-of-life and drug costs could also be explored further to determine the extent to which the severity of the chronic illness reduces quality-of-life compared to other potentially modifiable factors. Cost-effectiveness analyses could be undertaken to evaluate interventions that may reduce the demand for pharmaceuticals.

Conclusions

This study used data from a prospective randomized trial to estimate the predictors of pharmaceutical costs among patients suffering from CVD or DM. Our analysis found that factors such quality-of-life, BMI, HbA1c levels, and a history of cardiac events are significant predictors of costs. These results suggest that there may be a place for interventions that improve quality-of-life and concurrently reduce pharmaceutical costs in patients with CVD or DM.

Transparency

Declaration of funding

Jennifer Whitty is supported by a Fellowship from the Queensland Government Department of Employment, Economic Development and Innovation, Queensland Health and Griffith University. The project was funded through an ARC Discovery Project DP 0770716.

Declaration of financial/other relationships

No financial or other relationships to be declared for any of the authors.

Acknowledgments

Jacqui Young for her diligent extraction and coding of data. Prof Desley Hegney. Prof Elizabeth Patterson, Prof Chris Del Mar, Mr Paul Fahey, Mrs Jacqui Young, Dr Rosemary Mahomed, Prof Peter Baker.