A computer‐guided quality improvement tool for primary health care: cost‐effectiveness analysis based on TORPEDO trial data

High quality health systems are a major focus for health care reform in Australia.1 In July 2015, Medicare Locals were replaced by 31 primary health networks (PHNs) with the aim of improving population health. From 2016–17, PHNs have played a major service commissioning role, working with health service providers and others to eliminate gaps in health care delivery.2

In 2015–16, the level of health care expenditure related to cardiovascular disease (CVD) was the second highest for any disease group in Australia.3 It has been estimated that about one‐fifth of Australians aged 45–74 years (1.4 million people) are at high risk of a CVD event within five years.4 A large proportion of these events could be prevented by interventions that include assessment of absolute CVD risk.5

A major barrier to reducing CVD rates is the lack of evidence‐based risk assessment and prescribing of recommended medications for people at high risk.6,7 Clinical guidelines recommend assessing and managing a combination of risk factors (an absolute risk approach) rather than treating individual factors.8,9

Interventions that assist practitioners with clinical decision‐making and communicating with patients about CVD risk are needed. HealthTracker, a computer‐based decision support tool for improving CVD risk management at the practitioner, patient, and organisational levels, was evaluated in Australian primary health care by the TORPEDO (Treatment of Cardiovascular Risk using Electronic Decision Support) study, a randomised controlled trial including 38 725 people at 40 general practices in New South Wales (mostly metropolitan) and 20 Aboriginal Community Controlled Health Services in metropolitan and regional NSW and Queensland.10 CVD risk factor screening improved by 25% at intervention sites; prescribing rates for people at high risk of CVD not previously prescribed guideline‐recommended medicines improved by 33%, but this change was not statistically significant.11

Cardiovascular health‐related costs can be reduced considerably by health information technology (IT) strategies for managing major risk factors.12 Health IT is regarded internationally as a useful tool for improving health care services while reducing their cost,13,14 leading to considerable investment in the approach in Organisation for Economic Co‐operation and Development (OECD) countries.15 However, data on the cost‐effectiveness of health IT strategies that could inform decisions about their routine implementation are limited.16

We therefore undertook a modelled cost‐effectiveness analysis of the health and economic effects of HealthTracker were it implemented in all NSW PHNs. Based on this simulation, we estimated the total costs of the intervention and its cost‐effectiveness over five years for managing people with high CVD risk.

Methods

Study design and setting

We conducted a modelled cost‐effectiveness analysis, from the perspective of the Australian health system, of the HealthTracker intervention and of usual care for adults at high risk of CVD. The reporting of our economic evaluation conforms with the Consolidated Health Economic Evaluation Reporting Standards (CHEERS) 24‐item checklist.17

HealthTracker intervention

In the TORPEDO trial (ACTRN 12611000478910), HealthTracker was integrated into the electronic health record systems of the health care providers participating in the intervention.10 The key features of HealthTracker are real time decision support based on evidence‐based national guidelines, a patient risk communication interface, an automated clinical audit tool, and a website providing anonymised feedback comparing practice performance with that of the other trial sites. The primary outcomes of the TORPEDO trial were the proportion of patients screened for CVD risk factors (smoking, systolic blood pressure, total cholesterol and high‐density lipoprotein cholesterol levels), and the proportion of patients at high CVD risk prescribed guideline‐recommended medications by the end of the trial.

Modelled population

The modelled population comprised NSW adults who were at least 45 years old (Aboriginal and Torres Strait Islander Australians: at least 35 years) and at high risk of CVD. The proportions of people at high CVD risk used (20.3% of non‐Indigenous, 6.3% of Indigenous patients) were derived from TORPEDO trial prevalence data.11 Consistent with Australian guidelines,8 patients were deemed to be at high risk if they had a history of CVD, had been diagnosed with certain clinical conditions (diabetes, chronic kidney disease, hypertension, total serum cholesterol exceeding 7.5 mmol/L), or their 5‐year CVD risk (Framingham risk equation) exceeded 15% (Supporting Information, part 1). We calculated the mean number of people at high CVD risk per NSW PHN using 2015 PHN demographic data (Supporting Information, part 2).18

