Family Practice Advance Access originally published online on September 7, 2007
Family Practice 2007 24(6):628-635; doi:10.1093/fampra/cmm055
Diagnosis of left ventricular systolic dysfunction (LVSD): development and validation of a clinical prediction rule in primary care
a Division of Community Health Sciences, University of Dundee, Mackenzie Building, Dundee DD2 4BF
b Department of General Practice, Royal College of Surgeons in Ireland, 120 St Stephens Green, Stephen Street Lower, Dublin 2
c Division of Medicine and Therapeutics, University of Dundee, Ninewells Hospital, Dundee DD1 9SY, UK
Correspondence to Tom Fahey, Division of Community Health Sciences, University of Dundee, Mackenzie Building, Dundee DD2 4BF, UK; Email: tomfahey{at}rcsi.ie
Received 2 February 2007; Revised 27 July 2007; Accepted 30 July 2007.
 |
Abstract |
|---|
 Top Abstract IntroductionMethodsResultsDiscussionDeclarationReferences |
|---|
Background: Diagnosing suspected left ventricular systolic dysfunction (LVSD) in the community is a challenge for GPs. We developed and validated a clinical prediction rule (CPR) for LVSD based on history, examination and electrocardiogram (ECG).
Methods: Prospective cohort studies of 458 symptomatic patients (derivation cohort) and 535 patients (validation cohort) in 26 general practices in Tayside and Fife, Scotland. All patients underwent a structured clinical examination and ECG and the ‘reference standard’ investigation of echocardiography to establish the presence of LVSD.
Results: Four elements from the clinical history and examination were all independently associated with LVSD—male sex [adjusted odds ratio (OR) 2.5; 95% CI 1.1, 5.0], presence of orthopnoea (OR 5.4; 1.9, 13.8) history of myocardial infarction (OR 5.6; 2.3, 13.6) and elevated jugular venous pulsations (OR 15.1; 4.6, 49.3). Addition of ECG (OR 20.6; 2.7, 158.6) provides important diagnostic information in terms of probability of LVSD. A CPR based on the presence or absence of these five elements will generate probabilities ranging from 1% to 97% for LVSD when applied to an individual patient. In the validation cohort, the model under-predicted the probability of LVSD, particularly at lower levels of expected risk, reflecting differences in the risk-factor profiles of the derivation and validation cohorts.
Conclusions: The derived CPR provides quantitative estimates of post-test probability for LVSD. This rule requires further validation in other populations and settings because of the difficulties encountered in the validation cohort.
|
Introduction |
|---|
|
TopAbstractIntroductionMethodsResultsDiscussionDeclarationReferences |
|---|
Heart failure due to left ventricular systolic dysfunction (LVSD) is a common, costly, disabling and life-threatening condition. Around 120 000 hospital admissions per year are due to LVSD in the UK.1 The survival rate of LVSD from the time of diagnosis is 67% at 1 year and 24% at 3 years.2 As LVSD often carries a worse prognosis than several of the common cancers, it is important that patients who present to their family physician with symptoms suggestive of LVSD are identified and offered appropriate treatment.2
Identification of patients with LVSD is a challenge for family physicians. Around half of all patients suspected of having LVSD when assessed in primary care do not have this diagnosis confirmed on echocardiography.3–5 Clinical guidelines suggest that no individual symptom, sign or investigation can diagnose LVSD in a primary care patient. Though guidelines recommend performing a careful history and examination prior to further diagnostic testing, quantification concerning the predictive value of elements from history and examination and their impact on the probability of LVSD has been made in hospital-based settings but not in community-based settings.6–10
Echocardiography is the key investigation in the diagnosis of LVSD, but access is limited. Only 37% of GPs in the UK have direct access to echocardiography and there are delays between referral and final diagnosis.7 There is a need to develop a clinical prediction rule (CPR) that is based on clinical findings and diagnostic tests that are accessible to GPs. Implementation of such a rule would help GPs prioritize patients and improve onward referral of patients to limited echocardiography services.
