Study Subjects

Using administrative data, we retrospectively identified 1,235 consecutive first admissions with congestive HF (admission codes I500 [congestive HF], I501 [LV failure], and I509 [HF, unspecified]) at Royal Hobart Hospital and Launceston General Hospital, the two major referral hospitals in Tasmania, from July 2009 to June 2012. At these hospitals, patients with HF with other comorbidities are usually admitted under the care of the general medical team with related subspecialty input. The study was approved by the Tasmanian Health Research Ethics Committee.

Clinical Data

Clinical parameters (sociodemographic variables, comorbidity and medical history, specialty services during the index admission, medications at discharge, vital signs at discharge, serum markers at the closest discharge, and electrocardiographic data at the closest discharge) were comprehensively assessed by viewing each medical record. The Charlson comorbidity index was calculated as previously described.

Echocardiography

Transthoracic echocardiography was performed by experienced sonographers using a commercially available ultrasound machine (Vivid 7, Vivid 9, or Vivid i; GE Vingmed Ultrasound AS, Horten, Norway). Echocardiographic images were digitally recorded and downloaded as Digital Imaging and Communications in Medicine files to an imaging server for offline analysis. All echocardiographic parameters, including strain, were analyzed by a core laboratory at Menzies Research Institute Tasmania. Conventional echocardiographic parameters were measured according to the recommendations of the American Society of Echocardiography. Transmitral early diastolic velocity (E) was acquired in the apical four-chamber view using pulsed-wave Doppler at the level of the mitral valve tips during diastole. Early diastolic mitral annular tissue velocity (e′) was also measured in the apical four-chamber view, with the sample volume positioned at both the septal and lateral mitral annuli and being the average of these two values. The combined assessment of E and e′ was used to calculate the E/e′ ratio. Diastolic function grade was also assessed according to American Society of Echocardiography guidelines. Inferior vena cava diameter and respiratory collapse were measured and used for the estimation of right atrial pressure (RAP). Elevated RAP was defined as an estimated RAP of 15 mm Hg. The severity of valvular heart disease was assessed in accordance with American Society of Echocardiography guidelines. The median interval between echocardiography and discharge date was 4 days.

Strain

We measured the two strain parameters (global longitudinal strain [GLS] and global circumferential strain [GCS]) using standard methodologies for speckle-tracking (Research Arena; TomTec Medical Imaging, Unterschleissheim, Germany). After manual tracing of the LV endocardial border, the dedicated software automatically tracked the myocardium throughout the cardiac cycle. The peak values of segmental longitudinal strain were obtained from greyscale-recorded images in the apical four-chamber, two-chamber, and long-axis views, and GLS was obtained by averaging the peak values. The peak values of the six segmental circumferential strain curves were obtained from the short-axis view at the papillary muscle level and averaged to provide GCS. The mean frame rate was 56 ± 23 frames/sec. In patients with atrial fibrillation, strain parameters were measured if the ratio of preceding and pre-preceding intervals was 1.

Incomplete Data

All hospital admissions with new-onset HF were captured over the 3-year study period. The following measures are nonroutine and therefore incomplete: discharge B-type natriuretic peptide (BNP; performed in 41%), discharge troponin (82%), diastolic function evaluation (e′ in 75%, E/e′ in 74%, diastolic function grade in 74%), and inferior vena cava estimation of RAP (79%). Preadmission New York Heart Association class was ascertained from the notes in 85%. Missing data for all other variables were <5%, and apart from the incomplete variables mentioned above, missing data for continuous variables were imputed using the corresponding mean value, to minimize bias and optimize model inclusion.

Follow-Up

The primary outcome was 30-day all-cause death or hospital readmission after discharge. Outcomes were checked by data linkage to administrative databases from the Clinical Informatics and Business Intelligence Unit of the Department of Health and Human Services of Tasmania. This captures the death registry and all admission data in Tasmanian public hospitals. Because Tasmania is an island, it is unusual for a patient to be admitted to a hospital in another state, and public hospital patients are not commonly admitted to private hospitals. HF-specific readmission was decided when the admission code was I500 (congestive HF) or I501 (LV failure). Patients were censored at the time of readmission or death or at the end of follow-up (maximum, 30 days).

