Temporal Trends and Patterns in Mortality After Incident Heart Failure

Key Points Question Why has there been no improvement in the prognosis for patients with heart failure over the past 15 years when considerable advances in heart failure care have been introduced during the same period? Findings In this cohort study of patients who received a new diagnosis of heart failure between 2002 and 2013 in the United Kingdom, cardiovascular mortality declined by 27% and premature deaths from any cause declined by 21%. Improvements to overall mortality were hindered by noncardiovascular diseases, which represented most deaths and increased by 22% over time. Meaning Management strategies that solely target cardiovascular outcomes appear insufficient to improve the survival of patients with heart failure; the management of associated comorbidities, particularly infection prevention, appears as a major priority and opportunity.


eAppendix 1: Approach used to define disease categories for causes of deaths and hospitalizations
The CPRD data used in this study provides causes of death and hospitalizations in the form of clinical codes from the ICD-10 (International Classification of Diseases 10 th edition) system. We sought to categorize causes of death and hospitalizations into a set of unique and clinically meaningful disease categories. For that purpose, we used the following approach, which we applied to deaths and hospitalizations separately: (i) As a starting point, categories were defined as each code's overarching ICD chapter (n = 22); (ii) Chapters that presented similarities for the purpose of this study were grouped together (e.g. ICD chapters "Injury, poisoning and certain other consequences of external causes" and "External causes of morbidity and mortality" were grouped together and categorized as "Injuries"); (iii) Groups of conditions that, in 2013, represented at least 2% of deaths (for the categorization of causes of deaths) or either 2% of deaths or hospitalizations (for the categorization of causes of hospitalizations), e.g. kidney diseases or infections, were defined as independent categories; (iv) Any remaining category that represented less than 2% of deaths/hospitalizations in itself was classified as "other". In the UK, guidance for completing medical certificates states that heart failure is not a cause but a mode of death and discourages doctors from recording heart failure as the underlying cause of death. 1,2 Therefore, heart failure was defined as an individual disease category for hospitalizations, yet not for deaths.
We further report the care setting in which heart failure was first diagnosed. Diagnoses recorded during a hospital admission were further categorized based on whether heart failure was listed in primary or secondary diagnostic position. Diagnoses first recorded in primary care are likely to reflect both outpatient consultations by specialists and direct diagnoses by general practitioners.
To describe co-morbidities, we selected 17 common chronic conditions (anemia, asthma, atrial fibrillation, cancer, chronic kidney disease, chronic obstructive pulmonary disease, dementia, depression, diabetes, dyslipidemia, hypertension, ischemic heart disease, obesity, osteoarthritis, peripheral arterial disease, stroke, thyroid disease). Diagnosis code lists for the extraction of each condition were adapted from the CALIBER code repository. 3 To describe socioeconomic status, we used the Index of Multiple Deprivation (IMD) 2015 quintile, 4 a composite measure of relative deprivation at a small area level, ranked in ascending order of deprivation score and grouped in equal fifths.
Ethnicity is reported as recorded in patient's electronic health record. When ethnicity differed between primary and secondary care records, secondary care data was used. To assist readability, ethnicity was grouped into two categories, 'white' and 'other'.
Baseline characteristics are presented as frequencies (%) for categorical data, medians and interquartile range (IQR) for non-normally distributed continuous data, or means and standard deviation (SD) for normally distributed continuous data.

eAppendix 3: Validity of diagnoses recorded in electronic health record databases
Research using electronic health records databases is reliant on the accuracy of clinical coding input by physicians in primary care, as part of a consultation, or secondary care, as part of a hospital admission. The validity of diagnoses underlying our study has therefore been carefully assessed and was considered appropriate in light of the following arguments. Independent validation studies. Three studies are of major importance: (i) a systematic review, published in 2010, reports 212 validation studies over a broad range of conditions with an average positive predictive value of 89% 5 ; (ii) a study specifically investigating heart failure diagnoses, which despite it being conducted before the introduction of national care monitoring programs reports a positive predictive value of 82% 6 ; and (iii) a more recent study investigating the validity of chronic obstructive pulmonary disease (COPD), another major chronic condition managed in primary care, which reports an accuracy of 87% compared with specialist assessment. 7 National care monitoring programs. Two national clinical audit programs (in particular the 'quality and outcomes framework' (QOF) introduced in 2004 for primary care, and the 'national heart failure audit' (NHFA) introduced in 2007 for secondary care) ensure a stable quality of clinical coding practices and provide a solid support for the validity of recorded diagnoses. Indeed, these report that approximately 90% of recorded heart failure diagnoses in England are referred for echocardiography, specialist assessment, or B-type natriuretic peptide (BNP) measurement. 8,9 Clinical guidelines. Guidelines for the diagnosis and management of heart failure from the National Institute for Clinical Excellence (NICE) 10,11 provide additional consistency over the study period. Indeed, guidelines are largely consistent in regard to heart failure diagnostic criteria and recommended investigations. One important change is the availability of natriuretic peptides testing and the variability in assay accuracy; these are however mainly used to exclude suspected cases, as opposed to confirming diagnoses, and therefore unlikely to impact disease incidence rates. 10,11 Sensitivity analyses. Finally, to confirm the validity of heart failure cases included in our cohort, we performed the following sensitivity analyses. (a) case identification restricted to diagnostic codes included in national care monitoring programs. While for our main analysis we intentionally expanded the diagnostic codes from the national audit programs list with additional codes indicating a heart failure diagnosis, so as to ensure completeness; sensitivity analyses, restricting diagnostic codes to those used in the national audit programs, found that 97% of patients in our cohort had a record heart failure used in the national clinical audit programs, and led to no significant changes in the present results. (b) case identification restricted to diagnoses recorded in secondary care, or referred for specialist assessment or echocardiography. We further found that 92% of patients included in our cohort had a heart failure diagnosis recorded in secondary care, or either a referral to specialist cardiology service or echocardiography. While that proportion moderately increased over time, we found no significant change by sex or socio-economic status.

