Prognostic and predictive factors in patients with brain metastases from solid tumors: A review of published nomograms

Abstract

Objective

To review published nomograms that predict endpoints such as overall survival (OS) or risk of intracranial relapse in patients with brain metastases from solid tumors.

Methods

The methods and results of nomogram studies identified by a systematic search were extracted and compared, stratified by endpoint predicted by the respective nomograms. In particular, validation strategies (external/internal), concordance indices (cut-off 0.75) and comparisons to older models were analyzed.

Results

Six publications reported on prediction of OS. Most of these analyses focused on one particular primary tumor site, e.g., breast cancer or hepatocellular carcinoma, while the largest study included different primary tumor sites. The median number of patients was 244. Three of six studies included external validation cohorts. With few exceptions, concordance indices <0.75 were reported. In all studies reporting this endpoint, the nomogram outperformed older prognostic scores. Two nomograms focused on development of new brain metastases after radiosurgery (one externally validated), one on survival free from salvage whole brain radiotherapy (WBRT) after radiosurgery, and one on neurologic and non-neurologic death in patients receiving radiosurgery after WBRT failure. All concordance indices of these 4 nomograms were <0.70.

Conclusion

Taking into account concordance indices and comparisons to older prognostic models, the most promising, externally validated nomograms are the breast cancer and the non-small cell lung cancer nomogram predicting OS, and the distant brain failure after radiosurgery nomogram. Additional validation studies as well as continuous monitoring of the models' performance appear necessary to ensure their clinical applicability in the present era of rapidly changing treatment paradigms.

Introduction

A previous review (Nieder and Mehta, 2009) has addressed the strengths and weaknesses of 6 different prognostic indices, published between 1997 and 2008, i.e. after the Radiation Therapy Oncology Group (RTOG) developed and validated the widely adopted 3-tiered prognostic index known as recursive partitioning analysis (RPA) classes (Gaspar et al., 1997). In addition to these classic indices (Le Scodan et al., 2007; Lagerwaard et al., 1999; Weltman et al., 2000; Lorenzoni et al., 2004; Sperduto et al., 2008; Rades et al., 2008), other attempts were made, e.g., by combining three different models to predict individual patients' survival (Nieder et al., 2010) and by taking into account the specific hallmarks of individual primary tumor types, e.g., lung and breast cancer (Sperduto et al., 2010). The latter 4-tiered model, which is known as disease-specific graded prognostic assessment (DS-GPA), has recently been refined to also incorporate molecular features with high therapeutic impact in lung cancer (Sperduto et al., 2017a,b). Furthermore, this lung-mol GPA has now been validated in an independent database (Nieder et al., 2017a,b). Parallel to these developments, other groups have focused on cohorts of patients, which typically are underrepresented in large databases, e.g., those with rare primary tumors, surgical intervention or challenging location of their metastases (Rades et al., 2017a,b; Yamagishi et al., 2016; Sehmisch et al., 2017; Ferguson et al., 2017; Joshi et al., 2016). Most of these scores have yet to be validated. Different methods were used to create all these survival prediction models, and while the general usefulness of such models is widely acknowledged (Lambin et al., 2013) several challenges remain to be addressed. For example, poor prognosis strata with median survival of 2–3 months often contain a minority of patients who survive for at least 6-12 months (Nieder and Mehta, 2009), i.e. a time interval that is long enough to warrant active treatment interventions.

In contrast to 3- or 4-tiered scores that distinguish between groups of patients with relatively similar prognosis, nomograms are often used for individualized estimation of prognosis (Gittleman et al., 2017) or disease control and may be helpful for treatment planning (Kent et al., 2016). The purpose of this review was to identify and discuss published nomograms derived from cohorts of patients with brain metastases from solid tumors, stratified by endpoint, e.g., overall survival (OS) or risk of intracranial relapse.