Administrative health data

Organisations administering healthcare require records of their administrative activity. These organisations – hospitals, health insurers and health maintenance organisations – store these records in databases that can then be reviewed and used to inform their ongoing activities. With the development of computerised databases, the availability of these data has surged in the past number of years. ^{Reference Gavrielov-Yusim and Friger7} The data can include information on records of health services, diagnostic/medical procedures, prescription information and diagnostic information. ^{Reference Gavrielov-Yusim and Friger7} Although collected for administrative purposes, this information has increasingly been used as a data source for research. ^{Reference Gavrielov-Yusim and Friger7}

An advantage of administrative health data for research is that it contains large numbers of individuals such as the entire population of a geographic region/country, which may include the health information for several million individuals. ^{Reference Gavrielov-Yusim and Friger7} Furthermore, these data – particularly when collected at the population level – provide a more representative sample in real-world settings compared with RCTs, which may include restrictive eligibility criteria. As a result, clinical trial results may not apply to the target treatment population. ^{Reference Austin8} In contrast, administrative health data allows for assessing study outcomes of an intervention directly within the target population.

Administrative health data offer important strengths, but two major limitations exist. First, because these data are collected for healthcare administration rather than research, investigators cannot choose which variables are captured and must therefore shape questions around available data. Second, particularly when considering interventions, the researcher does not allocate individuals to an intervention. As a result confounding – especially confounding by indication – can occur when assessing associations between an intervention and an outcome. ^{Reference Austin8} With ECT, many of the characteristics associated with receiving ECT (e.g. suicidality, psychosis and depression severity) are associated with negative outcomes. ^{Reference Kaster, Blumberger, Gomes, Sutradhar, Dasklakis and Wijeysundera9} Given these limitations, the methods used in the analysis of administrative health data must minimise biases to produce valid estimates of intervention effects.

Methodological developments

In the past two decades, there has been increased awareness of statistical methods and frameworks to mitigate confounding that occurs when estimating treatment effects from observational data. It is now recognised that administrative health data can inform clinical care and serve as the basis for regulatory approvals as highlighted by the US Food and Drug Administration’s Real World Evidence Program. ¹⁰ When observational studies are designed with care, for example by using a target trial framework, ^{Reference Robins11} it is possible to develop treatment effect estimates where bias is sufficiently minimised such that it can inform clinical decision-making.

Although there are various methods for minimising confounding in observational studies, the most common approach is the use of regression-based models. In these models, the investigator will specify an outcome variable (e.g. hospital admission), an exposure variable (e.g. ECT exposure) and then ‘adjust’ for confounders, which are characteristics associated with both the exposure and outcome, but not along the causal pathway. ^{Reference Austin8} An important limitation of this approach is that, distinct from RCTs, this approach combines the ‘design’ and ‘analysis’ phase such that researchers may be tempted to modify their model until suitable results are obtained. ^{Reference Austin8}

One approach that has recently gained popularity is the use of propensity scores, which are the probability an individual is exposed to an intervention based on their pre-exposure characteristics. ^{Reference Austin8} Subject to several important assumptions, when propensity scores are used to ‘balance’ patient characteristics in a study sample at baseline, any differences in outcomes can then be attributed to the intervention. ^{Reference Austin8} This approach separates the design and analysis phase similar to a clinical trial. For example, a researcher may iteratively modify the propensity score until adequate balance is achieved on all baseline observed characteristics (design) before performing the primary study analysis of the outcome. Although propensity scores are useful tools, one limitation is that they only account for confounding captured in variables available in the data-set such that unobserved confounding remains a possibility. ^{Reference Austin8} However, to the extent that unobserved covariates correlate with observed variables (e.g. medical comorbidities are correlated with age), then propensity scores and adjustment can mitigate the effect of these confounders even if unobserved. ^{Reference Austin8}

Despite these limitations, rigorous application of advanced methods to large administrative databases can yield new insights into the long-term risks and benefits of ECT, providing answers to clinically important questions that cannot be addressed in clinical trials.

Approach

We conducted a systematic search of Medline on 9 October 2025, using keywords for electroconvulsive therapy and observational/cohort studies from 2010 to present, using the Ovid interface (Supplementary Material). Studies before 2010 were excluded because they were less likely to have used modern statistical methods to address confounding and/or have access to comprehensive administrative health databases. We included any comparative study using population-level administrative health data to examine the following outcomes: (a) dementia, (b) major adverse cardiovascular/cerebrovascular events, (c) suicide deaths and (d) all-cause mortality.

