Plasma Natriuretic Peptides for Community Screening for Left Ventricular Hypertrophy and Systolic Dysfunction: Title and subTitle BreakThe Framingham Heart Study FREE

It is widely acknowledged that asymptomatic left ventricular systolic dysfunction (LVSD) is a treatable precursor of congestive heart failure.^{1 - 2} This has generated enthusiasm for developing community-based screening tools for LVSD with the belief that early detection will be clinically beneficial.³ Similarly, the demonstration of echocardiographic left ventricular (LV) mass as a risk factor for cardiovascular events⁴ has spurred interest in screening select populations such as those with hypertension with a "limited" echocardiogram.⁵ The notion of community-wide screening for LVSD or left ventricular hypertrophy (LVH), however, has been challenged by concerns over the costs of echocardiography.⁶ In this context, the demonstration that patients with LVSD or LVH have elevated plasma natriuretic peptide levels,^{7 - 9} coupled with the availability of inexpensive assays, has spawned interest that these peptides may serve as useful initial screening tests.

Several studies have reported on the use of the brain natriuretic peptide (BNP) and the N-terminal component of proatrial natriuretic peptide (NT-ANP) for detecting asymptomatic LVSD and LVH.^{6 ,10 - 14} These prior investigations have been limited, however, by selection biases, small samples, sex-pooled analyses, inclusion of patients with both symptomatic and asymptomatic LVSD, and the use of less sensitive first-generation radioimmunoassays in some reports. Accordingly, we examined the usefulness of natriuretic peptides for screening for LVSD or elevated LV mass in a large community-based sample of asymptomatic subjects. We evaluated men and women separately, and we assessed the diagnostic performance of BNP and NT-ANP over a wide range of discrimination limits.

The design of the Framingham Offspring Study has been detailed previously.¹⁵ The 3532 subjects who attended the sixth examination cycle (1995-1998) were eligible for the present investigation. All attendees underwent a routine physical examination, laboratory assessment of vascular risk factors, electrocardiography, and echocardiography. Subjects were excluded from the present investigation because of unavailable natriuretic peptide levels (n = 80), creatinine ≥2.0 mg/dL [177 µmol/L] (n = 21), history of heart failure (n = 33), or inadequate echocardiograms (n = 221). After these exclusions, 3177 subjects (1707 women; 89.9% of attendees) remained eligible.

At baseline, all subjects underwent M-mode and 2-dimensional echocardiography (Sonos 1000, Hewlett-Packard, Palo Alto, Calif). Digitized images were stored on optical disks and measured using an offline analysis system. A sonographer or cardiologist, blinded to clinical information and natriuretic peptide results, read all echocardiograms according to standardized protocols. Left ventricular internal dimension (LVID) and the thicknesses of the interventricular septum (IVST) and the left ventricular posterior wall (PWT) were obtained by averaging digital M-mode measurements in at least 3 cardiac cycles using the leading-edge technique, according to the American Society of Echocardiography.¹⁶ Left ventricular mass was calculated using the formula LV mass (g) = 0.8[1.04(LVID + IVST + PWT)³ − (LVID)³] + 0.6, with all measurements made at end-diastole.¹⁷ Left ventricular ejection fraction (LVEF) was estimated by visual assessment of LV contractile performance and wall motion in multiple 2-dimensional views. The LVEF was categorized as normal (>55%), borderline (51%-55%), mildly reduced (41%-50%), moderately diminished (31%-40%), and severely impaired (≤30%). The accuracy of such visual assessment of LVEF has been validated previously.¹⁸ All studies with suspected LV systolic dysfunction (LVEF borderline or lower) were also read by a cardiologist.

For this investigation, we defined 3 primary echocardiographic variables of interest: (1) elevated LV mass, defined as a value greater than or equal to the sex-specific 90th percentile value of LV mass/(height)² for the entire sample; (2) LVSD, defined as the presence of mild or greater degree of impairment of LVEF on 2-dimensional assessment, or a fractional shortening (FS) less than 29% on M-mode. A value of 29% corresponds to an LVEF less than or equal to 50% and was the first percentile value of FS in a healthy reference sample; and (3) moderate to severe LVSD, defined as the presence of moderate or severe impairment of LVEF on 2-dimensional assessment, or an FS less than 22% (which corresponds to an LVEF of approximately ≤40%).^{11 ,19}

We chose a composite definition of LVSD that combined semiquantitative 2-dimensional assessment of global LV function with FS, a quantitative M-mode–derived measure of basal LV systolic function. Additional analyses were performed using only M-mode criteria for LVSD (FS<29%), and with the sex-specific 90th percentile of height-adjusted LV end-diastolic diameter as the cutpoint for defining LV dilation.

