Reporting guidelines for clinical trial reports for interventions involving artificial intelligence: the CONSORT-AI Extension

BMJ, Journal Year: 2020, Volume and Issue: unknown, P. m3164 - m3164

Published: Sept. 9, 2020

Abstract The CONSORT 2010 (Consolidated Standards of Reporting Trials) statement provides minimum guidelines for reporting randomised trials. Its widespread use has been instrumental in ensuring transparency when evaluating new interventions. More recently, there a growing recognition that interventions involving artificial intelligence (AI) need to undergo rigorous, prospective evaluation demonstrate impact on health outcomes. CONSORT-AI extension is guideline clinical trials with an AI component. It was developed parallel its companion trial protocols: SPIRIT-AI. Both were through staged consensus process, literature review and expert consultation generate 29 candidate items, which assessed by international multi-stakeholder group two-stage Delphi survey (103 stakeholders), agreed two-day meeting (31 stakeholders) refined checklist pilot (34 participants). includes 14 considered sufficiently important interventions, they should be routinely reported addition the core items. recommends investigators provide clear descriptions intervention, including instructions skills required use, setting intervention integrated, handling inputs outputs human-AI interaction providing analysis error cases. will help promote completeness assist editors peer-reviewers, as well general readership, understand, interpret critically appraise quality design risk bias

Language: Английский

---

Role of Artificial Intelligence Applications in Real-Life Clinical Practice: Systematic Review

Journal of Medical Internet Research, Journal Year: 2021, Volume and Issue: 23(4), P. e25759 - e25759

Published: March 9, 2021

Artificial intelligence (AI) applications are growing at an unprecedented pace in health care, including disease diagnosis, triage or screening, risk analysis, surgical operations, and so forth. Despite a great deal of research the development validation care AI, only few have been actually implemented frontlines clinical practice.The objective this study was to systematically review AI that real-life practice.We conducted literature search PubMed, Embase, Cochrane Central, CINAHL identify relevant articles published between January 2010 May 2020. We also hand searched premier computer science journals conferences as well registered trials. Studies were included if they reported had real-world settings.We identified 51 studies implementation evaluation practice, which 13 adopted randomized controlled trial design eight experimental design. The targeted various tasks, such screening (n=16), diagnosis analysis (n=14), treatment (n=7). most commonly addressed diseases conditions sepsis (n=6), breast cancer (n=5), diabetic retinopathy (n=4), polyp adenoma (n=4). Regarding outcomes, we found 26 examined performance settings, 33 effect on clinician 14 patient one economic impact associated with implementation.This indicates is still early stage despite potential. More needs assess benefits challenges through more rigorous methodology.

Language: Английский

---

Sources of bias in artificial intelligence that perpetuate healthcare disparities—A global review

PLOS Digital Health, Journal Year: 2022, Volume and Issue: 1(3), P. e0000022 - e0000022

Published: March 31, 2022

Background While artificial intelligence (AI) offers possibilities of advanced clinical prediction and decision-making in healthcare, models trained on relatively homogeneous datasets, populations poorly-representative underlying diversity, limits generalisability risks biased AI-based decisions. Here, we describe the landscape AI medicine to delineate population data-source disparities. Methods We performed a scoping review papers published PubMed 2019 using techniques. assessed differences dataset country source, specialty, author nationality, sex, expertise. A manually tagged subsample articles was used train model, leveraging transfer-learning techniques (building upon an existing BioBERT model) predict eligibility for inclusion (original, human, literature). Of all eligible articles, database source specialty were labelled. BioBERT-based model predicted first/last Author nationality determined corresponding affiliated institution information Entrez Direct. And sex evaluated Gendarize.io API. Results Our search yielded 30,576 which 7,314 (23.9%) further analysis. Most databases came from US (40.8%) China (13.7%). Radiology most represented (40.4%), followed by pathology (9.1%). Authors primarily either (24.0%) or (18.4%). First last authors predominately data experts (i.e., statisticians) (59.6% 53.9% respectively) rather than clinicians. majority male (74.1%). Interpretation U.S. Chinese datasets disproportionately overrepresented AI, almost top 10 nationalities high income countries (HICs). commonly employed image-rich specialties, predominantly male, with non-clinical backgrounds. Development technological infrastructure data-poor regions, diligence external validation re-calibration prior implementation short-term, are crucial ensuring is meaningful broader populations, avoid perpetuating global health inequity.

Language: Английский

---

MINIMAR (MINimum Information for Medical AI Reporting): Developing reporting standards for artificial intelligence in health care

Journal of the American Medical Informatics Association, Journal Year: 2020, Volume and Issue: 27(12), P. 2011 - 2015

Published: April 29, 2020

The rise of digital data and computing power have contributed to significant advancements in artificial intelligence (AI), leading the use classification prediction models health care enhance clinical decision-making for diagnosis, treatment prognosis. However, such advances are limited by lack reporting standards used develop those models, model architecture, evaluation validation processes. Here, we present MINIMAR (MINimum Information Medical AI Reporting), a proposal describing minimum information necessary understand intended predictions, target populations, hidden biases, ability generalize these emerging technologies. We call standard accurately responsibly report on care. This will facilitate design implementation promote development associated decision support tools, as well manage concerns regarding accuracy bias.

