Nature Medicine最新文献

Long COVID (LC) is a complex multisymptom condition with no known disease-modifying treatments. This wait-list-controlled open-label trial tested whether a remotely delivered structured weight management program could improve respective LC symptoms in people living with overweight. Adults with LC (symptoms >12 weeks) and body mass index >27 kg m⁻² (>25 kg m⁻² for South Asians) were randomized (n = 234, 1:1) to control (n = 116, usual care) or the remotely delivered structured weight management (n = 118, total diet replacement (850 kcal per day) for 12 weeks, followed by food reintroduction and weight loss maintenance support) via minimization and randomization (80:20) to balance dominant LC symptom, sex, age, ethnicity and postcode-based index of multiple deprivation between groups. The control group received the intervention after 6 months. Participants selected the dominant LC symptom they would most like to improve (fatigue, breathlessness, pain, anxiety/depression or other) as the prespecified respective primary outcome. Individual symptoms were assessed using validated questionnaires and a visual analog scale for those without prespecified scales. At 6 months, the primary outcome improved in the intervention group (change −1.16 (s.d. 1.42), n = 97 analyzed) compared with the control group (change −0.83 (s.d. 1.14), n = 117 analyzed) with a treatment effect of −0.34 (95% confidence interval −0.67 to −0.01), with no excess of serious adverse events. International Standard Randomised Controlled Trial Number Registry registration: ISRCTN12595520.

Up to 50–70% of patients with liver cirrhosis develop hepatic encephalopathy (HE), which is closely related to gut microbiota dysbiosis, with an unclear mechanism. Here, by constructing gut–brain modules to assess bacterial neurotoxins from metagenomic datasets, we found that phenylalanine decarboxylase (PDC) genes, mainly from Ruminococcus gnavus, increased approximately tenfold in patients with cirrhosis and higher in patients with HE. Cirrhotic, not healthy, mice colonized with R. gnavus showed brain phenylethylamine (PEA) accumulation, along with memory impairment, symmetrical tremors and cortex-specific neuron loss, typically found in patients with HE. This accumulation of PEA was primarily driven by decreased monoamine oxidase-B activity in both the liver and serum due to cirrhosis. Targeting PDC or PEA reversed the neurological symptoms induced by R. gnavus. Furthermore, fecal microbiota transplantation from patients with HE to germ-free cirrhotic mice replicated these symptoms and further corroborated the efficacy of targeting PDC or PEA. Clinically, high baseline PEA levels were linked to a sevenfold increased risk of HE after intrahepatic portosystemic shunt procedures. Our findings expand the understanding of the gut–liver–brain axis and identify a promising therapeutic and predictive target for HE.

The risk of T cell malignancies after chimeric antigen receptor (CAR) T cell therapy is a concern, although the true incidence remains unclear. Here we analyzed the DESCAR-T registry database, encompassing all pediatric and adult patients with hematologic malignancies who received CAR T cell therapy in France since 1 July 2018. Of the 3,066 patients included (2,536 B cell lymphoma, 162 B cell acute lymphoblastic leukemia (ALL) and 368 multiple myeloma), 1,680 (54.8%) received axicabtagene ciloleucel, 205 (6.7%) brexucabtagene autoleucel, 44 (1.4%) lisocabtagene maraleucel and 769 (25.1%) tisagenlecleucel. All multiple myeloma patients received idecabtagene vicleucel, with none receiving ciltacabtagene autoleucel. After a median follow-up of 12.7 months for B cell lymphoma, 17.7 months for B cell ALL and 6.3 months for multiple myeloma, only one (0.03%) patient developed a T cell malignancy after CAR T infusion. Specifically, the patient was diagnosed with a primary cutaneous CD30⁺ T cell lymphoproliferative disorder (anaplastic lymphoma kinase-negative) 3 years after receiving tisagenlecleucel therapy for diffuse large B cell lymphoma. This was associated with the integration of a CAR clone into the tumor suppressor gene PLAAT4 (phospholipase A and acyltransferase 4). Thus, the development of this secondary T cell malignancy might be linked to the use of CAR T cell therapy. In conclusion, our findings indicate a very low risk of T cell malignancy after CAR T cell therapy.

The delivery of accurate diagnoses is crucial in healthcare and represents the gateway to appropriate and timely treatment. Although recent large language models (LLMs) have demonstrated impressive capabilities in few-shot or zero-shot learning, their effectiveness in clinical diagnosis remains unproven. Here we present MedFound, a generalist medical language model with 176 billion parameters, pre-trained on a large-scale corpus derived from diverse medical text and real-world clinical records. We further fine-tuned MedFound to learn physicians’ inferential diagnosis with a self-bootstrapping strategy-based chain-of-thought approach and introduced a unified preference alignment framework to align it with standard clinical practice. Extensive experiments demonstrate that our medical LLM outperforms other baseline LLMs and specialized models in in-distribution (common diseases), out-of-distribution (external validation) and long-tailed distribution (rare diseases) scenarios across eight specialties. Further ablation studies indicate the effectiveness of key components in our medical LLM training approach. We conducted a comprehensive evaluation of the clinical applicability of LLMs for diagnosis involving artificial intelligence (AI) versus physician comparison, AI-assistance study and human evaluation framework. Our proposed framework incorporates eight clinical evaluation metrics, covering capabilities such as medical record summarization, diagnostic reasoning and risk management. Our findings demonstrate the model’s feasibility in assisting physicians with disease diagnosis as part of the clinical workflow.