For these reasons, we investigated the diagnostic utility of the clinical history, examination and electrocardiogram (ECG) in patients presenting to their GP with symptoms suggestive of LVSD. Our second objective was to develop and validate a CPR to quantify the probability of LVSD from these elements of clinical history, examination and ECG for use in individual patients presenting in primary care.
|
Methods |
|---|
|
TopAbstractIntroductionMethodsResultsDiscussionDeclarationReferences |
|---|
Practices and participants
Twenty-six GP surgeries from the Dundee, Tayside and North East Fife areas of Scotland were asked to participate. Assessment of patients occurred over 15 months from June 2002 to September 2003 at one of seven community settings. Consecutive patients over the age of 16 years who were suspected of having LVSD by their GP and had not been diagnosed previously with an echocardiogram as having LVSD were recruited into the study. Informed written consent was obtained for all patients. The Tayside research ethics committee and North East Fife research committee approved the study.
Data collection
A structured history and examination was performed and documented by SJ for each patient in the community. The collected variables are listed in Table 1. All patients had their clinical evaluation, electrocardiography (ECG) and echocardiography performed by SJ. Echocardiography was undertaken using a Siemens Cypress® system with a 2.5-MHz probe. Echocardiography was performed in the standard way, with the patients placed in the left lateral decubitus position and 2D, M Mode and colour Doppler techniques used. A fractional shortening, using M Mode echocardiography, of less than 25% or ejection fraction, using Simpson’s method, of less than 50% was used as the cut-off for the presence of LVSD. Visual assessment of left ventricular systolic function was used when fractional shortening or ejection fraction measurements where not possible. The ECGs were reported at the end of examination. SJ used the E-lite 6.34® system for 12 lead electrocardiography and interpreted the ECG immediately in the community. The ECG was considered normal if the reading showed sinus rhythm 60–100/min with an axis –30° to +90°, with the PR interval <0.2 s, QRS duration <0.12 s, QTc interval <0.42 s. The morphology of the P wave, QRS complex, ST segment and T wave had to be normal. Any ECG having any feature different from the stated definition was considered abnormal. A random sample of 91 ECGs were read by a GP (TF) blind to the symptoms, signs and echocardiography results of the patients as well as blind to the cardiology registrar's ECG interpretation.
|
Data analysis
Derivation. The first stage of the analysis was to investigate the univariable associations for the explanatory variables—clinical history, examination and ECG findings with the outcome of LVSD on echocardiography. These results are expressed as odd ratios (ORs) where values greater than 1 indicates increased odds of the presence of LVSD and values of less than 1 indicates decreased odds of the presence of LVSD. The explanatory variables were considered initially in three separate groups—clinical history, examination and ECG findings, with a final model constructed for clinical history and examination initially and then for clinical history, examination and ECG findings. For inclusion into the multivariable logistic regression model, explanatory variables had threshold P-values of
0.1. Multivariable logistic regression was used to identify the explanatory factors independently associated with LVSD for the two models, clinical history and examination alone and clinical history, examination and ECG finding. Age was fitted as a continuous variable in all regression models. The final models include explanatory variables with a P-value of 0.05. Final models were checked for interactions between explanatory variables. Validation. To assess whether the predicted probabilities agree with the observed probabilities for LVSD, we calibrated the model by applying the regression coefficients from the derivation cohort to the individuals in the validation cohort, generating expected and observed probabilities of LVSD. Quintiles of risk categories of expected and observed LVSD were generated and goodness-of-fit assessed by the Hosmer–Lemeshow test.11 All data analysis was carried out using Stata version 8 (Stata Corporation 2003).
Sample size. Assuming a positive predictive value of 22% and a prevalence of myocardial infarction (MI) of between 20% and 25% based on a previous primary care study,12 the sample size required to detect margins of error between 3%–5%, at a two-sided significance level of 0.05 ranging between 657 and 237 patients.
Derivation of the CPR. Estimations of the contributions of clinical history, examination and ECG on the post-test probability of LVSD were calculated. Receiver operating characteristic curves were generated for the final regression models with the respective area under the curves (AUCs) illustrating the overall diagnostic performance of the respective final regression models (clinical history and examination alone or clinical history, examination and ECG findings). We derived the post-test probability of LVSD using the equation P = 1/(1 + ey), where P is the post-test probability and y is the sum of the constant coefficient and the regression coefficients from the explanatory variables in the regression model.
