Survival Tree Analysis Research Articles

Abstract 1675: Lung cancer chemotherapy: genetic variations in the transforming growth factor beta pathway as predictors of survival

Abstract Treatments based on molecular determinants are likely to maximize therapeutic index for chemotherapy in non-small-cell lung cancer (NSCLC). The purpose of this study is to take a pathway-based approach to systematically evaluate the pharmacogenetic effects of transforming growth factor (TGF-β) pathway on lung cancer and identify potential genetic predictors of platinum-based chemotherapy. We genotyped 227 SNPS in 23 TGF-β pathways genes to evaluate their associations with overall survival in 598 late stage NSCLC patients receiving first-line platinum-based chemotherapy with or without radiotherapy. In individual analysis, 22 SNPs were significantly associated with overall survival, of which the strongest associations were found for BMP2:rs235756 (HR=1.45; 95%CI, 1.11-1.90; P=0.006) and SMAD3:rs4776342 (HR=1.25; 95%CI, 1.06-1.47; P=0.009). Cumulative effect analysis showed that individuals with multiple risk genotypes had a significant dose-dependent adverse affect on overall survival (Ptrend=2.44×10−15). Analysis stratified by treatment modality showed that SMAD3 haplotypes exhibited significant associations with outcomes of chemotherapy. Survival tree analysis categorized subgroups of patients with dramatically different survival pattern. The low-risk genetic profile group identified from patients treated with chemoradiation had a median survival time (MST) of 30.4 months. In comparison, the high-risk genetic profile group had a more than four-fold increased risk of death with MST of 6.8 months. The low-risk genetic profile group identified from patients treated with primary chemotherapy without definitive radiotherapy had a MST of 18.03 months, while the high-risk genetic profile group had a MST of 5.9 months and a four-fold increased risk of death. These results suggest that genetic variations in TGF-β pathway are potential predictors for clinical outcome in advanced NSCLC treated with platinum-based chemotherapy. We also found that genetic variations within the TGF-β signaling pathway may be able to identify potential subgroups of patients who will most benefit from platinum-based chemotherapy with or without radiotherapy. Citation Format: {Authors}. {Abstract title} [abstract]. In: Proceedings of the 101st Annual Meeting of the American Association for Cancer Research; 2010 Apr 17-21; Washington, DC. Philadelphia (PA): AACR; Cancer Res 2010;70(8 Suppl):Abstract nr 1675.

Abstract 2834: Genetic variants in TGFβ pathway genes and ovarian cancer risk

Abstract The transforming growth factor (TGF) β signaling pathway plays a crucial role in cell proliferation, differentiation, apoptosis, extracellular matrix production and other cellular functions. Dysregulation and genetic variations of major genes in this pathway may be involved in the process of carcinogenesis. In this study, 218 single nucleotide polymorphisms (SNPs) in 21 major genes of TGFβ pathway were genotyped in 339 Caucasian ovarian cancer cases and 349 age, gender, and ethnicity matched controls. We analyzed the association between these SNPs and the risk as well as clinical outcomes of ovarian cancer. In the risk analysis, 16 SNPs were found significantly associated with ovarian risk, among these SNPs, a dose-response effect of increased risk for ovarian cancer with increased unfavorable genotype (UFG: defined as the allele related with increased ovarian risk) was observed (P for trend < 0.0001). Compared with individuals carrying ≤ 6 UFGs, subjects carrying ≥ 9 UFGs conferred a 3.7-fold increased risk of ovarian cancer (ORs =3.71, 95%CI=2.52-5.45, P<0.0001). The most significant individual association was found for INHBC: rs2228225. Patients carrying at least one variant allele of this SNP conferred 63% increased risk for ovarian cancer (OR = 1.63, 95%CI= 1.20-2.22, P=0.0018). Potential high-order gene-gene interactions were identified using the classification and regression tree analysis (CART). Eight terminal nodes were identified representing eight groups of patients with different probability for ovarian cancer (ORs ranging from 2.87 to 6.68). In the analysis of clinical outcomes, 13 SNPs were significantly associated with the risk of death. The top two SNPs were from BMP1. Patients carried rare homozygous genotype of either one of the SNP has a nearly two-fold increased risk for death (rs3857979: HR= 1.80, 95%CI= 1.23- 2.65, P=0.0026; rs4075478: HR=1.89, 95%CI=1.22- 2.91, P=0.004). A cumulative effect was also observed (P for trend < 0.0001). The median survival times decreased from 83.3 months for patients with ≤5UFGs to 26.8 months for those with ≥ 9UFGs (log rank P < 0.0001). Survival tree analysis identified eight terminal nodes representing eight groups of patients with different survival probability. Interestingly, the two BPM1 SNPs that were most significant in survival analysis were also significantly associated with poor response for chemotherapy (rs3857979: OR= 1.90, 95%CI= 1.04-3.48, P=0.0375; rs4075478: HR=2.39, 95%CI=1.17-4.86, P=0.0166), indicating that these two SNPs were predictive markers for response. In conclusion, our result demonstrates that genetic variations within TGFβ pathway may have modulate the risk and clinical outcomes of ovarian cancer. Citation Format: {Authors}. {Abstract title} [abstract]. In: Proceedings of the 101st Annual Meeting of the American Association for Cancer Research; 2010 Apr 17-21; Washington, DC. Philadelphia (PA): AACR; Cancer Res 2010;70(8 Suppl):Abstract nr 2834.

