このページは自動翻訳されたものであり、翻訳の正確性は保証されていません。を参照してください。 英語版 ソーステキスト用。

AI-ECG Screening for Early Heart Failure Diagnosis (VISION-HF)

2026年11月15日 更新者:Antonio Luiz Pinho Ribeiro

Validated AI-ECG Screening and Intervention for Optimizing Heart Failure Diagnosis

The goal of this parallel-group cluster randomized clinical trial is to evaluate whether opportunistic screening for left ventricular systolic dysfunction (LVSD), using an artificial intelligence algorithm applied to electrocardiograms (AI-ECG), increases the detection of previously unrecognized LVSD among adults aged ≥40 years receiving routine care in primary healthcare centers within the Brazilian Public Health System. LVSD is a key finding in the diagnosis of heart failure with reduced ejection fraction (HFrEF), a condition for which effective guideline-directed medical therapies are available but which remains underdiagnosed and undertreated.

The study will also evaluate the effects of AI-ECG screening on clinical management, healthcare utilization, implementation outcomes, and healthcare costs.

The main questions this study aims to answer are:

  1. Does adding AI-ECG screening to usual clinical assessment increase the proportion of patients with newly identified LVSD (left ventricular ejection fraction <50%) within six months compared with clinical assessment alone?
  2. Does AI-ECG screening improve the use of echocardiography, the recognition by primary care teams, and the initiation or optimization of guideline-directed medical therapy for HFrEF?
  3. How do LVSD detection rates differ between patients identified through AI-ECG screening alone and those identified through clinical suspicion alone?
  4. Is this screening strategy feasible, acceptable, and cost-effective when implemented in primary healthcare?

Primary healthcare centers will be randomized to either an intervention group, in which AI-ECG screening results will be provided alongside the standard ECG report, or a control group receiving usual care without AI-ECG results.

Participants will:

  1. Undergo a routine 12-lead electrocardiogram (ECG), as clinically indicated by their primary care physician.
  2. Be referred for echocardiography based on clinical suspicion of LVSD in both groups or, additionally, a positive AI-ECG screening result in the intervention group.
  3. Receive educational materials about HFrEF if newly diagnosed, while participating healthcare professionals will receive educational interventions on heart failure diagnosis and management.
  4. Complete questionnaires evaluating educational interventions, when applicable.
  5. Be followed for six months to assess newly identified LVSD, treatment initiation and optimization, healthcare utilization (including emergency visits and hospitalizations), mortality, implementation outcomes, and healthcare costs.

調査の概要

状態

まだ募集していません

条件

介入・治療

詳細な説明

  1. Background: Cardiovascular diseases (CVD) are the main cause of death in Brazil. Among CVDs, heart failure with reduced ejection fraction (HFrEF) is the main cause of hospitalizations and costs for the public health system (Sistema Único de Saúde, SUS), which still tend to increase with population aging and greater survival rates from acute cardiovascular events. Guideline-directed medical therapy (GDMT) substantially reduces the risk of hospitalization and mortality among patients with HFrEF. Although several recommended therapies are available through the Brazilian Public Health System (SUS), their use remains suboptimal. Part of this is due to the lack of diagnosis, which requires an echocardiogram (ECHO) to diagnose left ventricular systolic dysfunction, an essential finding of HFrEF. ECHO is an expensive and complex exam, and therefore, with restricted access in primary health care (PHC), where patients can wait in lines more than 12 months to undergo an exam. Moreover, asymptomatic LVSD is also a condition that is considered preHF, which has specific GDMT to prevent it from evolving to clinical HFrEF.

    The electrocardiogram (ECG) is a screening tool for the evaluation of heart disease and is widely available in primary health care (PHC) centers in Brazil. Although it provides valuable clinical information, conventional ECG interpretation cannot determine left ventricular ejection fraction (LVEF). Sangha et al. demonstrated that a convolutional neural network model can accurately identify left ventricular systolic dysfunction (LVSD) from ECGs. Although this model has been externally validated, including in the community-based ELSA-Brasil cohort, there is currently no evidence that implementing AI-ECG screening for LVSD at scale will change clinical practice.

