Addressing the deficiencies in the evidence-base for primary practice in regional Australia - sentinel practices data sourcing (SPDS) project: a pilot study

In Australia, under the 2011 Australian Government’s National Health Reform Agreement there has been a recent move for primary care services to be reorganized from former Divisions of General Practice into Medicare Locals throughout the country. These are not-for-profit organisations funded by the Australian Government Department of Health and Ageing (DoHA), that offer a diverse range of primary health care services, through service brokerage and direct delivery and purchasing to improve the health outcomes of the community. The Australian Government also indicates that a Medicare Local is a regional body that undertakes planning to identify the needs of a population and works with local health professionals to provide more integrated care. A more localized surveillance system may improve the understanding of health care needs of local communities, resulting in a more targeted and coordinated approach to meeting these needs through delivery of a range of health, allied health and other services.

In addition, an important part of the National Health Reform was the development of General Practice (GP) Super Clinics [1], aimed at improving access to quality primary healthcare and taking the pressure off the hospital system. These Super Clinics include a greater focus on health promotion and illness prevention, with better coordination between GPs and allied health services, community health and other state and territory funded services. Entities, including Medicare Locals, were invited to apply for Federal government funding to establish and develop the GP Super Clinics at various sites across Australia. Amongst others, the Illawarra Shoalhaven Medicare Local (ISML), located in a regional area of New South Wales, obtained funding to establish the first GP Super Clinic in the region. The facility opened for patients on 6^th July 2011, trading as Shell Cove Family Health (SCFH) [2].

Valid data on morbidity, at the regional level, is essential for the purposes of primary healthcare services planning, that are specifically tailored to the needs, demands and requirements of the local population. Nationally representative data is available through the National Health Surveys (NHS) conducted by the Australian Bureau of Statistics (ABS) [3]; and regionally through the annual New South Wales Population Health Surveys [4]. However, extrapolations of these data to smaller geographical areas such as Local Government Areas (LGAs) and/or small area geographic regions within LGAs like suburbs and Statistical Local Areas (SLAs) is limited. The state of New South Wales (NSW) has 15 Local Health Districts (LHDs) and 17 Medicare Locals (MLs). While the LHDs run the public hospitals across the nation and are responsible for acute, sub-acute and palliative care services, the MLs are responsible for primary healthcare planning and delivery for their constituent LGAs. Some of the densely populated LGAs are further sub-divided into constituent SLAs. For illustration, a map of the NSW Local Health Districts and a breakout map of the Metropolitan LHDs is shown in Figure 1. Within the Illawarra Shoalhaven region, distribution of the constituent LGAs and SLAs within the ISML and the Illawarra Shoalhaven Local Health District (ISLHD) catchment areas is shown in Figure 2[5].

The Public Health Information Development Unit (PHIDU) at the University of Adelaide compiles and provides synthetic predictions of the prevalence of chronic diseases and associated risk factors at LGA and SLA levels but these are produced using modelling techniques based on the survey data collected in the 2007–08 ABS National Health Survey (NHS) and known characteristics of smaller areas at that time attained from the ABS Census. A synthetic prediction is provided according to SLA, and is argued as being the likely value for a ‘typical’ area with similar characteristics. These techniques provide reasonable estimates of disease burden, but are limited in terms of planning health care services for niche groups within further smaller regions represented by SLAs or suburbs.

Moreover, because the NHS sample excludes the most remote areas of Australia, synthetic predictions cannot be generated for smaller geographic local areas with relatively high proportions of their population in remote areas. Thus, while national representation of the NHS is excellent, regional and smaller area figures devised from the survey data are estimations at best and have a significant likelihood of being inaccurate for regions with diverse and disparately distributed populations.

Epidemiological indicators are also measured in the ongoing NSW Population Health Surveys that has been conducted by the NSW Ministry of Health since 1997 for adults and since 2001 for children. However they too are reported at a state-wide level and further broken down to Local Health District levels only. Measures for SLAs and LGAs are neither available, nor is it statistically accurate, to be extrapolated from state-based prevalence figures. The sampling frame for the NSW Population Health Surveys is approximately 1,000 persons in each of the health administrative areas (Local Health Districts). Hence for every indicator a respondent sample size of 8000–16000 is obtained for the entire state of NSW.

