Table 4

Tabular summarisation of search on the use of artificial intelligence in ICDs

Study	Number of participants	Study design	Algorithm used	Most accurate predictive factors	Potential limitations	Benefits
Wu et al56	382	Prospective registry analysis	Random Forest	HF hospitalisation CMR derived LA and LV volumes Larger total scar and grey zone extents Lower LA emptying fractions Serum IL-6	Observational study Long enrolment for cohort ICD programming parameters not prescriptive	Identification of predictive factors for appropriate ICD interventions in a cohort of patients suitable for primary prevention ICD insertion.
Van Hille et al57	62	Retrospective database analysis	Drools and ontology reasoning modules	With finer level of granularity DROOLS would be preferred	Small sample sizes Does not use specific instructions	Drools and ontology reasoning approaches are efficacious methods for the triage of AF alerts from ICD devices.
Shakibfar et al29	16 022	Retrospective database analysis	Logistic regression—model 1 Random forest—model 2	Total number of sustained episodes Shocks delivered Cycle length parameters	–	Prediction of electrical storm using machine learning models based on ICD remote monitoring summaries during episodes. Random forest superior to logistic regression (p<0.01).
Shakibfar et al 30	19 935	Retrospective cohort study	Random forest and logistic regression	Percentage of ventricular pacing during the day Activity of ICD during day Average ventricular HR during day Number of previously untreated tachycardias	Difficult to differentiate nsVT and VT US only (generalisability)	Use of large-scale random forest showed that daily summaries of ICD measurements in the absence of clinical information can predict short term risk of electrical storm.
Ross et al58	71 948	Retrospective registry analysis	Random forest and logistic regression	Family history of sudden death NYHA 4 Previous ICD Thoracic cardiac surgery and biventricular pacemaker insertion	Dual chamber ICDs only No information on leads Single rather than multiple imputation	Random forest can improve identification of mortality and adverse events by dual-chamber ICDs.

AF, atrial fibrillation; HF, heart failure; IL-6, interleukin-6; LA, left atrium; LV, left ventricle; nsVT, non-sustained ventricular tachycardia; NYHA-4, New York Heart Association Classification 4; VT, ventricular tachycardia.