  2. Objectives: The overarching aim is to evaluate whether opportunistic screening for **LVSD** using an artificial intelligence algorithm applied to tele-electrocardiography (**AI-ECG for left ventricular systolic dysfunction [LVSD]**) increases the rate of LVSD diagnosis in routine clinical practice within Brazil's Unified Health System (SUS) primary care network. Our hypothesis is that informing the AI-ECG result to the primary care team will increase the rate of LVSD diagnosis in six months.

  3. Study Design and Setting: VISION-HF is a multicenter, parallel-group, cluster-randomized clinical trial with a hybrid type 1 effectiveness-implementation design. The study will be conducted in primary healthcare centers (PHCs) in three or four Brazilian municipalities, each from a different Brazilian region. PHCs will be randomized to an intervention or control group.

  4. Cluster selection: Primary healthcare centers (PHCs) will serve as the units of randomization. Participating PHCs will be selected by municipal health authorities and subsequently randomized to the intervention or control group, stratifed by municipality. In each municipality, participating centers will represent more than 30% of all PHCs.

  5. Study Population: Individuals aged 40 years or older, with no previous diagnosis of left ventricular systolic dysfunction (LVSD), defined as left ventricular ejection fraction <50% who have been referred for an electrocardiogram (ECG) by their attending physician as part of routine clinical practice, will be invited to participate in the study. Patients who provide informed consent will be enrolled. Patients will be excluded if the ECG is uninterpretable.

  6. Study Flow: After informed consent is obtained, participants' baseline information will be collected according to the routine ECG acquisition protocol. Clinical suspicion of HF will be defined by the attending physician, when requesting the ECG. Data will be entered into the software by the nurse technician performing the ECG as part of routine clinical practice, using a dedicated system developed by the HC-UFMG Telehealth Center (CTS-HC-UFMG), to which the data will be transmitted and stored. In the intervention group, an artificial intelligence algorithm will analyze routinely acquired ECGs to identify patients at increased risk of LVSD. AI-ECG screening results will be provided alongside the standard ECG report. Echocardiography will be offered to patients with a positive AI-ECG result or clinical suspicion of LVSD. In the control group, physicians will receive the standard ECG report without AI-ECG screening results. Echocardiography will be offered based on clinical suspicion alone. The study will provide access to echocardiography for participants meeting referral criteria in both groups, following the same procedures.

  7. Sample Size Calculation: The sample size calculation assumes an intracluster correlation coefficient (ICC) of 0.01 and an average of 72 analyzable participants per primary healthcare center (PHC). The corresponding design effect is 1.71 [DE = 1 + (72 - 1) × 0.01]. The calculation is based on expected primary outcome rates of 1.8% in the control arm and 4.0% in the intervention arm, with a two-sided significance level of 5% and 90% statistical power. A total of 60 PHCs (30 per arm) are planned. To account for an anticipated 20% loss to follow-up, 90 participants will be recruited per PHC (5,400 overall), yielding an expected 72 analyzable participants per PHC (4,320 overall). The expected detection rates were based on data from the pilot study (10 PHCs in three municipalities, n=342), in which rates of 1.8% and 5.2% were observed in the control and intervention groups, respectively. A conservative estimate of 4.0% was used for the intervention group. The ICC assumption was informed by previous publications, including the EAGLE trial (Yao et al., Nature Medicine, 2021), and pilot study data. The power assessment for the key secondary outcome (diagnostic yield, defined as the number of newly identified LVSD cases divided by the number of echocardiograms performed in each arm) was based on pilot study data. The estimated statistical power was 100% across all scenarios.

