IN VITRO METHOD FOR PREDICTING THE RISK OF DEVELOPING A BREAST LATE EFFECT AFTER RADIOTHERAPY

Abstract

The present invention is drawn to a new diagnosis method and a calculator for predicting the risk of developing a breast late effect (BLE), which is defined as atrophic skin, telangiectasia, induration (fibrosis), necrosis or ulceration, in a subject after radiotherapy (RT), by using Radiation Induced late effect using T-Lymphocyte Apoptosis (RILA) and clinical parameters. The invention is also drawn to diagnosis kits for the implementation of the method and a nomogram.

Claims

1. An in vitro method for diagnosing the risk of developing breast late effects (BLE) after radiotherapy in a subject comprising the steps of: a. Determining the values of at least one biochemical marker relevant to assess an increasing risk of radiation toxicity in a subject, in particular selected from RILA, proteins and/or genes of radiosensitivity, from a biologic sample of said subject, preferably a blood sample of said subject; b. Determining the level of at least two clinical parameters relevant to assess an increasing risk of radiation toxicity in a subject, in particular selected from age, breast volume, adjuvant hormonotherapy, boost, node irradiation, and tobacco smoking; c. Combining said data through a multivariate Cox function to obtain an end value to determine the risk of developing BLE; wherein a multivariate Cox regression from said multivariate Cox function is obtained through the following method: i) Construction of the multivariate Cox regression by combination of said biochemical markers and said clinical parameters; and ii) Analyzing said multivariate Cox regression to assess the independent discriminative value of biochemical markers and clinical parameters.

2. The method according to claim 1, wherein said at least one biochemical marker consists in RILA which is based on the response of CD4 and/or CD8 after radiotherapy (RT), preferably based on the response of CD8 after radiotherapy (RT).

3. The method according to claim 1, wherein said at least two clinical parameters are at least tobacco smoking habits and adjuvant hormonotherapy.

4. The method according to claim 1, wherein said biochemical marker is used in combination with proteins and/or genes of radiosensitivity.

5. The method according to claim 4, wherein said proteins of radiosensitivity are selected from the group consisting of AK2, HSPA8, ANX1, APEX1 and ID2.

6. The method according to claim 4, wherein said genes of radiosensitivity are selected from the group consisting of TGFbeta, SOD2, TNFalpha and XRCC1.

7. The method according to claim 1, wherein said multivariate Cox function consist of:
Hazard (experiencing the BLE)=baseline hazard*exp(1*RILA+2*(Adjuvant Hormonotherapy [0=no; 1=yes])+3*(Tobacco smoking habits [0=no; 1=yes]), wherein: 1 is comprised between 0.077 and 0.010; 2 is comprised between 0.283 and 1.980; 3 is comprised between 0.063 and 0.965.

8. The method according to claim 1, wherein the end value of the said multivariate Cox function is used for the choice of a suitable treatment for the patient, such as an appropriate radiotherapy dosage regimen, wherein: if the patient presents a risk to develop breast late effect, the appropriate radiotherapy dosage regimen will be decreased for example by delivery of partial-breast hypofractionated treatment; if the patient presents low risk or no risk to develop breast late effect, the appropriate radiotherapy dosage regimen will be increased, for example by delivery of hypofractionated treatment (5 or 16 fractions).

9. The method according to claim 1, wherein the end value of the said multivariate Cox function is used to choose between a mastectomy or conserving surgery, preferably if said end value is more than 20% a decision of mastectomy instead of conserving surgery would be considered, and conversely.

10. The method according to claim 1, wherein the end value of the multivariate Cox regression is used in the decision of performing an immediate breast reconstruction after conserving surgery or mastectomy, preferably if said end value is less than 8% said immediate breast reconstruction after conserving surgery or mastectomy would be considered.

11. The method according to claim 1, wherein the functional representation is a nomogram.

12. A kit suitable for collecting data of a subject comprising: a box/container and bag suited for biological transportation of biological sample, in particular blood sample and forms to be completed by the patient and/or the nurse and/or the physician, specifically designed and necessary to run the radiosensitivity test and the nomogram analysis.

