Survival of patients with chronic heart failure in the community: a systematic review and meta‐analysis

Abstract Aim To provide reliable survival estimates for people with chronic heart failure and explain variation in survival by key factors including age at diagnosis, left ventricular ejection fraction, decade of diagnosis, and study setting. Methods and results We searched in relevant databases from inception to August 2018 for non‐interventional studies reporting survival rates for patients with chronic or stable heart failure in any ambulatory setting. Across the 60 included studies, there was survival data for 1.5 million people with heart failure. In our random effects meta‐analyses the pooled survival rates at 1 month, 1, 2, 5 and 10 years were 95.7% (95% confidence interval 94.3–96.9), 86.5% (85.4–87.6), 72.6% (67.0–76.6), 56.7% (54.0–59.4) and 34.9% (24.0–46.8), respectively. The 5‐year survival rates improved between 1970–1979 and 2000–2009 across healthcare settings, from 29.1% (25.5–32.7) to 59.7% (54.7–64.6). Increasing age at diagnosis was significantly associated with a reduced survival time. Mortality was lowest in studies conducted in secondary care, where there were higher reported prescribing rates of key heart failure medications. There was significant heterogeneity among the included studies in terms of heart failure diagnostic criteria, participant co‐morbidities, and treatment rates. Conclusion These results can inform health policy and individual patient advanced care planning. Mortality associated with chronic heart failure remains high despite steady improvements in survival. There remains significant scope to improve prognosis through greater implementation of evidence‐based treatments. Further research exploring the barriers and facilitators to treatment is recommended.


Introduction
One to two in every 100 adults in the general population, and more than one in 10 people aged over 70 years are diagnosed with heart failure (HF). 1,2 The true prevalence is likely closer to 4%, as HF often goes unrecognised or misdiagnosed, particularly in older people. 3,4 Prevalence has risen by almost 25% since 2002 due to factors such as population ageing, improved survival following coronary events and an increase in the prevalence of HF stability. 7 Previous research suggests 1-year survival in acute HF is between 55% and 65%, 8,9 compared to 80% to 90% in chronic HF. 10,11 The majority of patients have chronic HF and are treated in ambulatory settings. 12 This chronic phase should be a time to discuss advanced care planning and anticipated disease progression with patients and their families. These conversations rely on healthcare professionals providing accurate prognostic information, yet survival estimates for chronic HF vary significantly across studies. The pattern of disease progression in HF is also unpredictable and varies considerably between individuals. 13 Uncertainty over disease trajectory is one reason active HF treatment often persists into the terminal phases of illness, resulting in a large increase in resource use in the last 6 months of life. 14 It also explains why some clinicians lack confidence in discussing HF prognosis and so avoid the subject. 15,16 Not all patients wish to know or discuss their prognosis, but for those who do, the ambiguity around their future can be distressing and many would welcome more information. 17 Where patients are not informed of their expected prognosis, they tend to significantly overestimate their likely life expectancy. 18 Reliable prognostic estimates can help to promote advanced care planning, improve shared understanding of treatment goals and facilitate integrated treatment with specialist services, including palliative care. 16 The aim of this systematic review was to assimilate the existing evidence base to provide accurate survival estimates for people with chronic HF. We also aimed to identify key factors which explain the existing variation in prognostic estimates, including age at time of diagnosis, left ventricular ejection fraction (LVEF), decade of diagnosis, and study setting.

Methods
The protocol was published on PROSPERO (registration number CRD42017075680) and in Systematic Reviews. 19 Reporting adheres to the 'Meta-analysis Of Observational Studies in Epidemiology' (MOOSE) guidelines (online supplementary Methods S1). 20

Search strategy
We conducted a systematic search of relevant databases from inception to August 2018, incorporating Medical Subject Heading Indexation (MESH) terms and integrated validated search filters from the Scottish Intercollegiate Guidelines Network 21 (online supplementary Table S1). A hand search of the included papers' references and relevant review articles was completed to achieve literature saturation.