Polymorphisms of LIG4, BTBD2, HMGA2, and RTEL1 Genes Involved in the Double-Strand Break Repair Pathway Predict Glioblastoma Survival

Glioblastoma (GBM) is the most common and aggressive type of glioma and has the poorest survival. However, a small percentage of patients with GBM survive well beyond the established median. Therefore, identifying the genetic variants that influence this small number of unusually long-term survivors may provide important insight into tumor biology and treatment. Among 590 patients with primary GBM, we evaluated associations of survival with the 100 top-ranking glioma susceptibility single nucleotide polymorphisms from our previous genome-wide association study using Cox regression models. We also compared differences in genetic variation between short-term survivors (STS; <or= 12 months) and long-term survivors (LTS; >or= 36 months), and explored classification and regression tree analysis for survival data. We tested results using two independent series totaling 543 GBMs. We identified LIG4 rs7325927 and BTBD2 rs11670188 as predictors of STS in GBM and CCDC26 rs10464870 and rs891835, HMGA2 rs1563834, and RTEL1 rs2297440 as predictors of LTS. Further survival tree analysis revealed that patients >or= 50 years old with LIG4 rs7325927 (V) had the worst survival (median survival time, 1.2 years) and exhibited the highest risk of death (hazard ratio, 17.53; 95% CI, 4.27 to 71.97) compared with younger patients with combined RTEL1 rs2297440 (V) and HMGA2 rs1563834 (V) genotypes (median survival time, 7.8 years). Polymorphisms in the LIG4, BTBD2, HMGA2, and RTEL1 genes, which are involved in the double-strand break repair pathway, are associated with GBM survival.

Survival Tree and Meld to Predict Long Term Survival in Liver Transplantation Waiting List

MELD score is a formula based on laboratory variables used as a predictor of short-term mortality index in cirrhotic patients. It is applied to allocate patients in liver transplantation waiting list in many countries. However, MELD score cutoff point accuracy to predict long term mortality has not been statistically evaluated. The aim of this study was to analyze the MELD score and other variables related to long-term mortality using a new model: the Survival Tree analysis. The variables considered in this study were obtained at the time of liver transplantation list enrollment. The graphical representation of the survival trees showed that MELD 16 was the most statistically significant mortality cutoff point. The results were compatible with the MELD cutoff point reported in the clinical literature. This methodology can be extended to identify significant cutoff points related to other diseases whose severity is not necessarily expressed by MELD.

Abstract PR-03: Genetic variations in microRNA biogenesis genes as predictors for risk of second primary tumor (SPT) and/or recurrence in patients with early-stage head and neck cancer

