Diabetes mellitus is an independent prognostic factor for mid-term and long-term survival following transcatheter aortic valve implantation: a systematic review and meta-analysis

Summary

Prior studies have proposed the unfavourable roles of diabetes mellitus (DM) in surgical populations. For patients who underwent transcatheter aortic valve implantation, the prognostic value of DM remains controversial. This review summarizes the effects of DM on short-term, mid-term and long-term prognosis in patients who undergo transcatheter aortic valve implantation. We searched the PubMed database to identify eligible articles. The odds ratio and hazard ratio with the corresponding 95% confidence interval were adopted for synthesizing short-term and medium- to long-term survival outcomes, respectively. The heterogeneity level and publication bias between studies were also estimated. Finally, 20 observational studies enrolling 19 260 patients met the eligibility criteria and, thus, were included in this review. An overall analysis identified that DM was significantly associated with the poor medium- to long-term overall survival (hazard ratio 1.21, 95% confidence interval 1.03–1.41; P = 0.019). However, no significant impact of DM on 30-day mortality was observed (odds ratio 1.10, 95% confidence interval 0.86–1.41; P = 0.46) in patients undergoing transcatheter aortic valve implantation. Further subgroup analyses indicated that the prognostic value of DM for medium- to long-term overall survival remained significant in the subgroups of multivariable origins of incorporated data, duration of follow-ups (1-year/≥2-year follow-up), Western populations and insulin-dependent diabetes mellitus. This meta-analysis demonstrates that DM is a strongly independent predictor for poor medium- to long-term overall survival but shows no significant effect of DM on 30-day mortality. Our findings need to be further verified and modified by more worldwide studies.

INTRODUCTION

Rationale

Diabetes mellitus (DM) remains a major challenge to public health around the world, and its prevalence has dramatically increased over the last 3 decades [1]. According to official estimates, approximately 415 million adult people worldwide had DM in 2015, accounting for 8.8% of the adult population. Moreover, the number of patients with DM is predicted to continue to rise, reaching up to 642 million people in 2040 [2]. DM is generally considered an important cause of cardiovascular deaths and some severe complications, such as heart failure, coronary heart disease and renal insufficiency, which decrease life expectancy and increase health-care costs [3, 4]. Therefore, the increasing burdens of DM have been treated with caution in clinical practices.

Aortic stenosis (AS) is the most common valvular heart disease in Western developed countries. Its unfavourable effects on the quality of life and the mortality rate and its economic burden are expected to increase with the aging of Western populations [5]. In the past, surgical aortic valve replacement (SAVR) was regarded as the standard therapeutic option for AS [6]. However, elderly patients who are considered at high surgical risk were prohibited from undergoing SAVR because of their age and their comorbidities. In recent years, transcatheter aortic valve implantation (TAVI) has emerged as a minimally invasive transcatheter procedure to gain access to the aortic valve without opening the chest. Current evidence indicates that TAVI can be a safe and efficient alternative for elderly high-risk patients who are not candidates for SAVR. In addition, with the improvement of transcatheter techniques, TAVI indications are still expanding [7, 8].

Both DM and AS are common comorbidities in elderly populations. Besides, DM can accelerate the progression of sclera-calcific aortic valve diseases [9]. Previous studies had discovered significantly higher rates of mortality and morbidity in patients with DM undergoing SAVR. On the basis of the Society of Thoracic Surgeons’ database, DM was used as a reliable factor for predicting the poor prognosis of cardiac operations [10]. However, the findings on the prognostic value of DM for patients undergoing TAVI procedures are limited and controversial. The relationships between DM and post-TAVI survival outcomes are still not clear.

Objectives

In view of the foregoing issues, the objective of our study was to estimate the prognostic role of DM for in-hospital, mid-term and long-term survival rates of patients undergoing TAVI by performing a systematic review with meta-analysis.

MATERIALS AND METHODS

Protocol

The present study was conducted in compliance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) statement [11]. An additional PRISMA 2009 checklist is included as Supplementary Material, Table S1.

Eligibility criteria

Publications were included in this meta-analysis if they met the following eligibility criteria:

Study designs

Articles utilizing a quantitative comparative analysis of consecutive patients were eligible. Case studies, preclinical experiments, reviews, conference abstracts and letters were excluded.

