On Sequential Bayesian Inference for Continual Learning.

Sequential Bayesian inference can be used for continual learning to prevent catastrophic forgetting of past tasks and provide an informative prior when learning new tasks. We revisit sequential Bayesian inference and assess whether using the previous task's posterior as a prior for a new task can prevent catastrophic forgetting in Bayesian neural networks. Our first contribution is to perform sequential Bayesian inference using Hamiltonian Monte Carlo. We propagate the posterior as a prior for new tasks by approximating the posterior via fitting a density estimator on Hamiltonian Monte Carlo samples. We find that this approach fails to prevent catastrophic forgetting, demonstrating the difficulty in performing sequential Bayesian inference in neural networks. From there, we study simple analytical examples of sequential Bayesian inference and CL and highlight the issue of model misspecification, which can lead to sub-optimal continual learning performance despite exact inference. Furthermore, we discuss how task data imbalances can cause forgetting. From these limitations, we argue that we need probabilistic models of the continual learning generative process rather than relying on sequential Bayesian inference over Bayesian neural network weights. Our final contribution is to propose a simple baseline called Prototypical Bayesian Continual Learning, which is competitive with the best performing Bayesian continual learning methods on class incremental continual learning computer vision benchmarks.

Generic oncology drugs: are they all safe?

Although the availability of generic oncology drugs allows access to contemporary care and reduces costs, there is international variability in the safety of this class of drugs. In this Series paper, we review clinical, policy, safety, and regulatory considerations for generic oncology drugs focusing on the USA, Canada, the European Union (EU), Japan, China, and India. Safety information about generic formulations is reviewed from one agent in each class, for heavy metal drugs (cisplatin), targeted agents (imatinib), and cytotoxic agents (docetaxel). We also review regulatory reports from Japan and the USA, countries with the largest pharmaceutical expenditures. Empirical studies did not identify safety concerns in the USA, Canada, the EU, and Japan, where regulations and enforcement are strong. Although manufacturing problems for generic pharmaceuticals exist in India, where 40% of all generic pharmaceuticals used in the USA are manufactured, increased inspections and communication by the US Food and Drug Administration are occurring, facilitating oversight and enforcement. No safety outbreaks among generic oncology drugs were reported in developed countries. For developing countries, oversight is less intensive, and concerns around drug safety still exist. Regulatory agencies should collaboratively develop procedures to monitor the production, shipment, storage, and post-marketing safety of generic oncology drugs. Regulatory agencies for each country should also aim towards identical definitions of bioequivalence, the cornerstone of regulatory approval.

Regulatory and clinical considerations for biosimilar oncology drugs.

Biological oncology products are integral to cancer treatment, but their high costs pose challenges to patients, families, providers, and insurers. The introduction of biosimilar agents-molecules that are similar in structure, function, activity, immunogenicity, and safety to the original biological drugs-provide opportunities both to improve health-care access and outcomes, and to reduce costs. Several international regulatory pathways have been developed to expedite entry of biosimilars into global marketplaces. The first wave of oncology biosimilar use was in Europe and India in 2007. Oncology biosimilars are now widely marketed in several countries in Europe, and in Australia, Japan, China, Russia, India, and South Korea. Their use is emerging worldwide, with the notable exception of the USA, where several regulatory and cost barriers to biosimilar approval exist. In this Review, we discuss oncology biosimilars and summarise their regulatory frameworks, clinical experiences, and safety concerns.

Evaluation of gemtuzumab ozogamycin associated sinusoidal obstructive syndrome: Findings from an academic pharmacovigilance program review and a pharmaceutical sponsored registry.

BACKGROUND: In 2000, the Food and Drug Administration (FDA) approved gemtuzumab ozogamycin for monotherapy for older patients with relapsed AML. A 0.9% rate of hepatic sinusoidal obstructive syndrome (SOS) was noted in licensing trials. In 2001, FDA received reports of 14 GO-associated SOS cases from MD Anderson Cancer Center. A State of South Carolina/National Cancer Institute funded pharmacovigilance program and a manufacturer sponsored registry independently evaluated this concern. METHODS: The manufacturer's registry and the academic program focused on risk factors and incidence of GO-associated SOS in routine clinical practice and clinical trial settings, respectively. Comparisons were made of findings and dissemination efforts from the two studies. RESULTS: Retrospective analysis of clinical trials by the academic initiative identified 99 cases of SOS among 221 GO-treated stem cell patients and 649 patients who did not undergo HSCTs. SOS rates were 3% when GO was administered at doses ≤6 mg/m(2) as monotherapy or with non-hepatotoxic agents; 28% when administered with 6-thioguanine, a hepatotoxic agent; 15% when administered as monotherapy at doses at a dose of 9 mg/m(2), and between 15% and 40% if a stem cell transplant (SCT) was performed within 3 months of GO administration. Death from SOS occurred in 33% of the cases. The manufacturer's registry prospectively evaluated 482 GO-treated patients who received a mean dose of 7.8 mg/m(2). Overall, 41% received concomitant chemotherapy, 18% had undergone prior SCT, 9.1% developed SOS, and death from SOS occurred in 60% of the SOS cases. Findings from each initiative were disseminated at national conferences and in peer-reviewed manuscripts beginning in 2003. CONCLUSION: Retrospective review of clinical trials, case series, and FDA reports and prospective registries can provide important information on safety signals initially identified in licensing trials.

Forewarned is forearmed: erythropoietin use in CKD patients with cancer.

Nayak-Rao and McCormick raise important concerns about erythropoietin use in patients with chronic kidney disease and cancer, and appropriately build a case for caution in this setting (J Nephrol. 2013, in press). Their comprehensive and insightful review highlights both the clinical and the basic science considerations that form the basis for this warning.