Concordance Probability Estimate Research Articles

External validation of a prognostic score predicting overall survival for patients with brain metastases based on extracranial factors

The aim of our study was an external validation of the extracranial prognostic score predicting survival of patients with brain metastases receiving cranial irradiation on data from a single institution. A retrospective analysis of 524 patients with brain metastases treated with cranial radiotherapy in a single tertiary center was performed. Three predictive scores were calculated and assessed for their ability to discriminate prognostic groups: (i) The Recursive Partitioning Analysis (RPA) score (available for 524 patients); (ii) the Diagnosis-Specific Graded Prognostic Assessment (DS-GPA) score (464 patients); (iii) the extracranial score (EC-S) developed by Nieder et al. which is based on serum albumin, lactate dehydrogenase (LDH) and the number of extracranial organs involved (157 patients). Discrimination of each score was assessed by Gönen & Heller's concordance probability estimate (CPE). The calibration was checked by comparing median survival estimates of each risk group with the corresponding values of the datasets from which the scores were derived. Finally, a multivariable Cox regression model was built by using the least absolute shrinkage and selection operator on a large number of variables including all three scores. With a CPE = 0.626 ± 0.022, the EC-S had the best discriminatory power. The EC-S also appeared to be better calibrated and had the best ability to separate patients with a very poor prognosis: patients with combination of low albumin, elevated LDH and more than 1 extracranial organ with metastatic involvement had a median survival time of only 0.6 months (CI95% 0.1-1.1) and a hazard ratio for death of 6.36 (2.67-15.14) compared to patients with no extracranial metastases and normal levels of albumin and LDH. In the multivariable Cox model serum albumin, LDH, treatment modality, DS-GPA and EC-S were retained as prognostic factors. An ad hoc combination of both DS-GPA and EC-S into a new score was possible for 134 patients and indicated a slightly better discrimination (CPE = 0.636 ± 0.023) than either DS-GPA or EC-S alone. This study provides an independent validation of the prognostic EC-S which was the best prognostic model for defining the patients who obviously did not benefit from radiation therapy of brain metastases in terms of overall survival. The combination of the EC-S with the established DS-GPA score resulted in a slight increase in discriminatory ability. The new EC-GPA score needs further validation in larger patient cohorts.

A Simplified Risk Stratification Method for Women With Stage I Endometrial Carcinoma.

Available risk stratification methods for women with endometrial carcinoma are controversially defined. We sought to develop a simplified and an individualized prognostic index for cancer recurrence in women with International Federation of Gynecology and Obstetrics (FIGO) stage I endometrial carcinoma, solely of endometrioid histology. We identified 976 women who underwent a hysterectomy and did not receive any adjuvant therapy. Cox proportional hazards model was used to identify independent predictors of recurrence. Prognostic groups were created based on the number of independent predictors of recurrence (0, 1, or 2 or 3 risk factors). These groups were then validated using a separate cohort of 611 women treated at another academic institution. The model's performance for predicting cancer recurrence was measured by the concordance probability estimate along with a 95% confidence interval. Median follow-up was 65 months. The final recurrence model included 3 risk groups based on 3 independent predictors of recurrence (tumor grade 2 or 3, the presence of lymphovascular space invasion and stage IB). Five-year recurrence rates were 4%, 16%, and 44% for groups 0, 1, and 2 or 3, respectively. The performance of the model was very good with a concordance probability estimate of 0.72 and 0.80 for the development and validation cohorts, respectively. On the basis of 3 well-known prognostic factors, we have developed and externally validated a simplified prognostic model that accurately predicts cancer recurrence in women with stage I endometrial carcinoma. This simplified predictive tool may be helpful in estimating individualized risk of recurrence and guide counseling with regard to adjuvant treatment.

Validation of a Simplified International Prognostic Score (IPS-3) in Patients with Advanced Stage Hodgkin Lymphoma