Participants

The target disease was limited to AS. No specific restriction was imposed for patients’ basic characteristics, such as age, gender and other cardiovascular comorbidities.

Interventions

The TAVI operation was the only eligible treatment, no matter what approaches were used in current studies.

End points and outcome data

The outcomes of interest in our study were short-term and medium- to long-term survival rates of patients who underwent TAVI. We regarded the 30-day mortality rate as the primary end point for short-term survival, which was defined as any death within 30 days after TAVI. Overall survival (OS) with a follow-up ≥1 year was regarded as the primary end point for medium- to long-term survival. In further subgroup analysis, to be more specific, the 1-year OS was used as the mid-term survival rate, and the OS with a ≥2-year follow-up served as the long-term survival rate of patients who had TAVI.

Studies reporting any of the following outcome data were included in this meta-analysis. First, sufficient demographics or dichotomous statistics that were derived from the multivariable analysis should be available for the estimate of odds ratio (OR), relative risk or hazard ratio (HR) to reveal the prognostic roles of DM in TAVI. Second, the survival data with a log-rank P-value and Kaplan–Meier survival curves were also eligible.

Publication

Finally, the most recent studies were included if they were performed on overlapping patients. Only full-text papers published in English peer-reviewed journals were considered.

Search strategy

Three of our researchers were assigned to search the PubMed electronic database independently between 1 March 2017 and 15 March 2017, in order to identify the eligible articles available as of 1 March 2017.

The following 4 keywords, including 2 ‘TAVI’ terms and 2 ‘diabetes’ terms, were combined with 2 Boolean operators (‘AND’ and ‘OR’) to formulate 2 search strings for identification of eligible articles. These terms are listed as follows:

• ‘TAVI’-terms: ‘aortic valve implantation’, ‘aortic valve replacement’;

• ‘Diabetes’-terms: ‘diabetes mellitus’ and ‘diabetes’.

The search details are presented in Supplementary Material, Table S2. In addition, we also performed a manual search on the reference lists of retrieved publications to identify any possible study with no duplication.

Data extraction

We established a prespecified table in an Microsoft Office Excel spreadsheet and collected the following data items from each included study:

• Publication data including the authors, publication year and nation;

• Experimental data including the study design, study period, TAVI procedures, follow-ups and outcome measures;

• Demographic data including the total sample size, the number of patients with DM and other major baseline characteristics;

• Statistical data including the outcome statistics with their extractions, statistical analysis methods and adjusted confounding factors.

Quality assessment

We used the Newcastle–Ottawa Scale (NOS) to assess the methodological quality of non-randomized studies [12]. A semiquantitative estimation based on 3 perspectives of parameters including selection, comparability and exposure was performed. The ‘star system’ with a maximum number of 9 stars was used to grade all the included studies. We regarded 8–9 stars as good quality, 6–7 stars as fair quality and <6 stars as poor quality.

Statistical analysis

Summary measures

To assess the impact of DM on 30-day mortality, the OR with a 95% confidence interval (CI) was used as the estimate to be pooled for quantitative synthesis. It was our priority to integrate the OR outcomes derived from multivariable analysis because of the adequate elimination of confounding factors. Besides, we could also extrapolate the ORs from reported demographics in the literature if no multivariable statistics were included.

The HR with a 95% CI served as the summary estimate for mid-term and long-term OS because the HR was the only appropriate statistic compatible for both censoring and time-to-events values [13]. Similarly, incorporating the multivariable HR statistics was our first priority. If no multivariable statistics were reported, we could extrapolate the HRs from published survival data with their log-rank P-value in accordance with a practical method reported by Tierney et al. [14].

Synthesis of results

The heterogeneity level between studies was graded using Cochrane’s Q-test and the I²-statistic. Fine homogeneity was defined as I² < 50% and P > 0.1, suggesting that a standard fixed-effect model test would be used for synthesis of results. Otherwise, a random-effect model test would be required when significant heterogeneity was revealed if I² ≥ 50% or P ≤ 0.1 [15].

Additional analysis

In addition, we carried out a sensitivity analysis in which the impact of each included study on the overall results could be detected by omitting the individual study sequentially; thus the stability of the overall results was examined. The robustness of this meta-analysis was confirmed if there was no substantial variation between adjusted pooled estimates and primary pooled estimates [15].

