Individualized Weight Monitoring Using the Heartphone
Individualized Weight Monitoring Using the Heartphone
The present study confirms previous reports that population weight significantly increases in advance of many instances of clinical deterioration of HF and that conventional guideline approaches to weight monitoring are inadequate. It shows that apparently stable population weights observable over long periods of follow-up using a remote monitoring system mask considerable and dynamic individual variability in underlying weight over time. Accordingly, the use of an individualized approach to weight monitoring with the HeartPhone algorithm can significantly improve the sensitivity and specificity of weight alerts for clinical deterioration. Options of this algorithm have been developed for achievement of improved sensitivity or specificity, and further prospective studies using the HeartPhone algorithm as part of remote monitoring systems for HF are warranted.
Systems of remote or home monitoring are increasingly under evaluation for the management of HF patients in an attempt to reduce the clinical and health economic burden associated with >2 million annual hospitalizations in the USA and EU. While there is now an established body of literature underpinning the use of intensive disease management programmes in HF, particularly in the early post-discharge period, the largest scale evaluation of nurse-led disease management in HF, the COACH study, was neutral for a variety of possible reasons including scaling up a complex intervention. Similarly, despite numerous smaller studies describing clinical benefit, the largest scale evaluation of remote patient monitoring in HF, the TELE-HF study, reported no impact on outcome, and, taken together, these data underline the challenge for clinical researchers in understanding how best to integrate remote patient monitoring with HF disease management. While the design and execution flaws of the TELE-HF study have been well described, including the selection of low-risk patients for the study, its operation over 5 and not 7 days per week, and the lack of patient adherence to the system, a further and more fundamental criticism is that the weight criterion used has been a poorly validated and poorly understood biomarker of deterioration in HF. Indeed, the foundation of many remote patient monitoring and disease management studies in HF is daily weight monitoring using similar guideline approaches, and the present study confirms previous data from our unit that these are inadequate and, as argued here and elsewhere, add little value over the play of chance.
This view is supported by the recent randomized, prospective, controlled WISH study which included 344 high-risk patients in an evaluation of telemonitoring focused on European Society of Cardiology (ESC) guideline (2 kg over 2–3 days) weight criteria alone and demonstrated no impact on outcome despite an increased clinic workload. The present study adds to these data by highlighting a number of potential reasons for the inadequacy of this and other approaches to weight monitoring in HF. First, several groups have shown, and the present study confirms, that weight gain according to European guidelines has low sensitivity for clinical deterioration among patients who experience HF deterioration. Secondly, previous research from our unit and elsewhere has argued that using a single threshold (e.g. 2 kg over 2–3 days) cannot be reasonably applied to a wide variety of underlying body weights observed in clinical practice (from <60 kg to >110 kg in the present study). Importantly, our previous work also failed to demonstrate a benefit of application of relative instead of absolute change thresholds (1% or 2%), indicating that an individualized solution to weight monitoring is probably complex. Thirdly, the recognition in a landmark publication in 2007 that weight change preceding HF decompensation can be very gradual, observable more than a week before the event, has pointed to a need for lower thresholds and/or longer observation periods before weight alerts are triggered (e.g. 1.36 kg over 1 day or 1.36 kg over baseline). Finally, many approaches to weight monitoring in HF assume that underlying weight is stable and that weight thresholds can be determined over the 'baseline' weight, which is manually adjustable over time. The present study confirms that in most patients, underlying weight variability over time is significantly greater than the thresholds being applied and that a method of dynamically tracking these changes is essential. The importance of dynamically tracking weight change is recognized in the important contribution of Zhang and colleagues which evaluated a number of so-called 'rule of thumb' weight monitoring approaches based on guidelines and compared them with a weight monitoring algorithm using moving average data from the TENS-HMS data set. However, their analysis of a range of dynamic approaches highlights the challenges, and they concluded that their algorithm was unable to improve weight monitoring sensitivity/specificity balance over 'rule of thumb' approaches.
The HeartPhone weight monitoring algorithm significantly improves the detection of clinical deterioration of HF over guideline methods by individualizing the weight monitoring in three ways. First, the algorithm automatically adjusts for changes in underlying weight, which is of value because, in addition to day to day variability in an individual patient's weight, we demonstrate here significant (>2 kg) variability in underlying stable weight occurring in a majority of patients. This is not surprising given the complex relationship between obesity and HF and the occurrence of periods of cardiac cachexia in advanced disease. Secondly, the algorithm automatically generates a weight alert based on the individual patient's weight pattern rather than using an absolute threshold derived from population values as in guideline or 'rule of thumb' approaches. Thirdly, it allows for tracking of cumulative, subtle weight changes over time. The HeartPhone algorithm can provide significantly improved sensitivity at specificity similar to the ESC guideline approach in one form (HeartPhone A) or to the guideline alert 1.36 kg over 1 day in another (HeartPhone B). Interestingly, the data obtained with these guideline or 'rule of thumb' alerts are in agreement with previous reports including those of Zhang et al. In all cases (guideline and HeartPhone), the positive predictive value of weight monitoring is low and the negative predictive value is high, suggesting that weight monitoring can help clinicians focus intense follow-up on high-risk patients in periods of instability. However, negative and positive predictive values are somewhat distorted by the low prevalence of events (1.3% of weekly periods monitored in our study) and, from a clinical perspective, high negative predictive values are less useful when the sensitivity of the monitoring approach is poor.
