Abstract
Background:
Ultrasound enhancing agents (UEAs) are routinely used to improve transthoracic echocardiography (TTE) image quality, yet anticipation of UEA need is a barrier to their use.
Methods:
Structured report data from 171,509 consecutive TTEs on 97,515 patients who underwent TTE at our institution between 1/26/2000–9/20/2018, were analyzed. Trends in UEA use and suboptimal image quality were examined. Amongst outpatients (TTEs = 92,291, N = 56,479), the dataset was randomly split into a 75% derivation and 25% validation sample. Logistic regression was used to model the composite of either UEA receipt or suboptimal image quality (≥2 non-visualized segments) using only variables available at the start of the TTE. Model performance was tested in the validation sample.
Results:
A total of 4,444 TTEs (2.6%) on 3,827 patients (3.9%) received UEAs and 28,468 TTEs (16.6%) on 21,994 patients (22.5%) were suboptimal. UEA use increased over the observation period. Among TTEs with suboptimal image quality, UEA use was lower in females (p < 0.0001). Among outpatients referred for TTE, older age, greater weight, and higher heart rate best predicted UEA use or suboptimal image quality. Model performance in the validation sample was excellent (c-statistic [95% CI]: 0.74 [0.73–0.75]; calibration slope [95% CI]: 1.11 [1.06–1.15]).
Conclusions:
In this large single center retrospective study, UEA use remained substantially below rates of suboptimal image quality, despite increases over time. Among outpatients, a simple prediction rule using 3 routinely collected variables available prior to TTE image acquisition predicted potential benefit from UEAs with high accuracy. If confirmed in other cohorts, this rule may be used to identify patients who may benefit from IV placement for UEA administration, prior to TTE image acquisition, thus potentially improving workflow efficiency.
Keywords: ultrasound enhancing agents, image quality, transthoracic echocardiography
INTRODUCTION
Ultrasound enhancing agents (UEAs) have become an indispensable clinical tool to improve echocardiographic image quality. Consisting of microbubbles with permeable shells containing a high molecular weight gas, UEAs are intravenously administered during echocardiographic imaging to improve left ventricular (LV) blood pool and endocardial border delineation. Three commercial UEAs are Food and Drug Administration (FDA)-approved for LV opacification and are recommended for use in individuals with suboptimal image quality by the 2018 American Society of Echocardiography (ASE) Guidelines.1 Several studies support the cost-effectiveness of UEAs and their efficacy in technically challenging critical care and emergency-room settings.2–5
Currently, UEAs are recommended on a “case-by-case” basis, as determined by unenhanced image quality. This need for assessment of unenhanced images may lead to time delays in obtaining and administering UEAs due to insufficient numbers of available trained personnel to place peripheral intravenous (IV) access for UEA administration or to obtain, prepare, and administer the agent.6 Indeed, one of the largest barriers to UEA utilization is the need for IV access, especially in outpatient clinical sites that may lack the resources or time to start an IV.6–7 Therefore, an algorithm used prior to image acquisition to identify outpatient transthoracic echocardiogram (TTE) referrals who may benefit from UEA imaging may allow for streamlined clinical workflow, increase the rate of UEA utilization, and decrease the frequency of suboptimal studies. Additionally, limited data exist on long-term trends in utilization of UEAs and suboptimal image quality.
Therefore, we sought to: 1) describe the temporal trends in inpatient and outpatient UEA utilization and suboptimal image quality on TTE at a large tertiary care medical center and 2) develop and internally validate an algorithm to identify outpatients with a high likelihood of UEA utilization or suboptimal image quality using variables only available prior to image acquisition.
METHODS
Study Population
Structured echocardiographic report data were extracted from 173,323 consecutive TTEs on 98,814 unique patients collected over 18 years (1/26/2000 – 9/20/2018) at the Beth Israel Deaconess Medical Center (BIDMC) inpatient and outpatient echocardiography laboratories. At the time of TTE interpretation by National Board of Echocardiography (NBE) Level III certified faculty, TTE report data were recorded and stored in a large electronic database. The database is maintained by Clinical Informatics and is routinely evaluated for completeness and accuracy. The database includes TTE metrics as well as demographic details, information on image quality, and UEA receipt.
In order to remove implausible or outlier data, the dataset was restricted to patients with a weight 40–182 kg (approximately 90–400 lbs), heart rate 40–200 bpm, age 18–100 years, body mass index (BMI) 18–55 kg/m2, and height 142–203 cm (approximately 56–80 inches) which represents the 0.5th and 99.5th percentile for each variable distribution. Additionally, as UEA use is suggested by echocardiography laboratory protocol for individuals with suspected or identified LV apical dysfunction, apical hypertrophy, or suspicion of left ventricular thrombus, TTEs with apical hypokinesis, akinesis, or dyskinesis (N = 957; TTEs = 1,364), apical hypertrophy (N = 92; TTEs = 140), or left ventricular thrombus (N = 250; TTEs = 310) were excluded. All echocardiographic images were acquired using General Electric E-95, Vivid 7, Vivid Q, Vivid 9, Vivid i, Vivid S70 (2005–2018; General Electric Healthcare, Waukesha, WI) or Hewlett-Packard Medical Products 5000 and 5500 echocardiographs (2000–2005; Hewlett-Packard Medical Products, Andover, MA). All echocardiographs were equipped with very low mechanical index (VLMI) imaging software presets which were used for all UEA studies. The study was approved by the BIDMC institutional review board with a waiver of informed consent.
