Algorithmic decision tools (ADTs) are being introduced into public sector organizations to support more accurate and consistent decision-making. Whether they succeed turns, in large part, on how administrators use these tools. This is one of the first empirical studies to explore how ADTs are being used by Street Level Bureaucrats (SLBs). The author develops an original conceptual framework and uses in-depth interviews to explore whether SLBs are ignoring ADTs (algorithm aversion); deferring to ADTs (automation bias); or using ADTs together with their own judgment (an approach the author calls “artificing”). Interviews reveal that artificing is the most common use-type, followed by aversion, while deference is rare. Five conditions appear to influence how practitioners use ADTs: (a) understanding of the tool (b) perception of human judgment (c) seeing value in the tool (d) being offered opportunities to modify the tool (e) alignment of tool with expectations.

This systematic review and meta-analysis aimed to evaluate thrombocytopenia as a prognostic biomarker in patients with coronavirus disease 2019 (COVID-19). We performed a systematic literature search using PubMed, Embase and EuropePMC. The main outcome was composite poor outcome, a composite of mortality, severity, need for intensive care unit care and invasive mechanical ventilation. There were 8963 patients from 23 studies. Thrombocytopenia occurred in 18% of the patients. Male gender (P = 0.037) significantly reduce the incidence. Thrombocytopenia was associated with composite poor outcome (RR 1.90 (1.43–2.52), P < 0.001; I2: 92.3%). Subgroup analysis showed that thrombocytopenia was associated with mortality (RR 2.34 (1.23–4.45), P < 0.001; I2: 96.8%) and severity (RR 1.61 (1.33–1.96), P < 0.001; I2: 62.4%). Subgroup analysis for cut-off <100 × 109/l showed RR of 1.93 (1.37–2.72), P < 0.001; I2: 83.2%). Thrombocytopenia had a sensitivity of 0.26 (0.18–0.36), specificity of 0.89 (0.84–0.92), positive likelihood ratio of 2.3 (1.6–3.2), negative likelihood ratio of 0.83 (0.75–0.93), diagnostic odds ratio of 3 (2, 4) and area under curve of 0.70 (0.66–0.74) for composite poor outcome. Meta-regression analysis showed that the association between thrombocytopenia and poor outcome did not vary significantly with age, male, lymphocyte, d-dimer, hypertension, diabetes and CKD. Fagan's nomogram showed that the posterior probability of poor outcome was 50% in patients with thrombocytopenia, and 26% in those without thrombocytopenia. The Deek's funnel plot was relatively symmetrical and the quantitative asymmetry test was non-significant (P = 0.14). This study indicates that thrombocytopenia was associated with poor outcome in patients with COVID-19.

PROSPERO ID: CRD42020213974

A set of integers is primitive if it does not contain an element dividing another. Let f(n) denote the number of maximum-size primitive subsets of {1,…,2n}. We prove that the limit α = limn→∞f(n)1/n exists. Furthermore, we present an algorithm approximating α with (1 + ε) multiplicative error in N(ε) steps, showing in particular that α ≈ 1.318. Our algorithm can be adapted to estimate the number of all primitive sets in {1,…,n} as well.

We address another related problem of Cameron and Erdős. They showed that the number of sets containing pairwise coprime integers in {1,…n} is between ${2^{\pi (n)}} \cdot {e^{(1/2 + o(1))\sqrt n }}$ and ${2^{\pi (n)}} \cdot {e^{(2 + o(1))\sqrt n }}$. We show that neither of these bounds is tight: there are in fact ${2^{\pi (n)}} \cdot {e^{(1 + o(1))\sqrt n }}$ such sets.

