We undertook a retrospective cohort study based on analysis of existing medical records of adult patients cared for at participating general practices between 1 July 2013 and 30 June 2018. All practices provided fully subsidised care to their patients (which is known as ‘bulk-billing’). Participating practices were those within the Healius network (previously known as Primary Health Care) that used Medical Director software - this group comprised 43 centres from a network of 71. The remaining 28 practices were using software other than Medical Director and so data were not available for extraction and analysis. All centres in the network are transitioning over to the Medical Director software according to a schedule. There were no other regional or socioeconomic differences between those included in our study and those omitted for software reasons.
The use of prescribing and electronic health records in Australian general practices has been widely adopted such that by 2005, Australian general practice had achieved near-universal clinical computerisation7, 8. Medical Director software is one of two dominant providers of practice software in Australia, providing 4300 GP practices and 13600 GP users with its practice software9.
To identify patients with HF, a search of records was undertaken using Structured Query Language (SQL). The list of screening words was broad so that cases would be unlikely be missed. Search terms to identify a cohort for extraction and full analysis included HF diagnostic terms, HF-specific medication use, signs and symptoms of HF, pathology test results indicative of HF, the diagnosis of an aetiological condition for HF (Table 1) and a referral for cardiac imaging, principally echocardiography. These criteria were developed with expert opinion advice or from current Australian HF evidence-based guidelines10. All patients visiting the practices (with and without heart failure) were included and heart failure hospitalisation was not a prerequisite for being included in the study.
The cases were de-identified, removing all potentially identifiable data from the records, then provided to the researchers for analysis. Data were extracted from the following fields in the medical records: diagnosis, reason for presentation, prescriptions, vital signs, pathology results, specialist referrals and clinical notes. Chronic disease management item numbers billed to Medicare were also extracted. Each patient was allocated a unique study number so that re-identification would be possible by Healius for future scrutiny of records for any reason (for example, missing data). This allowed records belonging to the same patient to be linked through time so that GP visits and management for each patient could be identified.
To ensure data integrity, consistency and completeness of the data extraction, a detailed quality control process was performed. A registered nurse who was an experienced study coordinator employed by Healius examined the records of a random sample of 50 identified patients to ensure that the query collected the correct data from the correct patients. The study coordinator also performed a disease register search of HF to make sure that the query did not omit from the extract any potential HF patients. This quality control process confirmed that the data extraction produced the correct patient level results and showed that the query was comprehensive so that HF patients were very unlikely to be omitted.
The study’s primary endpoints were the prevalence and incidence of HF, stratified by age and gender, and standardised to the 2017 Australian population. We also sought to determine the demographics of the HF population and their clinical characteristics, including aetiological factors, comorbidities, symptoms of HF, examination findings and medication use. Other factors examined included the proportion receiving HF medications, the proportion receiving medications that are contraindicated in HF, the frequency of GP visits, the use of GP chronic disease management Medicare services, the use of mental health services, and the frequency of referrals from GPs to specific types of specialists.
Included patients were those who were aged 18 years and above, and who had one or more of the following criteria recorded in their medical record: i) a specific diagnosis of HF (Table 2);ii) were receiving ongoing treatment with a HF-specific medication (Table 3); iii) presented with signs or symptoms of HF (Table 4 and Table 5); or iv) had pathology test results indicative of HF (Table 6 and Table 7). In Australia, the HF-specific medications listed in Table 3 have a ‘Restricted Benefit’ in the Pharmaceutical Benefits Scheme (Australia’s list of subsidised medications) to ‘moderate to severe heart failure’. Furthermore, the restriction stipulates that patients must be stabilised on conventional therapy, which must include an angiotensin converting enzyme inhibitor or angiotensin II antagonist, if tolerated11. In the search of text fields, certain criteria were selected for common synonyms, which are listed in Appendix - Free text search terms. If certain words preceded the selected words in the notes, then the condition was considered not to be present in those notes. For example, if there was a mention of ‘shortness of breath (SOB)’, but this was preceded by ‘No’, ‘Nil’, or ‘denies’, then SOB was considered not to be a problem for the patient at that time.