  8. Study Procedures and Follow-up: Participants will be followed for 6±1months. The primary outcome will be the proportion of participants with newly identified LVSD, defined as LVEF <50% on echocardiography. Echocardiographic images will be stored in a standardized format in a centralized database. All echocardiograms indicating LVSD (LVEF <50%) will undergo independent review by an expert at the central echocardiography reading center. A sample of echocardiograms without LVSD (LVEF ≥50%) will be randomly selected for central review to assess diagnostic consistency and quality. Local examiners and central reviewers will be blinded to treatment allocation and AI-ECG screening results. The echocardiography database will also be used to explore additional echocardiographic parameters of potential clinical and research interest. Additional assessments will include diagnostic yield, echocardiography utilization, documentation of newly identified LVSD in medical records, initiation or optimization of guideline-directed medical therapy, healthcare utilization, and mortality.

  9. Statistical analysis: The participant will be the unit of analysis. Since randomization is performed at the public primary health care center (PHC) level, within-center correlation will be accounted for using generalized estimating equations (GEE), specifying the PHC as the clustering unit and an exchangeable working correlation structure. The primary analysis will follow the intention-to-treat principle, according to the randomized allocation of public primary health care centers (PHCs). The primary outcome, defined as a new diagnosis of left ventricular systolic dysfunction (LVSD; left ventricular ejection fraction [LVEF] <50%). Diagnostic yield will be considered a key secondary outcome and a power assessment has already been described under "sample size assumptions" section. All statistical analyses will be performed using R statistical software (R Foundation for Statistical Computing, Vienna, Austria), with a two-sided significance level of 5% (α = 0.05).

  10. Implementation and Economic Evaluation

Implementation outcomes will be assessed using the RE-AIM framework, including reach, effectiveness, adoption, implementation, and maintenance. Capacity-building activities will provide educational resources for healthcare professionals and patients diagnosed with LVSD.

A qualitative component will include PHC physicians and participants to explore barriers and facilitators to AI-ECG implementation. Patients with positive AI-ECG screening results will also be invited to participate, including patients with and without echocardiographically confirmed LVSD. These discussions will explore patients' experiences, feelings, and perceptions regarding AI-ECG screening and its results. Transcripts will be coded, organized into thematic categories, and analyzed using qualitative content analysis. An economic evaluation will assess the cost-effectiveness of AI-ECG screening compared with usual care, incorporating study-specific diagnostic and resource utilization data, Brazilian healthcare costs, and, where feasible, longer-term economic modeling.

研究の種類

介入

入学 (推定)

5400

段階

  • 適用できない

連絡先と場所

このセクションには、調査を実施する担当者の連絡先の詳細と、この調査が実施されている場所に関する情報が記載されています。

研究連絡先

研究連絡先のバックアップ

  • 名前:Antonio Luiz P Ribeiro, MD, PhD
  • 電話番号:55 31 987090451
  • メール:alpr1963br@gmail.com

研究場所

    • Minas Gerais
      • Belo Horizonte、Minas Gerais、ブラジル、30130100
        • Telehealth Center at Hospital das Clínicas
        • コンタクト:
        • コンタクト:
        • 主任研究者:
          • Luisa CC Brant, MD, PhD

参加基準

研究者は、適格基準と呼ばれる特定の説明に適合する人を探します。これらの基準のいくつかの例は、人の一般的な健康状態または以前の治療です。

適格基準

就学可能な年齢

40年歳以上 (大人、高齢者)

健康ボランティアの受け入れ

はい

受講資格のある性別

全て

説明

Participating primary care centers were selected by the municipal health authorities. For participants, the inclusion criteria are: (1) having an ECG ordered by a primary care physician during routine care; (2) age ≥40 years; (3) no known left ventricular systolic dysfunction (left ventricular ejection fraction <50%); and (4) provision of written informed consent. Participants with uninterpretable ECGs will be excluded.

研究計画

このセクションでは、研究がどのように設計され、研究が何を測定しているかなど、研究計画の詳細を提供します。

研究はどのように設計されていますか?