Abstract Second primary tumors (SPTs) and local-regional recurrence negatively impact the prognosis of patients with curatively treated early-stage head and neck cancer. Genetic variations in microRNA (miRNA) biogenesis genes have been found to be associated with the risk of multiple solid tumors. Thus, we hypothesized that these variations were also associated with the risk of SPT/recurrence in patients with early-stage head and neck cancer. We performed a nested case-control study including 150 early-stage head and neck cancer patients with SPT/recurrence and 300 patients without SPT/recurrence. We genotyped 71 tagging and potentially functional single nucleotide polymorphisms (SNPs) from eight miRNA biogenesis pathway genes (RNASEN, DGCR8, DICER1, EIF2C1, DDX20, GEMIN4, RAN, XPO5). Seven SNPs in RNASEN, one in XPO5, and one in RAN were significantly associated with the risk of SPT/recurrence. The most significant SNP was rs7735863, an intronic SNP in RANSEN. The variant-containing genotypes of this SNP were associated with a 1.72-fold increased risk (95% confidence interval [CI], 1.16–2.57; P=0.0076). A combined analysis showed that compared with the patients without any unfavorable genotypes, the SPT/recurrence risk was increased 1.50- (95% CI, 1.02–2.22) and 2.69-fold (95% CI, 1.59–4.53) for those with one and two unfavorable genotypes, respectively (P for trend&amp;lt;0.001). Survival tree analysis further identified potential higher-order interactions and categorized the study subjects into different risk groups. Compared to the low-risk group with an event-free median survival time (MST) over 96 months, patients in the high-risk group had a 5.02-fold increased risk (95% CI, 2.70–9.31; P=3.20×10−7) and an event-free MST of 31.4 months (log rank P=1.25 − 10−7). Our results suggest that genetic polymorphisms within the miRNA biogenesis pathway may be used individually and jointly to predict the risk of SPT/recurrence of early-stage head and neck cancer patients. Citation Information: Cancer Prev Res 2010;3(1 Suppl):PR-03.

Genetic variations in the PI3K/PTEN/AKT/mTOR pathway are associated with clinical outcomes in esophageal cancer patients treated with chemoradiotherapy.

The phosphoinositide-3-kinase (PI3K), phosphatase and tensin homolog (PTEN), v-akt murine thymoma viral oncogene homolog (AKT), and mammalian target of rapamycin (mTOR) signaling pathway has been implicated in resistance to several chemotherapeutic agents. In this retrospective study, we determined whether common genetic variations in this pathway are associated with clinical outcomes in esophageal cancer patients with adenocarcinoma or squamous cell carcinoma who have undergone chemoradiotherapy and surgery. Sixteen tagging single nucleotide polymorphisms (SNPs) in PIK3CA, PTEN, AKT1, AKT2, and FRAP1 (encoding mTOR) were genotyped in these patients and analyzed for associations with response to therapy, survival, and recurrence. We observed an increased recurrence risk with genetic variations in AKT1 and AKT2 (hazard ratio [HR], 2.21; 95% CI, 1.06 to 4.60; and HR, 3.30; 95% CI, 1.64 to 6.66, respectively). This effect was magnified with an increasing number of AKT adverse genotypes. In contrast, a predictable protective effect by PTEN genetic variants on recurrence was evident. Survival tree analysis identified higher-order interactions that resulted in variation in recurrence-free survival from 12 to 42 months, depending on the combination of SNPs. Genetic variations in AKT1, AKT2, and FRAP1 were associated with survival. Patients homozygous for either of the FRAP1 SNPs assayed had a more than three-fold increased risk of death. Two genes--AKT2 and FRAP1--were associated with a poor treatment response, while a better response was associated with heterozygosity for AKT1:rs3803304 (odds ratio, 0.50; 95% CI, 0.25 to 0.99). These results suggest that common genetic variations in this pathway modulate clinical outcomes in patients who undergo chemoradiotherapy. With further validation, these results may be used to build a model of individualized therapy for the selection of the optimal chemotherapeutic regimen.

Staging multiple myeloma patients with active disease using serum levels of β2m-free HLA class I heavy chain together with IgM or platelet count

Purpose: In multiple myeloma (MM), serum beta-2-microglobulin (β2m)-free heavy chains (FHC) of HLA class I has been shown to reflect disease activity. We investigated the possibility of stratifying patients with active disease according to FHC and other clinical parameters. Experimental design: We studied 146 patients with MM, including 100 at diagnosis, 31 in relapse and 15 unresponsive to therapy. Univariate and multivariate analyses were used to assess the prognostic significance of FHC together with continuous variables (age, albumin, creatinine, hemoglobin, erythrocyte sedimentation rate, β2m, calcium, IgM, platelet count) and categorical variables (Durie–Salmon disease stage, gender, bone lesion burden, heavy and light chain isotypes of M-component, clinical status). Survival tree analysis on significant variables was used to develop an MM staging system. Results: FHC, IgM, platelet count and hemoglobin were independent predictors of prognosis. Survival tree analysis of these variables defined 2 three-risk-group staging systems involving FHC and either IgM or platelet count. Median survival for FHC/IgM stages II and III was 41.5 and 27.8 months, whereas it was not reached for stage I patients ( p < 0.0001). In the FHC/platelets system, median survival was 93.2 (stage I), 44.1 (stage II) or 27.8 (stage III) months ( p < 0.0001). Similar results were obtained for the 117 MM patients without renal insufficiency (FHC/IgM p < 0.0001; FHC/platelets p = 0.001). For the 100 patients at diagnosis, FHC/IgM ( p = 0.001) was more effective than FHC/platelets ( p = 0.04). Conclusions: The independent prognostic markers FHC, IgM and platelets provide two staging systems unaffected by renal insufficiency. Both are effective in evaluating MM patients with active disease.

