Introduction

Cancer is the leading cause of death in economically developed countries and the second leading cause of death in developing countries1. In the United States, a total of 1,660,290 new cancer cases and 580,350 cancer deaths were projected to occur in 20132. In Europe, there were an estimated 3.45 million new cases of cancer (excluding non-melanoma skin cancer) and 1.75 million deaths from cancer in 20123. Furthermore, the global burden of cancer continues to increase, largely because of population growth and increased life-expectancy3. Invasion and metastasis are two important hallmarks of cancer and are responsible for the majority of cancer deaths4. Although much effort has been devoted to the diagnosis and therapy of cancers, the overall prognosis is still unsatisfactory. A lack of knowledge of molecular biomarkers in cancer has limited the development of personalized therapies and improvements in survival. Therefore, there is an urgent need for universal, effective, readily available and inexpensive biomarkers in solid tumors to identify patients with a poor prognosis so that novel treatments can be initiated earlier.

The metabolism of cancer cells differs from that of normal cells. This is largely because cancer cells exhibit metabolic alterations that are frequently associated with reprogramming. Unlike normal cells, cancer cells preferentially metabolize glucose by glycolysis to generate sufficient energy for the demands of rapid proliferation, even in the presence of adequate oxygen5.This phenomenon is known as the Warburg effect and is one of the predominant metabolicalterations that occur during malignant transformation. In this process, transcriptional programs regulated by oncogenes stabilize hypoxia-inducible factor 1 alpha (HIF-1α). HIF-1α contributes to the upregulation of most enzymes involved in the glycolytic pathway, including lactate dehydrogenase (LDH).In the final step of aerobic glycolysis, LDH converts pyruvate tolactate, which is coupled with the oxidation of NADH to NAD+. These metabolic changes are reflected by an elevated serum LDH level6(hereinafter LDH).

Elevated LDH has been recognized as a poor prognostic indicator in cancer for many years7,8,9,10. LDH has also been incorporated in prognostic scores for several types of cancer11. However, the consistency and magnitude of the prognostic impact of LDH are unclear12,13,14. The aim of this study was to review published studies and use standard meta-analytic techniques to quantify the prognostic value of LDH in various solid tumors.

Methods

Data sources and searches

This analysis was conducted in line with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses guidelines15. PubMed was searched for studies evaluating the LDH and survival in solid tumors from 1978 to 2014. We used various medical subject heading terms, including “l-lactate dehydrogenase”, “prognosis”, “multivariate analysis” and “proportional hazard model”. Title/abstract words included “lactate dehydrogenase”, “LDH”, “prognosis”, “prognose”, “prognostic”, “multivariate analysis”, “proportional hazard model”, “COX proportional hazard model” and “COX models”. The full search strategy is described in the Supplementary Methods (available online).

Study selection

Inclusion criteria for the primary analysis were as follows: 1) studies of people with solid tumors reporting on the prognostic impact of LDH; 2) prospective or retrospective cohort design with a clearly defined source population and justifications for all excluded eligible cases; 3) sample size greater than 200; 4)statistical analysis using multivariate proportional hazards modeling that adjusted for clinical prognostic factors; and 5) reporting of the resultant adjusted hazard ratios (HRs) and their 95% confidence intervals (CIs) or a P value for overall survival (OS). For the secondary analyses, studies providing a HR for cancer-specific survival (CSS), progression-free survival (PFS), disease-free survival (DFS), or recurrence-free survival (RFS) were included as well.

Data extraction

OS was the primary outcome of interest. CSS, PFS and DFS were secondary outcomes. Two authors (J.Z. and H.W.) independently extracted information using predefined data abstraction forms. The following details were extracted: name of first author, year of publication, number of patients included in analysis, disease site, disease stage (non-metastatic, metastatic, mixed [both non-metastatic and metastatic]), study type (prospective or retrospective), cutoff defining high LDH and HRs and associated 95% confidence intervals for OS, PFS, DFS, or RFS as applicable. HRs were extracted preferentially from multivariate analyses where available. Where several HR values were given in an article, the value adjusted for most confounders was used.

