When ICDs Do Not Deliver: The Unseen Burden of Untreated Arrhythmias

ICDs remain a cornerstone of sudden cardiac death prevention.¹ However, a paradox remains in that some patients die suddenly with a functioning ICD with no known therapy being delivered. While often perceived as a rare event, recent evidence suggests that untreated ventricular arrhythmias (VA) occur more frequently than expected, and their detection is hindered by a lack of systematic post-mortem device interrogation.^2–6 For clarity, the spectrum of patient outcomes ranges from appropriate shocks for true VA, to inappropriate shocks and patient-perceived ‘phantom shocks’ in which no device therapy is delivered. Phantom shocks are different from the scope of this article given that we require device-stored electrograms (EGMs) or logs to adjudicate untreated events.⁷

Hauser et al. analysed the Food and Drug Administration’s online Manufacturer and User Facility Device Experience (MAUDE) database to identify multiple incidences of VA that went untreated in normally functioning ICD systems due to undersensing, high-rate detection thresholds, and features that restrict detection of VA or inappropriate programming, resulting in adverse events or patient death.³ Importantly, the number of these reports has been increasing from 2019 to 2023 and is growing faster than the rise in ICD implantation rate. An important question is whether this is due to the programming of higher therapy zones with longer tachycardia detections.

Modern ICD programming practice has been driven by an ambition to reduce inappropriate therapy without compromising the device’s ability to detect true VA: ventricular tachycardia (VT) and VF, thereby improving the specificity of the devices without compromising the sensitivity.^8,9 A Heart Rhythm Society expert consensus statement was released in 2015 and updated with additional company-specific guidelines in 2019.^8,9 For primary-prevention patients, it recommends fast detection rates, long detection durations, and routine programming of supraventricular tachycardia (SVT)–VT discriminators (Table 1 ).

Since this consensus statement, our service at Barts Heart Centre has adopted a modified version of the programming protocols (Table 1 ). This enabled shock therapy from 200 BPM in a VT zone, and from 230 BPM in a VF therapy zone (lower than the consensus statement suggestion of 250 BPM). From our own internal review in 2020, we found no undetected VA in 849 patients programmed with these settings, over a follow-up period of 20 ± 10 months. This reassured us that our practice was not compromising sensitivity and therefore we continued to replicate this programming across the 4,000+ ICD patients in our clinic. However, in the ambulatory setting we have identified six patients who died unexpectedly without receiving appropriate shocks via incidental remote downloads post-mortem.⁴

This review explores the potential scope of the problem of normally functioning ICD systems failing to deliver indicated therapy, critiques current programming paradigms, and highlights the need for cultural and procedural change in how we investigate deaths of ICD patients.

The Missed Opportunity of Post-mortem Device Interrogation

Systematic Interrogation Is Rare and Not Standard

Despite the proliferation of ICD theory, there is no consistent approach to post-mortem device interrogation globally. In the UK, ICDs are deactivated ahead of cremation, if requested by mortuary teams, and are rarely interrogated after death, unless there is a forensic or research reason to do so. Interrogations are often initiated by a medical examiner (ME), coroner request or clinical suspicion, which creates a selection bias.^10,11

What We Can Learn from the Literature

Two systematic reviews of ICD therapy highlight this gap.^12,13 Nikolaidou et al. reviewed more than 300 post-mortem ICD interrogations from multiple studies and found that device data changed the presumed cause of death in nearly a quarter of cases.^12,13 Hauser et al.’s review of the MAUDE database entries further underscores the missed VF events and inappropriate device responses, with an increasing incidence since 2019.³ However, the data available to these investigators were limited, and the exact mechanism of device failure could not be determined when patient- and device-level data are lacking.

