Comparing Administrative Data and Resident Assessment Instrument Data for Mental Health Surveillance in Home Care and Long‐Term Care Populations Authors: Ruth Ann Marrie, James M Bolton, Xueyi Chen, Erind Dvorani, Okechukwu Ekuma, Zahra Goodarzi, Andrea Gruneir, Jin Luo, Nathan Herrmann, Matthias Hoben, David B Hogan, Andrea Iaboni, Dallas Seitz, Randy Walld, Walter P Wodchis, Colleen J Maxwell, EMHLTH‐CC Research Team Published: 2026 ABSTRACT ABSTRACTObjectivesTo compare mental illness prevalence, and agreement between mental illness measures, in home care (HC) and long‐term care (LTC) using administrative definitions and Resident Assessment Instrument (RAI)‐diagnoses.MethodsApplying a common protocol to linked administrative‐RAI datasets in Alberta, Manitoba, and Ontario, Canada, we selected adults receiving HC or LTC between 2012 and 2023. Cross‐sectionally, we compared administrative definitions with RAI‐diagnoses (depression, anxiety, bipolar disorder and schizophrenia) using kappa, sensitivity, specificity, positive and negative predictive value. We examined discordance‐associated factors using multivariable logistic regression.ResultsDepression was the most prevalent mental illness (administrative data, HC: 29%–34% of 493,895 people; LTC: 21%–51% of 357,117 people). Prevalence estimates were consistently lower using RAI‐diagnoses than administrative definitions. Agreement between data sources was moderate for depression (k = 0.33–0.57) and anxiety disorders (k = 0.35–0.63). Younger age, dementia, and more physician visits were associated with higher likelihood of discordance between data sources; male sex had a lower likelihood.ConclusionWhile mental illness prevalence is high in HC and LTC populations, RAI‐based estimates are lower than administrative data estimates. Discordance between data sources differs by individual characteristics. Concurrent comparison of RAI‐diagnoses and administrative data to a gold standard clinical interview would clarify the optimal surveillance strategy.Keywords: agreement, home care, long‐term care, mental illness, Resident Assessment Instrument Introduction Mental health disorders are common in home care (HC) and long‐term care (LTC) populations (Neufeld et al. 2021; Seitz et al. 2010). In a Canadian study half of HC clients had a mental health diagnosis (Poss et al. 2021). In LTC, the estimated prevalence of depressive disorder ranges from 5% to 25% and of depressive symptoms ranges from 14% to 82% (Neufeld et al. 2021; Seitz et al. 2010). Yet, mental health disorders are under‐diagnosed and undertreated among HC and LTC recipients (Dalby et al. 2008), due to patient‐level (e.g., stigma, comorbidities), provider‐level (e.g., insufficient staff training), and system‐related factors (e.g., staffing shortages, poor access to mental health services) (Azulai and Hall 2020; Poss et al. 2021). Untreated mental illness adversely affects lifespan, functional status, and quality of life (GBD 2019 Mental Disorders Collaborators 2022). This highlights the need for effective surveillance strategies and interventions for these populations. Herein, we refer to LTC as health and personal care services provided to individuals with complex support needs residing in designated LTC homes (also known as nursing homes, personal care homes, and residential care facilities) that provide 24‐hour support (Canadian Institute for Health Information 2026). National chronic disease surveillance often uses population‐based administrative (health claims) data. The Canadian Chronic Disease Surveillance System (CCDSS), for example, collects aggregate regional population‐based information (Lix et al. 2018) using claims‐based case definitions. While claims data are population‐based, relatively low cost, and sustainable, they only capture information about conditions for which care is sought, recognized and documented through the public health system and lack clinical details. This applies particularly to mental illness, for which concerns about validity and reliability of administrative data have been raised (Kisely, Lin, Gilbert, et al. 2009), and to HC and LTC settings where mental health services are fragmented (Poss et al. 2021), highlighting the need to evaluate the performance of alternative data sources to support surveillance. In many regions (Hirdes et al. 2008), individuals receiving HC or residential LTC are routinely assessed using a Resident Assessment Instrument (RAI) (Supporting Information S1: Table e1) (Gray et al. 2009; Hirdes et al. 2008), which explicitly captures diagnoses using a picklist. RAI assessments are mandated to occur for entire HC or LTC populations at consistent intervals; therefore, they may offer a cost‐effective, more inclusive means of surveillance because they are not dependent on health‐seeking behaviour unlike administrative data. Yet, we do not know the relationship between measures of mental illness using RAI diagnoses and the administrative data more commonly used for surveillance. We compared the prevalence of mental illness in HC and LTC homes using administrative (health claims‐based) definitions and diagnoses contained in the RAI, as well as the agreement between these measures to assess the potential value of using clinical data to enhance