Author + information
- Received August 3, 2005
- Revision received December 14, 2005
- Accepted December 16, 2005
- Published online June 6, 2006.
- Milton Packer, MD, FACC⁎,⁎ (, )
- William T. Abraham, MD, FACC†,
- Mandeep R. Mehra, MD, FACC‡,
- Clyde W. Yancy, MD, FACC⁎,
- Christine E. Lawless, MD, FACC§,
- Judith E. Mitchell, MD, FACC∥,
- Frank W. Smart, MD, FACC¶,
- Rachel Bijou, MD, FACC#,
- Christopher M. O’Connor, MD, FACC⁎⁎,
- Barry M. Massie, MD, FACC††,
- Ileana L. Pina, MD, FACC‡‡,
- Barry H. Greenberg, MD, FACC§§,
- James B. Young, MD, FACC∥∥,
- Daniel P. Fishbein, MD, FACC¶¶,
- Paul J. Hauptman, MD, FACC##,
- Robert C. Bourge, MD, FACC⁎⁎⁎,
- John E. Strobeck, MD, PhD, FACC†††,
- Srinvivas Murali, MD, FACC‡‡‡,
- Douglas Schocken, MD, FACC§§§,
- John R. Teerlink, MD, FACC††,
- Wayne C. Levy, MD, FACC¶¶,
- Robin J. Trupp, MSN, RN⁎,
- Marc A. Silver, MD, FACC∥∥∥,
- Prospective Evaluation and Identification of Cardiac Decompensation by ICG Test (PREDICT) Study Investigators and Coordinators
- ↵⁎Reprint requests and correspondence:
Dr. Milton Packer, University of Texas Southwestern Medical Center, 5323 Harry Hines Boulevard, Room E5.506P, Dallas, Texas 75390-8822.
Objectives This study sought to assess the potential utility of impedance cardiography (ICG) in predicting clinical deterioration in ambulatory patients with heart failure (HF).
Background Impedance cardiography uses changes in thoracic electrical impedance to estimate hemodynamic variables, but its ability to predict clinical events has not been evaluated.
Methods We prospectively evaluated 212 stable patients with HF and a recent episode of clinical decompensation who underwent serial clinical evaluation and blinded ICG testing every 2 weeks for 26 weeks and were followed up for the occurrence of death or worsening HF requiring hospitalization or emergent care.
Results During the study, 59 patients experienced 104 episodes of decompensated HF (16 deaths, 78 hospitalizations, and 10 emergency visits). Multivariate analysis identified 6 clinical and ICG variables that independently predicted an event within 14 days of assessment. These included three clinical variables (visual analog score, New York Heart Association functional class, and systolic blood pressure) and three ICG parameters (velocity index, thoracic fluid content index, and left ventricular ejection time). The three ICG parameters combined into a composite score was a powerful predictor of an event during the next 14 days (p = 0.0002). Visits with a high-risk composite score had 2.5 times greater likelihood and those with a low-risk score had a 70% lower likelihood of a near-term event compared with visits at intermediate risk.
Conclusions These results suggest that when performed at regular intervals in stable patients with HF with a recent episode of clinical decompensation, ICG can identify patients at increased near-term risk of recurrent decompensation.
The course of patients with chronic heart failure (HF) is marked by periodic episodes of clinical decompensation that not only impair the quality of life and may be fatal but also consume substantial health care resources, primarily because of the costs of hospitalization (1). Heart failure management programs have been developed to reduce the frequency and severity of these clinical events (2), but their effectiveness may be limited by physicians’ difficulty in identifying patients at imminent risk. Reliable prediction of these events may afford physicians the opportunity to intervene aggressively and potentially minimize the need for hospitalization or the risk of a serious adverse outcome.
Some investigators have hypothesized that the measurement of hemodynamic variables might identify patients likely to deteriorate clinically during follow-up (3–5). However, prior studies have shown the prognostic value of hemodynamic measurements only over periods too long to allow for immediate intervention to prevent imminent occurrence of a serious clinical event (5). However, periodic reassessment might be useful in identifying high-risk patients if cardiocirculatory variables could be estimated accurately and noninvasively.
Noninvasive impedance cardiography (ICG) uses changes in thoracic electrical impedance to measure thoracic fluid content, changes in the duration of cardiac ejection, and the velocity of blood flow within the aorta (6,7). Impedance cardiography has been used to estimate cardiac output and cardiac filling pressure in patients with and without HF (6–11), but little is known about its ability to predict episodes of clinical decompensation.