Statistical Analysis

Data are expressed as mean ± SD or median (interquartile range). The significance of differences between the groups was assessed using Student t tests when the distributions were normal. Mann-Whitney U tests were used for data that were not normally distributed. For categorical variables, χ ² tests or Fisher exact tests were used, as appropriate. For multiple comparisons, one-way analysis of variance or χ ² tests were used, followed by Bonferroni correction as appropriate. Univariate and multivariate Cox proportional hazards models were used to determine the contribution of 30-day death or readmission. Clinically relevant parameters were entered into the model on the basis of associations in the univariate analysis with P values < .10. Age and gender, Charlson comorbidity index, and hematocrit were forced into the model. The independence and robustness of GLS and GCS were examined using three different models: clinical (age, gender, systolic blood pressure, Charlson comorbidity index, and angiotensin-converting enzyme [ACE] inhibitor or angiotensin receptor blocker [ARB] use), laboratory (age, sex, hematocrit, blood urea nitrogen, and sodium), and echocardiography (age, gender, LV mass index, mitral regurgitation, and LV EF). GLS and GCS were included in the model with EF because their correlations were modest (GLS vs EF, R ² = 0.55; GCS vs EF, R ² = 0.48). Similarly, the independence of GLS was also assessed in the patient subgroups with LV EFs < 50% and ≥ 50%, and models were repeated to include subgroups with complete data for e′, E/e′ ratio, RAP, troponin, and BNP. The incremental value of GLS in the overall group was assessed in three modeling steps, using nested models. The first step consisted of fitting a multivariate model of age, sex, systolic blood pressure, ACE inhibitor or ARB use, blood urea nitrogen, and sodium. Then, LV EF was included in the second step. Finally, GLS was included in the third step. The change in overall log-likelihood ratio χ ² was used to assess the increase in predictive power after the addition of GLS. Harrell’s C statistic was used to evaluate model performance. Reclassification was evaluated to assess the incremental benefit of adding GLS to the model on the basis of clinical parameters and LV EF with categorical net reclassification improvement using the quartile boundaries of each model probability calculated from multivariable logistic regression. Receiver operating characteristic curves were used to determine the optimal cutoff values of GLS and GCS for the association of 30-day readmission or death and to compare the strength of the models. Comparisons of area under the curve (AUC) (clinical parameters plus LV EF vs clinical parameters plus GLS) in each group were performed with the method suggested by DeLong et al . Survival was estimated by the Kaplan-Meier method, and differences in survival between groups were assessed by the log-rank test.

Inter- and intraobserver variability was examined for GLS and GCS. Measurements were performed in a group of 20 randomly selected subjects by one observer and then repeated on 14 separate days by two observers. Data are presented as means of the absolute and relative differences between measurements and by the correlation coefficient. Statistical analysis was performed using standard statistical software packages (SPSS version 20.0 [SPSS, Inc, Chicago, IL] and R version 3.0.2 [ http://cran.r-project.org ]), and statistical significance was defined by P < .05.

Patient Characteristics

Of the 1,235 admissions, we identified 659 patients who underwent inpatient echocardiography. After the exclusion of patients with suboptimal echocardiographic image quality ( n = 45), inability to evaluate strain because of irregular heart rate in the acquired images (mainly atrial fibrillation; n = 132), and death during the index hospitalization ( n = 14), the final analysis was based on data from 468 patients ( Figure 1 ).

Selection of patients with heart failure for study, on the basis of availability of inpatient echocardiogram and adequate echocardiographic quality.

Events

Follow-up data were available in all 468 patients, with 92 events (20%) occurring within 30 days after discharge, including 15 deaths (nine cardiac deaths) and 77 readmissions (eight patients died after readmission within 30 days). Among the 77 readmitted patients, 51 (66%) were due to progressive HF. The common reasons for readmission other than HF were lung disease (seven patients) and coronary artery disease (five patients). When the patients were divided into groups with HFpEF (LV EF ≥ 50%) or HF with decreased LV EF (LV EF < 50%), the latter comprised 272 patients and 67 events, and the former included 196 patients and 25 events. Table 1 summarizes the baseline clinical, therapeutic, and echocardiographic parameters in patients with and without 30-day death or readmission. These findings conformed to an expected case mix of HF, with a mean age of 77 ± 12 years, a mean LV EF of 45 ± 17%, and mean GLS of −10.9 ± 4.7%. Patients readmitted for worsening HF and those dying of cardiac causes had similar impairment of GLS as patients readmitted for other causes and dying of noncardiac causes (−7.8 % [interquartile range, −5.5% to −10.4%] vs −9.0% [interquartile range, −6.3% to −11.7%], P = .29).

Associations with Overall Outcomes

Table 2 summarizes univariate Cox regression analyses for association with primary outcome, focused on clinical parameters, therapeutic parameters, and echocardiographic parameters. Thirty-day death or readmission was significantly associated with male sex, lower systolic blood pressure, ACE inhibitor or ARB use, New York Heart Association class ≥ III, higher troponin, higher blood urea nitrogen, lower sodium, higher LV mass index, lower e′, higher RAP, lower LV EF, and greater impairment of strain.

The independent association of GLS with outcome was examined using three different models ( Table 3 ). GLS was consistently significant in every model, and hazard ratios (HRs) were similar (1.13–1.18). No independent association between EF and readmission was observed in the echocardiography model. Similarly, the independent association of GCS with outcome was also assessed using the same models. GCS was independently associated with outcome in the clinical (HR, 1.05; 95% confidence interval [CI], 1.01–1.08; P = .01) and laboratory models (HR, 1.05; 95% CI, 1.02–1.08; P < .01), but not in the echocardiography model (HR, 1.02; 95% CI, 0.98–1.06; P = 0.30).

