The predictive value of PRDM2 in solid tumor: a systematic review and meta-analysis

Study registration and methodology

A protocol of this study is registered at the International Prospective Register of Systematic Reviews (PROSPERO) with the following registration number: CRD42019132156 (https://www.crd.york.ac.uk/prospero/display_record.php?RecordID=132156) (National Institute for Health Research, 2019). Preferred Reporting Items for Systematic Reviews and Meta-analyses (PRISMA) flow diagram was used as a guideline to conduct our systematic review and meta-analysis (Moher et al., 2009).

Search strategy and study selection

A comprehensive electronic search was done in PubMed, Cochrane Library, ProQuest, EBSCO and ScienceDirect from inception to July 2019 using the following search terms: (PRDM2 OR RIZ OR RIZ1 OR RIZ2 OR KMT8 OR KMT8A OR MTB-ZF OR HUMHOXY1) AND (Cancer OR Cancers OR Malignant OR Malignancy OR Malignancies OR Neoplasm OR Neoplasms OR Neoplasia OR Neoplasias OR Tumor OR Tumors OR Tumour OR Tumours). The search was performed by two independent reviewers (Alfredo Bambang and Indra Putra Wendi). Any differences were solved through a discussion with a third reviewer (Anton Sumarpo).

All of the search outputs were exported into the EndNote software. Duplicates were removed, and screening was performed based on the title and abstract of the study. Probable or included studies were identified and assessed for eligibility according to the criteria above. Finally, included studies were identified, and data extraction was performed.

A study is included if it meets the following criteria: (1) The study used human subjects; (2) The study investigated the relationship between PRDM2 expression level and solid tumor through the use of gene expression analysis; (3) The study used histopathological examination as a comparator; (4) The study is a clinical trial or cross-sectional study. A study is excluded if: (1) The study does not have a control group (people without cancer or non-cancer specimens); (2) The study did not use an appropriate or did not state the gene expression analysis method used; (3) The expression level of PRDM2 in the study is not clearly stated or unquantifiable; (4) The study is a review, case series, conference abstracts, in vitro or in vivo study. (5) The study is not written in English.

Data extraction

The included studies were then analyzed further and the following informations are extracted: First author, publication year, country of origin, age, gender, race, type of cancer, cancer differentiation state, stage of cancer, type of control, number of cases and controls, gene expression analysis method, PRDM2 expression level and conclusion of the study. In the case of missing data, the authors will be contacted via email to request access to those missing data.

Data synthesis and statistical analysis

Sensitivity and specificity of PDRM2 were assessed in order to elucidate the potential of PRDM2 expression level as a diagnostic biomarker in solid tumor. Sensitivity and specificity are said to be significant if >50%. Risk ratio (RR) with a 95% confidence interval (CI) was used to determine the relationship between PRDM2 expression level and risk of cancer, as well as the relationship between PRDM2 expression level and clinicopathological data. If heterogeneity is present, Random Effects Model (REM) will be used. However, if heterogeneity is absent, Fixed Effects Model (FEM) will be used instead.

Cochrane’s Q test (chi-squared test) and Higgins I² statistics were used to assess for the presence of heterogeneity statistically. Heterogeneity is said to be present if P < 0.10 or I² > 75% (Higgins & Green, 2011; Higgins et al., 2003). To assess for the presence of heterogeneity visually, a forest plot will be generated. Meta-regression and subgroup analysis will be conducted when there are at least 10 studies used in the meta-analysis (Baker et al., 2009). The possible causes of heterogeneity are: Age, gender, ethnicity, country of origin, type of cancer, cancer differentiation state, stage of cancer and genotyping method.

Funnel plot and Deek’s test will be used to assess publication bias when the number of included studies is at least 10. If the funnel plot is asymmetric, publication bias is present. If the P-value for Deek’s test is <0.10, there is funnel plot asymmetry (Deeks, Macaskill & Irwig, 2005). If publication bias is found, the trim and fill method will be used to correct this bias (Duval & Tweedie, 2000).

