• page_banner

News

YPE html PUBLIC “-//W3C//DTD XHTML+RDFa 1.0//EN” “http://www.w3.org/MarkUp/DTD/xhtml-rdfa-1.dtd”>
Objective To evaluate the diagnostic performance of N-terminal B-type natriuretic peptide precursor (NT-proBNP) threshold in acute heart failure, and to develop and validate a decision support tool that combines NT-proBNP concentration with clinical signs.
Conducted 14 studies from 13 countries, including randomized controlled trials and prospective observational studies.
Individual participant-level data from 10 to 369 patients with suspected acute heart failure were pooled for a meta-analysis to estimate NT-proBNP cut-offs. A decision support tool (Heart Failure Diagnosis and Evaluation Collaboration (CoDE-HF)), which combines NT-proBNP with clinical variables to report the likelihood of acute heart failure in individual patients, has been developed and validated.
Results. Overall, 43.9% (4549/10~369) of patients were diagnosed with acute heart failure (73.3% (2286/3119) and 29.0% (1802/6208) of patients with and without prior heart failure). the management’s recommended cut-off threshold of 300 pg/mL has a negative predictive value of 94.6% (95% confidence interval, 91.9% to 96.4%); despite use of age specific rule-in thresholds, the positive predictive value varied at 61.0% (55.3% to 66.4%), 73.5% (62.3% to 82.3%), and 80.2% (70.9% to 87.1%), in patients aged <50 years, 50-75 years, and >75 years, respectively. despite the use of age specific rule-in thresholds, the positive predictive value varied at 61.0% (55.3% to 66.4%), 73.5% (62.3% to 82.3%), and 80.2% (70.9% to 87.1%), in patients aged <50 years, 50-75 years, and >75 years, respectively. Несмотря на использование возрастных порогов правил, положительная прогностическая ценность варьировала в 61,0% (от 55,3% до 66,4%), 73,5% (от 62,3% до 82,3%) и 80,2% (от 70,9% до 87,1%) у пациентов в возрасте <50 лет, 50-75 лет и >75 лет соответственно. Despite the use of age thresholds in the rules, the positive predictive value varied in 61.0% (from 55.3% to 66.4%), 73.5% (from 62.3% to 82.3%) and 80.2% (from 70.9% to 87.1%) in patients aged <50 years, 50-75 years and >75 years, respectively. Despite the use of age thresholds in the rule, among older patients, the positive predictive values ​​were 61.0% (range 55.3% to 66.4%), 73.5% (range 62.3% to 82.3%) and 80. 2% (from 70.9% to 87.1%). ) change between. <50 岁、50-75 岁和>75 岁。 <50岁、50-75岁和>75岁。 <50 лет, 50-75 лет и >75 лет. <50 years, 50-75 years and >75 years. Clinical manifestations varied in most subgroups, especially in groups with obesity, renal insufficiency, or a history of heart failure. CoDE-HF was well calibrated and had excellent discrimination between patients with and without a history of heart failure (area under the receiver operating curve 0.846 (0.830 to 0.862) and 0.925 (0.919 to 0.932), respectively, and a Brier score of 0.130 and 0.099, respectively). ). In patients without prior heart failure, the diagnosis was consistent across all subgroups with a low probability of 40.3% (2502/6208) (negative predictive value 98.6%, 97.8% to 99.1%) and 28.0% ( 1737/6208) the likelihood of acute heart failure was high (positive predictive value 75.0%, 65.7% to 82.5%).
Conclusions In an international collaborative evaluation of the diagnostic performance of NT-proBNP, the recommended thresholds in the guidelines for diagnosing acute heart failure varied widely among important patient subgroups. The CoDE-HF Decision Support Tool integrates NT-proBNP into continuous measurements and other clinical variables, providing a more consistent, accurate and personalized approach.
Nearly 1 million people in the UK suffer from heart failure and the prevalence is expected to rise by around 50% over the next 25 years due to an aging population. 1 Decompensated acute heart failure accounts for 5% of all unplanned hospitalizations. 2 Accurate and timely diagnosis of acute heart failure can be challenging, and both national and international guidelines recommend testing for natriuretic peptides to aid in diagnosis. 345678 Despite these recommendations, testing of the N-terminal B-type natriuretic peptide precursor (NT-proBNP) has not routinely been performed, in part because of concerns about its clinical usefulness in the real world. Studies investigating the diagnostic performance of NT-proBNP have mainly been conducted in relatively small selected cohorts of patients, which limits the ability to generalize the results to clinically important subgroups, such as elderly patients and patients with renal insufficiency or obesity, where these characteristics vary positively. increasingly common in patients with heart failure. 91011 Statistical modeling approaches that take into account patient characteristics to provide more personalized estimates may have more consistent diagnostic performance across subsets of patients. 12
Although many models have been developed to predict prognosis in patients with heart failure, few models can help diagnose acute heart failure. 13141516171819 Previous attempts have had many advantages but have included subjective variables such as clinicians’ pre-test probabilities or patient descriptions of symptoms. In addition, they included NT-proBNP as a binary variable and did not take into account dynamic and non-linear interactions between NT-proBNP and other clinical variables. Previous attempts to develop and validate diagnostic scales have also included a limited number of patients from a single facility, which precluded evaluation of efficacy within subgroups and limited the possibility of external generalization.