デザインの詳細

  • 主な目的:診断
  • 割り当て:ランダム化
  • 介入モデル:並列代入
  • マスキング:独身

武器と介入

参加者グループ / アーム
介入・治療
実験的:Intervention

AI-ECG screening results for LVSD will be provided alongside the standard ECG report. AI-ECG results will be generated in real time through an application programming interface (API) connected to Yale University's Cardiovascular Data Science (CarDS) Lab (doi: 10.1161/CIRCULATIONAHA.122.062646). Physicians will be able to request an echocardiogram for patients with a positive AI-ECG screening result, as well as for patients with clinical suspicion of LVSD, defined prior to ECG acquisition. Given the limited availability of echocardiography services and prolonged waiting times in the participating municipalities, the study will provide echocardiograms for all eligible patients patients meeting the referral criteria in either study arm, following same referral procedures.

AI-ECG results will be generated in real time through an application programming interface (API) connected to Yale University's Cardiovascular Data Science (CarDS) Lab. The AI-ECG algorithm was developed by the Yale CarDS Lab and previously validated for detecting LVSD (Sangha et al. Circulation. 2023;148:765-777. doi: 10.1161/CIRCULATIONAHA.122.062646). A positive AI-ECG screening result will be generated when the predicted probability of LVSD exceeds a predefined threshold of 0.10. This threshold was selected to achieve a sensitivity greater than 90% in the original internal validation study to detect left ventricular ejection fraction (LVEF)<40%. In the present study, it will be used to detect a LVEF<50%, as current guidelines uses this threshold for LVSD, which leads to clinical actions.

介入なし:Control

The standard ECG report will be provided without AI-ECG screening results. The AI-ECG algorithm will run in the background, but its results will not be disclosed to treating physicians or used to guide clinical decisions during the six-month follow-up period. Physicians will be able to request an echocardiogram for patients with clinical suspicion of LVSD, , defined prior to ECG acquisition. Given the limited availability of echocardiography services and prolonged waiting times in the participating municipalities, the study will provide echocardiograms for all eligible patients meeting the referral criteria, following the same referral procedures as in the intervention group. For ethical reasons, participants in the control group with a positive AI-ECG screening result who have not undergone echocardiography during the study will be offered an echocardiogram after completion of the six-month follow-up period.

この研究は何を測定していますか?

主要な結果の測定

結果測定
メジャーの説明
時間枠
Proportion of new diagnosis of left ventricular systolic dysfunction
時間枠:6 months

The primary outcome will be the proportion of participants with newly identified left ventricular systolic dysfunction (LVSD), defined as a left ventricular ejection fraction <50% on echocardiography, within six months of enrollment. The proportion will be calculated as the number of participants with newly identified LVSD divided by the total number of participants enrolled in each study arm. The primary analysis will compare the intervention and control groups according to the intention-to-treat principle, accounting for clustering at the primary healthcare center level.

6 months

二次結果の測定

結果測定
メジャーの説明
時間枠
Key secondary outcome: Diagnostic yield
時間枠:6 months

Diagnostic yield, calculated as the number of newly identified LVSD cases divided by the number of echocardiograms performed in each study arm. The power assessment was based on data from the pilot study (10 PHC in three municipalities, n=342). Expected event proportions were estimated by combining participants from both study arms and were 2.5% for those with clinical suspicion only, 29.2% for those with a positive AI-ECG and no clinical suspicion, and 26.7% for those with both a positive AI-ECG and clinical suspicion. These estimates correspond to overall expected diagnostic yields of 17.7% for the intervention strategy (clinical suspicion or positive AI-ECG) and 9.25% for clinical suspicion alone. Power was assessed under scenarios including 60 or 74 PHCs, 72 or 100 participants per PHC, and echocardiography completion rates of 80% or or 100%. The estimated statistical power was 100% across all scenarios.

6 months
Number needed to diagnose a LVSD
時間枠:6 months

The number needed to diagnose will be calculated as the number of echocardiograms required to identify one case of LVSD.

6 months
Incremental diagnostic yield
時間枠:6 months

The incremental diagnostic yield will be assessed as the difference between study arms in the number of newly identified LVSD cases per 100 participants.

