Patient-level Characteristics and Inequitable Access to Inpatient Electroconvulsive Therapy for Depression: A Population-based Cross-sectional Study: Caractéristiques au niveau du patient et accès inéquitable à la thérapie électroconvulsive pour patients hospitalisés

Abstract

Objective:

A variety of patient characteristics drive the use of electroconvulsive therapy (ECT) in depression. However, the extent to which each characteristic influences the receipt of ECT, and whether they are appropriate, is unknown. The aim of this study is to identify patient-level characteristics associated with receiving inpatient ECT for depression.

Method:

We identified all psychiatric inpatients with a major depressive episode admitted to hospital ≥3 days in Ontario, Canada (2009 to 2017). The association between patient-level characteristics at admission and receipt of inpatient ECT was determined using logistic regression, where a generalized estimating equations approach accounted for repeat admissions.

Results:

The cohort included 53,174 inpatients experiencing 75,429 admissions, with 6,899 admissions involving ECT (9.2%). Among demographic factors, age was most associated with ECT—younger adults had reduced (OR = 0.30, 95%CI, 0.24 to 0.37; 18 to 25 years) while older adults had increased (OR = 3.08, 95%CI, 2.41 to 3.93; 85+ years) odds compared to middle-aged adults (46 to 55 years). The likelihood of ECT was greater for individuals who were married/partnered, had postsecondary education, and resided in the highest neighborhood income quintile. Among clinical factors, illness polarity was most associated with receiving ECT—bipolar depression had reduced odds of receiving ECT (OR = 0.62, 95%CI, 0.57 to 0.69) The likelihood of receiving ECT was greater in psychotic depression, more depressive symptoms, and incapable to consent to treatment and was reduced with comorbid substance use disorders and several medical comorbidities.

Conclusions:

Nearly 1 in 10 admissions for depression in Ontario, Canada, involve ECT. Many clinical factors associated with receiving inpatient ECT were concordant with clinical guidelines; however, nonclinical factors associated with its use warrant investigation of their impact on equitable access to ECT.

Introduction

Depression is a leading cause of illness and disability burden worldwide.¹ A particularly challenging clinical scenario for affected individuals is treatment-resistant depression (TRD), which is defined based on nonresponsiveness to initial pharmacotherapy or psychotherapy,² affecting more than 50% of individuals with depression.³ Of the individuals with TRD,³ only 13.0% and 13.7% achieve remission after 3 and 4 trials of pharmacotherapy.³ Among treatment options for these individuals, electroconvulsive therapy (ECT) offers great benefits in that it achieves remission in 60% of TRD patients,⁴ with some studies finding remission rates as high as 95% for individuals with depression with psychotic features.⁵ Furthermore, ECT is extremely effective at rapidly reducing suicidality.⁶ Consequently, practice guidelines recommend ECT as a first-line treatment for TRD and in situations where rapid treatment is critical—such as acute suicidality or psychotic depression.⁷

Despite its efficacy in TRD, ECT remains underutilized; indeed, in some regions, its use is declining,^8,9 suggesting clinical practice guidelines do not always reflect real-world application of important treatments. Furthermore, previous work has pointed out disparities in access to ECT treatment, as evidenced by nonclinical factors such as being unmarried, less educated, black race, and lack of proximity to ECT treatment facilities being associated with lower probability of receiving treatment.^10,11 Understanding the real-world implementation of ECT, and in particular characteristics of patients who do and do not receive it, may help identify populations to specifically target and improve access to ECT. Previous studies in this area have had shortcomings that limited their ability to achieve this goal. For example, many ECT studies have been too small to identify specific characteristics associated with real-world implementation.¹² Other studies have used large health care data sets but were limited to specific populations such as the U.S. Veterans Affairs health system^10,12 or to patients with private insurance.¹³ Two prior studies used population-level data in Ontario, Canada,^14,15 to characterize the use of ECT across the province but did not consider detailed clinical characteristics such as depression severity, cognitive functioning, medical/psychiatric comorbidities, and medications—all key characteristics influencing the decision to offer ECT.⁷

The goal of this study was to identify patient-level characteristics associated with receipt of ECT in a large, clinically representative cohort of patients with depression who might be candidates for ECT.