In this collaborative international analysis, we assessed the diagnostic performance of the guidelines’ recommended NT-proBNP thresholds for acute heart failure in a subset of patients. Subsequently, we developed and validated a decision support tool for patients with suspected acute heart failure that used a statistical model to combine NT-proBNP concentrations with clinical characteristics.
We conducted a systematic review to identify studies evaluating the diagnostic performance of NT-proBNP in patients with suspected acute heart failure. We updated a previous review by Roberts et al1 to include the keywords “heart failure” and “natriuretic peptides” by searching Embase, Medline, and the Cochrane Central Register of Controlled Trials for titles and abstracts published on 18 August 2021 (Supplementary Text 1). Studies were considered eligible if they met the following predefined inclusion criteria: enrollment of patients aged ≥18 years with suspected acute heart failure in the emergency setting, measurement of NT-proBNP in blood samples obtained during the patient’s initial assessment on the day of admission, and The diagnosis of acute heart failure was made using acceptable reference standards. Two investigators (KKL and MA) independently reviewed all studies identified by a systematic literature search, and a third (NLM) made a conflict decision using a predefined protocol (PROSPERO registry: CRD42019159407).
We contacted the respective authors for all eligible cohorts to request information on NT-proBNP concentrations, confirmed diagnosis of acute heart failure, demographics (age, gender, race), prior history (heart failure, coronary artery disease, anonymous individual patient level) . data on diabetes), hypertension, hyperlipidemia, smoking, asthma, chronic obstructive pulmonary disease, chronic kidney disease), physiological parameters (heart rate and blood pressure) at initial examination, clinical haematological and biochemical characteristics. We checked with all relevant authors for accuracy, definitions of variables, and completeness prior to agreement. All studies were conducted in accordance with the Declaration of Helsinki and were ethically approved to allow sharing of data at the individual patient level for this meta-analysis. Two investigators (KKL and MA) independently assessed the risk of bias for each study using the Study Quality Assessment Tool in Diagnostic Accuracy, version 2 (QUADAS-2), and 20 conflicts were resolved by a third party (NLM).
We derived meta-estimates with 95% confidence intervals of the sensitivity, specificity, negative predictive value, and positive predictive value of the guideline recommended NT-proBNP rule-out threshold (300 pg/mL)58 and age specific rule-in thresholds (450, 900, and 1800 pg/mL for patients aged <50, 50-75, and >75 years, respectively)7 for acute heart failure by using a two stage approach, with estimates calculated separately within each study and then pooled across studies in a binomial-normal random effects model using the DerSimonian and Laird method.21 We further evaluated the performance of these thresholds in pre-specified subgroups stratified by age, sex, ethnicity, body mass index, renal function, anaemia, and the presence of comorbidities (previous heart failure, hypertension, hyperlipidaemia, diabetes mellitus, atrial fibrillation, chronic obstructive pulmonary disease). Мы получили метаоценки с 95% доверительными интервалами чувствительности, специфичности, отрицательной прогностической ценности и положительной прогностической ценности рекомендуемого порога исключения NT-proBNP (300 пг/мл)58 и возрастных порогов исключения ( 450, 900 и 1800 пг/мл для пациентов в возрасте <50, 50-75 и >75 лет соответственно)7 для острой сердечной недостаточности с использованием двухэтапного подхода, при этом оценки рассчитываются отдельно в каждом исследовании, а затем объединяются по исследованиям. в модели биномиально-нормальных случайных эффектов с использованием метода ДерСимониана и Лэрда.21 Далее мы оценили эффективность этих пороговых значений в предварительно определенных подгруппах, стратифицированных по возрасту, полу, этнической принадлежности, индексу массы тела, функции почек, анемии и наличию сопутствующие заболевания (сердечная недостаточность в анамнезе, артериальная гипертензия, гиперлипидемия, сахарный диабет, мерцательная аритмия, хроническая обструктивная болезнь легких).我们对指南推荐的 NT-proBNP 排除阈值 (300 pg/mL)58 和年龄特定的排除阈值 (对于年龄 <50、50-75 和 >75 岁的患者,急性心力衰竭分别为 450、900 和 1800 pg/mL)7,采用两阶段方法,在每项研究中分别计算估计值,然后在研究中汇总在使用 DerSimonian 和 Laird 方法的二项式正态随机效应模型中。21 我们进一步评估了这些阈值在按年龄、性别、种族、体重指数、肾功能、贫血和存在合并症(既往心力衰竭、高血压、高脂血症、糖尿病、心房颤动、慢性阻塞性肺病)。我们 对 指南 的 nt-probnp 排除 阈值 (300 пг/мл) 58 和 特定 的 排除 阈值 阈值 (对于 年龄 <50、50-75 和> 75 岁 患者 , 急性 心力 分别 为 450、900 和 1800 стр. /мл) 7 , 采用 阶段 方法 , 在 每 项 研究 中 分别 计算 估计值 然后 在 研究 中 汇总 在 使用 使用 使用 和 和 方法 二 项式 正态 随机 效应 中 。21 我们 评估 阈值 在 随机 效应 中。。。。 我们 评估按 年龄 性别 、 种族 、 体重 指数 、 肾 功能 、 和 存在 合并症 (既往 心力 衰竭 、 高 血压 高脂血症 、 糖尿病 心 房 颤动 慢性 阻塞 性 肺病)。。。。使用 相同 方法 , 我们 随后 评估 了 nt-probnp 浓度 一系列 浓度 范围 的 诊断 性能 性能 , 以 排除 阈值 , 该 阈值 确定 最 比例 的 患者 具有 的 阴性 预测值 ≥ 98 % ≥ 75%。