Furthermore, all the estimates drawn from national and state level surveys are based on self-reported health and disease status data as provided by respondents of these surveys. The accuracy of self-reported data especially in terms of health services planning and mapping of regional disease epidemiology has always been questioned by health planners [7]. On the other hand disease diagnosis as noted in clinical softwares within hospitals and primary health care settings is clinically diagnosed and is supported by clinical diagnostic testing, examination and medical symptomatic evidences. Hence health data that is validated by the clinical judgement of medical practitioners offers a significantly higher level of accuracy of disease prevalence and is likely to be far more accurate than self-reported health information. Collation and analysis of clinical data thus provides a more evidence based platform for chronic disease surveillance for primary health care providers and for researchers to study regional epidemiology.

In much the same way that registries based in secondary care provide surveillance for cancers and other diagnosed diseases, there are examples of surveillance systems for chronic diseases located within primary care facilities from various parts of the world [8‐15]. Data collected routinely from sentinel sites (general practices) allows generalizable data to be collected from representative samples instead of the entire population. However, the success of sentinel site surveillance systems relies on motivation of participating practitioners, a well-functioning clinical practice database that allows for rapid data extraction, and timeous data collation and reporting.

The current pilot project aimed to assess the feasibility of implementing a sentinel site surveillance system within the newly formed GP Super Clinic, namely Shell Cove Family Health (SCFH), located in the Shellharbour LGA of the Illawarra-Shoalhaven catchment of New South Wales, Australia.

The Pen Computer Systems (PCS) Clinical Audit Tool™ (CAT) is a program that has been designed to collect and collate population-level health data for the purpose of chronic disease management. The program can be installed onto compatible desktop computer systems within General Practice and interfaces with the practice’s clinical software. It thus helps to simplify the review of clinical and practice management databases through analysis of practice data to identify chronic disease-related episodes of care, and associated risk factor prevalence. All practice clinical softwares utilise one of the several nationally validated health coding and medical classification systems, such as SNOMED-CT, DOCLE, PYEFINCH and ICPC2+. These are commonly known as medical vocabularies. Clinicians enter patient data using their practice software, according to these medical vocabularies, thus facilitating assessment of existing disease and health risk factor prevalence that is recorded during patient encounters at the practice. Higher level data analysis that is made possible through the program includes:

In all cases, judgments about the secondary use of practice data for these types of initiatives are made by General Practitioners. The PCS CAT has an automated de-identifying feature ensures privacy of patient information, allowing data to be extracted for research purposes.

The selected pilot site for the project was the newly formed GP Super Clinic - Shell Cove Family Health (SCFH) located in the Shellharbour LGA of the Illawarra-Shoalhaven catchment of New South Wales, Australia. The practice clinical software was Best Practice which uses the PYEFINCH medical coding vocabulary. Practice staff were informed about the aims and objectives of the study assisted the researchers in undertaking the data cleansing process on their clinical software system. The data cleansing phase of the study was conducted using the data maintenance utility tool which is available within all GP clinical software. Within Best Practice this is called the Cleanup history tool. Data cleansing included:

A cleaned, de-identified PCS CAT data extract was performed in July 2012 which included all information obtained from patient interactions in the preceding 15 months for all diagnosed pathologies, clinical variables such as anthropometric measures, patient demographic information such as age, sex, geographical location of residence (postcodes) and indigenous status. Extracted data was converted to usable database formats and then analysed using Microsoft Excel (V2007: Microsoft Corporation, Redmond Washington, USA) and IBM SPSS Statistical program (V19.0: 2010, IBM Corporation, New York, USA). The resultant SPSS and Excel databases hence included clinical diagnosis and patient demographic information as entered by GPs of the pilot site.

