Comparison of the prognostic values of three calculation methods for echocardiographic relative wall thickness in acute decompensated heart failure

This was a single-center, retrospective, observational study conducted at a Japanese community hospital. In total, 426 consecutive patients admitted due to ADHF through the clinic or emergency room were recruited between June 2014 and April 2016 and followed-up from June 2014 to September 2016. A total of 41 patients were excluded for any of the following reasons: no TTE on admission (n = 35); and RWT not measured (n = 6). Finally, 385 patients were eligible for the analysis (Fig. 1). We previously documented the enrolled patients in detail [5].

The present study followed the tenets of the Declaration of Helsinki and the Ethical Guidelines for Medical and Health Research Involving Human Subjects proposed by the Ministry of Health and Welfare in Japan. The institutional ethics committee at Tomishiro Central Hospital approved the present study and waived informed consent because of the observational nature of the study.

Comprehensive TTE (Vivid 7 ultrasound system, GE Vingmed Ultrasound, Horten, Norway) was performed during hospital admission by four medical technicians who had at least 5 years of experience performing TTE. Their measurements followed established and standardized methods recommended by the ASE and the European Society of Cardiology. At least two attending cardiologists certified by the Japanese Circulation Society and an experienced sonographer reviewed the echocardiography reports immediately after comprehensive TTE. LV geometry, including PWth, IVSth, and LVDd, was measured in M-mode in a parasternal long-axis view [6]. All measurements were performed from the leading edge to the leading edge [6]. RWTs were calculated by the three measurement methods and defined as follows: RWT_PW = 2 × PWth/LVDd; RWT_IVS + PW = (IVSth + PWth)/LVDd; and RWT_IVS = 2 × IVSth/LVDd. The patients were divided into two groups based on the median RWT_PW (low- and high-RWT_PW), median RWT_IVS + PW (low- and high-RWT_IVS + PW), or median RWT_IVS (low- and high-RWT_IVS).

Left ventricular ejection fraction (LVEF) was assessed using the biplane Simpson’s method [6]. Heart failure with preserved ejection fraction (HFpEF) was defined as an ejection fraction ≥50% [9]. LV mass was computed by the Cube formula [6]. LV end-diastolic volume (LVEDVI) was estimated by the Teichholz equation [10]. Peak transmitral early diastolic wave (E wave) velocity, atrial contraction wave (A wave) velocity, and deceleration time (DCT) were measured by the pulse wave Doppler signals of the mitral inflow in the apical four-chamber view [11]. Valvular diseases were evaluated using a semiquantitative 4-grade scale (none, mild, moderate, and severe) [12].

Cardiologists followed the patients at Tomishiro Central Hospital Clinic every 1–3 months after hospital discharge. Medical clerks confirmed the patients’ condition if the patients canceled the appointment.

Patients’ medical charts were reviewed to collect their demographic characteristics and clinical data, including medications, laboratory tests, and hemodynamic data on hospital admission. The primary outcome was all-cause death. Death was confirmed by the medical chart, telephone call with a patient’s family, or obituary in local newspapers.

Continuous variables with normal and skewed distributions are presented as means (SD) and medians [25th, 75th percentiles], respectively. Categorical variables are presented as numbers with a percentage.

During follow-up (235 [92, 425] days), 95/385 (25%) patients died. Survival analysis for all-cause death was performed. Kaplan-Meier curves were stratified by RWT_PW, RWT_IVS + PW, and RWT_IVS. The log-rank test was used to compare survival curves. High-RWT_PW, high-RWT_IVS + PW, and high-RWT_IVS were examined by univariate Cox proportional hazard models and a Cox proportional hazard model adjusted by the Get With The Guideline score (GWTG) [13, 14], an established risk score for mortality in patients with acute heart failure, to obtain hazard ratios (HRs) and 95% confidence intervals (95% CIs).

