约翰斯·霍普金斯大学

In this issue of CHEST (see page 257), another study from Canada by Skomro et al1Skomro RP Gjevre J Reid J et al.Outcomes of home-based diagnosis and treatment of obstructive sleep apnea.Chest. 2010; 138: 257-263Abstract Full Text Full Text PDF PubMed Scopus (92) Google Scholar shows that home-based limited-channel diagnostic testing for obstructive sleep apnea has equivalent short-term outcomes compared with in-laboratory polysomnography. This study is consistent with three other studies that were conducted similarly.2Whitelaw WA Brant RF Flemons WW Clinical usefulness of home oximetry compared with polysomnography for assessment of sleep apnea.Am J Respir Crit Care Med. 2005; 171: 188-193Crossref PubMed Scopus (126) Google Scholar, 3Mulgrew AT Fox N Ayas NT Ryan CF Diagnosis and initial management of obstructive sleep apnea without polysomnography: a randomized validation study.Ann Intern Med. 2007; 146: 157-166Crossref PubMed Scopus (297) Google Scholar, 4Berry RB Hill G Thompson L McLaurin V Portable monitoring and autotitration versus polysomnography for the diagnosis and treatment of sleep apnea.Sleep. 2008; 31: 1423-1431PubMed Google Scholar There are similar themes with these studies: All were performed at a single tertiary-care center, the patients were evaluated by physicians trained in sleep medicine, patients were at significant risk for obstructive sleep apnea, and those with comorbidities were excluded. This particular study is unique in that all patients underwent both home testing and laboratory testing but were treated according to the primary randomization (home testing with autotitrating positive airway pressure [autoPAP] or laboratory testing with continuous positive airway pressure [CPAP] titration); in the other studies, patients only got one test instead of both. All of these outcome studies demonstrated that, for this specific population of patients, it did not matter how their apnea was diagnosed; they used their CPAP device about the same amount of time and received about the same improvement from its use. With yet another validation of the equivalency of unattended limited-channel testing (ULCT), this must mean that this form of diagnostic testing should be available for widespread use in the United States. Well, I am still somewhat skeptical. I do believe that there is a role for ULCT in the diagnosis of obstructive sleep apnea; these studies clearly show the feasibility and potential success. However, these studies were all done by physicians who are familiar with the care of patients with obstructive sleep apnea and who are presumably facile in interpretation of both polysomnography and ULCT. These studies were done on patients who were adequately screened for the signs and symptoms of sleep apnea and who did not have confounding sleep or medical disorders that may reduce the sensitivity and specificity of the ULCT. Within these parameters, ULCT is a viable option. However, in the US health-care system, the use of ULCT cannot be as controlled as it was in these studies. Since Centers for Medicare and Medicaid Services approved ULCT as a viable diagnostic option for the use of CPAP, several companies have sprung up to market this technology. The business model I have seen most often is to offer it to primary care providers (PCPs) as a simple and easy way to diagnose sleep apnea “without the hassle” of using a sleep center. The patient undergoes the testing, the study is “interpreted” by a physician who presumably has some expertise in such interpretations but usually does not ever see the patient, and the report is sent to the ordering physician. He can then order a CPAP device, send the patient to a surgeon or dentist, or recommend some other form of therapy. The decision about which patient to perform this testing on is made by the PCP, and little is known currently about how they are making that determination. Indeed, it may be that this is a very reasonable approach and that most patients will be adequately diagnosed and treated. One could argue that with CPAP adherence being 50% in the best hands, why should sleep specialists think that PCPs will do any worse? Past investigations have shown that sleep and sleep disorders are rarely covered for more than 1 to 2 h in a medical school curriculum. And although little data are available, there is not thought to be much substantial coverage of the subject in most residency or fellowship programs either.5Gamaldo CE Salas RE Sleep medicine education: are medical schools and residency programs napping on the job?.Nat Clin Pract Neurol. 2008; 4: 344-345Crossref PubMed Scopus (9) Google Scholar Therefore, most physicians have little training in the diagnosis and treatment of sleep disorders. Therefore, I still worry about the patient. Do most physicians know enough about sleep apnea to do this correctly? For instance, if the ULCT is negative, is that adequate verification that the patient does not have sleep apnea, even if he has symptoms? Will the ordering physician know what to do next? Quite often patients come to my sleep clinic after having undergone a polysomnography study at an outside sleep center that was “negative for sleep apnea.” They are told they do not have sleep apnea, and that is the end of the evaluation. There are many other things that can cause sleep complaints, such as narcolepsy, depression, and restless legs syndrome. A negative test for sleep apnea may not rule out another sleep disorder. What about the opposite end of the spectrum? If the patient has severe apnea and CPAP is initiated, do we know enough about the initial use of CPAP in an unobserved setting to feel comfortable that it is set correctly without direct observation? Although autoPAP is often touted as “equivalent” to a CPAP titration study, most physicians without sleep medicine training are unaware of its pitfalls. Likewise, the machines themselves are more expensive than a CPAP machine, and there is no “extra” reimbursement for this device or for the added expense of getting it to and from the patient if one decides to use it for a few weeks and then switch to a fixed CPAP level. There is no reimbursement to the durable medical equipment companies for their added time and expense. And what of patients who have comorbid disorders of heart failure, COPD, or hypoventilation; will the ordering physician know that the results may be less accurate in these settings? Moreover, some patients will have complex disordered breathing syndromes with Cheyne-Stokes breathing, narcotic-induced central apneas, or sleep-induced hypoxemia. Will these be recognized and referred or treated appropriately? In conclusion, these studies show that ULCT can work in selected patients cared for by sleep specialists. We do not know how well it will work when there are no controls on who can use it. It is really not about the technology; it is about the initial and then chronic care of the patient through our health-care system. Thoughtful and pragmatic initiation of ULCT through accredited sleep centers with trained physicians has the potential to expand the number of patients being diagnosed and treated for obstructive sleep apnea. Unfortunately, in the United States market forces often control how these new technologies are used; only time will tell how this will play out for our patients in the long run.

