High-dimensional prediction of binary outcomes in the presence of between-study heterogeneity

Many prediction methods have been proposed in the literature, but most of them ignore heterogeneity between populations. Either only data from a single study or population is available for model building and evaluation, or when data from multiple studies make up the training dataset, studies are pooled before model building. As a result, prediction models might perform less than expected when applied to new subjects from new study populations. We propose a linear method for building prediction models with high-dimensional data from multiple studies. Our method explicitly addresses between-population variability and tends to select predictors that are predictive in most of the study populations. We employ empirical Bayes estimators and hence avoid selection bias during the variable selection process. Simulation results demonstrate that the new method works better than other linear prediction methods that ignore the between-study variability. Our method is developed for classification into two groups.

A new approach for modeling patient overall radiosensitivity and predicting multiple toxicity endpoints for breast cancer patients

INTRODUCTION

Evaluation of patient characteristics inducing toxicity in breast radiotherapy, using simultaneous modeling of multiple endpoints.

METHODS AND MATERIALS

In 269 early-stage breast cancer patients treated with whole-breast irradiation (WBI) after breast-conserving surgery, toxicity was scored, based on five dichotomized endpoints. Five logistic regression models were fitted, one for each endpoint and the effect sizes of all variables were estimated using maximum likelihood (MLE). The MLEs are improved with James-Stein estimates (JSEs). The method combines all the MLEs, obtained for the same variable but from different endpoints. Misclassification errors were computed using MLE- and JSE-based prediction models. For associations, p-values from the sum of squares of MLEs were compared with p-values from the Standardized Total Average Toxicity (STAT) Score.

RESULTS

With JSEs, 19 highest ranked variables were predictive of the five different endpoints. Important variables increasing radiation-induced toxicity were chemotherapy, age, SATB2 rs2881208 SNP and nodal irradiation. Treatment position (prone position) was most protective and ranked eighth. Overall, the misclassification errors were 45% and 34% for the MLE- and JSE-based models, respectively. p-Values from the sum of squares of MLEs and p-values from STAT score led to very similar conclusions, except for the variables nodal irradiation and treatment position, for which STAT p-values suggested an association with radiosensitivity, whereas p-values from the sum of squares indicated no association. Breast volume was ranked as the most significant variable in both strategies.

DISCUSSION

The James-Stein estimator was used for selecting variables that are predictive for multiple toxicity endpoints. With this estimator, 19 variables were predictive for all toxicities of which four were significantly associated with overall radiosensitivity. JSEs led to almost 25% reduction in the misclassification error rate compared to conventional MLEs. Finally, patient characteristics that are associated with radiosensitivity were identified without explicitly quantifying radiosensitivity.

Prevalence and prognosis of low-volume, oligorecurrent, hormone-sensitive prostate cancer amenable to lesion ablative therapy

OBJECTIVES

To describe the anatomical patterns of prostate cancer (PCa) recurrence after primary therapy and to investigate if patients with low-volume disease have a better prognosis as compared with their counterparts.

MATERIALS AND METHODS

Patients eligible for an 18-F choline positron-emission tomography (PET)-computed tomography (CT) were enrolled in a prospective cohort study. Eligible patients had asymptomatic biochemical recurrence after primary PCa treatment and testosterone levels >50 ng/mL. The number of lesions was counted per scan. Patients with isolated local recurrence (LR) or with ≤3 metastases (with or without LR) were considered to have low-volume disease and patients with >3 metastases to have high-volume disease. Descriptive statistics were used to report recurrences. Cox regression analysis was used to investigate the influence of prognostic variables on the time to developing castration-resistant PCa (CRPC).

RESULTS

In 208 patients, 625 sites of recurrence were detected in the lymph nodes (N1/M1a: 30%), the bone (18%), the prostate (bed; 11%), viscera (4%), or a combination of any of the previous (37%). In total, 153 patients (74%) had low-volume recurrence and 55 patients (26%) had high-volume recurrence. The 3-year CRPC-free survival rate for the whole cohort was 79% (95% confidence interval 43-55), 88% for low-volume recurrences and 50% for high-volume recurrences (P < 0.001). Longer PSA doubling time at time of recurrence and low-volume disease were associated with a longer time to CRPC.

CONCLUSIONS

Three out of four patients with PCa with a 18-F choline PET-CT-detected recurrence have low-volume disease, potentially amenable to local therapy. Patients with low-volume disease have a better prognosis as compared with their counterparts. Lymph node recurrence was the most dominant failure pattern.