Covariates and Outcomes
Covariates were obtained directly from the TTE report dataset and included all demographic and physiologic variables recorded at the time of the TTE—age, sex, weight, height, heart rate, blood pressure, and test location. Body mass index (BMI) and body surface-area (BSA) were calculated based on height and weight.8,9 Additional covariates collected included left atrial (LA) linear dimensions, right atrial linear dimensions, left ventricular (LV) ejection fraction, LV linear dimensions, right ventricle basal diastolic diameter, thoracic aortic size, aortic valve peak velocity, mitral valve peak E and A wave velocities, and peak estimated tricuspid regurgitation pressure gradient by the modified Bernoulli equation. LV ejection fraction was obtained via visual estimation, Simpson’s biplane method of disks, or 3-dimensional volumetric quantification when feasible. As indications are a free-text field in the dataset, natural language processing was used to query this field for any one of 23 indicator phrases for chest pain or coronary artery disease, 12 indicators for an abnormal electrocardiogram (ECG), and 32 indicator phrases of heart failure (Supplemental eTable 1).
The primary outcome was the composite of either UEA use or characterization of suboptimal image quality. UEAs included DEFINITY® (Lantheus Imaging, Billerica, MA), OPTISON® (GE Healthcare Inc., Princeton, NJ), and LUMASON® (Bracco Diagnostics Inc, Monroe Township, NJ). Per laboratory protocol (Supplemental eTable 2), the decision to give UEAs is made by the sonographer at the time of image acquisition based on their subjective assessment of image quality. If the resultant image quality is rendered adequate, as assessed by the attending TTE reader, image quality characterized as “adequate.” If the resultant image quality remains suboptimal, image quality is characterized as “suboptimal.” Suboptimal image quality is defined at the time of TTE interpretation by the attending TTE reader as ≥2 or more non-visualized ventricular segments per the ASE guidelines.1 As suboptimal image quality is considered separately from UEA receipt, a TTE could be both suboptimal and receive UEA. However, for the primary outcome of either UEA or suboptimal image quality, only the union of UEA receipt and suboptimal image quality was considered. All included outpatient TTEs were acquired at the BIDMC facility.
Statistical Analysis
Baseline demographic, physiologic, and TTE variables are presented as means and standard deviations (SDs) or medians and interquartile ranges (IQRs) for continuous variables and counts and percentages for categorical variables and compared with ANOVA or Chi-squared tests, respectively. Logistic regression was used to identify the predicted probabilities of the composite of either UEA receipt or suboptimal image quality by test year. Probabilities were plotted over the study period and the Mantel-Haenszel statistic used to evaluate significance of temporal trends. As we observed a large inflection in UEA use after 2008, corresponding to the lessening of national restrictions placed on UEA use, we separately evaluated trends pre- and post-2008. Logistic regression was used to evaluate the odds of UEA receipt or suboptimal image quality in each time period. Additionally, as we observed lower rates of UEA utilization among females, rate of change in UEA use and suboptimal image quality (as well as BMI) were stratified by sex and compared between sexes using z-tests for comparisons of rates and linear regression with a test year x sex interaction term for BMI. Continuous predictors were standardized, and the standardized univariable effect size for each variable on UEA receipt determined via logistic regression.
Subsequently, the outpatient TTE subset (N = 56,479; TTEs = 92,291), was randomly split into a 75% derivation and 25% validation cohort. To ensure random distribution of variables, all predictors were compared between the derivation and validation cohorts using univariate statistics. Among outpatient studies, a multivariable logistic regression model was built using all variables available to clinicians prior to image acquisition (e.g. age, sex, weight, height, BMI, BSA, heart rate, blood pressure, and location of test) and model performance tested in the validation cohort. For parsimony, variables were sequentially removed from the model if the concordance statistic (c-statistic) in the validation sample upon removal dropped by < 0.01. Receiver operating characteristic (ROC) curves for the parsimonious model were created in both derivation and validation cohorts. The validation cohort c-statistic was bootstrapped 1000 times to calculate a 95% confidence interval for the bootstrap sample.
To assess model calibration, predicted probabilities of UEA receipt or suboptimal image quality in the validation sample were outputted using the parsimonious model and binned into 20 categories, each with a 5% width. As the 99.5th percentile of the predicted probability distribution was 0.58, only predicted probabilities within the 0–75% range were examined. The mean predicted and observed probability in each bin was calculated. Subsequently, standard least squares regression was used to model the relationship between observed and predicted probabilities to create a calibration plot. The calibration slope was determined as the slope of this regression line.
As the mean number of TTEs per individual was low (1.76), no adjustment for clustering of TTEs within individuals was performed. All analyses were done in JMP v14.0 (SAS Institute, Cary, NC) using a two-tailed p < 0.05 to define significance.
