Heart failure (HF) is a leading cause of morbidity and mortality in the United States, mentioned in 1 out of 9 death certificates in 2011, and designated as the underlying cause in 58,309 out of 284,388 deaths [
The CardioMEMS™ HF system includes an implantable pulmonary artery pressure (PAP) sensor that was approved for use by the Food and Drug Agency (FDA) in 2014 for New York Heart Association (NYHA) functional class III patients with a prior HFH within the preceding 12 months. The CHAMPION (CardioMEMS Heart Sensor Allows Monitoring of Pressure to Improve Outcomes in NYHA Class III Heart Failure Patients) trial, open-access registry, and several subgroup analyses subsequently confirmed a reduction in HFH in patients using this sensor [
However, effectiveness of the CardioMEMS™ sensor requires not only safe sensor implantation but also appropriate stewardship by both patients and health care providers. Patients must upload their pulmonary artery pressure (PAP) data on a regular basis, and health care providers must subsequently review these data and react accordingly, typically through titration of medical therapy (Figure
The CardioMEMS system. It requires multiple inputs to achieve its aim of reducing heart failure hospitalizations. The sensor must be implanted correctly and safely, patients must transmit pulmonary artery pressure (PAP) data regularly, and health care providers must review those data and formulate treatment plans. Patients can transmit their PAP data at home or in other nonhospitalized settings, and health care providers review data in an independent process at a separate time interval.
Patients initiate the transmission of their pressure data by placing a handheld “wand” near their chests. This wand communicates with the intra-arterial pressure sensor, extracting PAP data at that time point and uploading them to Merlin.net, a secure online database. In a separate process, health care providers review uploaded pressure data and react by changing medication dosing or changing the timing of clinic follow-ups. If pressure data are not available, such as when a patient forgets to transmit data, health care providers can send reminders to patients to do so. Providers document their review of pressure data and their plans going forward on the Merlin.net website. To date, no study has taken a systems-based approach to understand the impact of the patient-specific and health care provider-specific uses of the CardioMEMS™ HF system on HFH.
The purpose of this study was to apply a systems-based approach to examine whether the frequency of patient pressure transmissions and the frequency of health care provider reviews of those data were associated with risk of HFH.
This was a single-center, retrospective cohort study of patients who received the CardioMEMS™ pressure sensor at Keck Medical Center of the University of Southern California from October 2014 to August 2017. Patients were included if they had had the sensor for at least 12 months at the time of data collection and regardless of left ventricular ejection fraction. Patients with a left ventricular assist device were included. This study was approved by the Institutional Review Board at the University of Southern California by waiver of consent.
A systems’ context diagram was constructed to reflect the key inputs affecting HFH within the CardioMEMS system and served as a guide for data collection. Demographics, medical comorbidities, and the number of HFH days in the one year prior to CardioMEMS implant were obtained through the chart review. These data were handled as covariates in statistical modeling. The frequency of patient pressure transmissions and the frequency of health care provider reviews were the key independent variables. These data were collected from the Merlin.net website. Patient pressure transmission was defined as receipt and documentation of that pressure on the Merlin.net website. A health care provider review was defined as a note on the Merlin.net website that documented data reviews and a treatment plan, which could include reminding a patient to transmit data.
The frequency of patient pressure transmissions was characterized in two ways by (1) the mean number of days between pressure transmissions and (2) the number of times patients did not transmit pressures for more than 7 days during one year following CardioMEMS sensor implantation. The frequency of health care provider reviews was characterized by (1) the mean number of days between health care provider reviews and (2) the number of times there was no review of pressure data for more than 7 days. A secondary hypothesis was that the patient’s initial pressure response to the remote hemodynamic monitoring management strategy would predict future HFH. To explore this, a pressure-response variable was constructed and defined as the difference in mean PA diastolic pressure between the first and second two weeks following sensor implantation. This pressure-response variable was an additional independent variable explored in our statistical model.
The primary outcome was the total number of days a patient spent hospitalized for heart failure in the 1 year following CardioMEMS™ implant. HFH data were obtained from an individual patient chart review. HFH was defined as any hospitalization with reason for admission being directly related to an acute heart failure exacerbation. Each case was reviewed by two independent physicians (JST and AMW), who adjudicated the cause of hospitalization and determined whether or not the primary reason for hospitalization was due to HF.