6 months
Agreement between clinical suspicion and AI-ECG positivity in identifying patients with LVSD
時間枠:6 months

Agreement between clinical suspicion and AI-ECG positivity will be assessed using Cohen's kappa coefficient, with 95% confidence intervals accounting for clustering at the primary healthcare center level. The overlap between the two screening strategies will also be described by calculating the proportions of participants identified exclusively through clinical suspicion, exclusively through AI-ECG positivity, or through both approaches. Among participants who undergo echocardiography, the proportion with confirmed LVSD (LVEF <50%) will be calculated for each subgroup to assess the diagnostic yield of the different screening strategies.

6 months
Implementation outcome: Reach
時間枠:6 months

Reach will be assessed as the proportion of eligible individuals who agree to participate in the study.

6 months
Implementation outcome: Effectiveness
時間枠:6 months

Effectiveness will be evaluated based on the primary study outcome, defined as the proportion of participants with newly identified left ventricular systolic dysfunction (LVSD) within six months.

6 months
Implementation outcome: Adoption
時間枠:6 months

Adoption will be assessed as the proportion of physicians who request an echocardiogram after it is offered, stratified by the indication for echocardiography.

6 months
Implementation outcome: Implementation
時間枠:6 months

Implementation will be evaluated through: (1) the proportion of indicated echocardiograms that are performed, (2) the proportion of participants with confirmed LVSD (left ventricular ejection fraction <50%) whose diagnosis is documented in the primary health center's electronic health record, and (3) the proportion of participants with confirmed LVSD who receive appropriate treatment, defined as initiation or optimization of guideline-directed medical therapy (GDMT). ACE inhibitors/ARBs and beta-blockers will be the primary treatment indicators, given their recommendations for LVSD even if asymptomatic, as well as their availability in Brazilian public health system. Other recommended therapies, including ARNIs, MRAs, and SGLT2 inhibitors, will also be assessed according to clinical indications and availability. These outcomes will be stratified by the indication for echocardiography (positive AI-ECG screening result or clinical suspicion).

6 months
Implementation outcome: Maintenance
時間枠:6 months

Maintenance will be assessed by comparing the proportion of indicated echocardiograms that are completed during the first three months versus the last three months of participant enrollment. Analyses will be conducted stratified by the indication for echocardiography (positive AI-ECG screening result or clinical suspicion)

6 months

その他の成果指標

結果測定
メジャーの説明
時間枠
Exploratory: Diagnosis performance of AI-ECG for LVSD only vs. Clinical suspicion only
時間枠:6 months

Exploratory analyses will be performed for the primary outcome and the secondary outcomes 2, 3, and 4 by comparing participants with positive AI-ECG screening results in the intervention arm with those referred for echocardiography exclusively on the basis of clinical suspicion in the control arm.

6 months
Exploratory: Hospitalizations and deaths
時間枠:6 months

Among participants newly diagnosed with left ventricular systolic dysfunction (LVSD), number and proportion of all-cause hospitalizations and all-cause mortality will be assessed during the six-month follow-up period using standardized questionnaires administered by digital tools and review of administrative data at the municipality's level. Information on hospitalizations will include the occurrence, date, and reason for admission, if available. Hospitalization and mortality outcomes will be compared between the intervention and control groups, accounting for the cluster-randomized study design.

6 months
Evaluation of short educational videos by healthcare providers and patients
時間枠:6 months

Videos will be evaluated by a simplified question if they were useful, with a binary answer, followed by a rank from 1 to 10.

6 months
Acceptability of AI-ECG Screening Among Healthcare Professionals
時間枠:6 months

Healthcare professionals' perceptions of the acceptability of AI-ECG screening will be explored through focus groups. Data will be analyzed using thematic analysis to identify perceived benefits, concerns, and experiences with the screening strategy.

6 months
Barriers and Facilitators to AI-ECG Implementation
時間枠:6 months

Barriers and facilitators to implementing AI-ECG screening in primary healthcare will be explored through focus groups with healthcare professionals and other relevant stakeholders. Data will be analyzed thematically to identify factors influencing implementation.