eAppendix 4: Accuracy of hospital episodes data
Accuracy of diagnostic coding in routinely collected hospital records in the United Kingdom has been widely studied and findings show that data are sufficiently robust to be used in healthcare research and decision-making. Specifically, a recent systematic review identified 32 studies were that compared routinely collected data with case or operation notes. Although accuracy presented significant variation between studies, their findings show that since the 2002 introduction of the 'Payment by Results' program, accuracy has improved, and for primary discharge diagnoses accuracy was 96.0%. 12

eAppendix 5: Validity of causes of death records
The present research is reliant on the accuracy of death certificates. The validity of death records underlying our study has therefore been carefully assessed. In England and Wales, information collected at death registration is normally supplied by the informant (usually a close relative of the deceased), underdoes automatic validation checks and is verified by the registrar. The cause of death is usually obtained from the 'Medical Certificate of Cause of Death', completed by a medical practitioner when the death is certified. The final underlying cause of death takes account of additional information received from medical practitioners or coroners after the death has been registered; around 40% of deaths are referred to the coroner. 13 The Office for National Statistics (ONS) collects information on all deaths that occur in England and Wales as well as deaths of all ordinary residents. 13 The dataset used in this study was restricted to patients whose record was linked to death information from ONS death certificates. Mortality data from the Office for National Statistics is used by academics, demographers and health researchers, as well as major national and international health organizations (including Public Health England, Eurostat, and the United Nations) for disease surveillance and epidemiological research. A recent study performed by the ONS has examined causes of death recorded on death certificates in five pilot areas in the UK. The study found on scrutiny of an independent medical examiner, the broad underlying cause of death (as defined by International Classification of Diseases chapter (ICD)) remained unchanged in 88 per cent of cases. 14 In the present study, the 22 ICD chapters were grouped into higher-level disease categories (respectively 9 and 11 disease categories for death and hospitalizations), so that consistency is likely to be even higher than in the aforementioned ONS study. Moreover, no national coding reform has been introduced over the study period and there is no evidence to suggest recording practices to have changed considerably over time or by age, sex, and socioeconomic sub-groups. 13 In light of this information, we conclude that UK death certificates data may present some level of inaccuracy that must be taken into account in the design and interpretation of studies relying on death registration data; yet that the information is appropriate for the study longterm temporal trends of cause-specific mortality in large populations.

eAppendix 6: Literature review
We searched Pubmed for reports published from 1 January 2012 to 15 February 2019 that included "heart failure" and "mortality" in their title, reviewed references from clinical practice guidelines and consulted with experts for relevant studies. We found numerous studies that investigated mortality following a hospital admission for heart failure, though few studies reported survival rates after incident diagnosis. In an attempt to compare various reports, we selected studies that reported 1-year mortality rates following a new diagnosis of heart failure and (i) referred to European or North American cohorts, (ii) included at least 1,000 patients, (iii) were not restricted to clinical trials, special care management programs, certain age-groups or associated conditions, and (iv) reported trends over time. A few relevant studies were identified (eTable 1). Overall these reported improvements in mortality up to around 2005 but stable rates thereafter, despite increasing uptake of new treatments, in particular beta-blockers. Most studies confined investigations to all-cause mortality, with only one study distinguishing between cardiovascular and non-cardiovascular mortality. No study investigated underlying patterns and cause-specific mortality. eTable 1: Selected studies reporting heart failure mortality rates following incident heart failure.

Causes of death Causes of hospitalization
Cardiovascular disorders: ICD chapter 'Diseases of the circulatory system' (code range: I00-I99), excluding codes relating to infections.
Other cardiovascular disorders: ICD chapter 'Diseases of the circulatory system' (code range: I00-I99), excluding codes relating to heart failure or infections.
Neoplasms: ICD chapter 'Neoplasms' (C00-D48). Other: any code not falling into any of the above categories.

Infections
Other: ICD chapter 'Symptoms, signs and abnormal clinical and laboratory findings, not elsewhere classified' (R00-R99) as well as any code not falling into any of the above categories.