Sensitivity Analysis
As a sensitivity analysis, we evaluated if a more complex machine learning model would improve upon predictive performance. Using only predictors available prior to image acquisition (age, sex, weight, height, BMI, BSA, heart rate, blood pressure, and test location), a random forest model was built in the derivation sample and tested in the validation sample using 100 trees with 10–2000 splits per tree, a minimum split size of 92 (0.1% of the total sample), a bootstrap sampling rate of 1, and 2–5 terms sampled per split. Using similar predictors, a boosted tree model was built with 50 layers, 1–3 splits per tree with a minimum split size of 9 (0.01% of the total sample), a maximum learning rate of 0.1, an overfit penalty of 0.0001, and stochastic boosting with a 50% row and column sampling rate. The loss function for both models was set to minimize root mean square error. Model c-statistics and misclassification rates were compared.
RESULTS
Overall Results
Of 171,509 TTEs on 97,515 unique patients included (Figure 1; mean [SD] age, 62.9 (17.1) years; 49.2% female), 30,733 TTEs (17.9%) on 23,387 individuals (24.0%) were coded as either suboptimal image quality or receiving UEAs, with 4,444 TTEs (2.6%) on 3,827 individuals (3.9%) receiving UEAs and 28,468 TTEs (16.6%) on 21,994 individuals (22.5%) considered suboptimal. Thus, 2,139 TTEs (1.2%) on 2,434 individuals (2.5%) had both UEA receipt and suboptimal image quality. Of the TTEs excluded for apical dysfunction (N = 957; TTEs = 1364), left ventricular thrombus (N = 92; TTEs = 310), or apical hypertrophy (N = 250; TTEs = 140), UEAs were used in 11.2%, 28.1%, and 17.1% respectively. If these TTEs were included, UEAs would be used in 2.7% of overall studies and 7.9% of suboptimal studies.
Figure 1:
Flow Diagram Illustrating Study Inclusion The diagram above illustrates numbers included and excluded from the sample. As indicated, outpatient TTEs were randomly divided into a 75% derivation and a 25% validation sample for model development.
Of the 4,444 UEA TTEs, DEFINITY® was used in 858 (19.3%), LUMASON® in 1,873 (42.1%), OPTISON® in 1,576 (35.5%) and an unspecified agent in 137 (3.1%). Supplemental eTable 3 details use of UEA brand by year. UEAs were used in 2179 (7.7%) suboptimal TTEs compared to 2265 (1.6%) adequate quality TTEs (p < 0.0001). Of the 10,878 TTEs from 2014–2018 considered suboptimal (32.8% of total number suboptimal), UEAs were used in 1539 (14.1%) compared to 1406 (3.0%) of adequate quality TTEs (p < 0.0001).
After applying exclusions, over 18 years, the mean (SD) annual rate of UEA use was 2.6 (2.0) %/year and increased from 0% in 2000 to 5.5% in 2018 (p for trend < 0.0001). The mean (SD) annual rate of suboptimal image quality was 16.6 (2.4) %/year and increased from 13.8% in 2000 to 19.9% in 2018 (p for trend < 0.0001). Thus, the mean rate of increase in suboptimal image quality (0.32 %/year) increased at a similar rate to UEA use (0.29 %/year) (Figure 2; p = 0.98). The mean (SD) annual rate of UEA use was higher among individuals with multiple TTEs vs. one TTE during the study period (2.6 %/year [2.0] vs. 2.5 %/year [2.0], p < 0.0001). UEA use increased by a mean of 0.16 %/year prior to 2008 and 0.49 %/year after (odds ratio [OR] for post-2008 vs. pre-2008: 2.91 [95% CI 2.65–3.20]; p < 0.0001). Suboptimal image quality increased by a mean of 0.07 %/year prior to 2008 and 0.3 %/year after (OR for post-2008 vs. pre-2008: 1.35 [95% CI 1.31–1.39]; p < 0.0001). Over the study period, the mean (SD) BMI increased from 27.5 (6.3) kg/m2 in 2000 to 28.9 (6.6) kg/m2 2018 (p for trend < 0.0001). Stratified by sex, the mean rate of increase in UEA use was 0.18 %/year in females and 0.39 %/year in males (p = 0.74). The mean rate of increase in suboptimal image quality was 0.22 %/year in females and 0.43 %/year in males (p = 0.97). The mean BMI increased by 1.5 kg/m2 in females and 1.3 kg/m2 in males (p for test year x sex interaction < 0.0001).
Figure 2:
Probability of UEA Receipt or Suboptimal Image Quality on Transthoracic Echocardiography by Study Year The blue shaded area represents the probability of UEA receipt and the red shaded area represents the probability of suboptimal optimal image quality. The mean increase in the proportion of UEA use was 0.29% per year (p for trend < 0.0001). The mean increase in the proportion of TTEs with suboptimal image quality was 0.32% per year (p for trend < 0.0001; p = 0.98 for difference in trend).