There is growing interest in other biomarkers of clinical deterioration in HF, including non-invasive methods such as electrocardiogram and blood pressure which have been components of previous studies, as well as newer methods such as sleep monitoring, spirometry, and transthoracic bioimpedance. These are non-invasive methods like the HeartPhone algorithm, which in turn is mathematical and can be deployed on other systems. The telemonitoring system used in our study is based on conventional mobile phone software linked to commercially available Bluetooth scales which is inexpensive to deploy and is acceptable to patients. On average, valid weight data were transmitted on >90% of monitored days, which is as good as the best reports from non-invasive telemonitoring systems and significantly better than reported in the TELE-HF study. Other more invasive telemonitoring systems with implantable devices such as CRT and ICDs are well established, and the HOME BNP project will provide valuable information on the value of daily plasma biomarker measurement in HF patients. However, it is noteworthy that the use of a specific invasive haemodynamic monitoring device for prediction of early deterioration of HF was initially rejected by an advisory committee of the Food and Drug Administration (FDA) on the basis of unproven efficacy. The present study suggests that the novel HeartPhone algorithm can improve the performance of non-invasive weight evaluation in remote monitoring of HF and that non-invasive telemonitoring in HF, which is more widely applicable, can be optimized.
There are a number of limitations to the present study. First, this prospective study was designed to evaluate remotely transmitted weight patterns in a controlled environment and mathematically to compare guideline weight monitoring with a novel individualized algorithm in the generation of true positives and false positives related to clinical deterioration. It must be emphasized that it requires further validation in a prospective trial where clinical staff intervene based on the alerts generated. Secondly, while the clinical staff involved in the telemonitoring work were blinded to the HeartPhone algorithm, weight data were available to the cardiologist during assessment of clinical deterioration, and this may have influenced the diagnosis of clinical deterioration. However, this would most probably have increased the sensitivity of the ESC guideline alert, and, at 21%, there is no evidence that this occurred. Thirdly, the alerts were generated using weight as a single marker, and ongoing work is evaluating the integration of the algorithm with patient-reported outcomes and other non-invasive and minimally invasive biomarkers. Finally, the data were evaluated in a pre-defined high-risk HF cohort and the results obtained here cannot be directly applied to lower risk HF patients.
Discussion
The present study confirms previous reports that population weight significantly increases in advance of many instances of clinical deterioration of HF and that conventional guideline approaches to weight monitoring are inadequate. It shows that apparently stable population weights observable over long periods of follow-up using a remote monitoring system mask considerable and dynamic individual variability in underlying weight over time. Accordingly, the use of an individualized approach to weight monitoring with the HeartPhone algorithm can significantly improve the sensitivity and specificity of weight alerts for clinical deterioration. Options of this algorithm have been developed for achievement of improved sensitivity or specificity, and further prospective studies using the HeartPhone algorithm as part of remote monitoring systems for HF are warranted.
Systems of remote or home monitoring are increasingly under evaluation for the management of HF patients in an attempt to reduce the clinical and health economic burden associated with >2 million annual hospitalizations in the USA and EU. While there is now an established body of literature underpinning the use of intensive disease management programmes in HF, particularly in the early post-discharge period, the largest scale evaluation of nurse-led disease management in HF, the COACH study, was neutral for a variety of possible reasons including scaling up a complex intervention. Similarly, despite numerous smaller studies describing clinical benefit, the largest scale evaluation of remote patient monitoring in HF, the TELE-HF study, reported no impact on outcome, and, taken together, these data underline the challenge for clinical researchers in understanding how best to integrate remote patient monitoring with HF disease management. While the design and execution flaws of the TELE-HF study have been well described, including the selection of low-risk patients for the study, its operation over 5 and not 7 days per week, and the lack of patient adherence to the system, a further and more fundamental criticism is that the weight criterion used has been a poorly validated and poorly understood biomarker of deterioration in HF. Indeed, the foundation of many remote patient monitoring and disease management studies in HF is daily weight monitoring using similar guideline approaches, and the present study confirms previous data from our unit that these are inadequate and, as argued here and elsewhere, add little value over the play of chance.