While many problems in machine learning focus on learning mappings between finite-dimensional spaces, scientific applications require approximating mappings between function spaces, i.e., operators. We study the problem of learning collections of operators and provide both theoretical and empirical advances. We distinguish between two regimes: (i) multiple operator learning, where a single network represents a continuum of operators parameterized by a parametric function, and (ii) learning several distinct single operators, where each operator is learned independently. For the multiple operator case, we introduce two new architectures, MNO and MONet, and establish universal approximation results in three settings: continuous, integrable, or Lipschitz operators. For the latter, we further derive explicit scaling laws that quantify how the network size must grow to achieve a target approximation accuracy. For learning several single operators, we develop a framework for balancing architectural complexity across subnetworks and show how approximation order determines computational efficiency. Empirical experiments on parametric PDE benchmarks confirm the strong expressive power and efficiency of the proposed architectures. Overall, this work establishes a unified theoretical and practical foundation for scalable neural operator learning across multiple operators.

Transformers are increasingly adopted for modeling and forecasting time-series, yet their internal mechanisms remain poorly understood from a dynamical systems perspective. In contrast to classical autoregressive and state-space models, which benefit from well-established theoretical foundations, Transformer architectures are typically treated as black boxes. This gap becomes particularly relevant as attention-based models are considered for general-purpose or zero-shot forecasting across diverse dynamical regimes. In this work, we do not propose a new forecasting model, but instead investigate the representational capabilities and limitations of single-layer Transformers when applied to dynamical data. Building on a dynamical systems perspective, we interpret causal self-attention as a linear, history-dependent recurrence and analyze how it processes temporal information. Through a series of linear and nonlinear case studies, we identify distinct operational regimes. For linear systems, we show that in the single-head attention-only setting, the convexity constraint imposed by softmax attention restricts the class of autoregressive operators that can be represented, leading to oversmoothing when the target dynamics require mixed-sign lag coefficients. For nonlinear systems under partial observability, attention instead acts as an adaptive delay-embedding mechanism, enabling effective state reconstruction when sufficient temporal context and latent dimensionality are available. These results help bridge empirical observations with classical dynamical systems theory, providing insight into when and why Transformers succeed or fail as models of dynamical systems.

Finite element (FE) methods are the benchmark for solid mechanics simulations, yet their computational cost becomes prohibitive in problems exhibiting localized nonlinearities, fine-scale features, or long-time dynamic evolution. In our earlier FE-neural operator (FE-NO) hybrid framework (Wang et al., 2025), physics-informed deep operator networks were coupled with FE solvers through overlapping domain decomposition with Dirichlet-Dirichlet interface exchange, accelerating computationally intensive subdomains employing NOs while preserving FE fidelity elsewhere. Two limitations remained: the overlapping formulation required redundant interface computations that increased inner Schwarz iteration counts, and the convolutional feature extractor restricted the NO subdomain to structured grids, precluding irregular geometries. This work addresses both limitations. A non-overlapping Schwarz alternating method with Neumann-Dirichlet interface exchange replaces the overlapping formulation, transmitting traction boundary conditions from the NO to FE rather than exchanging displacement. This eliminates the overlap layer and reduces the number of inner Schwarz iterations, while maintaining bounded error accumulation across all tested time horizons. To enable simulations on arbitrarily shaped subdomains, a Point-DeepONet is introduced, which directly operates on unstructured FE point clouds without interpolation, extending the framework to non-convex and irregular subdomain geometries. Furthermore, strain and stress operators are derived analytically from the displacement operators through the governing kinematic equations, rather than trained as independent networks, significantly reducing the number of trainable parameter sets, while enforcing mechanical consistency by construction. The framework is validated on three solid mechanics benchmarks: static linear elasticity, quasi-static hyperelasticity, and elastodynamics with both regular and irregular subdomain geometries. These results establish a non-overlapping FE-NO coupling paradigm that is geometry-flexible, parameter-efficient, and convergence-stable, providing a principled pathway for hybrid physics-based and operator-learning solvers in large-scale dynamic solid mechanics.

This study proposes and operationalizes a new typology based on the relative strength of the regional main center (CBD) and subcenters to classify and track changes in urban spatial structure across around 360 metropolitan areas in the U.S. from 2010 to 2020. Using employment and population data, we first applied spatial statistical techniques to identify CBDs and subcenters across metropolitan areas. We then developed composite CBD and subcenter indices and classified metropolitan areas into four categories for 2010 and 2020: monocentric, balanced polycentric, subcenter-led polycentric, and dispersed. The results show that in 2020, subcenter-led polycentric metropolitan areas emerged as the dominant urban spatial structure, accounting for approximately 43% of all cases. The most notable transition between 2010 and 2020 occurred among balanced polycentric MSAs. Of the 48 metropolitan areas in this category, 20 shifted to a monocentric structure, 8 transitioned to a subcenter-led form, and one transitioned to a dispersed form. Overall, the proposed empirical classification framework provides a replicable approach for systematically classifying and tracking changes in metropolitan spatial structure over time. Also, this framework can support future research on urban form, transportation, and regional development by enabling researchers to systematically evaluate how changes in CBDs and subcenters influence metropolitan spatial structure over time.