Eligibility criteria
Eligible studies reported survival time for adult patients with a diagnosis of HF in the 'chronic' or 'stable' phase. 7 Survival times were calculated from diagnosis, or from point of study recruitment if this information was unavailable. Studies with under 1-year follow-up were excluded given the lack of information on long-term prognosis. We included studies reporting outcomes for both acute and chronic HF where it was possible to extract survival rates for chronic HF. If the results were combined, we attempted to contact study authors. As our aim was to report survival time in the context of usual care, we . . excluded interventional studies, service evaluations and studies where participants had been recruited on the basis of another co-morbidity. Conference abstracts were excluded as having insufficient detail for quality assessment.

Data analysis
Two authors (N.R.J., I. A.) independently completed two rounds of screening, the first based on titles and abstracts and the second a full text review. Foreign language papers were translated before assessment. Disagreements were checked with a third reviewer (C.J. T.). Two authors (N.R.J., I.A.) also completed independent duplicate data extraction.
Pooled survival rates were calculated at pre-specified time points using a random effects model given the anticipated variability in study methods. We used the metaprop command in Stata 14, designed for meta-analysis of binomial data. 22 We calculated the study-specific 95% confidence intervals using the score statistic via the cimethod(score) function and used the ftt command to perform the Freeman-Turkey double arcsine transformation and stabilise variance in our weighted pooled estimates. 22 Heterogeneity and consistency were assessed using Chi-squared and I 2 statistics respectively. Sources of heterogeneity were explored using pre-specified sensitivity and subgroup analyses.
We conducted subgroup analyses and meta-regression for study date, setting, age and LVEF. To pool study dates, we categorised each included study or relevant subgroup by the decade of participant recruitment. Mean participant age was used to categorise results as either < 65, 65-74 or ≥ 75 years. Study setting was determined by point of recruitment and majority of management. Where there was evidence of significant input across both primary and secondary care, studies were classified as 'cross-discipline'. HF was categorised as HF with preserved ejection fraction (HFpEF) if LVEF ≥ 50%, HF with mid-range ejection fraction (HFmrEF) with LVEF in the range 40-49%, and HF with reduced ejection fraction (HFrEF) if LVEF < 40%. Some earlier studies did not include a mid-range group and so categorised HFpEF as LVEF ≥ 40%. Studies reporting pooled outcomes for all three groups or not measuring LVEF were grouped as 'mixed' ejection fraction. Data were unavailable to allow all subgroups of interest to be included together as covariates in a meta-regression analysis, therefore each covariate was considered separately in meta-regression models of survival rates at 1 and 5 years.
Two authors (N.R.J., I.A.) independently completed a risk of bias assessment for each study using the Quality in Prognosis Studies (QUIPS) tool, recommended by the Cochrane Prognosis Methods Group. 23 We conducted a sensitivity analysis excluding studies at moderate or high risk of bias. We report a Grading of Recommendations Assessment, Development and Evaluation (GRADE) score to provide an estimate of confidence in the cumulative outcomes (online supplementary Methods S2). 24

Study characteristics
We included 60 studies after screening, 5423 studies at the title and abstract stage and 97 full texts (online supplementary Figure S1). A number of studies reported survival rates from the same dataset. Where these provided relevant information for our pre-specified subgroup analyses, we included both studies in the review but only one in any single meta-analysis. Two studies met the inclusion criteria but reported survival rates at time points which could not be pooled; these are reported narratively. 16,25 The majority of included studies were conducted in Europe or North America and recruited participants from primary care (n = 23), cardiology outpatient clinics (n = 20), or both (n = 15). Over half were longitudinal cohort studies (n = 34) but many recent studies have analysed big databases of routinely collected patient information. 9 HF diagnosis was most frequently captured using validated database codes (n = 19), though many studies also defined HF using Framingham (n = 12), or European Society of Cardiology (n = 10) criteria (Table 1). 1,10,11, In eight studies the criteria for defining HF was unspecified or relied on a clinical diagnosis. There were insufficient data to conduct a meaningful analysis comparing outcomes by sex.
Demographic and baseline participant characteristics differed significantly between studies (online supplementary Table S2). Reporting of this information was inconsistent with ethnicity and deprivation indices only rarely included. However, co-morbid cardiovascular disease was common, with hypertension the most frequent co-morbidity, followed by diabetes and ischaemic heart disease. Treatment rates of key HF medications including angiotensin-converting enzyme inhibitors/angiotensin receptor blockers, beta-blockers and mineralocorticoid receptor antagonists improved over time. Some recent studies reported treatment rates close to 90%. Detailed prescribing information was lacking, meaning it was not possible to determine how many participants were treated with optimum dosage or the recommended combination of all three agents.