Independence of GLS in Complete Data Sets

In addition, because the number of patients with missing values for e′, E/e′ ratio, RAP, troponin, and BNP could have led the above results to overstate the importance of GLS, we repeated the Cox models in subsets with all observations. GLS was again significantly associated with primary outcome after adjustment of age, gender, ACE inhibitor or ARB use, e′, E/e′ ratio, RAP, troponin, and BNP (HR, 1.14–1.16).

Independent Association of GLS and Outcome in Patients with LV EF < 50% and ≥ 50%

In the Cox univariate analyses, outcome was associated with atrial fibrillation, higher blood urea nitrogen, lower sodium, ACE inhibitor or ARB use, and greater impairment of GLS in patients with HF with LV EFs < 50% and was associated with New York Heart Association class ≥ III, aortic stenosis, higher LV mass index, lower e′, and greater impairment of GLS in patients with HF with LV EFs ≥ 50% ( P < .10 for all). In patients with HF with LV EFs < 50%, GLS was independently associated with outcome (HR, 1.12; 95% CI, 1.02–1.22; P = .02) after adjusting for age, male gender, atrial fibrillation, blood urea nitrogen, sodium, and ACE inhibitor or ARB use. In patients with HF with LV EFs ≥ 50%, GLS was also independently associated with outcome ( Table 4 ).

Discrimination by GLS and GCS

The receiver operating characteristic curve analysis for the associations of 30-day death or readmission (AUC, 0.67) showed GLS ≥ −10.5% to show optimal sensitivity (74%) and specificity (55%) in the overall group. When the patients were divided into two groups on the basis of LV EF < 50% or ≥ 50%, GLS appeared to be more associated with readmission and death in patients with preserved (≥50%) LV EFs (AUC, 0.70; P < .01) than in those with LV EFs < 50% (AUC, 0.60; P = .01). Cutoffs were also different; GLS ≥ −14.8% showed optimal sensitivity (88%) and specificity (50%) in patients with preserved EFs, whereas GLS ≥ −7.2% showed optimal sensitivity (54%) and specificity (63%) in patients with reduced EF. Patients with GLS below each cutoff had significantly worse survival than those with less severe deterioration of GLS in the group overall, in patients with LV EFs < 50%, and even patients with LV EFs ≥ 50% ( Figure 2 ).

Kaplan-Meier event-free curves in the overall group (A) , patients with LV EFs < 50% (B) , and patients with LV EFs ≥ 50% (C) stratified in two groups according to the optimal cutoff value of GLS (−10.5%).

Similarly, GCS ≥ −14.9% showed optimal sensitivity (65%) and specificity (56%) in overall patients for the association with outcome. However, AUCs for GCS in patients with preserved (AUC, 0.59; P = .17) and reduced (AUC, 0.53; P = .42) LV EFs were not significantly different.

Incremental Value of GLS over Clinical Parameters and LV EF in the Overall Group

In sequential Cox models, the model based on clinical and laboratory variables was significantly improved by the addition of LV EF and furthermore improved by adding GLS ( Figure 3 ). Moreover, the addition of GLS to the model based on clinical parameters and LV EF led to a further significant reclassification improvement in the overall group (categorical net reclassification improvement, 0.15; P = .04) ( Table 5A ). In addition, the AUC of the association of clinical parameters and GLS with outcome was significantly better than that of the clinical and LV EF model ( Figure 4 A).

Incremental value of GLS over clinical parameters and LV EF as a correlate of 30-day death or readmission. ACEi , Angiotensin-converting enzyme inhibitor; ARB , angiotensin receptor blockers.

Table 5A

Net reclassification table: overall

Readmission ( n = 86)	Clinical + LV EF + GLS				Reclassified		Net correctly reclassified (%) †
	Quartile 1 (<12.3%)	Quartile 2 (12.4%–16.2%)	Quartile 3 (16.3%–21.3%)	Quartile 4 (>21.4%)	Increased risk ∗	Decreased risk ∗
Clinical + LV EF
Quartile 1 (<12.3%)	8	3	0	1	16	15	1.2
Quartile 2 (12.4%–16.2%)	2	8	3	1
Quartile 3 (16.3%–21.3%)	2	3	8	8
Quartile 4 (>21.4%)	0	1	7	31

Only gold members can continue reading. Log In or Register to continue

Share this:

• Click to share on Twitter (Opens in new window)
• Click to share on Facebook (Opens in new window)

Related

Related posts:

1. Myocardial Damage Detected by Two-Dimensional Speckle-Tracking Echocardiography in Patients with Extracardiac Sarcoidosis: Comparison with Magnetic Resonance Imaging
2. Echocardiographic Predictors of Progression to Persistent or Permanent Atrial Fibrillation in Patients with Paroxysmal Atrial Fibrillation (E6P Study)
3. Author Index
4. Two Years In Review

Stay updated, free articles. Join our Telegram channel

Join