We calculated a value (0-100) corresponding to the probability of developing acute heart failure in an individual patient using statistical modeling. Due to significant differences in the prevalence of comorbidities and acute heart failure, we developed and validated models for patients with and without heart failure, respectively. We used NT-proBNP concentrations as a continuous measure and selected simple objective clinical variables known to be associated with acute heart failure that had the highest relative importance during the training phase of our model (age, estimated glomerular filtration rate, hemoglobin, mass index bodies). , heart rate, blood pressure, peripheral edema, chronic obstructive pulmonary disease and ischemic heart disease) (Supplementary text 2).
In developing Code-HF, we evaluated four different statistical models: Generalized Linear Mixed Models, Naive Bayes, Random Forest, and Extreme Gradient Boost (XGBoost) (Supplementary Text 2). 222324 To account for missing data in the studies (Supplementary Figure A), we multiplied the imputed 10 datasets using jointly modeled multiple imputation with a randomized study-specific covariance matrix corresponding to a Monte Carlo Markov chain algorithm. 25 We performed multiple imputation for all variables included in the model except for NT-proBNP. We performed 10 iterations of 10-fold cross-validation for each model and used the median estimate of the iteration and imputed datasets as the CoDE-HF estimate for each patient. Subsequently, we identified scores that classified the largest proportion of patients with high or low probability of acute heart failure, with the best performance for exclusion (75% positive predictive value and 90% specificity) and for exclusion (98% negative predictive value and 90% specificity) % sensitivity) in acute heart failure.
We assessed the performance of each model on a range of diagnostic metrics (area under the receiver operating curve, Brier score, proportion of patients achieving high and low probability optimal criteria, and positive and negative predictive values ​​for subgroups of patients). The Brier score is a discrimination and calibration measure calculated by taking the standard error between predicted probabilities and observations. 26 We chose the most efficient model for the Code-HF decision support tool. We evaluate the performance of CoDE-HF using decision curve analysis and internal and external cross-validation. Briefly, this approach iteratively ignores one study at a time for external validation and uses the remaining studies to develop the model. 27 We did not enter values ​​into the externally validated datasets and therefore did not externally validate for most of the studies. The variable was completely absent (Supplementary Figure A). We used R version 4.1.2 for all analyses.
Patients and members of the public commission participated in the interpretation of the results. There are plans to disseminate the results to the relevant patient community.
We contacted investigators from 30 eligible studies, of which 19 responded. Fourteen studies (12 prospective cohort studies and two randomized controlled trials) provided individual patient-level data from 10 to 369 patients with suspected acute heart failure (mean age 69.3 years; 53.3% men) from 13 countries (Table 1 ). Figure B; Supplementary Tables A and B) 1528293031323334353637383940 All studies were conducted in the emergency department, with the exception of one study that included cardiac and pulmonary inpatients (mean 488 patients per study (quartile. Bit spacing 322–1053)). Overall, 43.9% (4549/10,369) of patients had a confirmed diagnosis of acute heart failure (median study prevalence 46% (31–54%)). In patients with prior heart failure, the incidence of acute heart failure was higher than in patients without heart failure (73.3% (2286/3119) vs. 29.0% (1802/6208)) (Supplementary Table C).
Baseline characteristics of patients stratified by diagnosis of acute heart failure. Values ​​are numbers (percentages) unless otherwise noted
At the guideline’s recommended exclusion threshold of 300 pg/mL, the combined meta-estimation of negative predictive value, sensitivity, positive predictive value, and specificity of NT-proBNP in the general population was 94.6% (95% confidence interval, 91.9%). to 96.4%), 96.8% (from 94.6% to 98.1%), 62.9% (from 51.3% to 73.3%) and 49.3% (from 35.4 % to 63.4%) (Figure 1; Supplementary Table D). Overall, 30.4% (3148/10,369) of patients had NT-proBNP levels below 300 pg/mL. However, there was marked heterogeneity between patient subgroups and studies (Figure 2; Figure 3; Supplementary Figures C and D). Negative predictive values ​​were lower in patients ≥75 years of age (88.2%, from 83.5% to 91.8%), as well as in patients with a history of heart failure (79.4%, from 68.4% to 87.3%) and obesity (90.4%, from 84.5% to 87.3%). 94.2%.
