Showing papers by "Eric J. Topol published in 2019"

High-performance medicine: the convergence of human and artificial intelligence

Abstract: The use of artificial intelligence, and the deep-learning subtype in particular, has been enabled by the use of labeled big data, along with markedly enhanced computing power and cloud storage, across all sectors. In medicine, this is beginning to have an impact at three levels: for clinicians, predominantly via rapid, accurate image interpretation; for health systems, by improving workflow and the potential for reducing medical errors; and for patients, by enabling them to process their own data to promote health. The current limitations, including bias, privacy and security, and lack of transparency, along with the future directions of these applications will be discussed in this article. Over time, marked improvements in accuracy, productivity, and workflow will likely be actualized, but whether that will be used to improve the patient-doctor relationship or facilitate its erosion remains to be seen.

A comparison of deep learning performance against health-care professionals in detecting diseases from medical imaging: a systematic review and meta-analysis

Abstract: Summary Background Deep learning offers considerable promise for medical diagnostics. We aimed to evaluate the diagnostic accuracy of deep learning algorithms versus health-care professionals in classifying diseases using medical imaging. Methods In this systematic review and meta-analysis, we searched Ovid-MEDLINE, Embase, Science Citation Index, and Conference Proceedings Citation Index for studies published from Jan 1, 2012, to June 6, 2019. Studies comparing the diagnostic performance of deep learning models and health-care professionals based on medical imaging, for any disease, were included. We excluded studies that used medical waveform data graphics material or investigated the accuracy of image segmentation rather than disease classification. We extracted binary diagnostic accuracy data and constructed contingency tables to derive the outcomes of interest: sensitivity and specificity. Studies undertaking an out-of-sample external validation were included in a meta-analysis, using a unified hierarchical model. This study is registered with PROSPERO, CRD42018091176. Findings Our search identified 31 587 studies, of which 82 (describing 147 patient cohorts) were included. 69 studies provided enough data to construct contingency tables, enabling calculation of test accuracy, with sensitivity ranging from 9·7% to 100·0% (mean 79·1%, SD 0·2) and specificity ranging from 38·9% to 100·0% (mean 88·3%, SD 0·1). An out-of-sample external validation was done in 25 studies, of which 14 made the comparison between deep learning models and health-care professionals in the same sample. Comparison of the performance between health-care professionals in these 14 studies, when restricting the analysis to the contingency table for each study reporting the highest accuracy, found a pooled sensitivity of 87·0% (95% CI 83·0–90·2) for deep learning models and 86·4% (79·9–91·0) for health-care professionals, and a pooled specificity of 92·5% (95% CI 85·1–96·4) for deep learning models and 90·5% (80·6–95·7) for health-care professionals. Interpretation Our review found the diagnostic performance of deep learning models to be equivalent to that of health-care professionals. However, a major finding of the review is that few studies presented externally validated results or compared the performance of deep learning models and health-care professionals using the same sample. Additionally, poor reporting is prevalent in deep learning studies, which limits reliable interpretation of the reported diagnostic accuracy. New reporting standards that address specific challenges of deep learning could improve future studies, enabling greater confidence in the results of future evaluations of this promising technology. Funding None.

A decade of digital medicine innovation.

Abstract: The field of digital medicine has matured over the past decade, but validation will require careful randomized, controlled clinical trials.

Re-analysis of whole-exome sequencing data uncovers novel diagnostic variants and improves molecular diagnostic yields for sudden death and idiopathic diseases.

Abstract: Whole-exome sequencing (WES) has become an efficient diagnostic test for patients with likely monogenic conditions such as rare idiopathic diseases or sudden unexplained death. Yet, many cases remain undiagnosed. Here, we report the added diagnostic yield achieved for 101 WES cases re-analyzed 1 to 7 years after initial analysis. Of the 101 WES cases, 51 were rare idiopathic disease cases and 50 were postmortem “molecular autopsy” cases of early sudden unexplained death. Variants considered for reporting were prioritized and classified into three groups: (1) diagnostic variants, pathogenic and likely pathogenic variants in genes known to cause the phenotype of interest; (2) possibly diagnostic variants, possibly pathogenic variants in genes known to cause the phenotype of interest or pathogenic variants in genes possibly causing the phenotype of interest; and (3) variants of uncertain diagnostic significance, potentially deleterious variants in genes possibly causing the phenotype of interest. Initial analysis revealed diagnostic variants in 13 rare disease cases (25.4%) and 5 sudden death cases (10%). Re-analysis resulted in the identification of additional diagnostic variants in 3 rare disease cases (5.9%) and 1 sudden unexplained death case (2%), which increased our molecular diagnostic yield to 31.4% and 12%, respectively. The basis of new findings ranged from improvement in variant classification tools, updated genetic databases, and updated clinical phenotypes. Our findings highlight the potential for re-analysis to reveal diagnostic variants in cases that remain undiagnosed after initial WES.