Publication bias

Begg’s test was used to detect the potential publication bias within this meta-analysis. The presence of bias would be suggested by the visual asymmetry of Begg’s funnel plot, in which log ORs or log HRs were plotted against their corresponding standard errors [16].

All of the above statistical analyses were performed using Stata 12.0.

RESULTS

Study selection

The process of literature retrieval is illustrated in Fig. 1. In total, 467 citations were identified after searching PubMed, and 5 additional studies were found from other sources. After an initial screening of titles and abstracts, 82 papers were immediately excluded due to their inappropriate styles (Fig. 1). After carefully reading through the remaining 390 articles, we excluded 370 because they were related to irrelevant issues or were conducted on overlapping patients. Finally, 20 studies met all the eligibility criteria and were included in this meta-analysis [17–36].

Study characteristics

The baseline characteristics of the 20 included studies are summarized in Table 1.

Study designs

All of these studies were observational studies, including 1 prospective study [25] and 19 retrospective studies [17–24, 26–36]. They were published between 2010 and 2016. Sample sizes ranged from 30 to 3813. Three studies [17, 18, 20] were designed to evaluate the effects of DM on the survival of patients who underwent TAVI. Eight [22, 28–31, 34–36] of 20 included studies were designed to identify significant prognostic factors for TAVI. Five studies [23, 25–27, 33] investigated the prognostic roles of an individual clinical parameter in TAVI. The remaining 4 studies [19, 21, 24, 32] were designed to establish a scoring system for predicting the short-term or medium- to long-term survival after TAVI.

Participants and interventions

A total of 19 260 patients who underwent TAVI were enrolled in our meta-analysis, including 14 452 patients from Europe (75.0%), 2522 patients from North America (13.1%) and 2286 patients from Asia (11.9%). They were considered at high surgical risk according to their mean logistic EuroSCORE (20.5%). The prevalence of DM ranged from 20.0% to 50.8% across these studies [21, 25]. There were 5551 patients diagnosed with DM, with an overall prevalence rate of 28.8%. The other available demographics in each study, including the ages, genders and other preoperative comorbidities, are presented in Table 1. Only 3 of the 20 studies detailed the baseline characteristics of patients with and without DM separately [17, 18, 20]. The pooled baseline characteristics based on 3 available studies showed that the patients with DM were younger but were more likely to have hypertension, coronary artery disease and a higher Society of Thoracic Surgeons score than the patients without DM.

Two studies reported the post-TAVI outcomes in 439 patients receiving a transfemoral TAVI approach [25, 26]. The remaining 18 studies reported patients treated with a variety of TAVI approaches but analysed their outcome data as a whole [17–24, 27–36].

Outcome measures

Eighteen studies directly reported 28 multivariable or univariable statistics revealing the survival outcomes of TAVI, including 6 ORs for 30-day mortality [19, 22, 24, 28, 34], 15 HRs for mid-term OS [17, 21–23, 25, 27–29, 32, 33] and 7 HRs for long-term OS [18, 26, 28, 31, 34–36]. The follow-up periods for long-term OS ranged from 2 to 5 years.

Risk of bias within studies

The quality level of each study was graded by an NOS score and then represented by the appropriate number of stars (see Supplementary Material, Table S3). Finally, their mean NOS score was 8.1 (range 7–9), suggesting they were of fairly good quality.

Overall analysis

For 30-day mortality, a random-effect model was applied to process the data because of high heterogeneity (I² = 38.5%, P = 0.070). Finally, the pooled analysis showed no significant relationship between DM and 30-day mortality of TAVI (OR 1.10, 95% CI 0.86–1.41; P = 0.46; Table 2; Fig. 2A) [17–20, 22, 24, 28–30, 34].

For medium- to long-term OS, a random-effect model was also adopted because of the significant heterogeneity (I² = 60.9%, P = 0.001). The pooled data demonstrated that DM was significantly associated with the lower OS of TAVI (HR 1.21, 95% CI 1.03–1.41; P = 0.019; Table 2; Fig. 2B) [17, 18, 22, 23, 25–29, 31–34, 36].

Subgroup analysis

To further assess the effect of DM on the prognosis of TAVI in detail, we divided the data regarding the medium- to long-term OS into different subgroups according to the follow-up duration, statistical analyses, origins of patients and subtypes of DM, as shown in Table 2.