Language: Английский

---

Designing deep learning studies in cancer diagnostics

Nature reviews. Cancer, Journal Year: 2021, Volume and Issue: 21(3), P. 199 - 211

Published: Jan. 29, 2021

Language: Английский

---

Reporting guideline for the early-stage clinical evaluation of decision support systems driven by artificial intelligence: DECIDE-AI

Nature Medicine, Journal Year: 2022, Volume and Issue: 28(5), P. 924 - 933

Published: May 1, 2022

Language: Английский

---

Predicting COVID-19 Pneumonia Severity on Chest X-ray With Deep Learning

Cureus, Journal Year: 2020, Volume and Issue: unknown

Published: July 28, 2020

Introduction The need to streamline patient management for coronavirus disease-19 (COVID-19) has become more pressing than ever. Chest X-rays (CXRs) provide a non-invasive (potentially bedside) tool monitor the progression of disease. In this study, we present severity score prediction model COVID-19 pneumonia frontal chest X-ray images. Such can gauge lung infections (and in general) that be used escalation or de-escalation care as well monitoring treatment efficacy, especially ICU. Methods Images from public database were scored retrospectively by three blinded experts terms extent involvement degree opacity. A neural network was pre-trained on large (non-COVID-19) datasets is construct features images which are predictive our task. Results This study finds training regression subset outputs predicts geographic (range 0-8) with 1.14 mean absolute error (MAE) and opacity 0-6) 0.78 MAE. Conclusions These results indicate model's ability could To enable follow up work, make code, labels, data available online.

Language: Английский

---

Guidelines for clinical trial protocols for interventions involving artificial intelligence: the SPIRIT-AI extension

The Lancet Digital Health, Journal Year: 2020, Volume and Issue: 2(10), P. e549 - e560

Published: Sept. 9, 2020

The SPIRIT 2013 statement aims to improve the completeness of clinical trial protocol reporting by providing evidence-based recommendations for minimum set items be addressed. This guidance has been instrumental in promoting transparent evaluation new interventions. More recently, there a growing recognition that interventions involving artificial intelligence (AI) need undergo rigorous, prospective demonstrate their impact on health outcomes. SPIRIT-AI (Standard Protocol Items: Recommendations Interventional Trials-Artificial Intelligence) extension is guideline protocols evaluating with an AI component. It was developed parallel its companion reports: CONSORT-AI (Consolidated Standards Reporting Intelligence). Both guidelines were through staged consensus process literature review and expert consultation generate 26 candidate items, which consulted upon international multi-stakeholder group two-stage Delphi survey (103 stakeholders), agreed meeting (31 stakeholders) refined checklist pilot (34 participants). includes 15 considered sufficiently important These should routinely reported addition core items. recommends investigators provide clear descriptions intervention, including instructions skills required use, setting intervention will integrated, considerations handling input output data, human-AI interaction analysis error cases. help promote transparency Its use assist editors peer reviewers, as well general readership, understand, interpret, critically appraise design risk bias planned trial.

Language: Английский

---

Risk of bias in studies on prediction models developed using supervised machine learning techniques: systematic review

BMJ, Journal Year: 2021, Volume and Issue: unknown, P. n2281 - n2281

Published: Oct. 20, 2021

Abstract Objective To assess the methodological quality of studies on prediction models developed using machine learning techniques across all medical specialties. Design Systematic review. Data sources PubMed from 1 January 2018 to 31 December 2019. Eligibility criteria Articles reporting development, with or without external validation, a multivariable model (diagnostic prognostic) supervised for individualised predictions. No restrictions applied study design, data source, predicted patient related health outcomes. Review methods Methodological was determined and risk bias evaluated assessment tool (PROBAST). This contains 21 signalling questions tailored identify potential biases in four domains. Risk measured each domain (participants, predictors, outcome, analysis) (overall). Results 152 were included: 58 (38%) included diagnostic 94 (62%) prognostic model. PROBAST 19 validations. Of these 171 analyses, 148 (87%, 95% confidence interval 81% 91%) rated at high bias. The analysis most frequently models, 85 (56%, 48% 64%) an inadequate number events per candidate predictor, 62 handled missing inadequately (41%, 33% 49%), 59 assessed overfitting improperly (39%, 31% 47%). Most used appropriate develop (73%, 66% 79%) externally validate based (74%, 51% 88%). Information about blinding outcome predictors was, however, absent 60 (40%, 32% 47%) 79 (52%, 44% 60%) respectively. Conclusion show poor are Factors contributing include small size, handling data, failure deal overfitting. Efforts improve conduct, reporting, validation such necessary boost application clinical practice. review registration PROSPERO CRD42019161764.

Language: Английский

---

Detecting caries lesions of different radiographic extension on bitewings using deep learning

Journal of Dentistry, Journal Year: 2020, Volume and Issue: 100, P. 103425 - 103425

Published: July 4, 2020

Language: Английский

---