Baseline clinical characteristics, patient compliance with transmission, health care provider review, and hospitalization statistics were calculated and presented as percent when classified categorically, mean/standard deviation when normally distributed, and quartiles when nonnormally distributed. Initial descriptive analysis of outcome data revealed a high number of patients with a zero count of days spent in HFH and wide distribution of nonzero count data, so data were fit to a zero-inflated negative binomial (ZINB) regression. The initial binomial component identified characteristics of patients not likely to have spent any days in HFHs, and the subsequent negative binomial component assessed factors predictive of the count of days spent in HFH, drawn from the subgroup of patients determined to be at risk for an HFH event based on the initial binomial component. The ZINB model was constructed in a stepwise elimination fashion. A
Between October 2014 and August 2017, 105 patients received a CardioMEMS™ sensor, and 78 patients met criteria for inclusion. Twenty-six patients were excluded because they had less than one year of pressure data, and one patient with congenital heart disease was excluded. Baseline patient characteristics are shown in Table
Baseline descriptive characteristics.
|
|
Age | 64.4 [14.8] |
Male | 52 (66.7%) |
White | 42 (53.8%) |
Left ventricular ejection fraction >40% | 20 (24.1%) |
Ischemic cardiomyopathy | 36 (46.2%) |
Implantable cardiac defibrillator | 51 (65.4%) |
Left ventricular assist device | 11 (14.1%) |
|
|
III | 76 (97.4%) |
IV | 2 (2.6%) |
|
|
|
|
Hypertension | 47 (60.2%) |
Coronary artery disease | 38 (48.7%) |
Diabetes mellitus | 38 (48.7%) |
Atrial fibrillation | 46 (59.0%) |
Chronic obstructive pulmonary disease | 8 (10.3%) |
Chronic kidney disease stage IV or V | 13 (16.7%) |
|
|
| |
Beta-blocker | 69 (88.5%) |
ACE inhibitor/ARB | 32 (41.0%) |
Aldosterone antagonist | 34 (43.6%) |
Loop diuretic | 67 (85.9%) |
Angiotensin receptor-neprilysin inhibitor | 6 (7.7%) |
Organic nitrate | 13 (16.7%) |
Hydralazine | 19 (24.4%) |
Inotrope (home infusion) | 8 (10.3%) |
|
|
| |
Systolic blood pressure | 114 [17.6] |
Diastolic blood pressure | 64 [11.5] |
Heart rate | 76 [11.9] |
Baseline pulmonary artery diastolic pressure | 24.7 [7.9] |
Note: categorical data are presented as number (percent), and continuous data are presented as mean [standard deviation].
Patient and health care provider’s sensor utilization practices.
|
|
Mean number of days between transmissions | 1.1/1.5/2.3 |
Count where time between transmissions >7 days | 0/2/4 |
|
|
|
|
Mean number of days between reviews | 4.4/6.3/10.3 |
Count where time between reviews >7 days | 6/14/20 |
Note: continuous data are presented as 25th/50th/75th quartiles, as data were not normally distributed; count data are presented as 25th/50th/75th quartiles.
The 78 patients in this study spent a total of 538 patient-days hospitalized for HF in the 1 year after CardioMEMS™ implant. Fifty-three patients did not spend any time in HFH, and the remaining 25 patients demonstrated a wide distribution of the number of days spent in HFH (alpha coefficient = 0.75, likelihood ratio
Heart failure hospitalization statistics.
|
|
Number of patients with a zero count of HFH | 53 (67.9%) |
Days spent in HFH out of the first year with CardioMEMS™ sensor (all patients) | 0/0/5 |
Days spent in HFH out of the first year with CardioMEMS™ sensor (nonzero count patients) | 7/11/30 |
|
|
|
|
Number of days spent in HFH 1 year before CardioMEMS™ | 0/7/12 |
Number of days spent hospitalized for any cause 1 year before CardioMEMS™ | 4/13/28 |
Note: count data are presented as 25th/50th/75th quartiles, categorical data are presented as number (percent).