In the subgroups classified by length of follow-up, the pooled HR estimates based on 12 studies for 1-year OS (HR 1.19, 95% CI 1.03–1.38; P = 0.022; I² = 49.7%, P = 0.007) [17, 18, 20–23, 25, 27–29, 32, 33] and 7 studies for long-term OS (≥2-year follow-up) (HR 1.16, 95% CI 1.03–1.31; P = 0.013; I² = 32.4%, P = 0.18) [18, 26, 28, 31, 34–36] showed that DM could be predictive of both poor mid-term and long-term survival of patients undergoing TAVI.

In the subgroups classified by statistical analyses, a total of 10 studies provided the outcome data derived from multivariable analysis [17, 18, 23, 26, 28, 29, 31, 32, 34, 36]. The pooled HR of multivariable data indicated that DM was significantly associated with a lower medium- to long-term OS (HR 1.24, 95% CI 1.06–1.46; P = 0.006; I² = 54.3%, P = 0.016). Another 8 included studies that reported the univariable demographics or statistics and in which the pooled estimates showed no statistical significance but a tendency towards a poorer OS in patients with DM (HR 1.10, 95% CI 0.93–1.29; P = 0.25; I² = 44.1%, P = 0.044) [18, 20–22, 25, 27, 28, 33].

In the subgroups classified by populations, 3 studies were carried out on Asian patients [18, 26, 36] and 11 studies, on Western patients [17, 22, 23, 25, 27–29, 31–34], respectively. The summarized estimates showed a significantly lower medium- to long-term OS in the Western patients with DM (HR 1.22, 95% CI 1.02–1.46; P = 0.030; I² = 65.0%, P = 0.001). However, no significant relationship was observed between DM and the prognosis of TAVI in Asian populations (HR 1.22, 95% CI 0.93–1.58; P = 0.15; I² = 44.3%, P = 0.166).

In the subgroups classified by subtypes of DM, 4 studies based on 3077 patients evaluated the prognostic value of insulin-dependent diabetes mellitus (IDDM) [17, 18, 20, 21], and the integrated results showed that IDDM was significantly associated with a lower OS of patients undergoing TAVI (HR 1.53, 95% CI 1.10–2.13; P = 0.011; I² = 13.5%, P = 0.328). Likewise, 4 studies individually reported the medium- to long-term OS of 3077 patients with non-insulin-dependent diabetes mellitus (NIDDM) [17, 18, 20, 21]. Further pooled analyses revealed no significant impact of NIDDM on the prognosis of patients undergoing TAVI (HR 1.18, 95% CI 0.94–1.48; P = 0.16; I² = 0.0%, P = 0.51).

Sensitivity analysis

We performed a sensitivity analysis by omitting the individual studies one at a time to test the robustness of our findings. We found that the sequential removal of each study did not alter the results of primary overall analyses, as shown in Fig. 3A and B. The robustness of our meta-analysis was thus confirmed.

Publication bias

Begg’s funnel plots estimating the publication bias between the included studies are shown in Fig. 4A and B. Finally, a symmetrical appearance was observed in both Fig. 4A (P = 0.44) and B (P = 0.053), showing that no significant evidence for publication bias was detected within the present meta-analysis.

DISCUSSION

Summary of evidence

As previous large-scale registries reported, more than 30% of patients with AS who were elderly and considered at high surgical risk had DM. Such a high proportion of DM in patients with AS may be explained by the following 2 reasons. On the one hand, both DM and AS develop linearly as one ages. On the other hand, DM can accelerate the sclera-calcific process of the aortic valve, resulting in the formation of AS. As a minimally invasive alternative therapy, TAVI has played an increasingly important role in the treatment for AS among high-risk elderly patients. Many large-scale trials have confirmed the safety and efficacy of TAVI and concluded that TAVI could be an alternative to conventional SAVR in a well-chosen, high-risk subgroup of patients with AS. However, as an important comorbidity for patients with AS, the role of DM in determining the prognosis for patients undergoing TAVI remains controversial because previous studies showed inconsistent results that have not yet been definitively explained.

