Dr. Agent: Clinical predictive model via mimicked second opinions.
Ontology highlight
ABSTRACT: Objective
Prediction of disease phenotypes and their outcomes is a difficult task. In practice, patients routinely seek second opinions from multiple clinical experts for complex disease diagnosis. Our objective is to mimic such a practice of seeking second opinions by training 2 agents with different focuses: the primary agent studies the most recent visit of the patient to learn the current health status, and then the second-opinion agent considers the entire patient history to obtain a more global view.Materials and methods
Our approach Dr. Agent augments recurrent neural networks with 2 policy gradient agents. Moreover, Dr. Agent is customized with various patient demographics information and learns a dynamic skip connection to focus on the relevant information over time. We trained Dr. Agent to perform 4 clinical prediction tasks on the publicly available MIMIC-III (Medical Information Mart for Intensive Care) database: (1) in-hospital mortality prediction, (2) acute care phenotype classification, (3) physiologic decompensation prediction, and (4) forecasting length of stay. We compared the performance of Dr. Agent against 4 baseline clinical predictive models.Results
Dr. Agent outperforms baseline clinical prediction models across all 4 tasks in terms of all metrics. Compared with the best baseline model, Dr. Agent achieves up to 15% higher area under the precision-recall curve on different tasks.Conclusions
Dr. Agent can comprehensively model the long-term dependencies of patients' health status while considering patients' demographics using 2 agents, and therefore achieves better prediction performance on different clinical prediction tasks.
SUBMITTER: Gao J
PROVIDER: S-EPMC7647368 | biostudies-literature |
REPOSITORIES: biostudies-literature
Similar Datasets
Project description:BackgroundSecond medical opinions have become commonplace and even mandatory in some health-care systems, as variations in diagnosis, treatment or prognosis may emerge among physicians.ObjectiveTo evaluate whether physicians' judgment is affected by another medical opinion given to a patient.DesignOrthopedic surgeons and neurologists filled out questionnaires presenting eight hypothetical clinical scenarios with suggested treatments. One group of physicians (in each specialty) was told what the other physician's opinion was (study group), and the other group was not told what it was (control group).ParticipantsA convenience sample of 332 physicians in Israel: 172 orthopedic surgeons (45.9% of their population) and 160 neurologists (64.0% of their population).MeasurementsScoring was by choice of less or more interventional treatment in the scenarios. We used ?(2) tests and repeated measures ANOVA to compare these scores between the two groups. We also fitted a cumulative ordinal regression to account for the dependence within each physician's responses.ResultsOrthopedic surgeons in the study group chose a more interventionist treatment when the other physician suggested an intervention than those in the control group [F (1, 170)?=4.6, p=0.03; OR=1.437, 95% CI 1.115-1.852]. Evaluating this effect separately in each scenario showed that in four out of the eight scenarios, they chose a more interventional treatment when the other physician suggested an intervention (scenario 1, p=0.039; scenario 2, p<0.001; scenario 3, p=0.033; scenario 6, p<0.001). These effects were insignificant among the neurologists [F (1,158)?=0.44, p=0.51; OR=1.087, 95% CI 0.811-1.458]. In both specialties there were no differences in responses by level of clinical experience [orthopedic surgeons: F (2, 166)?=0.752, p=0.473; neurologists: F (2,154)?=1.951, p=0.146].ConclusionsThe exploratory survey showed that in some cases physicians' judgments may be affected by other physicians' opinions, but unaffected in other cases. Weighing previous opinions may yield a more informed clinical decision, yet physicians may be unintentionally influenced by previous opinions. Second opinion has the potential to improve the clinical decision-making processes, and mechanisms are needed to reconcile discrepant opinions.
