Palo Alto Research Center

This Guide to Statistics and Methods provides an overview of image-based big data research.

Importance It is unclear whether the effects of intensive vs standard blood pressure (BP) targets seen in clinical trials generalize to patients with chronic kidney disease (CKD) encountered in everyday practice due to differences in the distribution of cardiovascular risk factors and coexisting conditions. Objective To evaluate whether the beneficial and adverse effects of intensive vs standard BP control observed in the Systolic Blood Pressure Intervention Trial (SPRINT) are transportable to a target population of adults with CKD in clinical practice. Design, Setting, and Participants This comparative effectiveness study identified 2 populations with CKD who met the eligibility criteria for SPRINT between January 1 and December 31, 2019, in the Veterans Health Administration (VHA) and Kaiser Permanente of Southern California (KPSC). Baseline covariate, treatment, and outcome data from SPRINT were combined with covariate data from these populations to estimate the treatment effects in the target population, applying models that estimated outcomes using distributions in the trial. Analysis was performed between May 2023 and October 2024. Main Outcomes and Measures The main outcomes were major cardiovascular events, all-cause death, cognitive impairment, CKD progression, and adverse events at 4 years. Results A total of 85 938 patients (mean [SD] age, 75.7 [10.0] years; 81 628 [95.0%] male) from the VHA and 13 983 patients (mean [SD] age, 77.4 [9.6] years; 5371 [38.4%] male) from KPSC were included. Compared with 9361 SPRINT participants (mean [SD] age, 67.9 [9.4] years; 6029 [64.4%] male), these patients were older, had less prevalent cardiovascular disease, higher albuminuria, and used more statins. The associations of intensive vs standard BP control with major cardiovascular events, all-cause death, and adverse events were transportable from the trial to the VHA and KPSC populations; however, the trial’s effects on cognitive and CKD outcomes were not transportable in 1 or both clinical populations. Intensive vs standard BP treatment was associated with lower absolute risks for major cardiovascular events at 4 years by 5.1% (95% CI, −9.8% to 3.2%) in the VHA population and 3.0% (95% CI, −6.3% to 0.3%) in the KPSC population and higher risks for adverse events by 1.3% (95% CI, −5.5% to 7.7%) in the VHA population and 3.1% (95% CI, −1.5% to 8.3%) in the KPSC population. Conclusions and Relevance In this comparative effectiveness study, the reduction in fatal and nonfatal cardiovascular end points and the increase in adverse events observed in SPRINT were largely transportable to trial-eligible CKD populations from clinical practice, suggesting benefits of implementing intensive BP targets.

Primary sclerosing cholangitis (PSC) is a risk factor for cholangiocarcinoma. When a child is diagnosed with both PSC and inflammatory bowel disease (IBD), evidence‐based information on counseling families and risk management of developing cholangiocarcinoma is limited. In this case series (PubMed/collaborators), we included patients with PSC‐IBD who developed cholangiocarcinoma and contacted authors to determine an event curve specifying the time between the second diagnosis (IBD or PSC) and a diagnosis of cholangiocarcinoma. Review of n = 175 studies resulted in a cohort of n = 21 patients with pediatric‐onset PSC‐IBD‐cholangiocarcinoma. The median time to development of cholangiocarcinoma was 6.95 years from the second diagnosis. Despite the small number, 38% of cholangiocarcinoma developed within the first 2 years, and 47% of patients developed cholangiocarcinoma in the transition period to adult care (age 14–25). Our findings highlight the importance of screening that extends over the so‐called transition period from pediatric to adult care.

Background: Self-efficacy is a modifiable intervention target in behavioral weight loss interventions. However, its role in the context of digital interventions is less clear. Purpose: To determine change in self-efficacy in a digital weight loss intervention, and whether self-efficacy is associated with engagement in self-monitoring diet or weight loss. Methods: This is a secondary analysis of the GoalTracker study among 100 adults with overweight or obesity enrolled in a 12-week standalone digital weight loss intervention emphasizing daily self-monitoring. At baseline, 1 month, and 3 months, we assessed self-efficacy for controlling eating (via the Weight Efficacy Lifestyle Questionnaire; WELQ) and self-efficacy for tracking diet. Dietary self-monitoring engagement data were collected from the MyFitnessPal app. Weight was collected in person on a calibrated scale. Analyses included participants with complete data (N range: 72-99). Results: Positive change from baseline to 1 month in self-efficacy for controlling eating was associated with higher dietary self-monitoring engagement (r = 0.21, P = .008) but not with 3-month weight change (r = -0.20, P = .052). Meanwhile, positive change from baseline to 1 month in self-efficacy for tracking diet was associated in a beneficial direction with both outcomes (r = 0.57, P < .001; r = -0.35, P < .001, respectively). However, on average, self-efficacy for controlling eating did not change over time while self-efficacy for tracking diet decreased (P < .001). Conclusion: Improvements in self-efficacy-particularly for tracking diet-early on in a digital weight loss intervention served as a mechanism of greater engagement and weight loss, highlighting the need for strengthening intervention strategies that promote early self-efficacy within a digital context.