Health economic model

We developed a decision analytic model to estimate the costs and benefits of HealthTracker over five years, comparing TORPEDO‐based findings of reduced CVD risk factor levels (LDL‐C, systolic blood pressure) with frequency of cardiovascular disease‐related events and health care costs (Supporting Information, part 3). We distinguished between people at high CVD risk who had not had a CVD event (primary prevention group) and those with a prior CVD event (ie, established CVD; secondary prevention group). The model was run in annual cycles and a 3% annual discount rate applied to costs.19 Incremental cost‐effectiveness ratios (ICERs) — the difference in mean costs divided by the difference in mean effects (CVD events) between HealthTracker and usual care — were calculated to estimate the cost over five years for each averted CVD event.

Major health outcome

The major health outcome for our cost‐effectiveness analysis was the estimated number of major CVD events — coronary heart disease, cerebrovascular disease, peripheral vascular disease — during the five years following the end of the TORPEDO trial, a period selected to match the 5‐year absolute CVD risk estimated with the Framingham risk equation.8 Estimated 5‐year CVD risk was adjusted according to measured changes in the levels of intermediate risk factors — low‐density lipoprotein cholesterol (LDL‐C) and systolic blood pressure levels — during the TORPEDO trial (Box 1). Small but statistically significant differences between the intervention and usual care groups in the reduction in LDL‐C level from baseline were found for both the primary and secondary prevention groups; for systolic blood pressure, the differences were not statistically significant.11 The relative risk reductions associated with reduced LDL‐C and systolic blood pressure levels were based on systematic reviews and meta‐analyses (Supporting Information, parts 4 and 5).20,21

Model assumptions

We assumed that the effects of the HealthTracker intervention (compared with usual care) on prescribing patterns and changes in intermediate risk factors in people with high CVD risk at the end of the TORPEDO trial were maintained during the five years following the trial. It was assumed that the price of lipid‐lowering drugs (statins) would drop by 50% with expiry of the applicable patents.

Baseline 5‐year absolute CVD risk was estimated at randomisation according to Framingham risk scores derived from mean population characteristics: 20% for the primary prevention group and 30% for the secondary prevention group.

By focusing on blood pressure and lipid levels at baseline and at the end of the TORPEDO trial, incomplete adherence to blood pressure‐ and lipid‐lowering medication prescribing was taken into account.11 The adherence rate for aspirin recommended as secondary prevention was assumed to be 66%, based on a meta‐analysis of 376 162 patients in 20 studies that assessed adherence according to prescription refill frequency.22

Cost estimation

Intervention costs associated with implementing and maintaining HealthTracker in health services and with patient‐level costs to the health system (hospitalisation with major CVD, medications) were derived from TORPEDO data (Box 2). Estimated annual software licensing costs were provided by the PenCS, the developer, and were assumed to be borne by the PHNs. The licence cost, estimated at $500 per health service per year or $141 500 per PHN per year, covered long term implementation costs (implementation, training, support, and maintenance), but excluded development and research costs. CVD medications were categorised as lipid‐lowering, blood pressure‐lowering, and antiplatelet and anticoagulant drugs. We used the dispensed price for maximum quantity and number of dispensed scripts listed by the Pharmaceutical Benefits Schedule23 to estimate the annual weighted mean cost for each medication group during the 2015 calendar year. Estimated hospital costs related to major vascular events were based on Australian Institute for Health and Welfare data for 2015–16.24

Sensitivity analyses

Model inputs were varied in one‐way sensitivity analyses: baseline 5‐year absolute risk (5% lower or higher for primary and secondary prevention); cost discounting rate (0%, 7%); the effect of patent protection expiry on the price of statins (25%, 75% reduction); and substituting the mean difference between the HealthTracker and usual care groups in LDL‐C level change with the lower or upper limits of the 95% confidence interval (CI) for the difference. Corresponding sensitivity analyses for systolic blood pressure values were not conducted because the difference in change during the TORPEDO trial was not statistically significant.

Ethics approval

This study was approved by the University of Sydney Human Research Ethics Committee (reference, 2012/2183) and the Aboriginal Health and Medical Research Council of New South Wales (reference, 778/11).