A fasting blood sample was obtained with the subjects in a supine position. Blood specimens were centrifuged immediately, and the plasma stored at –70° C without freeze-thaw cycles until the natriuretic peptides were measured. Both BNP and NT-ANP were measured from unextracted plasma using highly sensitive noncompetitive immunoradiometric assays (ShionoRIA, Shionogi Inc, Osaka, Japan). All measurements were made in duplicate and averaged. The lower limits of detection for these assays were 4 pg/mL for BNP and 94 pmol/L for NT-ANP. The average interassay coefficients of variation were 12.2% for BNP and 12.7% for NT-ANP.

Receiver Operating Characteristic Curves: Test Accuracy We used natural logarithmic (log) transformation for BNP and NT-ANP because of skewed distributions. Sex-specific logistic regression²⁰ was used with each echocardiographic phenotype as the binary dependent variable, and the log natriuretic peptide as the continuous independent variable. Separate analyses were performed for BNP and NT-ANP and for each echocardiographic variable.

Sex-specific receiver operating characteristic (ROC) curves were used to examine the performance characteristics of the natriuretic peptides over their entire range of values.²¹ The c statistic from the logistic regression model corresponds to the area under the ROC curve (AUC).²² The AUC was used as an index of global test performance, with an AUC of 0.5 indicating no discrimination ability.²¹ For each peptide and each echocardiographic variable, we compared the AUC for men and women.²³ Additionally, we compared the sex-specific AUC for BNP vs NT-ANP for each echocardiographic trait.²³

Test Performance in High-Risk Subgroups Investigators have suggested that natriuretic peptide screening may be especially cost-effective if individuals with increased pretest probability of LVH or LVSD are targeted.⁶ Accordingly, we examined the performance of the natriuretic peptides in 4 prespecified subgroups: age 60 years or older; hypertensives²⁴ ; participants with known cardiovascular disease; and subjects with 2 or more of these "high-risk" features. Cardiovascular disease was defined as coronary heart disease (other than a recognized myocardial infarction), cerebrovascular or peripheral vascular disease.²⁵ Subjects with a prior recognized myocardial infarction were not included in this group because echocardiography is indicated in them regardless of natriuretic peptide levels.

Defining Optimal Discrimination Limits for Screening: Test Efficacy The efficacy of a screening test is dependent not only on its overall accuracy (assessed by the AUC) but also on the prevalence of the condition screened for, the costs of the follow-up confirmatory test, and the consequences of misclassification (costs associated with false-positive and false-negative tests).²⁶ Accordingly, we evaluated 3 different strategies, all defined a priori, for choosing the optimal natriuretic peptide discrimination limits by examining diagnostic performance at 3 cutpoints: (1) yielding specificity values of 0.95 (detection rate for a 5% false-positive rate)²⁷ ; a high specificity was chosen to assess the "rule in" ability of the peptides; (2) maximizing the sum of sensitivity and specificity (Youden index)²⁸ ; such a cutpoint weighs true positives and false negatives equally (point on the ROC curve closest to the upper left corner); and (3) corresponding to the age- and sex-specific 95th percentile values of the natriuretic peptides in a large healthy reference sample.²⁹

We evaluated the sensitivity, specificity, positive and negative predictive values, and accuracy for select cutoff values of log BNP and log NT-ANP, and likelihood ratios (LRs) were calculated to evaluate the ability of a threshold value of the peptide to rule in (LR positive) or rule out (LR negative) the presence of LVSD or elevated LV mass.³⁰

Multivariable Prediction of LVH and LVSD Since in practice clinicians evaluate patients prior to obtaining imaging studies or screening tests, we evaluated the association of natriuretic peptides and echocardiographic outcomes in multivariable models that incorporated known clinical risk factors for elevated LV mass and LVSD. The incremental contribution of natriuretic peptides to a multivariable model incorporating clinical variables was evaluated by the increase in the c statistic of the logistic models.²²

All analyses were performed with SAS 6.12 (SAS Institute Inc, Cary, NC), and P<.05 indicated statistical significance.

Baseline characteristics of 3177 study subjects are summarized in Table 1. Overall, 179 subjects (137 men) had LVSD, of whom 70 (60 men) had moderate to severe LVSD. BNP levels were higher in individuals with elevated LV mass and in subjects with LVSD (Figure 1) compared with participants without these features. Plasma BNP and NT-ANP levels were correlated (Spearman correlation coefficients: men, 0.70; women, 0.62; P<.001 for both).