Measures of quality in resident training in plastic and reconstructive surgery (PRS) programs are scarce and often methodologically inconsistent. Our research provides insights from current PRS trainees globally, mapping their training inputs, expected outputs, and recommendations for program improvements. A global online survey was conducted among PRS residents across 70 countries to gauge their satisfaction with residency training, capturing training inputs such as the number of surgeries attended and seminars they participated in. We also extracted residents’ proposed recommendations for program improvement. We investigated the explanatory role of training inputs, demographics, hospital characteristics, and country income on resident satisfaction and graduate competence. The analysis incorporated data from 518 PRS residents. On average, residents attended 9.8 surgeries and 1.3 seminars per week. Simultaneously, there was a positive correlation between the perceived level of professional competency and training inputs, particularly seminars attended (p − value = 0.001). Male residents tended to report higher satisfaction (p − value = 0.045) with their training (67%) compared with their female counterparts (58%), while those with family responsibilities also demonstrated slightly higher satisfaction levels. Our analysis expands the evidence base regarding a “global hunger” for more comprehensive seminar-based and hands-on surgical training. Resident recommendations on program improvement reveal the need to address gaps, particularly in aesthetic surgery training. The development of healthcare business models that allow for aesthetic procedures in training institutions is crucial in the promotion of aesthetic surgery training during residency. Policymakers, program directors, and stakeholders across the world can leverage these findings to formulate policies addressing the weaknesses of training programs. This journal requires that authors assign a level of evidence to each article. For a full description of these Evidence-Based Medicine ratings, please refer to the Table of Contents or the online Instructions to Authors www.springer.com/00266.

For the young plastic surgeon, the quantity of first-author peer-reviewed publications plays a prominent role in job offers and promotions. Women surgeons carry a disproportionate share of family responsibilities, contributing to their lower representation in positions of leader- ship and influence. Policies protecting reproductive rights and mandating paid family leave (PFL) boost women’s participation and productivity in the workplace. However, these policies vary by U.S., state and territory. Web-scraped publication data from all PubMed-indexed plastic surgery journals from 2010 to 2022 were evaluated by first-author gender and affiliated state reproductive rights policy and PFL. Female first authors were further compared with men by publication output (1 article; ≥ 2; ≥ 5) by gender and by affiliated state policies. Protective reproductive rights policies were associated with greater representation of female first authors (3.3 percentage points; p value = 0.003). Protective reproductive rights policies and PFL were associated with a decreased publication gender gap (0.13 articles, p value < 0.001, and 0.18 articles, p value < 0.001, respectively). Protective reproductive policies and PFL had an even greater correlation with higher publication output among female first authors. Protective reproductive rights and mandatory PFL are not only correlated with women’s representation among early-career researchers but with a reduction in the publication gender gap. Legislation and policies aimed at supporting women’s family responsibilities are associated with higher research productivity among women and likely play a significant role in attracting more women to higher academic ranks and improving gender equity in professional success in plastic surgery. This journal requires that authors assign a level of evidence to each article. For a full description of these Evidence-Based Medicine ratings, please refer to the Table of Contents or the online Instructions to Authors www.springer.com/00266.

Artificial intelligence (AI) has become an omnipresent topic in the media. Lively discussions are being held on how AI could revolutionize the global healthcare landscape. The development of innovative AI models, including in the medical sector, is increasingly dominated by large high-tech companies. As a global technology epicenter, Silicon Valley hosts many of these technological giants which are muscling their way into healthcare provision with their advanced technologies. The annual conference of the American College of Obstetrics and Gynecology (ACOG) was held in San Francisco from 17 – 19 May 2024. ACOG celebrated its AI premier, hosting two sessions on current AI topics in gynecology at their annual conference. This paper provides an overview of the topics discussed and permits an insight into the thinking in Silicon Valley, showing how technology companies grow and fail there and examining how our American colleagues perceive increased integration of AI in gynecological and obstetric care. In addition to the classification of various, currently popular AI terms, the article also presents three areas where artificial intelligence is being used in gynecology and looks at the current developmental status in the context of existing obstacles to implementation and the current digitalization status of the German healthcare system.