Taking the current evidence together, we got the initial impression that most authors reported a more favourable medium- to long-term OS in patients without DM undergoing TAVI than in those with DM. The largest current study, based on 3813 patients undergoing TAVI from the UK registry, demonstrated that DM was an excellent predictor of poor mid-term survival following the TAVI procedure [28]. This conclusion was supported by Tamburino et al. [34] in their small study of 663 patients undergoing TAVI. In addition, those authors also found that DM was significantly associated with the long-term mortality of TAVI. However, 1 large multicentre analysis of the Italian Transcatheter Balloon-Expandable Valve Implantation Registry, which included 1904 patients, concluded that DM had no significant impact on the long-term survival according to multivariable analysis [31]. In another Italian multicentre registry, Conrotto et al. [37] prospectively analysed the clinical data of 511 consecutive patients undergoing TAVI in different subgroups stratified by subtypes of DM. They found that insulin-treated DM was significantly associated with the poor mid-term survival of patients undergoing TAVI but orally treated DM was not. These findings agreed with what Abramowitz et al. [17] reported in their large retrospective study based on 802 patients undergoing TAVI, showing that only insulin-treated DM had significant adverse effects on the mid-term prognosis of patients undergoing TAVI. Only Tamburino et al. [34] found significantly lower 30-day mortality after the TAVI procedure in the patients with DM in a cohort of 663 patients. More than half of the remaining studies just showed a common tendency towards higher 30-day mortality in the patients with DM but the results were without statistical significance.

Therefore, we proposed that a key point to be addressed was whether the relationship between DM and prognosis of TAVI was statistically reliable. Meta-analysis is a well-designed statistical method that quantitatively integrates the appropriate estimates from homogeneous studies to get a conclusion. By applying evidence-based methods to a large sample size, a quantitative synthesis may help to clarify this issue [13]. By synthesizing the current evidence, we found that DM could be an independent predictor of poor medium- to long-term OS for patients undergoing TAVI. Such a significant association did not change across the subgroup analyses of multivariable data, length of follow-up, Western populations and IDDM. However, no significant impact of DM was observed on 30-day mortality. Additional analyses, including sensitivity analysis and publication bias tests, were also performed to confirm the robustness and accuracy of pooled estimates.

The possible reasons for the unfavourable prognostic value of DM in patients treated with TAVI for AS are not fully understood. We speculated that this phenomenon might be explained by the following pathophysiological mechanisms.

Firstly, DM plays an important role in accelerating the degeneration of the aortic valve, showing a close relationship with calcification of the aortic valve [9]. Evidence indicates that an increasing amount of aortic valve calcium correlates closely with the post-TAVI paravalvular regurgitation because calcific clumps hinder the sufficient apposition of the prosthesis to the native annulus [38]. Ultimately, above-moderate regurgitation can easily cause a large decrease in diastole time and pressure, which are essential to myocardial perfusion, and then further impair the delivery of oxygen to the myocardium and aggravate myocardial ischaemia, resulting in deterioration of left ventricular functions [39]. Therefore, cardiovascular mortality may be increased in patients with DM undergoing TAVI.

Secondly, prior studies have demonstrated that DM may predispose patients with AS to hypertrophic remodelling and left ventricular diastole dysfunction. The major causes of these conditions may be severe myocardial fibrosis, advanced deposition of glycation end products adhering to the intramyocardial vessels and high resting tension of isolated stretched cardiomyocytes [40]. Therefore, the DM-induced myocardial dysfunction may serve as another possible factor for increased medium- to long-term mortality.

Finally, because of advances in device profiles and vascular access sheaths, the incidence of major vascular and bleeding complications has decreased dramatically in recent years [41]. Evidence from previous studies showed a close relationship between IDDM and these vascular complications [42]. We speculate that the heavy burden of these post-TAVI morbidities may significantly worsen the survival outcomes of patients with DM.

The subgroup analyses of follow-up data indicated that DM was significantly associated with poor mid-term and long-term OS following the TAVI procedure. However, in our results, no significant effect of DM was observed on 30-day mortality, which means that the predictive value of DM for 30-day mortality may be far less than for medium- to long-term OS. We suspected that the following 2 reasons might explain such differences in the prognostic roles of DM. First, patients with DM could achieve better glycaemic control and could be given more aggressive treatment of underlying DM-related disorders in the hospital than when they were discharged. Thus, the development of lethal events was more likely to be sufficiently controlled during hospitalization. Another interpretation suggested that the effects of DM on the sclera-calcific process of AS and ventricular remoulding might need a certain period to appear after TAVI intervention. Thus, DM was found to have adverse effects on the medium- to long-term OS instead of on 30-day mortality.

