BackgroundIntellectual Disabilities (ID) and Attention Deficit Hyperactivity Disorder (ADHD) are recognized psychological vulnerabilities in police interviews and court proceedings in England and Wales. The aims of this study were to investigate: (a) the prevalence of ID and/or ADHD among suspects detained at a large London metropolitan police station and their relationship with conduct disorder (CD), (b) the impact of their condition on police staff resources, (c) the effectiveness of current custody assessment tools in identifying psychological vulnerabilities, and (d) the use of ‘Appropriate Adults’ in interviews.MethodA total of 200 individuals in a police custody suite were interviewed and screened for ID, ADHD (current symptoms) and CD.ResultsThe screening rates for these three disorders were 6.7%, 23.5% and 76.3%, respectively. ADHD contributed significantly to increased requests being made of staff after controlling for CD and duration of time in custody. This is a novel finding. Reading and writing difficulties and mental health problems were often identified from the custody risk assessment tools, but they were not used effectively to inform on the need for the use of an Appropriate Adult. The frequency with which Appropriate Adults were provided to support detainees in police interviews (4.2%) remains almost identical to that found in a similar study conducted 20 years previously.ConclusionsThe current findings suggest that in spite of reforms recently made in custodial settings, procedures may not have had the anticipated impact of improving safeguards for vulnerable suspects. Detainees with ID and ADHD require an Appropriate Adult during police interviews and other formal custody procedures, which they commonly do not currently receive. The findings of the current study suggest this may be due, in large part, to the ineffective use of risk-assessment tools and healthcare professionals, which represent missed opportunities to identify such vulnerabilities.

Whatever their philosophical bias, few policymakers would argue the importance of adequate information describing differences among health care providers in quality, cost, and availability of services. In particular, such information is widely viewed as essential to improving the quality of health through consumer choice. Thus, advocates of many points of view have pressed for the development of new mechanisms for collecting and disseminating comparative data about health services.The uses of comparative provider information, however, have rarely been subject to empirical test. No one knows very much about how consumers and their agents respond to such information or how such responses influence provider behavior in turn. This DataWatch reports the results of a "natural experiment" in the impact of public release of provider-specific quality data. The HCFA "Death List"In 1985 and 1986, the Health Care Financing Administration (HCFA) analyzed Medicare data as part of an effort to refine and better target peer review organization (PRO) and administrative oversight of Medicare providers. The analyses involved constructing multivariate models to predict hospital-specific mortality rates for Medicare patients discharged in 1984 and comparing those predicted rates with actual rates. Those comparisons identified 269 hospitals as "outliers," approximately half of which had higher-than-predicted mortality, and half lower. Additional analyses, less widely reported, applied the same methodology to nine specific diagnosis-related groups (DRGs).

Perioperative infusion of lidocaine has been reported to decrease the incidence of postsurgical pain at 3 and 6 months following mastectomy using dichotomous (yes/no) scoring. Although intravenous lidocaine reduced the reported incidence of pain at rest at 6 months, pain with activity, pain qualities, and the physical or emotional impact of the pain were unaffected. Future studies evaluating postsurgical persistent pain should adhere to the IMMPACT recommendations in order to more accurately describe the effect of an intervention on persistent pain.

Recently, there has been much interest in the question of whether deep natural language understanding models exhibit systematicitygeneralizing such that units like words make consistent contributions to the meaning of the sentences in which they appear. There is accumulating evidence that neural models often generalize non-systematically. We examined the notion of systematicity from a linguistic perspective, defining a set of probes and a set of metrics to measure systematic behaviour. We also identified ways in which network architectures can generalize non-systematically, and discuss why such forms of generalization may be unsatisfying. As a case study, we performed a series of experiments in the setting of natural language inference (NLI), demonstrating that some NLU systems achieve high overall performance despite being non-systematic.