This view is supported by the recent randomized, prospective, controlled WISH study which included 344 high-risk patients in an evaluation of telemonitoring focused on European Society of Cardiology (ESC) guideline (2 kg over 2–3 days) weight criteria alone and demonstrated no impact on outcome despite an increased clinic workload. The present study adds to these data by highlighting a number of potential reasons for the inadequacy of this and other approaches to weight monitoring in HF. First, several groups have shown, and the present study confirms, that weight gain according to European guidelines has low sensitivity for clinical deterioration among patients who experience HF deterioration. Secondly, previous research from our unit and elsewhere has argued that using a single threshold (e.g. 2 kg over 2–3 days) cannot be reasonably applied to a wide variety of underlying body weights observed in clinical practice (from <60 kg to >110 kg in the present study). Importantly, our previous work also failed to demonstrate a benefit of application of relative instead of absolute change thresholds (1% or 2%), indicating that an individualized solution to weight monitoring is probably complex. Thirdly, the recognition in a landmark publication in 2007 that weight change preceding HF decompensation can be very gradual, observable more than a week before the event, has pointed to a need for lower thresholds and/or longer observation periods before weight alerts are triggered (e.g. 1.36 kg over 1 day or 1.36 kg over baseline). Finally, many approaches to weight monitoring in HF assume that underlying weight is stable and that weight thresholds can be determined over the 'baseline' weight, which is manually adjustable over time. The present study confirms that in most patients, underlying weight variability over time is significantly greater than the thresholds being applied and that a method of dynamically tracking these changes is essential. The importance of dynamically tracking weight change is recognized in the important contribution of Zhang and colleagues which evaluated a number of so-called 'rule of thumb' weight monitoring approaches based on guidelines and compared them with a weight monitoring algorithm using moving average data from the TENS-HMS data set. However, their analysis of a range of dynamic approaches highlights the challenges, and they concluded that their algorithm was unable to improve weight monitoring sensitivity/specificity balance over 'rule of thumb' approaches.
The HeartPhone weight monitoring algorithm significantly improves the detection of clinical deterioration of HF over guideline methods by individualizing the weight monitoring in three ways. First, the algorithm automatically adjusts for changes in underlying weight, which is of value because, in addition to day to day variability in an individual patient's weight, we demonstrate here significant (>2 kg) variability in underlying stable weight occurring in a majority of patients. This is not surprising given the complex relationship between obesity and HF and the occurrence of periods of cardiac cachexia in advanced disease. Secondly, the algorithm automatically generates a weight alert based on the individual patient's weight pattern rather than using an absolute threshold derived from population values as in guideline or 'rule of thumb' approaches. Thirdly, it allows for tracking of cumulative, subtle weight changes over time. The HeartPhone algorithm can provide significantly improved sensitivity at specificity similar to the ESC guideline approach in one form (HeartPhone A) or to the guideline alert 1.36 kg over 1 day in another (HeartPhone B). Interestingly, the data obtained with these guideline or 'rule of thumb' alerts are in agreement with previous reports including those of Zhang et al. In all cases (guideline and HeartPhone), the positive predictive value of weight monitoring is low and the negative predictive value is high, suggesting that weight monitoring can help clinicians focus intense follow-up on high-risk patients in periods of instability. However, negative and positive predictive values are somewhat distorted by the low prevalence of events (1.3% of weekly periods monitored in our study) and, from a clinical perspective, high negative predictive values are less useful when the sensitivity of the monitoring approach is poor.
There is growing interest in other biomarkers of clinical deterioration in HF, including non-invasive methods such as electrocardiogram and blood pressure which have been components of previous studies, as well as newer methods such as sleep monitoring, spirometry, and transthoracic bioimpedance. These are non-invasive methods like the HeartPhone algorithm, which in turn is mathematical and can be deployed on other systems. The telemonitoring system used in our study is based on conventional mobile phone software linked to commercially available Bluetooth scales which is inexpensive to deploy and is acceptable to patients. On average, valid weight data were transmitted on >90% of monitored days, which is as good as the best reports from non-invasive telemonitoring systems and significantly better than reported in the TELE-HF study. Other more invasive telemonitoring systems with implantable devices such as CRT and ICDs are well established, and the HOME BNP project will provide valuable information on the value of daily plasma biomarker measurement in HF patients. However, it is noteworthy that the use of a specific invasive haemodynamic monitoring device for prediction of early deterioration of HF was initially rejected by an advisory committee of the Food and Drug Administration (FDA) on the basis of unproven efficacy. The present study suggests that the novel HeartPhone algorithm can improve the performance of non-invasive weight evaluation in remote monitoring of HF and that non-invasive telemonitoring in HF, which is more widely applicable, can be optimized.
There are a number of limitations to the present study. First, this prospective study was designed to evaluate remotely transmitted weight patterns in a controlled environment and mathematically to compare guideline weight monitoring with a novel individualized algorithm in the generation of true positives and false positives related to clinical deterioration. It must be emphasized that it requires further validation in a prospective trial where clinical staff intervene based on the alerts generated. Secondly, while the clinical staff involved in the telemonitoring work were blinded to the HeartPhone algorithm, weight data were available to the cardiologist during assessment of clinical deterioration, and this may have influenced the diagnosis of clinical deterioration. However, this would most probably have increased the sensitivity of the ESC guideline alert, and, at 21%, there is no evidence that this occurred. Thirdly, the alerts were generated using weight as a single marker, and ongoing work is evaluating the integration of the algorithm with patient-reported outcomes and other non-invasive and minimally invasive biomarkers. Finally, the data were evaluated in a pre-defined high-risk HF cohort and the results obtained here cannot be directly applied to lower risk HF patients.
Source...