N-terminal threshold of pro-B-type natriuretic peptide (NT-proBNP) in acute heart failure. Top left: Negative predictive value of NT-proBNP concentration to exclude the diagnosis of acute heart failure. Bottom left: Cumulative proportion of patients with suspected acute heart failure with NT-proBNP concentrations below each threshold. Top right: Positive predictive value of NT-proBNP concentration for the diagnosis of acute heart failure. Bottom right: Cumulative proportion of patients with suspected acute heart failure with NT-proBNP concentrations above each threshold.
Diagnostic performance of the guidelines-recommended N-terminal thresholds for pro-B-type natriuretic peptide in patient subgroups: negative predictive value threshold of 300 pg/mL. COPD = chronic obstructive pulmonary disease; eGFR = estimated glomerular filtration rate
Diagnostic performance of guideline recommended NT-proBNP thresholds across patient subgroups: positive predictive value of age specific thresholds across patient subgroups (450, 900, and 1800 pg/mL for <50, 50-75, and >75 years, respectively). Diagnostic performance of guideline recommended NT-proBNP thresholds across patient subgroups: positive predictive value of age specific thresholds across patient subgroups (450, 900, and 1800 pg/mL for <50, 50-75, and >75 years, respectively). Диагностическая эффективность рекомендованных в руководстве порогов NT-proBNP для подгрупп пациентов: положительная прогностическая ценность возрастных порогов для подгрупп пациентов (450, 900 и 1800 пг/мл для <50, 50-75 и >75 лет соответственно). Diagnostic performance of guideline-recommended NT-proBNP thresholds for patient subgroups: positive predictive value of age-specific thresholds for patient subgroups (450, 900, and 1800 pg/mL for <50, 50-75, and >75 years, respectively).指南推荐的跨患者亚组的NT-proBNP 阈值的诊断性能:跨患者亚组的年龄特异性阈值的阳性预测值(分别为450、900 和1800 pg/mL,<50、50-75 和>75 岁)。指南 推荐 的 跨患者 的 nt-Probnp 阈值 的 性能 : 跨患者 亚组 的 年龄 特异性 的 阳性 (分别 为 为 450、900 和 1800 pg/ml , <50、50-75 和> 75岁)。 Диагностическая эффективность порогов NT-proBNP, рекомендованных руководством, для подгрупп пациентов: положительная прогностическая ценность возрастных порогов для подгрупп пациентов (450, 900 и 1800 пг/мл, <50, 50-75 и >75 соответственно возрасту) . Diagnostic performance of guidelines-recommended NT-proBNP thresholds for patient subgroups: positive predictive value of age-specific thresholds for patient subgroups (450, 900, and 1800 pg/mL, <50, 50-75, and >75, respectively for age) . COPD = chronic obstructive pulmonary disease; eGFR = estimated glomerular filtration rate
The pooled meta-estimations of the positive predictive value of the age cut-offs of the NT-proBNP 450, 900, and 1800 pg/mL rule were 61.0% (55.3% to 66.4%), 73.5% (62.3% to 82 .3%) and 80.2%, respectively (70.9% to 87.1%) (Table 2). The corresponding specificities were 87.8% (79.5% to 93.0%), 81.1% (72.6% to 87.5%), and 73.1% (65.2% to 79. eight%). Overall, 48.7% (5052/10,369) of patients with suspected acute heart failure had NT-proBNP above these age thresholds. Despite heterogeneity across age groups, kidney function, and the prevalence of acute heart failure, within subgroups, the age cut-offs of the rules had positive predictive values ​​above a single cut-off of 300 pg/mL (Supplementary Figure EI).
Diagnostic performance of N-terminal B-type natriuretic peptide precursor (NT-proBNP) age threshold for acute heart failure
Overall, we identified seven studies at high risk of bias (Supplementary Table A). In sensitivity analyzes limited to studies blinded to NT-proBNP concentrations for adjudication of acute heart failure and studies with a low risk of bias, the guidelines’ recommended diagnostic characteristics and age cut-offs for NT-proBNP remained unchanged (Supplementary Tables E and F). .