PurposeSince its publication in 1998, the International Prognostic Score (IPS-7) has been widely adopted as a risk stratification tool in patients with advanced stage classical Hodgkin lymphoma (cHL).1 In this study, the 5-y freedom-from-progression (FFP) ranged from 42 to 84% and 5-y overall survival (OS) from 56 to 89%. This index has demonstrated utility in the modern era, but with a significantly narrowed prognostic range.2 Further, missing factors can be problematic, limiting utility in clinical practice. A novel prognostic score (IPS-3), comprised of three of the seven IPS-7 indicators (age ≥ 45, stage IV, hemoglobin < 105), was proposed with data derived from advanced stage cHL patients enrolled on the E2496 clinical trial comparing ABVD to Stanford V.3 This model was reported to outperform the IPS-7 in predicting 5-y FFP and OS. We aimed to validate the IPS-3 model in advanced stage cHL treated with ABVD or ABVD equivalent chemotherapy in British Columbia.Patients and MethodsThe BC Cancer Lymphoid Cancer database was used to identify all advanced stage cHL patients (stage 1/2 bulky, stage 2B, stage 3/4), age ≥ 16 years, diagnosed between January 1, 1980 and June 6, 2018, treated with curative intent ABVD or an ABVD-equivalent regimen with available information for all seven IPS variables. Kaplan-Meier method and Cox proportional hazard models were used to estimate survival rates, hazard ratios (HRs), and 95% CIs. FFP was defined as time from date of diagnosis to disease progression or relapse. Log-rank testing was used to compare the survival curves between groups. As previously described3, prognostic performance and predictive accuracy of IPS-7 and IPS-3 were evaluated using concordance probability estimates (CPEs) with a Cox proportional hazard model based on risk groups.Results1191 patients were identified. Median age was 33 years (range 16 to 85), 30% were ≥45 years of age and 9% were >65 years of age, 55% were male and 22% had a high risk IPS-7 (≥4). Patient characteristics were similar to the E2946 cohort used to create the IPS-3 (Table 1). Furthermore, estimates of 5-y FFP and 5-y OS in the risk groups were very similar to those in the original report3 (Table 2). In the BC IPS-7 patients, 5-y FFP ranged from 65% to 91% (p<.001) and 5-y OS ranged from 68 to 99% (p<.001) and the IPS-3 model predicted a 5-y FFP of 84% ±2 for a score of 0, 76% ±2 for a score of 1 (HR 1.5, 95% CI 1.2 to 2.1), 72% ±4 for a score of 2 (HR 1.9, 95% CI 1.4 to 2.7) and 68% ±7 for a score of 3 (HR 2.4, 95% CI 1.4 to 4.2) and corresponding values for 5-y OS were 95% ±1, 87% ±2 (HR 3.5, 95% CI 2.4 to 5.1), 80% ±3 (HR 5.1 95% CI 3.4 to 7.7) and 61% ±8 (HR 9.0 95% CI 5.2 to 15.8). Restricting the analysis to patients 65 years of age and younger in our cohort (N = 1080), values for 5-y FFP were similar to the full cohort, ranging from 67 to 84% for IPS-3 and 69 to 91% for IPS-7. Values were slightly improved for OS, ranging from 70 to 95% for IPS-3 and 76 to 99% for IPS-7. Both the IPS-7 and the IPS-3 scores were not effective for predicting 5-y FFP or OS when applied to patients older than 65 years of age (all P≥0.54, N=111).Predictive accuracy and discriminatory performance were evaluated by CPE with higher scores associated with greater accuracy. CPEs for OS were 0.63 (SE 0.014) and 0.66 (SE 0.014) for IPS-7 and IPS-3, respectively. This result may suggest a better concordance between the observed data and IPS-3; however, there was a reversal in performance when analyzing FFP, as the CPEs for FFP were 0.59 (SE 0.014) and 0.57 (SE 0.015) for IPS-7 and IPS-3 respectively.ConclusionThis population-based study confirms that both IPS-3 and IPS-7 are prognostic in advanced stage cHL patients treated with ABVD. Unlike the original study evaluating the IPS-3, we did not find overwhelming evidence to suggest that the IPS-3 was more accurate for predicting prognosis than the IPS-7; however, given its simplicity and comparable performance to the IPS-7, IPS-3 may be more appealing for application in the clinical setting.

A Comparison of Prognostic Indices in Diffuse Large B-Cell Lymphoma within the UK NCRI R-CHOP 14 Versus 21 Phase III Trial

A Comparison of Prognostic Indices in Diffuse Large B-Cell Lymphoma within the UK NCRI R-CHOP 14 Versus 21 Phase III Trial

The prognostic value of 3 commonly measured blood parameters and geriatric assessment to predict overall survival in addition to clinical information in older patients with cancer.

The current study was performed to evaluate the prognostic value of laboratory parameters and geriatric assessment (GA) in addition to a baseline model with clinical information regarding overall survival (OS) in patients with cancer. GA was systematically performed in patients aged ≥70 years. The baseline model consisted of age, tumor type, and stage of disease. The incremental prognostic values of the GA as a whole (10-item GA) and laboratory parameters were assessed separately and combined. The parameters included hemoglobin (Hb), albumin, C-reactive protein (CRP), and the Glasgow Prognostic Score (GPS). Analyses were conducted with continuous and dichotomized variables. Cox models were compared based on Akaike information criterion (ΔAIC) and their discriminatory ability was assessed using the concordance probability estimate (CPE). A total of 328 patients were considered for this analysis. The baseline model had a CPE of 0.725. The addition of CRP, albumin, and Hb combined resulted in the best performing model (ΔAIC: 40.12 and CPE: 0.757) among the laboratory parameters. However, the 10-item GA improved the baseline model even more (ΔAIC: 46.03 and CPE: 0.769). Similar results were observed in the analysis with dichotomous variables. The addition of the 3 laboratory parameters (CRP, albumin, and Hb) improved the CPE by 1.4% compared with the baseline model already extended with the 10-item GA. The CPE increase (1.7%) was the highest with the GPS in the analysis with dichotomous variables. GA appears to add slightly more prognostic information than laboratory parameters in addition to clinical information. The laboratory parameters have an additional prognostic value beyond clinical and geriatric information.

