Simulated case management of home telemonitoring to assess the impact of different alert algorithms on work-load and clinical decisions.

BACKGROUND: Home telemonitoring (HTM) of chronic heart failure (HF) promises to improve care by timely indications when a patient's condition is worsening. Simple rules of sudden weight change have been demonstrated to generate many alerts with poor sensitivity. Trend alert algorithms and bio-impedance (a more sensitive marker of fluid change), should produce fewer false alerts and reduce workload. However, comparisons between such approaches on the decisions made and the time spent reviewing alerts has not been studied. METHODS: Using HTM data from an observational trial of 91 HF patients, a simulated telemonitoring station was created and used to present virtual caseloads to clinicians experienced with HF HTM systems. Clinicians were randomised to either a simple (i.e. an increase of 2 kg in the past 3 days) or advanced alert method (either a moving average weight algorithm or bio-impedance cumulative sum algorithm). RESULTS: In total 16 clinicians reviewed the caseloads, 8 randomised to a simple alert method and 8 to the advanced alert methods. Total time to review the caseloads was lower in the advanced arms than the simple arm (80 ± 42 vs. 149 ± 82 min) but agreements on actions between clinicians were low (Fleiss kappa 0.33 and 0.31) and despite having high sensitivity many alerts in the bio-impedance arm were not considered to need further action. CONCLUSION: Advanced alerting algorithms with higher specificity are likely to reduce the time spent by clinicians and increase the percentage of time spent on changes rated as most meaningful. Work is needed to present bio-impedance alerts in a manner which is intuitive for clinicians.

A method to adapt thoracic impedance based on chest geometry and composition to assess congestion in heart failure patients.

Multi-frequency trans-thoracic bioimpedance (TTI) could be used to track fluid changes and congestion of the lungs, however, patient specific characteristics may impact the measurements. We investigated the effects of thoracic geometry and composition on measurements of TTI and developed an equation to calculate a personalized fluid index. Simulations of TTI measurements for varying levels of chest circumference, fat and muscle proportion were used to derive parameters for a model predicting expected values of TTI. This model was then adapted to measurements from a control group of 36 healthy volunteers to predict TTI and lung fluids (fluid index). Twenty heart failure (HF) patients treated for acute HF were then used to compare the changes in the personalized fluid index to symptoms of HF and predicted TTI to measurements at hospital discharge. All the derived body characteristics affected the TTI measurements in healthy volunteers and together the model predicted the measured TTI with 8.9% mean absolute error. In HF patients the estimated TTI correlated well with the discharged TTI (r=0.73,p <0.001) and the personalized fluid index followed changes in symptom levels during treatment. However, 37% (n=7) of the patients were discharged well below the model expected value. Accounting for chest geometry and composition might help in interpreting TTI measurements.

Early Indication of Decompensated Heart Failure in Patients on Home-Telemonitoring: A Comparison of Prediction Algorithms Based on Daily Weight and Noninvasive Transthoracic Bio-impedance.

BACKGROUND: Heart Failure (HF) is a common reason for hospitalization. Admissions might be prevented by early detection of and intervention for decompensation. Conventionally, changes in weight, a possible measure of fluid accumulation, have been used to detect deterioration. Transthoracic impedance may be a more sensitive and accurate measure of fluid accumulation. OBJECTIVE: In this study, we review previously proposed predictive algorithms using body weight and noninvasive transthoracic bio-impedance (NITTI) to predict HF decompensations. METHODS: We monitored 91 patients with chronic HF for an average of 10 months using a weight scale and a wearable bio-impedance vest. Three algorithms were tested using either simple rule-of-thumb differences (RoT), moving averages (MACD), or cumulative sums (CUSUM). RESULTS: Algorithms using NITTI in the 2 weeks preceding decompensation predicted events (P<.001); however, using weight alone did not. Cross-validation showed that NITTI improved sensitivity of all algorithms tested and that trend algorithms provided the best performance for either measurement (Weight-MACD: 33%, NITTI-CUSUM: 60%) in contrast to the simpler rules-of-thumb (Weight-RoT: 20%, NITTI-RoT: 33%) as proposed in HF guidelines. CONCLUSIONS: NITTI measurements decrease before decompensations, and combined with trend algorithms, improve the detection of HF decompensation over current guideline rules; however, many alerts are not associated with clinically overt decompensation.