Cardiovascular disease burden in Australia

Despite recent gains, cardiovascular disease (CVD) remains Australia's biggest killer accounting for 18% of the total disease burden and 11% of health system expenditure in Australia.1 Aboriginal and Torres Strait Islander peoples experience around five times greater CVD burden than other Australians.2 Current estimates project that by 2030 annual CVD expenditure will rise by around 100% to $16 billion.3 Primary care-based strategies that improve the uptake of best-practice recommendations could substantially reduce both the Indigenous and non-indigenous CVD burden and help improve health system efficiencies.

Evidence-practice gaps in CVD prevention

In addition to lifestyle modification, a number of drug therapies have been shown to be highly effective in preventing cardiovascular events, primarily through modification of blood pressure, lipids and platelet function.4–8 However, there is compelling evidence of the failure of current clinical practice to adequately implement such treatments, and to translate current knowledge into maximally improved health outcomes. Three recently completed cross-sectional studies of CVD risk management in Australian general practice and Aboriginal Community Controlled Health Service (ACCHS) settings9–11 demonstrated that 50% of routinely attending adults lacked sufficient recorded information to comprehensively evaluate vascular risk. For those identified at high vascular risk, only around 40% were prescribed guideline-indicated medicines. Similar findings have been noted in other Australian studies.12–15 These surveys have demonstrated failure to adequately implement the ‘absolue risk’ paradigm for CVD prevention. Numerous tools are now available to estimate an individual's 5-year or 10-year absolute risk of coronary heart disease and/or CVD.16–21 Despite their availability, only a minority of Australian general practitioners (GPs) use these risk assessment tools, and then primarily for patient education, rather than to guide management decisions.9 ,22 ,23 Australia's first absolute risk assessment guideline was released in 2009 by the National Vascular Disease Prevention Alliance (NVDPA)17 and in 2012 this was augmented by a single management guideline.24 Despite these guidelines now becoming available, there remain substantial challenges in effectively implementing their recommendations. We have found that CVD-risk assessment and treatment work best when negotiated as part of a shared decision-making approach, taking an average of 15 min even where only one guideline needs to be consulted.25

The role of electronic decision support in closing evidence practice gaps

Electronic decision support (EDS) systems are among the most promising interventions to improve uptake of guideline-based recommendations in clinical practice. In five systematic reviews on the effectiveness of EDS, around two-thirds of studies demonstrated improvement in practitioner performance.26–30 One systematic review identified four decision support system features associated with improved performance: incorporation in routine work flow, provision at the time and location of patient consultation, use of computer-based tools and provision of treatment recommendations rather than just assessments.28 Of 32 systems that incorporated all of these elements, significant improvements in performance were noted in 30. There are relatively few controlled evaluations of EDS systems that are integrated with electronic health records (EHRs) in the area of CVD.31–35 Effect sizes vary greatly depending on the variables studied and the type of EDS system. In one systematic review of on-screen point-of-care reminder systems the absolute improvements ranged from 1% to 24% for test ordering and from 3% to 28% for medication prescribing.27 In New Zealand, an EDS system that is fully integrated with the country's most popular primary care software has been successfully implemented.36 To date, we are unaware of an EDS system aimed at assisting comprehensive cardiovascular risk management based on Australian guidelines. Furthermore, we are not aware of any randomised evaluations of such systems in Australian primary care settings. Globally, few examples exist and the evidence base remains poor.

Algorithm development and validation

A single screening and management algorithm was developed based on a synthesis of recommendations from several primary care screening and management guidelines (table 1). The algorithm calculates a person's 5-year absolute CVD risk based on the Framingham risk equation and NVDPA recommendations 17 ,20 and provides management recommendations based on the guidelines listed in table 1. In 2008–2009, a β-version of HealthTracker was developed in a stand-alone software system and independently validated for accuracy and compliance with the prevailing guidelines.22 In 2011, the algorithm was extensively revised to incorporate recommendations from newly published guidelines. A similar validation process was then conducted consisting of three levels:

• Level 1 was an iterative process where each of the calculations programmed in the algorithm were tested to ensure they were consistent with recommendations from the guidelines. This was conducted using deidentified data from 337 patients involved in the pilot. Programming modifications were made where necessary and all variables were retested to ensure they were programmed correctly.

• Level 2 involved giving a plain language summary of the algorithm to a research fellow who had not been involved in the development of the algorithm. She independently programmed the algorithm into a statistical software package. Using data from 9077 patients from three representative cross-sectional general practice surveys,9–11 we then assessed whether the outputs from HealthTracker correlated with those generated from the independently programmed version. For 60 of the 63 output variables HealthTracker achieved perfect correlation with the independently programmed version. For the remaining three variables minor programming errors were identified and corrected.