Sensitivity analysis
The majority of studies were rated at low (n = 26) or moderate (n = 27) overall risk of bias (online supplementary Table S3).
Excluding the studies at moderate or high risk of bias in a sensitivity analysis did not alter the results. The pooled survival rate at 1 year across the remaining studies was 85.9% (84.1-87.7) and at 5 years 56.9% (52.1-61.7). The GRADE assessment suggests there is 'high' certainty in the summary findings (online supplementary Table S4).

Subgroup analysis by age
Evidence from the forest plots and meta-regression suggests survival rates decreased with increasing age at diagnosis (1- Table S5).
The trend towards a worse prognosis in relation to age at diagnosis was also reported within individual studies. 11,41 In a recent analysis of survival rates within the UK THIN database, 5-year survival rates were 50% amongst participants aged 75-84 years, compared to 81% amongst the youngest participants aged 45-54 years. 11 In both cases, survival rates were significantly worse than for age-matched participants of 72% and 98%, respectively. 11

Subgroup analysis by study setting
The pooled 1and 5-year survival rates were significantly better for participants in secondary care studies compared to cross-discipline studies ( Figure 3). There was some evidence of improved survival in secondary care studies compared to primary care, with around 5% more participants alive at 1 year and 10% more at 5 years. The association between survival and setting was confirmed by meta-regression (online supplementary Table S5). Individual secondary care studies with the poorest survival rates were those that purposively recruited either elderly frail participants, or those with a significant reduction in LVEF. 30,31 The primary care studies reporting the best survival rates used screening to detect incident HF cases. 48,63 Rates of key HF medication prescribing were consistently better in secondary care.
The four studies 36,45,48,52 conducted in South-East Asia reported better survival rates compared to Europe and North America, despite recruiting participants of comparable age and co-morbid disease burden. One of these studies 48 used screening to detect incident cases and the proportion of participants prescribed HF medication was also relatively high, which may explain this survival difference.

Subgroup analysis by left ventricular ejection fraction
The pooled survival rate at 5 years was better for patients with HFrEF than mixed ejection fraction ( Figure 4). There was no significant difference in the pooled survival rates for HFpEF compared to HFrEF at either 1 or 5 years (online supplementary Table S5). A number of studies compared the risk of death by LVEF in their individual populations and found a preserved ejection fraction was associated with improved survival. Survival analysis from a community-based screened cohort found patients with a LVEF <40% compared to LVEF >50% had a 1.80 (1.55-2.10) times greater risk of death over the study period, when adjusted for key factors such as age and sex. 60 Other studies found the risk of death to be even greater for those with HFrEF, with hazard ratio of 2.62 (1. 45-4.75), 50 and 3.72 (1. 80-7.68) reported. 39 In every study reporting cause of death data categorised by LVEF, the proportion of total mortality attributed to cardiovascular disease     and HF-related mortality was greater for people with HFrEF than HFpEF ( Table 2).  29.5% (28.9-30.2). 11 A number of studies have also demonstrated improving survival rates over time within their individual population. Framingham data show an improvement in 5-year survival between 1950-1969 to 1990-1999 from 30% to 41% for men and from 43% to 55% for women. 33 This trend is also seen in the Rochester Epidemiology Project. 40 Recently, there have been more modest improvements in survival. A database study of over 400 000 people with HF in Ontario, found 1-year mortality fell amongst outpatients with HF from 17.7% in 1997 to 16.2% in 2007. 59 A study of 600 000 Medicare patients with incident HF reported a reduction in mortality from 67.5% to 64.9% for men and from 61.7% to 60.2% for women between 1994 and 2003. 49