The 100 pg/mL NT-proBNP threshold met our best exclusion criteria with a combined negative predictive value of 97.8% (range 95.8% to 98.8%) and a sensitivity of 99.3% (range 98.5% to 99.7 %) (Supplementary Table D). However, only 17.9% (1851/10~369) of patients had NT-proBNP concentrations below 100 pg/mL, and they were negative in elderly patients and patients with heart failure, coronary artery disease, and a history of disorders Predictions remain poor . . Kidney function (Supplementary Figure J). Similarly, the 1000 pg/mL NT-proBNP cut-off met our best evaluation criteria with a positive predictive value of 74.9% (64.4% to 83.2%) and a specificity of 76.1% (65.6% to 84.2%). was lower. Difference. It was also lower in patient subgroups, especially those with no previous history of heart failure (positive predictive value 62%, 41% to 79%) (Supplementary Table D; Supplementary Figure K).
The extreme gradient boosting (XGBoost) model and the generalized linear mixed model were the best performing models (area under the curve in the total training cohort 0.925 (95% CI 0.919 to 0.932) and 0.931 (0.925 to 0.937), respectively) (Supplementary Text 2) . Although the performance of XGBoost is similar to generalized linear mixed models, the key advantage of XGBoost is its ability to calculate scores when there are missing values. This is an important feature that we hope to implement in the CoDE-HF decision support tool to facilitate its implementation in clinical practice, which is why we chose the XGBoost model as the final model for CoDE-HF.
CoDE-HF was well calibrated and had excellent discrimination in patients with and without heart failure (area under the receiver operating curve 0.846 (0.830 to 0.862) and 0.925 (0.919 to 0.932) and a Brier score of 0.130 and 0.130, respectively). 0.099) (Fig. 4; Supplementary Fig. L). A CoDE-HF score of 4.7 provides a negative predictive value of 98.6% (97.8% to 99.1%) and a sensitivity of 98.1% (96.9% to 98.9%) (Supplementary Table G) , and a score of 51.2 provides a positive predictive value. value 75.0% (65.7%) 82.5%), specificity was 92.2% (87.5% to 95.2%) of patients without a history of heart failure. These inclusion and exclusion rates had similar diagnostic performance in all subgroups (Figure 5, Figure 6, Figure 7). If these scores were applied in patients with suspected acute heart failure, CoDE-HF would identify 40.3% (2502/6208) at low probability (<4.7) and 28.0% (1737/6208) at high probability (≥51.2) of acute heart failure. If these scores were applied in patients with suspected acute heart failure, CoDE-HF would identify 40.3% (2502/6208) at low probability (<4.7) and 28.0% (1737/6208) at high probability (≥51.2) of acute heart failure. Если бы эти показатели применялись к пациентам с подозрением на острую сердечную недостаточность, CoDE-HF выявил бы 40,3% (2502/6208) при низкой вероятности (<4,7) и 28,0% (1737/6208) при высокой вероятности (≥51,2) сердечной недостаточности. If these rates were applied to patients with suspected acute heart failure, CoDE-HF would detect 40.3% (2502/6208) with a low probability (<4.7) and 28.0% (1737/6208) with a high probability (≥51.2) heart failure. acute heart failure.如果将这些评分应用于疑似急性心力衰竭的患者,CoDE-HF 将识别出40.3% (2502/6208) 的低概率(<4.7) 和28.0% (1737/6208) 的高概率(≥51.2)急性心力衰竭。如果 将 这些 评分 应用 于 急性 心力 衰竭 的 , , code-hf 识别 出 出 40.3% (2502/6208) 的 低概率 低概率 (<4.7) 和 28.0% (1737/6208) 的 高概率 高概率 (≥51.2) 急性 心力 心力 急性 急性 急性 急性 急性 急性 急性 急性 急性 急性 急性 急性 急性 急性衰竭。 Если бы эти оценки применялись к пациентам с подозрением на острую сердечную недостаточность, CoDE-HF выявил бы 40,3% (2502/6208) низкой вероятности (<4,7) и 28,0% (1737/6208) высокой вероятности (≥51,2) острой сердечной недостаточности. If these scores were applied to patients with suspected acute heart failure, CoDE-HF would reveal 40.3% (2502/6208) low probability (<4.7) and 28.0% (1737/6208) high probability (≥ 51.2) acute heart failure. exhaustion. Among patients with pre-existing heart failure, none of the scores in the training cohort met our target exclusion criteria. The CoDE-HF score was 84.5, the positive predictive value was 92.7% (89.1% to 95.2%), and the specificity was 90.2% (84.0% to 94.1%). This assessment will identify 45.5% (1420/3119) of patients with a high probability of developing acute heart failure (Fig. 8). In the decision curve analysis at all threshold probabilities, CoDE-HF had a higher net gain than NT-proBNP alone (Supplementary Figure M). CoDE-HF scores were slightly reduced with no history of training (area under the receiver working curve was 0.922 (0.916 to 0.929) and 0.841 (0.825 to 0.825 in patients without heart failure and pre-heart failure) 0.857)). Internal and external cross-validation performed well in the cohort of both models (Supplementary Figure N).
The Heart Failure Joint Diagnosis and Evaluation Scale (CoDE-HF) was calibrated to the observed proportion of patients with acute heart failure. The dotted line indicates the ideal calibration. Each point corresponds to 100 patients. Top: CoDE-HF calibration in a patient without prior heart failure. Bottom: CoDE-HF calibration in a patient with a history of heart failure.