Data synthesis

The meta-analysis was conducted initially for all included studies for each of the endpoints of interest. Subgroup analyses were conducted for predefined parameters such as disease site, disease stage and LDH cutoff and all data were limited to multivariate analyses. Disease site subgroups were generated if at least three studies on that site were available; the remaining studies were pooled in a subgroup termed “other.” LDH cutoff subgroups were < 250 U/L, 250–300 U/L, 301–400 U/L and >400 U/L. In three studies, the effect of LDH was reported as a continuous variable; we pooled those studies separately. Univariate meta-regression model analysis was performed to evaluate the relationship between covariates (LDH cutoff) and the HR for OS.

Statistical analyses

The meta-analysis was performed with RevMan 5.2 analysis software (Cochrane Collaboration, Copenhagen, Denmark). Estimates of HRs were weighted and pooled using the generic inverse-variance and random-effect model16. Analyses were conducted for all studies and differences between the subgroups were assessed using methods described by Deeks et al.17. Publication bias was assessed by visual inspection of the funnel plot. Heterogeneity was assessed using Cochran Q and I2 statistics. Meta-regression analysis was conducted using Stata12.0 software. All statistical tests were two-sided and statistical significance was defined as P less than 0.05. No correction was made for multiple testing.

Results

Description of studies

Sixty-eight studies were included in the meta-analysis. The selection process for the systematic review is shown in Figure S1 and the characteristics of the included studies are shown in Table 1. A total of 31,857 patients were included and the median trial sample size was 363.

Table 1 Baseline Characteristics of Included Studies
Full size table

Overall survival

Sixty-three studies comprising 29,620 patients reported HRs for OS. All studies analyzed LDH as a dichotomous variable. The studies have clearly shown that upper limit of normal (ULN) remains common for high LDH. The median cutoff for high LDH was 250U/L (range = 200–1000).

Two of the 63 eligible studies (3.2%) reported a non-statistically significant HR. A forest plot of all studies is presented in Figure 1. Overall, LDH greater than the cutoff was associated with a HR for OS of 1.48 (95% CI = 1.43 to 1.53; P < 0.00001). As the heterogeneity among studies was significant (P < 0.00001; I2 = 93%), a random-effects model was applied. To explore potential sources of heterogeneity, we performed subgroup analysis in the following subgroups: disease site, tumor stage and LDH subdivided by predefined cutoffs.

Figure 1
figure 1

Forest plots showing HR for OS for LDH greater than or less than the cutoff.

HRs for each study are represented by the squares, the size of the square represents the weight of the study in the meta-analysis and the horizontal linecrossing the square represents the 95% confidenceinterval (CI). All statistical tests were two-sided.

Full size image

The effect of LDH on OS among disease subgroups is shown in Figure 2. The prognostic effect of LDH was highest in renal cell carcinoma (HR = 1.84, 95% CI = 1.35 to 2.51), followed by nasopharyngeal carcinoma (HR = 1.82, 95% CI = 1.48 to 2.24), sarcoma (HR = 1.79, 95% CI = 1.30 to 2.47), melanoma (HR = 1.76, 95% CI = 1.56 to 1.98), prostate cancer (HR = 1.55, 95% CI = 1.06 to 2.26), colorectal cancer (HR = 1.52, 95% CI = 1.29 to 1.79) and lung cancer (HR = 1.50, 95% CI = 1.27 to 1.78). The HR for the subgroup of other unselected solid tumors was 1.69 (95% CI = 1.44 to 2.00). For the eight disease-site subgroups analyzed, there was statistically significant heterogeneity between disease sites (P < 0.00001), but no significant differences in the prognostic values of LDH between the subgroups (P for subgroup difference = 0.68).

Figure 2
figure 2

Forest plots showing HRs by disease subgroups.