Validation of CPR. A separate cohort of 536 patients was assessed by three nurses as part of a separate study on open access echocardiography between January 2004 and December 2005. The nurses were trained in taking a focussed history and clinical examination. Echocardiography using the same criteria as in the derivation cohort was carried out and reported by a radiographer trained in echocardiographic assessment of LVSD. The radiographer was blind to the clinical features of the patients. A GP (TF) assessed ECG readings in 503 of these patients, blind to the patient's clinical status. Twenty-three patients in the validation cohort did not consent to ECG recording and were excluded. ECG criteria were the same for both the derivation and validation cohorts.
|
Results |
|---|
|
TopAbstractIntroductionMethodsResultsDiscussionDeclarationReferences |
|---|
Patient characteristics
Four hundred and forty-eight symptomatic patients suspected of having LVSD were underwent clinical assessment and echocardiography. The characteristics of the 458 patients are shown in Table 1. Sixty percent of patients referred were female and the majority were over 65 years of age. As recruitment was of symptomatic patients, all patients complained of dyspnoea, lethargy or ankle swelling. LVSD was diagnosed by echocardiography in 37 (8%) of referred patients. ECG recording was abnormal in 223 (54%) of the 412 patients who had this investigation. Agreement between the cardiology registrar and a GP blind to patient's symptoms and signs was very good (kappa 0.85) in 91 randomly selected individuals.
Derivation—univariable and multivariable associations
On univariable analysis, patients who were male, complained of orthopnoea, paroxysmal nocturnal dyspnoea or wheeze, who had a history of angina, MI, coronary revascularization, previous echocardiogram or reported having an elevated cholesterol level, were taking diuretics, spironolactone, digoxin, aspirin or a statin, had a displaced apex beat, a heart murmur, an elevated jugular venous pulsation (JVP) and an abnormal ECG were more likely to have LVSD (Table 1). Conversely, patients with known hypertension, who were overweight or who had a systolic blood pressure 
150 mmHg were less likely to have LVSD. These explanatory variables were fitted into regression models in sequence of clinical history, examination and ECG finding.
After adjustment for age, the factors independently associated with LVSD were male sex, orthopnoea, history of MI and raised JVP (Table 2). When abnormal ECG was added to the model, male gender was not significantly associated with LVSD at the 5% level (Table 2). Tests for interactions between all explanatory variables in the final model were not significant. The diagnostic performance of the model remained high irrespective of which of the four clinical variables—male sex, orthopnoea, history of MI or rasied JVP was excluded, with the AUC varying between 0.75 and 0.83 for clinical variables alone and 0.86 and 0.88 for clinical variables plus ECG (Table 3).
|
|
CPR
Figure 1 shows post-test probabilities for the presence of LVSD using different combinations associated symptoms and signs. The range of post-test probabilities based on symptoms or signs is substantial, varying between 1% and 89%. The effect on post-test probabilities of a positive or negative ECG is also displayed in Figure 1. The impact of ECG was greatest in the intermediate probability range, particularly with a positive ECG result.
|
Validation
Of the 503 patients who underwent complete assessment and who consented to ECG reading, LVSD was diagnosed by echocardiography in 73 (14.5%). The risk-factor profile in the validation cohort was lower, with fewer men 195 (37.4%), fewer numbers of patients who had suffered an MI 80 (15.9%) and ECG abnormalities 168 (33.4%) than in the derivation cohort (Table 1). In contrast, a larger number of validation patients reported orthopnoea as a symptom 121 (24%). Nobody in the validation cohort was identified as having a raised JVP. For this reason, we generated expected quintiles of risk of LVSD in the validation cohort using the following regression function, age x 0.0076126 + male gender x 0.5896059 + 1.753037 x presence of orthopnoea + 1.162851 x past history of MI + 3.331099 x abnormal ECG – 6.642794, and omitting raised JVP.
The observed and expected quintiles of risk in the validation cohort are summarized in Figure 2. Observed risk of LVSD being near to predicted quintile of risk only in the highest quintile risk category, with evidence of poor goodness-of-fit between the observed and expected quintiles (P < 0.001).