Diagnostic performance of the Heart Failure Collaborative Diagnosis and Evaluation Scale (CoDE-HF) in patient subgroups. The CoDE-HF exclusion score had a negative predictive value of 4.7 in the subgroup of patients without a history of heart failure. CoDE-HF uses N-terminal natriuretic peptide type B precursor concentrations as continuous measurements and predefined simple objective clinical variables (age, estimated glomerular filtration rate (eGFR), hemoglobin, body mass index, heart rate, blood pressure, peripheral edema, chronic obstructive pulmonary disease (COPD) and coronary heart disease) provides an individual assessment of the likelihood of a diagnosis of acute heart failure.
Diagnostic performance of the CoDE-HF scale on the Collaboration for the Diagnosis and Evaluation of Heart Failure scale in subgroups of patients. The CoDE-HF rule score had a positive predictive value of 51.2 in the subgroup of patients without a history of heart failure. CoDE-HF pooled NT-proBNP concentrations as continuous measurements and predefined simple objective clinical variables (age, estimated glomerular filtration rate (eGFR), hemoglobin, body mass index, heart rate, blood pressure, peripheral edema, chronic obstructive pulmonary disease ( COPD)). coronary artery disease) provide an individual assessment of the likelihood of a diagnosis of acute heart failure
Diagnostic performance of the Collaboration for Diagnosis and Evaluation of Heart Failure (CoDE-HF) scale in patient subgroups. The CoDE-HF rule score had a positive predictive value of 84.5 in patients with a history of heart failure in a subgroup of patients. CoDE-HF pooled NT-proBNP concentrations as continuous measurements and predefined simple objective clinical variables (age, estimated glomerular filtration rate (eGFR), hemoglobin, body mass index, heart rate, blood pressure, peripheral edema, chronic obstructive pulmonary disease ( COPD)). coronary artery disease) provide an individual assessment of the likelihood of a diagnosis of acute heart failure
The Heart Failure Joint Diagnosis and Assessment Scale (CoDE-HF) is not diagnostically effective in patients with a history of heart failure. Top: Negative and positive predictive values ​​for CoDE-HF scores. The blue vertical dotted line indicates the target elimination score of 4.7. The red vertical dotted line indicates the target rule score of 51.2. Bottom: density map of CoDE-HF scores in patients without a history of heart failure. Exclusion and rule targets identified 40.3% of patients with low probability and 28.0% with high probability, respectively.
Patients identified as low-probability by CoDE-HF had significantly lower all-cause and CV mortality at 30 days and 1 year than patients identified as intermediate and high-probability (30-day all-cause mortality: 1. 0% compared to 4.0% and 10.4%). mortality from all causes within one year: 5.9% versus 17.8% and 33.4%, respectively; 30-day mortality from cardiovascular diseases: 0.2% vs. 0.8% and 4.1%; annual mortality from cardiovascular diseases: 1.4% versus 3.4% and 16.3%, respectively) (Fig. 9). In patients with NT-proBNP concentrations <300 pg/mL compared with those ≥300 pg/mL, the all cause mortality rates were 0.8% versus 7.6% at 30 days and 5.9% versus 26.6% at one year, respectively, and the cardiovascular mortality rates were 0.1% versus 2.6% at 30 days and 1.3% versus 10.2% at one year, respectively (supplementary table H; supplementary figure O). In patients with NT-proBNP concentrations <300 pg/mL compared with those ≥300 pg/mL, the all cause mortality rates were 0.8% versus 7.6% at 30 days and 5.9% versus 26.6% at one year, respectively, and the cardiovascular mortality rates were 0.1% versus 2.6% at 30 days and 1.3% versus 10.2% at one year, respectively (supplementary table H; supplementary figure O). У пациентов с концентрацией NT-proBNP <300 пг/мл по сравнению с таковой ниже 300 пг/мл смертность от всех причин составила 0,8% по сравнению с 7,6% через 30 дней и 5,9% по сравнению с 26,6% через один год, соответственно, и показатели смертности от сердечно-сосудистых заболеваний составили 0,1% по сравнению с 2,6% через 30 дней и 1,3% по сравнению с 10,2% через один год соответственно (дополнительная таблица H; дополнительный рисунок O). In patients with an NT-proBNP concentration <300 pg/ml compared with that below 300 pg/ml, all-cause mortality was 0.8% compared with 7.6% at 30 days and 5.9% compared with 26, 6% at one year, respectively, and CV mortality rates were 0.1% versus 2.6% at 30 days and 1.3% versus 10.2% at one year, respectively (Supplementary Table H; Supplementary Figure O). NT-proBNP 浓度<300 pg/mL 的患者与≥300 pg/mL 的患者相比,30 天全因死亡率分别为0.8% 和7.6%,一年时分别为5.9% 和26.6%,以及心血管死亡率在30 天时分别为0.1% 和2.6%,一年时分别为1.3% 和10.2%(补充表H;补充图O)。 NT-PROBNP 浓度 <300 pg/ml 的 与 ≥ ≥300 pg/ml 的 相比 , , 30 天全 因 分别 为 为 为 为 0.8% 和 7.6% , 年 分别 为 为 5.9% 和 26.6% , 心血管 心血管 以及 以及 以及 以及 以及 以及 以及 以及 以及 以及 以及 以及 以及 以及 以及 以及 以及 以及死亡率在30 天时分别为0.1% 和2.6%,一年时分别为1.3% 和10.2%(补充表H;补充图O)。 Пациенты с концентрацией NT-proBNP <300 пг/мл по сравнению с ≥300 пг/мл имели 30-дневную смертность от всех причин 0,8% и 7,6% соответственно, 5,9% и 26,6% в течение одного года, а также сердечно-сосудистую смертность. Patients with NT-proBNP concentrations <300 pg/mL compared with ≥300 pg/mL had a 30-day all-cause mortality of 0.8% and 7.6%, respectively, 5.9% and 26.6% within one years, and cardiovascular mortality. were 0.1% and 2.6% at 30 days and 1.3% and 10.2% at 1 year (Supplementary Table H; Supplementary Figure O).