| S-EPMC3445697 | biostudies-literature
Project description:BACKGROUND:To date, the outcomes of second opinions in internal medicine in terms of diagnostic yield and patient benefit have not been studied extensively. This retrospective study explores the outcomes of second opinions at a general internal medicine outpatient clinic in an academic hospital. METHODS:A register of all patients referred to the general internal medicine outpatient clinic of the University Medical Center in Utrecht for a second opinion, was kept. All 173 patients referred between June 2016 and August 2018 were selected. Case records were analyzed for patient characteristics, referring doctor, chief complaint, performed investigations, follow-up time and, established diagnosis, additional diagnoses, initiated treatment and reported benefit. RESULTS:A new diagnosis was established in 13% of all patients. A new treatment was initiated in 56% of all patients: 91% and 51% of patients with and without a new diagnosis respectively (p < 0.001). Of all patients, 19% received an effective treatment (52% vs 14% of patients with vs without a new diagnosis, p < 0.001). Regardless of treatment, resolution or improvement of the chief complaint was achieved in 28% of all patients (52% vs 25% of patients with vs without a new diagnosis, p = 0.006). Regarding diagnostics, 23-33% of radiology, endoscopy and pathology tests performed during second opinion were a repetition of previously conducted investigations. Conventional blood tests were a repetition in 89% of cases. Median time to diagnosis was 64 days (IQR: 25-128 days) and median time to discharge was 75 days (IQR: 31-144 days). CONCLUSION:Second opinions in general internal medicine lead to the establishment of a new diagnosis in a small proportion of patients. However, the value of second opinions may not be limited to the establishment of diagnoses, as new treatments are often initiated and overall patients report improved symptomatology in 28% of cases.
| S-EPMC7347190 | biostudies-literature
Project description:BackgroundAlthough patient-driven second opinions are increasingly sought in oncology, the desirability of this trend remains unknown. Therefore, this systematic review assesses evidence on the motivation for and frequency of requests for second opinions and examines how they evolve and their consequences for oncological practice.Materials and methodsRelevant databases were sought using the terms "cancer," "second opinion," and "self-initiated." Included were peer-reviewed articles that reported on patient-initiated second opinions within oncology. Selection, data extraction, and quality assessment were performed and discussed by two researchers.ResultsOf the 25 included studies, the methodological designs were qualitative (n?=?4), mixed (n?=?1), or quantitative (n?=?20). Study quality was rated high for 10 studies, moderate for eight, and low for seven studies. Reported rates of second opinion seeking ranged from 1%-88%. Higher education was most consistently related to seeking a second opinion. Patients' primary motivations were a perceived need for certainty or confirmation, a lack of trust, dissatisfaction with communication, and/or a need for more (personalized) information. Reported rates of diagnostic or therapeutic discrepancies between the first and second opinions ranged from 2%-51%.DiscussionAdditional studies are required to further examine the medical, practical, and psychological consequences of second opinions for patients and oncologists. Future studies could compare the potential advantages and disadvantages of second opinion seeking, and might offer guidance to patients and physicians to better facilitate the second opinion process. Some practical recommendations are provided for oncologists to optimally discuss and conduct second opinions with their patients. The Oncologist 2017;22:1197-1211 IMPLICATIONS FOR PRACTICE: Although cancer patients increasingly seek a second opinion, the benefits of this process remain unclear. Results of this systematic review suggest that the available studies on this topic are highly variable in both methodology and quality. Moreover, reported rates for a second opinion (1%-88%) as well as for disagreement between the first and second opinion (2%-51%) range widely. The primary motivations of patients are a need for certainty, lack of trust, dissatisfaction with communication, and/or a need for more (personalized) information. Additional research should evaluate how unnecessary second opinions might be avoided. Practical suggestions are provided for oncologists to optimize second opinions.
| S-EPMC5634767 | biostudies-literature
Project description:OBJECTIVES:The importance of clinical outcome prediction models using artificial intelligence (AI) is being emphasized owing to the increasing necessity of developing a clinical decision support system (CDSS) employing AI. Therefore, in this study, we proposed a "Dr. Answer" AI software based on the clinical outcome prediction model for prostate cancer treated with radical prostatectomy. METHODS:The Dr. Answer AI was developed based on a clinical outcome prediction model, with a user-friendly interface. We used 7,128 clinical data of prostate cancer treated with radical prostatectomy from three hospitals. An outcome prediction model was developed to calculate the probability of occurrence of 1) tumor, node, and metastasis (TNM) staging, 2) extracapsular extension, 3) seminal vesicle invasion, and 4) lymph node metastasis. Random forest and k-nearest neighbors algorithms were used, and the proposed system was compared with previous algorithms. RESULTS:Random forest exhibited good performance for TNM staging (recall value: 76.98%), while k-nearest neighbors exhibited good performance for extracapsular extension, seminal vesicle invasion, and lymph node metastasis (80.24%, 98.67%, and 95.45%, respectively). The Dr. Answer AI software consisted of three primary service structures: 1) patient information, 2) clinical outcome prediction, and outcomes according to the National Comprehensive Cancer Network guideline. CONCLUSION:The proposed clinical outcome prediction model could function as an effective CDSS, supporting the decisions of the physicians, while enabling the patients to understand their treatment outcomes. The Dr. Answer AI software for prostate cancer helps the doctors to explain the treatment outcomes to the patients, allowing the patients to be more confident about their treatment plans.