Patients were included if they had chronic HF attributable to an ischemic or nonischemic cause (regardless of ejection fraction); had New York Heart Association (NYHA) functional class II, III, or IV symptoms; and were receiving appropriate medications for heart failure. All patients had a hospitalization, emergency department visit, or unscheduled clinic visit for worsening HF within three months, but had no meaningful change in symptoms or medications for heart failure within seven days.
Patients were excluded if they had any of the following: height <47 or >91 inches; weight <66 or >342 lbs; HF caused by myocarditis, cor pulmonale, congenital heart disease, constrictive pericarditis, or hypertrophic or restrictive cardiomyopathy; hemodynamically significant aortic regurgitation; acute coronary syndrome or coronary revascularization within 2 months; history of resuscitated sudden death or symptomatic or sustained ventricular fibrillation or ventricular tachycardia (unless within 24 h of a myocardial infarction or treated with a implantable cardioverter-defibrillator that had not fired within 2 months); second-degree Mobitz type II or third-degree heart block (unless treated with a pacemaker); left ventricular assist device or an activated minute ventilation pacemaker; any planned use of intravenous medications for HF (diuretics, vasodilators, or positive inotropic agents); serum creatinine >5 mg/dl; any liver function test >3 times the upper limit of normal or dialysis within 2 weeks; pulmonary disease that contributed to the limitation of exercise or required long-term corticosteroids; hypersensitivity to sensor gel or adhesive; skin lesions that prohibited sensor placement; or a disorder other than heart failure that might compromise survival within 6 months. The study was approved by the institutional review board at each participating institution. All patients provided written, informed consent.
After an initial assessment, patients were scheduled to undergo a clinical assessment and an ICG test as an outpatient every 2 weeks for 26 weeks. The clinical assessment included vital signs and weight, patient self-assessment of HF using a visual analog score, and NYHA functional class. The ICG measurements (Table 1)were performed in a blinded manner using a BioZ ICG Monitor (CardioDynamics, San Diego, California) and stored in the monitor’s electronic memory for subsequent analysis. Throughout the study, patients were managed according to the judgment of their usual physicians, who had no knowledge of the ICG test results.
All patients were followed up throughout the study for the occurrence of a HF event, which was prospectively defined in the original protocol as death from any cause or hospitalization or emergency department visit for worsening HF that required intensification of treatment. A committee of three cardiologists adjudicated all hospitalizations and emergency visits without knowledge of their clinical or ICG variables to determine whether events were related to worsening HF.
The prespecified primary hypothesis of the study was that changes in ICG variables combined into a composite score would predict the occurrence of a major clinical event. The variables were to be identified in an initial cohort of approximately 200 patients and then validated in a second cohort of similar size. This report describes the results in the first patient cohort; the second cohort has not yet been studied.
To test the primary study hypothesis, patients were categorized regarding whether they had or had not experienced one or more HF events during the study. Visits were categorized into those immediately preceding and those not immediately preceding a HF event. Differences in variables measured at visits preceding and not preceding an event were tested for significance by ttest. A 14-day window was used retrospectively to define “immediate” because the study visits were scheduled every 2 weeks. General estimation equation modeling was used to predict the occurrence of a HF event. Dependence in repeated data within a subject was modeled with an autoregressive covariance structure and a logistic link function to address the binary nature of the end point. A backward stepwise approach (which included all baseline, clinical, change in clinical variables from the prior visit, and ICG variables) was used to identify independent predictor variables, which were then validated with a forward stepwise procedure and ranked according to chi-square value.
To assess the ability of multiple ICG variables to simultaneously predict a HF event, the independently associated ICG variables were combined into a single composite score using weights derived from the multivariate model. The regression equation generated a value for each patient visit that was translated into the log of odds for a HF event and then converted to a numeric score between 0 and 10 with a higher score denoting higher risk.
Event rates for each group were calculated by dividing the number of visits that preceded an event by the total number of visits. Event rates were converted to relative risks (with 95% confidence intervals) by dividing the reference event rate by the comparator event rate. Differences in event rates and relative risks were assessed by two-tailed Fisher exact tests.