A recent patient-level analysis of the MADIT-CRT trial showed that 25% of patients (6/24) who had sudden death and had their device interrogated post-mortem had ‘unsatisfactory device response’ to VA recorded on their device; the trial enrolled 1,820 patients overall.² The subset included four device failures to shock, and two cases of fatal pacing-induced tachycardia VA. Pacing-induced tachycardia is also an under-reported phenomenon with a variety of mechanisms that can cause VA.¹⁴

Further evidence from the Paratz et al. study, in which they reviewed 260 post-mortem device interrogations over 15 years, highlighted that post-mortem cardiac device interrogation of any type was successful in 98.5% of cases, raising the concern that not all devices can be interrogated post-mortem. Of the devices that could be interrogated, a potential device malfunction was identified in 7.7% of cases.¹¹ However, in a subset of ICD patients (n=56), a total of 23% (13 patients) died because of device faults. These included ‘major technical issues’ such as lead failure, or failure to treat VT/VF in normally functioning systems (‘VT/VF undersensed and inadequate therapy delivered’; ‘VT/VF below programmed therapy thresholds, and therapy withheld’). Reviewing these cases without EGM analysis and no company-specific information is challenging. Of the 13 patients, four received a minimum of 1 ICD shock, but they either ‘failed’ or failed to re-detect VT/VF. The description suggests that only two of these 13 patients had ‘successful’ shocks. The prevalence of ‘device failure’ in these studies could be underestimated compared with our own analysis of post-mortem device downloads, in which 29% (6 of 21 ICD patients) did not receive ICD shocks during VT/VF. Another important factor to note, as Paratz et al. highlighted, is that device clocks may be out of synchronisation with the local time zone, resulting in misclassification of fatal events as not fatal.¹¹

The literature highlights the disparity and lack of a systematic approach to post-mortem interrogations, and the prevalence of selection and reporting bias. Furthermore, the use of standardised description and labelling of events in ICD post-mortem analysis requires the standardisation of systematic assessment.

Undersensing and Misclassification: Classifying Events in Post-mortem ICD Analysis

Despite continual improvements in detection algorithms, ICDs are not foolproof. The most concerning failure mode is undersensing, particularly of fine or low-amplitude VF, when the device fails to enter a shock therapy zone. Undersensing is mostly a combination of signal amplitude size, EGM amplitude variability, programmed sensitivity, manufacturer-specific differences in band-pass filters and dynamically adjusted sensitivity, counter mechanism and rate zone detection. Several studies have underscored this issue, mostly after the fact.

Signal-related Undersensing

As early as 1994, Swerdlow et al. identified low R-wave amplitude and signal variability as a risk for VF undersensing.⁵ This led to assessments of the sinus R-wave amplitude to predict these occurrences. This appears logical: smaller R waves during sinus rhythm may reflect a scarred or diseased myocardium, which may also produce small amplitudes in VF. However, this did not appear to be an associated factor, but the authors continued to exercise caution with R wave measurements <3 mV.¹⁵ Importantly, that study was focused on dedicated bipolar sensing using Medtronic ICD devices and low-sensitivity nominals (0.3 mV). This is a sensitivity that does not reflect real-world programming across all manufacturers, given that band-pass filtering characteristics vary by manufacturer. Examples of signal-related undersensing are shown in Figure 1.

Functional Undersensing

The nature of signal amplitude variability during VF can cause functional undersensing, for example, when automatic sensitivity adjustments by the device result in it being too slow to sense the low-amplitude signals after larger high-amplitude signals, even if the low-amplitude signals exceed the minimum programmed sensitivity. This results in the device detection being slower than the detection criteria (Figure 2), which can lead to functional undersensing. The primary prevention consensus recommends that the lowest rate zone with shock therapy enabled would be programmed at a rate of 180–200 BPM to avoid this phenomenon, given that most VF detection studies have shown that the median cycle length of VF is commonly above 200 BPM. Importantly, this includes data only from Medtronic devices.¹⁶ However, it does not include VT below 200 BPM: that could be haemodynamically compromising and fatal. This is one of the main trade-offs when programming higher zones, and the expectation is that haemodynamically compromising VT that is below the 200 BPM threshold will degenerate quickly to VF, which is then detected with a device-defined rate above the 200 BPM threshold and defibrillated. This expectation could be why some physicians can program devices with rate criteria faster than those in the expert consensus statement, which should be documented clearly given that it can result in functional undersensing below the rate zone criteria (Figure 3). This is particularly important with regard to devices in which monitor zones are not used in the re-detection of arrhythmia.