Methods

Setting and Data Sources

We conducted a population-based cross-sectional study in Ontario, Canada, using linked administrative health care databases available through ICES, an independent nonprofit research organization evaluating health care services across Ontario. These data sets were linked using unique encoded identifiers and analyzed at ICES. Data sources used for this study included (1) Registered Persons Database (RPDB) that contains date of birth, sex, and postal code for all Ontario Residents; (2) Canadian Institute for Health Information Discharge Abstract Database (CIHI-DAD) that contains clinical information about medical hospital admissions and discharges coded using standard diagnoses (ICD-10-CA); (3) CIHI Ontario Mental Health Reporting System (OMHRS) that contains mental health hospital data from all facilities with designated adult inpatient mental health beds; (4) CIHI National Ambulatory Care Reporting System (NACRS) that contains clinical information about emergency department visits coded using standard diagnoses (ICD-10-CA); (5) Ontario Health Insurance Plan (OHIP) Claims Database that contains information related to physician services submitted for reimbursement; and (6) Ontario Drug Benefit (ODB) that includes details on all publicly funded prescription medications dispensed through retail pharmacies to eligible individuals including those >65, residents of long-term care, or recipients of home-care services, social assistance (Ontario Works or Ontario Disability Support Program), or medication payment assistance (Trillium Drug Program). The use of data in this project was authorized under section 45 of Ontario’s Personal Health Information Protection Act, which does not require review by a Research Ethics Board.

Study Population

Using the OMHRS data set, we identified all adult (≥18 years) residents of Ontario admitted for at least 3 days to a designated psychiatric inpatient unit at an Ontario hospital between April 1, 2007, to March 31, 2017. We excluded individuals with short-stay admissions (<3 days) because these patients receive abbreviated assessments that would not contain detailed clinical information required to assess their suitability for ECT. Any patient with a discharge diagnosis of a major depressive episode—including both unipolar major depressive disorder and bipolar affective disorder—was included in this study (diagnostic codes are in Supplemental Table S1). Patients with non-affective psychotic disorders (e.g., schizophrenia, schizoaffective disorder) were included, provided they had a discharge diagnosis of depression.

Receipt of ECT Treatment

Treatment with ECT during each eligible admission was determined using OHIP physician billing codes for inpatient ECT procedures (G478). OHIP procedure codes are considered to be reliable and valid sources of information with respect to physician billing for procedures and visits.¹⁶ Receipt of ECT was treated as a binary variable throughout the analyses.

Covariates

We included covariates assessed at each admission to the psychiatric inpatient unit including sociodemographic, clinical, psychiatric/functional symptoms, and service utilization characteristics. The RPDB data set was used to ascertain age, neighborhood income quintile, and rural residence status. From the OMHRS data set, we included marital status, education level, employment status, reason for admission, type of admission (voluntary vs. involuntary), substance use, psychiatric comorbidities, capacity to consent to psychiatric treatment, and body mass index. The OMHRS data set also includes standardized psychiatric and functional scales adapted from the Resident Assessment Instrument Minimum Dataset.¹⁷ Scales included for the current study were the Activities of Daily Living (ADL) Hierarchy Scale (assesses daily functioning of ADLs), Aggressive Behavior Scale (measure of aggressive behavior), Social Withdrawal Scale (assesses occurrence of anhedonia, lack of motivation and reduction in socialization), Cognitive Performance Scale (a clinician rated objective measure that provides a functional measure of a person’s cognitive status and correlates with the Mini-Mental Status Examination¹⁸), Depressive Severity Index (measure of depression severity), Mania Scale (measure of elevated mood disturbances), Positive Symptoms Scale (measure of severity of psychosis), Risk of Harm to Others Scale (measure of the level of risk of harm to others), Self Care Index (measure of the risk that a person will be unable to care for themselves due to presence of psychiatric symptoms), and Severity of Self-harm Score (measure of severity of person’s risk of self-harm). From the ODB data set, we extracted information on outpatient medications dispensed 120 days prior to admission for eligible individuals. The remaining databases (OHIP, NACRS, CIHI-DAD) provided information on medical comorbidities, as well as past-year number of psychiatric and/or medical emergency department visits, psychiatric and/or medical inpatient admissions, outpatient visits to psychiatrist, and outpatient visits to family physician for medical and/or psychiatric reasons.