Furthermore, sensitivity analysis was performed to elucidate the effect and stability of a single study on the pooled estimates by deleting one study at a time. Additionally, sensitivity analysis was also conducted to compare the pooled estimates using odds ratio (OR) and RR, as well as using REM and FEM. All statistical analyses were generated using RevMan 5.3 and STATA 12.0.

Quality assessment and data reliability

In order to claim that the meta-analysis conducted has been conclusive, the required information size has to be achieved. Thus, a trial sequential analysis (TSA) was performed using TSA software in order to determine the required information size (Wetterslev, Jakobsen & Gluud, 2017). Quality of evidence will be assessed using Quality Assessment of Diagnostic Accuracy Studies-2 (QUADAS-2) which consists of the following key domains: patient selection, index test, reference standard, as well as flow and timing (Whiting et al., 2011).

Confidence in cumulative evidence

Grading of Recommendations, Assessment, Development, and Evaluations (GRADE) was used to evaluate the confidence in cumulative evidence. Overall certainty of evidence can be written as high, moderate, low or very low (Schünemann et al., 2013).

Search results

Using variants of the keywords “PRDM2” and “cancer”, we performed a search from inception to July 2019 in PubMed, Cochrane Library, ProQuest, EBSCO and ScienceDirect. After duplicate removal, a total of 3,928 records was obtained. Titles and abstracts were screened and 58 potential studies were identified. Out of these 58 studies, 52 were excluded due to the studies being unable to meet the inclusion criteria (ineligible), in vitro and/or in vivo, used unsuitable methods, written in non-English, or is a review. The remaining six studies (Akahira et al., 2007; Dong et al., 2012; Ge et al., 2015; Geli et al., 2005; Jiang et al., 1999; Tan et al., 2018) were included in the systematic review while only five studies (Akahira et al., 2007; Dong et al., 2012; Geli et al., 2005; Jiang et al., 1999; Tan et al., 2018) were included in the meta-analysis. This is because Ge et al. (2015) did not mention the number of samples and controls that expressed PRDM2 downregulation in renal cell carcinoma. Thus, only five studies were included in the meta-analysis. The Preferred Reporting Items for Systematic Reviews and Meta-analyses (PRISMA) flow diagram for this study is shown in Fig. 1.

The studies that were eligible for systematic review were published from 1999 to 2015. There were a total of 314 samples of solid tumors and 225 controls obtained from patients in China (Dong et al., 2012; Ge et al., 2015; Tan et al., 2018), Japan (Akahira et al., 2007), Sweden (Geli et al., 2005) and United States of America (Jiang et al., 1999). All of these six studies are cross-sectional studies. The solid tumors included in this study are ovarian cancer (Akahira et al., 2007), esophageal squamous cell carcinoma (Dong et al., 2012), renal cell carcinoma (RCC) (Ge et al., 2015), pheochromocytoma (Geli et al., 2005), abdominal paraganglioma (Geli et al., 2005), hepatoma (Jiang et al., 1999), lung squamous cell carcinoma (LSCC) (Tan et al., 2018) and lung adenocarcinoma (LAC) (Tan et al., 2018). Out of these six studies, one used immunohistochemistry (IHC) only (Akahira et al., 2007), three used reverse transcription-polymerase chain reaction (RT-PCR) only (Ge et al., 2015; Geli et al., 2005; Jiang et al., 1999) and two used both IHC and RT-PCR (Dong et al., 2012; Tan et al., 2018). A summary of the main characteristics of the included studies for systematic review and meta-analysis is presented in Tables 1 and 2, respectively.