A total of 212 patients were enrolled at 21 sites in the U.S. The patients were 21 to 91 years of age (mean 59 years); 64% were men and 35% were black. The cause of HF was coronary artery disease in 46%; the mean ejection fraction was 0.27 ± 0.14 (range 0.10 to 0.76), and 17% had an ejection fraction ≥0.40. Functional status was class II in 32%, class III in 66%, and class IV in 3%. Treatment included diuretics (96%), angiotensin-converting enzyme inhibitors or angiotensin receptor blockers (84%), beta-blockers (72%), digoxin (62%), and aldosterone antagonists (41%).
The 212 patients completed 2,316 study visits (mean 10.9 ± 3.8 visits per patient). Twenty patients did not complete the study (15 because of a patient request, 4 because of a physician request, and 1 because of a serious adverse event unrelated to ICG measurement), but all available data in these 20 patients are included in the analysis.
During the study, there were 104 HF events in 59 patients (27.8%), which included 16 deaths, 78 hospitalizations for worsening HF (in 50 patients), and 10 emergency department visits for worsening HF (in 10 patients). Of the 104 events, 80 events (8 deaths, 64 hospitalizations, and 8 emergency visits) occurred within 14 days of a study visit, and of the 2,316 visits, 77 visits preceded an event within 14 days, equating to a 14-day heart failure event rate of 3.3%. For these visits, the mean time from the visit to the event was 5.9 ± 4.5 days.
There were no significant differences in baseline demographic, clinical, or ICG variables in patients with or without a HF event during the study. However, when the analysis was confined to visits within 14 days of an event, several variables had predictive value (Table 2).The patient visual analog scores and systolic blood pressures were lower and the heart rates and NYHA functional class were higher at visits preceding an event compared with visits not preceding an event. In addition, ICG assessment indicated more severe cardiocirculatory abnormalities (lower stroke volume, cardiac output, stroke work indices, velocity index [VI], and left ventricular ejection time [LVET], and higher vascular resistance, systolic time ratio index, and thoracic fluid content [TFC] index [TFCI]) at visits immediately preceding an event compared with those not preceding an event.
Multivariate regression analysis identified three clinical variables (visual analog score, NYHA functional class, and systolic blood pressure) and three ICG variables (TFCI, VI, and LVET) that were independently associated with the occurrence of a heart failure event within the next 14 days (Table 3).These three ICG variables were then combined into a single ICG composite score ranging from 0 to 10, with a higher score indicating higher risk for a heart failure event. The logit for the ICG composite score was calculated based on the following equation: (−α) + (β) TFCI + (−χ) VI + (−δ) LVET, where α, β, χ, and δ are constants reflecting the unique weighting of each variable. The ICG composite score was significantly higher in visits immediately preceding an event when compared with visits not immediately preceding an event (6.1 ± 2.0 vs. 4.4 ± 1.9, p < 0.0001). (Visits closer to an event were not disproportionately weighted in the ICG score because the ICG scores in the 51 visits within 7 days of an event were not different than the scores in the 26 visits within 8 to 14 days of an event [6.1 ± 1.6 vs. 6.0 ± 2.2].) When the multivariate analysis was repeated using the composite ICG score along with the three clinical predictors, the composite score was the most powerful predictor of an event within the next 14 days (Table 3).
Table 4shows the HF event rate and relative risks for each ICG score. Because most study visits did not precede an event, there was a nonlinear distribution of risk across score values, and thus the ICG scores were grouped into low-, intermediate-, and high-risk categories (corresponding to scores of 0 to 3, 4 to 6, and 7 to 10, respectively). Visits with a high-risk ICG score had an 8.4% event rate during the next 14 days; high scores were present in only 16.5% of visits, but they predicted 41.6% of the events. In contrast, visits with a low-risk ICG score had only a 1.0% event rate within 14 days; low scores were present in 38.6% of visits but were followed by only 11.7% of the events. Visits with a high-risk ICG score had a 2.5 times higher likelihood and those with a low-risk score had a 70% lower likelihood of a near-term HF event when compared with visits at intermediate risk (Table 4). Visits with a high-risk ICG score were 8.3 (95% CI 5.7 to 11.5) times more likely to be followed by an event within 14 days than those with a low-risk score. When NYHA functional class was added to the ICG composite score model, it did not improve the ability to discern high- or low-risk visits.