We used a critical lens in reviewing these studies, but not a formal meta-analytic or risk of bias framework in examining them, as the intent of this work is to provide an up-to-date summary of recent studies focusing only on administrative health data to answer clinically relevant questions related to ECT outcomes. However, we did separately include any relevant meta-analyses previously reporting on these outcomes. We also cross-referenced the articles included in the meta-analyses with those from our systematic search, to minimise the risk of missing relevant articles.

ECT and dementia

Although imaging evidence does not support any indication of neurological damage associated with ECT, ^{Reference Ousdal, Brancati, Kessler, Erchinger, Dale and Abbott12} given the known cognitive side-effects that occur with ECT treatment (e.g. anterograde/retrograde amnesia or subjective cognitive complaints), ^{Reference McClintock, Choi, Deng, Appelbaum, Krystal and Lisanby13} a common concern for patients is the eventual development of dementia. Administrative health data is poorly suited to quantifying the nuanced cognitive side-effects related to ECT, but it is an ideal approach for identifying any potential association between ECT and subsequent development of dementia, years or even decades after treatment. Quantifying this association is particularly challenging given that the indication for ECT treatment itself (i.e. depression) is a known risk factor for developing dementia. ^{Reference Stafford, Chung, Sommerlad, Kirkbride and Howard14}

Our systematic search identified four studies examining the association between ECT and dementia and no systematic reviews (Fig. 1). Several of these studies found that before confounding was considered, there was a higher incidence of dementia among those receiving ECT versus not receiving ECT. ^{Reference Osler, Rozing, Christensen and Andersen15,Reference Hjerrild, Kahlert, Buchholtz, Rosenberg and Videbech16} However, once confounding was accounted for all studies found that there was no indication ECT increased the risk of developing dementia, except for the younger age subgroup in one study, which had few events and a small sample size. ^{Reference Kirov, Simmonds, Kaster and Escott-Price17} This was the case both in the short ^{Reference Chu, Chien, Chung, Chao, Chang and Kao18} and long term, ^{Reference Hjerrild, Kahlert, Buchholtz, Rosenberg and Videbech16,Reference Kirov, Simmonds, Kaster and Escott-Price17} and among those with schizophrenia. ^{Reference Chu, Chien, Chung, Chao, Chang and Kao18}

Fig. 1

Association of electroconvulsive therapy with dementia, major adverse cardiovascular/cerebrovascular events, suicide deaths and all-cause mortality. *Point estimate and confidence interval were inverted from original publication to use no electroconvulsive therapy exposure as reference group. **Acute cardiac events: acute myocardial infarction, cardiac arrest, paroxysmal tachycardia, atrial fibrillation/flutter and other cardiac arrhythmias. ***Reported odds ratio. ****90-day outcome. Note: Ahmadi et al, ^{Reference Ahmadi, Moss, Simon, Nemeroff and Atre-Vaidya26} Liang et al ^{Reference Liang, Chung, Ho, Tsai and Chien32} and Jørgensen et al ^{Reference Jørgensen, Rozing, Kellner and Osler31} are not included in this figure because of methodological limitations. PSM, propensity score matching; PSW, propensity score weighting.

ECT and major adverse cardiovascular/cerebrovascular events

Administering ECT results in a series of alternating parasympathetic and sympathetic events. The electrical stimulus causes a parasympathetic reaction through indirect electrical stimulation of the vagal nerve, ^{Reference Sartorius, Kellner and Karl19} which is followed by increased sympathetic activity when the seizure occurs, and finally, upon seizure termination, there is once again increased parasympathetic activity. ^{Reference Suzuki, Miyajima, Ohta, Yoshida, Okumura and Nakamura20} In addition to the physiological effects of ECT, there can also be haemodynamic consequences of the general anaesthetics used for ECT. ^{Reference Kaster and Blumberger3} As a result, patients and physicians are often concerned as to whether ECT significantly increases the risk of major adverse cardiovascular or cerebrovascular events.