Germline genetic variations in drug action pathways predict clinical outcomes in advanced lung cancer treated with platinum-based chemotherapy

Genetic polymorphisms contribute to interindividual variation in drug response. However, a single polymorphism is likely to exhibit a modest effect. Therefore, we applied a pathway-based approach to evaluate the cumulative effect of multiple polymorphisms on clinical outcome of patients with non-small cell lung cancer. We genotyped 25 functional polymorphisms in 16 key genes involved in cisplatin metabolism and action and evaluated their associations with overall survival in 229 non-small cell lung cancer patients receiving first-line cisplatin-based chemotherapy. Several biologically plausible main effects were identified in individual analysis. More importantly, when six polymorphisms in nucleotide excision repair genes were analyzed jointly, a significant trend of reduced risk of death with decreasing number of putative unfavorable genotypes was observed (P for trend < 0.001 and log rank P < 0.001). Survival tree analysis revealed potential higher-order gene-gene interactions and categorized subgroups with dramatically different survival experiences, based on distinct genotype profiles. The median survival time was 78.5 months for terminal node 1 in the low-risk group, 15.1 months for terminal node 10 in the medium-risk group, and 6.7 months for terminal node 9 in the high-risk group (log rank P < 0.001). We also constructed a prediction hazard model. The area under the curve increased from 0.71 (using clinical variables only) to 0.84 (using clinical, epidemiological, and genetic variations from survival tree analysis). Our results highlight the clinical potential of taking a pathway-based approach and using survival tree analytic approach to identify subgroups of patients with distinctly differing outcomes.

Prognostic relevance of the mTOR pathway in renal cell carcinoma

AbstractBACKGROUND.The mammalian target of rapamycin (mTOR) pathway is up‐regulated in many human cancers, and agents targeting the mTOR pathway are in various stages of clinical development. The goal of the study was to evaluate the potential and limitations of targeting the mTOR pathway in renal cell carcinoma (RCC).METHODS.Immunohistochemical analysis using antibodies against pAkt, PTEN, p27, and pS6 was performed on a tissue microarray constructed from paraffin‐embedded specimens from 375 patients treated by nephrectomy for RCC. The expression was associated with pathological parameters and survival.RESULTS.The mTOR pathway was more significantly altered in clear‐cell RCC, high‐grade tumors, and tumors with poor prognostic features. PS6 and PTEN showed the strongest associations with pathological parameters. Survival tree analysis regarding expression of cytoplasmic pAkt, nuclear pAkt, PTEN, cytoplasmic p27, and pS6 identified staining percentages of 40%, 10%, 75%, 7%, and 70%, respectively, as ideal cutoff values for stratification, with corresponding P‐values of .03, .001, .02, .005, and &lt;.0001, respectively. Interestingly, high nuclear pAkt expression was associated with a favorable prognosis, whereas high cytoplasmic pAkt expression was associated with a poor prognosis. In multivariate Cox regression analysis, ECOG PS, T classification, N classification, M classification, cytoplasmic Akt, nuclear pAkt, PTEN, and pS6 were independent prognostic factors of DSS.CONCLUSIONS.Components of the mTOR pathway are significantly associated with pathological features and survival. Not all RCC tumor types seem to be equally amenable to mTOR targeted therapy. PTEN, pAkt, p27, and pS6 may serve as surrogate parameters for patient selection and predicting prognosis. Patients with a highly activated mTOR pathway should benefit most from this therapy. External validation of our results is recommended. Cancer 2007. © 2007 American Cancer Society.