Two ICG parameters, stroke index (SI) and TFC, were used to create four unique cardiocirculatory quadrants in a manner similar to that originally proposed by Forrester et al. (3) to identify risk subsets after an acute myocardial infarction (Fig. 1).Visits with an SI >35 ml/m2and a TFC ≤35 /kOhm were followed by a risk of only 0.9% of a heart failure event as compared with a risk of 6.5% for visits with an SI ≤35 ml/m2and a TFC >35 /kOhm. Visits in the highest-risk quadrant had seven times (95% confidence interval 4.7 to 9.9) the risk of an event as did visits in the low-risk quadrant.
The flow of blood within the thorax can produce striking but transient changes in electrical impedance, which may be detected by ICG (6,7). By assessing the change in impedance of an alternating current applied across the chest, this technique can measure the baseline impedance as well as the magnitude and duration of the change in impedance during systole (Fig. 2).These measurements lead directly to the calculation of TFC, VI, and LVET, which can be used alone or in combination to estimate hemodynamic parameters (6–11). However, ICG may complement rather than replicate the information provided by right heart catheterization. Although estimates of cardiac output and stroke volume by ICG seem to be related to estimates of these variables by thermodilution (9,10), ICG may not reliably estimate right and left ventricular filling pressures (10,12) because cardiac filling pressures are influenced by changes in cardiac contractility and loading conditions, whereas TFC varies with radioisotopic estimates of plasma volume and thus may reflect the quantity of intravascular and extravascular fluid within the chest (13–16). Furthermore, the invasive nature of cardiac catheterization restricts its repetitive use, whereas ICG can be used to perform serial noninvasive assessments, thus allowing it to potentially detect the emergence of potentially adverse hemodynamic changes during periods of apparent clinical stability.
Despite these theoretical advantages, little is known about the ability of ICG to predict changes in cardiac function or clinical status or the occurrence of clinical events in HF. In one retrospective chart review of 13 patients (17), a relationship was observed between changes in impedance parameters and changes in ejection fraction over six months. In a second retrospective chart review of 64 patients (18), changes in impedance variables were related to changes in NYHA functional class, corridor walk distance, visual analog score, and quality of life over 6 months. In a third study of 98 men seen in an emergency department (19), a combination of brain natriuretic peptide and ICG variables was reported to identify patients at risk of death and hospitalization, but the independent contribution of ICG measurements was not assessed (20). Until the present report, no study had prospectively evaluated the ability of ICG to predict clinical events.
In the current prospective study, we found that ICG can identify patient visits likely to be followed in the near-term but not necessarily in the long-term by a clinical event. Over a period of six months, three variables directly assessed by ICG (TFCI, VI, LVET) were each powerful independent predictors of the occurrence of death or a worsening HF event during the 14 days after testing. Furthermore, when combined into a single composite score, these three variables provided powerful short-term prognostic information that was incremental to that available from a physician’s clinical evaluation. However, neither clinical nor ICG variables measured at the start of the study identified patients who deteriorated, suggesting that the predictive value of both clinical and ICG measurements may wane over time. This observation underscores the need for periodic reassessment and close follow-up for the optimal management of patients.
Impedance cardiography can generate a wide range of cardiocirculatory variables (Table 1), including those that are directly measured from the impedance waveforms, those that are derived from two or more ICG measurements, and those that are calculated from a combination of clinical and ICG assessments. Some investigators have suggested that measurements derived from combinations of ICG variables might increase the utility of the device (21,22), but this hypothesis has not been evaluated. It is therefore noteworthy that in the current study the three ICG variables that provided strong and independent prognostic information (TFCI, VI, and LVET) were not derived measurements, but were determined directly from the raw impedance waveforms. Simultaneous consideration of more than one of these independent variables seemed to improve their predictive value not when they were combined to generate a derived variable, but when their individual contributions were retained in a composite score or were used to formulate Forrester-like risk quadrants (3,4,22). These combined analyses were prognostically more powerful and discriminating than any variable considered alone.
Do ICG variables provide prognostic information that is incremental to that already available to the practicing physician? Clinicians assess the status of patients and gauge the need for therapeutic adjustments by taking a history of recent symptoms and performing a physical examination to assess fluid retention. This evaluation leads to an overall physician-based assessment (e.g., NYHA functional class) and an overall patient-based assessment (e.g., visual analog score), which are considered together with vital signs and body weight. The current study confirms the independent contribution made by each component of a physician’s typical clinical assessment (NYHA functional class, patient visual analog score, and systolic blood pressure) (23). However, in this study, the prognostic information provided by ICG seemed to complement that provided by the clinical assessment.