|
|
Discussion |
|---|
|
TopAbstractIntroductionMethodsResultsDiscussionDeclarationReferences |
|---|
Summary of main findings
The derivation study shows that elements of the clinical history and examination—male sex, presence of orthopnoea, history of MI and elevated jugular venous pulsation can be used together to quantify the probability of LVSD in an individual patient (Table 2 and Figure 1). This finding is consistent with evidence that diagnostic testing has greatest value when prior probability of disease lies within an intermediate range.13,14
Implications for clinical practice
Previous studies and guidelines suggest that a normal ECG result effectively ‘rules out’ a diagnosis of LVSD.6–8,15,16 The derivation study is congruent with these findings. ECG has a high sensitivity (97% in this study), making a diagnosis of LVSD highly unlikely when the ECG is normal. Each item from the clinical history and examination represents a diagnostic test, modifying the pre-test probability of LVSD.14 Therefore, in the low range of probabilities, a normal ECG adds little additional diagnostic information, while in the intermediate range (two or three symptoms and signs), the diagnostic value of a normal ECG reading is more substantial. These findings emphasize the importance of making an assessment of the overall probability of LVSD based on the whole history and examination.17 Our findings provide quantitative estimates of post-test probability, when combined with a non-invasive diagnostic test (ECG) may enable GPs to base their decision to refer for echocardiography on explicit, quantifiable grounds.
Comparison with existing literature
A study of clinical judgement analysis has shown that the diagnosis of mild LVSD is most difficult for GPs when case vignettes present symptoms and signs in such as way that probability of LVSD is intermediate.18 The prevalence of clinical findings in the derivation study is broadly consistent with previous reports. For example, past history of MI appears to occur in about 10–20% of patients presenting to their GP with symptoms of LVSD,12,15 while the symptom of orthopnoea occurs in 10–24% of patients.4,12,15 The findings are also consistent with previous studies that have evaluated the predictive value of different individual items from clinical history, examination and ECG. For example, past history of MI has found to be independently associated with LVSD in previous cross-sectional and referral studies,12,19,20 while an abnormal ECG is also independently associated with LVSD.12,20,21 Hospital-based studies have shown that raised JVP and abnormal ECG (particularly q waves or left bundle branch block) are associated with LVSD.22 In hospital-based studies, other signs and investigations significantly associated with LVSD such as abnormal apex beat and radiographic redistribution on chest X-ray have limited value in a primary care setting either because they occur too infrequently in patients (Table 1) or because they are diagnostic tests that GPs have limited access to.
Validation study
The CPR did not perform particularly well in the validation cohort. The research nurses, though trained in taking a history and examination, did not identify any patient with a raised JVP, despite a higher prevalence of LVSD in the validation cohort. The most likely explanation is that this difficult clinical sign was missed. It is interesting to note that the one explanatory variable that is not dependent on subjective examiner assessment, male gender, produced similar associations in the derivation and validation cohorts. We feel that this CPR requires further validation in community-based populations.
Strengths and limitations of this study
The findings from the CPR are important for several reasons. Most patients who present to GPs will have one or two symptoms or signs suggestive of LVSD. In this cohort, 255 (56%) patients had either one or two symptoms or signs, with 32 (87%) of cases of LVSD occurring in this group. When trying to detect LVSD, GPs deal with probabilities in a low to intermediate range (2–35%). Contrary to current guideline recommendations,6–9 we have found that the diagnostic value of individual symptoms or signs when combined appropriately may be able to provide important quantitative information that differs in magnitude in terms of the probability of LVSD. Furthermore, different combinations of symptoms and signs produce quite different probabilities of LVSD (Figure 1). Standard 11 of the National Service Framework for Coronary Heart Disease recommends that ‘Doctors should arrange for people with suspected LVSD to be offered appropriate investigations (ECG or echocardiography) that will confirm or refute the diagnosis’.23 Use of a CPR would be valuable to GPs when considering onward referral for echocardiography.
