Eligible patients participating in the study (n = 405, 34%) were 1.9 years younger than non-responders (n = 781, 66%) and healthier overall (). The median age of the study population was 73 (SD 5.3) years, and 55% were male.
Characteristics of eligible patients with general practitioner's diagnosis of COPD, according to participation in study. Values are numbers (percentages) of patients unless stated otherwise
In 83 patients (20.5%) the consensus panel set a new diagnosis of heart failure. Of these, 42 patients had systolic and 41 had isolated diastolic heart failure. There were no cases of isolated right sided heart failure. All 41 patients with isolated diastolic heart failure had symptoms indicative of heart failure. In addition to these symptoms, 22 patients had also indicative signs of heart failure, four patients had atrial fibrillation, eight patients had echocardiographic left ventricular hypertrophy, two patients had angina and five patients had a combination of atrial fibrillation, left ventricular hypertrophy, or angina.
Re-presenting (blinded to the original decision) a random sample of 41 (10%) patients to the panel showed disagreement in one case only (Cohen's κ = 0.92). This patient had moderatesevere dyspnoea, indicative symptoms and signs of heart failure, atrial fibrillation, a slightly impaired echocardiographic left ventricular ejection fraction of 45-50%, and a normal diastolic function. This patient was originally classified as not having heart failure and subsequently reclassified as having heart failure.
Of all participants, only three had an S3-gallop, and 11 patients had signs of pulmonary fluid on chest radiography. One participant had a serum creatinine concentration > 200 μmol/l (243 μmol/l), and no participant had a blood urea > 20 mmol/l.
shows the univariable associations. Electrocardiographic abnormalities were more common in those with heart failure, mostly ST or T wave abnormalities, or both (22.7%), left bundle branch block (complete or incomplete) (16.1%), and Q waves suggesting a previous myocardial infarction (7.7%). Results from pulmonary function tests were similar in patients with and without heart failure.
Characteristics of participants according to presence or absence of heart failure and results of univariable analysis. Values are numbers (percentages) unless stated otherwise
Of the variables from history and physical examination with a univariable P < 0.15, only history of ischaemic heart disease (odds ratio 2.16), a laterally displaced apex beat (2.34), body mass index (BMI) (1.11 per kg/m2), and heart rate (1.26 per 10 beats/minute) were independent clinical predictors of presence of heart failure in multivariable analysis and included in the “clinical” model (). Cardiovascular medication (such as diuretics or angiotensin converting enzyme inhibitors) was not an independent predictor.
Independent contribution according to multivariable analysis of tests from history, physical examination, and additional tests to diagnosis of heart failure in patients with general practitioner's diagnosis of COPD
NT-proBNP (odds ratio 1.06 per 5 pmol/l) was the best single diagnostic test when applied without information from the clinical assessment with an ROC area of 0.72 (0.66 to 0.79, P < 0.001). Addition of NT-proBNP to the four clinical items significantly increased the ROC area from 0.70 to 0.77 (). Addition of electrocardiography to the clinical model increased the ROC area significantly from 0.70 to 0.75, and addition of C reactive protein or cardiothoracic ratio increased it to 0.73 (). Addition of an abnormal electrocardiogram to the clinical+NT-proBNP model was also significantly associated with presence of heart failure (odds ratio 2.75) (). When added to the clinical plus NT-proBNP model, C reactive protein (1.03 per mg/l, 1.00 to 1.07, P = 0.09) and cardiothoracic ratio (1.05 per unit, 1.00 to 1.07, P = 0.11) were borderline associated with presence of heart failure and did not change the ROC area. Hence, our final model included a history of ischaemic heart disease, a laterally displaced apex beat, high body mass index, raised heart rate, NT-proBNP, and abnormal electrocardiogram (). shows the diagnostic accuracy of the independent predictors in the final model.
Unadjusted sensitivity, specificity, and predictive values of variables in final model
Using the formula of the final model in , we can estimate a patient's probability of heart failure based on his or her clinical profile and the NT-proBNP and result of electrocardiography. The ROC area was 0.76 (). To facilitate use in daily care, we simplified the final model to an easy applicable scoring rule. Regression coefficients of the variables of the final model were rounded to integers according to their relative contribution to the risk estimation (see , last column). Subsequently, for each patient the total points were estimated by this scoring rule. The total score of the patients ranged from 0-14 points. The observed prevalence of heart failure among very low risk patients (0 points) was 4.9% (4 out of 81 patients), 10.6% (15 out of 142 patients) among low risk (2-5 points), 25.4% (32 out of 126 patients) among medium risk (6-9 points), and 57.1% (32 out of 56 patients) among high risk patients (10-14 points) (). Dichotomising the scale at, for example, 9 points (at ≤ 9 points the diagnosis is negative and > 9 it is positive) yielded a positive predictive value of 57.1% and a negative predictive value of 85.4% ().
Distribution of presence and absence of heart failure per score category of final model and corresponding sensitivity, specificity, and predictive values when dichotomised at different score thresholds*