Methods
Search Strategy
We conducted a search strategy for this study in adherence to the PRISMA (preferred reporting items for systematic reviews and meta-analyses) statement for reporting systematic reviews and meta-analysis. This literature search of PubMed, EBSCO HOST, and EMBASE via Medline was performed using the key words 'strain', 'speckle tracking', 'mortality', 'cardiovascular events', 'survival' outcome', and 'prognosis'. We limited the paper search to English language and human studies. Reference lists were thoroughly searched for additional studies. We contacted the authors wherever study data were incomplete.
Inclusion and Exclusion Criteria
We included studies in which GLS was evaluated as a predictor of mortality and major adverse cardiac events. Due to the limitations of Doppler based strain, we only included studies that calculated strain by 2D strain. Due to paucity of studies analysing global circumferential strain and wall motion score index (WMSI), we excluded these modalities from the final draft. We excluded studies where strain analysis was restricted to the RV.
Endpoints
The studies were included only if their primary and secondary endpoints matched the specified inclusion criteria. The primary outcome was all cause mortality. Secondary outcomes included the composite of cardiac death, HF hospitalisation, and malignant arrhythmias (sustained ventricular tachycardia, ventricular fibrillation, and recorded automatic implantable cardioverter defibrillator (AICD) shocks).
Data Extraction
Data were independently extracted and reviewed by two of the investigators (KK and THM). All discrepancies were mutually reviewed and resolved by consensus. For systematic review, we extracted the following population data: study design, months of follow-up, age, sex, number, primary cardiac condition, outcomes, GLS, LVEF, and clinical variables.
Echocardiography
All the patients in the studies underwent standard echocardiographic assessment by experienced physicians or sonographers using commercially available ultrasound systems. LVEF was measured using the biplane Simpson's approach and LV mass was mostly calculated from LV linear dimension in the parasternal long axis view. LA volume index was calculated by the area length method. Diastolic function and LV filling pressures were investigated by assessment of transmitral flow. Colour and spectral Doppler were used to assess valvular regurgitation and stenosis.
GLS
Most studies recorded strain from standard 2D grey scale recordings at a frame rate >50/s. GLS was calculated using a variety of proprietary software (EchoPAC, GE Medical Systems, Milwaukee, Wisconsin, USA; Syngo velocity vector imaging, Siemens, Mountain View, California, USA; LV analysis, TomTec GmbH, Unterschlessheim, Germany). The common steps were definition of the systolic ejection phase, allocation of fiducial points, outlining a region of interest, and automatic tracking of speckles from frame-to-frame in the apical four-chamber, twochamber, and long axis views. The final results of GLS were displayed on a 16-segment polar map with segmental strain values and a mean GLS (figure 1).
(Enlarge Image)
Figure 1.
Calculation of regional and global strain in a normal (A) and abnormal ventricle (B). Normal strain is characterised by uniform waveforms of comparable timing, uniform colour on the polar map display, and a global strain of approximately −20%. Impaired strain often shows variation in the timing and magnitude of contraction, with a reduction of global strain, −12% being analogous to an EF<35%.
Statistical Analysis
Pooled HRs and 95% CIs were computed using random effect models of the association of GLS and LVEF with mortality and the composite endpoints. We only included studies that had estimated HRs for both GLS and LVEF to the same endpoint. Some analyses showed very low heterogeneity; however, for consistency and due to the diverse nature of the populations and adjustment factors, a random effects method was chosen for all pooled analyses. Most studies reported HR effect sizes relative to a 1% increase in GLS and LVEF, but some studies reported effects relative to larger changes in GLS or LVEF. To allow comparison of GLS and LVEF, all studies were rescaled by the within-study SD to represent a standardised change in the absolute value of each parameter (either GLS or LVEF) before pooling. Scaled in this way, each SD change represents a reduction in hazard. We used bivariate meta-regression to compare directly the effect sizes for GLS and LVEF and examine heterogeneity. Within-study correlations were unknown for these bivariate meta-regressions and we therefore used the Riley method for estimation. In one model that did not converge, we used a within-study correlation estimate of 0.5. We report the absolute strength of GLS relative to LVEF when directly comparing the two effects; a value >1 indicates that a change of 1 SD in GLS is associated with a greater hazard reduction relative to 1 SD change in LVEF. Cochrane's Q test for heterogeneity between the studies and the I measure of inconsistency is reported. Heterogeneity was explored from proportions of sample characteristics within studies, such as female gender, and diagnoses of hypertension, diabetes mellitus, and dyslipidaemia, as well as for four study subgroups according to the underlying cardiac condition: HF, acute coronary syndrome (ACS), VHD, and other miscellaneous cardiac conditions (MC). Begg's funnel plot, Egger's test, and Duval and Tweedie's trim and fill were used to examine publication bias. Forest plots were constructed to show the overall effect of GLS and LVEF. Statistical analysis was performed using Stata V.12.1 (Stata Statistical Software, release 12, StataCorp 2011, College Station, Texas, USA) with two-tailed p values; a value of p<0.05 was considered significant.