Full size image

The effect of LDH on OS among different disease stages is shown in Figure 3. The HRs were 1.54 (95% CI = 1.32 to 1.80) for non-metastatic disease, 1.70 (95% CI = 1.59 to 1.82) for metastatic disease and 1.20 (95% CI = 1.16 to 1.24) for a mixed group consisting of studies that included both metastatic and non-metastatic patients. There was statistically significant heterogeneity between disease stages (P < 0.00001). The prognostic value of LDH also varied significantly between different disease stages (P for subgroup difference < 0.00001).

Figure 3
figure 3

Forest plots showing HRs by stage subgroups.

Full size image

The effect of LDH on OS among different cutoffs for LDH is shown in Figure 4. The HRs were 1.71 (95% CI = 1.38 to 2.12) for LDH cutoff < 250U/L, 1.67(95% CI = 1.52 to 1.84) for LDH cutoff 250 to 300U/L, 1.69 (95% CI = 1.27 to 2.24) for LDH cutoff 301 to 400U/L and 1.72(95% CI = 1.45 to 2.05)for LDH cutoff > 400 U/L. There was no statistically significant heterogeneity between the different cutoffs for LDH (P for subgroup difference = 0.99).

Figure 4
figure 4

Forest plots showing HRs by LDH cutoffs.

Full size image

The scatter plot for the univariate meta-regression analysis is shown in Figure 5.A total of 63 studies was included in the meta-regression analysis. Overall, there was no statistically significant association between LDH cutoff and the HR for OS (P = 0.614).

Figure 5
figure 5

Study-level (i.e., at the individual publication level) association of the cutoff used to define LDH and the HR for overall survival.

Each study is represented by a circle and the area of the circleis proportional to the number of patients enrolled in each study. The gradient of the dashed line represents the results of the meta-regression (β = 1.000138).

Full size image

There was evidence of publication bias, with fewer small studies reporting negative results than would be expected (Supplementary Figure S2).

Three studies, comprising 1,766 patients, analyzed LDH as a continuous variable and reported HRs for OS. The pooled summary HR of these studies was 2.11 (95% CI, 1.35–3.28; P = 0.0003; I2 = 84%) per incremental LDH unit (Supplementary Figure S5).

Progression-free survival

Six studies, comprising 2,451 patients, reported HRs for PFS. Overall, LDH greater than the cutoff was associated with a HR for PFS of 1.70 (95% CI = 1.44 to 2.01; P < 0.00001; I2 = 13%). A forest plot is presented as Figure S3.

Disease-free (Recurrence-free) survival

A total of five trials, comprising 1,992 patients, reported HRs for DFS. Overall, LDH greater than the cutoff was associated with a HR for the endpoints of 1.86 (95% CI = 1.15 to 3.01; P = 0.01; I2 = 88%). A forest plot is presented in Figure S4.

Discussion

This is the first comprehensive meta-analysis of the prognostic relevance of LDH in solid tumors and it is based on a large pool of clinical studies (31,857 patients). We found a consistent effect of an elevated LDH on OS (HR = 1.48, 95%CI = 1.43 to 1.53) across all disease subgroups and stages. In addition, there is a trend toward a stronger prognostic value of LDH in metastatic disease compared with non-metastatic disease, which may reflect greater tumor burden. The prognostic impact of LDH on PFS and DFS (or RFS) is also robust. Interestingly, different cutoffs of LDH for different disease sites were reported in the included studies. However, the result of subgroups analysis for LDH cutoff showed that there was no association between LDH cutoff and reported HR for OS. This result was confirmed by meta-regression of LDH cutoff and HR for OS. Moreover, LDH was also related to poor prognosis in solid tumors when analyzed as a continuous variable. Our conclusions are supported by the fact that our selected studies were confined to those that used proportional hazards modeling to adjust for clinical prognostic factors and where the sample size was greater than 200.