| S-EPMC7406030 | biostudies-literature
Project description:The long-time goal is to understand what decision-making process and patient characteristics factor into a patient deciding to stay with their initial physician versus seeking treatment with a second opinion.
| 2394118 | ecrin-mdr-crc
Project description:Second opinions in pathology improve patient safety by reducing diagnostic errors, leading to more appropriate clinical treatment decisions. Little objective data are available regarding the factors triggering a request for second opinion despite second opinion consultations being part of the diagnostic system of pathology. Therefore we sought to assess breast biopsy cases and interpreting pathologists characteristics associated with second opinion requests.Collected pathologist surveys and their interpretations of 60 test set cases were used to explore the relationships between case characteristics, pathologist characteristics and case perceptions, and requests for second opinions. Data were evaluated by logistic regression and generalised estimating equations.115 pathologists provided 6900 assessments; pathologists requested second opinions on 70% (4827/6900) of their assessments 36% (1731/4827) of these would not have been required by policy. All associations between case characteristics and requesting second opinions were statistically significant, including diagnostic category, breast density, biopsy type, and number of diagnoses noted per case. Exclusive of institutional policies, pathologists wanted second opinions most frequently for atypia (66%) and least frequently for invasive cancer (20%). Second opinion rates were higher when the pathologist had lower assessment confidence, in cases with higher perceived difficulty, and cases with borderline diagnoses.Pathologists request second opinions for challenging cases, particularly those with atypia, high breast density, core needle biopsies, or many co-existing diagnoses. Further studies should evaluate whether the case characteristics identified in this study could be used as clinical criteria to prompt system-level strategies for mandating second opinions.
| S-EPMC5849252 | biostudies-literature
Project description:BackgroundSecond medical opinions (SOs) can assist patients in making informed treatment decisions and improve the understanding of their diagnosis. In Germany, there are different approaches to obtain a structured SO procedure: SO programs by health insurers and SOs according to the SO Directive. Through a direct survey of the population, we aimed to assess how structured SOs should be provided to fulfil patients' needs.MethodsA stratified sample of 9990 adults (≥18 years) living in the federal states of Berlin and Brandenburg (Germany) were initially contacted by post in April and sent a reminder in May 2020. The survey results were analyzed descriptively.ResultsAmong 1349 participants (response rate 14%), 56% were female and the median age was 58 years (interquartile range (IQR) 44-69). Participants wanted to be informed directly and personally about the possibility of obtaining an SO (89%; 1201/1349). They preferred to be informed by their physician (93%; 1249/1349). A majority of participants would consider it important to obtain an SO for oncological indications (78%; 1049/1349). Only a subset of the participants would seek an SO via their health insurer or via an online portal (43%; 577/1349 and 16%; 221/1349). A personally delivered SO was the preferred route of SO delivery, as 97% (1305/1349) would (tend to) consider this way of obtaining an SO. Participants were asked to imagine having moderate knee pain for years, resulting in a treatment recommendation for knee joint replacement. They were requested to rate potential qualification criteria for a physician providing the SO. The criteria rated to be most important were experience with the recommended diagnosis/treatment (criterion (very) important for 93%; 1257/1349) and knowledge of the current state of research (criterion (very) important for 86%; 1158/1349). Participants were willing to travel 60 min (median; IQR 60-120) and wait 4 weeks (median; IQR 2-4) for their SO in the hypothetical case of knee pain.ConclusionIn general, SOs were viewed positively. We found that participants have clear preferences regarding SOs. We propose that these preferences should be taken into account in the future design and development of SO programs.