Risk modeling in prospective diabetes studies: Association and predictive value of anthropometrics

AimsThis study aimed to introduce and apply modern statistical techniques for assessing association and predictive value of risk factors in first-degree relatives (FDR) of patients with diabetes from repeatedly measured diabetes data. MethodsWe used data from 1319 FDR’s of patients with diabetes followed for 8 years. Association and predictive performance of weight (Wt), body mass index (BMI), waist and hip circumferences (WC and HC) and their ratio (WHR), waist-height ratio (WHtR) and a body shape index (ABSI) in relation to future diabetes were evaluated by using Cox regression and joint longitudinal-survival modeling. ResultsAccording to Cox regression, in total sample, WC, HC, Wt, WHtR and BMI had significant direct association with diabetes (all p < 0.01) with the best predictive ability for WHtR (concordance probability estimate = 0.575). Joint modeling suggested direct associations between diabetes and WC, WHR, Wt, WHtR and BMI in total sample (all p < 0.05). According to LPML criterion, WHtR was the best predictor in both total sample and females with LPML of −2666.27 and −2185.67, respectively. However, according to AUC criteria, BMI had the best predictive performance with AUC-JM = 0.7629 and dAUC-JM = 0.5883 in total sample. In females, both AUC criteria indicated that WC was the best predictor followed by WHtR. ConclusionWC, WHR, Wt, WHtR and BMI are among candidate anthropometric measures to be monitored in diabetes prevention programs. Larger multi-ethnic and multivariate research are warranted to assess interactions and identify the best predictors in subgroups.

Evaluation of the MD Anderson Tumor Score and Their Tumor Related Prognostic Variables in the Rituximab Era