Statistical Analyses

We first described characteristics of individual patients who did and did not receive ECT during the study period. We compared groups using standardized differences, which describes between-group differences in units of standard deviation.¹⁹ This measure was selected, rather than a P value, because it is not influenced by sample size and therefore is preferred in large cohorts where small, insignificant differences would be statistically significant using a P value approach.¹⁹ An absolute SD of >0.10 is considered a meaningful difference.¹⁹

Next, to identify patient-level characteristics associated with receiving inpatient ECT, we used hospital admission data—which allows for multiple admissions per patient—and applied a multivariable logistic regression model with receipt of ECT (yes/no) at each admission being the outcome. A generalized estimating equation (GEE) with an exchangeable correlation structure was incorporated to account for correlation among outcomes arising from multiple admissions from the same patient. Continuous covariates were discretized into clinically meaningful categories (e.g., deciles for age) to account for nonlinear relationships. All categorical covariates were assessed for sparse data. The main multivariable model included all covariates except prescription medication use prior to admission as this was available only for individuals eligible for public drug benefits. We assessed model discrimination using the c-statistic and conducted a complete case analysis because missing data were uncommon (<1%). We also assessed for multicollinearity in our model to ensure tolerance was <0.2 for all covariates included in the model.

Using identical methods as the main multivariable model, two additional models, restricted to individuals for whom all medication prescription data were available and stratified by age, were also generated that included prescription medication as a covariate. Models for individuals aged ≥65 (all Ontarians ≥65 receive public drug coverage) and individuals aged <65 (mainly individuals who receive social assistance or disability support) were generated separately given differences in reasons for public drug coverage differ by age-group.

All analyses were conducted in SAS 9.4 (SAS Institute Inc.) with referent categories for multilevel categorical variables being the largest group and for binary variables being the absence of the condition. We did not adjust for multiple comparisons as this was a descriptive study and our focus was on effect size rather than statistical significance.²⁰ This study was reported in accordance with STROBE guidelines.²¹

Role of Funding Source

The study sponsor had no role in study design or decision to publish this study. The corresponding author had full access to all data and had final responsibility for the decision to submit for publication.

Results

Descriptive Characteristics

From 75,429 unique hospital admissions for depression, 6,899 involved treatment with ECT (9.2%). This included 53,174 unique patients, of whom 5,043 (9.5%) received ECT during at least 1 admission. In univariate analyses, patients who received ECT were older, more likely to be female, married or partnered, and living in a high-income neighborhood than individuals who did not receive ECT (Table 1). Those who received ECT, more frequently had unipolar (vs. bipolar or unspecified) depression, were more frequently admitted due to inability to care for self, were older at their first hospital admission, and were more frequently incapable of consenting to some form of psychiatric treatment (ECT or otherwise) than those who never received ECT. They more frequently had depression with psychotic features and cognitive disorders but less frequently had a substance use disorder. We attempted to include catatonia as a comorbidity; however, <1% of admissions had this diagnostic code listed. They had higher rates of most medical comorbidities and were more likely to have publicly funded drug coverage. ECT-treated patients also had higher (worse) scores on ADL, Self-care, and Depressive Symptom Scales. They had more outpatient visits to psychiatrists and psychiatric hospitalizations within 12 months preceding admission. Among those with information on prescription medication use prior to admission, ECT-treated patients had more psychotropic medication prescriptions (e.g., antidepressant, antipsychotic, benzodiazepine, and lithium) with the exception of anticonvulsants (Table 2).

Table 1.

Comparing Distributions of Characteristics among Individuals Who Received versus Who Did Not Receive Electroconvulsive Therapy (ECT).

