Exploration of IMDC model in patients with metastatic renal cell carcinoma using targeted agents: a meta-analysis

ABSTRACT Purpose: To explore the International Metastatic Renal Cell Carcinoma Database Consortium (IMDC) model application for predicting outcome of patients with metastatic renal cell carcinoma using targeted agents. Materials and Methods: We performed a literature review of 989 articles. The selecting process used preferred reporting items for systematic reviews and meta-analyses (PRISMA). All included studies were assessed by Newcastle-Ottawa scale. Results of individual studies were pooled using Stata 14.0 software. Results: A total of 17 articles were included. Most articles provided univariate and multivariate analysis of IMDC model prognosis. Combined HRs were 1.58 (95% CI 1.34-1.82) and 3.74 (95% CI 2.67-4.81) for univariate PFS of intermediate to favorable and poor to favorable respectively. In the category of multivariate PFS, combined HRs were 1.27 (95% CI 0.99-1.56) and 2.29 (95% CI 1.65-2.93) with intermediate to favorable and poor to favorable respectively. Regarding univariate OS, combined HRs were 1.93 (95% CI 1.62-2.24) and 6.25 (95% CI 4.18-8.31) with intermediate to favorable and poor to favorable respectively. With multivariate OS, combined HRs were 1.32 (95%CI 1.04-1.59) and 2.35 (95%CI 1.69-3.01) with intermediate to favorable and poor to favorable respectively. Conclusion: In summary, analysis of currently available clinical evidence indicated that IMDC model could be applied to classify patients with metastatic renal cell carcinoma using targeted agents. However, different types of targeted agents and various areas could affect the accuracy of the model. There was also a difference in predicting patients' PFS and OS.


INTRODUCTION
Renal cell carcinoma (RCC) represents approximately 3% of all cancers, with the highest incidence occurring in western countries. Generally, during the last two decades, there has been an annual increase of 2% in incidence both worldwide and in Europe, leading to approximately 99,200 new RCC cases and 39.100 kidney cancer-related deaths within the European Union in 2018 (1). According to the 2019 tumor statistics, there were 44.120 new kidney cancer men and 29.700 women in the United States, with the incidence rates being third and eighth respectively (2). Although most RCC cases are diagnosed at an early stage, approximately 20% of patients undergoing curati-ve nephrectomy will subsequently develop metastasis during the follow-up period (3). Many new therapeutic drugs have emerged, such as immune checkpoint drugs based on PD-1/PD-L1 or CTLA4 as representative drugs, targeted agents are still the mainstream drugs for the treatment of metastatic renal cell carcinoma. Because of the poor prognosis of metastatic renal cell carcinoma, it is important to choose appropriate prognostic factors for communication with patients and their families, to determine treatment options, and to group people in clinical trials. The most widely used prognostic models for the prognosis of metastatic renal cancer is International Metastatic Renal Cell Carcinoma Database Consortium (IMDC) model (4). IMDC model was based on prognostic data from populations treated with various targeted drugs (5). Although the applicability of the model has been verified by some articles like Kwon's article (6), there are also articles like Peltola's (7) article that provide different conclusions. Therefore, we conducted this study to explore the IMDC model application for predicting outcome in patients with metastatic renal cell carcinoma using targeted agents.

Search strategy
We performed a literature review of articles published before June 31, 2019 using the PubMed, Web of Sciences and Embase Databases. The main search terms used were "metastatic renal carcinoma", "prognosis", "TKI", "mTORi", "sunitinib", "sorafenib", "pazopanib", "axitinib", "bevacizumab", "everolimus", " temsirolimus" et al. and their combinations. Additional references were identified from the reference list of each article. Two reviewers carried out this process independently. The selecting process using preferred reporting items for systematic reviews and meta-analyses (PRISMA) (8) statement was exhibited in Figure-1 following the inclusion and exclusion criteria.

Poor to favorable
The combined HR was 2.29 (95% CI 1.65-2.93) and the forest plot is shown in Figure-

Univariate OS
In all 10 articles (6, 12-14, 16, 18, 18, 22-24) including 2419 patients in the category, among these patients, 1667 were clear cell RCC and 196 were non clear cell RCC. It was unfortunate that Kim's article did not provide specific number of patients with different pathological types. Favorable, intermediate and poor risk group had 565, 1227, and 419 patients respectively.

Intermediate to favorable
The combined HR was 1.93 (95% CI 1.62-2.24) and the forest plot is shown in Figure-3. According to funnel plot and Egger's test (p=0.194), no publication bias was detected. Sensitivity analysis showed the result was robust. Subgroup analysis showed the model was applicable in both Asia and other areas (Supplementary Figure-3). Whether the cohort of patients all took sunitinib alone or part of patients took sorafenib or pazopanib or temsirolimus, the model was efficient to classify favorable and intermediate-risk group (Supplementary Figure-4).