Evaluation of glutathione S‐transferase polymorphisms and mutagen sensitivity as risk factors for the development of second primary tumors in patients previously diagnosed with early‐stage head and neck cancer

The objective of this study was to evaluate the effects of polymorphisms in 2 genes in the glutathione S-transferase (GST) family and the mutagen-sensitivity phenotype on the risk of second primary tumors (SPTs) in patients with previously diagnosed early-stage head and neck squamous cell carcinoma. Data were available for 303 patients who were enrolled in a placebo-controlled chemoprevention trial of low-dose 13-cis-retinoic acid to reduce the occurrence of SPTs. A Cox proportional hazards model and survival tree analysis were used to evaluate the association between specified genetic variations and the development of SPTs. The average number of bleomycin-induced chromatid breaks per cell was used to quantify mutagen sensitivity as an individual patient's degree of sensitivity to genotoxicity. The GST-M1 null genotype was associated with an increased risk for any SPTs (hazard ratio [HR], 1.99; 95% confidence interval [95% CI], 1.11-3.56) and for tobacco-related SPTs (HR, 2.16; 95% CI, 1.01-4.62) after adjusting for covariates. The GST-T1 null genotype and bleomycin-induced chromatid breaks were not associated with a statistically significant increased risk for SPTs or tobacco-related SPTs after similar adjustment. Simultaneous nonnull status for both GST genotypes was associated with a decreased risk for any SPTs (HR, 0.52; 95% CI, 0.28-0.96) and tobacco-related SPTs (HR, 0.50; 95% CI, 0.22-1.11) compared with null status for GST-M1 accompanied by nonnull status for GST-T1. An association was observed between the development of SPTs and the GST-M1 null genotype after successful treatment for early-stage head and neck squamous cell carcinoma. The GST-T1 null genotype and bleomycin-induced chromatid breaks were not associated with an increased risk, and no significant interactions were identified.

Predictors of prostate cancer in men undergoing serial prostate biopsies: results from 371 patients

4589 Background: Due to a high rate of false negative prostate biopsies, a repeat biopsy is often recommended to men if there is a suspicion of occult carcinoma. We sought to determine the rate of prostate cancer (PC) detection and identify risk factors for cancer detection in men who despite suspicion of PC had a negative biopsy initially and then underwent serial prostate biopsies. Methods: Records of all men undergoing at least one repeat biopsy of the prostate from 1992 to 2003 were reviewed. Variables recorded were: age, race, family history of PC (FH), body mass index, biopsy indication, all PSA values, digital rectal exam (DRE), PSA density (PSAD; i.e. PSA/prostate volume), ultrasound findings, total number of cores obtained at all procedures, and the presence of high-grade PIN on initial biopsy. Outcome was defined as time from the initial biopsy to cancer detection or last follow-up. The data were analyzed by Kaplan-Meier method, log-rank test (univariate analysis), stepwise Cox regression (multivariate analysis) and binary recursive partitioning for censored data (survival tree). Results: 373 patients underwent 977 biopsy procedures. Median FU was 3.1 years for all patients and 3.6 years for patients without PC. PC was detected in 105 of 373 (28.7%) patients during follow-up. The estimated median time to PC detection for the entire cohort was 10.3 years. Factors significantly associated (p < 0.05) with a positive biopsy in the univariate analysis were PSADT, PSAD, biopsy indication, and PIN. The survival tree analysis identified 4 distinct risk groups with estimated 5-year PC detection rates (±S.E.) of 21 ± 4%, 40 ± 7%, 58 ± 8% and 100%. The Cox regression analysis showed: Conclusions: Patients who are referred for suspicion of PC but have a negative biopsy can be stratified into clinically meaningful risk groups. Prospective validation of this model is warranted. No significant financial relationships to disclose.

Predictors of prostate cancer in men undergoing serial prostate biopsies: results from 371 patients

4589 Background: Due to a high rate of false negative prostate biopsies, a repeat biopsy is often recommended to men if there is a suspicion of occult carcinoma. We sought to determine the rate of prostate cancer (PC) detection and identify risk factors for cancer detection in men who despite suspicion of PC had a negative biopsy initially and then underwent serial prostate biopsies. Methods: Records of all men undergoing at least one repeat biopsy of the prostate from 1992 to 2003 were reviewed. Variables recorded were: age, race, family history of PC (FH), body mass index, biopsy indication, all PSA values, digital rectal exam (DRE), PSA density (PSAD; i.e. PSA/prostate volume), ultrasound findings, total number of cores obtained at all procedures, and the presence of high-grade PIN on initial biopsy. Outcome was defined as time from the initial biopsy to cancer detection or last follow-up. The data were analyzed by Kaplan-Meier method, log-rank test (univariate analysis), stepwise Cox regression (multivariate analysis) and binary recursive partitioning for censored data (survival tree). Results: 373 patients underwent 977 biopsy procedures. Median FU was 3.1 years for all patients and 3.6 years for patients without PC. PC was detected in 105 of 373 (28.7%) patients during follow-up. The estimated median time to PC detection for the entire cohort was 10.3 years. Factors significantly associated (p < 0.05) with a positive biopsy in the univariate analysis were PSADT, PSAD, biopsy indication, and PIN. The survival tree analysis identified 4 distinct risk groups with estimated 5-year PC detection rates (±S.E.) of 21 ± 4%, 40 ± 7%, 58 ± 8% and 100%. The Cox regression analysis showed: Conclusions: Patients who are referred for suspicion of PC but have a negative biopsy can be stratified into clinically meaningful risk groups. Prospective validation of this model is warranted. No significant financial relationships to disclose.