There is a good biologic rationale for the use of LDH as a prognostic marker for cancer patients; however, the exact mechanism is not understood. One potential mechanism may be an association between LDH and the well-established phenomenon of oncogenicanaerobic glycolysis, or the Warburg effect5. This metabolic reprogramming is regulated by HIF-1α, as well as myc, through the transcriptional activation of key genes encoding metabolic enzymes; these include LDH, which converts pyruvate to lactate. This process is closely associated with an increased risk of invasion, metastasis and patient death77.

These analyses have several important implications. First, they show that a high LDH is associated with worse outcome, which suggests that LDH may be a useful biomarker to direct therapeutic selection78,79.This is because LDH is under the translational control of HIF-1α, as well as myc and thus is regulated by key oncogenic processes, such as the phosphatidylinositol 3-kinase/Akt/TORC1/hypoxia-inducible factor (PI3K/Akt/TORC1/HIF) pathway80,81,82. A recent study has demonstrated that the TORC1 inhibitor, temsirolimus, could provide therapeutic benefit in patients with RCC and high LDH79. Further work to investigate the predictive value of pretreatment LDH in other solid tumors may provide a more general insight into which patients derive benefit from TORC1 inhibition. Second, they show that increased LDH may be interpreted as reflecting high tumor burden or tumor aggressiveness. This suggests that dynamic changes of LDH level may be useful for predicting the prognosis in cancer patients after a primary operation, adjuvant chemotherapy, hormonal therapy, or radiotherapy65. Third, LDH allows the identification of a subgroup of tumors with a worse outcome. It is essential in the treatment of cancer to distinguish between low- and high-risk patients, thereby allowing stratification for standard or intensified treatment protocols. It has been shown that LDH can be used as an effective biomarker to guide the selection of regorafenib in patients with colorectal cancer; patients with high LDH may not be optimal candidates for regorafenib83.To adequately address these issues and dissect the complex relationship between LDH and cancer, future studies should be conducted within tumor- and stage-specific cohorts.

The strengths of this meta-analysis include the large sample size, estimation of HR using multivariate proportional hazards modeling that adjusted for clinical prognostic factors and analysis of a massive dataset comprising a large pool of clinical studies. LDH is also likely to be a cancer-specific biomarker, given that it is rarely increased in patients without cancer84. Thus, LDH may be a universal prognostic marker in cancer. To improve research in this area, studies with a more specific focus, such as those that address the impact of an individual LDH level on the prognosis of a homogeneous population of cancer patients (i.e., patients with the same cancer stage and subtype), would likely be more informative.

These analyses have limitations. One of the main limitations is the significant heterogeneity between studies, although we used random-effects models when pooling subgroup data. The heterogeneity in these studies could be explained by different patient characteristics or study designs. To facilitate interpretation, we grouped the patients by tumor type and tumor stage. Another limitation is that this is a literature-based analysis. It is compromised by the potential for publication bias, in which there is a tendency for predominantly positive results to have been published, thus inflating our estimate for the association between LDH and outcome. Our strict inclusion criteria (study size greater than 200, the requirement for HRs and a requirement for a 95% CI or P value) may have introduced selection bias. Most of the included studies were retrospective, which may have introduced reporting bias. Finally, different cutoffs used to assess high LDH level in these studies might also have contributed to the heterogeneity because it is possible that more false-positive cases were obtained with a cutoff of < 300 U/L than with a cutoff of >300 U/L. However, there is no accepted and validated absolute LDH level above which high LDH can be assigned. Instead, we used a cutoff of ULN. This may have introduced substantial heterogeneity, which may not have been fully accounted for by our use of sensitive analyses. The use of ULN is less robust; however, this was the only feasible method with the data available. An internationally accepted and validated LDH cutoff is warranted.

In summary, our data suggest that pretreatment LDH is a simple, cost-effective prognostic factor that can be considered as a criterion to consider patients in different prognostic groups. LDH is also a potential predictive marker to guide individual therapy decisions in solid tumors. Further, adequate, multi-center prospective studies are required to explore the clinical utility of LDH in solid tumors.