Basic epidemiological measures in the form of age specific prevalence and total prevalence were then calculated for all individual diagnosed conditions. These prevalence figures were matched and analysed with comparable indicators for same age groups as reported by the Australian Health Survey (AHS) 2011–12 conducted by the Australian Bureau of Statistics (ABS) [17]. The age specific disease prevalence figures obtained from the sample and the estimated national prevalence figures reported by the AHS 2011–12 were then age standardised using the 2011 estimated resident population of Australia [18]. Comparisons across age standardised prevalence were conducted for all major chronic conditions that the SPDS project is targeting for regular surveillance namely, obesity; overweight; diabetes mellitus; hypertension; asthma; mental health disorders such as clinically diagnosed depression and anxiety disorder; Coronary Heart Disease; Stroke; and Chronic Bone Diseases such as Osteoarthritis and Osteoporosis. Both Microsoft Excel (V2007: Microsoft Corporation, Redmond Washington, USA) and the PCS CAT tool (v.3.1: pencs.com.au) were used for graphical illustration of demographic data and age specific disease prevalence.

Other patient interactions that were recorded and analysed for relevant age specific cohorts were the uptake of preventive health checks as endorsed by the Australian Governments’ Medicare Benefit Scheme (MBS) Items [19] under the group: Health Assessments. Using the MBS schedule of January 2012 [19] and the MBS items claiming patterns of the practice for the actual health assessments undertaken, a potential revenue estimation matrix was created using Microsoft Excel (V2007: Microsoft Corporation, Redmond Washington, USA) and estimations of potential revenue lost were calculated in order to illustrate the added benefits of improving preventive health item uptake for the practice.

The study was performed with the approval of the Human Research Ethics Committee (Health and Medical) of the University of Wollongong (HE 12/447).

The number of patients that had visited the General Practice within the previous 15 months (30^th April 2011 to 31^st July 2012) was 3623 (1574 men; 2041 women; 8 no gender identified).

Median age for the sample was = 32 (IQR = 11–48) years. Children aged 0–4 years comprised the largest age group at 12.9% of the total sample, followed by the 35–39 year age group (9.9%), and 5–9 year old children (9.7%). Older adults aged 65 years and above comprised 8.5% of the sample. The age distribution of the population of the Illawarra Shoalhaven Medicare Local (ISML) coverage region is compared against that of the total Australian population in Figure 3, and the population pyramid of the sample is shown in Figure 4. The proportion of the local population who consulted the pilot practice during the study period is shown in Table 1. Proportions are based on the residential postcodes of the persons within the sample. A significant majority of the sample (72.3%) resided within the practice location postcode area and most of the remainder came from surrounding postcodes within the Shellharbour, Kiama and Wollongong LGAs.

The age specific population and disease counts within the sample and the age standardised prevalence comparisons of the sample and the Australian National estimates are shown in Tables 2 and 3 respectively. An illustration of age specific burden of disease (Figure 5) indicates that asthma and depression prevalence is much higher amongst younger age groups compared to older adults. While the proportion of older aged persons was comparatively smaller than middle aged adults (Figure 4), the proportional share of conditions such as hypertension, osteoarthritis, osteoporosis and diabetes are much higher in older age groups.

The number of preventive health assessment items, reimbursable by the Medicare Benefits System, that were performed during the recording period are shown in Table 4. While the sub-optimal uptake of preventive health checks indicates a lost opportunity to check on the health and well-being of identified vulnerable cohorts within the practice patient population, it also reflects on the missed revenue earning potentials for the practice. Potential revenue lost quantified using the assumptions drawn from the MBS items claiming patterns of the health assessments undertaken by the practice are illustrated in Table 5.

The Pen Computer Systems (PCS) Clinical Audit Tool™ (PCS CAT) offers the ability to collect and collate population-level health data for the purpose of chronic disease management. Through engagement with primary health services, data extraction may facilitate better planning for preventive as well as curative health services. To date, the desktop computerised system has been adopted by 62 Divisions of General Practice (now merged into lesser or equivalent numbers of Medicare Locals) throughout Australia. Its primary purpose is to provide Divisions the ability to audit their members’ National Performance Indicator (NPI) requirements, for funding allocations through Medicare reimbursements. Additional uses include the measurement of epidemiological indicators at a local and regional level for planning purposes, the uptake of various services by patients, including preventive strategies, and the ability to measure the effectiveness of programs delivered through general practice. This pilot study compared the prevalence of chronic diseases, and related risk factors, such as overweight and obesity and mental health indicators with data collected in recent national and regional surveys. It also assessed the uptake of primary prevention initiatives, including cervical screening, immunisations, medication profiling and home medicine review.