Cyclospora cayetanensis is a parasite causing cyclosporiasis (an illness in humans). Produce (fruits, vegetables, herbs), water and soil contaminated with C. cayetanensis have been implicated in human infection. The objective was to conduct a scoping review of primary research in English on the detection, epidemiology and control of C. cayetanensis with an emphasis on produce, water and soil. MEDLINE® (Web of ScienceTM), Agricola (ProQuest), CABI Global Health, and Food Science and Technology Abstracts (EBSCOhost) were searched from 1979 to February 2020. Of the 349 relevant primary research studies identified, there were 75 detection-method studies, 40 molecular characterisation studies, 38 studies of Cyclospora in the environment (33 prevalence studies, 10 studies of factors associated with environmental contamination), 246 human infection studies (212 prevalence/incidence studies, 32 outbreak studies, 60 studies of environmental factors associated with non-outbreak human infection) and eight control studies. There appears to be sufficient literature for a systematic review of prevalence and factors associated with human infection with C. cayetanensis. There is a dearth of publicly available detection-method studies in soil (n = 0) and water (n = 2), prevalence studies on soil (n = 1) and studies of the control of Cyclospora (particularly on produce prior to retail (n = 0)).

Tick-borne encephalitis (TBE) is a vector-borne infection associated with a variety of potentially serious complications and sequelae. Vaccination against TBE is strongly recommended for people living in endemic areas. There are two TBE vaccination schemes – standard and rapid – which differ in the onset of protection. With vaccination in a rapid schedule, protection starts as early as 4 weeks after the first dose and is therefore especially recommended for non-immune individuals travelling to endemic areas. Both schemes work reliably in immunocompetent individuals, but only little is known about how TBE vaccination works in people with HIV infection. Our aim was to assess the immunogenicity and safety of the rapid scheme of TBE vaccination in HIV-1 infected individuals. Concentrations of TBE-specific IgG > 126 VIEU/ml were considered protective. The seroprotection rate was 35.7% on day 28 and 39.3% on day 60. There were no differences between responders and non-responders in baseline and nadir CD4 + T lymphocytes. No serious adverse events were observed after vaccination. The immunogenicity of the TBE vaccination was unsatisfactory in our study and early protection was only achieved in a small proportion of vaccinees. Therefore, TBE vaccination with the rapid scheme cannot be recommended for HIV-1 infected individuals.

In the group testing problem the aim is to identify a small set of k ⁓ nθ infected individuals out of a population size n, 0 < θ < 1. We avail ourselves of a test procedure capable of testing groups of individuals, with the test returning a positive result if and only if at least one individual in the group is infected. The aim is to devise a test design with as few tests as possible so that the set of infected individuals can be identified correctly with high probability. We establish an explicit sharp information-theoretic/algorithmic phase transition minf for non-adaptive group testing, where all tests are conducted in parallel. Thus with more than minf tests the infected individuals can be identified in polynomial time with high probability, while learning the set of infected individuals is information-theoretically impossible with fewer tests. In addition, we develop an optimal adaptive scheme where the tests are conducted in two stages.

Let ${\mathbb{P}}(ord\pi = ord\pi ')$ be the probability that two independent, uniformly random permutations of [n] have the same order. Answering a question of Thibault Godin, we prove that ${\mathbb{P}}(ord\pi = ord\pi ') = {n^{ - 2 + o(1)}}$ and that ${\mathbb{P}}(ord\pi = ord\pi ') \ge {1 \over 2}{n^{ - 2}}lg*n$ for infinitely many n. (Here lg*n is the height of the tallest tower of twos that is less than or equal to n.)

Bollobás and Riordan, in their paper ‘Metrics for sparse graphs’, proposed a number of provocative conjectures extending central results of quasirandom graphs and graph limits to sparse graphs. We refute these conjectures by exhibiting a sequence of graphs with convergent normalized subgraph densities (and pseudorandom C4-counts), but with no limit expressible as a kernel.

This paper develops an asymptotic theory for nonlinear cointegrating power function regression. The framework extends earlier work on the deterministic trend case and allows for both endogeneity and heteroskedasticity, which makes the models and inferential methods relevant to many empirical economic and financial applications, including predictive regression. A new test for linear cointegration against nonlinear departures is developed based on a simple linearized pseudo-model that is very convenient for practical implementation and has standard normal limit theory in the strictly exogenous regressor case. Accompanying the asymptotic theory of nonlinear regression, the paper establishes some new results on weak convergence to stochastic integrals that go beyond the usual semimartingale structure and considerably extend existing limit theory, complementing other recent findings on stochastic integral asymptotics. The paper also provides a general framework for extremum estimation limit theory that encompasses stochastically nonstationary time series and should be of wide applicability.