The findings of the current study should be interpreted cautiously. First, although our findings are consistent with our original hypothesis, the hypothesis for this pilot trial was not specific and could have been confirmed by a number of possible outcomes, including the one observed. To address this concern, the original protocol specified that the findings in the first cohort had to be validated in a second cohort; this validation has not yet occurred. Second, the risk models that we reported in the present study did not include the ICG data collected before 24 of the 104 (23%) events seen in the study because their ICG tests were delayed beyond our 14-day window. Widening of the window to 16 or 18 days so to include some of these excluded events would not have qualitatively altered our findings but would have reduced our estimates of the magnitude of incremental risk associated with a high-risk score. Third, we did not have complete proximate data for clinical events for periods longer than 14 days for patients enrolled in first and final 2 weeks of the study, thus our data sets were not sufficiently complete to evaluate the prognostic value of ICG beyond 14 days. Fourth, although specific values for certain ICG variables were associated with a higher risk, this association does not consider the relative frequency of visits with low- or high-risk scores. As in the case of other prognostic variables, the majority of patients with high-risk scores did not experience near-term events and the majority of events took place in patients with low- or intermediate-risk scores. Finally, although our findings suggest that the ICG test adds information to certain clinical variables used alone, in clinical practice physicians assess patients using a combination of clinical variables—some objective and measured (e.g., body weight and systolic blood pressure), some subjective and measured (e.g., NYHA functional class and patient global assessment), and some subjective and unmeasured (e.g., the overall impression and prior experience with the patient). The current study could not determine whether ICG adds value to a weighted composite of all clinical information typically available to clinicians during their routine management of patients.
As a result, our finding that specific clinical and ICG variables had predictive value should not be construed to imply that such variables can be used to titrate therapeutic agents or monitor their effectiveness. Although one study suggested that knowledge of ICG parameters can alter a physician’s treatment plans (24), and others have suggested that ICG parameters improve when patients receive effective drugs for heart failure (25,26), it is not clear whether ICG-directed modifications improve clinical outcomes beyond that expected if physicians responded appropriately to clinical signals in the absence of ICG data. In the Evaluation Study of Congestive Heart Failure and Pulmonary Artery Catheterization Effectiveness (ESCAPE) (27), the use of in-hospital invasive hemodynamic measurements to titrate treatment failed to alter long-term outcomes, but such assessments were not repeated during the period of follow-up. A large-scale clinical trial is now being planned to assess whether treatment guided by repeated assessment of a combination of clinical and ICG data reduces the risk of an adverse event when compared with clinical data alone.
In conclusion, our results suggest that when performed at regular intervals in patients with heart failure with a recent episode of clinical decompensation, noninvasive assessment by ICG can identify patients at near-term risk of recurrent decompensation. The clinical importance of these findings is currently being tested in a large-scale trial.
The authors thank Nandikishore Gurram, MD, Peggy Hardesty, MSN, and Lynn Rayl-Miller, RN, for their expert data collection.
Some authors have received consulting fees and honoraria from CardioDynamics, and all authors received research grants from CardioDynamics to support the study. Mihai Gheorgiade, MD, FACC, served as Guest Editor for this paper.
- Abbreviations and Acronyms
- heart failure
- impedance cardiography
- left ventricular ejection time
- New York Heart Association
- stroke index
- thoracic fluid content
- thoracic fluid content index
- velocity index
- Received August 3, 2005.
- Revision received December 14, 2005.
- Accepted December 16, 2005.
- American College of Cardiology Foundation
- McAlister F.A.,
- Stewart S.,
- Ferrua S.,
- McMurray J.J.
- Nohria A.,
- Tsang S.W.,
- Fang J.C.,
- Lewis E.F.,
- Jarchon J.A.,
- Mudge G.H.,
- Stevenson L.W.
- Spinale F.G.,
- Hendrick D.A.,
- Crawford F.A.,
- Carabello B.A.
- Greenberg B.H.,
- Hermann D.D.,
- Pranulis M.F.,
- et al.
- Albert N.M.,
- Hail M.D.,
- Li J.,
- Young J.B.
- Peacock W.F.,
- Albert N.M.,
- Kies P.,
- White R.D.,
- Emerman C.L.
- Snitzler R.N.,
- Paparelli V.
- Zewail A.,
- Broom C.,
- Delgado R.