Univariable Results
TTE demographic, physiologic, and TTE variables, stratified by receipt of UEAs and presence of suboptimal image quality, are presented in Table 1. The mean age of individuals receiving UEA was higher than those not receiving UEAs (p < 0.0001). Among those receiving UEAs, the mean age of individuals with suboptimal image quality was higher than those with adequate image quality (p < 0.0001). Both suboptimal image quality (male vs. female, 18.6% vs. 14.6%, p < 0.0001) and UEA use were more common in males (male vs. female, 3.6% vs. 1.5%, p < 0.0001). However, among TTEs with suboptimal image quality, UEAs were more frequently used in males than females (male vs. female, 9.8% vs. 4.9%, p < 0.0001). A total of 61.7% of studies involving UEA receipt and 78.2% of studies with suboptimal image quality were inpatients and 40.9% and 53.7% were bedside studies respectively (all p < 0.0001) (Supplemental eTable 4). Of studies involving UEA receipt in the echocardiography lab, 55.6% were inpatients and 44.4% were outpatients (p < 0.0001). Use of UEAs in the ICUs (6.3% in CCU, 4.0% in MICU, 5.2% in SICU) was proportionally greater. Use of UEAs was more common in those with the indication of abnormal ECG or chest pain (p < 0.0001 for both). BMI and BSA were higher in those receiving UEAs (both p < 0.0001). BSA and inpatient status were the most important univariable predictors of UEA receipt (Supplemental eTable 5).
Table 1:
Distribution of Demographic and TTE Characteristics by Receipt of Ultrasound Enhancing Agents or Presence of Suboptimal Image Quality*
| Suboptimal Image Quality (TTEs = 28468) | Adequate Image Quality (TTEs = 143041) | ||||
|---|---|---|---|---|---|
| Variable | UEA Receipt (TTEs = 2197) | No UEA Receipt (TTEs = 26289) | UEA Receipt (TTEs = 2265) | No UEA Receipt (TTEs = 140776) | p-value |
| Age (years) – mean (SD) | 66.8 (12.5) | 67.3 (14.6) | 65.0 (13.8) | 61.9 (17.5) | < 0.0001 |
| Female – no. (%) | 1584 (72.1) | 14598 (55.5) | 1587 (70.1) | 69423 (49.3) | < 0.0001 |
| Bedside – no. (%) | 1194 (54.3) | 10445 (39.7) | 1191 (52.6) | 33571 (23.8) | < 0.0001 |
| Inpatient – no. (%) | 1693 (77.1) | 15884 (60.4) | 1782 (78.7) | 59859 (42.5) | < 0.0001 |
| Location – no. (%) | < 0.0001 | ||||
| CCU | 137 (6.2) | 1150 (4.4) | 208 (9.2) | 4005 (2.8) | |
| Echo Lab-Inpatient | 536 (24.4) | 5843 (22.2) | 639 (28.2) | 27660 (19.6) | |
| ED | 33 (1.5) | 395 (1.5) | 32 (1.4) | 1636 (1.2) | |
| MICU | 249 (11.3) | 2777 (10.6) | 172 (7.6) | 7236 (5.1) | |
| Other | 14 (0.6) | 145 (0.6) | 17 (0.8) | 517 (0.4) | |
| Study indications – no. (%) | |||||
| Heart Failure | 425 (19.3) | 2827 (10.8) | 368 (16.2) | 9918 (7.0) | < 0.0001 |
| Other | 1640 (74.6) | 22179 (84.4) | 1757 (77.6) | 122401 (86.9) | < 0.0001 |
| Systolic Blood | 126.6 (32.9) | 127.6 (33.4) | 125.8 (32.1) | 127.5 (29.9) | 0.03 |
| Pressure (mmHg) – mean (SD) | |||||
| Diastolic Blood | 70.2 (25.6) | 71.7 (35.6) | 73.0 (39.6) | 72.8 (31.6) | < 0.0001 |
| Pressure (mmHg) – mean (SD) | |||||
| Heart Rate (bpm) – mean (SD) | 78.3 (16.7) | 79.3 (18.1) | 76.9 (16.7) | 73.7 (15.9) | < 0.0001 |
| Height (cm) – mean (SD) | 172.8 (10.0) | 169.4 (10.4) | 171.8 (10.3) | 168.4 (10.3) | < 0.0001 |
| Weight (kg) – mean (SD) | 106.6 (25.9) | 89.6 (23.8) | 95.1 (24.7) | 78.3 (18.7) | < 0.0001 |
| BMI (kg/m2) – mean (SD) | 35.6 (7.9) | 31.1 (7.5) | 32.1 (7.5) | 27.5 (5.7) | < 0.0001 |
| BSA (m2) – mean (SD) | 2.25 (0.31) | 2.04 (0.30) | 2.12 (0.31) | 1.90 (0.26) | < 0.0001 |
| Left atrial size (cm) – mean (SD) | |||||
| Superoinferior | 5.7 (0.9) | 5.4 (0.9) | 5.6 (0.9) | 5.3 (0.9) | < 0.0001 |
| Right atrial length (cm) – mean (SD) | 5.3 (0.9) | 5.1 (0.9) | 5.3 (0.9) | 5.0 (0.9) | < 0.0001 |
| Left ventricular diameter (cm) – mean (SD) | |||||
| End-systolic | 3.4 (1.0) | 3.0 (0.8) | 3.7 (1.3) | 3.0 (0.7) | < 0.0001 |
| Left ventricular wall thickness (cm) – mean (SD) | |||||
| Inferolateral wall | 1.1 (0.2) | 1.1 (0.2) | 1.1 (0.2) | 1.1 (0.2) | < 0.0001 |
| LVEF (%) – mean (SD) | 56.0 (19.9) | 61.2 (16.3) | 47.3 (21.2) | 62.1 (15.8) | < 0.0001 |
