Systematic review and individual patient data meta-analysis of diagnosis of heart failure, with modelling of implications of different diagnostic strategies in primary care.

Jonathan Mant, J Doust, Andrea Roalfe, Pelham Barton, MR Cowie, P Glasziou, D Mant, Richard McManus, Roger Holder, Jonathan Deeks, Kate Fletcher, M Qume, S Sohanpal, S Sanders, Frederick Hobbs

Research output: Contribution to journalReview article

162 Citations (Scopus)


OBJECTIVES: To assess the accuracy in diagnosing heart failure of clinical features and potential primary care investigations, and to perform a decision analysis to test the impact of plausible diagnostic strategies on costs and diagnostic yield in the UK health-care setting. DATA SOURCES: MEDLINE and CINAHL were searched from inception to 7 July 2006. 'Grey literature' databases and conference proceedings were searched and authors of relevant studies contacted for data that could not be extracted from the published papers. REVIEW METHODS: A systematic review of the clinical evidence was carried out according to standard methods. Individual patient data (IPD) analysis was performed on nine studies, and a logistic regression model to predict heart failure was developed on one of the data sets and validated on the other data sets. Cost-effectiveness modelling was based on a decision tree that compared different plausible investigation strategies. RESULTS: Dyspnoea was the only symptom or sign with high sensitivity (89%), but it had poor specificity (51%). Clinical features with relatively high specificity included history of myocardial infarction (89%), orthopnoea (89%), oedema (72%), elevated jugular venous pressure (70%), cardiomegaly (85%), added heart sounds (99%), lung crepitations (81%) and hepatomegaly (97%). However, the sensitivity of these features was low, ranging from 11% (added heart sounds) to 53% (oedema). Electrocardiography (ECG), B-type natriuretic peptides (BNP) and N-terminal pro-B-type natriuretic peptides (NT-proBNP) all had high sensitivities (89%, 93% and 93% respectively). Chest X-ray was moderately specific (76-83%) but insensitive (67-68%). BNP was more accurate than ECG, with a relative diagnostic odds ratio of ECG/BNP of 0.32 (95% CI 0.12-0.87). There was no difference between the diagnostic accuracy of BNP and NT-proBNP. A model based upon simple clinical features and BNP derived from one data set was found to have good validity when applied to other data sets. A model substituting ECG for BNP was less predictive. From this a simple clinical rule was developed: in a patient presenting with symptoms such as breathlessness in whom heart failure is suspected, refer directly to echocardiography if the patient has a history of myocardial infarction or basal crepitations or is a male with ankle oedema; otherwise, carry out a BNP test and refer for echocardiography depending on the results of the test. On the basis of the cost-effectiveness analysis carried out, such a decision rule is likely to be considered cost-effective to the NHS in terms of cost per additional case detected. The cost-effectiveness analysis further suggested that, if likely benefit to the patient in terms of improved life expectancy is taken into account, the optimum strategy would be to refer all patients with symptoms suggestive of heart failure directly for echocardiography. CONCLUSIONS: The analysis suggests the need for important changes to the NICE recommendations. First, BNP (or NT-proBNP) should be recommended over ECG and, second, some patients should be referred straight for echocardiography without undergoing any preliminary investigation. Future work should include evaluation of the clinical rule described above in clinical practice.
Original languageEnglish
Pages (from-to)1-207, iii
JournalHealth Technology Assessment
Issue number32
Publication statusPublished - 1 Jul 2009


Dive into the research topics of 'Systematic review and individual patient data meta-analysis of diagnosis of heart failure, with modelling of implications of different diagnostic strategies in primary care.'. Together they form a unique fingerprint.

Cite this