Results

We estimated the mean number of people in NSW with high CVD risk in 2015 to be 62 723 per PHN (Supporting Information, part 2). The prescribed medications profile of our model population, based on TORPEDO data, is summarised by treatment category in Box 3. In the primary prevention category, 7732 of 15 963 people (48.4%) in the HealthTracker group and 7065 of 15 963 (44.3%) in the usual care group received optimal therapy (lipid‐ and blood pressure‐lowering medications); in the secondary prevention category, 8517 of 15 398 (55.3%) in the HealthTracker group and 7445 of 15 398 in the control group (48.5%) received optimal therapy (all three of lipid‐lowering, blood pressure‐lowering, and antiplatelet/anticoagulant therapies).

The estimated number of major CVD events per 1000 patients over 5 years was 168 in primary prevention using HealthTracker and 176 for usual care; in secondary prevention, the estimated numbers of events were 259 with the intervention and 267 with usual care (Box 4).

Medication costs were higher for the HealthTracker intervention than for usual care, but the hospitalisation costs were lower for both primary and secondary prevention. The net difference in health care costs was estimated to be $59.25 per patient for primary and $142.11 per patient for secondary prevention. The estimated ICER for one averted CVD event was $7406 for primary and $17 988 for secondary prevention (Box 4).

One‐way sensitivity analyses indicated that ICERs were sensitive to changes in LDL‐C level, because LDL‐C level influences the number of CVD events and therefore hospitalisation costs. Should statin costs decline by 75% following patent expiry, the ICER would fall to $5104 per CVD event averted by primary prevention and $14 661 per CVD event averted by secondary prevention (Box 5).

Discussion

We evaluated the impact of the modest but statistically significant reduction in LDL‐C levels measured in the TORPEDO trial, with potentially important consequences for the incidence of CVD events. Our modelled cost‐effectiveness analysis of introducing HealthTracker in all NSW PHNs estimated that the health system cost per patient would be $59.25 for primary and $142.11 for secondary prevention; the estimated cost for averting one CVD event (ICER) was $7406 for primary and $17 988 for secondary prevention. One‐way sensitivity analyses indicated that the ICER was sensitive to changes in LDL‐C level and change in statin price after patent expiry.

Evidence for the cost‐effectiveness of quality improvement interventions at the PHN level that could inform investment decisions is scant. Funding for PHNs is based on local health needs and is intended to improve health outcomes through population health planning and the appropriate commissioning of services. Our study provides important evidence about the cost‐effectiveness of investing in quality improvement interventions such as HealthTracker. The strength of our study was its basis in high quality clinical trial data; it provides population‐level estimates of the costs and effects of implementing a computer‐supported quality improvement intervention in NSW. Our findings can assist PHN decision makers considering investing in health IT services.

Limitations

The daily operating costs of the HealthTracker intervention, including those for staff and other resources, were not included in our analysis because of their complexity and variation between health services. However, the HealthTracker intervention could improve care processes and reduce resource demand once it has been fully established. Our model estimated changes to CVD events over a 5‐year period, not taking into account CVD‐ and non‐CVD‐related mortality. Most importantly, we assumed that the risk factor improvements during the clinical trial would be sustained after its conclusion. We also assumed no subsequent changes in compliance with prescribing; medication adherence primarily declines during the first six months of treatment.25 Finally, primary health care organisations would incur transition costs during the introduction of HealthTracker, distinct from those for infrastructure and staffing, and such costs are not easily measured with cost‐effectiveness analytic methods. We have documented the determinants of transition costs for HealthTracker in a previous article,26 but they could not be included in this evaluation.

Conclusions

Our economic evaluation of the HealthTracker computer‐guided quality improvement intervention provides insights for decision makers considering strategies for achieving better value care for people at high risk of CVD. This could lead to broader benefits for the health system by reducing CVD‐related disability and improving system performance. However, the effect sizes for health outcomes achieved with HealthTracker were modest, and system performance could be improved further by more intensive quality improvement programs. This, however, would require consideration of health system complexity at the macro‐, meso‐ and micro‐health system levels and of strategies that can influence adoption and sustainability.