| Right ventricular diastolic diameter (cm) – mean (SD) | 3.9 (0.8) | 3.6 (0.9) | 3.7 (0.9) | 3.5 (0.8) | < 0.0001 |
| Aortic size (cm) – mean (SD) | |||||
| Arch | 2.8 (0.4) | 2.8 (0.4) | 2.8 (0.4) | 2.7 (0.4) | < 0.0001 |
| AV peak velocity (m/s) – mean (SD) | 1.6 (0.7) | 1.7 (0.7) | 1.5 (0.7) | 1.6 (0.7) | < 0.0001 |
| MV peak E wave velocity (m/s) – mean (SD) | 0.9 (0.3) | 0.9 (0.3) | 0.9 (0.3) | 0.9 (0.3) | 0.051 |
| MV peak A wave velocity (m/s) – mean (SD) | 0.8 (0.4) | 0.9 (0.3) | 0.8 (0.4) | 0.8 (0.3) | < 0.0001 |
| Peak estimated TR gradient (mmHg) – mean (SD) | 30.6 (10.4) | 30.2 (11.8) | 30.2 (10.6) | 28.4 (11.2) | < 0.0001 |
p-value < 0.05 for all between-group comparisons. AV = aortic valve; BMI = body mass index; BSA = body surface area; CAD = coronary artery disease; CCU = coronary care unit; ECG = electrocardiogram. ED = emergency department; EP = electrophysiology; floor = inpatient floor; LVEF = left ventricular ejection fraction; MICU = medical intensive care unit; MV = mitral valve; OR = operating room; SD = standard deviation; SICU = surgical intensive care unit; TR = tricuspid regurgitant.
Multivariable Results
Among outpatients (TTEs = 92,291, N = 56,479), the dataset was randomly split into a 75% derivation cohort (69,218 TTEs on 45,724 individuals) and a 25% validation cohort (23,073 TTEs on 19,105 individuals) (Figure 1). No parameters were statistically different between derivation and validation cohorts. The proportion of TTEs in the derivation and validation cohort were similar before and after 2008 (p = 0.93). Sex, test location, blood pressure, BMI, BSA, and height were included in the initial prediction model but removed due to parsimony. Model parameters are listed in Supplemental eTable 6. The final model (Table 2) included weight (adjusted OR for a 10-kg increase in weight: 1.43, 95% CI 1.41–1.44; p < 0.0001), age (OR for 10-year increase in age: 1.38, 95% CI 1.36–1.40; p < 0.0001), and heart rate (OR for a 10-bpm increase in heart rate: 1.21, 95% CI 1.19–1.23; p < 0.0001). A formula to calculate predicted risk with this model is provided in Table 2.
Table 2:
Adjusted Odds Ratios for the Final Parsimonious Multivariable Logistic Regression Model to Predict the Composite of UEA Receipt or Suboptimal Image Quality in Outpatients Referred for TTE
| Variable | Adjusted OR (95% CI) | p-value |
|---|---|---|
| Weight* | 1.43 (1.41–1.44) | < 0.0001 |
| Age | 1.38 (1.36–1.40) | < 0.0001 |
| Heart rate | 1.21 (1.19–1.23) | < 0.0001 |
| Final model formula: log odds (of UEA receipt or suboptimal image quality) = −8.38 + 0.032 * age (in years) + 0.036 * weight (in kg) + 0.019 * heart rate (in bpm) | ||
Odds ratios represent the multivariable-adjusted relative increase in odds of the primary outcome (i.e. receipt of UEAs or presence of suboptimal image quality) for a 10-kg increase in weight, a 10-year increase in age, or a 10 bpm increase in heart rate). Model estimates are adjusted for all variables listed (e.g. weight, age, heart rate).
Model Prediction Performance
The parsimonious model had a c-statistic of 0.73 in the derivation cohort and 0.74 (95% CI 0.73–0.75) in the validation cohort. ROC curves for derivation and validation cohorts are presented in Supplemental eFigure 1 and Figure 3 respectively. The median (IQR) predicted probability was 0.10 (0.06–0.16) (Supplemental eFigure 2). The model calibration slope in the validation sample was 1.11 (95% CI 1.06–1.15) with an intercept of −0.014 (Figure 4).
Figure 3:
Receiver Operating Characteristic (ROC) Curve for the Composite Outcome of UEA Receipt or Presence of Suboptimal Image Quality in the Validation Cohort This represents the ROC curve for the primary outcome (i.e. receipt of UEAs or presence of suboptimal image quality), in the validation cohort. The area under the ROC curve was 0.74 (95% CI 0.73–0.75). The blue line represents the ROC curve for the probability of a positive response and the red line represents the ROC curve for the probability of a negative response.
