Machine learning platforms have emerged as a new promissory technology that some argue will revolutionize work practices across a broad range of professions, including medical care. During the past few years, IBM has been testing its Watson for Oncology platform at several oncology departments around the world. Published reports, news stories, as well as our own empirical research show that in some cases, the levels of concordance over recommended treatment protocols between the platform and human oncologists have been quite low. Other studies supported by IBM claim concordance rates as high as 96%. We use the Watson for Oncology case to examine the practice of using concordance levels between tumor boards and a machine learning decision-support system as a form of evidence. We address a challenge related to the epistemic authority between oncologists on tumor boards and the Watson Oncology platform by arguing that the use of concordance levels as a form of evidence of quality or trustworthiness is problematic. Although the platform provides links to the literature from which it draws its conclusion, it obfuscates the scoring criteria that it uses to value some studies over others. In other words, the platform "black boxes" the values that are coded into its scoring system.

Adverse events (AEs) are common during disease-modifying antirheumatic drug (DMARD) treatment, but their influence on treatment results is unclear. We studied AEs in relation to disease activity in early rheumatoid arthritis (RA). Ninety-nine patients started intensive treatment with three conventional synthetic DMARDs (csDMARDs) and oral prednisolone, and were randomized to a 6-month induction treatment with infliximab or placebo. All AEs during the first 12 months of treatment were recorded. We scored each AE based on severity (scale 1-4) and defined the burden of AEs as the sum of these scores. Patients were divided into tertiles according to the burden of AEs. As outcomes, we assessed 28-joint disease activity score (DAS28) levels and remission rates at 12 and 24 months. Three hundred thirty-one AEs in 99 patients were reported, and 27 (8%) were categorized as severe or serious. Mean burden of AEs per patient was 5.4 +/- 4.3. Seventy-nine AEs (24%) led to temporary (n = 52) or permanent (n = 27) csDMARD discontinuation. Of discontinuations, 1, 21, and 57 were detected in the first, second, and third tertiles, respectively. DAS28 remission rates decreased across tertiles at 12 months (94, 94, and 76%; p for linearity 0.029) and at 24 months (90, 86, and 70%; p for linearity 0.021). Mean DAS28 levels increased across tertiles at 12 months (1.5 +/- 1.0, 1.7 +/- 0.9, and 1.9 +/- 1.2; p for linearity 0.021) and at 24 months (1.4 +/- 0.8, 1.6 +/- 1.0, and 1.9 +/- 1.1; p for linearity 0.007). High burden of AEs is associated with higher disease activity and lower likelihood of remission in early RA.

Background Neurotrophic factors are endogenous proteins promoting the survival of different neural cells. Therefore, they elicited great interest as a possible treatment for neurodegenerative disorders, including Parkinson's Disease (PD). PD is the second most common neurodegenerative disorder, scientifically characterized more than 200 years ago and initially linked with motor abnormalities. Currently, the disease is viewed as a highly heterogeneous, progressive disorder with a long presymptomatic phase, and both motor and non-motor symptoms. Presently only symptomatic treatments for PD are available. Neurohistopathological changes of PD affected brains have been described more than 100 years ago and characterized by the presence of proteinaceous inclusions known as Lewy bodies and degeneration of dopamine neurons. Despite more than a century of investigations, it has remained unclear why dopamine neurons die in PD. Methods This review summarizes literature data from preclinical studies and clinical trials of neurotrophic factor based therapies for PD and discuss it from the perspective of the current understanding of PD biology. Results Newest data point towards dysfunctions of mitochondria, autophagy-lysosomal pathway, unfolded protein response and prion protein-like spreading of misfolded alpha-synuclein that is the major component of Lewy bodies. Yet, the exact chain of events leading to the demise of dopamine neurons is unclear and perhaps different in subpopulations of patients. Conclusions Gaps in our understanding of underlying disease etiology have hindered our attempts to find treatments able to slow down the progression of PD. Graphic abstract

Background: Healthcare costs are rising, and a substantial proportion of medical care is of little value. De-implementation of low-value practices is important for improving overall health outcomes and reducing costs. We aimed to identify and synthesize randomized controlled trials (RCTs) on de-implementation interventions and to provide guidance to improve future research. Methods: MEDLINE and Scopus up to May 24, 2021, for individual and cluster RCTs comparing de-implementation interventions to usual care, another intervention, or placebo. We applied independent duplicate assessment of eligibility, study characteristics, outcomes, intervention categories, implementation theories, and risk of bias. Results: Of the 227 eligible trials, 145 (64%) were cluster randomized trials (median 24 clusters; median follow-up time 305 days), and 82 (36%) were individually randomized trials (median follow-up time 274 days). Of the trials, 118 (52%) were published after 2010, 149 (66%) were conducted in a primary care setting, 163 (72%) aimed to reduce the use of drug treatment, 194 (85%) measured the total volume of care, and 64 (28%) low-value care use as outcomes. Of the trials, 48 (21%) described a theoretical basis for the intervention, and 40 (18%) had the study tailored by context-specific factors. Of the de-implementation interventions, 193 (85%) were targeted at physicians, 115 (51%) tested educational sessions, and 152 (67%) multicomponent interventions. Missing data led to high risk of bias in 137 (60%) trials, followed by baseline imbalances in 99 (44%), and deficiencies in allocation concealment in 56 (25%). Conclusions: De-implementation trials were mainly conducted in primary care and typically aimed to reduce low-value drug treatments. Limitations of current de-implementation research may have led to unreliable effect estimates and decreased clinical applicability of studied de-implementation strategies. We identified potential research gaps, including de-implementation in secondary and tertiary care settings, and interventions targeted at other than physicians. Future trials could be improved by favoring simpler intervention designs, better control of potential confounders, larger number of clusters in cluster trials, considering context-specific factors when planning the intervention (tailoring), and using a theoretical basis in intervention design. Registration: OSF Open Science Framework hk4b2.

Objectives: To provide a perspective on the current practice of randomized clinical trials (RCTs) of diagnostic strategies focusing on patient-important outcomes. Study Design and Setting: We conducted a comprehensive search of MEDLINE and included RCTs published in full-text reports that evaluated alternative diagnostic strategies. Results: Of 56,912 unique citations, we sampled 7,500 and included 103 eligible RCTs, therefore suggesting that MEDLINE includes approximately 781 diagnostic RCTs. The 103 eligible trials reported on: mortality (n = 41; 39.8%); morbidities (n = 63; 61.2%); symptoms/quality of life/functional status (n = 14; 13.6%); and on composite end points (n = 10; 9.7%). Of the studies that reported statistically significant results (n = 12; 11.6%), we judged 7 (58.3%) as at low risk of bias with respect to missing outcome data and 4 (33.3%) as at low risk of bias regarding blinding. Of the 41 RCTs that reported on mortality, only one (2.4%) reported statistically significant results. Of 63 RCTs addressing morbidity outcomes, 11 (17.5%) reported statistically significant results, all of which reported relative effects of greater than 20%. Conclusion: RCTs of diagnostic tests are not uncommon, and sometimes suggest benefits on patient-important outcomes but often suffer from limitations in sample size and conduct. (C) 2017 Elsevier Inc. All rights reserved.