6 months
Budget Impact of AI-ECG Screening
時間枠:Six-month trial data informing longer-term budget impact projections

The budget impact of implementing AI-ECG screening in Brazilian primary healthcare will be estimated by comparing projected healthcare expenditures with and without the screening strategy, including screening, confirmatory testing, treatment, and potential downstream healthcare costs.

Six-month trial data informing longer-term budget impact projections
Cost-Effectiveness of AI-ECG Screening
時間枠:Six-month follow-up, with medium- and long-term projections based on economic modeling.

Cost-effectiveness will be assessed by comparing opportunistic AI-ECG screening with usual care in Brazilian primary healthcare. The analysis will include the costs of AI-ECG screening, confirmatory echocardiography, and subsequent clinical management, using study-specific data and Brazilian microcosting studies. Potential long-term reductions in disease progression and associated healthcare costs will be estimated using evidence from published clinical trials. If available, study data on heart failure hospitalizations, emergency visits, and cardiovascular mortality will inform medium- and long-term projections using Markov models or microsimulation. The analysis will estimate the potential cost savings and health benefits associated with earlier LVSD detection and treatment.

Six-month follow-up, with medium- and long-term projections based on economic modeling.

協力者と研究者

ここでは、この調査に関係する人々や組織を見つけることができます。

スポンサー

協力者

捜査官

  • 主任研究者:Luisa C Brant, MD, PhD、Federal University of Minas Gerais

出版物と役立つリンク

研究に関する情報を入力する責任者は、自発的にこれらの出版物を提供します。これらは、研究に関連するあらゆるものに関するものである可能性があります。

一般刊行物

研究記録日

これらの日付は、ClinicalTrials.gov への研究記録と要約結果の提出の進捗状況を追跡します。研究記録と報告された結果は、国立医学図書館 (NLM) によって審査され、公開 Web サイトに掲載される前に、特定の品質管理基準を満たしていることが確認されます。

主要日程の研究

研究開始 (推定)

2026年11月19日

一次修了 (推定)

2028年2月1日

研究の完了 (推定)

2028年5月1日

試験登録日

最初に提出

2026年11月15日

QC基準を満たした最初の提出物

2026年11月15日

最初の投稿 (実際)

2026年11月20日

学習記録の更新

投稿された最後の更新 (実際)

2026年11月20日

QC基準を満たした最後の更新が送信されました

2026年11月15日

最終確認日

2026年11月1日

詳しくは

本研究に関する用語

その他の研究ID番号

  • 93551425.6.0000.0444
  • 406938/2024-5 (その他の助成金/資金番号:Conselho Nacional de Desenvolvimento Científico e Tecnológico CNPq)
  • U1111-1312-5443 (レジストリ識別子:WHO International Clinical Trials Registry Platform (ICTRP))

個々の参加者データ (IPD) の計画

個々の参加者データ (IPD) を共有する予定はありますか?

はい

IPD 共有時間枠

Data will be available beginning 12 months after publication of the primary study results and will remain available for 5 years.

IPD 共有アクセス基準

De-identified individual participant data underlying the results reported in the primary publication, including relevant baseline characteristics, ECG findings, echocardiographic results, and study outcomes, will be available upon reasonable request, subject to approval by the study investigators and applicable ethical and data protection requirements.

IPD 共有サポート情報タイプ

  • STUDY_PROTOCOL
  • SAP
  • ICF
  • ANALYTIC_CODE
  • CSR

医薬品およびデバイス情報、研究文書

米国FDA規制医薬品の研究

いいえ

米国FDA規制機器製品の研究

いいえ

この情報は、Web サイト clinicaltrials.gov から変更なしで直接取得したものです。研究の詳細を変更、削除、または更新するリクエストがある場合は、register@clinicaltrials.gov。 までご連絡ください。 clinicaltrials.gov に変更が加えられるとすぐに、ウェブサイトでも自動的に更新されます。