13. A kit suitable for detecting the risk of developing of breast late effect (BLE) in a subject comprising: reagents for determining the values of at least one biochemical markers according to claim 1; and optionally means of collecting information on at least two clinical parameters according to claim 1.

14. A kit according to claim 13, comprising: some or all specific reagents required to run the RILA assay in an independent laboratory, where the irradiation of the sample will be run by a linear accelerator or a lab irradiator, and optionally specific forms to be completed by the patient and/or the nurse and/or the physician, specifically designed and required to run the radiosensitivity test and the nomogram analysis.

15. A system including a machine-readable memory, such as a computer or/and a calculator, and a processor configured to compute said multivariate Cox function according to claim 1 and preferably additionally a module for executing a software to build a nomogram and calculate the instantaneous risk for the subject to develop a BLE after radiotherapy.

Description

DESCRIPTION OF THE FIGURES

[0145] FIG. 1: ROC Curve for pooled data. The ROC curve was drawn by plotting the sensitivity versus 1-specificity after classification of patients and according to the values obtained for the logistic function for different thresholds (from 0 to 1).

[0146] FIG. 2: Scatter plot of RILA according to breast late effect status for pooled data.

EXAMPLES

[0147] Materials and Methods

[0148] Patient Analysis

[0149] Analysis (pooled data) was performed among 502 breast cancer patients included in the multicentre PHRC (Programme Hospitalier de Recherche Clinique) study evaluating the predictive RILA test in patients treated by radiotherapy and conserving surgery in terms of BLE during the follow-up. 434 patients were eligible for this study. Patients were included between 15 Jan. 2007 and 11 Jul. 2011 and followed with a median follow-up of 38.6 months. A specific questionnaire (Case Report Form) was filled out for each patient including general social demographics administrative data, risk factors, and at each visit, clinical and biological and treatment items, and histologic data described in Table 1. At baseline, 16 patients were excluded for major deviation, one patient withdrew consent before any treatment and blood samples were not technically available for 29 participants. All these patients (n=46) were treated according to current guidelines and were not included for analysis since no data were collected. Thus, RILA and complete RT were performed for 456 patients (90.8%) before entering follow-up. 434 (86.5%) patients were followed for at least 36 months according to the protocol. The 22 other patients interrupted the planned follow-up before 36 months.

[0150] Multivariate model was built for a total of 415 patients (reference population) with completed data for the selected parameters.

[0151] Radiation-Induced CD8 T-Lymphocyte Apoptosis (RILA) Procedure

[0152] The protocol was adapted from studies of Ozsahin et al. (Ozsahin, Crompton et al. 2005). Before RT one blood sample was collected from each patient in a 5-ml heparinized tube. 200 L of blood was aliquoted into a 6-well plate. All tests were carried out in triplicate for both 0 and 8 Gy. Irradiations (single dose of 8 Gy in a 25 cm25 cm field size at a dose rate of 1 Gy/min) were delivered after 24 h (H24) using a linear accelerator (2100 EX, 200 UM/min, Varian, US) in the Radiation Department. Control cells were removed from the incubator and placed for the same period of time under the Linac but without radiation treatment. After irradiation, the flasks were immediately incubated at 37 C. (5% CO.sub.2). After a further forty-eight hours (H72), it was labeled with anti-human CD8-FITC antibody (10 L/tests, Becton Dickinson, USA). After addition of lysis buffer (Becton Dickinson, USA), propidium iodide (Sigma, France) and RNAse (Qiagen, France) was added to each tube and prepared for flow cytometry (FACS).

[0153] Preparation and Delivery of Radiotherapy

[0154] RT was delivered in the supine position to ensure reproducibility during simulation and treatment. The planning target volume included the whole breast (WB) and the regional lymph nodes (RLN) if necessary. Only photons were allowed for WB irradiation thus allowing standardization of treatment across centers.