mental illness surveillance in these settings. Finally, we examined factors associated with discordance between the classification of mental illness in the two data sources. Methods Design We conducted retrospective cross‐sectional analyses of prospectively collected linked administrative and clinical datasets including HC and LTC populations. We included adults of all ages in these populations because of our interest in mental illness surveillance, and the high prevalence of mental illness in younger and older adults. We conducted the analysis in parallel within three Canadian provinces (Alberta, Manitoba, and Ontario) due to privacy regulations. Together, these provinces account for ∼53% of the Canadian population. Study Period Due to differences in data availability, the maximum observation (accrual) period was 1‐April‐2013 to 31‐March‐2023 in Alberta and 1‐April‐2012 to 31‐March‐2023 in Manitoba and Ontario. Data Sources We used administrative and clinical assessment databases (capturing interRAI instruments (Hirdes et al. 2008) and RAI‐MDS 2.0 (Canadian Institute for Health Information 2012; Morris et al. 1997)) for HC and LTC available in all provinces unless specified otherwise. In Alberta we accessed these databases through the Alberta Strategy for Patient‐Oriented Research SUPPORT Unit while we accessed them from the Manitoba Population Research Data Repository at the Manitoba Centre for Health Policy in Manitoba and through ICES in Ontario. In each province databases were linked using unique encoded identifiers and analysed where the data were accessed (e.g., ICES). In all provinces, medically necessary health services are publicly funded and universal, and captured in administrative databases. The administrative databases (data elements used) included: population registries (sex, dates of birth, death and health care coverage, postal code of residence); Discharge Abstract Database (DAD, hospital admission and separation dates, discharge diagnoses, intervention codes); medical/physician services (the Ontario Health Insurance Plan in Ontario, service date, physician‐assigned diagnosis); prescription claims (Pharmaceutical Information Network in Alberta, Drug Prescription Information Network in Manitoba, Ontario Drug Benefit Database in Ontario; drug dispensed, dates dispensed, and drug identification number which is linked to the WHO's Anatomical Therapeutic Chemical Classification System). In Ontario only, we also used the Ontario Mental Health Reporting System (OMHRS, admission and discharge dates, discharge diagnoses to designated adult inpatient mental health beds) (Canadian Institute for Health Information 2024). In the DAD, diagnoses are recorded using 5‐digit International Classification of Disease (ICD)‐10‐Canadian enhancement codes (Canadian Institute for Health Information 2015), and interventions using the Canadian Classification of Interventions. In medical/physician claims, diagnoses are recorded using ICD‐9‐clinical modification (ICD‐9‐CM) in Alberta and Manitoba and a modified ICD‐9 version in Ontario. HC and LTC are extended health services which may be publicly or privately funded. A standardized tool is used to assess clinical, functional and psychosocial characteristics of individuals receiving publicly funded HC. The Resident Assessment Instrument‐Home Care (RAI‐HC) was used in all study years in Alberta. In Manitoba and Ontario, the RAI‐HC was used until April 2018, when the interRAI Home Care Assessment Tool (interRAI‐HC) was adopted (Supporting Information S1: Table e2). The HC assessments are captured in the Alberta Continuing Care Information System, in the Winnipeg Regional Health Authority MDS in Manitoba, and the HC database and RAI‐HC in Ontario. In Manitoba, clinical assessments are only captured in the Winnipeg Regional Health Authority, where 57% of the provincial population resides. For LTC residents, another tool, the Resident Assessment Instrument Minimum Data Set (RAI‐MDS) version 2.0 is used in all three provinces (all LTC homes). A full assessment is completed within 14 days of admission, annually and when a significant health status change occurs. In March 2021, Alberta began switching to the new interRAI LTC Assessment, but this affected only 0.7% of the Alberta cohort, which we excluded to minimize heterogeneity. In Alberta, these data are captured in the Alberta Continuing Care Information System, whereas they are in the LTC MDS in Manitoba (Winnipeg Regional Health Authority), and in the Continuing Care Reporting System—LTC in Ontario. RAI assessments are intended to be recorded by a single assessor based on multiple sources including the medical record, direction observation, and multi‐disciplinary input, when relevant to their current health status and care (Supporting Information S1: Table e1); conditions and diagnoses that are resolved or no longer relevant are not captured (Canadian Institute for Health Information 2012). The RAI‐MS and interRAI‐HC demonstrate acceptable interrater reliability (Hirdes et al. 2008; Morris et al. 1997). Diagnostic validity as compared to hospital discharge diagnoses varies by condition (Lix et al. 2014; Mor et al. 2011; Wodchis et al. 2008). Study Populations We identified adults receiving HC or LTC in