Cumulative all-cause mortality rate stratified by Collaborative for the Diagnosis and Evaluation of Heart Failure (CoDE-HF) probability group
We performed a meta-analysis of individual patient-level data to evaluate the diagnostic performance of NT-proBNP thresholds in more than 10 patients with suspected acute heart failure included in 14 prospective studies from 13 countries that we designed and implemented using NT-proBNP. proBNP as a decision support tool for continuous measurement. We report several important findings. First, the guideline’s recommended thresholds for excluding acute heart failure are not uniform across important patient subgroups. 3 Although the general population and several subgroups, including younger patients and women, performed well, older patients and women had significantly lower negative predictive values. In patients with obesity or prior heart failure, the false negative rate ranged from one in ten to one in five. Secondly, age-stratified thresholds have shown themselves well in the diagnosis of acute heart failure. However, the positive predictive value was lower in younger patients. Third, although our optimized NT-proBNP cut-offs of 100 pg/mL to rule out acute heart failure and 1000 pg/mL to rule on acute heart failure have excellent negative and positive predictive value in the general population, older patients are worse off. in patients with acute heart failure. Previous heart failure and obesity. Finally, we have developed and validated a decision support tool, the CoDE-HF score, with excellent diagnostic performance in all patient subgroups. This decision support tool excluded and ruled out acute heart failure more accurately than any method using only the NT-proBNP threshold.
To our knowledge, this is the largest study to date evaluating the diagnostic performance of NT-proBNP in acute heart failure. All included studies were prospective and final diagnoses were made by a panel of clinicians using all available information. It is important to note that the availability of data at the individual patient level in large study populations allows reliable assessment of the diagnostic performance of all possible NT-proBNP thresholds in subgroups of patients, as well as the development and validation of new diagnostic scales.
Most national and international guidelines recommend using an NT-proBNP cut-off value of 300 pg/mL to rule out acute heart failure58 based on numerous previous studies344142 reporting a negative predictive value of 98% at this cut-off. the diagnostic performance of important subgroups of patients could not be assessed. Our study enrolled three times as many patients as previous study-level meta-analyses,3 which showed low overall negative predictive value at a cut-off of 300 pg/mL with a pooled meta-estimation of 94.6%. More importantly, the negative predictive value was significantly lower in key subgroups such as elderly patients and patients with pre-existing heart failure, coronary artery disease, and obesity. In addition, nearly 70% of patients had NT-proBNP concentrations above the 300 pg/ml cut-off point, highlighting the limitations of using a single cut-off point in practice. Although the lower cutoff of 100 pg/mL achieved an overall negative predictive value of 98%, it performed poorly in an important subgroup of patients. In addition, age and optimized thresholds for acute heart failure showed heterogeneity across patient subgroups, especially among those with no prior history of heart failure. This heterogeneity in diagnostic performance is of particular concern as our patient population ages and has more comorbidities. This raises the question of whether clinical guidelines should continue to recommend the use of uniform cut-offs when NT-proBNP is affected by many risk factors and comorbidities.
