Bayesian information criterion approximations to Bayes factors for univariate and multivariate logistic regression models.

Schwarz's criterion, also known as the Bayesian Information Criterion or BIC, is commonly used for model selection in logistic regression due to its simple intuitive formula. For tests of nested hypotheses in independent and identically distributed data as well as in Normal linear regression, previous results have motivated use of Schwarz's criterion by its consistent approximation to the Bayes factor (BF), defined as the ratio of posterior to prior model odds. Furthermore, under construction of an intuitive unit-information prior for the parameters of interest to test for inclusion in the nested models, previous results have shown that Schwarz's criterion approximates the BF to higher order in the neighborhood of the simpler nested model. This paper extends these results to univariate and multivariate logistic regression, providing approximations to the BF for arbitrary prior distributions and definitions of the unit-information prior corresponding to Schwarz's approximation. Simulations show accuracies of the approximations for small samples sizes as well as comparisons to conclusions from frequentist testing. We present an application in prostate cancer, the motivating setting for our work, which illustrates the approximation for large data sets in a practical example.

Repeated kidney re-transplantation-the Eurotransplant experience: a retrospective multicenter outcome analysis.

In Eurotransplant kidney allocation system (ETKAS), candidates can be considered unlimitedly for repeated re-transplantation. Data on outcome and benefit are indeterminate. We performed a retrospective 15-year patient and graft outcome data analysis from 1464 recipients of a third or fourth or higher sequential deceased donor renal transplantation (DDRT) from 42 transplant centers. Repeated re-DDRT recipients were younger (mean 43.0 vs. 50.2 years) compared to first DDRT recipients. They received grafts with more favorable HLA matches (89.0% vs. 84.5%) but thereby no statistically significant improvement of patient and graft outcome was found as comparatively demonstrated in 1st DDRT. In the multivariate modeling accounting for confounding factors, mortality and graft loss after 3rd and ≥4th DDRT (P < 0.001 each) and death with functioning graft (DwFG) after 3rd DDRT (P = 0.001) were higher as compared to 1st DDRT. The incidence of primary nonfunction (PNF) was also significantly higher in re-DDRT (12.7%) than in 1st DDRT (7.1%; P < 0.001). Facing organ shortage, increasing waiting time, and considerable mortality on dialysis, we question the current policy of repeated re-DDRT. The data from this survey propose better HLA matching in first DDRT and second DDRT and careful selection of candidates, especially for ≥4th DDRT.

Interest in, willingness-to-pay for and willingness-to-recommend genetic testing for prostate cancer among affected men after radical prostatectomy.

Knowledge about interest in genetic testing and willingness-to-pay for a genetic test among men affected from prostate cancer (PCa) is limited. This study aimed to gain insight into men's attitudes in genetic testing for PCa. 4699 men with PCa from the German multicenter prospective database "Familial Prostate Cancer" were included. Interest in, Willingness-to-pay for and Willingness-to-recommend a genetic test for PCa were quantified. Associations with several sociodemographic and psychosocial variables were evaluated by logistic regression. 76.8% of the affected men with a median follow-up of 12.9 years were interested in a genetic test for PCa. Newly identified variables significantly associated with interest were having sons (OR 1.66, p < 0.001) and a high perceived severity of the PCa (OR 1.40, p < 0.001). 19% of men were willing to pay more than 500 € for a genetic test. Men with higher education, men with a better self-reported economic situation and men with a lethal PCa in their family were more likely to be willing to pay a larger sum for a test. 84.9% of men were willing to recommend a test to their relatives. Interest in genetic testing for PCa among affected men was generally high with most men willing to recommend a test to their relatives. Various characteristics associated with interest and willingness-to-pay larger sums for genetic testing were uncovered and need to be addressed when designing both future educational material and genetic tests for PCa.

A Contemporary Prostate Biopsy Risk Calculator Based on Multiple Heterogeneous Cohorts.

BACKGROUND: Prostate cancer prediction tools provide quantitative guidance for doctor-patient decision-making regarding biopsy. The widely used online Prostate Cancer Prevention Trial Risk Calculator (PCPTRC) utilized data from the 1990s based on six-core biopsies and outdated grading systems. OBJECTIVE: We prospectively gathered data from men undergoing prostate biopsy in multiple diverse North American and European institutions participating in the Prostate Biopsy Collaborative Group (PBCG) in order to build a state-of-the-art risk prediction tool. DESIGN, SETTING, AND PARTICIPANTS: We obtained data from 15 611 men undergoing 16 369 prostate biopsies during 2006-2017 at eight North American institutions for model-building and three European institutions for validation. OUTCOME MEASUREMENTS AND STATISTICAL ANALYSIS: We used multinomial logistic regression to estimate the risks of high-grade prostate cancer (Gleason score ≥7) on biopsy based on clinical characteristics, including age, prostate-specific antigen, digital rectal exam, African ancestry, first-degree family history, and prior negative biopsy. We compared the PBCG model to the PCPTRC using internal cross-validation and external validation on the European cohorts. RESULTS AND LIMITATIONS: Cross-validation on the North American cohorts (5992 biopsies) yielded the PBCG model area under the receiver operating characteristic curve (AUC) as 75.5% (95% confidence interval: 74.2-76.8), a small improvement over the AUC of 72.3% (70.9-73.7) for the PCPTRC (p<0.0001). However, calibration and clinical net benefit were far superior for the PBCG model. Using a risk threshold of 10%, clinical use of the PBCG model would lead to the equivalent of 25 fewer biopsies per 1000 patients without missing any high-grade cancers. Results were similar on external validation on 10 377 European biopsies. CONCLUSIONS: The PBCG model should be used in place of the PCPTRC for prediction of prostate biopsy outcome. PATIENT SUMMARY: A contemporary risk tool for outcomes on prostate biopsy based on the routine clinical risk factors is now available for informed decision-making.

A prospective evaluation of clinical and genetic predictors of weight changes in breast cancer survivors.

BACKGROUND: Postdiagnosis weight gain in patients with breast cancer has been associated with increased cancer recurrence and mortality. This study was designed to identify risk factors for weight gain and create a predictive model to identify a high-risk population for targeted interventions. METHODS: The weight of 393 patients with breast cancer from the Northwestern Robert H. Lurie Cancer Center was measured over a 2-year period from diagnosis, with body mass index (BMI) change over 18 months as the primary endpoint. Demographics, clinical factors, treatment methods, as well as tumor characteristics were also recorded; and a lifestyle questionnaire was conducted. Blood samples were genotyped for 16 single nucleotide polymorphisms in FTO, adiponectin pathway genes (ADIPOQ, ADIPOR1), and FNDC5. Serum leptin, adiponectin, and irisin levels also were measured. RESULTS: Mean ± standard deviation 18-month BMI changes were 0.68 ± 1.42, 0.98 ± 1.62, 0.79 ± 1.74, and -0.44 ± 1.58 kg/m2 for patients ages <40, 40 to 49, 50 to 59, and ≥60 years, respectively. The optimal multivariable model for 18-month BMI change contained the predictors age, height, and endocrine therapy, but only age was statistically significant, with a 0.04 kg/m2 increase in 18-month BMI change per younger year of age. Single nucleotide polymorphisms in ADIPOR1, FTO, and FNDC5 were associated with 18-month BMI change, and the first 2 remained significant after adjusting for the optimal clinical model (all P < .05). CONCLUSIONS: Women age 60 years and younger at the time of breast cancer diagnosis who have an obesity genetic risk model are at increased risk for weight gain after treatment and should be targeted for weight-maintenance interventions. Cancer 2017;123:2413-21. © 2017 American Cancer Society.