| S-EPMC8760563 | biostudies-literature
Project description:Background:Clinical predictive models (CPMs) estimate the probability of clinical outcomes and hold the potential to improve decision-making and individualize care. The Tufts Predictive Analytics and Comparative Effectiveness (PACE) CPM Registry is a comprehensive database of cardiovascular disease (CVD) CPMs. The Registry was last updated in 2012, and there continues to be substantial growth in the number of available CPMs. Methods:We updated a systematic review of CPMs for CVD to include articles published from January 1990 to March 2015. CVD includes coronary artery disease (CAD), congestive heart failure (CHF), arrhythmias, stroke, venous thromboembolism (VTE), and peripheral vascular disease (PVD). The updated Registry characterizes CPMs based on population under study, model performance, covariates, and predicted outcomes. Results:The Registry includes 747 articles presenting 1083 models, including both prognostic (n = 1060) and diagnostic (n = 23) CPMs representing 183 distinct index condition/outcome pairs. There was a threefold increase in the number of CPMs published between 2005 and 2014, compared to the prior 10-year interval from 1995 to 2004. The majority of CPMs were derived from either North American (n = 455, 42%) or European (n = 344, 32%) populations. The database contains 265 CPMs predicting outcomes for patients with coronary artery disease, 196 CPMs for population samples at risk for incident CVD, and 158 models for patients with stroke. Approximately two thirds (n = 701, 65%) of CPMs report a c-statistic, with a median reported c-statistic of 0.77 (IQR, 0.05). Of the CPMs reporting validations, only 333 (57%) report some measure of model calibration. Reporting of discrimination but not calibration is improving over time (p for trend < 0.0001 and 0.39 respectively). Conclusions:There is substantial redundancy of CPMs for a wide spectrum of CVD conditions. While the number of CPMs continues to increase, model performance is often inadequately reported and calibration is infrequently assessed. More work is needed to understand the potential impact of this literature.
| S-EPMC6460840 | biostudies-other
Project description:BACKGROUND:The second victim effect is defined as emotional distress experienced by providers involved in mistakes. This study characterises events contributing to the second victim effect among a diverse sample of physician mothers, describes the impact on both provider and patient and seeks to determine the association between experiencing a mistake and burnout. METHODS:In this mixed-methods study, an anonymous, cross-sectional survey was posted to an online network of over 65 000 physician mothers on 17 June 2016. Self-reported involvement in a mistake provided opportunity to describe the error and impact on both provider and patient. Free-text responses were qualitatively coded to identify error types. Hypothesising that making a mistake contributes to burnout, self-reported burnout was examined using a single question. We used logistic regression to estimate the association between involvement in a mistake and burnout, adjusting for practice years, setting and specialty. RESULTS:5782 members completed the survey for an estimated response rate of 16.5% based on 34956 active users during the survey period. 2859 respondents reported involvement in a mistake (49%), which was associated with higher reported burnout (p<0.0001). 56% of those reporting a mistake provided descriptions. Qualitative analysis revealed that self-reported treatment errors were more common and diagnostic errors were most often reported to result in greater patient harm. Of those involved in a mistake, 82% reported feelings of guilt; 2.2% reported reducing clinical workload, taking leave or leaving the profession. CONCLUSIONS:Physician mothers involved in errors experience negative outcomes and may be at increased risk for burnout. Additional research should focus on strategies to mitigate burnout associated with the second victim effect, particularly among women physicians and those with family responsibilities.
| S-EPMC7173705 | biostudies-literature
Project description:Comprehensive analysis of the complex nature of the Human Leukocyte Antigen (HLA) class II ligandome is of utmost importance to understand the basis for CD4(+)T cell mediated immunity and tolerance. Here, we implemented important improvements in the analysis of the repertoire of HLA-DR-presented peptides, using hybrid mass spectrometry-based peptide fragmentation techniques on a ligandome sample isolated from matured human monocyte-derived dendritic cells (DC). The reported data set constitutes nearly 14 thousand unique high-confident peptides,i.e.the largest single inventory of human DC derived HLA-DR ligands to date. From a technical viewpoint the most prominent finding is that no single peptide fragmentation technique could elucidate the majority of HLA-DR ligands, because of the wide range of physical chemical properties displayed by the HLA-DR ligandome. Our in-depth profiling allowed us to reveal a strikingly poor correlation between the source proteins identified in the HLA class II ligandome and the DC cellular proteome. Important selective sieving from the sampled proteome to the ligandome was evidenced by specificity in the sequences of the core regions both at their N- and C- termini, hence not only reflecting binding motifs but also dominant protease activity associated to the endolysosomal compartments. Moreover, we demonstrate that the HLA-DR ligandome reflects a surface representation of cell-compartments specific for biological events linked to the maturation of monocytes into antigen presenting cells. Our results present new perspectives into the complex nature of the HLA class II system and will aid future immunological studies in characterizing the full breadth of potential CD4(+)T cell epitopes relevant in health and disease.
| S-EPMC4824864 | biostudies-literature