[0155] A median dose of 50 Gy to the target volume was recommended. The field arrangement involved the use of an anterior photon field in the supraclavicular region and a combination of anterior electrons/photons to the internal mammary nodes at 44-50 Gy. A daily dose of 50 Gy to the WB was delivered by two opposed tangential fields; a boost in the surgical bed up to 10-16 Gy was given when necessary. Fractionation was 2 Gy per fraction, 5 days a week. Calculation used 3-D dosimetry. The ICRU report 62 prescription points were used for prescribing dose. As a minimum, on-line portal imaging was obtained each day for the first three days and once a week during the rest of the course of treatment.

[0156] Adjuvant Systemic Therapies

[0157] Chemotherapy (CT) regimen when indicated (in case of node positivity and grade 3) consisted either of 6 cycles of FEC 100 [5 FU (500 mg/m 2), epirubicin (100 mg/m 2), cyclophosphamide (500 mg/m 2)] on day 1 and repeated every 21 days or 3 cycles of FEC 100 followed by 3 cycles of docetaxel (100 mg/m 2) every three weeks. In case of HER2 overexpression or gene amplification, trastuzumab (beginning with a loading dose of 8 mg/kg) was added to the protocol (6 mg/kg every 3 weeks for 1 year). Hormonotherapy (HT: tamoxifen or aromatase inhibitor) was started after surgery or after the end of RT and given daily for five years.

[0158] End-Point Assessments: Identification of Biomarkers and Relevant Covariables as Prognostic Factors of BLE on Reference Population

[0159] The primary objective was the predictive role of RILA in radiation-induced grade BLE (defined as atrophic skin, telangiectasia, induration (fibrosis), necrosis or ulceration). Secondary objectives were the incidence of acute side effects, local recurrence, relapse-free survival (RFS), breast fibrosis-free survival (BF-FS), breast fibrosis-relapse-free survival (BF-RFS) and overall survival (OS). Acute and late side effects were assessed and graded according to the CTC v3.0 scale (Trotti, Colevas et al. 2003).

[0160] Toxicity evaluations were performed at baseline, every week during RT, one, three and six months after the last RT fraction, every 6 months up to month 36. Each evaluation was assessed by the physicians blinded for RILA. The most severe BLE observed during the follow-up after RT was considered as the primary endpoint. The most severe late effects (lung, cardiac) observed from 12 weeks to 3 years post RT and the most severe acute side effects (skin and lung mainly) observed from the start of RT to 12 weeks post RT were considered as the secondary endpoints. Toxicities were evaluated using all the possible definitions described in the scale Dermatology/skin area, pulmonary/upper respiratory and cardiac general (Trotti, Colevas et al. 2003).

[0161] All endpoints were defined as the interval between the start of RT and following the first events: death for OS, local or contralateral or distant recurrence or death for RFS, grade BLE for BF-FS, and first event of RFS and BF-FS for BF-RFS (Peto, Pike et al. 1977). Censoring patients were patients alive at the last follow-up visit for OS, patients alive and without relapse for RFS, patients alive who never experienced a grade 2 BLE for BF-FS and patients alive who never experienced grade BLE or relapse for BF-RFS.

[0162] Sample Size Calculation and Statistical Analysis

[0163] To test the prognostic value of RILA rate on the occurrence of BLE, we started from the results of our preliminary study (Azria, Gourgou et al. 2004). Details are presented in the following protocol. Briefly, based on a2=0.54, an estimated complication rate of 15% with a two-sided error of 0.05 and a error of 0.05 (power=0.95), 430 patients had to be included. The number of patients was increased by at least 15% (n=494) to take into account loss to follow-up and the impact of the boost on BLE.

[0164] 2 is the variance of the studied variable (log CD8) and is the rate of complication/toxicity expected events.