A total of 48 patients who were lost to follow-up were excluded to evaluate the risk of 90-day mortality. Logistic regression models were used to obtain the odds ratios (ORs) of 90-day mortality and 95% CIs. High-RWT_PW, high-RWT_IVS + PW, and high-RWT_IVS were examined in univariate logistic regression models and a logistic regression model adjusted by GWTG.

Receiver operating curves for 90-day mortality were drawn using RWT_PW, RWT_IVS + PW, and RWT_IVS to obtain c-statistics, and the best RWT cut-off values were determined by the maximum Youden index [15].

To confirm the consistency of the survival analysis, the participants were divided based on the best RWT cut-off value derived from the Youden index.

Survival analysis was performed to compare low and high-RWTs. High-RWT_PW, high-RWT_IVS + PW, and high-RWT_IVS were also examined with univariate and adjusted proportional Cox hazard models.

Spearman’s correlation coefficient (ρ) was used to identify significant associations between RWTs and clinical characteristics: age, the natural logarithm of brain natriuretic peptide (logBNP), LVEF, LVEDV, and systolic blood pressure (SBP).

The reliabilities of the TTE measurements of PWth, IVSth, and LVDd were examined in 25 patients whose TTE image quality was good, and all of the patients underwent TTE performed by the same one of four medical technicians. The medical technician and two other examiners re-measured PWth and IVSth in the TTE image stored in the local server on hospital admission, using an off-line image analysis system (Nahri Aqua, Mehergen Group, Fukuoka, Japan). Comparing every two examiners’ measurements, Bland-Altman plots were used to assess the agreement between the measurement by the same examiner and different examiners [16]. The inter-class coefficient (ICC) was computed to assess agreement [17].

The reliabilities of RWT_PW, RWT_IVS + PW, and RWT_IVS were also examined. RWTs were computed using PWth, IVSth, and LVDd measured by three examiners. Bland-Altman plots were drawn, and the ICC and P values were calculated.

The statistical software used was R 3.4.3 (R Foundation for Statistical Computing, Vienna, Australia). All reported P values are two-tailed, and a P value < 0.05 was considered significant.

The participants’ median age was 81 years, and there were 181/385 (47%) men in the overall population. Comparing low- and high-RWT_PW, high-RWT_PW had more elderly patients and more females, whereas in comparisons between low- and high-RWT_IVS + PW and between low- and high-RWT_IVS, there were no significant differences in baseline characteristics (Table 1).

The mean RWT_PW, RWT_IVS + PW, and RWT_IVS values in the overall population were 0.36 ± 0.12, 0.37 ± 0.13, and 0.38 ± 0.14, respectively.

On comparing the three RWTs (low- vs. high- RWT_PW, RWT_IVS + PW, RWT_IVS), high-RWTs had thicker IVSth and PWth, smaller LVDd, greater LVEF, smaller LV end-diastolic volume, high LVM/LVEDV, and less severe mitral regurgitation than low-RWTs (Table 1).

During follow-up (235 [92, 425] days), 95/385 (25%) patients died in the overall population.

Comparing low- and high-RWT_PW, there was a significant difference in the incidence of all-cause death (low 36/193 (19%) vs. high-RWT_PW 59/192 (31%), P = 0.007). Kaplan-Meier curves showed that high-RWT_PW had worse survival than low-RWT_PW (P for log-rank test = 0.009; Fig. 2a).

Comparing low- and high-RWT_IVS + PW, there was no significant difference in all-cause death (low 40/193 (21%) vs. high-RWT_PW 55/192 (29%), P = 0.077) or survival (P for log-rank test = 0.074; Fig. 2b).

In a comparison between low- and high-RWT_IVS, there was no significant difference in all-cause death (low 42/193 (22%) vs. high-RWT_IVS 53/192 (28%), P incidence = 0.2) or survival (P for log-rank test = 0.19; Fig. 2c).