Pitfalls in Prediction Modeling for Normal Tissue Toxicity in Radiation Therapy: An Illustration With the Individual Radiation Sensitivity and Mammary Carcinoma Risk Factor Investigation Cohorts

PURPOSE

To identify the main causes underlying the failure of prediction models for radiation therapy toxicity to replicate.

METHODS AND MATERIALS

Data were used from two German cohorts, Individual Radiation Sensitivity (ISE) (n=418) and Mammary Carcinoma Risk Factor Investigation (MARIE) (n=409), of breast cancer patients with similar characteristics and radiation therapy treatments. The toxicity endpoint chosen was telangiectasia. The LASSO (least absolute shrinkage and selection operator) logistic regression method was used to build a predictive model for a dichotomized endpoint (Radiation Therapy Oncology Group/European Organization for the Research and Treatment of Cancer score 0, 1, or ≥2). Internal areas under the receiver operating characteristic curve (inAUCs) were calculated by a naïve approach whereby the training data (ISE) were also used for calculating the AUC. Cross-validation was also applied to calculate the AUC within the same cohort, a second type of inAUC. Internal AUCs from cross-validation were calculated within ISE and MARIE separately. Models trained on one dataset (ISE) were applied to a test dataset (MARIE) and AUCs calculated (exAUCs).

RESULTS

Internal AUCs from the naïve approach were generally larger than inAUCs from cross-validation owing to overfitting the training data. Internal AUCs from cross-validation were also generally larger than the exAUCs, reflecting heterogeneity in the predictors between cohorts. The best models with largest inAUCs from cross-validation within both cohorts had a number of common predictors: hypertension, normalized total boost, and presence of estrogen receptors. Surprisingly, the effect (coefficient in the prediction model) of hypertension on telangiectasia incidence was positive in ISE and negative in MARIE. Other predictors were also not common between the 2 cohorts, illustrating that overcoming overfitting does not solve the problem of replication failure of prediction models completely.

CONCLUSIONS

Overfitting and cohort heterogeneity are the 2 main causes of replication failure of prediction models across cohorts. Cross-validation and similar techniques (eg, bootstrapping) cope with overfitting, but the development of validated predictive models for radiation therapy toxicity requires strategies that deal with cohort heterogeneity.

A let-7 microRNA polymorphism in the KRAS 3'-UTR is prognostic in oropharyngeal cancer

INTRODUCTION

This study aimed to investigate the effect of genetic polymorphisms in miRNA sequences, miRNA target genes and miRNA processing genes as additional biomarkers to HPV for prognosis in oropharyngeal squamous cell carcinoma (OPSCC) patients. Secondarily, the prevalence of HPV-associated OPSCC in a European cohort was mapped.

METHODS

OPSCC patients (n=122) were genotyped for ten genetic polymorphisms in pre-miRNAs (pre-mir-146a, pre-mir-196a2), in miRNA biosynthesis genes (Drosha, XPO5) and in miRNA target genes (KRAS, SMC1B). HPV status was assessed by p16 immunohistochemistry (IHC) and high-risk HPV in situ hybridization (ISH) or by p16 IHC and PCR followed by enzyme-immunoassay (EIA). Overall and disease specific survival were analysed using Kaplan-Meier plots (log-rank test). Cox proportional hazard model was used to calculate hazard ratios (HR).

RESULTS

The overall HPV prevalence rate in our Belgian/Dutch cohort was 27.9%. Patients with HPV(+) tumours had a better 5-years overall survival (78% vs. 46%, p=0.001) and a better 5-years disease specific survival (90% vs. 70%, p=0.016) compared to patients with HPV(-) tumours. In multivariate Cox analysis including clinical, treatment and genetic parameters, HPV negativity (HR=3.89, p=0.005), advanced T-stage (HR=1.81, p=0.050), advanced N-stage (HR=5.86, p=0.001) and >10 pack-years of smoking (HR=3.45, p=0.012) were significantly associated with reduced overall survival. The variant G-allele of the KRAS-LCS6 polymorphism was significantly associated with a better overall survival (HR=0.40, p=0.031).

CONCLUSIONS

Our results demonstrate that OPSCC patients with the KRAS-LCS6 variant have a better outcome and suggest that this variant may be used as a prognostic biomarker for OPSCC.