Logistic regression analyses confirmed that BNP and NT-ANP were positively associated with elevated LV mass and LVSD in both sexes. In stepwise models examining the 2 peptides together, BNP entered the regression models first, whereas NT-ANP did not enter subsequently.

Table 2 shows the AUC for the natriuretic peptides for identifying echocardiographic abnormalities of interest. Sex-specific ROC plots for BNP for detection of elevated LVM and LVSD are displayed in Figure 2. With the exception of moderate to severe LVSD, AUC was at or below 0.75 for both peptides. For detection of elevated LV mass and LVSD, AUC was higher in men compared with women (P<.005), and marginally higher for BNP than for NT-ANP.

The AUC for both peptides remained unchanged when elevated LV mass was defined as a value in the top quintile, or when values exceeding the sex- and height-specific 95th percentile of a healthy reference sample were used. The AUC for the peptides for detecting elevated LV mass without LVSD were nearly identical to that for identifying elevated LV mass regardless of LV systolic function. The AUC varied from 0.52 to 0.56 in women and from 0.70 to 0.72 in men when FS less than 29% was used to define LVSD, and from 0.58 in women to 0.67-0.69 in men for identification of LV dilation.

In the 4 high-risk subgroups of men, diagnostic performance of the natriuretic peptides was similar to that for the entire group for all echocardiographic findings (Table 2). In women, diagnostic performance for detecting LVSD improved considerably in the high-risk groups, approaching values obtained in men. This was particularly striking for women with prevalent cardiovascular disease. However, the AUC for detecting elevated LV mass was similar to that for the entire group of women.

Since AUCs for BNP were slightly higher compared with NT-ANP in initial analyses, we focused on BNP when defining optimal discrimination limits using the 3 strategies outlined earlier. Plasma BNP levels of 45 pg/mL in men and 50 pg/mL in women yielded 95% specificity for detecting LVSD (Table 3). Similarly, a plasma BNP level of 46 pg/mL in men and 47 pg/mL in women corresponded to sensitivities of 0.27 (for men) and 0.13 (for women) for detecting elevated LV mass with 95% specificity.

Lower levels of BNP (21-24 pg/mL in men and 27-34 pg/mL in women) were associated with maximization of the sums of sensitivity and specificity for detecting elevated LV mass and LVSD (Table 3). However, false positives far outnumbered true positives at these thresholds. Use of age- and sex-specific reference limits (95th percentile values) yielded higher sensitivities for detecting elevated LV mass and LVSD (compared with 95% specificity-based limits), but with the trade-off of lower specificities, and lower LR-positive and positive predictive values (Table 3).

Table 4 displays the usefulness of known clinical risk factors alone and in combination with natriuretic peptides for identifying elevated LV mass and LVSD. The addition of BNP and NT-ANP to multivariable models incorporating clinical risk factors resulted in a minimal increase in the c statistic (0.02 to 0.03) that was statistically significant in men only.

The association of LVSD and LVH with elevation of plasma natriuretic peptide levels has lent support to the notion of mass screening with an inexpensive test.³ In prior investigations examining the diagnostic usefulness of plasma natriuretic peptide levels, the AUC for detecting LVSD varied from 0.83 to above 0.90,^{6 ,10 - 12 ,31 - 33} values that exceed the performance of accepted screening tests such as the Papanicolaou smear,³⁴ mammography,³⁵ and the prostate-specific antigen test.³⁶ Fewer studies have evaluated the usefulness of natriuretic peptides for diagnosis of LVH and have reported widely varying AUC values (0.67-0.91).^{9 ,11 ,37}

Prior investigations have several limitations. First, BNP discrimination limits for identifying LVSD vary 4-fold across studies, from a value of 17.9 pg/mL to 75 pg/mL,^{6 ,12} in part related to selection bias (referral bias, inclusion of symptomatic and asymptomatic individuals, post–myocardial infarction patients), varying definitions of LVSD, and different natriuretic peptide assays. Second, previous studies combined men and women, overlooking the substantial differences in the distribution of BNP levels between the sexes,²⁹ the considerably lower prevalence of LVSD in women,³⁸ and sex-specific differences in natriuretic peptide responses to LV dysfunction³⁹ ; these factors can influence the performance characteristics of natriuretic peptides and the choice of discrimination limits in men vs women. Third, the diagnostic performance of natriuretic peptides in select high-risk groups has not been adequately evaluated. Finally, prior investigations have not adequately addressed a key criterion for the selection of discrimination limits for screening tests performed on asymptomatic individuals in the general population, ie, the importance of high specificity ("rule in" strategy) as opposed to a high sensitivity ("rule out" strategy), which is required of a diagnostic test performed on symptomatic patients presenting to clinicians.⁴⁰ In the case of natriuretic peptide screening for asymptomatic LVSD or elevated LV mass, because of the size of the screened population and the low prevalence of the condition, the false-positive rate of the screening test must be low to reduce the burden of expensive follow-up tests (echocardiography).