each province during the accrual period who were aged ≥ 18 years on their first available RAI assessment during the study period; this constituted their index date. The index year included the 6 months on either side of the index date. Exclusion criteria for both cohorts were: (i) not alive and in the population registry on their index date; (ii) invalid unique identifier; (iii) invalid sex; (iv) age < 18 or age > 110 years on the index date; (v) non‐resident in the relevant province at the index date; (vi) lacked health care coverage for ≥ 180 days within the 5‐year period before the index date; and (vii) RAI assessment in the 12 months preceding their index year for those with an index date in fiscal year 2012 (to capture the first assessment). For this analysis, we further restricted the HC cohorts in Manitoba and Ontario to individuals assessed using the interRAI‐HC (starting April 2018) because it is the current tool and provides more granular mental illness diagnoses. Mental Illness We derived mental illness measures three ways. First, we identified mental illness (depression, anxiety, bipolar disorder, schizophrenia) by applying administrative definitions that relied on combinations of hospital, physician or prescription claims. The definitions were developed by the CCDSS or validated in chronic disease populations (Supporting Information S1: Table e3) and included definitions to estimate mental illness occurrence (Kisely, Lin, Gilbert, et al. 2009; Kisely, Lin, Lesage, et al. 2009; Lix et al. 2018; Marrie et al. 2013, 2016; Vanasse et al. 2012). For example, one hospitalization or two physician visits ≥ 30 days apart in a 2‐year period with ICD‐9/ICD‐10‐CA diagnostic codes 295, F20, F21, F23, or F25 met the schizophrenia definition. Since the bipolar disorder definition required a 5‐year period to optimize performance, all definitions were applied over a 5‐year lookback period to ensure consistency, beginning 6 months after the index date to ensure overlap between the administrative data and the RAI data. Individuals who met the case definition at any point in the 5‐year period were considered affected during the index year. For depression and anxiety we included definitions with/without prescription claims (identified by suffix ‐Rx) (Marrie et al. 2013, 2016). Second, we identified the mental illness diagnoses recorded on the RAI including depression, anxiety disorder, bipolar disorder and schizophrenia (Supporting Information S1: Table e4). Hereinafter, all RAI‐diagnoses are identified by the prefix RAI. Covariates Covariates from the administrative datasets used to describe the cohorts, or used in the discordance analyses included age, sex, Canadian Index of Multiple Deprivation (CIMD) (Statistics Canada 2019), urban/rural residence, comorbidities, fiscal year of the index date, and number of non‐mental health physician visits in the year preceding the index period. The CIMD is derived from linkage of postal code to dissemination area‐level census data and comprises four factor scores covering the dimensions of residential instability, economic dependence, ethno‐cultural composition and situational vulnerability which can be categorized into quintiles. We grouped those quintiles as 1–3 (lower) or 4–5 (higher) marginalization. Missing values were assigned to the lower category. Rural residence was defined as population of < 10,000 (Barkey and Renaud 2025). Using validated definitions from the CCDSS or other sources we identified morbidities including hypertension, ischemic heart disease (Tu et al. 2010), stroke, diabetes mellitus (Blanchard et al. 1996), multiple sclerosis (Culpepper et al. 2019), epilepsy,(Tu et al. 2014) Parkinsonism (Butt et al. 2014), and dementia (Supporting Information S1: Table e5) (Jaakkimainen et al. 2016). We applied these definitions to the 5‐year period preceding (not including) the index date. We summarized the morbidities as a count (0, 1, 2, 3+); we excluded dementia from the count because dementia was of distinct interest in HC and LTC care settings. Covariates from the RAI datasets used to describe the cohorts included marital status, Activities of Daily Living (ADL) Hierarchy Scale, and Pain Scale. Marital status was classified as not married, married/partnered or widowed. The ADL Hierarchy Scale measures degree of dependence in personal hygiene, locomotion in home, toilet use, and eating. Scores range from 0 (independent) to 6 (total dependence) (Morris et al. 1999). In HC, the ADL score was categorized as relatively independent (0, 1), mild impairment (2), moderate impairment (3), and more severe impairment (4–6). Due to skewing of the distribution to more severe impairment in LTC, we combined the two lowest categories. The Pain Scale describes the frequency and intensity of pain (Fries et al. 2001). Scores range from 0 (no pain) to 3 (severe daily pain) or 0 to 4 depending on the version of the RAI used in HC, and were categorized as no pain (0), less than daily pain (1), mild‐moderate daily pain (2) and severe daily pain (3–4). Analysis We summarized participant characteristics using descriptive statistics including mean (standard deviation [SD]), median (interquartile range [IQR]), and frequency (percent) as appropriate. Within each province we