Despite offering early promise, Deep Reinforcement Learning (DRL) suffers from several challenges in adaptive bitrate streaming stemming from the uncertainty and noise in network conditions. However, in this paper, we find that although these challenges complicate the training process, in practice, we can substantially mitigate their effects by addressing a key overlooked factor: the skewed input trace distribution in DRL training datasets. We introduce a generalized framework, Plume , to automatically identify and balance the skew using a three-stage process. First, we identify the critical features that determine the behavior of the traces. Second, we classify the traces into clusters. Finally, we prioritize the salient clusters to improve the overall performance of the controller. We implement our ideas with a novel ABR controller, Gelato , and evaluate the performance against state-of-the-art controllers in the real world for more than a year, streaming 59 stream-years of television to over 280,000 users on the live streaming platform Puffer. Gelato trained with Plume outperforms all baseline solutions and becomes the first controller on the platform to deliver statistically significant improvements in both video quality and stalling, decreasing stalls by as much as 75%.

Pretrained Foundation Models (PFMs) are regarded as the foundation for various downstream tasks across different data modalities. A PFM (e.g., BERT, ChatGPT, GPT-4) is trained on large-scale data, providing a solid parameter initialization for a wide range of downstream applications. In contrast to earlier methods that use convolution and recurrent modules for feature extraction, BERT learns bidirectional encoder representations from Transformers, trained on large datasets as contextual language models. Similarly, the Generative Pretrained Transformer (GPT) method employs Transformers as feature extractors and is trained on large datasets using an autoregressive paradigm. Recently, ChatGPT has demonstrated significant success in large language models, utilizing autoregressive language models with zero-shot or few-shot prompting. The remarkable success of PFMs has driven significant breakthroughs in AI, leading to numerous studies proposing various methods, datasets, and evaluation metrics, which increases the demand for an updated survey. This study provides a comprehensive review of recent research advancements, challenges, and opportunities for PFMs in text, image, graph, and other data modalities. It covers the basic components and existing pretraining methods used in natural language processing, computer vision, and graph learning, while also exploring advanced PFMs for different data modalities and unified PFMs that address data quality and quantity. Additionally, the review discusses key aspects such as model efficiency, security, and privacy, and provides insights into future research directions and challenges in PFMs. Overall, this survey aims to shed light on the research of the PFMs on scalability, security, logical reasoning ability, cross-domain learning ability, and user-friendly interactive ability for artificial general intelligence.

In this chapter, we provide definitions and examples for key concepts that are utilized throughout the book, including knowledge graphs and knowledge graph reasoning tasks, from the perspective of both graph learning and symbolic logic. In the end, we provide a brief and broader introduction to symbolic logic to facilitate the readers to better understand the symbolic component of this book.

KGs usually suffer from severe incompleteness. Knowledge graph completion aims to predict missing facts by reasoning with existing facts. In this chapter, we give a comprehensive introduction to various methods for KG completion, including (1) traditional symbolic reasoning methods, (2) recent representation learning-based methods, and (3) neuro-symbolic integration-based methods. Although many attempts have been made to combine neural and symbolic methods through probabilistic programming frameworks, they are limited by scalability issues. To address these limitations, UniKER [1] algorithm is introduced as a state-of-the-art (SOTA) method that effectively and efficiently combines symbolic reasoning and representation learning for KG completion tasks.

Although many knowledge graphs represent two views: (1) an ontology view for meta-level abstraction, and (2) an instance view for instance-level instantiation, Chapters 3-5 leverage merely instance view knowledge graphs for KG reasoning. In this chapter, we focus on the methods which incorporate additional information in the ontology view with the goal of further improving KG reasoning performance by jointly considering two-view KGs. It is important to note that both cross-view connections and intra-view structures in KG ontologies are essential for KG reasoning. We introduce JOIE [1] as the most representative work that effectively utilizes both pieces of information in a joint manner.