Most of the current complex networks that are of interest to practitioners possess a certain community structure that plays an important role in understanding the properties of these networks. For instance, a closely connected social communities exhibit faster rate of transmission of information in comparison to loosely connected communities. Moreover, many machine learning algorithms and tools that are developed for complex networks try to take advantage of the existence of communities to improve their performance or speed. As a result, there are many competing algorithms for detecting communities in large networks. Unfortunately, these algorithms are often quite sensitive and so they cannot be fine-tuned for a given, but a constantly changing, real-world network at hand. It is therefore important to test these algorithms for various scenarios that can only be done using synthetic graphs that have built-in community structure, power law degree distribution, and other typical properties observed in complex networks. The standard and extensively used method for generating artificial networks is the LFR graph generator. Unfortunately, this model has some scalability limitations and it is challenging to analyze it theoretically. Finally, the mixing parameter μ, the main parameter of the model guiding the strength of the communities, has a non-obvious interpretation and so can lead to unnaturally defined networks. In this paper, we provide an alternative random graph model with community structure and power law distribution for both degrees and community sizes, the Artificial Benchmark for Community Detection (ABCD graph). The model generates graphs with similar properties as the LFR one, and its main parameter ξ can be tuned to mimic its counterpart in the LFR model, the mixing parameter μ. We show that the new model solves the three issues identified above and more. In particular, we test the speed of our algorithm and do a number of experiments comparing basic properties of both ABCD and LFR. The conclusion is that these models produce graphs with comparable properties but ABCD is fast, simple, and can be easily tuned to allow the user to make a smooth transition between the two extremes: pure (independent) communities and random graph with no community structure.

This paper studies the uniform convergence rates of Li and Vuong’s (1998, Journal of Multivariate Analysis 65, 139–165; hereafter LV) nonparametric deconvolution estimator and its regularized version by Comte and Kappus (2015, Journal of Multivariate Analysis 140, 31–46) for the classical measurement error model, where repeated noisy measurements on the error-free variable of interest are available. In contrast to LV, our assumptions allow unbounded supports for the error-free variable and measurement errors. Compared to Bonhomme and Robin (2010, Review of Economic Studies 77, 491–533) specialized to the measurement error model, our assumptions do not require existence of the moment generating functions of the square and product of repeated measurements. Furthermore, by utilizing a maximal inequality for the multivariate normalized empirical characteristic function process, we derive uniform convergence rates that are faster than the ones derived in these papers under such weaker conditions.

Under a Mundlak-type correlated random effect (CRE) specification, we first show that the average likelihood of a parametric nonlinear panel data model is the convolution of the conditional distribution of the model and the distribution of the unobserved heterogeneity. Hence, the distribution of the unobserved heterogeneity can be recovered by means of a Fourier transformation without imposing a distributional assumption on the CRE specification. We subsequently construct a semiparametric family of average likelihood functions of observables by combining the conditional distribution of the model and the recovered distribution of the unobserved heterogeneity, and show that the parameters in the nonlinear panel data model and in the CRE specification are identifiable. Based on the identification result, we propose a sieve maximum likelihood estimator. Compared with the conventional parametric CRE approaches, the advantage of our method is that it is not subject to misspecification on the distribution of the CRE. Furthermore, we show that the average partial effects are identifiable and extend our results to dynamic nonlinear panel data models.

People in prison are disproportionately affected by viral hepatitis. To examine the current epidemiology of and responses targeting hepatitis B virus (HBV) in prisons across the European Union, European Economic Area and United Kingdom, we analysed HBV-specific data from the World Health Organization's Health in Prisons European Database and the European Centre for Disease Prevention and Control's hepatitis B prevalence database. Hepatitis B surface antigen seroprevalence ranged from 0% in a maximum-security prison in United Kingdom to 25.2% in two Bulgarian juvenile detention centres. Universal HBV screening on opt-out basis and vaccination were reported available in 31% and 85% of 25 countries, respectively. Disinfectants, condoms and lubricants were offered free of charge in all prisons in the country by 26%, 46% and 15% of 26 countries, respectively. In 38% of reporting countries, unsupervised partner visits with the possibility for sexual intercourse was available in all prisons. The findings are suggestive of high HBV prevalence amidst suboptimal coverage of interventions in prisons. A harmonised monitoring system and robust data at national and regional levels are needed to better understand the HBV situation in prisons within the framework of the European action plan and Global Health Sector Strategy on viral hepatitis.