Baseline Characteristics: n (%)	No ECT Treatment (n = 48,131)	ECT Treatment (n = 5,043)	Absolute Standardized Difference
Sociodemographic
Mean age ± SD (years)	44 ± 17	56 ± 17	0.75
Female sex	27,632 (57.4%)	3,241 (64.3%)	0.14
Marital status
Never married	20,532 (42.7%)	1,198 (23.8%)	0.41
Widowed/separated/divorced	10,193 (21.2%)	1,327 (26.3%)	0.12
Married/partnered	17,401 (36.2%)	2,518 (49.9%)	0.28
Education level
Less than high school	9,243 (19.2%)	912 (18.1%)	0.03
High school	13,132 (27.3%)	1,220 (24.2%)	0.07
Any postsecondary	20,686 (43.0%)	2,335 (46.3%)	0.07
Unknown	5,064 (10.5%)	575 (11.4%)	0.03
Employment status
Employed	14,604 (30.3%)	954 (18.9%)	0.27
Unemployed, seeking employment	4,355 (9.0%)	221 (4.4%)	0.19
Unemployed, not seeking employment	18,874 (39.2%)	2,210 (43.8%)	0.09
Other	9,232 (19.2%)	1,559 (30.9%)	0.27
Unknown	1,065 (2.2%)	99 (2.0%)	0.02
Urban dwelling	43,211 (89.8%)	4,554 (90.3%)	0.02
Neighborhood income quintile
1 (lowest)	13,243 (27.5%)	1,118 (22.2%)	0.12
2	10,360 (21.5%)	1,009 (20.0%)	0.04
3	9,044 (18.8%)	967 (19.2%)	0.01
4	8,340 (17.3%)	977 (19.4%)	0.05
5 (highest)	6,915 (14.4%)	952 (18.9%)	0.12
Clinical characteristics
Depression type
Unipolar	30,556 (63.5%)	4,109 (81.5%)	0.41
Bipolar	9,645 (20.0%)	733 (14.5%)	0.15
Unspecified	7,930 (16.5%)	201 (4.0%)	0.42
Reason for admission
Problem with addiction/dependency	8,660 (18.0%)	354 (7.0%)	0.34
Forensic/justice/other	2,753 (5.7%)	228 (4.5%)	0.05
Specific psychiatric symptoms	36,751 (76.4%)	4,185 (83.0%)	0.17
Threat to others	5,158 (10.7%)	227 (4.5%)	0.24
Threat to self	30,454 (63.3%)	2,562 (50.8%)	0.25
Inability to care for self due to illness	14,748 (30.6%)	2,033 (40.3%)	0.2
Involuntary admission	7,860 (16.3%)	601 (11.9%)	0.13
Incapable to consent to treatment	1,573 (3.3%)	413 (8.2%)	0.21
Age at first psychiatric hospitalization
0 to 14 years	1,463 (3.0%)	84 (1.7%)	0.09
15 to 24 years	12,722 (26.4%)	669 (13.3%)	0.33
25 to 44 years	19,158 (39.8%)	1,827 (36.2%)	0.07
45 to 64 years	11,564 (24.0%)	1,528 (30.3%)	0.14
65+ years	3,217 (6.7%)	934 (18.5%)	0.36
Eligible for public drug coverage	20,391 (42.4%)	3,005 (59.6%)	0.35
Mean body mass index ± SD (kg/m2)	27.1 ± 7.4	27.4 ± 7.3	0.04
Psychiatric comorbidities
Anxiety disorder	8,226 (17.1%)	1,065 (21.1%)	0.1
Depression with psychotic features	5,945 (12.4%)	1,105 (21.9%)	0.26
Nonaffective psychotic disorder	1,443 (3.0%)	78 (1.5%)	0.1
Personality disorder	7,282 (15.1%)	797 (15.8%)	0.02
Substance use disorder	8,346 (17.3%)	348 (6.9%)	0.32
PTSD or trauma-related disorder	1,384 (2.9%)	148 (2.9%)	0
Eating disorder	472 (1.0%)	62 (1.2%)	0.02
Cognitive disorder	1,074 (2.2%)	383 (7.6%)	0.25
Medical comorbidities
Respiratory
Asthma	10,525 (21.9%)	974 (19.3%)	0.06
Chronic obstructive pulmonary disease	5,545 (11.5%)	781 (15.5%)	0.12
Cardiac
Congestive heart failure	1,090 (2.3%)	219 (4.3%)	0.12
Myocardial infarction	688 (1.4%)	100 (2.0%)	0.04
Ischemic heart disease	5,000 (10.4%)	869 (17.2%)	0.2
Atrial fibrillation	578 (1.2%)	133 (2.6%)	0.1
Cardiac arrhythmia	3,412 (7.1%)	609 (12.1%)	0.17
Gastrointestinal
Inflammatory bowel disease	244 (0.5%)	24 (0.5%)	0
Gastrointestinal bleed	883 (1.8%)	114 (2.3%)	0.03
Neurologic
Parkinson’s disease	240 (0.5%)	94 (1.9%)	0.13
Epilepsy	2,291 (4.8%)	259 (5.1%)	0.02
Stroke	1,855 (3.9%)	282 (5.6%)	0.08
Endocrine, hematologic, and autoimmune
Hypothyroidism	3,484 (7.2%)	613 (12.2%)	0.17
Diabetes	6,520 (13.5%)	1,046 (20.7%)	0.19
Thromboembolic disease	814 (1.7%)	117 (2.3%)	0.04
Rheumatoid arthritis	509 (1.1%)	102 (2.0%)	0.08
Renal
Chronic kidney disease	1,981 (4.1%)	351 (7.0%)	0.12
Musculoskeletal
Hip fracture	167 (0.3%)	50 (1.0%)	0.08
Osteoporosis	1,727 (3.6%)	397 (7.9%)	0.19
Osteoarthritis	8,150 (16.9%)	1,300 (25.8%)	0.22
Other
HIV	183 (0.4%)	19 (0.4%)	0
Hypertension	11,709 (24.3%)	2,172 (43.1%)	0.4
Visual impairment	299 (0.6%)	37 (0.7%)	0.01
Hearing impairment	2,237 (4.6%)	349 (6.9%)	0.1
Falls	3,593 (7.5%)	323 (6.4%)	0.04
Psychiatric and functional scales (Mean ± SD)
Activity of Daily Living Hierarchy Scale	0.17 ± 0.66	0.45 ± 1.10	0.31
Aggressive Behavior Scale	0.57 ± 1.62	0.41 ± 1.33	0.11
Anhedonia Scale	4.49 ± 4.12	6.68 ± 4.27	0.52
Cognitive Performance Scale	0.39 ± 0.84	0.77 ± 1.23	0.35
Depressive Severity Index	5.01 ± 4.00	6.28 ± 4.28	0.31
Mania Scale	2.26 ± 3.39	1.54 ± 2.53	0.24
Positive Symptoms Scale	2.22 ± 3.48	1.79 ± 2.95	0.13
Risk of Harm to Others	1.35 ± 1.52	1.02 ± 1.20	0.24
Self-care Index	1.44 ± 1.55	1.90 ± 1.70	0.28
Severity of Self-harm Scale	2.56 ± 1.81	2.32 ± 1.83	0.13
Past-year health service use (Median IQR)
Mental health
Outpatient visits to psychiatrist	1 (0 to 5)	6 (2 to 12)	0.81
Outpatient visits to family physician	1 (0 to 4)	2 (0 to 5)	0.14
Emergency visits	0 (0 to 1)	0 (0 to 1)	0.09
Hospitalizations	0 (0 to 0)	0 (0 to 1)	0.57
Nonmental health
Outpatient visits to family physician	4 (2 to 9)	5 (2 to 10)	0.1
Emergency visits	1 (0 to 2)	0 (0 to 2)	0.08
Hospitalizations	0 (0 to 0)	0 (0 to 0)	0.1