[0165] The cumulative incidences of complications as a function of the prognostic variables were calculated using a non-parametric model (Pepe and Mod 1993). The main statistical procedure included a multivariate analysis using the Fine et al. model of competing risks (Fine 2001) for the assessment of the impact of RILA rate on the occurrence of BLE in the presence of other events (such as relapse or death) that are considered as competing risk events in this pathology. For multivariate analysis, selected factors were the baseline parameters with a p-value (statistical significance) less than 0.20 in univariate analysis. Final model was defined using backward stepwise selection (p<0.15) and a step by step method was used to include only the significant parameters (p<0.05) or clinically relevant and/or (p<0.10).

[0166] Data were summarized by frequency and percentage for categorical variables and by median and range for continuous variables. Absolute changes in RILA counts before and after irradiation were evaluated as continuous and categorical variables. Three categories were constructed around the 33% quantiles (<12, 12-20, and 20) and then merged in two categories (<12 and 12).

[0167] OS, RFS, BF-FS and BF-RFS rates were estimated by the Kaplan-Meier method (Kaplan and Meier 1958). Ninety-five percent confidence intervals (95% C1) were also determined.

[0168] Univariate analysis and multivariate analysis were performed using the Cox proportional hazard's regression model (Cox, et al. 1984) to estimate the hazard ratio including baseline characteristics and treatment parameters. Comparisons were performed using the log-rank test for univariate analysis. Independent effects were evaluated from the likelihood ratio statistics.

[0169] Impact of RILA on breast fibrosis-relapse-free survival (BF-RFS) was assessed. The cumulative incidence of BLE and relapse or death were estimated from a competing risk model using estimates obtained from the cause-specific hazard functions and the composite RFS and BF-FS distribution (Arriagada, Rutqvist et al. 1992) and compared using Gray's test.

[0170] Median follow-up was estimated with the inverse Kaplan-Meier method. A p-value less than was considered as significant. All statistical tests were two-sided. Stata was used for all statistical analyses (version 13.0) and the SAS macro % cif was used for Gray's test.

[0171] To complement analysis, receiver-operator characteristic (ROC) curve analyses for RILA were performed to identify patients who experienced at least a grade 2 BLE during the follow-up (Kramar, Faraggi et al. 2001). The empirical areas under the ROC curves (AUC) and the respective 95% CI were used for RILA to determine the sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV).

Example 1: Descriptive Statistical Analysis for Modelling

[0172] The analysis was performed on selected data from studies for all patients observing independent variables, as presented in Table 1 below.

TABLE-US-00001 TABLE 1 Prognostic factors characteristics of BLE N = 415 (reference population) Age (Year) Mean (std) 56.52 (9.853) Median (range) 56.00 (29.00:88.00) Baseline RILA Mean (std) 15.96 (8.597) Median (range) 15.39 (0.74:52.82) Breast Volume (cm.sup.3) n = 410 Mean (std) 7.47 (13.126) Median (range) 4.99 (0.00:228.15) Age (Year) 55 (200/415) 48.2% >55 (215/415) 51.8% Adjuvant Hormonotherapy No (99/415) 23.9% Yes (316/415) 76.1% Boost (complementary dose of irradiation) No (6/415) 1.4% Yes (409/415) 98.6% Node irradiation No (104/415) 25.1% Yes (311/415) 74.9% Tobacco smoking Non smoker (277/415) 66.7% Ex-smoker or Smoker (138/415) 33.3%

[0173] According to the CTCAE V3.0 (Trotti, Colevas et al. 2003), toxicity was assessed and patients were divided in two groups (with or without BLE).

[0174] The major endpoint was the identification of patients with or without BLE during follow-up. The first stage consisted in identification of factors which differed significantly between these groups by unidimensional analysis and using the log rank test.

[0175] The second stage consisted in analysis of multivariate Cox proportional hazard model to assess the independent parameters for the diagnosis of BLE and to estimate the effect size defined as Hazard ratio (HR).

[0176] Moreover, the overall diagnosis values were estimated by the Receiving Operating Characteristic curves (ROC Curves). The diagnostic value of RILA was assessed by sensitivity (Se), specificity (Sp), positive and negative predictive values (PPV, NPV).