This data obtained from patient interactions provides an accurate and timely picture of the major primary health care needs of patients that access general practitioner services within an identified smaller geographic region within a local government area. For example, it is clear that activities that target weight reduction and prevention of hypertension are required in the catchment area of this General Practice. Compared to estimates obtained from the ABS Australian Health Survey 2011–12, this pilot study identified higher than national figures for the prevalence for clinically diagnosed anxiety, hypertension, and overweight and obesity, but a lower prevalence of asthma, COPD, chronic heart disease, depression, osteoarthritis and osteoporosis. The prevalence of diabetes was similar to Australian national estimates. While these prevalence figures represent only the section of the population that goes to a GP and can thereby be argued to be a biased estimate of regional population disease prevalence; the figures do help primary care initiatives to target vulnerable groups that present to general practices with an identified need to address specific chronic conditions. They hence enable primary health planners to tailor healthcare services to meet local needs of the populations that actively visit general practices to receive primary healthcare services. It is not possible to determine from this single pilot practice data source whether the age-standardized prevalence figures are truly reflective of regional differences, or whether they may be a result of some disease classification and coding anomalies, nor is it possible to hypothesize why such differences may exist. This will be a topic of further larger studies.

The collection of morbidity data in primary care can be based on either (a) episodes of care, whereby patient interactions are recorded, or (b) a limited number of specific disorders that are standardized using case definitions, usually defined according to the nationally recognised disease coding systems. The former provides information on patterns of clinical management, such as the method employed in the English General Practice Research Database Programme [11, 20], whereas the latter provides a picture of the burden of disease in the population, as in the Morbidity Sentinel Stations Programme that operates in several European countries [13‐15, 21, 22]. In both these surveillance models, data generated from general practice settings have been shown to be useful [23]. Our pilot study obtained data using both of these models, however broader application of the PCS CAT system for the purpose of surveillance will require further streamlining of the type of data to be extracted. It could be argued that morbidity data should be based on individuals rather than the patient-provider consultations as the latter reflects workloads and case management rather than burden of disease in the epidemiological sense [11].

In Australia, as in most other countries, general practitioners are the first source of referral within the larger health care system, and thereby provide an optimal opportunity for effective monitoring of morbidity data of the population. The internal validity of the surveillance programme will depend on the accuracy of classification of disease at the point of data entry by practitioners, together with a reliable source of denominator data. An example of how inconsistencies in definitions of hypertension can impact on variations in age-adjusted prevalence figures and estimates of hypertension control collected in surveillance systems is described by Crim and colleagues [24]. Different definitions, using data collected in the National Health and Nutrition Examination Surveys (NHANES) (2003/4 onwards) has resulted in varied estimates of prevalence, ranging from 29% to 32%, and levels of hypertension control, from 35% to 64%.

In France, the Public Health Act of August 9, [25] listed 100 health targets for a 5-year period, and is currently under review for updated health targets. Evaluation of the effectiveness of public health programs to meet these targets requires representative data on indicators of infectious and chronic diseases chronic diseases at a regional level. The existence of a national health identification number and the development of a national information system inter-operability framework that has the capability to interchange data between heterogeneous systems, paves the way for implementation of such an epidemiological surveillance system. The Rhône-Alpes regional health platform that began in the 2000s and by August 2011, had 2.6 million patients in its repository, provides a feasibility study for widespread implementation of regional platforms for managing electronic health records [26].