Trends in Prices of Popular Brand-Name Prescription Drugs in the United States.

Abstract: Importance High and continually increasing pharmaceutical drug spending is a major health and health policy concern in the United States. Objective To demonstrate trends in prices among popular brand-name prescription drugs. Design, Setting, and Participants This economic evaluation of drug prices focuses on 49 top-selling brand-name medications in the United States. Pharmacy claims data from January 1, 2012, through December 31, 2017, were obtained from Blue Cross Blue Shield Axis, a database that includes data from more than 35 million individuals with private pharmaceutical insurance. Drugs that exceeded $500 million in US sales or $1 billion in worldwide sales were examined. Main Outcomes and Measures The median sum of out-of-pocket and insurance costs paid by patients or insurers for common prescriptions, presented annually and monthly, was the primary outcome. Results In total, 132 brand-name prescription drugs were identified in 2017 that met the inclusion criteria. Of this total, the study focused on 49 top-selling drugs that exceeded 100 000 pharmacy claims. Substantial cost increases among these drugs was near universal, with a 76% median cost increase from January 2012 through December 2017, and almost all drugs (48 [98%]) displaying regular annual or biannual price increases. Of the 36 drugs that have been available since 2012, 28 (78%) have seen an increase in insurer and out-of-pocket costs by more than 50%, and 16 (44%) have more than doubled in price. Insulins (ie, Novolog, Humalog, and Lantus) and tumor necrosis factor inhibitors (ie, Humira and Enbrel) demonstrated highly correlated price increases, coinciding with some of the largest growth in drug costs. Relative price changes did not differ between drugs that entered the market in the past 3 to 6 years and those that have been on the market longer (number of drugs, 13 vs 36; median, 29% increase from January 2015 through December 2017;P = .81) nor between drugs with or without a Food and Drug Administration–approved therapeutic equivalent (number of drugs, 17 vs 32; median, 79% vs 73%;P = .21). Changes in prices paid were highly correlated with third-party estimates of changes in drug net prices (ρ = 0.55;P = 3.8 × 10−5), suggesting that the current rebate system, which incentivizes high list prices and greater reliance on rebates, increases overall costs. Conclusions and Relevance The growth of drug spending in the United States associated with government-protected market exclusivity is likely to continue; greater price transparency is warranted.

Mechanical activation of noncoding-RNA-mediated regulation of disease-associated phenotypes in human cardiomyocytes

Abstract: How common polymorphisms in noncoding genome regions can regulate cellular function remains largely unknown. Here we show that cardiac fibrosis, mimicked using a hydrogel with controllable stiffness, affects the regulation of the phenotypes of human cardiomyocytes by a portion of the long noncoding RNA ANRIL, the gene of which is located in the disease-associated 9p21 locus. In a physiological environment, cultured cardiomyocytes derived from induced pluripotent stem cells obtained from patients who are homozygous for cardiovascular-risk alleles (R/R cardiomyocytes) or from healthy individuals who are homozygous for nonrisk alleles contracted synchronously, independently of genotype. After hydrogel stiffening to mimic fibrosis, only the R/R cardiomyocytes exhibited asynchronous contractions. These effects were associated with increased expression of the short ANRIL isoform in R/R cardiomyocytes, which induced a c-Jun N-terminal kinase (JNK) phosphorylation-based mechanism that impaired gap junctions (particularly, loss of connexin-43 expression) following stiffening. Deletion of the risk locus or treatment with a JNK antagonist was sufficient to maintain gap junctions and prevent asynchronous contraction of cardiomyocytes. Our findings suggest that mechanical changes in the microenvironment of cardiomyocytes can activate the regulation of their function by noncoding loci.