[0177] For such diagnosis or prognosis of a responding phenotype, a responding condition or test outcome is considered a positive result, while a non-responding condition or test outcome is considered a negative result. True and false positive results, NPV, PPV, specificity, sensitivity are defined and calculated as follows:

TABLE-US-00002 Condition (responding) Positive Negative Test outcome Positive True Positive (TP) False positive (FP) (responding) Negative False negative (FN) True negative (TN)

PPV=TP/(TP+FP)

NPV=TN/(TN+FN)

Specificity=TN/(TN+FP)

Sensitivity=TP/(TP+FN)

[0178] ROC or ROC curve is a tool for diagnostic test evaluation, wherein the true positive rate (Sensitivity) is plotted in function of the false positive rate (1Specificity) for different cut-off points of a parameter (FIG. 1) after classification of patients. Each point on the ROC curve represents a sensitivity/specificity pair corresponding to a particular decision threshold (from 0 to 1). The area under the ROC curve (AUC) is a measure of how well a parameter can distinguish between two diagnostic groups (diseased/normal). The accuracy of the test depends on how well the test separates the group being tested into those with and without the disease in question. Accuracy is measured by the area under the ROC curve. An area of 1 represents a perfect test; an area of 0.5 represents a worthless test.

[0179] It is usually acknowledged that a ROC area under the curve has a value superior to 0.7 is a good predictive curve for diagnosis. The ROC curve has to be acknowledged as a curve allowing prediction of the diagnosis quality of the method.

[0180] The diagnostic value (area under the curve) of RILA marker is presented in table 2.

TABLE-US-00003 TABLE 2 Diagnosis value (area under the ROC curve sd) of biochemical markers for significant BLE for the patients during the follow-up; Sensitivity, specificity and predictive value of the RILA as a BLE function Pooled data Nb of events (AE) 60/415 AUC 0.6119 CI95% [0.531-0.692] cut-off 20 Se 0.80 Sp 0.34 PPV 0.17 NPV** 0.91 cut-off <12 Se 0.55 Sp 0.67 PPV** 0.22 NPV 0.90 Se: Sensitivity; Sp: Specificity; PPV: positive predictive value; NPV: negative predictive value.

[0181] Results

[0182] In the analysis, it has been confirmed that the negative predictive value (RILA with cut-off 20) was excellent with more than 90% and may be useful for patients.

[0183] In terms of clinical application, patients with high RILA (RILA 20) will not observe a BLE and will be proposed to hypofractionation regimen.

[0184] Conclusion

[0185] The RILA marker is a good marker of BLE during the follow-up leading to develop a personalized treatment according to the patient profile.

[0186] The inventors also demonstrated that the combination of RILA marker and two clinical parameters being tobacco smoking habit and adjuvant hormonotherapy, the AUC is improved 0.68 IC95% [0.608-0.749] in comparison to RILA alone (0.61), and for optimal threshold: [0187] Se: 0.80 [0188] Sp: 0.487 [0189] VPP: 0.209 [0190] VPN: 0.935

[0191] These results showed that the combination of the three parameters (RILA and two clinical parameters being tobacco smoking habit and adjuvant hormonotherapy), the specificity of the in vitro method is improved and the negative predictive value is even more than 93%.

[0192] The following example made with a referenced population of 415 patients, is an illustrative example without limiting the scope of the invention.

Example 2: Determination of the Multivariate Cox Function

[0193] A total of 415 patients with breast cancer and treated by adjuvant radiotherapy after conserving surgery were selected by multivariate Cox regression using independent parameters.

[0194] The overall prevalence of BLE was 14.5% (60 events among 415 patients).

[0195] Diagnosis of Significant Breast Late Effect (BLE)

[0196] Scatter plots were drawn for each study to compare the level of RILA according to BLE status. Patients with BLE presented a low level of RILA.

[0197] Moreover, the risk of BLE was higher with low value of RILA. The risk of BLE was significantly increased combined with several clinical parameters (tobacco smoking habits and adjuvant hormonotherapy; Table 3).