compared LTC and HC cohorts using standardized differences; standardized differences of 0.20 are small effects, 0.5 are medium, and 0.8 are large (Andrade 2020). Using the administrative definitions and RAI‐diagnoses, we estimated 12‐month prevalence of mental illness in the index year, across all study years. First, to measure agreement between the various administrative and clinical metrics (Supporting Information S1: Table e4) where neither was considered the reference standard, we used a kappa statistic with 95%CI (Sim and Wright 2005). Kappa indicates the proportion of agreement beyond chance and is affected by the prevalence of the condition of interest such that if prevalence is very high or very low, chance agreement is high, and kappa is reduced with a maximum value of < 1 (Sim and Wright 2005). Because prevalence and bias influence the magnitude of kappa, we also calculated the prevalence index, bias index, and the prevalence and bias‐adjusted kappa (Sim and Wright 2005). We interpreted κ as: slight (0–0.20), fair (0.21–0.40), moderate (0.41–0.60), substantial (0.61–0.80), and almost perfect agreement (0.81–1.0) (Landis and Koch 1977). Second, we also report sensitivity, specificity, positive and negative predictive value where the RAI mental illness measures were considered as the reference standard given the use of clinical records in the RAI assessment, to better understand the implications if the RAI were used alone, without additional administrative definitions which national surveillance relies upon. Recognizing that multiple interrelated factors could affect discordance, where the data sources were discordant regarding the presence of diagnosis (i.e., present in administrative data but absent in RAI‐diagnoses or absent in administrative data but present in RAI‐diagnoses), we examined factors associated with discordance where ‘not discordant’ (i.e., both data sources classified a diagnosis as present, or both classified a diagnosis as absent) was the reference group, using a binary logistic regression model for depression, anxiety, bipolar disorder and schizophrenia. We restricted HC analyses to Manitoba and Ontario but included all provinces for LTC analyses. Characteristics of a priori interest were age (18–64, 65–74, 75–84, ≥ 85 [reference]), sex (male, female [reference]), dementia status (no as reference), number of comorbidities (0–1 [reference], 2, ≥ 3), and number of non‐mental health physician visits (continuous) (Wang et al. 2024). We focused on these characteristics because they were associated with mental illness and health care use, were measured at the level of the individual (rather than area), and the expected consistency in their determination across provinces. Covariates included only as potential confounders or predictors based on their associations with mental illness risk, detection, and/or need for HC or LTC services were: residence (rural, urban [reference]), the four CIMD measures (more marginalized, less marginalized [reference], marital status (not married, widowed, married [reference]), ADL (relatively independent [reference in HC; modified for LTC as noted earlier], mild, moderate, more severe impairment, and pain (none [reference], less than daily, more than daily/more severe) (Wang et al. 2024). We used variance inflation factors to assess collinearity. Statistical analyses were conducted using SAS V9.4 (SAS Institute Inc., Cary, NC). Results Cohorts Over the entire study period we identified 493,895 people in HC (AB: 159,112, MB: 14,719, ON: 320,064) and 357,117 people in LTC (AB: 55,475, MB: 20,556, ON: 281,086) (Table 1, Supporting Information S1: Table e6). In all provinces, the LTC population was older than the HC population, had more ADL impairments and a higher prevalence of dementia. When we compared HC cohorts within Manitoba and Ontario that were included (interRAI‐HC) or excluded (RAI‐HC), we noted few meaningful differences apart from fiscal year of assessment (Supporting Information S1: Table e7). In Ontario the percentage of people who were relatively independent with respect to ADL was 10% lower in the interRAI‐HC cohort than the RAI‐HC cohort. TABLE 1Characteristics of the home care and long‐term care populations.AlbertaManitoba a OntarioTotal LTCHome careTotal LTCHome careTotal LTCHome careVariable N = 55,475 N = 159,112Std diff N = 20,556 N = 14,719Std diff N = 281,086 N = 320,064Std diffFiscal year of index, n (%)20122757 (13.4)0 (0.0)0.5634,761 (12.4)0 (0.0)0.5320137315 (13.2)27,932 (17.6)0.121810 (8.8)0 (0.0)0.4427,062 (9.6)0 (0.0)0.4620145442 (9.8)18,336 (11.5)0.061884 (9.2)0 (0.0)0.4526,483 (9.4)0 (0.0)0.4620155121 (9.2)15,139 (9.5)0.011798 (8.7)0 (0.0)0.4425,388 (9.0)0 (0.0)0.4520165134 (9.3)15,020 (9.4)0.011771 (8.6)0 (0.0)0.4325,565 (9.1)0 (0.0)0.4520175242 (9.4)13,990 (8.8)0.021692 (8.2)0 (0.0)0.4225,763 (9.2)0 (0.0)0.4520185619 (10.1)14,108 (8.9)0.041697 (8.3)2978 (20.2)0.3525,865 (9.2)66,298 (20.7)0.3320195538 (10.0)13,769 (8.7)0.051966 (9.6)3266 (22.2)0.3525,728 (9.2)70,866 (22.1)0.3620204838 (8.7)11,143 (7.0)0.061411 (6.9)2550 (17.3)0.3313,448 (4.8)40,248 (12.6)0.2820215694 (10.3)13,828 (8.7)0.051861 (9.1)2967 (20.2)0.3224,540 (8.7)66,047 (20.6)0.3420225532 (10.0)15,847 (10.0)01909 (9.3)2958 (20.1)0.3126,483 (9.4)76,605 (23.9)0.4Age at index dateMedian (IQR)85 (77–91)81 (72–88)0.3686 (79–91)80 (71–87)0.5585 (78–90)80 (71–87)0.45Age group, n (%)18–644144 (7.5)21,467 (13.5)0.2988 (4.8)2043 (13.9)0.3217,414 (6.2)47,949 (15.0)0.2965–746559 (11.8)25,904 (16.3)0.132310 (11.2)2830 (19.2)0.2233,296 (11.8)59,685 (18.6)0.1975–8415,676 (28.3)51,378 (32.3)0.095955 (29.0)4891 (33.2)0.0988,026 (31.3)106,587 (33.3)0.0485+29,096 (52.4)60,363 (37.9)0.2911,303 (55.0)4955 (33.7)0.44142,350 (50.6)105,843 (33.1)0.36SexFemale32,197 (58.0)91,912 (57.8)0.0113,064 (63.6)8623 (58.6)0.1177,838 (63.3)180,087 (56.3)0.14Male23,278 (42.0)67,200 (42.2)0.017492 (36.4)6096 (41.4)0.1103,248 (36.7)139,977 (43.7)0.14Urban/Rural residence, n (%)Missing2244 (0.8)1072 (0.3)0.06Urban46,296 (83.5)131,923 (82.9)0.0119,985 (97.2)14,344 (97.5)0.01241,090 (85.8)277,983 (86.9)0.03Rural9179 (16.5)27,189 (17.1)0.01571 (2.8)375 (2.5)0.0137,752 (13.4)41,009 (12.8)0.02Canadian Index of Multiple Deprivation, n (%)Residential instabilityLess marginalized27,126 (48.9)79,338 (49.9)0.027242 (35.2)5734 (39.0)0.08144,556 (51.4)172,774 (54.0)0.05Most marginalized28,349 (51.1)79,774 (50.1)0.0213,314 (64.8)8985 (61.0)0.08136,530 (48.6)147,290 (46.0)0.05Ethno‐cultural compositionLess marginalized29,860 (53.8)83,860 (52.7)0.029355 (45.5)6646 (45.2)0.01159,644 (56.8)161,319 (50.4)0.13Most marginalized25,615 (46.2)75,252 (47.3)0.0211,201 (54.5)8073 (54.8)0.01121,442 (43.2)158,745 (49.6)0.13Economic dependencyLess marginalized36,631 (66.0)106,872 (67.2)0.0210,969 (53.4)7970 (54.1)0.02130,628 (46.5)140,502 (43.9)0.05Most marginalized18,844 (34.0)52,240 (32.8)0.029587 (46.6)6749 (45.9)0.02150,458 (53.5)179,562 (56.1)0.05Situational vulnerabilityLess marginalized35,683 (64.3)99,222 (62.4)0.0411,179 (54.4)7630 (51.8)0.05187,543 (66.7)207,831 (64.9)0.04Most marginalized19,792 (35.7)59,890 (37.6)0.049377 (45.6)7089 (48.2)0.0593,543 (33.3)112,233 (35.1)0.04Marital status, n (%)Not married4635 (22.5)3479 (23.6)0.0356,331 (20.0)66,916 (20.9)0.02Married5748 (28.0)5896 (40.1)0.2688,182 (31.4)141,901 (44.3)0.27Widowed10,173 (49.5)5344 (36.3)0.27136,573 (48.6)111,247 (34.8)0.28Activities of Daily Living, n (%)Relatively independent3984 (7.2)118,937 (74.8)1.892804 (13.6)10,722 (72.8)1.4924,306 (8.6)185,562 (58.0)1.23Mild impairment7053 (12.7)17,595 (11.1)0.055619 (27.3)1982 (13.5)0.3542,009 (14.9)51,696 (16.2)0.03Moderate impairment14,061 (25.3)10,132 (6.4)0.545564 (27.1)1271 (8.6)0.583,033 (29.5)41,076 (12.8)0.42More severe impairment30,377 (54.8)12,448 (7.8)1.176569 (32.0)744 (5.1)0.74131,738 (46.9)41,730 (13.0)0.79PainNo pain30,850 (55.6)61,656 (38.8)0.3413,689 (66.6)5754 (39.1)0.57167,456 (59.6)109,228 (34.1)0.53Less than daily pain17,079 (30.8)21,391 (13.4)0.433909 (19.0)2696 (18.3)0.0280,503 (28.6)53,388 (16.7)0.29Mild/mod daily pain6324 (11.4)57,749 (36.3)0.612610 (12.7)4510 (30.6)0.4527,311 (9.7)109,454 (34.2)0.62Severe daily pain1222 (2.2)18,316 (11.5)0.37348 (1.7)1759 (12.0)0.425816 (2.1)47,994 (15.0)0.48Dementia, n (%)32,771 (59.1)30,206 (19.0)0.911,396 (55.4)2607 (17.7)0.85157,505 (56.0)59,910 (18.7)0.84Comorbidity count, n (%)Median (IQR)2 (1–2)1 (1–2)0.192 (1–2)1 (1–2)0.151 (1–2)1 (1–2)0.16Comorbidity level, n (%)Low (0,1)25,149 (45.3)84,803 (53.3)0.169862 (48.0)7948 (54.0)0.12150,242 (53.5)192,361 (60.1)0.13Moderate (2)18,012 (32.5)48,781 (30.7)0.046670 (32.4)4600 (31.3)0.0380,553 (28.7)84,341 (26.4)0.05High (3+)12,314 (22.2)25,528 (16.0)0.164024 (19.6)2171 (14.7)0.1350,291 (17.9)43,362 (13.5)0.12Hypertension, n (%)42,500 (76.6)114,143 (71.7)0.1115,895 (77.3)10,830 (73.6)0.09179,471 (63.8)185,259 (57.9)0.12Ischemic Heart Disease15,451 (27.9)39,463 (24.8)0.075526 (26.9)2946 (20.0)0.1671,223 (25.3)69,354 (21.7)0.09Diabetes, n (%)18,044 (32.5)49,025 (30.8)0.046009 (29.2)4740 (32.2)0.0688,987 (31.7)105,844 (33.1)0.03Multiple Sclerosis, n (%)560 (1.0)1414 (0.9)0.01193 (0.9)118 (0.8)0.012198 (0.8)1815 (0.6)0.03Epilepsy, n (%)1842 (3.3)3693 (2.3)0.06397 (1.9)239 (1.6)0.026846 (2.4)5820 (1.8)0.04Parkinsonism, n (%)3362 (6.1)5041 (3.2)0.141179 (5.7)546 (3.7)0.115,605 (5.6)10,701 (3.3)0.11Stroke, n (%)11,728 (21.1)21,623 (13.6)0.24160 (20.2)2136 (14.5)0.1555,203 (19.6)43,498 (13.6)0.16# non‐MH physician visits in 1 year pre‐indexMedian (IQR)22 (10–46)16 (8–29)0.3115 (7–28)15 (8–26)0.0115 (6–34)13 (6–26)0.13Advance Directive b , n (%)Missing0 (0.0)159,112 (100.0)0 (0.0)14,719 (100.0)00 (0.0)320,064 (39.1)No45,955 (82.8)0 (0.0)20,499 (99.7)0 (0.0)26.82211,413 (75.2)0 (0.0)2.46Yes9520 (17.2)0 (0.0)57 (0.3)0 (0.0)0.0769,673 (24.8)0 (0.0)0.81Abbreviations: IQR = interquartile range, SD = standard deviation, Std = Standardized Difference. a Restricted to Winnipeg Regional Health Authority. b Do not hospitalize (only captured in long‐term care). Characteristics of the home care and long‐term care populations. Abbreviations: IQR = interquartile range, SD = standard deviation, Std = Standardized Difference. Restricted to Winnipeg Regional Health Authority. Do not hospitalize (only captured in long‐term care). Prevalence of Mental Illness Using administrative data, we observed largely the same pattern of prevalence in HC and LTC settings, and across provinces (Table 2, Figure 1), although absolute estimates were higher in Alberta and Manitoba than in Ontario. Using the RAI‐diagnoses, the