▪IntroductionDiffuse large B cell lymphoma (DLBCL) is a heterogeneous group of aggressive B cell lymphomas, considering their biologic, pathological and clinical backgrounds. However, treatment of DLBCL is relatively homogeneous and standard, mainly based in the R-CHOP regimen. Several prognostic scores have been proposed for categorizing the risk and finding those with worse results with standard treatment, suitable to be treated with new schemes or drugs. The most important and widely used is the International Prognostic Index (IPI) proposed in 1993 and lately validated in the rituximab era (R-IPI). However, being a good prognostic score lacks the ability to identify a very high risk prognostic subset in the rituximab era. Trying to improve this situation several attempts have been made including NCCN-IPI or GELTAMO-IPI. In 1992, the MD Anderson Cancer Centre (MDACC) reported a prognostic score considering exclusively variables related to the tumor: the Tumor Score (TS). Two of them already present in IPI: high LDH and Ann Arbor stage III-IV but three different: high beta-2-microglobulin (B2M), bulky mass and presence of B symptoms. This index has not been studied in the rituximab era. Our aim is to validate the TS in the rituximab era and analyze its current potential role.METHODSFrom the nation-wide database of DLBCL of the Grupo Español de Linfoma y Trasplante de Médula Ósea (GELTAMO), we included for the validation those patients homogenously treated with R-CHOP and with all five TS variables available (n=1294). Patients had to be ≥ 18 years-old and a minimum of 1 year of follow-up; all histological subtypes of DLBCL and primary extranodal cases were acceptable, with the only exclusion of primary testicular or CNS sites. Failure-free survival (FFS) (including disease progression, no response to treatment or death events) and overall survival (OS) were analyzed with the Kaplan-Meier method and compared with the log-rank test. Cox Regression models were used for univariate and multivariate analysis. Comparisons between scores were performed with concordance probability estimates (CPEs).RESULTSMedian follow-up was 60 months (12-176). 5y-FFS and OS of the series were 62% and 74%, respectively. All the variables of the original TS retained an independent prognostic role in our series. The TS in the rituximab era (R-TS) remains predictive and clearly identifies four different risk groups (Figure 1A), finding a particularly high risk subset with a worse outcome (5y-FFS of 29%). Comparison between TS and the other indexes (IPI, NCCN-IPI or GELTAMO-IPI) showed similar CPEs for FFS in our series: 0.64 vs 0.64, 0.64 and 0.65, respectively (Figure 1B). However, TS has a better discrimination of the higher risk subgroup than IPI (5y-FFS of 47%) and NCCN-IPI (5y-FFS of 39%), and the same that GELTAMO-IPI but with a better consistency between points of the score. As R-IPI and R-TS have complementary information, another option was combining R-IPI and R-TS: patients within high risk by R-IPI (5y-FFS of 47%) when tested with R-TS can be further subdivided in three risk categories (intermediate low, intermediate high and high with 5y-FFS of 75%, 61% and 37%, respectively). To further improve the ability of finding a very high risk subset with the variables included in TS, we tested categorizing B2M normalized (normal, >1 and >3), LDH normalized (normal, >1 and >3) and AA stage (I, II and III-IV). With these changes the new enhanced TS could identify a higher risk group with a 5y-FFS of 20% and a median FFS of 7 months (Figure 1C and 1D).CONCLUSIONS1) All variables included in the original MD. Anderson TS retain an independent prognostic role in the rituximab era, including B symptoms, B2M and bulky mass; 2) TS remains predictive of FFS and OS in the rituximab era with a similar CPE in discrimination when compared to previously reported prognostic scores; 3) TS and GELTAMO-IPI showed a better identification of patients with high risk prognosis compared to IPI or NCCN-IPI; 4) R-IPI and R-TS may be combined in order to easily improve risk classification of DLBCL patients; 5) Further categorization of LDH, B2M and AA stage increased the ability of TS to identify high risk subsets of DLBCL; 6) TS could be a backbone for introducing other new molecular or biologic tumor related prognostic factors.Figure 1. FFS using R-TS (1A), R-IPI (1B) and enhanced TS (1C). OS using enhanced TS (1D). [Display omitted] DisclosuresGutierrez:SERVIER: Speakers Bureau; GILEAD: Honoraria; TAKEDA: Speakers Bureau; PFIZER: Consultancy; JANSSEN: Consultancy, Research Funding, Speakers Bureau; ROCHE: Research Funding, Speakers Bureau. Diaz-Lopez:TFS: Employment. Lopez-Guillermo:Roche: Consultancy, Other: Research grant; Celgene: Consultancy; Gilead: Consultancy; Janssen: Consultancy; Novartis: Consultancy. Martín:Roche: Consultancy, Honoraria; Celgene: Consultancy, Honoraria; Servier: Consultancy, Honoraria; Janssen: Honoraria; Gilead: Consultancy. Salar:Roche: Speakers Bureau; Janssen: Speakers Bureau; Servier: Speakers Bureau.

Clinical and Pathological Staging Validation in the Eighth Edition of the TNM Classification for Lung Cancer: Correlation between Solid Size on Thin-Section Computed Tomography and Invasive Size in Pathological Findings in the New T Classification

IntroductionThe aim of this study was to validate the new eighth edition of the TNM classification and to elucidate whether radiological solid size corresponds to pathological invasive size incorporated in this T factor. MethodsWe analyzed the data on 1792 patients who underwent complete resection from 2003 to 2011 at the National Cancer Center Hospital East, Japan. We reevaluated preoperative thin-section computed tomography (TSCT) to determine solid size and pathological invasive size using the fourth edition of the WHO classification and reclassified them according to the new TNM classification. The discriminative power of survival curves by the seventh edition was compared with that by the eighth edition by using concordance probability estimates and Akaike’s information criteria calculated using a univariable Cox regression model. Pearson’s correlation coefficient was calculated to elucidate the correlation between radiological solid size using TSCT and pathological invasive size. ResultsThe overall survival curves in the eighth edition were well distinct at each clinical and pathological stage. The 5-year survival rates of patients with clinical and pathological stage 0 newly defined were both 100%. The concordance probability estimate and Akaike’s information criterion values of the eighth edition were higher than those of the seventh edition in discriminatory power for overall survival. Solid size on TSCT scan and pathological invasive size showed a positive linear relationship, and Pearson’s correlation coefficient was calculated as 0.83, which indicated strong correlation. ConclusionsThis TNM classification will be feasible regarding patient survival, and radiological solid size correlates significantly with pathological invasive size as a new T factor.