Surgically defined prognostic parameters in patients with early cervical carcinoma. A multivariate survival tree analysis.

This study was performed to identify a statistical combination of independent pathologic and clinical features that best predict 5-year disease free survival (DFS) in patients with early stage cervical carcinoma treated by radical hysterectomy. The main goal of the study was to identify subsets of patients based on risk factors with maximal differences in DFS. Three hundred and seventy patients were found for whom complete clinical and pathologic material, including cone and cervical biopsies, were available for analysis. Variables studied included age, weight, race, marital status, economic status, tumor size (TS), depth of invasion (DI), lymph-vascular space involvement (LVSI), cell type, tumor grade, lymph node metastasis (LNM), and number of lymph nodes removed. Patients with LNM, parametrial involvement, and positive or close surgical margins were offered postoperative radiation. After excluding patients with microinvasive and small cell carcinoma, data from the remaining 301 patients were submitted to univariate and multivariate analyses to define those variables that best predict DFS. Univariate analysis showed that, ranked by degree of significance, DI, TS, LVSI, LNM, tumor volume (TV) and clinical stage were significant in predicting survival. Significant (P < 0.05) single parameters and other variables considered important were chosen for multivariate analysis, including the creation of a survival tree. With this method, DI (< or = 6 mm and > 2 cm), LVSI, age (> or = 40 yrs), and LNM were found to be the best combination of risk factors to define prognosis. The multivariate survival tree analysis maximally separates patients with early stage invasive carcinoma of the cervix into 3 subgroups with 5-year DFS of 91%, 68%, and 43%, respectively. The authors excluded patients with microinvasive carcinoma (SGO, Society of Gynecologic Oncologists), who have an excellent DFS of 100%, and patients with small carcinoma, who have a poor DFS of 36.4% based on cell type alone, to define independent risk factors that maximally separate the remaining patients by DSF. The survival tree prognostic scoring system is easy to apply, and only requires DI (mm), LVSI (+, -), LNM, and age to assign an individual patient to one of three risk groups.

Survival tree analysis for fracture risk

Survival tree analysis for fracture risk

Predictors of survival in systemic sclerosis (Scleroderma)

We conducted followup of 264 patients with definite systemic sclerosis (SSc) who were entered into the multicenter Scleroderma Criteria Cooperative Study (SCCS) during 1973-1977. At the end of the study (average 5.2 years of followup), 38% were known to be alive, 50% were dead (68% of these deaths definitely related to SSc), and 12% were lost to followup. Survival analyses of 484 demographic, clinical, and laboratory items recorded at entry into the SCCS (within 2 years of physician diagnosis of SSc) were performed. Survival declined linearly, and the cumulative survival rate was less than 80% at 2 years, 50% at 8.5 years, and 30% at 12 years after entry. Analysis using combinations of entry variables identifying organ system involvement confirmed that renal, cardiac, pulmonary, and gastrointestinal involvement in SSc predicted reduced survival; however, data on organ system involvement at study entry could not be used to consistently predict which organ system would ultimately be involved as the primary cause of death. By survival tree analysis, the individual entry variables best predicting reduced survival included older age (greater than 64 years), reduced renal function (blood urea nitrogen greater than 16 mg/dl), anemia (hemoglobin less than or equal to 11 gm/dl), reduced pulmonary diffusing capacity for carbon monoxide (less than or equal to 50% of predicted), reduced total serum protein level (less than or equal to 6 gm/dl), and reduced pulmonary reserve (forced vital capacity less than 80% with hemoglobin greater than 14 gm/dl or forced vital capacity less than 65% with hemoglobin less than or equal to 14 gm/dl). Cox proportional hazards model analysis confirmed these results. Different combinations of variables led to markedly different survival rates. The poorest prospects for survival were in patients with SSc who were less than or equal to 64 years old with a hemoglobin level less than or equal to 11 gm/dl, and in those greater than 64 years old with a blood urea nitrogen level greater than 16 mg/dl. These results may be useful in predicting individual patients at risk for shortened survival.