[0198] Breast late effects (BLE) were evaluated clinically by expert clinicians and graded using the international grading score for toxicity CTCAE V3.0 well known from man skilled in the art (Trotti, Colevas et al. 2003). The NCI Common Terminology Criteria for Adverse Events (CTCAE) v3.0 is a descriptive terminology which can be utilized for Adverse Event (AE) reporting. A grading (severity) scale is provided for each AE term.

TABLE-US-00004 TABLE 3 Multivariate analyses to detect independent prognostic factors for Breast Fibrosis free survival (BF-FS) using the Proportional hazards Cox model (Pooled data). Pooled data N* = 535/703 Nb of events (AE): 136 Median follow-up HR CI95% p-value Univariate model N = 702 Nb of events (AE): 60 RILA 0.96 0.940-0.981 <0.001 Univariate model N = 415 RILA 0.96 0.927-0.990 0.011 Mulivariate model N* = 415 Concordance Harrell'C = 0.6876 RILA 0.96 0.926-0.990 0.01 Tobacco Smoking 0.085 No 1 Active/former 1.57 0.939-2.625 Adjuvant Hormonotherapy 0.009 No 1 Yes 3.10 1.327-7.243 Mulivariate model** N* = 415 Concordance Harrell'C = 0.7004 RILA 0.96 0.927-0.992 0.015 Tobacco Smoking 0.080 No 1 Active/former 1.59 0.946-2.671 Adjuvant Hormonotherapy 0.015 No 1 Yes 2.88 1.229-6.7573 HR = Hazard ratio estimated by multivariate Cox regression. RILA = radiation-induced CD8 T-lymphocyte apoptosis *Number of patients included in the model/included population **adjusted multivariate model on age (55), Boost(N/Y), node irradiation(N/Y)

[0199] Results

[0200] Multivariate models identified three parameters (RILA, tobacco smoking and adjuvant hormonotherapy) as independent parameters with an increased risk of BLE for active/former tobacco smoking patients (HR=1.57 CI95%[0.939-2.625]) and for patients treated by hormonotherapy (HR=3.10 CI95%[1.327-7.243]) and a decrease of risk for elevated level of RILA (HR=0.96 CI95% [0.926-0.990]). The other clinical parameters (age, boost and node irradiation) were integrated in multivariate model for adjustment because of clinically relevance. Finally theses parameters were not selected for definitive model.

[0201] Conclusion

[0202] The combination of clinical parameters (tobacco smoking habits and adjuvant hormonotherapy) and RILA allowed prediction of the probability to develop a breast late effect and integration of clinical and treatment parameters. All these parameters improve the estimation of the risk evaluated only by RILA.

[0203] It has been confirmed that the negative predictive value was excellent (more than 90%) and may be useful for patients. In terms of clinical application, patients with low risk of breast recurrences and high RILA will be proposed hypofractionation regimen or partial breast irradiation. The number of fractions will be reduced and higher dose per fraction will be proposed. This scheme will be delivered safely thanks to the RILA assay.

[0204] In addition, patients who desire immediate breast reconstruction and radiotherapy will be offered this strategy only in case of low risk (risk less than about 8%) of breast fibrosis evaluated with a high value of RILA.

Example 3: Construction of a Nomogram Determining the Probability for a Patient of Developing Breast Fibrosis During the Follow-Up after Radiotherapy

[0205] Nomogram was built according to the method described by Iasonos et al. (2008) using, as an illustrative example, the estimated parameters by the multivariate Cox function including selected parameters identified as being relevant according to Example 2:

TABLE-US-00005 RILA p-value = 0.01 Tobacco smoking p-value = 0.085 Adjuvant hormonotherapy p-value = 0.009.

[0206] The optimal beta-coefficients may be obtained by classical statistical analysis and a nomogram may be easily built by a man skilled in the art based on these coefficients and resulted hazard (experiencing BLE).