Development and validation of a major adverse transplant event (MATE) score to predict late graft loss in pediatric heart transplantation

There is inadequate power to perform a valid clinical trial in pediatric heart transplantation (HT) using a conventional end-point, because the disease is rare and hard end-points, such as death or graft loss, are infrequent. We sought to develop and validate a surrogate end-point involving the cumulative burden of post-transplant complications to predict death/graft loss to power a randomized clinical trial of maintenance immunosuppression in pediatric HT. Pediatric Heart Transplant Study (PHTS) data were used to identify all children who underwent an isolated orthotopic HT between 2005 and 2014 who survived to 6 months post-HT. A time-varying Cox model was used to develop and evaluate a surrogate end-point comprised of 6 major adverse transplant events (MATEs) (acute cellular rejection [ACR], antibody-mediated rejection [AMR], infection, cardiac allograft vasculopathy [CAV], post-transplant lymphoproliferative disease [PTLD] and chronic kidney disease [CKD]) occurring between 6 and 36 months, where individual events were defined according to international guidelines. Two thirds of the study cohort was used for score development, and one third of the cohort was used to test the score. Among 2,118 children, 6.4% underwent graft loss between 6 and 36 months post-HT, whereas 39% developed CKD, 34% ACR, 34% infection, 9% AMR, 4% CAV and 2% PTLD. The best predictive score involved a simple MATE score sum, yielding a concordance probability estimate (CPE) statistic of 0.74. Whereas the power to detect non-inferiority (NI), assuming the NI hazard ratio of 1.45 in graft survival was 10% (assuming 200 subjects and 6% graft loss rate), the power to detect NI assuming a 2-point non-inferiority margin was >85% using the MATE score. The MATE score reflects the cumulative burden of MATEs and has acceptable prediction characteristics for death/graft loss post-HT. The MATE score may be useful as a surrogate end-point to power a clinical trial in pediatric HT.

Incorporation of CEA Improves Risk Stratification in Stage II Colon Cancer

High-risk features are used to direct adjuvant therapy for stage II colon cancer. Currently, high-risk features are identified postoperatively, limiting preoperative risk stratification. We hypothesized carcinoembryonic antigen (CEA) can improve preoperative risk stratification for stage II colon cancer. The National Cancer Database (NCDB 2004–2009) was reviewed for stage II colon adenocarcinoma patients undergoing curative intent resection. A novel risk stratification including both traditional high-risk features (T4 lesion, <12 lymph nodes sampled, and poor differentiation) and elevated CEA was developed. Unadjusted Kaplan-Meier and adjusted Cox proportional hazards analyzed overall survival. Concordance Probability Estimates (CPE) assessed discrimination. Seventy-four thousand nine hundred forty-five patients were identified; 40,844 (54.5%) had CEA levels reported and were included. Chemotherapy administration was similar between normal and elevated CEA groups (23.8 vs. 25.1%, p = 0.003). Compared to patients with CEA elevation, 5-year overall survival in patients with normal CEA was improved (74.5 vs. 63.4%, p < 0.001). Restratification incorporating CEA resulted in reclassification of 6912 patients (16.9%) from average to high risk. CPE increased for novel risk stratification (0.634 vs. 0.612, SE = 0.005). The routinely available CEA test improved risk stratification for stage II colon cancer. CEA not only may improve staging of colon cancer but may also help guide additional therapy.

Poorly differentiated clusters as a prognostic marker at the invasive front of colon cancer.

621 Background: Histology at the invasive front of colon cancer can predict malignant potential. However, the optimal histological marker is yet to be established. This study compares the various invasive front histologic markers. Methods: A single-institution prospective database was queried for consecutive patients who underwent curative resection for Stage I-III colon adenocarcinoma from 2007-14. Histologic features were reviewed by a pathologist, including poorly differentiated clusters (PDC), tumor budding (BD), perineural invasion (PN), desmoplastic reaction (DR) and Crohn’s like reaction (CLR) at the invasive front, and WHO grade of the whole tumor. PDC was defined as cancer clusters of ≥ 5 cancer cells that lack a gland-like structure, and was graded into G1 ( &lt; 5), G2 (5-9) and G3 ( ≥ 10) by the highest number of the clusters /HPF. Clinical outcome included recurrence free survival (RFS) and peak hazard function of recurrence and death identified using the kernel-smoothing method. Predictive accuracy was measured with concordance probability estimate (CPE) for proportional hazards regression. Inter-observer agreement was assessed by weighted kappa values on diagnoses rendered by 3 pathologists on 50 randomly selected cases. Results: The study cohort consisted of 851 patients with a median follow up of 36 months. PDC, BD, PN, DR and CLR at the invasive front were significantly associated with RFS. When analyzed by stage, PDC, BD and PN were associated with RFS both in Stage II and Stage III, while the others were prognostic only in Stage III. CPE was the highest in PDC (0.642), indicating the best predictive accuracy, while it was the lowest in WHO grade (0.526). Weighted kappa was also the highest for PDC, indicating the best inter-observer agreement (PDC: 0.824, WHO grade: 0.568). The smoothed graph of the hazard function showed that the risk of recurrence was not only the highest but peaked earlier for PDC G3 (between0 and 12 months) than PDC G2 (between12 and 24 months) and G1 (no evident peak). Conclusions: Of the commonly evaluated histologic markers at the invasive front, PDC grade is the most predictive and reproducible. Further confirmatory investigations are warranted to determine if PDC can replace WHO grade.