The negative binomial count component of the ZINB (Figure
Forest plot illustration of the count model component of the zero-inflated negative binomial model. Regression coefficients and 95% confidence intervals are presented as incidence rate ratios. The interaction term is defined as the product of the average number of days between patient transmissions and the average number of days between health care provider reviews; it captures the synergistic impact of a patient’s stewardship and a health care provider’s stewardship of the CardioMEMS sensor. The constant term indicates the
The inflate component of the ZINB model (Figure
Forest plot illustration of the logit model component of the zero-inflated negative binomial model. Regression coefficients and 95% confidence intervals are presented as odds ratios. The 1-month pressure response is defined as the difference in mean pulmonary artery diastolic pressure between the first and second two weeks following sensor implantation (
Results of the ZINB model projections can be found in Figure
Heart failure hospitalization projections from the zero-inflated negative binomial model. Using these projections, this figure illustrates the exponential relationship between the number of HFH days a patient is projected to spend in the 1 year after sensor implant, based on the frequency of that patient’s transmissions of PAP data. Three separate projections are presented to demonstrate the impact of health care provider reviews. These projections were calculated at the 25th/50th/75th quartiles of health care provider reviews of PAP data, which correspond to health care provider reviews at a mean of 4.4/6.3/10.3 days, respectively. The gray shaded area illustrates the 95% confidence interval for the projection made at the 50th quartile of the health care provider review. HFH: heart failure hospitalization; PAP: pulmonary artery pressure; HCP: health care provider; CI: confidence interval.
The main finding of our study of patients receiving hemodynamic-guided therapy was that patient and provider utilization of the CardioMEMS™ HF system was associated with risk of future HFH. Patients who transmit PAP data more frequently appear to spend less time hospitalized for HF. Similarly, patients whose health care providers review PAP data more frequently appear to spend fewer days in HFH. Although these findings seem intuitive, this is the first study to implement a systems-based approach to quantify the impact that patients and health care providers can have on future HFH.
Model projections were presented to demonstrate the ease and feasibility of translating our statistical results into clinical outcomes. The projections directly translate patients and health care providers’ daily actions into days patients may spend hospitalized for HF, which is a much more tangible outcome compared to a regression coefficient. The projections presented in this study are not intended to provide estimates for a generalized population but rather to serve as a proof of concept.
Other nonmodifiable risk factors associated with an increase in the number of days spent in HFH were HF with reduced ejection fraction and a history of diabetes mellitus. There was a suggestion that LVEF >40% was associated with spending fewer days in HFH (IRR = 0.37, 95% CI: 0.13–1.01,
It should be noted that the interaction term between patient transmissions and health care provider reviews suggested a protective effect against HFH, although the term failed to meet statistical significance (IRR = 0.988,
Our results also suggest there is a group of patients who are at low risk of HFH after sensor implant, largely independent of patient and health care provider use of the sensor. The zero-inflated logit aspect of the ZINB is classically used to define a phenotype of subjects not at risk of the outcome of interest and filter those subjects out of the count model aspect of the ZINB, thereby increasing sensitivity of the overall model. While all patients in this study had been clinically identified to be at risk for HFHs, our model did identify a low-risk phenotype associated with the patients who did not have any HFHs. This low-risk phenotype patient did not have atrial fibrillation and exhibited a pressure response.
The characteristics associated with low risk for HFH identified in this study are consistent with results of previous studies. Pressure data analysis of the COMPASS-HF trial, which studied the Medtronic Chronicle implantable hemodynamic monitoring sensor, revealed prognostic implications of a low average pulmonary artery diastolic pressure (PADP). These analyses showed that patients with no HFH events had a significantly lower average PADP compared to patients with one or more HFH events [
Our study has several limitations: First, the overall study cohort is relatively small with all patients enrolled from a single academic institution. Second, variability of health care provider response to patient pressure trends was not captured nor implemented by our model; each provider had their own unique strategy for patient management. Future multicenter studies with a larger sample size will be necessary to more definitively quantify the impact that patients and health care providers have on future HFH, as well as a more refined patient phenotyping for patients that may or may not respond favorably to a remote hemodynamic monitoring strategy.
A systems-based analysis of patients managed with a remote hemodynamic monitoring strategy may be a useful approach for isolating specific factors that enhance efficacy of a remote hemodynamic monitoring management strategy.
The patient data used to support the findings of this study are restricted by the USC Institutional Review Board in order to protect patient privacy. Data are available from the corresponding author (
The preliminary findings of this study were presented as a moderated poster presentation at the 2017 Annual Scientific Meeting of the Heart Failure Society of America in Dallas, TX.
Dr. Shavelle is a consultant for Abbott Vascular, Inc. All other authors report no conflicts of interest.