Sharing of electronic health records between care facilities to improve health care coordination has been the object of major investment in many countries. Programs have been launched, and are at various phases in development and implementation in Australia (the national HealthConnect program), New Zealand (national program), the United Kingdom (national program managed within the National Health Service), France (personal medical record or DMP project) and the United States. Projects in these countries are currently in various phases of development and/or implementation with none being fully operational as yet. An added, but still to be explored, major benefit of these interfaced data recognition systems will be epidemiological monitoring and surveillance. In the US, the Centers for Disease Control and Prevention (CDC) set up a National Center for Public Health Informatics in 2005, but reports that the “fragmentation of population health data collection, and data stewardship responsibilities among federal, state and local governments” [27] remains the greatest barrier to a creation of a population health record (popHR) in the United States. The definition of a popHR is “…aggregated and usually de-identified data. It may be obtained directly from EHRs or created de novo from other electronic health repositories. It is used for public health and other epidemiological purposes, research, health statistics, policy development, and health services management” [28, 29]. We propose that the PCS CAT system provides a feasible method by which to obtain a popHR in the Australian primary care setting.

The PCS CAT system also offers potential to monitor uptake of preventive health activities that are offered through the existing Medicare Benefits Scheme (MBS). Reimbursement is provided to general practitioners for 4 time based MBS health assessment items: 701 (brief), 703 (standard), 705 (long) and 707 (prolonged) [30]. Under these items a medical practitioner is able to undertake a range of health assessments including: Healthy Kids Check (children aged 3 – 5 years, or who are receiving their 4 year old immunisation); a health assessment for people aged 45–49 years who are at risk of developing chronic disease; a type 2 diabetes risk evaluation for people aged 40–49 years with a high risk of developing type 2 diabetes as determined by the Australian Type 2 Diabetes Risk Assessment Tool (AUSDRISK) [31]; and a health assessment for people aged 75 years and older (75 + HA) [30]. Our data has identified a low uptake of these MBS-funded items by eligible patients in the practice, despite a recent focus on preventive health care, as evidenced by significant government investment in the establishment of an Australian National Preventive Health Agency [32]. Australia is one of a number of countries that have recognised the need to re-orient health systems from a treatment model to a preventive health-promoting model. However, our data suggests that within primary care, there is a need to encourage individuals to access these health checks as a starting point for dialogue around healthy lifestyles.

The main limitation of the study relates to validity and generalizability of the prevalence data. This pilot study was undertaken to assess the feasibility of obtaining prevalence data on chronic disease conditions using data that is recorded by practitioners during episodes of care, rather than providing a comprehensive snapshot of the population. Generalizability of data collected from sentinel sites requires representativeness of the general practice clientele to the broader population in that geographical area, which will require statistical techniques to identify practices, both large and small, that are relative to the size of the populations that they serve. This approach will be adopted in the same geographical region as a follow-on from this pilot project.

One of the main benefits of using general practice databases for disease surveillance is the ability to access data from large patient populations across a wide population coverage. However, the data are collected primarily for clinical and routine use, rather than for surveillance or research purposes. Hence, data quality and reliability may be compromised [33] and additional data cleaning is required before being extracted to a usable format, as was performed in the present study. The SPDS study relies on the clinical judgement of GPs for accuracy of disease diagnosis and assumes inclusiveness of GP history-taking, clinical inquiry and data recording skills. The validity of medical terminology and coding systems such as PYEFINCH, DOCLE and ICPC2+ also need further standardisation. With the introduction and implementation of new e-health requirements in Australia in February 2013, the National E-Health Transition Authority (NEHTA) has advised all General Practice clinical software vendors to start working towards mapping their local medical vocabulary against SNOMED-CT, that has been identified by NEHTA as the preferred nationally recognised disease classification or terminology system [34]. In due course, coded General Practice clinical data in Australia may be considered a reasonably valid and reliable source of information for mapping regional disease prevalence and conducting surveillance of chronic conditions. Such data is particularly useful for planning of primary care services to meet the needs of local populations.

This pilot project has demonstrated that systematic data that is routinely entered into desktop software in General Practice could form the basis of a valid and sensitive surveillance system on chronic diseases, provided that sufficiently representative samples of sentinel sites are recruited within Medicare Local regions.