Application of concordance probability estimate to predict conversion from mild cognitive impairment to Alzheimer's disease

ABSTRACTSubjects with mild cognitive impairment (MCI) have a substantially increased risk of developing dementia due to Alzheimer's disease (AD). Identifying MCI subjects who have high progression risk to AD is important in clinical management. Existing risk prediction models of AD among MCI subjects generally use either the AUC or Harrell's C-statistic to evaluate predictive accuracy. AUC is aimed at binary outcome and Harrell's C-statistic depends on the unknown censoring distribution. Gönen and Heller's K-index, also known as concordance probability estimate (CPE), is another measure of overall predictive accuracy for Cox proportional hazards (PH) models, which does not depend on censoring distribution. As a comprehensive example, using Alzheimer's Disease Neuroimaging Initiative (ADNI) data-set, we built a Cox PH model to predict the conversion from MCI to AD where the prognostic accuracy was evaluated using K-index.

Multi-institutional Validation Study of the American Joint Commission on Cancer (8th Edition) Changes for T and N Staging in Patients With Pancreatic Adenocarcinoma.

The aim of this study was to evaluate and validate the proposed 8th edition American Joint Committee on Cancer (AJCC) system for T and N staging of pancreatic adenocarcinoma. Investigators have questioned the clinical relevance and reproducibility of previous AJCC staging for pancreatic adenocarcinoma. Prospective databases at Memorial Sloan Kettering (MSK), Massachusetts General Hospital (MGH), and Johns Hopkins Hospital (JHH) were queried for patients who had undergone resection for pancreatic adenocarcinoma. Patients who underwent a margin-negative (R0) resection, and who had previously undergone pathologic review, were included. Patients were staged according to 7th edition AJCC criteria, as well as the proposed 8th edition system that includes different definitions of tumor size (T) and nodal status (N). The dataset was randomly split into training and test sets. Two thousand three hundred eighteen patients were identified who met inclusion criteria. Recursive partitioning on the training set (n = 1551) identified statistically appropriate cutoffs for tumor size (<2.2 cm, ≥4.8 cm,) and nodal status (no positive nodes, 1 to 3 positive nodes, ≥4 positive nodes) that supported the proposed 8th edition changes. Median survival in patients staged as T3, N0 by the 7th edition definitions was different between institutions (median Center 1, 24 mo; Center 2, 37 mo; Center 3, 29 mo; P = 0.054). This difference was not observed when patients were staged as T3, N0 by 8th edition criteria. Stage, and stage-specific outcome (7th edition), on the test set revealed a predominance of patients (68%) within the IIB subgroup, and a concordance probability estimate (CPE) of 0.57 for stage-specific survival. When assessed with 8th edition criteria, no stage subgroup had a majority of patients, and the CPE was 0.58. The proposed 8th edition changes for T and N classification were statistically valid and may allow a more reproducible system of T staging. This system also stratifies patients more evenly across stages without sacrificing prognostic accuracy.

Validation of the Web-Based IBTR! 2.0 Nomogram to Predict for Ipsilateral Breast Tumor Recurrence After Breast-Conserving Therapy

Toevaluate the IBTR! 2.0 nomogram, which predicts 10-year ipsilateral breast tumor recurrence (IBTR) after breast-conserving therapy with and without radiation therapy for breast cancer, by using a large, external, and independent cancer center database. We retrospectively identified 1898 breast cancer cases, treated with breast-conserving therapy and radiation therapy at the University Hospital Leuven from 2000 to 2007, with requisite data for the nomogram variables. Clinicopathologic factors were assessed. Two definitions of IBTR were considered where simultaneous regional or distant recurrence were either censored (conform IBTR! 2.0) or included as event. Validity of the prediction algorithm was tested in terms of discrimination and calibration. Discrimination was assessed by the concordance probability estimate and Harrell's concordance index. The mean predicted and observed 10-year estimates were compared for the entire cohort and for 4 risk groups predefined by nomogram-predicted IBTR risks, and a calibration plot was drawn. Median follow-up was 10.9years. The 10-year IBTR rates were 1.3% and 2.1%, according to the 2 definitions of IBTR. The validation cohort differed from the development cohort with respect to the administration of hormonal therapy, surgical section margins, lymphovascular invasion, and tumor size. In univariable analysis, younger age (P=.002) and a positive nodal status (P=.048) were significantly associated with IBTR, with a trend for the omission of hormonal therapy (P=.061). The concordance probability estimate and concordance index varied between 0.57 and 0.67 for the 2 definitions of IBTR. In all 4 risk groups the model overestimated the IBTR risk. In particular, between the lowest-risk groups a limited differentiation was suggested by the calibration plot. The IBTR! 2.0 predictive model for IBTR in breast cancer patients shows substandard discriminative ability, with an overestimation of the risk in all subgroups.