In the unadjusted and adjusted Cox proportional hazard models, high-RWT_PW was a significant risk factor for all-cause death (unadjusted Cox model, HR (95% CI), 1.72 (1.41–2.61), P = 0.01; adjusted Cox model, 1.95 (1.28–2.98), P = 0.02; Table 2).

High-RWT_IVS + PW was not a significant risk factor for all-cause death in the unadjusted Cox proportional model (unadjusted Cox model, HR, 1.45 (0.96–2.17), P = 0.075), but it was in the adjusted Cox proportional hazard model (adjusted Cox model, 1.53 (1.01–2.32), P = 0.045; Table 2).

High-RWT_IVS was not a significant factor in either the unadjusted or the adjusted Cox proportional hazard model (Table 2).

The OR of high- to low-RWT_PW was significant (univariate, OR, 2.19, 95%CI, 1.15–2.19, P = 0.017; adjusted, OR, 2.26, 95%CI, 1.16–4.4, P = 0.017) on univariate analysis and the adjusted logistic regression model (Table 3). In contrast, the OR of neither high to low-RWT_IVS + PW nor RWT_IVS was significant on univariate analysis or the adjusted logistic regression models.

A total of 48 (13%) patients died within 90 days from hospital admission. Figure 3 shows the receiver operating characteristic (ROC) curves for 90-day mortality using the RWTs. The c-statistic of the ROC curve using RWT_PW was 62.6%, and the best cut-off value was 0.35. The c-statistic of the ROC curve using RWT_IVS + PW was 59.7%, and the best cut-off value was 0.55. The c-statistic of the ROC curve using RWT_IVS was 43.1%, and the best cut-off value was 0.36.

Additional file 1: Table S1 shows the demographic data and echocardiographic data with stratification by the best RWT cut-off. High-RWT_PW had worse survival than low-RWT_PW (P for log-rank test = 0.03; Additional file 2: Figure S1a). High-RWT_IVS + PW also had a worse prognosis than low-RWT_IVS + PW (P for log-rank test < 0.001; Additional file 2: Figure S1b). In contrast, there was no significant difference in survival between low- and high-RWT_IVS (P for log-rank test = 0.077; Additional file 2: Figure S1c).

In the unadjusted and adjusted Cox proportional hazard models, high-RWT_PW and high-RWT_IVS + PW were associated with mortality (high-RWT_PW, unadjusted Cox model, HR (95% CI), 1.55 (1.04–2.33), P = 0.033; adjusted Cox model, 1.72 (1.14–2.59), P = 0.01; high-RWT_IVS + PW, unadjusted Cox model, HR (95% CI), 3.88 (2.34–6.43), P <  0.001; adjusted Cox model, 3.42 (2.04–5.72), P <  0.001; Additional file 3: Table S2). High-RWT_IVS was not a significant risk factor in the unadjusted and adjusted Cox proportional hazard models.

There were significant positive correlations between the three RWTs and age and LVEF, and negative correlations between the RWTs and LogBNP and LVEDV (Table 4). RWT_IVS + PW and RWTI_VS did not have significant correlations with systolic blood pressure, but RWT_PW did (ρ = 0.15, P = 0.004).

Intra-observer agreement of TTE measurement of PWth was significant (ICC = 0.73, P <  0.001; Fig. 4a). Inter-observer agreements of TTE measurement of PWth were also significant (observer 1 vs. 2, ICC = 0.76, P <  0.001; observer 1 vs. 3, ICC = 0.6, P <  0.001; observer 2 vs. 3, ICC = 0.7, P <  0.001; Fig. 3a). There were no systematic biases in the intra- and inter-observer agreements in PWth measurement (Fig. 4a).

Intra-observer agreement of TTE measurement of IVSth was significant (ICC = 0.88, P <  0.001; Fig. 4b). Inter-observer agreements of TTE measurement of IVSth were also significant (observer 1 vs. 2, ICC = 0.81, P <  0.001; observer 1 vs. 3, ICC = 0.77, P <  0.001; observer 2 vs. 3, ICC = 0.73, P <  0.001; Fig. 4b). There were no systematic biases in the intra- and inter-observer agreements in IVSth measurement (Fig. 4b).