In the subgroup analyses based on DM subtypes, we found that IDDM was an independent predictor for poor medium- to long-term OS but NIDDM was not. Similarly, the patients with IDDM undergoing TAVI showed a tendency towards higher 30-day mortality, although the relationship was not statistically significant, but patients with NIDDM did not. These findings were in agreement with those reported by Abramowitz et al. [17] and Conrotto et al. [37], which might be explained by the following perspectives. In general, insulin therapy was given to patients suffering from a late course of DM when optimal glycaemic control could not be achieved with 2 or more oral hypoglycaemic agents [43]. Therefore, IDDM could be considered a more advanced stage of DM, predisposing the individual to a variety of complications that had unfavourable effects on the medium- to long-term OS following TAVI. In addition, evidence indicates that insulin itself can directly facilitate atherosclerosis [44], which impairs the perfusion of many essential organs, thereby influencing the prognosis of patients with IDDM.

Our study demonstrated that DM was significantly associated with poor medium- to long-term OS in Western patients undergoing TAVI but not in Asian patients undergoing TAVI. Prior studies have recognized the difference in DM-susceptible gene expression between various ethnic groups [45]. Asian people appear to be less sensitive to insulin than Western people and thus have an ethnic propensity to DM-related complications even when they have an equal body mass index [46]. These discoveries suggest that Asian patients may have a higher probability of experiencing DM-induced complications than Western patients. However, this phenomenon was not supported by our findings in the subgroup analysis of the 2 populations. We suspected that the limited number of studies included in this subgroup analysis was likely to result in a weakened persuasive power of their pooled estimates. Regardless, the potential ethnic differences in the prognostic roles of DM on the prognosis of TAVI need to be further clarified in future studies.

Limitations

Several major limitations of the present study merit further consideration.

Firstly, no randomized controlled trials were included in our meta-analysis because of the intrinsic limitations of the topic itself: Grouping patients randomly according to their diabetes status was not possible. Thus, the strength of the evidence in our meta-analysis was attenuated by the inherent nature of observational studies, which are subject to recall bias and confounding factors. Unfortunately, due to the scarcity of raw data, we could not sufficiently evaluate the bias risks caused by potential confounders. Furthermore, the baseline characteristics of the DM and non-DM groups could not be compared completely.

Secondly, most of the studies analysed the clinical data of NIDDM and IDDM as a whole [19, 22–36]. Thus, we did not have enough eligible studies to perform a subgroup analysis of diabetic status. For populations with diabetes, different levels of glycaemic control can result in different comorbidities related to DM. Several scoring systems, such as those proposed by Charlson et al. [47] and Elixhauser et al. [48], have been widely used in clinical practices to measure these comorbidities. These methods place a major emphasis on the severity of DM and consider it a principal risk factor for mortality. However, there was a scarcity of details concerning DM conditions in the studies we identified, including the duration of DM status and glycaemic control. We could not rule out the fact that the glycaemic level might be more closely associated with the prognosis of TAVI.

Thirdly, studies reporting positive results are more likely to be published. Potential publication bias could not be easily ignored, although no significant publication bias was observed after funnel plot inspection.

Finally, only English articles were included, causing some selection bias. More eligible studies may be identified by searching more databases in other languages.

CONCLUSION

DM was a strongly independent predictor for unfavourable medium- to long-term OS of patients undergoing TAVI. Such prognostic roles of DM were not substantially altered across the subgroups of multivariable data, duration of follow-up, Western populations and IDDM. However, no significant association was observed between DM and 30-day mortality of TAVI. Several major limitations may weaken the validity of these conclusions in different clinical settings. Our findings need to be further verified and modified by more worldwide studies in the future.

SUPPLEMENTARY MATERIAL

Supplementary material is available at ICVTS online.

ACKNOWLEDGEMENTS

We thank Jing Liu, from the Institution of Medical Statistics, West China School of Public Health, Sichuan University, Chengdu, China, for her help proofreading the statistical analyses reported in this manuscript. We thank Stanley Crawford from the Institution of Medical English, West China Medical Center, Sichuan University, Chengdu, China, for his help with the English language editing of this manuscript.

Funding

This work was supported by the National Natural Science Foundation of China [81370219 and 81400267].

Conflict of interest: none declared.

REFERENCES

Author notes