An International Multi-Institutional Validation of Age 55 Years as a Cutoff for Risk Stratification in the AJCC/UICC Staging System for Well-Differentiated Thyroid Cancer.

Age is a critical factor in outcome for patients with well-differentiated thyroid cancer. Currently, age 45 years is used as a cutoff in staging, although there is increasing evidence to suggest this may be too low. The aim of this study was to assess the potential for changing the cut point for the American Joint Committee on Cancer/Union for International Cancer Control (AJCC/UICC) staging system from 45 years to 55 years based on a combined international patient cohort supplied by individual institutions. A total of 9484 patients were included from 10 institutions. Tumor (T), nodes (N), and metastasis (M) data and age were provided for each patient. The group was stratified by AJCC/UICC stage using age 45 years and age 55 years as cutoffs. The Kaplan-Meier method was used to calculate outcomes for disease-specific survival (DSS). Concordance probability estimates (CPE) were calculated to compare the degree of concordance for each model. Using age 45 years as a cutoff, 10-year DSS rates for stage I-IV were 99.7%, 97.3%, 96.6%, and 76.3%, respectively. Using age 55 years as a cutoff, 10-year DSS rates for stage I-IV were 99.5%, 94.7%, 94.1%, and 67.6%, respectively. The change resulted in 12% of patients being downstaged, and the downstaged group had a 10-year DSS of 97.6%. The change resulted in an increase in CPE from 0.90 to 0.92. A change in the cutoff age in the current AJCC/UICC staging system from 45 years to 55 years would lead to a downstaging of 12% of patients, and would improve the statistical validity of the model. Such a change would be clinically relevant for thousands of patients worldwide by preventing overstaging of patients with low-risk disease while providing a more realistic estimate of prognosis for those who remain high risk.

Evaluation of a breast cancer nomogram to predict ipsilateral breast relapse after breast-conserving therapy

Background and purposeA nomogram to predict for the 10-year ipsilateral breast relapse (IBR) after breast-conserving therapy (BCT) for breast cancer (BC) was developed based on the ‘boost-no-boost’-trial with a concordance probability estimate (CPE) of 0.68. The aim of our study was to validate that algorithm. Material and methodsWe retrospectively identified 1787 BC cases, treated with BCT and radiotherapy at the University Hospitals Leuven from 2000 to 2007, without missing data of the nomogram variables. Clinicopathologic factors were assessed. Validity of the prediction model was tested in terms of discrimination and calibration. ResultsMedian follow-up time was 10.75years. The validation cohort differed with respect to the administration of a radiation boost, chemo- or hormonal therapy, age, tumour diameter or grade, ductal carcinoma in situ and hormone receptor positivity. On multivariable analysis, the omission of the boost was a significant prognosticator of IBR (p<0.01). The 10-year IBR-rate was 1.4%. The nomogram demonstrated suboptimal discrimination (CPE 0.54) and calibration, with an overestimation of the IBR-risk in general. ConclusionsThe predictive model for IBR in BC is imperfect in this more recent study population.

A scoring model for predicting survival following primary tumour resection in stage IV colorectal cancer patients with unresectable metastasis

Stage IV colorectal cancer patients with unresectable metastasis who undergo elective primary tumour resection experience heterogeneous post-operative survival. We aimed to develop a scoring model for predicting post-operative survival using pre-operative variables to identify patients who are least likely to experience extended survival following the procedure. Survival data were collected from stage IV colorectal cancer patients who had undergone elective primary tumour resection between January 1999 and December 2007. Coefficients of significant covariates from the multivariate Cox regression model were used to compute individual survival scores to classify patients into three prognostic groups. A survival function was derived for each group via Kaplan-Meier estimation. Internal validation was performed. Advanced age (hazard ratio, HR 1.43 (1.16-1.78)); poorly differentiated tumour (HR 2.72 (1.49-5.04)); metastasis to liver (HR 1.76 (1.33-2.33)), lung (HR 1.37 (1.10-1.71)) and bone (HR 2.08 ((1.16-3.71)); carcinomatosis (HR 1.68 (1.30-2.16)); hypoalbuminaemia (HR 1.30 (1.04-1.61) and elevated carcinoembryonic antigen levels (HR 1.89 (1.49-2.39)) significantly shorten post-operative survival. The scoring model separated patients into three prognostic groups with distinct median survival lengths of 4.8, 12.4 and 18.6 months (p < 0.0001). Internal validation revealed a concordance probability estimate of 0.65 and a time-dependent area under receiver operating curve of 0.75 at 6 months. Temporal split-sample validation implied good local generalizability to future patient populations (p < 0.0001). Predicting survival following elective primary tumour resection using pre-operative variables has been demonstrated with the scoring model developed. Model-based survival prognostication can support clinical decisions on elective primary tumour resection eligibility.