There are several shortcomings to the derivation study that may account for the relatively poor validation. A cardiology registrar took the history and examination and also interpreted the ECG rather than a GP. This was due to limited availability of research funding but has produced a potential incorporation bias in the derivation study, and may be a further reason why the rule did not perform well in the validation study. To minimize this bias, a GP independently assessed a random sample of ECGs in the derivation study and we found that agreement between registrar and GP was very high. Of the other variables independently associated with LVSD in the derivation cohort, raised JVP is the only element that is dependent on prior experience and clinical background.22,24 The diagnostic value of the three readily available elements of the CPR—male sex, past history of MI and presence of dyspnoea had an AUC of 0.81 compared with an AUC of 0.83 when raised JVP was added to the regression model (Table 3). The examiner also performed the reference standard assessment, portable echocardiography, unblinded to the findings of physical examination.25 This was because of limited availability of research funds. Independent hospital validation of portable echocardiography was undertaken in 136 (30%) of the cohort with a misclassification of 5 (4%) of patients overall with four additional patients classified as having and one as not having LVSD. Our findings may be an underestimate of LVSD in referred primary care patients, but it is not known whether portable echocardiography by a cardiology registrar or hospital-based echocardiography by a technician is the reference standard assessment for LVSD. The validation study used nurses to assess patients. We believe further validation is warranted with other clinicians such as GPs in community settings. Lastly, the CPR study is based on 4 or 5 independent/explanatory variables with 37 patients experiencing the outcome of interest. Ideally, there should be 10 outcome events (cases of LVSD) for each independent/explanatory variable in a final logistic regression model when deriving a CPR, so the derivation study would have benefited from a larger sample size.26
Implications for future research and clinical practice
Further studies are needed to validate the CPR in other populations with GPs performing the clinical evaluation and ECG interpretation.27 The role of other diagnostic tests such as natriuretic peptides have shown promising results in a primary care setting.12,20,28,29 Future studies should evaluate near patient natriuretic peptide testing in terms of the additional diagnostic information it provides, after clinical examination and ECG readings have produced a revised post-test probability of LVSD, particularly in the context of ruling out LVSD.30 Otherwise, uncritical application of additional diagnostic test information without appropriate revision of post-test probability based on clinical examination and ECG finding may lead to misleading results.14 Lastly, further studies are required to determine the probability threshold for LVSD above which referral for echocardiography is both cost effective and maximizes patient's survival and quality of life. Future studies should also examine the detection of diastolic dysfunction and the value of using BNP or NT-proBNP as a further diagnostic test.
Conclusions
We have found that a simple four to five point CPR, based on clinical history, examination and ECG findings may enable GPs to quantify the probability of LVSD in a primary care setting.24 Once fully validated, this CPR could help target overstretched echocardiography resources to the most deserving patients.
|
Declaration |
|---|
|
TopAbstractIntroductionMethodsResultsDiscussionDeclarationReferences |
|---|
Funding: British Heart Foundation.
Ethical approval: None.
Conflicts of interest: None.
|
Acknowledgments |
|---|
We would like to thank all participating family doctors and patients.
|
Notes |
|---|
Fahey T, Jeyaseelan S, McCowan C, Carr E, Goudie BM, Pringle SD, Donnan PT, Sullivan FM and Struthers AD. Diagnosis of left ventricular systolic dysfunction (LVSD): development and validation of a clinical prediction rule in primary care. Family Practice 2007; 24: 628–635.
|
References |
|---|
|
TopAbstractIntroductionMethodsResultsDiscussionDeclarationReferences |
|---|
1 Mair FS, Crowley TS, Bundred PE. Prevalence, aetiology and management of heart failure in general practice. Br J Gen Pract (1996) 46:77–79.[Web of Science][Medline]
2 Hobbs FDR. Unmet need for diagnosis of heart failure: the view from primary care. Heart (2002) 88:ii9–ii11.
3 Wheeldon NM, MacDonald TM, Flucker CJ, McKendrick AD, McDevitt DG, Struthers AD. Echocardiography in chronic heart failure in the community. QJM (1993) 86:17–23.
4 Sparrow N, Adlam D, Cowley A, Hampton JR. The diagnosis of heart failure in general practice: implications for the UK National Service Framework. Eur J Heart Fail (2003) 5:349–354.[CrossRef][Web of Science][Medline]
5 Remes J, Miettinen H, Reunanen A, Pyorala K. Validity of clinical diagnosis of heart failure in primary health care. Eur Heart J (1991) 12:315–321.
6 Prodigy. Chronic Heart Failure (2004) http://www.cks.library.nhs.uk/heart_failure (last accessed on August 2007).
7 Scottish Intercollegiate Guidelines Network. Diagnosis and Treatment of Heart Failure due to Left Ventricular Systolic Dysfunction (1999) Edinburgh: Royal College of Physicians in Edinburgh.
8 Task force for the diagnosis and treatement of chronic heart failure ESoC. Guidelines for the diagnosis and treatment of chronic heart failure. Eur Heart J (2001) 22:1527–1560.
