Participants were 217 patients with newly diagnosed metastatic RCC at the University of Texas MD Anderson Cancer Center, a life expectancy of greater than 4 months, a Zubrod performance status of less than or equal to 2, and no serious intercurrent illnesses. Informed consent was obtained from each participant prior to enrollment in the study. Period of protocol accrual was from April 2000 through November 2005. At the time of enrollment patients completed a battery of questionnaires, provided a blood sample, and collected five saliva samples per day for the subsequent three consecutive days (upon awakening, 45 minutes later, 8 and 12 hours after waking, and at bedtime). The study was approved by the Surveillance Committee for the Protection of Human Subjects at MD Anderson. Appropriate material transfer agreements were in place for the assessment of the blood and saliva samples.
At study entry, patients completed several psychosocial questionnaires. The Centers for Epidemiologic Studies - Depression was used to assess depressive symptoms, with scores of 16 or above classified as meeting screening criteria for depressive symptoms with further evaluation recommended 
. We chose to dichotomize the CES-D based on the well-established cut-off scores for screening criteria to avoid artificially creating cut off scores for a continuous measure, as well as for ease of reporting and interpreting results. Moreover, our interest was in determining whether patients who meet a minimum threshold for depressive symptoms would have worse outcomes and not examining symptom severity on a continuum, which would have less clinical relevance. Furthermore, a number of past meta-analyses examining the role of depressive symptoms on biological and economic outcomes have excluded studies that reported depressive symptoms as a continuous outcome 
. The SF-36 Health Status Survey assessed quality of life 
(Physical Component Scores and Mental Health Component Scores are reported). The Duke Social Support Index 
contains two subscales and measures size and structure of social network, and perceived satisfaction with support obtained from the network 
and the Coping Operations Preference Enquiry (Brief-COPE) 
which measures preferences for certain coping mechanisms, were also completed. For the B-COPE, we calculated subscales for Engagement and Avoidant coping 
. We also used a multi-modal assessment of religiosity/spirituality including measuring organized religious activity, non-organized religious activity 
, and intrinsic religiosity 
. Medical information was abstracted from medical charts.
Saliva samples were frozen and then shipped to Dr. Clemens Kirschbaum, Department of Psychology, Dresden University of Technology, for cortisol assay. Levels of cortisol were determined using a time-resolved immunoassay with fluorescence detection.
Blood samples were collected in sterile heparinized tubes (30 ml total) between 7–11 am and PBMCs (peripheral blood mononuclear cells) were isolated by Ficoll-Hypaque gradient centrifugation and cryopreserved. Gene expression profiling was carried out using total RNA extracted (Qiagen RNeasy) from PBMCs from a subgroup of 31 patients (15 patients with highest CES-D scores (≥16) and 16 patients with the lowest CES-D scores (<16) – matched on: sex, age at metastatic disease, smoking history, and disease risk group - low, intermediate, and high based on the following risk factors: Karnofsky <80%; corrected calcium >
10; serum hemoglobin <
13 mg/dl for males and <
11. 5 mg/dl for females; serum lactate dehydrogenase (1.5 times the upper limit of normal –upper limit of normal was 618 IU/L), previous radiation therapy; number of metastatic sites >
2; and interval between date of diagnosis and date of registration <
1 year 
. Those with 0 or 1 risk factor were classified at low risk, those with 2 risk factors were classified at intermediate risk, and those with more than 2 risk factors were classified at high risk) 
All samples met quality assurance standards for RNA mass and integrity, and whole genome transcriptional profiles were assayed by Illumina Human Ref-8 BeadArrays in the UCLA Social Genomics Core and the UCLA Southern California Genotyping Consortium, following the manufacturer’s specified protocol (Illumina Inc., San Diego CA). Data were quantile normalized and log-2 transformed for differential expression analyses identifying transcripts showing ≥50% difference in average expression across groups while controlling False Discovery Rates at 5%. Data are posted as Gene Expression Omnibus series GSE36957. Differential gene expression was determined using a fold-change cut-off because previous studies have shown that fold-change thresholds yield more replicable results than do p-value-based thresholds 
. To avoid potential confounding with correlates of CES-D scores, all differential gene expression analyses controlled for patient age, sex, ethnicity, education, marital status, body mass index, and disease risk index. Functional commonalities among differentially expressed genes were analyzed by GOstat Gene Ontology analysis (with default parameter settings) 
, and promoter-based bioinformatics analyses were carried out to identify transcription control pathways mediating the observed effects 
(both controlling False Discovery Rates at ≤5%). Promoter-based bioinformatics focused on the hypothesis that high CES-D scores would be associated with increased expression of genes with promoters bearing predicted binding sites for the pro-inflammatory NF-κB and STAT transcription factors (assessed by Transfac V$NFKB_Q6 and V$STAT1_01 nucleotide weight matrices, respectively) and transcription factors involved in monocyte/macrophage activation (V$EGR1_01-V$EGR3_01, V$NGFIC_01, V$MEF2_02, and V$MZF1_01) 
. Results represent the mean fold-difference in promoter response elements for each of those 8 transcription factor-binding motifs averaged over 9 different combinations of 3 promoter lengths (−300, −600, and −1000 to +200 bp relative to gene transcription start site) and 3 transcription factor motif detection stringencies (Transfac mat_sim
.80, .90, .95) 
Formalin-fixed, paraffin-embedded tissues were used for immunohistochemical-peroxidase staining for macrophages (CD68), HIF1α, MMP-2, MMP-9, or COX-2, as previously described 
Statistical Analyses: The database was locked for analyses May 2010. Correlation coefficients between study variables were first determined. For the purposes of this paper, we examined baseline psychosocial factors and cortisol slope as predictors of survival. Cox semi-parametric regression models were utilized to model survival from the diagnoses with metastatic disease of the participants and examine the univariate association of psychosocial factors, cortisol slope, demographic, and medical factors. As mortality is commonly associated with the metastasis of disease 
rather than the primary tumor, we chose to conduct our analysis from the time of diagnosis of metastatic disease versus initial diagnosis with earlier stage of disease. In order to have the cortisol data normally distributed, cortisol raw score levels were log-transformed. Cortisol slopes were calculated by regressing log-transformed cortisol levels on saliva collection time (five times a day for three days) for each patient. For all subsequent analyses, we included RCC risk factor classified as low, intermediate, or high risk as described above. We then analyzed separately models including CES-D, SF-36 PCS, and cortisol slopes, as each were associated with survival in the univariate analyses. We also ran Cox regression models to determine the associations between cortisol and depressive symptoms on survival in the same model in order to examine cortisol slope as a potential mediator. Model assumptions were evaluated for all variables using standard residual-based diagnostic procedures.