Continuous compensation for cerebral dopamine deficiency represents an ideal treatment for Parkinson’s disease. Dopamine does not cross the digestive and blood–brain barriers and is rapidly oxidized. The new concept is the intracerebroventricular administration of anaerobic dopamine (A-dopamine) using an abdominal pump connected to a subcutaneous catheter implanted in the third ventricle, near the striatum. An open-label phase 1 study showed no serious adverse reactions induced by A-dopamine in 12 patients. A randomized, controlled, open-label, crossover phase 2 study of 1 month of A-dopamine versus 1 month of optimized oral antiparkinsonian therapy was conducted in 9 patients. The primary endpoint, a blinded assessment of the percentage over target (that is, time with dyskinesia or bradykinesia), recorded by home actimetry using a wristwatch, was significantly reduced on A-dopamine compared with that on oral treatment alone (P = 0.027), with a median within-patient difference of −10.4 (Hedge g = −0.62 (95% confidence interval: −1.43, −0.08)). Home diaries were also significantly improved. These initial data on the feasibility, safety and effects of this new device-assisted therapy suggest validation by a large randomized double-blind trial. ClinicalTrials.gov registration: NCT04332276.

Suraxavir marboxil (GP681) is an antiviral drug inhibiting the polymerase acidic protein (PA) of RNA polymerase, of influenza. It has shown therapeutic activity against influenza A and B virus infections in preclinical studies. In this multicenter randomized, double-blind, placebo-controlled, phase 3 trial, we aimed to investigate the efficacy and safety of single-dose suraxavir marboxil (40-mg oral dose) in otherwise healthy outpatients aged 5–65 years with uncomplicated influenza unaccompanied by severe issues. From 28 July 2022 to 31 October 2023, 591 outpatients aged 5–65 years with uncomplicated influenza underwent randomization in 46 research centers in China and were randomly assigned in a 2:1 ratio to receive suraxavir marboxil (40 mg) or placebo within 2 days of symptom onset. The primary outcome was time to alleviation of influenza symptoms (TTAS) (from the start of treatment until body temperature returned to 37.2 °C or less and all seven influenza symptoms (cough, sore throat, headache, nasal congestion, feverishness or chills, muscle or joint pain, and fatigue) resolved for at least 21.5 h) within 15 days by treatment. The secondary endpoints included virological indicators, system and respiratory symptoms, PA variant mutation and adverse events. The median TTAS was significantly shorter in the group that received suraxavir marboxil compared to the placebo group (42.0 h versus 63.0 h, P = 0.002). Suraxavir marboxil was associated with more rapid decrease in viral load from baseline than placebo by 1 day after administration, with a mean change of −2.2 ± 1.3 compared to −1.3 ± 1.7 log₁₀ copies per ml (P < 0.001) in the placebo group. Adverse events were reported in 28.4% (112 of 395) of suraxavir marboxil recipients and 23.3% (45 of 193) of placebo recipients, most of which were mild or moderate. The incidences of PA variants with the I38T mutation in the H1N1pdm and H3N2 subtypes were 0.7% (1 of 138) and 0.9% (2 of 213), respectively. Low acquired drug resistance was observed. In this trial, timely single-dose suraxavir marboxil was effective in shortening TTAS and reducing the influenza viral load in patients aged 5–65 years with uncomplicated influenza safely. ClinicalTrials.gov registration: NCT05474755.

Endometrial decidualization resistance (DR) is implicated in various gynecological and obstetric conditions. Here, using a multi-omic strategy, we unraveled the cellular and molecular characteristics of DR in patients who have suffered severe preeclampsia (sPE). Morphological analysis unveiled significant glandular anatomical abnormalities, confirmed histologically and quantified by the digitization of hematoxylin and eosin-stained tissue sections. Single-cell RNA sequencing (scRNA-seq) of endometrial samples from patients with sPE (n = 11) and controls (n = 12) revealed sPE-associated shifts in cell composition, manifesting as a stromal mosaic state characterized by proliferative stromal cells (MMP11 and SFRP4) alongside IGFBP1⁺ decidualized cells, with concurrent epithelial mosaicism and a dearth of epithelial–stromal transition associated with decidualization. Cell–cell communication network mapping underscored aberrant crosstalk among specific cell types, implicating crucial pathways such as endoglin, WNT and SPP1. Spatial transcriptomics in a replication cohort validated DR-associated features. Laser capture microdissection/mass spectrometry in a second replication cohort corroborated several scRNA-seq findings, notably the absence of stromal to epithelial transition at a pathway level, indicating a disrupted response to steroid hormones, particularly estrogens. These insights shed light on potential molecular mechanisms underpinning DR pathogenesis in the context of sPE.

Artificial intelligence (AI) in mammography screening has shown promise in retrospective evaluations, but few prospective studies exist. PRAIM is an observational, multicenter, real-world, noninferiority, implementation study comparing the performance of AI-supported double reading to standard double reading (without AI) among women (50–69 years old) undergoing organized mammography screening at 12 sites in Germany. Radiologists in this study voluntarily chose whether to use the AI system. From July 2021 to February 2023, a total of 463,094 women were screened (260,739 with AI support) by 119 radiologists. Radiologists in the AI-supported screening group achieved a breast cancer detection rate of 6.7 per 1,000, which was 17.6% (95% confidence interval: +5.7%, +30.8%) higher than and statistically superior to the rate (5.7 per 1,000) achieved in the control group. The recall rate in the AI group was 37.4 per 1,000, which was lower than and noninferior to that (38.3 per 1,000) in the control group (percentage difference: −2.5% (−6.5%, +1.7%)). The positive predictive value (PPV) of recall was 17.9% in the AI group compared to 14.9% in the control group. The PPV of biopsy was 64.5% in the AI group versus 59.2% in the control group. Compared to standard double reading, AI-supported double reading was associated with a higher breast cancer detection rate without negatively affecting the recall rate, strongly indicating that AI can improve mammography screening metrics.