Note: For individuals who never received ECT, we used their characteristics from the first admission of the study period, and for individuals who received ECT, we used their characteristics from the first ECT they received ECT. PTSD = post-traumatic stress disorder; HIV = human immunodeficiency virus; SD = standard deviation.

Table 2.

Psychotropic Medication Prescriptions for Patients Eligible for Public Drug Coverage.

Medication	No ECT Treatment	ECT Treatment	Absolute Standardized Difference
Greater than or equal to 65 years old	n = 5,479	n = 1,681
Anticonvulsant	647 (11.8%)	194 (11.5%)	0.01
Antidepressant	3988 (72.8%)	1475 (87.8%)	0.38
Antipsychotic	2216 (40.4%)	1125 (66.9%)	0.55
Benzodiazepine	2737 (50.0%)	937 (55.7%)	0.12
Cholinesterase inhibitor	317 (5.8%)	127 (7.6%)	0.07
Lithium	349 (6.4%)	152 (9.0%)	0.1
Less than 65 years old on social assistance	n = 14,912	n = 1,324
Anticonvulsant	3323 (22.3%)	307 (23.2%)	0.02
Antidepressant	9306 (62.4%)	960 (72.5%)	0.22
Antipsychotic	7744 (51.9%)	847 (64.0%)	0.25
Benzodiazepine	6939 (46.5%)	782 (59.1%)	0.25
Cholinesterase inhibitor	26 (0.2%)	8 (0.6%)	0.07
Lithium	1446 (9.7%)	174 (13.1%)	0.11

Note: For individuals who never received ECT, we used their characteristics from the first admission of the study period, and for individuals who received ECT, we used their characteristics from the first ECT they received ECT. ECT = electroconvulsive therapy.