Our systematic search identified two studies examining cerebrovascular events, two studies examining cardiovascular outcomes, two studies examining both (Fig. 1) and one relevant systematic review. Regarding the risk of ECT associated with cerebrovascular events, there did not appear to be any increased risk associated with ECT for both new and recurrent cerebrovascular events ^{Reference Hsieh, Tsai, Chou and Chou21,Reference Rozing, Jørgensen and Osler22} or in those with pre-existing physical health conditions. ^{Reference Osler, Rozing, Jorgensen and Jorgensen23} Although three studies suggested a reduced risk associated with ECT, ^{Reference Hsieh, Tsai, Chou and Chou21,Reference Rozing, Jørgensen and Osler22,Reference Nordenskjöld, Güney and Nordenskjöld24} a sensitivity analysis that used the Fine and Gray method ^{Reference Austin8} to account for competing risks found that there was no association (subdistribution hazard ratio (sdHR): 0.86, 95% CI 0.67–1.10) between ECT and risk of incident stroke, suggesting the observed reduction may be attributable to the competing risk of mortality. ^{Reference Rozing, Jørgensen and Osler22}

For the risk of ECT associated with major adverse cardiovascular events, the evidence is more conflicting (Fig. 1). The systematic review examining the association between ECT and major adverse cardiovascular events reported the incidence of cardiac events to be quite low (i.e. one in 50 patients); ^{Reference Duma, Maleczek, Panjikaran, Herkner, Karrison and Nagele25} however, this meta-analysis included studies without comparator groups such that these meta-analytic results cannot be used to draw inferences about the effect of ECT on these outcomes.

In terms of comparative studies, one of the first studies used data from the Department of Veterans Affairs in the USA and found a significant reduction in cardiovascular mortality for individuals with post-traumatic stress disorder and major depressive disorder receiving ECT compared with not receiving ECT. ^{Reference Ahmadi, Moss, Simon, Nemeroff and Atre-Vaidya26} However, this study had several methodological limitations that hinder the interpretability of these results and was excluded from our reporting (Supplementary Material). A more recent study found that, among individuals with mood disorders who had physical comorbidities, there was no significantly increased risk of major adverse cardiovascular event after ECT. ^{Reference Osler, Rozing, Jorgensen and Jorgensen23} Interestingly, those without physical comorbidities did have a significantly increased risk of major adverse cardiovascular event after ECT, although the authors indicate this was based on very few events and the increased risk did not persist beyond 30 days. ^{Reference Osler, Rozing, Jorgensen and Jorgensen23} In the only study conducted to date that focused on major adverse cardiovascular events and ECT and used propensity score matching and regression adjustment, the authors found that individuals with affective disorders who received ECT did not have an increased risk of myocardial infarction, but did have a reduced risk of cardiovascular death in the year following admission. No study used a Fine and Gray model for competing risk.

ECT and suicide deaths

One of the main indications for ECT is treatment of psychiatric conditions (most commonly severe depression) in the presence of acute or severe suicidal ideation. ^{Reference Kaster and Blumberger3} ECT has been associated with both substantial and rapid reduction of suicidal thoughts, which has often been assumed to translate to reduction in suicide death. ^{Reference Espinoza and Kellner1} However, one of the challenges with testing this assertion is that the severity of illness for those receiving ECT is often such that they are at greater risk of dying from suicide entirely independent from ECT. ^{Reference Kaster, Blumberger, Gomes, Sutradhar, Dasklakis and Wijeysundera9} Therefore, determining whether the symptom reduction observed with ECT is associated with reduced suicide deaths is one of the most critical questions related to ECT.

Our search identified four relevant systematic reviews, all of which were published in 2025. These studies reported a significant association between ECT and reduced risk of suicide death in the range of 30–50%. ^{Reference Rhee, Shim, Nasir, McIntyre, Kaster and Wilkinson27–Reference Odermatt, Sarlon, Schaefer, Ulrich, Ridder and Schneider29} One meta-analysis did report discrepant results, with no association identified between ECT and suicide deaths ^{Reference Naismith, Wilson, Costello, Davies, Pitman and Howard30} and another demonstrating heterogeneity regarding the association between ECT and suicide deaths based on follow-up time and geographic region. ^{Reference Rhee, Shim, Nasir, McIntyre, Kaster and Wilkinson27} Of note, the one meta-analysis that also examined the effect of antidepressant treatment (versus no treatment) found that there was no association between antidepressant treatment and the risk of suicide deaths. ^{Reference Chan, Solmi, Lo, Chan, Choo and Lai28}