Poor to favorable
The combined HR was 6.25 (95% CI 4.18-8.31) and the forest plot is shown in Figure-3. According to funnel plot and Egger's test (p=0.596), no publication bias was detected. Sensitivity analysis showed the result was robust. Subgroup analysis showed the model was applicable in both Asia and other areas (Supplementary Figure-3). Whether the cohort of patients all took sunitinib alone or part of    Figure-4).

Multivariate OS
A total of 9 articles (7, 12, 16, 19-22, 25, 26) including 1950 patients in the category, among these patients, 1180 were clear cell RCC and 192 were non clear cell RCC. Kim's article not providing specific number of patients with different pathological types was also included. Favorable, intermediate and poor risk groups had 457, 1122, and 363 patients, respectively.

Intermediate to favorable
The combined HR was 1.32 (95% CI 1.04-1.59) and the forest plot is shown in Figure-3. According to funnel plot and Egger's test (p=0.551), no publication bias was detected. Sensitivity analysis showed the result was not robust. When Cai's article and You's article were omitted respectively, combined HR became not significant. Subgroup analysis showed the model was applicable in Asia. However, in other areas the model could not differentiate patients sufficiently (95% CI 0.80-1.

Poor to favorable
The combined HR was 2.35 (95% CI 1.69-3.01) and the forest plot is shown in Figure-3. According to funnel plot and Egger's test (p=0.555), no publication bias was detected. Sensitivity analysis showed the result was robust. Subgroup analysis showed the model was applicable in both Asia and other areas (Supplementary Figure-3). The model's efficiency was not reliable when it was applied to different types of targeted agents in the cohort of patients (Supplementary Figure-4).

DISCUSSION
IMDC model including six independent factors such as KPS <80%, time from diagnosis to treatment <1 year; hemoglobin <LLN, Calcium >ULN, Neutrophils <ULN, and Platelets >ULN was first set in 2009 (5). After its occurrence, many studies applied it to make risk stratification of patients using targeted agents. However, there was not a systematic evaluation for the model. In-depth analysis of the existing literature was performed to explore the application of IMDC model. Interestingly, it was found that the model was also utilized to predict patient's PFS though it was first set to predict patient's overall survival. Actually, its application in predict patient's PFS had not been explored. This was the first study to validate their application in the area.
Most incorporated articles provide univariate and multivariate analysis of prognostic factors. For meta-analysis, univariate pooling can best reflect potential valuable prognostic factors despite the possibility of combining confounding factors leading to repetitive effects. Multivariate merging may be inherently heterogeneous due to the inconsistencies in the variables included in each article. Conversely, the statistically significant prognostic factors obtained through this combination may be able to withstand the challenges of different conditions and could be widely used.
According to our analysis, IMDC model was able to classify patients to different risk group with various PFS and OS except in the category of intermediate to poor risk group for PFS (95% CI 0.99-1.56). Simultaneously, the combined HR was larger in the category of univariate analysis than those in the category of multivariate analysis. It possibly suggested that IMDC model was affected by other existing factors. In other words, it should be taken into account when the model is incorporated as one independent prognostic factor to reform a new prognostic model. In addition, we also explored the applicability of this model in different drugs and different populations. There are a variety of targeted drugs, and we have included studies that simply use sunitinib as a treatment, as well as a combination of sorafenib, pazopanib, and even mTORi, such as temsirolimus. Based on the subgroup analysis, IMDC model was reliable on the univariate analysis of PFS and OS and multivariate analysis of PFS limited in the poor to favorable risk group. Its applicability was not stable in the category of multivariate analysis of PFS located in the intermediate to favorable risk group and multivariate analysis of OS. When it came to the area targeted agents were used, various results existed in different conditions. IMDC model was reliable on the univariate analysis of PFS and OS and multivariate analysis of PFS and OS limited in the poor to favorable risk group both in Asia and other areas. It was not reliable in the category of multivariate analysis of PFS located in the intermediate to favorable risk group both in Asia and other areas. However, it could be used in the multivariate analysis of OS in Asia not in other area. There were two main explanations for the difference. On one hand, unstable results were concentrated on the intermediate to favorable risk group, indicating the classification was not accurate enough. On the other hand, PFS results were more stable than OS results, indicating that OS was easier to be affected by other factors other than targeted drug therapy. There was no doubt that the number of studies included is an important factor affecting the outcome. More high-quality clinical studies could provide more robust results.

Limitation and prospection
The findings of this systematic review should be considered in the context of the available evidence, which may be limited by selection bias and follow-up as reflected in the strength of evidence ratings. Due to there was not enough articles available, the application of the model for specific country or race was not explored. Meanwhile, most of the involved patients were ccRCC, the reliability of the model for nccRCC needed more studies to verify. Additionally, most articles used targeted agents as first line therapy except Keizman's, Auclin's and Kwon's articles (6,16,26), whether first line or second line of targeted therapy would influence the model was not explored. Although Heng's article (5) showed that there was no difference. And many other targeted agents such as axitinib were not covered in the included studies, leading to that the analysis was not particularly comprehensive. According to our analysis, the number of patients in the intermediate risk group was almost twice that of the other two groups, which was consistent with its initiative results (5). It indicated that a more specific subdivision could be made in the intermediate risk group.