Figure 4:
Calibration Plot of Observed vs. Predicted Probabilities of the Composite Outcome of UEA Receipt or Presence of Suboptimal Image Quality on Transthoracic Echocardiography Root mean square error = 0.0087; R-squared value = 0.99; The model calibration slope in the validation sample was 1.11 (95% CI 1.06–1.15) with an intercept of −0.014. The blue line represents the mean observed probability.
Sensitivity Analysis
The random forest model was optimized with 44 trees per forest and 5 terms per split with a misclassification rate of 12.1% and a root mean square error of 0.31. The c-statistics in the derivation and validation model were 0.80 and 0.74 respectively. The boosted tree model was optimized with 50 layers, 3 splits per tree, at a learning rate of 0.1 with a misclassification rate of 12.1% and a root mean square error of 0.31. The c-statistics in the derivation and validation model were 0.74 and 0.74 respectively. There was no significant difference in c-statistic between the logistic regression, the random forest, and the boosted tree models (p = 0.43; Supplemental eFigure 3; Supplemental eTable 7).
DISCUSSION
Despite compelling indications and guidelines for use, UEA utilization in TTE remains low. In the current retrospective study, at a single large academic medical center, UEAs were used in only 2.6% of TTEs and 7.7% of suboptimal TTE studies. Among those with suboptimal image quality, UEA use was less common in females. UEA use was more common in older individuals, inpatients, overweight individuals, and those with depressed LV systolic function. Weight and inpatient status were the most important predictors of the composite outcome of UEA use or suboptimal image quality. Among outpatients, a prediction model based on 3 parameters available prior to image acquisition (e.g. age, weight, and heart rate) predicted the composite outcome with very good discrimination and near-perfect calibration. If externally validated, this algorithm could be used to identify outpatients prior to TTE image acquisition that may benefit from UEAs and thus may streamline workflow to allow for IV placement prior to scanning to improve appropriate UEA utilization.
Use of UEAs in Stress Echocardiography
Our study extends prior work to develop a clinical prediction algorithm for UEA use in stress echocardiography. Bernier et al. used an 11-point clinical score using 8 variables (age, sex, BMI, > 2 risk factors for coronary artery disease [CAD], use of dobutamine stress echocardiography, abnormal ECG, history of smoking, and history of CAD) to accurately predict UEA use in individuals undergoing stress echocardiography.10 The current prediction rule differs by predicting the composite of UEA receipt or suboptimal image quality (i.e. potential benefit from UEAs) in TTE, not stress echocardiography, using only 3 variables—all available to clinicians prior to image acquisition. We similarly identify age and weight as significant risk factors for the composite outcome for TTE. However, as no testing or clinical knowledge are required for the current algorithm, it could be easily implemented with minimal a priori clinical knowledge.
Value of UEAs in Clinical Management
Current evidence supports the use and cost-effectiveness of UEAs for LV opacification for studies of suboptimal image quality. In 2009, Kurt et al. assessed the impact of UEAs on image quality and clinical management in 632 patients with technically difficult TTE studies, prospectively enrolled at a single center.4 The impact on clinical management was related to improved endocardial visualization and was greatest for ICUs patients followed by non-ICU inpatients and outpatients. Accounting for downstream cost savings from avoidance of additional diagnostic procedures, a cost benefit analysis suggested a $122 per patient cost savings when a UEA was used.4 A retrospective study by Dolan et al. (42,028 patients from 3 institutions) found UEA use was associated with improved diagnostic accuracy and no adverse harm.11 Based on the available data, UEAs are recommended to improve the accuracy and reproducibility of LV functional assessment and wall motion analysis, not satisfactorily assessed on unenhanced TTE.1
Trends in UEA Utilization
Despite potential benefits, UEA utilization was low in the current study, with only 7.7% of those with suboptimal TTEs receiving UEAs, despite similar rates of growth in suboptimal image quality and UEA use over time. However, after 2008, we observed a large inflection in the growth of use of UEA, corresponding to relaxing of the FDA’s black box safety warning for perflutren-containing UEAs that year. In October 2007, an FDA black box warning was placed on UEAs with contraindications for severe cardiopulmonary disease and mandating a 30-minute post-injection monitoring period.12 Subsequently, compelling safety data have emerged to support UEA use.13 In 2008, the FDA downgraded contraindications for individuals with severe cardiopulmonary disease to warnings and removed the 30-minute monitoring period from labeling, reflecting emerging safety data. Coincidentally, we observed a nearly threefold increase in UEA use in 2008. However, despite the existence of extensive data supporting the safety and efficacy of UEAs in the current era (2014–2018), UEAs use remained at only 14.1% in those with suboptimal quality TTEs. Reasons for the growth in suboptimal images on TTE are uncertain but may relate to increasing obesity among individuals receiving a TTE as the mean weight of individuals increased by 4.7 kg (approximately 10 lbs) over the 18-year period as well as changes in the perception of suboptimal image quality among TTE readers. Of note, although males had suboptimal image quality at a higher rate than females, amongst those with suboptimal image quality, UEA was less frequently used for females. Additionally, despite females having greater increases in preconditions for UEA use (e.g. BMI) over time, there were no sex differences observed in temporal changes in suboptimal image quality or uptake of UEAs. In totality, these results support observations that suggesting women may receive less optimal cardiovascular care and reasons for this disparity should be explored in future study.14
Barriers to UEA Adoption
Robust data currently exist to support UEA efficacy, safety, and cost-effectiveness in practice. Historical perceived barriers to utilization included costs, staffing, IV access, and the additional time needed for study planning, preparation, and acquisition. Subsequent research has found UEAs to be cost-effective by reducing downstream utilization4 and associated with a reduction in acquisition time when implemented properly through reduction of sonographer “struggle time.”6 Current barriers to utilization include insufficient awareness of the utility and benefit of UEAs, lack of training in administration of UEAs, absence of a formal training pathway for development of skills in IV placement, and mechanisms to identify which patients a priori may require IVs for UEA administration.