[0207] For example, a patient with a RILA=10% (determined as disclosed above on a blood sample), non-smoker and treated by adjuvant hormonotherapy, we can calculate the following risk step by step:

TABLE-US-00006 1/RILA = 10% => 82 points 2/Non-smoker/Tobacco = 0 => 0 points 3/Treated by Adjuvant Hormonotherapy/Adj_HRM = 1 => 46 points 4/Total points: 82 + 0 + 46 = 128 points 5/Probability of a 3y-BF-FS is between 0.85 and 0.80 corresponding to a risk of developing a breast late effect between 15 and 20%.

REFERENCES

[0208] Al-Ghazal, S. K., L. Fallowfield, et al. (1999). Does cosmetic outcome from treatment of primary breast cancer influence psychosocial morbidity? European journal of surgical oncology: the journal of the European Society of Surgical Oncology and the British Association of Surgical Oncology 25(6): 571-573. [0209] Arriagada, R., L. E. Rutqvist, et al. (1992). Competing risks determining event-free survival in early breast cancer. Br J Cancer 66(5): 951-957. [0210] Azria, D., M. Betz, et al. (2012). Identifying patients at risk for late radiation-induced toxicity. Critical reviews in oncology/hematology 84 Suppl 1: e35-41. [0211] Azria, D., S. Gourgou, et al. (2004). Concomitant use of tamoxifen with radiotherapy enhances subcutaneous breast fibrosis in hypersensitive patients. Br J Cancer 91(7): 1251-1260. [0212] Azria, D., M. Ozsahin, et al. (2008). Single Nucleotide Polymorphisms, Apoptosis, and the Development of Severe late Adverse Effects After Radiotherapy. Clin Cancer Res 2008 October; 14(19): 6284-6288. [0213] Azria, D., O. Riou, et al. (2015). Radiation-induced CD8 T-lymphocyte Apoptosis as a Predictor of Breast Fibrosis After Radiotherapy: Results of the Prospective Multicenter French Trial. EBioMedicine 2(12): 1965-1973. [0214] Cox, D. R., et al., Eds. (1984). Analysis of Survival Data., New York: Chapman & Hall. [0215] Fine, J. P. (2001). Regression modeling of competing crude failure probabilities. Biostatistics 2(1): 85-97. [0216] Harrell, F. E., Jr. and Y. C. Shih (2001). Using full probability models to compute probabilities of actual interest to decision makers. Int J Technol Assess Health Care 17(1): 17-26. [0217] Iasonos, A., D. Schrag et al. (2008). How to build and interpret a Nomogram for Cancer Prognosis. Journal of Clinical Oncology. Review Article. Volume 26, Number 8, March 2008. [0218] Kaplan, E. L. and P. Meier (1958). Non-parametric estimation from incomplete observation. J Am Stat Assoc 53: 457-481. [0219] Kramar, A., D. Faraggi, et al. (2001). mROC: a computer program for combining tumour markers in predicting disease states. Comput Methods Programs Biomed 66(2-3): 199-207. [0220] Ozsahin, M., N. E. Crompton, et al. (1997). Rapid assay of intrinsic radiosensitivity based on apoptosis in human CD4 and CD8 T-lymphocytes. Int. J. Radiation Oncology Biol. Phys., Vol. 38, N.sup.o 2, pp. 429440. [0221] Ozsahin, M., N. E. Crompton, et al. (2005). CD4 and CD8 T-lymphocyte apoptosis can predict radiation-induced late toxicity: a prospective study in 399 patients. Clin Cancer Res 11(20): 7426-7433. [0222] Pepe, M. S. and M. Mod (1993). Kaplan-Meier, marginal or conditional probability curves in summarizing competing risks failure time data? Stat Med 12(8): 737-751. [0223] Peto, R., M. C. Pike, et al. (1977). Design and analysis of randomized clinical trials requiring prolonged observation of each patient. II. analysis and examples. Br J Cancer 35(1): 1-39. [0224] Trotti, A., A. D. Colevas, et al. (2003). CTCAE v3.0: development of a comprehensive grading system for the adverse effects of cancer treatment. Semin Radiat Oncol 13(3): 176-181.