Digital recruitment and enrollment in a remote nationwide trial of screening for undiagnosed atrial fibrillation: Lessons from the randomized, controlled mSToPS trial.

Abstract: Objectives The advent of large databases, wearable technology, and novel communications methods has the potential to expand the pool of candidate research participants and offer them the flexibility and convenience of participating in remote research. However, reports of their effectiveness are sparse. We assessed the use of various forms of outreach within a nationwide randomized clinical trial being conducted entirely by remote means. Methods Candidate participants at possibly higher risk for atrial fibrillation were identified by means of a large insurance claims database and invited to participate in the study by their insurance provider. Enrolled participants were randomly assigned to one of two groups testing a wearable sensor device for detection of the arrhythmia. Results Over 10 months, the various outreach methods used resulted in enrollment of 2659 participants meeting eligibility criteria. Starting with a baseline enrollment rate of 0.8% in response to an email invitation, the recruitment campaign was iteratively optimized to ultimately include website changes and the use of a five-step outreach process (three short, personalized emails and two direct mailers) that highlighted the appeal of new technology used in the study, resulting in an enrollment rate of 9.4%. Messaging that highlighted access to new technology outperformed both appeals to altruism and appeals that highlighted accessing personal health information. Conclusions Targeted outreach, enrollment, and management of large remote clinical trials is feasible and can be improved with an iterative approach, although more work is needed to learn how to best recruit and retain potential research participants. Trial registration Clinicaltrials.gov NCT02506244 . Registered 23 July 2015.

Real world usage characteristics of a novel mobile health self-monitoring device: Results from the Scanadu Consumer Health Outcomes (SCOUT) Study.

Abstract: A wide range of personal wireless health-related sensor devices are being developed with hope of improving health management. Factors related to effective user engagement, however, are not well-known. We sought to identify factors associated with consistent long-term use of the Scanadu Scout multi-parameter vital sign monitor among individuals who invested in the device through a crowd-funding campaign. Email invitations to join the study were sent to 4525 crowd-funding participants from the US. Those completing a baseline survey were sent a device with follow-up surveys at 3, 12, and 18 months. Of 3872 participants receiving a device, 3473 used it during Week 1, decreasing to 1633 (47 percent) in Week 2. Median time from first use of the device to last use was 17 weeks (IQR: 5–51 weeks) and median uses per week was 1.0 (IQR: 0.6–2.0). Consistent long-term use (defined as remaining in the study at least 26 weeks with at least 3 recordings per week during at least 80% of weeks) was associated with older age, not having children in the household, and frequent use of other medical devices. In the subset of participants answering the 12-month survey (n = 1222), consistent long-term users were more likely to consider the device easy to use and to share results with a healthcare provider. Thirty percent of this subset overall reported improved diet or exercise habits and 25 percent considered medication changes in response to device results. The study shows that even among investors in a device, frequency of device usage fell off rapidly. Understanding how to improve the value of information from personal health-related sensors will be critical to their successful implementation in care.

Deep Learning Under Scrutiny: Performance Against Health Care Professionals in Detecting Diseases from Medical Imaging - Systematic Review and Meta-Analysis