Intra-observer agreement of TTE measurement of LVDd was significant (ICC = 0.94, P <  0.001; Fig. 4c). Inter-observer agreements of TTE measurement of LVDd were also significant (observer 1 vs. 2, ICC = 0.71, P <  0.001; observer 1 vs. 3, ICC = 0.92, P <  0.001; observer 2 vs. 3, ICC = 0.65, P <  0.001; Fig. 4c). There were no systematic biases in the intra- and inter-observer agreements in LVDd measurement (Fig. 4c).

Intra-observer agreement of RWT_PW was significant (ICC = 0.77, P <  0.001; Fig. 5a). Inter-observer agreements of RWT_PW were significant (observer 1 vs. 2, ICC = 0.74, P <  0.001; observer 1 vs. 3, ICC = 0.63, P <  0.001; observer 2 vs. 3, ICC = 0.8, P <  0.001). There were no systematic biases in the intra- and inter-observer agreements in RWT_PW.

Intra-observer agreement of RWT_IVS + PW was significant (ICC = 0.89, P <  0.001; Fig. 5b). Inter-observer agreements of RWT_PW were also significant (observer 1 vs. 2, ICC = 0.82, P <  0.001; observer 1 vs. 3, ICC = 0.74, P <  0.001; observer 2 vs. 3, ICC = 0.83, P <  0.001). There were no systematic biases in the intra- and inter-observer agreements in RWT_IVS + PW.

Intra-observer agreement of RWT_IVS was significant (ICC = 0.84, P <  0.001; Fig. 5c). Inter-observer agreements of RWT_IVS were also significant (observer 1 vs. 2, ICC = 0.77, P <  0.001; observer 1 vs. 3, ICC = 0.75, P <  0.001; observer 2 vs. 3, ICC = 0.72, P <  0.001). There were no systematic biases in the intra- and inter-observer agreements in RWT_IVS.

High-RWT_PW was associated with a poor prognosis on survival analysis and Cox proportional hazard models (Fig. 2a; Table 2). High-RWT_IVS + PW was not associated with poor survival on survival analysis (Fig. 1b), whereas high-RWT_IVS + PW was a significant risk only in the Cox proportional hazard model adjusted by GWTG, not in the unadjusted model (Table 2). High-RWT_IVS did not show worse survival than low-RWT_IVS (Fig. 1c; Table 2). The equations of RWT_PW and RWT_IVS + PW contain PWth. PWth or the ratio of PWth to LVDd, therefore, may represent the LV remodeling related to a worse prognosis better than IVSth or IVSth to LVDd in patients with ADHF. Patients with high-RWT_PW had higher systolic blood pressure than those with low-RWT_PW (Table 1), while there was no such difference either between low- and high-RWT_IVS + PW or between low- and high-RWT_IVS. RWT_PW had a positive correlation with systolic blood pressure (Table 4), while either RWT_IVS + PW or RWT_IVS did not. This may suggest that thickening of PWth, rather than IVSth, is likely to counterbalance pressure overload and may lead to LV diastolic dysfunction leading to a poor prognosis. A higher A wave in high RWT_pw patients than in low RWT_pw patients may support this assumption (Table 1).

In terms of methodological validity, there were no differences in inter- and intra-observer agreements for each RWT. Given that fairly good reproducibility was observed in all measurements, differences in prognostic values among the three RWTs may not result from technical aspects of TTE.

Paradoxically, high-RWT_PW patients had lower BNP than low-RWT_PW patients (Table 1). High-RWT_PW included 101 (53%) patients with HFpEF. Generally, BNP increases modestly in HFpEF [18]. Furthermore, the prognostic value of BNP has not been confirmed in patients with HFpEF [19]. High RWT_PW might be of clinically utility, especially, in patients with HFpEF.