Artificial Intelligence to Prevent Mobile Heart Failure Patients Decompensation in Real Time: Monitoring-Based Predictive Model

Table 1

Summary of decompensation detection studies.

Study	Data type	Dataset	Method	Results
Zhang et al. [7]	Weight	135 patients; 1964 days monitoring	RoT	Se = 58.3%, Sp = 54.1%
			MACD	Se = 20.4%, Sp = 89.4% (AUC = 0.55%)
Gyllensten et al. [6]	Weight	91 patients; 10 months	RoT	Se = 20%, Sp = 90%
			MACD	Se = 33%, Sp = 91%
			CUMSUM	Se = 13%, Sp = 91%
	Noninvasive transthoracic bioimpedance	91 patients; 10 months	RoT	Se = 13%, Sp = 91%
			MACD	Se = 13%, Sp = 91%
			CUMSUM	Se = 13%, Sp = 91%
Adamson et al. [10]	Blood pressure	274 patients	CUMSUM	Se = 83.1%, FA = 4.1/pt-y
Abraham et al. [5]	Intrathoracic impedance	156 patients; 537 ± 312 days	RoT	Se = 76.4%; FA = 1.9/pt-y
	Weight	156 patients; 537 ± 312 days	RoT	Se = 21%; FA = 4.3/pt-y
Ledwidge et al. [8]	Weight	87 patients; 23.9 ± 12 weeks	RoT	Se = 21%; Sp = 86%
			HeartPhone algorithm (based on MA)	Se = 82%; Sp = 68%
Gilliam et al. [11]	Multivariate	201 patients		Se = 41%; FA = 2/pt-y