Reasons for low UEA utilization at our institution may include historical requirements for verbal approval for use from the attending TTE reader or cardiology fellow (policy no longer in place), storing of UEAs in the echocardiography laboratory (versus sites of most common use), lack of use of UEAs for off-hours emergent studies by fellows or on weekends, and insufficient training of nurses and other health professionals for administration of UEAs.15
An additional barrier to use is the insufficient number of trained personnel to place IVs. The ASE supports training of sonographers on IV placement and UEA administration.1 However, for many hospital-based echocardiography labs, this requires hospital approval and educational support to train personnel in venous anatomy, technique, and the risks and benefits of the procedure.1 Multiple studies (including the current one) demonstrate UEA use is higher in ICUs and inpatient floors. While this may, in part, be due to the greater prevalence of suboptimal image quality in these settings, it may also reflect the larger proportion of inpatients with pre-existing IVs, thus removing a major barrier to UEA use.4 In the current study, UEA was more common in inpatients, even when the TTE was performed in the echocardiography lab. Given both rates of suboptimal image quality and potential efficacy of UEAs are greater among inpatients, it is reasonable to consider UEA use for all inpatients, and to plan workflow accordingly to ensure appropriate access to UEAs for inpatient studies. Additionally, further research is necessary to understand factors related to underutilization of UEAs in circumstances where IVs are already in place, such as inpatients.
In outpatient settings, use of the current prediction rule to identify patients likely needing UEA a priori could help improve uptake of UEAs by streamlining workflow and pre-emptive IV placement in appropriate individuals. The current prediction rule could be implemented using a web-based or smartphone-app-based calculator, or integrated into the electronic ordering system, to determine an individual’s likelihood of potentially benefitting from UEAs. Thus, individuals at a high probability of benefitting from UEAs could have IVs placed prior to scanning, to improve workflow. Importantly, this algorithm is not meant to be proscriptive or indicate necessity of UEA use (e.g. all patients with sinus tachycardia do not benefit from UEAs). Rather, it is intended to be used to identify individuals at a sufficiently high post-test probability of either UEA receipt or suboptimal image quality that IV placement is reasonable to consider a priori to improve workflow. In the example used, as heart rate was higher in both those receiving UEA and having suboptimal image quality (p < 0.001 for both), heart rate may serve as a proxy for other unmeasured factors that influence image quality and UEA use.
Limitations
Though large and spanning nearly two decades, the current study has several limitations. First, as a retrospective analysis, causal inference cannot be inferred using the current methods. Second, the model prediction accuracy may differ when applied to a dissimilar population, that uses UEAs more frequently or has different rates of suboptimal image quality. External validation at a center with greater UEA use is planned. Additionally, temporal trends in UEA use and suboptimal image quality, observed locally, may not generalize to other clinical settings or regions. Third, the presence of a suboptimal study does not imply that UEA administration would have improved the study’s diagnostic yield, though suggests potential benefit. Fourth, as individuals with extremes of age, weight, height, and heart rate were excluded, the current results may not generalize to these populations. Fifth, it is possible that a different machine learning technique (e.g. artificial neural networks, support vector machines) may improve upon model performance. Sixth, as detailed information on timing and presence of IV placement was not available, the conclusion that IVs represented a major barrier to use is inferred. Finally, among studies where UEA was not given, it was not possible to discern the sonographer’s assessment of image quality, and thus agreement between physicians and sonographers on the decision to give UEA cannot be assessed.
CONCLUSIONS
In this large, single-center retrospective TTE study, UEA use was infrequent and less common in females with suboptimal image quality. Among outpatients referred for TTE, an algorithm with 3 routinely collected variables, available prior to TTE image acquisition – age, weight, and heart rate accurately predicted the composite of UEA receipt or suboptimal image quality. Predicting which outpatients TTE referrals may benefit from UEAs prior to image acquisition, could be used to select patients for IVs prior to imaging, thus improving workflow efficiency and diagnostic study yield. Further research is necessary to understand reasons for underutilization of UEAs in circumstances where IVs are already in place.
Supplementary Material
HIGHLIGHTS.
Ultrasound enhancing agents (UEAs) can improve suboptimal image quality on TTE.
While safe and effective, UEA utilization is limited by need for anticipation of use.
In this study, 3 variables predicted potential benefit from UEA in outpatients accurately.
This algorithm may help identify outpatients who benefit from IV placement for UEAs.