• Level 3 involved user acceptance testing and scrutiny of the algorithm by the study investigators, 20 health professionals working in both General Practice and ACCHSs and three national professional organisations—the NVDPA, the Royal Australian College of General Practitioners and the National Prescribing Service. Following this feedback, a number of minor algorithm and user interface changes were incorporated into the final version of the tool. The level 2 testing process was repeated following these changes and perfect correlation between HealthTracker and the independently programmed version was achieved for all variables.

Integration of HealthTracker with the primary care electronic health record and quality improvement tools

HealthTracker interfaces with the two Australian clinical practice software systems most commonly used in general practice and ACCHS settings (Medical Director and Best Practice). There are four components to the system:

• Point-of-care decision support: HealthTracker is built in the Pen Computer Systems PrimaryCareSidebar, third-party software that interacts with the primary EHR system. Figure 1 shows the HealthTracker user interface and its integration with the PrimaryCareSidebar and the EHR. A prompt function is used to encourage health professionals to conduct a cardiovascular assessment if guideline recommended. Where possible, the tool populates with information from the patient's record. If essential information required for the calculation of absolute risk is missing or out-of-date, a traffic light prompt alerts the health professional and updated information can be entered. If the patient is receiving suboptimal treatment then a traffic light recommendation is made to consider initiation of treatment or additional agents. Information about eligibility for the Australian Government Pharmaceutical Benefits Scheme subsidy is provided if lipid lowering medicines are recommended. All outputs are qualified by statements emphasising that the final decision to commence or change therapy should be made by the health professional based on all available information.

• A patient-oriented risk communication interface: A key finding from the pilot evaluation was the role of the output in risk communication. GPs wanted to interactively alter current risk factors and perform ‘what if’ scenarios to demonstrate to patients the effects of current and altered risk over time. This functionality has been built into HealthTracker and uses the concept of ‘Heart Age’ to demonstrate to patients the discrepancy between current risk and an ideal risk based on well-controlled risk factor levels. Figure 2 shows an example of how a patient's heart age changes with the effect of smoking cessation.

• A data extraction tool: This provides health professionals with immediate feedback on their performance on screening and management of CVD risk for their entire patient population. Figure 3 shows an example of screening performance for a range of CVD risk factors. Practitioners can use this tool to identify specific patients in whom there may be a particular risk factor measurement missing or a potential prescribing gap. Customised point-of-care prompts can then be created. When a patient record is opened an alert is provided to notify the practitioner of the particular management issue and this can then be actioned.

• A quality improvement (QI) component: This is aligned with the methods of the Improvement Foundation of Australia (IFA) Australian Primary Care Collaboratives (APCC) programme. Deidentified data extracts of clinical performance are securely exported to a web-based central repository managed by the IFA. This repository provides access to site-specific feedback reports on performance compared with other anonymised sites. Figure 4 shows an example of how this information is presented.

• Download figure
• Open in new tab
• Download powerpoint

Figure 1

HealthTracker_user interface.

• Download figure
• Open in new tab
• Download powerpoint

Figure 2

HealthTracker—cardiovascular disease-risk communication interface.

• Download figure
• Open in new tab
• Download powerpoint

Figure 3

Sample output of performance in cardiovascular disease-risk factor screening.

• Download figure
• Open in new tab
• Download powerpoint

Figure 4

Sample display from the quality improvement portal.

Hypotheses

Using a cluster randomised, controlled trial design, two specific hypotheses will be tested. Compared with control practices, those practices randomised to receive HealthTracker will have:

1. An increased proportion of patients receiving appropriate (guideline-indicated) measurements of their CVD risk factors.

2. An increased proportion of patients at high risk receiving appropriate (guideline-indicated) prescriptions for the management of their CVD risk.

These aims will be augmented by formal economic and process evaluations to provide crucial information on large-scale implementation and sustainability.

Eligibility criteria

Randomisation

Sixty services (40 general practices and 20 ACCHSs) will be randomised in a 1:1 allocation to use HealthTracker or ‘usual care’ for 12 months. Clusters will be stratified at three levels:

1. ACCHS versus general practices.

2. Service size (<500 patients meeting eligibility criteria vs >=500).

3. Participation in existing QI programmes (current involvement in one of five national and state programmes involving regular audit and feedback versus past or never involved in these programmes).

A site assessment survey will be administered to all sites to assess for service eligibility and these stratifying variables. Permuted block randomisation will be centrally performed using a web-based form. As this is a pragmatic trial, allocation will be single blinded with outcome analyses conducted blinded to treatment allocation.

Primary outcomes

• Change in the proportion of eligible patients receiving appropriate measurements of their CVD risk in the previous 12 months (measured at randomisation and at 12 months).