Systematic review results

All six studies concluded that PRDM2 gene expression is significantly decreased in solid tumor compared to control, with the P-value ranging from <0.05 to <0.001 using CI 95%. Akahira et al. (2007) stated that there was a significant correlation between PRDM2 downregulation with cancer grade (P < 0.0345) and stage (P < 0.0153) in ovarian cancer. On the other hand, Ge et al. (2015) stated otherwise, concluding that there was no significant relationship between RCC with tumor progression (P = 0.19). A study by Geli et al. (2005) reported that decreased PRDM2 gene expression was not correlated significantly with gender and tumor size, but was found to be weakly correlated with younger age (Spearman rank-order correlations; R = 0.4). Other clinicopathological data were either absent or not investigated in the studies. Hence, the role of PRDM2 downregulation in cancer grade, stage, gender, age and other clinicopathological data is still unclear. Due to the lack of sufficient clinicopathological data, only PRDM2 gene expression and its sensitivity and specificity were further analysed in the meta-analysis.

Meta-analysis results

Five studies were included in this meta-analysis to further investigate the relationship between PRDM2 downregulation with solid tumor (Akahira et al., 2007; Dong et al., 2012; Geli et al., 2005; Jiang et al., 1999; Tan et al., 2018). The pooled analysis suggested that PRDM2 gene expression is decreased in solid tumor (RR 4.29, 95% CI [2.58–7.13], P < 0.00001; Fig. 2). Based on this pooled analysis, three sensitivity analyses were conducted to evaluate the stability of our findings: with and without the deletion of Jiang et al. (1999) (Fig. 3), RR or OR (Fig. 4), and FEM or REM (Fig. 5). All three sensitivity analyses did not have meaningful differences, proving that our results are stable.

The sensitivity and specificity of PRDM2 downregulation in solid tumor were also assessed in order to investigate its potential as a diagnostic biomarker. A split forest plot displaying the sensitivity and specificity of the included studies is shown in Fig. 6. As demonstrated in the summary receiver operating characteristic (SROC) curve (Fig. 7), the summary sensitivity and specificity of decreased PRDM2 gene expression in solid tumor is 84% (95% CI [39–98]%) and 86% (95% CI [71–94]%), respectively. This result is in favor of PRDM2 downregulation as a potential diagnostic biomarker. However, the confidence interval for PRDM2 downregulation is wide, suggesting that there is marked imprecision. This was later confirmed on TSA (Fig. 8). In Fig. 8, the line representing the cumulative Z-curve failed to cross the significance boundary and did not reach the required number of studies which is 7743. Therefore, it can be concluded that the usage of PRDM2 downregulation as a diagnostic biomarker in solid tumor is still inconclusive.

Quality assessment of included studies

The quality of the included studied was evaluated using the QUADAS-2 tool, and a summary of the results can be viewed in Table 3. As shown in Table 3, in the index test domain there are four studies (Ge et al., 2015; Geli et al., 2005; Jiang et al., 1999; Tan et al., 2018) having an unclear risk of bias. These four studies did not directly state whether the index test (gene expression analysis) was interpreted independently from the reference standard (histopathological examination). Thus, we decided that unclear was most fit as the risk of bias. One of the studies, Jiang et al. (1999) also had missing information on how the patients were recruited, leading to an unclear risk of bias for one other domain. In general, the quality of the included studies was robust, ensuring the reliability of our systematic review and meta-analysis.

Confidence in cumulative evidence

By assessing five domains, including the risk of bias (by using the results from QUADAS-2 risk of bias assessment), indirectness, inconsistency, imprecision (by using the results from TSA) and risk of publication bias, a GRADE evidence profile was constructed as shown in Table 4. To be noted, all of the included studies used diagnostic accuracy test as their design, whereby all of the samples and controls will undergo both the index test and reference standard. Ideally, diagnostic studies should randomize which of the samples and controls will undergo the index test only and which will undergo the reference standard only. Hence, this made the design susceptible to indirectness. In addition, most of the included studies have wide confidence interval and inconclusive TSA results. Thus, serious was placed in the imprecision domain. As for publication bias, since the number of included studies is <10, publication bias could not be evaluated. Unfortunately, this does not entirely rule out the possibility of publication bias being present in our study, and thus we decided to downgrade the quality of evidence further. Overall, we have very low confidence in the pooled estimates obtained for our meta-analysis.