We evaluated the diagnostic performance of BNP and NT-ANP for detection of elevated LV mass and varying degrees of LVSD using high-sensitivity assays, performing sex-specific analyses and examining several screening strategies.

Overall, the AUC for both peptides was at or below 0.75 for elevated LV mass and LVSD in both sexes. While the AUC for moderate to severe LVSD exceeded 0.80 in women, this estimate was based on only 10 individuals, had wide confidence intervals, and should be interpreted with caution. Generally, BNP performed slightly better than NT-ANP for all echocardiographic phenotypes and in both sexes. Other investigators have reported similar findings.^{6 ,41}

For both natriuretic peptides, AUC for men exceeded that for women for both elevated LV mass and LVSD. The AUC in women was indicative of poor diagnostic performance, and is likely related to a greater degree of overlap of natriuretic peptide values in groups with and without elevated LV mass or LVSD. A lesser BNP response to LVSD could be a contributing factor.³⁹ We could not identify any high-risk subgroup in men in which the natriuretic peptides performed better than in the overall sample. In women, however, the AUC for LVSD increased when the test was considered in high-risk groups.

Discrimination limits based on 95% specificity outperformed thresholds derived from maximizing the sum of sensitivity and specificity, and those based on age- and sex-specific reference limits, when the criteria of high specificity, positive predictive value, and LR positivity were used. However, sensitivity was low for the resultant thresholds.

Of note, the c statistic of models with clinical risk factors was higher than that for models incorporating the natriuretic peptides alone. Furthermore, the incremental contribution of BNP to models incorporating clinical variables was low, as reported by others in referral samples.³³

Our community-based sample reduces the possibility of spectrum bias, while the evaluation of all subjects with both echocardiography and natriuretic peptide levels eliminates verification bias. The independent analysis of echocardiography and natriuretic peptide test results blinded to each other and to all clinical data removes interpretation bias and avoids context bias.

Several limitations must be noted. It may be argued that echocardiography is an imperfect gold standard, and that some of the false-positive natriuretic peptide elevations may represent true positives due to conditions such as LV diastolic dysfunction that were not directly assessed in our study. In this context, it is noteworthy that the natriuretic peptides performed poorly for detection of elevated LV mass in the absence of systolic dysfunction, a surrogate for LV diastolic dysfunction.⁴² The use of visual 2-dimensional assessment of LV systolic function instead of a quantitative method (such as Simpson's formula) might also be an additional limitation of our study. It is important to note that such visual assessment of LVEF has been shown to be accurate for diagnostic purposes¹⁸ and provides important prognostic information.⁴³ It is also important to note that the very low prevalence of moderate to severe LVSD in women hampers our ability to examine the diagnostic performance of natriuretic peptides as screening tests for this condition among women. The Framingham Study cohort is overwhelmingly white, reducing the generalizability of our findings to other ethnicities. Lastly, it is important to note that our findings cannot be extrapolated to the use of natriuretic peptides for diagnosis of heart failure in subjects with acute dyspnea. In such a situation, a rule-out strategy that maximizes sensitivity of a test is preferable (higher pretest probability of LVSD) and natriuretic peptides have been reported to perform well for this very different purpose.⁷

Contrary to some prior reports, the performance of natriuretic peptides for detection of elevated LV mass or LVSD in the community-based cohort was only fair in men, and poor in women. Even at cutpoints yielding 95% specificity, for which negative predictive values are high, the resultant low sensitivity combined with the low prevalence of LVSD may offset any potential usefulness of screening. The better performance of clinical variables (compared with natriuretic peptides alone) and minimal incremental usefulness of natriuretic peptides raise the possibility that a clinical risk score for identifying LVH or LVSD may be an alternative strategy to screening with natriuretic peptides.³³ Overall, our findings do not support use of natriuretic peptides for screening for elevated LV mass and LVSD in the general population. Additional investigations to confirm our findings are warranted.