Multivariable Models

Estimates of adjusted associations from the multivariable logistic GEE model are presented in Figure 1. The c-statistic for the primary regression model was 0.82, and no variable had tolerance <0.2 indicating multicollinearity was not problematic.

Figure 1.

Characteristics associated with receiving inpatient electroconvulsive therapy (ECT)—as represented by adjusted odds ratios (ORs) and 95% confidence interval (CI)—for (A) sociodemographic characteristics, (B) clinical characteristics, (C) psychiatric comorbidities, (D) psychiatric and functional assessment scales, and (E) health service utilization. Greater OR indicates higher likelihood an individual receives ECT.

Sociodemographic factors

Adjusted odds of ECT treatment increased with age and were also higher for individuals who were married or partnered versus never married, who had achieved any postsecondary education versus high school or less, who lived in a rural versus urban dwelling, who lived in the highest versus lowest neighborhood income quintile, and were eligible versus ineligible for public drug coverage.

Illness severity

Adjusted odds of ECT treatment increased with number of prior psychiatric hospitalizations and number of outpatient visits to a psychiatrist in the year preceding hospital admission. Adjusted odds of receiving ECT were also higher for individuals with more severe depressive symptoms (assessed by Depressive Severity Index), more social withdrawal (assessed by Anhedonia Scale), worse impairment in their ADLs (assessed by Activity of Daily Living Hierarchy Scale), and worse functional impairment due to baseline cognition (assessed by Cognitive Performance Scale). Adjusted odds of receiving ECT were lower for individuals with more severe mania (assessed by the Mania Scale), more psychotic symptoms (assessed by Positive Symptoms Scales), and more aggressive behavior toward others (assessed by Risk of Harm to Others Scale) or themselves (assessed by Severity of Self-harm Scale). The odds of ECT were also greater for those admitted due to inability to care for self due to illness versus other reasons, and for those incapable versus capable to consent to treatment.

Diagnosis and comorbidity

Adjusted odds of receiving ECT were higher for individuals with unipolar depressive disorder versus bipolar or unspecified depression, as well as a diagnosis of depression with psychotic features versus without psychotic features. Factors associated with reduced odds of receiving ECT were admission due to substance use (vs. other reasons), involuntary admission (vs. voluntary), diagnosis of substance use disorder (vs. not), diagnosis of chronic obstructive pulmonary disease, congestive heart failure, previous stroke, or falls (Supplemental Figure S1).

Additional Analyses

For individuals with prescription medication information (Figure 2), in those ≥65 years, a prescription for antidepressant or antipsychotic was associated with increased odds of ECT, while for individuals <65 years on social assistance, a prescription for antipsychotic medication was associated with increased odds of ECT, and a prescription for anticonvulsant medication was associated with reduced odds of ECT. The subgroup analysis for individuals <65 years on social assistance does not account for clustering as the GEE algorithm failed to converge in this subgroup.

Figure 2.

Prescription medications associated with receiving inpatient electroconvulsive therapy (ECT)—as represented by odds ratios (ORs) and 95% confidence interval (CI)—for (A) individuals 65 years and older and (B) individuals under 65 years old on social assistance (naive analysis). Greater OR indicates higher likelihood an individual receives ECT.

Discussion

The current work provides valuable information regarding the real-world use of ECT and is based on robust analyses using population-level data from a large jurisdiction with detailed clinical information rarely available in administrative health databases. We found that nearly 10% of patients admitted for depression for longer than 3 days ultimately received inpatient ECT. Patient-level characteristics differed markedly between those who did and did not receive ECT on an inpatient psychiatric unit. While some characteristics including diagnosis, illness severity and comorbidity were consistent with best clinical practice and clinically sensible, several nonclinical factors were also strongly associated with receipt of ECT. These latter characteristics are surprising and require further investigation as they may represent important and modifiable disparities in access to ECT treatment.