Pre-treatment tumor-specific growth rate as a temporal biomarker that predicts treatment failure and improves risk stratification for oropharyngeal cancer

To assess the relationship between tumor-specific growth rate (TSGR) and oropharyngeal cancer (OPC) outcomes in the HPV era. Primary tumor volume differences between a diagnostic and secondary scan separated ⩾7days without interval treatment were used to estimate TSGR, defined as percent volume growth/day derived from primary tumor volume doubling time for 85 OPC patients with known p16 status and smoking pack-years managed with (chemo)radiation. Variables were analyzed using Kruskal-Wallis or Fisher's exact test as appropriate. Log-rank tests and Cox proportional models analyzed endpoints. Using concordance probability estimates (CPE), TSGR was incorporated into RTOG 0129 risk grouping (0129RG) to assess whether TSGR could improve prognostic accuracy. Median time between scans was 35days (range 8-314). Median follow up was 26months (range 1-76). The 0129RG classification was: 56% low, 25% intermediate, and 19% high risk. Median TSGR was 0.74%/day (range 0.01-4.25) and increased with 0129RG low (0.41%), intermediate (0.57%) and high (1.23%) risk, respectively (p=0.015). TSGR independently predicted for TF (TSGR: HR (95%CI)=2.79, 1.67-4.65, p<0.001) in the Cox model. On CPE, prognostic accuracy for TF, disease-free survival and overall survival was improved when 0129RG was combined with TSGR. Dichotomizing 0129RG by median TSGR yielded no observed recurrences in low risk patients with TSGR<0.74% and demonstrated significant difference for intermediate risk (8% vs. 50% for TSGR<0.74% vs. ⩾0.74%, respectively, p<0.001). Tumor-specific growth rate correlates with increasing 0129RG and predicts treatment failure, potentially improving the prognostic strength and risk stratification of established 0129 risk groups.

Incorporation of postoperative CT data into clinical models to predict 5-year overall and recurrence free survival after primary cytoreductive surgery for advanced ovarian cancer

PurposeThe use of multivariable clinical models to assess postoperative prognosis in ovarian cancer increased. All published models incorporate surgical debulking. However, postoperative CT can detect residual disease (CT-RD) in 40% of optimally resected patients. The aim of our study was to investigate the added value of incorporating CT-RD evaluation into clinical models for assessment of overall survival (OS) and progression free survival (PFS) in patients after primary cytoreductive surgery (PCS). Methods212 women with PCS for advanced ovarian cancer between 01/1997 and 12/2011, and a contrast enhanced abdominal CT 1–7weeks after surgery were included in this IRB approved retrospective study. Two radiologists blinded to clinical data, evaluated all CT for the presence of CT-RD, and Cohen's kappa assessed agreement. Cox proportional hazards regression with stepwise selection was used to develop OS and PFS models, with CT-RD incorporated afterwards. Model fit was assessed with bootstrapped Concordance Probability Estimates (CPE), accounting for over-fitting bias by correcting the initial estimate after repeated subsampling. ResultsReaders agreed on the majority of cases (179/212, k=0.68). For OS and PFS, CT-RD was significant after adjusting for clinical factors with a CPE 0.663 (p=0.0264) and 0.649 (p=0.0008). CT-RD was detected in 37% of patients assessed as optimally debulked (RD<1cm) and increased the risk of death (HR: 1.58, 95% CI: 1.06–2.37%). ConclusionCT-RD is a significant predictor after adjusting for clinical factors for both OS and PFS. Incorporating CT-RD into the clinical model improved the prediction of OS and PFS in patients after PCS for advanced ovarian cancer.

Estimating the concordance probability in a survival analysis with a discrete number of risk groups.

A clinical risk classification system is an important component of a treatment decision algorithm. A measure used to assess the strength of a risk classification system is discrimination, and when the outcome is survival time, the most commonly applied global measure of discrimination is the concordance probability. The concordance probability represents the pairwise probability of lower patient risk given longer survival time. The c-index and the concordance probability estimate have been used to estimate the concordance probability when patient-specific risk scores are continuous. In the current paper, the concordance probability estimate and an inverse probability censoring weighted c-index are modified to account for discrete risk scores. Simulations are generated to assess the finite sample properties of the concordance probability estimate and the weighted c-index. An application of these measures of discriminatory power to a metastatic prostate cancer risk classification system is examined.