Purpose: To develop a prognostic model and nomogram using baseline clinical variables to predict death among men with metastatic hormone-refractory prostate cancer (HRPC). Experimental Design: TAX327 was a clinical trial that randomized 1,006 men with metastatic HRPC to receive every three week or weekly docetaxel or mitoxantrone, each with prednisone. We developed a multivariate Cox model and nomogram to predict survival at 1, 2, and 5 years. Results: Ten independent prognostic factors other than treatment group were identified in multivariate analysis: (a) presence of liver metastases [hazard ratio (HR), 1.66; P = 0.019], (b) number of metastatic sites (HR, 1.63 if z2 sites; P = 0.001), (c) clinically significant pain (HR, 1.48; P < 0.0001), (d) Karnofsky performance status (HR, 1.39 if V70; P = 0.016), (e) type of progression (HR, 1.37 for measurable disease progression and 1.29 for bone scan progression; P = 0.005 and 0.01, respectively), (f) pretreatment prostate-specific antigen (PSA) doubling time (HR, 1.19 if <55 days; P = 0.066), (g) PSA (HR, 1.17 per log rise; P < 0.0001), (h) tumor grade (HR, 1.18 for high grade; P = 0.069), (i) alkaline phosphatase (HR, 1.27 per log rise; P < 0.0001), and (j) hemoglobin (HR, 1.11 per unit decline; P = 0.004). Despite the recent demonstration of palliative and survival benefits with docetaxel-based regimens, men with metastatic hormone-refractory prostate cancer (HRPC) have a poor prognosis, with a median survival of 16 to 20 months (1, 2). The biology of prostate cancer is heterogeneous, with expected survival depending on performance status, presence of visceral metastases, baseline prostate-specific antigen (PSA), advanced primary Gleason sum, baseline hemoglobin, lactate dehydrogenase, albumin, and alkaline phosphatase (3, 4). Nomograms have been developed as stratification tools for phase III clinical trials evaluating cytotoxic chemotherapy based on data from multiple studies that assessed prognostic factors (3 -5).Baseline PSA kinetics [PSA doubling time (PSADT) or PSA velocity] have not been shown conclusively to be an independent prognostic factor in HRPC, with most analyses based on retrospective reviews of relatively small numbers of patients (6 -9). These analyses were limited by the number of controllable covariates, the duration of follow-up, potential confounders, and measurement bias. Furthermore, past prognostic analyses are limited by the inclusion of various types of noncytotoxic therapy and did not include patients treated with docetaxel-based therapy (3, 4).In 2004, docetaxel was approved for the treatment of men with metastatic HRPC based on large multicenter randomized clinical trials (1, 2). The TAX327 study randomized 1,006 patients to receive one of two schedules of docetaxel or mitoxantrone, each given with low-dose prednisone, and showed an extension of overall survival, improvement in quality of life, pain control, PSA decline, and objective tumor response (2). A second phase III study using estramustine phosphate in combination w...

In the TAX327 trial, a PSA decline of > or = 30% within 3 months of chemotherapy initiation had the highest degree of surrogacy for overall survival, confirming data from the Southwest Oncology Group 9916 trial. However, given the wide CIs around the estimate of this moderate surrogate effect, overall survival should remain the preferred end point for phase III trials of cytotoxic agents in HRPC.

Contextual pretrained language models, such as BERT (Devlin et al., 2019), have made significant breakthrough in various NLP tasks by training on large scale of unlabeled text resources. Financial sector also accumulates large amount of financial communication text. However, there is no pretrained finance specific language models available. In this work, we address the need by pretraining a financial domain specific BERT models, FinBERT, using a large scale of financial communication corpora.Experiments on three financial sentiment classification tasks confirm the advantage of FinBERT over generic domain BERT model. The code and pretrained models are available at https://github.com/yya518/FinBERT. We hope this will be useful for practitioners and researchers working on financial NLP tasks.