Our findings indicate that illness severity characteristics impact whether an individual will receive inpatient ECT. Higher depression severity, more social withdrawal, more functional impairment in ADLs, more psychiatric hospitalizations, and more outpatient psychiatrist visits were associated with increased odds of receiving ECT. Similarly, other clinical factors such as depression with psychotic features increased the likelihood of receiving ECT and several medical comorbidities or comorbid substance use disorders reduced the likelihood of receiving ECT. These findings align with clinical practice guideline recommendations in which ECT is recommended for severe, refractory depression (e.g., TRD) or for depression with psychotic features.⁷ Unexpectedly, we found admissions due to self-harm risk and higher scores on a scale assessing self-harm risk were associated with lower likelihood of receiving ECT. One potential explanation for this finding is that these results may be capturing self-harm risk due to personality pathology (e.g., borderline personality disorder), for which ECT is noticeably less effective.^22,23 Similarly surprising is our finding that bipolar disorder was strongly associated with reduced likelihood of receiving ECT despite evidence that bipolar and unipolar depression respond similarly to ECT.²⁴ One potential explanation is that individuals with bipolar depression are often treated with mood stabilizers, such as lithium or anticonvulsants,²⁵ which may interfere with the efficacy of ECT or increase risk of complications.^26,27 While there are no absolute contraindications to ECT, medical comorbidities or active substance use may increase the safety risk of providing treatment with ECT, likely due to risks associated with general anesthesia, and reduced odds of use in the presence of these characteristics reflect appropriate practice patterns.^7,28

A noteworthy finding in this study is that incapacity to consent to treatment was associated with increased odds of receiving ECT. To our knowledge, though studies have examined the role of capacity in treatment outcomes for depressed patients, there are no studies examining capacity on the likelihood of ECT.²⁹ In Ontario, Canada, there is a well-developed legal framework designed to protect the right of individuals to make autonomous treatment decisions. When an individual is declared incapable to consent to a specific psychiatric treatment (ECT or otherwise), they are notified with a legislatively defined form and have the opportunity to challenge this finding at an independent tribunal in which they will receive legal representation at no cost. If the finding of incapacity is upheld or the individual does not challenge the finding of incapacity then informed consent is provided by a substitute decision maker defined by legislation. As a result of this rigorous process, it is unlikely that the presence of incapacity to consent to a treatment (ECT or otherwise) is causally related to an increased use of ECT. Instead, treatment incapacity more likely serves as a marker for severe illness or psychotic symptoms (both clinical indications for ECT⁷) that are not captured by other measured covariates. An important indication for ECT, and one associated with treatment incapacity, is catatonia, a syndrome characterized by psychomotor disturbances, which often occurs in severe mood disorders.^30,31 We could not assess the association of this comorbidity with ECT because the diagnostic code had a very low prevalence (<1%), which likely reflects both clinical underrecognition³¹ and undercoding in administrative health databases.

Factors like illness severity or comorbid conditions indicate clinical need or relative contraindications for ECT treatment. However, several characteristics that impacted the odds of receiving ECT in this study are considered “enabling resources” by health care utilization models that reflect logistical aspects of obtaining care.³² In this study, being unmarried, less educated, and living in a lower income area (i.e. enabling resources) each independently reduced the odds that a person would receive ECT, even after accounting for clinical factors and despite all patients being treated in the common environment of a psychiatric inpatient unit in a universal health care system where ECT and all inpatient care is publicly funded. These are important findings because, when access to medical care is determined by enabling resources, as opposed to need factors (e.g., illness severity), this may indicate inequitable access to medical care.³² Inequitable access to ECT has previously been suggested in the U.S. Veterans Affairs health system, where black race, particular geographic regions (e.g., South and West U.S.), and living near a facility that does not perform ECT were all associated with reduced odds of receiving treatment.¹⁰ Our findings are also consistent with a recent study from Denmark, which has a similar universal health care system to Canada, that found age and depression severity increase the likelihood of ECT, while alcohol use, being unmarried, and having low education decrease the likelihood of ECT.¹¹ These consistent results from a variety of jurisdictions support our hypothesis that ECT use is driven by a combination of appropriate clinical factors and questionable nonclinical factors. Although our study used Canadian data, this finding that enabling resources influence receipt of ECT also likely applies to other health care systems who are seeking to improve equitable access to this important treatment.