Sources of Funding:
Dr. Strom is funded by a grant from the NIH/NHLBI (1K23HL144907) outside of the submitted work.
Disclosures: Dr. Strom reports grant funding from Edwards Lifesciences, consulting from Philips Healthcare, and speaker fees from Northwest Imaging Forums, unrelated to the submitted work. All other authors report no disclosures.
ABBREVIATIONS:
- ASE
American Society of Echocardiography
- BIDMC
Beth Israel Deaconess Medical Center
- BMI
Body Mass Index
- BSA
Body surface area
- CAD
Coronary artery disease
- ECG
Electrocardiogram
- FDA
Food and Drug Administration
- IV
peripheral intravenous
- LA
Left atrial
- LV
Left ventricular
- NBE
National Board of Echocardiography
- TTE
Transthoracic echocardiography
- UEA
Ultrasound enhancing agent
- VLMI
Very low mechanical index
Footnotes
Publisher's Disclaimer: This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.
REFERENCES
- 1.Porter TR, Mulvagh SL, Abdelmoneium SS, Becher H, Belcik JT, Bierig M, et al. Clinical applications of ultrasonic enhancing agents in echocardiograpy: 2018 ASE guidelines update. J Am Soc Echocardiogr 2018;31:241–74. [DOI] [PubMed] [Google Scholar]
- 2.Exuzides A, Main ML, Colby C, Grayburn PA, Feinstein SB, Goldman JH. A retrospective comparison of mortality in critically ill hospitalized patients undergoing echocardiography with and without an ultrasound contrast agent. JACC Cardiovasc Imaging 2010;3:578–85. [DOI] [PubMed] [Google Scholar]
- 3.Main ML, Hibberd MG, Ryan A, Lowe TJ, Miller P, Bhat G. Acute mortality in critically ill patients undergoing echocardiography with or without an ultrasound contrast agent. JACC Cardiovasc Img 2014;7:408. [DOI] [PubMed] [Google Scholar]
- 4.Kurt M, Shaikh KA, Peterson L, Kurrelmeyer KM, Shah G, Nagueh SF, et al. Impact of contrast echocardiography on evaluation of ventricular function and clinical management in a large prospective cohort. J Am Coll Cardiol 2009;53:802–10. [DOI] [PubMed] [Google Scholar]
- 5.Tong KL, Kaul S, Wang XQ, Rinkevich D, Kalvaitis S, Belcik T, et al. Myocardial contrast echocardiography versus thrombolysis in myocardial infarction score in patients presenting to the emergency department with chest pain and a nondiagnostic electrocardiogram. J Am Coll Cardiol 2005;46:920–7. [DOI] [PubMed] [Google Scholar]
- 6.Castello R, Bella JN, Rovner A, Swan J, Smith J, Shaw L. Efficacy and time-efficiency of a “sonographer-driven” contrast echocardiography protocol in a high-volume echocardiography laboratory. Am Heart J 2003;145:535–41. [DOI] [PubMed] [Google Scholar]
- 7.Tang A, Chiew SK, Rashkovetsky R, Becher H, Choy JB. Feasibility of sonographer-administered echocontrast in a large-volume tertiary-care echocardiography laboratory. Can J Cardiol 2013;29:391–5. [DOI] [PubMed] [Google Scholar]
- 8.Nuttall FQ. Body Mass Index. Nutr Today 2015;50:117–28. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9.Mosteller RD. Simplified calculation of body-surface area. N Engl J Med 1987;317:1098. [DOI] [PubMed] [Google Scholar]
- 10.Bernier M, Abdelmoneim, Moir S, McCully RB, Pellikka PA, Mulvagh SL. Pretest score for predicting microbubble contrast agent use in stress echocardiography: a method to increase efficeicny in the echo laboratory. Cardiol Res Pract 2009; 2009:1–6. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11.Dolan MS, Gala SS, Dodla S, Abdelmoneim SS, Xie F, Cloutier D, et al. Safety and efficacy of commercially available ultrasound contrast agents for rest and stress echocardiography: a multicenter experience. J Am Coll Cardiol 2009;53:32–8. [DOI] [PubMed] [Google Scholar]
- 12.Grayburn PA. Product safety compromises patient safety (an unjustified black box warning on ultrasound contrast agents by the Food and Drug Administration). Am J Cardiol 2008;101:892–3. [DOI] [PubMed] [Google Scholar]
- 13.Khawaja OA, Shaikh KA, Al-Mallah MH. Meta-analysis of adverse cardiovascular events associated with echocardiographic contrast agents. Am J Cardiol 2010;106:742–7. [DOI] [PubMed] [Google Scholar]
- 14.Tanguturi VK, Hidrue MK, Picard MH, Atlas SJ, Weilburg JB, Ferris TG, et al. Variation in the echocardiographic surveillance of primary mitral regurgitation. Circ Cardiovasc Imaging. 2017. August; 10 (8): e006495. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15.Castello R, Bella JN, Rovner A, Swan J, Smith J, Shaw L. Efficacy and time-efficiency of a “sonographer driven” contrast echocardiography protocol in a high-volume echocardiography laboratory. Am Heart J. 2003;145 (3): 535–541. [DOI] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.