38.2
Only studies reporting cause of mortality included. Blank cells indicate data were not reported in the original study. All figures refer to proportion of total mortality within the study. Selected subgroups of both cardiovascular and non-cardiovascular mortality were reported in some studies, meaning in some cases the sum of the subgroup results are not equal to the combined mortality result.
± HF cases recorded as either 'definite' or 'probable'. In Taylor

Discussion
This is the first systematic review of prognosis in chronic HF and provides contemporary survival estimates applicable across high income countries. The analyses draw on survival data from 1.5 million people with chronic HF across 60 studies. Survival rates have improved over time and 20% more people survive at both 1and 5-year follow-up today compared to between 1950 and 1969. Survival rates improved sharply from the 1970s to 1990s, but there has been only a modest reduction in mortality in the past two decades. Increasing age at diagnosis is one key factor associated with a poor prognosis. Survival rates amongst people aged ≤ 65 years were almost 10% better at 1 year and over .
30% better at 5 years, when compared to people aged ≥ 75 years. Survival rates were higher in studies recruiting participants from cardiology outpatient settings compared to cross-discipline or primary care.
There was no significant difference in survival between HFrEF and HFpEF in our pooled analysis, though individual studies reported improved survival rates and lower rates of hospital admission and cardiovascular mortality for people with HFpEF. Both survival rates and prescribing of HF medication were significantly lower for patients where LVEF was not reported or analysed. This may be due to older trials with worse survival rates not reporting LVEF. It may also reflect certain populations, such as nursing home residents or older patients, are less likely to have LVEF measured despite having a worse prognosis. Nevertheless, recognising that patients who are not categorised by LVEF have a poorer outlook may have important implications for future assessment and treatment pathways.
The search strategy and eligibility criteria were designed to be inclusive, drawing studies from a wide range of geographical and healthcare settings. Source data from developing countries were less abundant but landmark cross-continental studies provide data for these healthcare settings. Internationally, the lowest mortality rates were in South-East Asian studies.

Limitations
The diversity in study design and setting captured by the inclusive search strategy resulted in high levels of heterogeneity in each individual meta-analysis. This 82 Not all studies reported HF survival from time of diagnosis. Whilst primary care studies generally used routinely collected data sources to identify a first coded episode of HF, secondary care studies tended to calculate survival from first clinic visit, which may have been several years after diagnosis. Studies were categorised by setting to account for this potential time lag, though this was not apparent in our results. In practice, most patients with a confirmed diagnosis of HF will have input at some point from a cardiologist, except for some very frail patients who may be limited by cognitive or mobility issues. It is possible the differences seen in survival between settings reflect such variation in participant characteristics, though secondary care studies also reported higher rates of prescribing for key HF treatment. We plan to report more detail on prescribing rates in a separate paper. The definitions    of cardiovascular and non-cardiovascular death varied between studies as did the categories used in the cause of death subgroup analyses, making it difficult to compare these outcomes directly.
Outcome data are pooled from across a wide time period to capture changing survival rates over time. However, survival rates may not be directly comparable across these studies given there have been significant changes in HF management in the past 70 years, including the introduction of medications proven to improve prognosis for people with HFrEF. The statistical heterogeneity also reflects the large sample sizes of the included studies, which resulted in narrow confidence intervals. Even small differences in survival rates resulted in non-overlapping confidence intervals and .
high I 2 scores, a recognised limitation of this statistical measure in observational meta-analysis.
The review included observational studies to present the real-world outcomes for people with HF, outside of trial settings. Confounding is a recognised problem in these non-randomised trials and reporting of important covariates was inconsistent. Missing data were a particular problem in earlier studies and those drawing on data from large primary care databases. Some meta-regression results rely on data from a small number of studies, such as for HFmrEF and general secondary care clinics. However, similar results were observed when these small subgroups were combined with adjacent categories. Few studies reported echocardiogram .
( -, -  findings or categorisation of HF by LVEF, despite the prognostic significance of this information. Accurate coding of HF is also a recognised limitation in routinely collected datasets. 83,84 However, this approach to epidemiological research is still felt to be valid and coding has been improving in line with performance payments and better access to diagnostic tests in primary care. 85