9 Cowie MR. B type natriuretic peptide testing: where are we now? Heart (2004) 90:725–726.
10 Wang C, Fitzgerald J, Schulzer M, Mak E, Ayas N. Does this dyspneic patient in the emergency department have congestive heart failure? J Am Med Assoc (2006) 294:1944–1956.[Web of Science]
11 Hosmer D, Hosmer T, Le Cessie S, Lemeshow S. A comparison of goodness-of-fit tests for the logistic regression model. Stat Med (1997) 16:965–980.[CrossRef][Web of Science][Medline]
12 Nielsen OW, Hansen JF, Hilden J, Larsen CT, Svanegaard J. Risk assessment of left ventricular systolic dysfunction in primary care: cross sectional study evaluating a range of diagnostic tests. BMJ (2000) 320:220–224.
13 Jaeschke R, Guyatt G, Sackett DL. Users' guides to the medical literature. III. How to use an article about a diagnostic test. A. Are the results of the study valid? J Am Med Assoc (1994) 271:389–391.
14 Jaeschke R, Guyatt G, Sackett DL. Users' guides to the medical literature. III. How to use an article about a diagnostic test. B. What are the results and will they help me caring for my patients? J Am Med Assoc (1994) 271:703–707.
15 Fox KF, Cowie MR, Wood DA, Coats AJS, Poole-Wilson PA, Sutton GC. A Rapid Access Heart Failure Clinic provides a prompt diagnosis and appropriate management of new heart failure presenting in the community. Eur J Heart Fail (2000) 2:423–429.[CrossRef][Web of Science][Medline]
16 The National Collaborating Centre for Chronic Conditions. Chronic Heart Failure. National Clinical Guideline for Diagnosis and Management in Primary and Secondary Care (2002) London: NHS Executive.
17 Struthers AD. Heart failure: the diagnosis of heart failure. Heart (2000) 84:334–338.
18 Skaner Y, Bring J, Ullman B, Strender LE. Heart failure diagnosis in primary health care: clinical characteristics of problematic patients. A clinical judgement analysis study. BMC Fam Pract (2003) 4.
19 Davie AP, Francis CM, Caruana L, Sutherland GR, McMurray JJ. Assessing diagnosis in heart failure: which features are any use? QJM (1997) 90:335–339.
20 Landray MJ, Lehman R, Arnold I. Measuring brain natriuretic peptide in suspected left ventricular systolic dysfunction in general practice: cross-sectional study. BMJ (2000) 320:985–986.
21 Davie AP, Francis CM, Love MP, et al. Value of the electrocardiogram in identifying heart failure due to left ventricular systolic dysfunction. BMJ (1996) 312:222–224.
22 Badgett RG, Lucey CR, Mulrow CD. Can the clinical examination diagnose left-sided heart failure in adults? JAMA (1997) 277:1712–1719.
23 Department of Health. National Service Framework for Coronary Heart Disease. Modern Standards and Service Models (2000) London: HMSO.
24 McAlister FA, Straus S, Sackett DL. Why we need large, simple studies of the clinical examination: the problem and a proposed solution. Lancet (1999) 354:1721–1724.[CrossRef][Web of Science][Medline]
25 Laupacis A, Sekar N, Stiell I. Clinical prediction rules: a review and suggested modifications of methodological standards. J Am Med Assoc (1997) 277:488–494.
26 Katz MH. Multivariable analysis: a primer for readers of medical research. Ann Int Med (2003) 138:644–650.
27 McGinn TG, Guyatt GH, Wyer PC, Naylor CD, Richardson WS. Users' guides to the medical literature. XXII: how to use articles about clinical decision rules. J Am Med Assoc (2000) 284:79–84.
28 Smith JA, Bruusgaard D, Bodd E, Hall C. Relations between medical history, clinical findings and plasma N-terminal proatrial natriuretic peptide in patients in primary health care. Eur J Heart Fail (2001) 3:307–313.[CrossRef][Web of Science][Medline]
29 Ng LL, Loke I, Davies JE, et al. Identification of previously undiagnosed left ventricular systolic dysfunction: community screening using natriuretic peptides and electrocardiography*1. Eur J Heart Fail (2003) 5:775–782.[CrossRef][Web of Science][Medline]
30 Battaglia M, Pewsner D, Juni P, Egger M, Bucher HC, Bachmann LM. Accuracy of B-type natriuretic peptide tests to exclude congestive heart failure: systematic review of test accuracy studies. Arch Intern Med (2006) 166:1073–1080.
CiteULike 
Connotea 
Del.icio.us What's this?
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||