Rough sleeping is a chronic experience faced by some of the most disadvantaged people in modern society. This paper describes work carried out in partnership with Homeless Link (HL), a UK-based charity, in developing a data-driven approach to better connect people sleeping rough on the streets with outreach service providers. HL's platform has grown exponentially in recent years, leading to thousands of alerts per day during extreme weather events; this overwhelms the volunteer-based system they currently rely upon for the processing of alerts. In order to solve this problem, we propose a human-centered machine learning system to augment the volunteers' efforts by prioritizing alerts based on the likelihood of making a successful connection with a rough sleeper. This addresses capacity and resource limitations whilst allowing HL to quickly, effectively, and equitably process all of the alerts that they receive. Initial evaluation using historical data shows that our approach increases the rate at which rough sleepers are found following a referral by at least 15% based on labeled data, implying a greater overall increase when the alerts with unknown outcomes are considered, and suggesting the benefit in a trial taking place over a longer period to assess the models in practice. The discussion and modeling process is done with careful considerations of ethics, transparency, and explainability due to the sensitive nature of the data involved and the vulnerability of the people that are affected.

Focusing on seven major agricultural commodities with a long history of trade, this study employs data-driven analytics to decipher patterns of trade, namely using supervised machine learning (ML), as well as neural networks. The supervised ML and neural network techniques are trained on data until 2010 and 2014, respectively. Results show the high relevance of ML models to forecasting trade patterns in near- and long-term relative to traditional approaches, which are often subjective assessments or time-series projections. While supervised ML techniques quantified key economic factors underlying agricultural trade flows, neural network approaches provide better fits over the long term.

State preferences play an important role in international politics. Unfortunately, actually observing and measuring these preferences are impossible. In general, scholars have tried to infer preferences using either UN voting or alliance behavior. The two most notable measures of state preferences that have flowed from this research area are ideal points (Bailey et al., 2017) and S-scores (Signorino & Ritter, 1999). The basis of both these models is a spatial weighting scheme that has proven useful but discounts higher-order effects that might be present in relational data structures such as UN voting and alliances. We begin by arguing that both alliances and UN voting are simply examples of the multiple layers upon which states interact with one another. To estimate a measure of state preferences, we utilize a tensor decomposition model that provides a reduced-rank approximation of the main patterns across the layers. Our new measure of preferences plausibly describes important state relations and yields important insights on the relationship between preferences, democracy, and international conflict. Additionally, we show that a model of conflict using this measure of state preferences decisively outperforms models using extant measures when it comes to predicting conflict in an out-of-sample context.

This study was a retrospective multicentre cohort study of patients with coronavirus disease 2019 (COVID-19) diagnosed at 24 hospitals in Jiangsu province, China as of 15 March 2020. The primary outcome was the occurrence of acute respiratory failure during hospital stay. Of 625 patients, 56 (9%) had respiratory failure. Some selected demographic, epidemiologic, clinical and laboratory features as well as radiologic features at admission and treatment during hospitalisation were significantly different in patients with and without respiratory failure. The multivariate logistic analysis indicated that age (in years) (odds ratio [OR], 1.07; 95% confidence interval [CI]: 1.03–1.10; P = 0.0002), respiratory rate (breaths/minute) (OR, 1.23; 95% CI: 1.08–1.40; P = 0.0020), lymphocyte count (109/l) (OR, 0.18; 95% CI: 0.05–0.69; P = 0.0157) and pulmonary opacity score (per 5%) (OR, 1.38; 95% CI: 1.19–1.61; P < 0.0001) at admission were associated with the occurrence of respiratory failure. Older age, increased respiratory rate, decreased lymphocyte count and greater pulmonary opacity score at admission were independent risk factors of respiratory failure in patients with COVID-19. Patients having these risk factors need to be intensively managed during hospitalisation.