To improve the clinical usefulness of NT-proBNP, we developed and externally validated the CoDE-HF evaluation of a clinical decision support tool. This score combines NT-proBNP as a continuous measure with simple objective clinical variables to provide an individual assessment of the likelihood of a diagnosis of acute heart failure. We show that the diagnostic performance of the CoDE-HF score is robust in subgroups of patients. CoDE-HF was able to rule out and rule out the diagnosis of acute heart failure in a larger proportion of patients than the optimized NT-proBNP threshold alone. Furthermore, in our decision curve analysis, we found that CoDE-HF has a higher net benefit than NT-proBNP alone, across the entire threshold probability range. We believe that this conclusion is intuitive since NT-proBNP is a continuous marker of risk and its concentration depends on other patient-related factors such as body mass index, age, and renal function. 434445 While these ratios are based on predefined performance criteria, we recognize that these goals may not be universally supported and that different healthcare facilities may have different risk tolerances. The advantage of using decision support tools such as CoDE-HF is that clinicians or institutions can select diagnostic performance criteria to be used for local decision making based on their priorities and the availability of echocardiography or heart failure specialists. .
We expect that our new decision support tool, Code-HF, can improve the triage of patients with suspected acute heart failure seen in various medical specialties and transform their care, facilitating more accurate diagnosis. Previous studies have shown that timely and accurate evidence-based treatment of patients with acute heart failure can significantly reduce mortality and length of hospital stay, and delay is associated with worse outcomes. 46 In addition, routinely collected CoDE-HF uses variables and can therefore be incorporated into clinical workflows as part of the emergency department triage pathway to enable more efficient evaluation. Currently, the vast majority of patients with suspected acute heart failure have echocardiography on admission to determine their treatment, but only a subset of patients are ultimately diagnosed. 2 Echocardiography is a relatively time-consuming and resource-intensive specialty study We expect that the use of CoDE-HF for more accurate and informed use of specialty services such as echocardiography can lead to significant cost savings and efficiency for the healthcare system. . In addition, cost savings can be achieved through the outpatient treatment of low-risk patients. A prospective study is currently needed to evaluate the clinical and cost-effectiveness of different CoDE-HF decision thresholds in clinical practice.
We acknowledge several limitations. First, we were able to obtain individual patient-level data for 14 of the 30 studies that met our eligibility criteria, so selection bias can be introduced. However, eligible studies that were not included had similar prevalence of acute heart failure, publication dates, and geographic coverage, and populations had similar demographic and clinical characteristics to the included populations. Second, when information from multiple studies was pooled, some studies were missing data for some variables. To maximize the use of information, we used a hierarchical method of multiple imputation. Third, we did not record ECG and chest X-ray data sequentially to include them in our model. Interpretation of NT-proBNP in patients with suspected acute heart failure should be done in conjunction with these studies, 47 and further studies are needed to determine whether methods combining these studies can improve CoDE-HF scores. Fourth, not all studies made diagnoses without taking into account the results of the NT-proBNP test. In our sensitivity analysis, when we excluded two studies with unblinded definition, there was no change in diagnostic performance. Fifth, the established diagnosis of acute heart failure did not allow to differentiate between heart failure with reduced ejection fraction and heart failure with preserved ejection fraction. 48 The increasing prevalence of HF with preserved ejection fraction in elderly patients may explain some of the heterogeneity observed with age, but current guidelines recommend HF with reduced ejection fraction and preserved EF. Heart failure uses the same NT-ProBNP threshold. 58 Sixth, although most studies consistently enrolled patients with acute dyspnoea, the prevalence of acute heart failure was high and selection bias may have been present. However, the effectiveness of the guidelines-recommended NT-proBNP cut-offs and age limits did not change in sensitivity analyses, except for studies with a high risk of bias. Finally, acute heart failure is a clinical syndrome, and the diagnosis itself has inherent uncertainty and research variability. This uncertainty may be greater in the elderly, which may partly explain the observed heterogeneity in diagnostic results.
We have shown that the diagnostic performance of the NT-proBNP cut-off values ​​recommended in the guidelines for acute heart failure varies across an important subgroup of patients. We have developed and validated the CoDE-HF score, which combines NT-pro-BNP as a continuous measure with clinical variables to determine the likelihood of acute heart failure in individual patients using a statistical model. This decision support instrument accurately ruled out and ruled out acute heart failure and was consistently performed in all subgroups. Prospective studies are currently needed to evaluate the impact of implementing this decision support tool on the use of healthcare resources and patient outcomes.
Diagnosis of acute heart failure can be difficult because patients often present with non-specific symptoms.
Most national and international guidelines recommend testing the N-terminal B-type natriuretic peptide precursor (NT-proBNP) for the diagnosis of acute heart failure.
NT-proBNP testing has not been universally applied due to problems with diagnostic performance in clinically important subgroups of patients.
The recommended NT-proBNP thresholds for acute heart failure in the guidelines have relatively poor diagnostic performance in important patient subgroups.
A validated decision support tool has been developed that combines NT-pro-BNP as a continuous measure with clinical variables using statistical modeling.
This tool more accurately ruled out and ruled out acute heart failure than any method using the NT-proBNP threshold alone and was performed consistently across all subgroups.
All studies were conducted in accordance with the Declaration of Helsinki and were ethically approved to allow sharing of patient-level data for this analysis.
The R code and anonymous data used to develop and validate the CoDE-HF score are available to researchers at the request of the respective author.


Post time: Sep-23-2022