Abstract: Background: Deep learning offers considerable promise for medical diagnostics. In this review, we evaluated the diagnostic accuracy of deep learning (DL) algorithms versus health care professionals (HCPs) in classifying diseases from medical imaging. Methods: We searched (Pre-)Medline, Embase, Science Citation Index, Conference Proceedings Citation Index, and arXiv from 01 January 2012 until 31 May 2018. Studies comparing the diagnostic performance of DL models and HCPs, for any pre-specified condition based on medical imaging material, were included. We extracted binary diagnostic accuracy data and constructed contingency tables at the reported thresholds to derive the outcomes of interest: sensitivity and specificity. Studies undertaking an out-of-sample validation were included in a meta-analysis. Results: 24 studies, from a starting number of 19889, compared DL models with HCPs. 22 studies provided enough data to construct contingency tables, enabling calculation of test accuracy. The mean sensitivity for DL models was 78% (range 13 - 100%), and mean specificity was 86% (range 51 - 100%). An out-of-sample external validation was performed by 5 studies and were therefore included in the meta-analysis. We found a pooled sensitivity of 86% (95% CI: 84 - 88%) for DL models and 93% (95% CI: 87 - 97%) for HCPs, and a pooled specificity of 88% (95% CI: 84 - 92%) for DL models and 87% (95% CI: 84 - 89%) for HCPs. Conclusion: Our review found the diagnostic performance of deep learning models to be similar to health care professionals. A major finding was the poor reporting and potential biases arising from study design that limited reliable interpretation of the reported diagnostic accuracy. New reporting standards which address specific challenges of deep learning could improve future studies, enabling greater confidence in the results of future evaluations of this promising technology. Funding Statement: The authors state: "None" Declaration of Interests: All authors have completed the ICMJE uniform disclosure form online (available on request from the corresponding author) and declare: no support from any organization for the submitted work; no financial relationships with any organizations that might have an interest in the submitted work in the previous three years; no other relationships or activities that could appear to have influenced the submitted work. Ethics Approval Statement: The authors state: "Not required." The authors utilized PRISMA and MOOSE protocols.

Direct to Consumer Fitness DNA Testing.

Abstract: The direct to consumer (DTC)2 genetic testing market was initiated in 2007 with the introduction of 23andMe. Since then, there has been marked expansion and, at the same time, considerable efforts to scrutinize the potential benefits and harms of such testing (1). The various tests are heterogeneous, with output ranging from medical data to ancestry to lifestyle. Along the way, fitness-centered DTC genetic tests have cropped up, encouraging consumers to pay, as we will briefly describe here, high costs for what we feel to be low value information. These tests promise insights about nutrition, healthy aging, weight loss, and athletic aptitude (even for specific sports). They market faux scientific authority, engendering confusion and the potential for harm, no less diminishing consumer confidence in validated, actionable, genetic tests. The genetic testing market is currently divided into 2 main categories with a couple of exceptions that operate in both arenas. Clinical tests require that a physician order each test, are subject to regulatory oversight, and are aimed at diagnosing or screening for mutations related to specific medical conditions. In contrast, DTC tests are purchased without any need for physician involvement. Current surveys identify approximately 40 companies offering DTC fitness genetic tests (2) and several companies that serve as a marketplace for purchasing a variety of such tests interpreted by third parties. Although registries exist to provide some assistance for medical genetics professionals to choose between the approximately 75000 clinical genetic tests available (3), there is no such resource for consumers. DTC genetic test prices range from under $30 to well over $1000. The less expensive tests are generally for individual single nucleotide variants, and marketing by some DTC companies drives sales of multiple tests per customer. The DTC genetic testing market is estimated to total about $100 million in 2017, with …

Deep learning detects impending organ injury in the clinic.

Abstract: Organ damage is often detected late, when treatment options are limited. The use of artificial intelligence to continuously monitor a patient’s medical data can identify people at risk of imminent kidney injury. Artificial intelligence identifies patients at risk of kidney damage.

On the Effectiveness of Deep Representation Learning: The Atrial Fibrillation Case

Abstract: The automatic, unsupervised analysis of biomedical time series is vital for diagnostic and preventive medicine, enabling fast and reliable data processing that reveals clinical insights without human intervention. This article explores and quantifies the benefits of modern deeplearning architectures of varying degrees of complexity.

It takes a planet.

Abstract: To take full advantage of deep-learning solutions in healthcare, the United States and China should collaborate, not compete.

Dispense With Supplements for Improving Heart Outcomes.

Abstract: Khan and colleagues reported an evidence review to identify supplements and dietary interventions associated with improved cardiovascular outcomes The editorialists discuss the findings and limita

What Should Clinicians Do to Engage the Public About Gene Editing

Abstract: Genome editing holds tremendous promise for preventing, ameliorating, or even curing disease, but a thorough discussion of its bioethical and social implications is necessary to protect humankind against harm, a central tenet of the original Hippocratic Oath. It is therefore essential that medical students, physicians, and all health care workers have a working understanding of what gene editing entails, the controversy surrounding its use, and its far-reaching clinical and ethical implications.