Functional undersensing may also be a by-product of the use of concurrent device systems. An example is the use of subcutaneous and extravascular ICDs devices with leadless or concurrent pacemaker technology, which could result in pacemaker undersensing and pacing outputs being overdetected on ICDs.¹⁷ With the community’s enthusiasm for multi-manufacturer non-integrated ICD and pacemaker technology, physicians should be cautious regarding the absence of sensitivity evidence and limitations of case reports.

Algorithmic Undersensing (Misclassification of Ventricular Tachycardia/VF)

Misclassification of VA as supraventricular tachycardia (SVT) or artefact is a rarely documented phenomenon, but an important cause of undersensing. These phenomena have usually been noted only retrospectively and are often missed entirely without post-mortem ICD analysis. Algorithms for SVT discrimination have been routinely used in ICDs since the 1990s and although they are tested extensively during development (often with offline EGM modelling), most validation studies rely on surviving patients’ analysable episodes.^5,18,19 This introduces a critical bias, given that fatal events of device misclassification are almost never included or mentioned in the publications.

Sensitivity and specificity figures are often derived from these registries and yet contrasting reports on sensitivity are seen. One example of this is the assessment of the Medtronic Wavelet discriminator: Klein et al. demonstrated a sensitivity of 98.6% after review of 2,235 available episodes from single-chamber ICD patients.¹⁸ However, most rejected episodes were relatively slow sinus tachycardia and Wavelet was not applied above 187 BPM.

In contrast, Frontera et al. included atrial lead data to adjudicate true VT and found Wavelet sensitivity to be 90%.²⁰ Therefore, clinicians should be aware of the added error from these conclusions if post-mortem analysis is not performed. Figure 4 shows two examples of algorithmic misclassification: one that did not result in death, but delayed therapy, and one that resulted in patient death.

Did the Shock Fail? (Incessant or Recurrent Ventricular Arrhythmia)

Recurrent or incessant VA, particularly electrical storm, is a well-recognised phenomenon in patients with ICDs.¹ Although a single shock may successfully terminate a VF or VT episode, rapid re-initiation of arrhythmia is common and may lead to multiple shocks within minutes. Importantly, this is successful device therapy but can be misreported to suggest device failure to defibrillate, which, if it has occurred, may be caused by either patient or device issues. If the device does not cardiovert the rhythm this is a concern and should be highlighted. Figure 5 shows examples of successful and unsuccessful ICD shocks.

Failure of the Local Sensing EGM to Represent the Global Ventricular Rate

A unique challenge with ICD function arises when ventricular depolarisation occurs with different timing and amplitude between the right and left ventricle: so-called ‘dissimilar rhythms.’ This has been observed in up to 17% of patients during ICD therapy, particularly those with severe cardiomyopathy or structural heart disease.²¹ The phenomenon stems from regional delays in conduction and repolarisation between the ventricles, often due to scarring or desynchrony, and can result in one ventricular EGM on the sensed lead producing a lower rate than others.²² Thus, the conduction delay produces an entrance block or ‘myocardial undersensing’ of the VA (Figure 6).

Despite accurate device counting of EGM deflections, it is not representative of the rate of the entire ventricles. To alleviate this, one algorithm, Abbott VF Therapy Assurance (VFTA), has been produced to detect these events and increase the chance of therapy being delivered.²³ An example of this in action is shown in Figure 7. Algorithms such as this can increase the sensitivity of ICDs to VA but currently are exclusive to one manufacturer.

We have attempted to facilitate cardiac scientists’ and clinicians’ analysis of post-mortem ICD episodes, but there are also other factors to consider. The timing of such events becomes important when attempting to determine whether VA occurs as the cause of death or as part of the natural dying process.

Dying with Ventricular Arrhythmias: Cause or Consequence?

In a patient who dies with an ICD in situ, a clinical dilemma is whether documented VA at, or around, the time of death represents the cause of death or merely its epiphenomenon. VT/VF may occur as a terminal rhythm after a period of pulseless electrical activity, profound bradycardia or asystole. These agonal rhythms are physiologically unsurprising in end-of-life contexts such as acute MI, multi-organ failure or withdrawal of care.^10,12

To try and determine this, the timing of diagnostic parameters that are concurrently recorded by the device may be informative. Heart rate trend graphs, activity monitors and device automated tests may provide the timing to show that the VA occurred after the onset of cardiac arrest. Determination of this, however, requires device interrogation soon after (within 24 hours of) the event, to ensure the granularity of data needed to make these conclusions. This poses a challenge for device services.