most prevalent condition was depression, followed by anxiety disorders. Whether bipolar disorders or schizophrenia was more prevalent differed by province and measure, though schizophrenia prevalence was generally higher in LTC. We observed the same rank ordering of prevalence for the administrative case definitions that did not incorporate prescription claims. Prevalence estimates based on RAI‐diagnoses were lower than prevalence estimates based on administrative definitions for all conditions; the gap was widest for administrative case definitions that incorporated prescription claims and narrowest for the rarest disorders. TABLE 2Agreement between mental illness measures from administrative data and RAI diagnoses.Test measurePrevalence (per 100) (95% CI)ReferencePrevalence (per 100) (95% CI)Kappa (95% CI)PABAK (95% CI)Home CareManitobaDepression28.97 (28.11, 29.85)RAI‐depression15.08 (14.45, 15.72) 0.3349 (0.3182, 0.3515) 0.5303 (0.5166, 0.544) Depression‐Rx34.47 (33.52, 35.43)RAI‐depression15.08 (14.45, 15.72) 0.3610 (0.3460, 0.3761) 0.4997 (0.4857, 0.5137) Anxiety22.59 (21.83, 23.37)RAI‐anxiety13.45 (12.86, 14.05) 0.3011 (0.2827, 0.3196) 0.5812 (0.5681, 0.5944) Anxiety‐Rx31.84 (30.93, 32.76)RAI‐anxiety13.45 (12.86, 14.05) 0.2564 (0.2408, 0.2721) 0.4539 (0.4395, 0.4683) Bipolar5.08 (4.72, 5.45)RAI‐bipolar1.41 (1.23, 1.62) 0.3093 (0.2712, 0.3475) 0.9124 (0.9057, 0.9190) Schizophrenia2.19 (1.96, 2.44)RAI‐schizophrenia1.77 (1.56, 2.00) 0.7708 (0.7327, 0.8089) 0.9822 (0.9792, 0.9852) OntarioDepression20.25 (20.1, 20.41)RAI‐depression18.66 (18.52, 18.81) 0.3080 (0.3040, 0.3120) 0.5660 (0.5632, 0.5689) Depression‐Rx26.75 (26.57, 26.93)RAI‐depression18.66 (18.52, 18.81) 0.3889 (0.3852, 0.3926) 0.5670 (0.5641, 0.5698) Anxiety20.03 (19.87, 20.18)RAI‐anxiety14.41 (14.28, 14.54) 0.2484 (0.2443, 0.2524) 0.5691 (0.5662, 0.5719) Anxiety‐Rx29.13 (28.95, 29.32)RAI‐anxiety14.41 (14.28, 14.54) 0.2313 (0.2278, 0.2349) 0.4597 (0.4566, 0.4627) Bipolar2.57 (2.51, 2.62)RAI‐bipolar1.70 (1.65, 1.74) 0.4711 (0.4606, 0.4815) 0.9558 (0.9548, 0.9568) Schizophrenia2.11 (2.06, 2.16)RAI‐schizophrenia1.63 (1.59, 1.67) 0.6373 (0.6272, 0.6474) 0.9733 (0.9726, 0.9741) Long‐term careAlbertaDepression41.65 (41.12, 42.19)RAI‐depression27.81 (27.37, 28.25) 0.2914 (0.2835, 0.2992) 0.3439 (0.3361, 0.3517) Depression‐Rx51.19 (50.59, 51.78)RAI‐depression27.81 (27.37, 28.25) 0.3320 (0.3251, 0.3390) 0.3250 (0.3171, 0.3329) Anxiety24.60 (24.19, 25.01)RAI‐anxiety11.67 (11.39, 11.95) 0.2487 (0.2396, 0.2577) 0.5413 (0.5343, 0.5483) Anxiety‐Rx33.81 (33.32, 34.29)RAI‐anxiety11.67 (11.39, 11.95) 0.2126 (0.2050, 0.2201) 0.4081 (0.4005, 0.4157) Bipolar10.20 (9.93, 10.46)RAI‐bipolar1.60 (1.50, 1.71) 0.1990 (0.1863, 0.2118) 0.8162 (0.8114, 0.8210) Schizophrenia3.26 (3.11, 3.41)RAI‐schizophrenia1.80 (1.69, 1.91) 0.6252 (0.6041, 0.6463) 0.9629 (0.9607, 0.9651) ManitobaDepression34.04 (33.25, 34.85)RAI‐depression20.32 (19.71, 20.95) 0.3149 (0.3014, 0.3283) 0.4446 (0.4324, 0.4569) Depression‐Rx44.36 (43.46, 45.28)RAI‐depression20.32 (19.71, 20.95) 0.348 (0.3365, 0.3594) 0.3916 (0.379, 0.4042) Anxiety20.01 (19.4, 20.63)RAI‐anxiety12.42 (11.95, 12.92) 0.3295 (0.3131, 0.3458) 0.6317 (0.6211, 0.6423) Anxiety‐Rx29.21 (28.47, 29.96)RAI‐anxiety12.42 (11.95, 12.92) 0.2765 (0.2628, 0.2903) 0.5026 (0.4908, 0.5144) Bipolar4.58 (4.29, 4.88)RAI‐bipolar1.25 (1.10, 1.41) 0.2834 (0.2499, 0.3169) 0.9182 (0.9128, 0.9236) Schizophrenia2.76 (2.54, 3)RAI‐schizophrenia2.01 (1.82, 2.21) 0.7380 (0.7069, 0.7692) 0.9756 (0.9726, 0.9786) OntarioDepression26.22 (26.03, 26.41)RAI‐depression23.67 (23.49, 23.85) 0.2422 (0.2383, 0.2462) 0.4320 (0.4287, 0.4353) Depression‐Rx39.17 (38.94, 39.4)RAI‐depression23.67 (23.49, 23.85) 0.3178 (0.3143, 0.3212) 0.3956 (0.3922, 0.399) Anxiety23.82 (23.64, 24)RAI‐anxiety11.41 (11.29, 11.54) 0.1734 (0.1694, 0.1773) 0.5074 (0.5042, 0.5106) Anxiety‐Rx34.93 (34.71, 35.15)RAI‐anxiety11.41 (11.29, 11.54) 0.1503 (0.1471, 0.1535) 0.3479 (0.3445, 0.3514) Bipolar3.01 (2.95, 3.07)RAI‐bipolar1.74 (1.70, 1.79) 0.4640 (0.4535, 0.4746) 0.9502 (0.9490, 0.9513) Schizophrenia3.00 (2.94, 3.06)RAI‐schizophrenia2.00 (1.95, 2.06) 0.6568 (0.6476, 0.6659) 0.9665 (0.9655, 0.9674) Abbreviations: PABAK = prevalence and bias‐adjusted kappa; RAI = Resident Assessment Instrument; Rx = case definition that includes prescription claims. Agreement between mental illness measures from administrative data and RAI diagnoses. 0.3349 (0.3182, 0.3515) 0.5303 (0.5166, 0.544) 0.3610 (0.3460, 0.3761) 0.4997 (0.4857, 0.5137) 0.3011 (0.2827, 0.3196) 0.5812 (0.5681, 0.5944) 0.2564 (0.2408, 0.2721) 0.4539 (0.4395, 0.4683) 0.3093 (0.2712, 0.3475) 0.9124 (0.9057, 0.9190) 0.7708 (0.7327, 0.8089) 0.9822 (0.9792, 0.9852) 0.3080 (0.3040, 0.3120) 0.5660 (0.5632, 0.5689) 0.3889 (0.3852, 0.3926) 0.5670 (0.5641, 0.5698) 0.2484 (0.2443, 0.2524) 0.5691 (0.5662, 0.5719) 0.2313 (0.2278, 0.2349) 0.4597 (0.4566, 0.4627) 0.4711 (0.4606, 0.4815) 0.9558 (0.9548, 0.9568) 0.6373 (0.6272, 0.6474) 0.9733 (0.9726, 0.9741) 0.2914 (0.2835, 0.2992) 0.3439 (0.3361, 0.3517) 0.3320 (0.3251, 0.3390) 0.3250 (0.3171, 0.3329) 0.2487 (0.2396, 0.2577) 0.5413 (0.5343, 0.5483) 0.2126 (0.2050, 0.2201) 0.4081 (0.4005, 0.4157) 0.1990 (0.1863, 0.2118) 0.8162 (0.8114, 0.8210) 0.6252 (0.6041, 0.6463) 0.9629 (0.9607, 0.9651) 0.3149 (0.3014, 0.3283) 0.4446 (0.4324, 0.4569) 0.348 (0.3365, 0.3594) 0.3916 (0.379, 0.4042) 0.3295 (0.3131, 0.3458) 0.6317 (0.6211, 0.6423) 0.2765 (0.2628, 0.2903) 0.5026 (0.4908, 0.5144) 0.2834 (0.2499, 0.3169) 0.9182 (0.9128, 0.9236) 0.7380 (0.7069, 0.7692) 0.9756 (0.9726, 0.9786) 0.2422 (0.2383, 0.2462) 0.4320 (0.4287, 0.4353) 0.3178 (0.3143, 0.3212) 0.3956 (0.3922, 0.399) 0.1734 (0.1694, 0.1773) 0.5074 (0.5042, 0.5106) 0.1503 (0.1471, 0.1535) 0.3479 (0.3445, 0.3514) 0.4640 (0.4535, 0.4746) 0.9502 (0.9490, 0.9513) 0.6568 (0.6476, 0.6659) 0.9665 (0.9655, 0.9674) Abbreviations: PABAK = prevalence and bias‐adjusted kappa; RAI = Resident Assessment Instrument; Rx = case definition that includes prescription claims. Prevalence of mental illness in (A) home care (B) long‐term care population according to measure used, stratified by province. Alberta did not use interRAI‐HC during study period, thus excluded from home care analyses for agreement. All analyses in home care setting in Manitoba and Ontario used only the interRAI‐HC. RAI = Resident Assessment Instrument Confidence intervals in Table 2. Agreement Considering the prevalence and bias‐adjusted kappa, agreement was generally moderate for the depression and anxiety diagnoses in Manitoba and Ontario in the HC setting (Table 2). Agreement was almost perfect for bipolar disorder and schizophrenia diagnoses in all provinces. Findings were similar in LTC except that agreement was only fair for depression (Table 2). When we compared administrative definitions to RAI‐diagnoses as the reference standard, performance of the administrative definitions varied substantially by diagnosis (Supporting Information S1: Table e8‐e9). In the HC setting, sensitivity of the depression definition compared to the RAI‐depression was moderate in Manitoba but fair in Ontario (Supporting Information S1: Table e8). Adding prescription claims to the depression definition improved sensitivity by 13%–17% with minimal (< 5%) loss of specificity. We similarly observed relatively modest sensitivity but good specificity for the anxiety definition without prescription claims; the addition of prescription claims to the definition effected a more modest improvement in sensitivity with nearly double the loss of specificity. In Manitoba and Ontario, sensitivity was higher for bipolar disorder (0.74 and 0.61, respectively) and schizophrenia (0.87 and 0.74, respectively) definitions with specificities exceeding 0.96. Findings were broadly similar in the LTC setting to those observed in the HC setting (Supporting Information S1: Table e9) although the loss of specificity associated with case definitions including drug claims was slightly larger (7%–9%). Alberta's findings more closely resembled those in Manitoba for sensitivity than for Ontario except for bipolar disorder where Alberta had the highest sensitivity. Factors Associated With Discordance Home Care In HC, several factors were associated with increased odds of discordance for all mental illnesses examined in Manitoba and Ontario including being younger (18–64, 65–74 years) versus older (85+ years), having dementia, and a higher number of non‐mental health physician visits (although confidence intervals included 1 for bipolar and schizophrenia in Manitoba and for schizophrenia in Ontario) (Figure 2A/B, Tables e10‐e11). Male sex was associated with decreased odds of discordance for all mental illnesses in both provinces. Higher morbidity level was associated with discordance only in Ontario, where higher morbidity was associated with increased odds of discordance for depression but decreased odds for bipolar disorder and schizophrenia. Forest plot: Association of age, sex, dementia, morbidity, and health care use with discordance between administrative and RAI measures of mental illness in home care: (A) Depression/Anxiety (B) Bipolar disorder/Schizophrenia. HC = Home care, MB = Manitoba, ON = Ontario. Long‐Term Care In LTC, findings were similar for age, male sex, dementia as for HC (Supporting Information S1: Figure e1, Table e12‐e14) although confidence intervals included 1 for the association of dementia with schizophrenia in Manitoba. A higher number of non‐mental health visits was associated with increased odds of discordance for all mental illnesses in all provinces; the confidence interval included 1 for schizophrenia in Manitoba. Associations with morbidity and discordance were inconsistent. In Alberta, higher morbidity was associated with increased odds of discordance for bipolar disorder but decreased odds for schizophrenia. In Manitoba, higher morbidity was associated with increased odds of discordance for depression only. In Ontario, higher morbidity was associated with increased odds of discordance for depression and anxiety and decreased odds for schizophrenia. Discussion In this retrospective cross‐sectional study of HC and LTC populations in three Canadian provinces, we compared prevalence estimates and agreement between mental illness status as identified using administrative case definitions and RAI‐diagnoses. Mental illness prevalence estimates varied across provinces, consistent with interprovincial differences reported by the Canadian Community Health Survey (Palay et al. 2019). As anticipated, we found that