The underlying mechanisms that explain the relationship between enabling resources and receipt of ECT are unknown but may involve factors at the patient, physician, and system level. For example, the finding that individuals with higher education are more likely to receive ECT—even after accounting for other factors—is unexpected and has a number of possible explanations. It is possible that the manner in which informed consent for ECT is obtained may not be suitable for individuals with a lower literacy level or that lower levels of education may reflect less trust in the health care system.³³ Alternatively, individuals with less education may have less insight into their psychiatric symptoms,³⁴ which may result in these patients being less likely to consent to an intensive treatment such as ECT. There may also be a provider bias against individuals with less education. Individuals with higher levels of education have previously been demonstrated to be more satisfied following ECT treatment,³⁵ which over time may result in physicians preferentially selecting individuals with higher levels of education. It is also possible that the seemingly disparate characteristics increasing the odds of receiving inpatient ECT (e.g., living in a rural area, higher neighborhood income, being married) are connected by the fact that they reflect an individual’s membership in a community with higher levels of social capital. Social capital, which has been defined as “features of social life—networks, norms, and trust—that enable participants to act together more effectively to pursue shared objectives,”³⁶ has been demonstrated to influence an individual’s ability to access care.³⁷ Therefore, it may be that individuals living in communities with higher levels of social capital may be able to more effectively navigate a complex and challenging health care system to access effective treatments. Finally, providers, due to their concerns about whether people who have an initial course of ECT as inpatients will be able to navigate outpatient treatment, may also be more likely to recommend the treatment for people who have higher levels of social stability and resources.

Although our findings potentially indicate a concern with respect to equitable access to ECT treatment, we found that nearly 10% of inpatients with depression eventually received ECT. The prevalence of use in our study is similar to a study using data from 1993 in the United States, which found that 9.4% of inpatients with depression received ECT.³⁸ However, a more recent study using administrative data from 2015 to 2017 in 9 different U.S. states, found that ECT was used in only 1.5% of inpatients with depression.³⁹ A detailed analysis examining the declining use of ECT in general hospitals in the United States found the decline was due to fewer hospitals conducting ECT (70.5% in 1993 vs. 44.7% in 2009) as opposed to changing clinical characteristics over time (e.g., less severely ill patients being admitted).⁹ Although the current study did not examine temporal trends of ECT our results suggest that at least in Ontario, Canada, ECT is not underutilized in this severely ill population. However, there appear to be opportunities for improving equitable access to this important treatment.

Limitations

There are several limitations of this work to consider. First, we maintained our focus on only patient-level characteristics. Institutional- and provider-level characteristics and geographic variation may also play an important role in the receipt of ECT; however, the c-statistic of 0.82 does suggest that the discrimination of who will get ECT is quite high even using only the patient-level characteristics.⁸ Future work examining the use of ECT should include institutional characteristics (e.g., number of yearly ECT procedures) as well as geographic features (e.g., patient proximity to hospital with ECT machines) to better characterize the delivery and utilization of this important procedure. Second, we considered only inpatients with depression, which limits the generalizability of our findings to other conditions that can be treated with ECT, such as schizophrenia,⁴⁰ or for other treatment locations (e.g., outpatients). Given that 80% of the ECT delivered in Ontario is provided to inpatients and that depression is the main indication for the procedure, we felt the emphasis on this population to be reasonable.¹⁵ Third, while we used data from the population of Ontario, our results may not be generalizable to different health care models, such as the United States, where insurance status may have a significant impact on the receipt of ECT.¹³ Last, while we had data on a wide variety of characteristics, we did not have detailed information on important characteristics, such as treatment resistance, which may result in residual confounding impacting our findings.⁴¹ In particular, our results could be partially impacted by unmeasured covariates including factors such as patient preferences, previous experience with ECT, or degree of treatment resistance. Given the stigma associated with ECT, further qualitative research on the role of patient preference will be important. However, many characteristics associated with receiving inpatient ECT (presence of psychotic features, increased age, symptom severity, etc.) are consistent with best clinical practice, which supports the validity of our overall model. This in turn lends greater confidence to our unexpected findings regarding characteristics suggestive of inequitable access to ECT treatment.

Conclusion

We found that nearly 1 in 10 patients hospitalized for depression in Ontario, Canada, received ECT. While receipt of ECT was associated with many clinically sensible factors, several unexpected sociodemographic characteristics (e.g., marital status, education level, neighborhood income quintile) that likely reflect logistical aspects in obtaining care, were associated with reduced likelihood of receiving ECT. Further research is needed to define mechanisms and identify solutions to ensure equitable access to ECT.