The previous chapter focuses on the problem of KG completion, which aims at predicting the related entities based on a given entity and a specific relation. KG completion queries can be considered as one-hop queries, as the answers are just one-hop away from the query entities. In this chapter, we discuss how to address complex queries, which are defined in the form of First-Order Logic (FOL) and involve multiple entities and relations. The ability to perform complex query answering over KGs is essential for enabling advanced applications, such as dialogue systems, search engines, and recommender systems. In this section, we introduce two main approaches for complex query answering in KGs: (1) traditional subgraph matching-based methods, and (2) more recent logical query embedding methods. Although logical query embedding approaches have shown significant power, many existing models fail to satisfy logical laws with their logical operations, resulting in inferior performance. To address this issue, FuzzQE [1] has been proposed. By carefully designing embeddings for entities and sets and utilizing fuzzy logic to define logical operators, FuzzQE ensures logical laws to be satisfied, leading to improved performance with fewer training labels.

Logical rules are widely used to represent domain knowledge and hypothesis, which is fundamental to symbolic reasoning-based methods. Despite the potential benefits of logical rules, they are usually obtained through a labor-intensive process in the early days. In this section, we look into the problem of learning logic rules automatically from KGs, which can be broadly divided into two categories: (1) traditional search-based methods and (2) more recent neuro-symbolic methods. Despite achieving remarkable performance in learning logical rules, neuro-symbolic integration approaches heavily depend on observed data to identify rules. This makes it challenging for these methods to identify rules that lack sufficient instances to support them. To overcome this limitation, the state-of-the-art approach called RLogic [1] has been introduced. RLogic does not solely rely on rule instances but suggests learning logical rules directly at the schema level and pushing deductive reasoning deep into the learning process.

In this book, we have conducted an extensive investigation into the fascinating domain of integrating neural networks with symbolic reasoning for Knowledge Graph (KG) reasoning. Our focus has been directed toward three primary categories of reasoning tasks: knowledge graph completion, complex query answering, and logical rule learning.

Background: In the phase 3 randomized controlled study, ATTRibute-CM, acoramidis, a transthyretin (TTR) stabilizer, demonstrated significant efficacy on the primary endpoint. Participants with transthyretin amyloid cardiomyopathy (ATTR-CM) who completed ATTRibute-CM were invited to enroll in an open-label extension study (OLE). We report efficacy and safety data of acoramidis in participants who completed ATTRibute-CM and enrolled in the ongoing OLE. Methods: Participants who previously received acoramidis through Month 30 (M30) in ATTRibute-CM continued to receive it (continuous acoramidis), and those who received placebo through M30 were switched to acoramidis (placebo to acoramidis). Participants who received concomitant tafamidis in ATTRibute-CM were required to discontinue it to be eligible to enroll in the OLE. Clinical efficacy outcomes analyzed through Month 42 (M42) included time to event for all-cause mortality (ACM) or first cardiovascular-related hospitalization (CVH), ACM alone, first CVH alone, ACM or recurrent CVH, change from baseline in N-terminal pro-B-type natriuretic peptide (NT-proBNP), 6-minute walk distance (6MWD), serum TTR, and the Kansas City Cardiomyopathy Questionnaire Overall Summary score (KCCQ-OS). Safety outcomes were analyzed through M42. Results: Overall, 438 of 632 participants in ATTRibute-CM completed treatment and 389 enrolled in the ongoing OLE (263 continuous acoramidis, 126 placebo to acoramidis). The hazard ratio (HR) (95% CI) for ACM or first CVH was 0.57 (0.46, 0.72) at M42 based on a stratified Cox proportional hazards model ( P -value < 0.0001) favoring continuous acoramidis. Similar analyses were performed on ACM alone and first CVH alone, with HRs (95% CI) of 0.64 (0.47, 0.88) and 0.53 (0.41, 0.69), respectively, at M42. Treatment effects for NT-proBNP and 6MWD also favored continuous acoramidis. Upon initiation of open-label acoramidis in the placebo-to-acoramidis arm there was a prompt increase in serum TTR. Quality of life assessed by KCCQ-OS was well preserved in continuous acoramidis participants compared with the placebo to acoramidis participants. No new clinically important safety issues were identified in this long-term evaluation. Conclusions: Early initiation and continuous use of acoramidis in the ATTRibute-CM study through M42 of the ongoing OLE study was associated with sustained clinical benefits in a contemporary ATTR-CM cohort, with no clinically important safety issues newly identified.