The search term ‘PND’ was found to produce a lot of false positive results (also being used for other conditions, such as post-natal depression, and post-nasal drip). A review of 2000 records with ‘PND’ was undertaken and this included 1151 with nasal symptoms, 659 with upper respiratory tract infection (URTI), 515 with sinusitis, 169 with lower respiratory tract infection (LRTI) and 63 with depression. However, the term ‘PND’ was still included, but non-HF causes were excluded and further supporting evidence (ejection fraction data, BNP data, or loop diuretic use) was required in order for a case to be classified as definite or probable HF.
The analysis assessed the number and combinations of relevant terms and cut-off criteria in a hierarchical approach. The population was then stratified into three groups based on a hierarchy of selection criteria: (1) definite HF, (2) probable HF and (3) possible HF (Table 8). The eligibility criteria for ‘definite HF’ were: HF coded in the field of diagnosis codes; any mention of HF diagnoses in the free text fields; prescription of HF-specific drugs; BNP/ NT-ProBNP above HF cut-offs; recorded ejection fraction (EF) <40%, EF ≥40 - <50% and typical symptoms and signs recorded in the notes; and EF ≥40 - <50% & use of a loop diuretic The criteria for ‘probable HF’ were recorded EF ≥40 - <50%, or typical symptoms and signs of HF recorded in the notes & any of the following: BNP/ NT-ProBNP in the inconclusive ranges, use of a loop diuretic, or documented EF > 50%.
In Australia, the Federal Government mandates through the Pharmaceutical Benefits Scheme (PBS) that the HF-specific drugs used in our search and analysis are for to be prescribed for the management of heart failure only. General practitioners are unlikely to stray from these restrictions – education must be in line with the PBS listing and Medicare can perform audits on GPs’ practices. Also, as HF is a clinical diagnosis that can also be inferred from response to treatment, it would be highly likely that patients with the constellation of symptoms described plus prescriptions for diuretic medication/s would have heart failure, even if no specific diagnosis has been entered or other more specific HF medications not initiated.
Data analysis was conducted using SAS for windows (version 9.4). For laboratory and other data, the most recent measurement for each patient of each parameter was selected for analysis. If any of the selected drugs were taken at any time by a patient during the whole period under study, then that patient was identified as having been prescribed that drug. Medications prescribed following the diagnosis of HF was also reviewed. Referrals (to a cardiologist, endocrinology or renal physician) were recorded for a patient only if the referral occurred around the time of diagnosis of HF, or later. That period started one month prior to HF diagnosis and then onwards. This presumed that the referral to the specialist was the time when the GP was suspecting HF and was seeking specialist involvement. We also assessed referrals starting from seven months prior to diagnosis, which allowed for patients to have been seen by a specialist, provided with six months of prescriptions and so only needed a GP consultation after this period. In this case, the diagnosis may only appear in the GP records up to a maximum of seven months after the specialist visit.
The point prevalence and annual incidence of HF were calculated, along with their 95% confidence intervals. From the age- and gender-specific rates of HF, and estimates of the Australian population in these subgroups, prevalence and incidence were age-standardised to the 2017 Australian population overall, and by gender.
The calculation of prevalence and incidence of HF involved only ‘active’ patients; that is, those patients with at least three visits per two-year period12. This approach avoided the under-estimation of prevalence and incidence that would have otherwise arisen from over-inflation of the denominator data by one-off or infrequent GP visits. Such visits would be more common in bulk billing centres. Furthermore, among people who are not regular patients of the centres, medical records may not contain sufficient information on which to assess the presence of a HF diagnosis. In secondary analyses, denominators were estimated from the total number of patients seen at the participating GP clinics during each calendar year for the period under study. Overall prevalence within gender and age groups was calculated, along with the proportion of cases within each of the gender and age-groups.
The numerator for the prevalence of HF was obtained by tabulating the numbers of HF cases by age group and gender over the five-year period under study. Annual incidence of HF was reported similarly to prevalence, except that only new cases were included, based on the date of first diagnosis of HF. In an attempt to remove from the file the cases with pre-existing HF, we identified cases where the diagnosis of definite and probable HF was made or was present during the first year of data collection and removed these from the file. This meant that cases which remained in the incidence calculation had no mention of HF during the first year of the data collection.