Comparison with existing literature
A recent European secondary care study reported 1-year mortality rates for people with acute and chronic HF of 23% and 6%, respectively, compared to 3% for matched controls. 69 In our pooled .
analysis, 1-year mortality in chronic HF was above 10%. This may be because some people with a very poor prognosis are never admitted to hospital or referred to secondary care. Categorisation of HF has changed over time to recognise the importance of LVEF when considering treatment options and prognosis. Survival rates are better for people with HFpEF compared to HFrEF, once adjusted for key covariates including age, sex, and aetiology of HF. 86 However, people with HFpEF are more likely to be older and have significant co-morbid disease, meaning the unadjusted HFrEF and HFpEF survival rates are similar. This may explain why there was no significant difference in survival in our subgroup analysis based on LVEF.

Research implications
Our results provide a reference source for clinicians, patients and policy makers, to inform population prognostic estimates. The subgroup analyses help to provide adjusted survival estimates based on key variables, such as age at time of diagnosis. Further work is needed to refine prognostic models for individuals with chronic HF. Existing tools, such as the Seattle Heart Failure Model and MAGGIC HF risk tool, lack specificity and sensitivity data that are applicable to clinical practice. 86,87 Reducing uncertainty and confusion about the outcomes in HF could lead to improvements in advanced care planning, treatment adherence and integration with wider healthcare teams such as palliative care. 16,88 Survival rates in HF remain poor despite modest improvements over time. Investment in healthcare infrastructure and public health initiatives for conditions with similar outcomes such as cancer and stroke have seen improvements in morbidity and mortality. 89,90 This review suggests that targeted allocation of resources towards improving early diagnosis, prescribing and treatment adherence and multi-disciplinary models of care may lead to further reductions in mortality for people with HF.

Conclusion
There have been modest improvements in survival rates for people with chronic HF over the past 70 years. Despite this, the 5-year survival rate is close to 50% and many people will die directly from HF or from related cardiovascular disease. Older populations are at the greatest risk of death, presenting a looming challenge to healthcare systems given changing global demographics. Our results draw from very heterogeneous data sources and when applying survival estimates to any individual, consideration should be given to factors such as their age, co-morbid disease, treatment, and LVEF. Further research is needed to develop the evidence base around key prognostic indicators for patients with chronic HF that will enable population estimates to be refined for individuals. Greater understanding and awareness of chronic HF survival rates can facilitate better multi-disciplinary team working and inform advanced care planning between patients and healthcare professionals.

Supplementary Information
Additional supporting information may be found online in the Supporting Information section at the end of the article. Methods S1. MOOSE (Meta-analyses Of Observational Studies in Epidemiology) checklist. Methods S2. Risk of bias and quality assessment. Table S1. Search strategy. Table S2. Prevalence of co-morbid disease, cardiovascular risk factors and heart failure medication across studies .  Table S3. Risk of bias assessment using the Quality in Prognosis Studies tool. Table S4. GRADE risk of bias assessment across studies. Table S5. Subgroup and meta-regression analyses by age at diagnosis, setting, left ventricular ejection fraction, and date. .
. Figure S1. Survival of people with heart failure at 1 month. Figure S2. Survival of people with heart failure at 1 year. Figure S3. Survival of people with heart failure at 2 years. Figure S4. Survival of people with heart failure at 5 years. Figure S5. Survival of people with heart failure at 10 years. Figure S6. PRISMA flow diagram of study selection.