Systematic Challenge of Post-mortem ICD Interrogation

Current Practice

Post-mortem interrogation of ICDs is not routinely performed in the UK or globally. While cremation protocols often include ICD shock therapy deactivation for safety reasons, this creates its own selection bias and is not always performed. There is no international mandate or widespread infrastructure for this practice. Mortuary teams are required to liaise with their local cardiology department to arrange a cardiac scientist to attend the mortuary. With many services facing staffing and logistical challenges during daily schedules, services are limited in what they can provide.

Barriers to Routine Interrogation

As of writing, post-mortem ICD analysis has no National Health Service healthcare resource group code to log activity to provide validation for this clinically relevant assessment of the cause of death. Operational limitations are not the only challenge for post-mortem analysis; technical limitations also exist. All ICDs have a limited memory storage capacity for EGMs and may overwrite the most critical episodes if multiple arrhythmic or artefactual events occur peri-mortem or post-mortem, leading to the loss of the true index event. Additionally, if the device is interrogated after explant and lead severance, the act of lead disconnection can introduce electrical noise, potentially obscuring the meaningful data.

Global Post-mortem Interrogation: Models and Potential Future Infrastructure

There are emerging pathways, but there are few that are reliably operational end-to-end. In England and Wales, the new statutory medical examiner review creates a logical triage point to flag deaths in people with cardiac devices; however, this is not routine given the limitations already discussed.²⁴ In Europe, pathology guidance recommends device removal or interrogation at autopsy for suspected sudden cardiac death, which is useful, but creates a selection bias not founded in evidence but practicality.²⁵ Two operational models show feasibility: a regional core-laboratory approach for blinded post-mortem interrogation, and long-running medicolegal workflows in Victoria, Australia.^11,26

To standardise outcome classification in these services an international core minimum dataset of programming and parameters should be reported to medical examiners to ensure comprehensive governance of device findings. For each case capture: device and lead identifiers; actual programming at death (zones/discriminators); clock-sync method; full episode logs with EGMs and markers; therapy counters and charge/abort events; battery/impedance/self-tests; and any safety mode or algorithm events. These would align the European, UK and the established core-laboratory workflows.^{10,11,24–26}

Conducting this review highlighted many limitations, including the sparsity of academic literature, and the unknown denominators and reporting/population bias in some published reports. Cross-vendor and different geography comparisons are imperfect. To try and improve this we have added our institutional observations, which are small and illustrative rather than prevalence estimates.

Considering these limitations, remote interrogation, performed as soon as death is identified, may represent the most accurate and feasible strategy for capturing the peri-mortem cause. The availability of this globally may be challenging, given that differing geographies have varying adoption of remote monitoring for all cardiac device patients, and not all device manufacturers have an upload capability from a generic monitor/machine. The Medtronic CareLink Express mechanism used for off-site interrogation could provide an example workflow for mortuary downloads.²⁷ The gradual transition to more Bluetooth, app-based remote monitoring may improve this, with future ‘on demand’ downloads that could be used as mortuary full memory pulls.

Conclusion

The widespread adoption of high-rate detection zones and prolonged detection times in ICD therapy has been supported by landmark trials and international guidelines. However, these strategies may inadvertently leave a subset of patients untreated. The drive for an increased specificity and reduction of inappropriate shocks may have compromised sensitivity to detect true VA. The literature suggests that this problem is under-recognised and a challenge to report; we have provided some guidance to the cardiac scientists and clinicians who review these ICD interrogations.

To draw a parallel: in aviation, the black box is sacred (Figure 8). Every crash is extensively analysed, not to assign blame, but to understand what went wrong and ensure that it does not happen again. ICDs are life-critical systems, but we fail to recover the ‘black box’ when patients die. Without reviewing these cases, our understanding remains skewed toward the survivors. We must look to adopt a new standard, not only as a medicolegal and technical tool, but also to provide families with answers. Doing so may drive improvements in programming strategies, algorithm development and, ultimately, patient survival.