• Change in the proportion of eligible patients assessed at high CVD risk receiving appropriate prescriptions for their CVD risk factors in the previous 12 months (measured at randomisation and 12 months).

Appropriate measurement of CVD risk factors is defined as having recorded or updated all the essential risk factors for the measurement of CVD risk (smoking status, blood pressure (BP) in the previous 12 months, total cholesterol and high-density lipoprotein (HDL) cholesterol in the previous 24 months) among those in whom risk assessment is guideline indicated. Unless explicitly recorded, diagnoses of diabetes or left ventricular hypertrophy will be assumed to be absent.

High CVD risk is defined as a calculated 5-year CVD risk of >15%, a history of CVD or the presence of any clinically high-risk conditions (as per NVDPA recommendations). Based on audit data, this is expected to comprise ∼30% of the patient population.10 ,11

Appropriate prescriptions is defined as a prescription for one or more BP lowering drugs and a statin for people at high risk without CVD; or a prescription for one or more BP lowering drugs and a statin and an antiplatelet agent (unless contraindicated by oral anticoagulant use) for people with established CVD.

Secondary outcomes

• Change in the measurement of individual risk factors separately (smoking status, BP, cholesterol, other non-Framingham risk factors—BMI, chronic kidney disease (CKD) screening with urinary Albumin to Creatinine ratio, estimated Glomerular Filtration Rate);

• Intensification of existing medication regimes among patients at high CVD risk (additional BP and lipid-lowering agents);

• Changes in mean systolic BP, total cholesterol, LDL cholesterol and HDL cholesterol;

• New CVD and CKD diagnoses.

Statistical considerations

Randomisation of 60 services (30 per arm) will provide 90% power to detect a ≥10% absolute higher occurrence in each primary study outcome among practices receiving HealthTracker. The following assumptions are based on our three audits in ACCHSs and mainstream general practices9–11 and include an assumed improvement of 10% in the two primary outcomes in control practices as a result of study participation.

1. Cluster size of eligible population will range from 200 in a small service through to 2000 in a large service. An average cluster size of 750 is assumed.

2. Recording rates of essential risk factors needed for risk assessment in the target group (first primary outcome) average 50%.10 ,11

3. Thirty per cent of the cluster will be either be at high CVD risk or have established CVD (n=250) and prescription of appropriate medicines to high-risk patients (second primary outcome) is 50%. This is based on our published data on drug prescriptions for individuals with and without established CVD.9–11

4. An intraclass correlation coefficient of 0.05 for both primary outcomes based on our audit data.9–11

5. Two-sided α=0.05

Data analysis will be performed on an intention-to-treat basis using generalised estimating equations.38 Subgroup analyses will be carried out using the three prespecified strata: (1) ACCHS versus general practices, (2) service size (small vs large) and (3) current participation in QI programmes vs past or no involvement in these programmes.

Economic evaluation

The economic evaluation will have a trial-based component and a modelled evaluation of long-term costs and outcomes. The incremental cost will be based on software, training and other costs incurred with transitioning practices to using HealthTracker. This will help determine the cost barriers experienced by different practices in adopting the system. Data on medications, laboratory tests, consultations and newly recorded diagnoses of CVD events incurred by eligible patients during the trial will be obtained from the data extraction tools. Costs will be calculated from prevailing Medicare rates and standard Australian National Diagnosis Related Groups cost weights for CVD hospitalisations. The incremental cost consequences of the HealthTracker system in achieving each of the primary outcomes will then be estimated, for example, cost per eligible patients assessed at high CVD risk receiving appropriate prescriptions. Trial-based data, however, cannot capture costs and outcomes beyond the trial. To address this, a modelled economic evaluation will enable quality of life and survival to be examined and allow incremental cost-effectiveness ratios to be calculated in terms of cost per Quality Adjusted Life Years gained. Using a Markov model, the eligible patient population in both study arms will be hypothetically tracked over an extended period. Transition between various defined health states, costs and quality of life attached to various health states and the projected long-term intervention effects from that observed in the trial will be based on published evidence. With appropriate discounting, estimates of long-term costs and outcomes will fold out of the model. Sensitivity analyses will be conducted on discount rate, uncertainty in outcome estimates and assumptions made in costing (eg, varying efficiencies with different patient practice ratios to those of the trial setting). This will better inform policy makers as to the resource consequences of rolling out this programme to scale.

Process evaluation

The qualitative evaluation of the β-version of HealthTracker suggested that a critical factor affecting the uptake of EDS interventions is whether and how they become embedded in routine healthcare.22 In the TORPEDO trial we will build on this observation through a detailed process evaluation to better appreciate the factors that might influence sustainability beyond the trial setting. Two qualitative methods will be used to explore these factors.