In our systematic search, we identified eight population-level studies examining the association between suicide deaths (Fig. 1) that, consistent with the systematic reviews, typically identified a significant reduction in suicide-related mortality associated with ECT. Similar to other outcomes, we did not include the study by Ahmadi et al because of its methodological limitations. ^{Reference Ahmadi, Moss, Simon, Nemeroff and Atre-Vaidya26} Also of note is that we did not include the study by Jørgensen et al, as the design of this study likely results in bias of the effect of ECT on suicide deaths (Supplementary Material). ^{Reference Jørgensen, Rozing, Kellner and Osler31} We similarly excluded Liang et al because of study methodological limitations and the resulting risk of bias (Supplementary Material). ^{Reference Liang, Chung, Ho, Tsai and Chien32} Although most studies did identify a significantly reduced risk of suicide mortality associated with ECT, there was some variability in these findings in that longer follow-up periods (e.g. a year after ECT) were not always associated with reductions in suicide death, ^{Reference Rhee, Sint, Olfson, Gerhard, H Busch and Wilkinson33,Reference Peltzman, Shiner and Watts34} but this was not the case for all studies. ^{Reference Kaster, Blumberger, Gomes, Sutradhar, Wijeysundera and Vigod35} Some of this variability in findings may be attributable to not fully accounting for differences between ECT and non-ECT groups (despite propensity score matching) on important characteristics such as illness severity or pharmacotherapy, ^{Reference Peltzman, Shiner and Watts34} or differences in past number of hospital admission. ^{Reference Liang, Chung, Ho, Tsai and Chien32} Despite these limitations, no study – with the exception of Jørgensen et al ^{Reference Jørgensen, Rozing, Kellner and Osler31} – identified a significantly increased risk of suicide death once confounding was addressed.

ECT and all-cause mortality

The single most important metric to determine whether ECT is a ‘life-saving’ procedure as has been claimed ^{Reference Espinoza and Kellner1} is the rate of all-cause mortality, which is a single metric that encompasses both the risks and benefits of ECT.

Our search identified four systematic reviews that are the same reviews as those for suicide death, which often also consider all-cause mortality. Each one yielded a remarkably consistent 25–30% reduction in all-cause mortality associated with ECT despite differences in analytic techniques. ^{Reference Rhee, Shim, Nasir, McIntyre, Kaster and Wilkinson27–Reference Naismith, Wilson, Costello, Davies, Pitman and Howard30} This is a particularly important finding as all-cause mortality is not susceptible to competing risk or diagnostic misclassification, which can arise in the context of cause-specific mortality. ^{Reference Austin8}

Regarding specific studies assessing all-cause mortality, our search identified 11 population-level studies assessing the association between all-cause mortality and ECT. Similar to other outcomes, the studies by Ahmadi et al and Jørgensen et al were not included. ^{Reference Ahmadi, Moss, Simon, Nemeroff and Atre-Vaidya26,Reference Jørgensen, Rozing, Kellner and Osler31} Studies assessing the risk of all-cause mortality have consistently found a reduced risk of death associated with ECT and have come from regions including Taiwan, ^{Reference Liang, Chung, Tsai and Chien36} USA, ^{Reference Rhee, Sint, Olfson, Gerhard, H Busch and Wilkinson33,Reference Watts, Peltzman and Shiner37} Canada, ^{Reference Kaster, Blumberger, Gomes, Sutradhar, Wijeysundera and Vigod35} Denmark, ^{Reference Osler, Rozing, Jorgensen and Jorgensen23,Reference Jørgensen, Rozing, Kellner and Osler31} Sweden ^{Reference Hedna, Jonson, Sigström, Levinsson, Nordenskjöld and Waern38,Reference Rönnqvist, Nilsson and Nordenskjöld39} and the UK. ^{Reference Kirov, Simmonds, Kaster and Escott-Price17} Although it is possible these findings may be related to confounding (i.e. healthier individuals receive ECT), differences in healthcare systems or access to ECT, the multitude of investigators in different populations, with distinct and rigorous methods, finding similar results lends greater support to the position that ECT may be a life-saving procedure. The potential mechanism for this – rather than due to fewer medical comorbidities among those receiving ECT, which is typically accounted for – is that given its significant reduction in suicide death, combined with a benign medical profile, ^{Reference Kaster, Vigod, Gomes, Sutradhar, Wijeysundera and Blumberger40} ECT treatment may result in a net reduction in all-cause mortality for those with severe psychiatric illness.