depression was the most common mental illness in HC and LTC (Poss et al. 2021; Seitz et al. 2010). Prevalence estimates based on administrative data were consistently higher than those based on RAI‐diagnoses. Agreement measured using administrative and RAI data was moderate for depression and anxiety disorders, and excellent for bipolar disorder and schizophrenia; findings regarding concordance were consistent across provinces. Younger age, dementia, and more physician visits were generally associated with increased likelihood of discordance between data sources whereas male sex was associated with a lower likelihood. Administrative definitions developed and validated in the general population might not perform well in HC or LTC populations due to differences in population characteristics, morbidity burden, polypharmacy and patterns of health care use. However, our findings provide some reassurance about applying administrative definitions in these settings. We found excellent agreement between data sources for schizophrenia and bipolar disorder; however, agreement was moderate and more variable for depression and anxiety. Some discordance between data sources was expected based on differences in the underlying constructs. While the administrative definitions used were validated against medical records, self‐reported diagnoses, and/or structured diagnostic interviews (Marrie et al. 2013, 2016), they capture individuals seeking health services and are affected by limitations in access to care, diagnosis, recognition and coding of mental illness within the health system; thus, more severe mental illnesses are more likely to be captured. Non‐specific 3‐digit ICD codes in physician claims makes it difficult to discriminate depression and anxiety (Kisely, Lin, Gilbert, et al. 2009). Bipolar disorder and schizophrenia are more specific diagnoses with more specific treatments whereas medications used for depression and anxiety may overlap with those used for other conditions. Also, depression and anxiety disorders relapse and remit, unlike bipolar disorder and schizophrenia. Collectively, this may account for higher agreement between data sources for bipolar disorder and schizophrenia. Treatment from non‐physician mental health providers is not captured. In contrast, RAI assessments are intended to be completed by a trained assessor using all available information sources. Unlike prior studies comparing RAI‐diagnoses and administrative data which used only hospital data, we used validated case definitions that combined hospital, physician and sometimes prescription claims. Nonetheless, our observations are consistent with prior studies. A Manitoba study comparing administrative data to RAI‐diagnoses for physical conditions found that agreement was variable across conditions, and that RAI‐based prevalence estimates were lower than those from administrative data (Lix et al. 2014). An Ontario study of 80,664 individuals admitted to complex continuing care hospitals from acute care hospitals between 1997 and 2005 (Wodchis et al. 2008) found that sensitivity of the RAI‐MDS for mental illness discharge diagnoses varied by condition, ranging from 67% for anxiety when it was the primary diagnosis and 81% for depression; sensitivity was lower when the mental illness was not primary (Wodchis et al. 2008). A United States Medicare claims study found a sensitivity of 78% and specificity of 83% for RAI‐diagnoses of depression but PPV of only 2% (Mor et al. 2011). Younger age, more physician visits, and dementia were consistently associated with increased likelihood of discordance between data sources, but male sex was associated with a decreased likelihood of discordance, which may reflect several factors. Women aged 75 years and older use more health care services, increasing the opportunity for recording of mental illness in administrative data, although men in HC in Ontario have more comorbidity, impairment, and hospitalizations, and are more likely to be admitted to LTC (Gruneir et al. 2013). While men are more likely to have atypical presentations, they are also more likely to have a partner who could advocate to facilitate recognition and diagnosis; this warrants further evaluation. Younger age and dementia are associated with increased likelihood of psychiatry visits in HC and LTC (Poss et al. 2021). A prior study in HC and complex continuing care settings suggested that depressive symptoms were less common at older ages, but that undertreatment occurred more often at older ages (Szczerbińska et al. 2012). Evaluating mental illness may be more difficult in individuals with dementia who have less ability to report symptoms, and whose dementia symptoms may overlap those of mental illness (Canadian Coalition for Seniors’ Mental Health 2024). Our findings regarding the factors associated with discordance highlight the complexity of effective surveillance among all sub‐populations and point to the need for further work to determine if surveillance strategies should be tailored based on age and specific morbidities. Strengths of this work include the use of large population‐based cohorts from three Canadian provinces with differing geography, population sizes and health delivery organization. We examined a 10‐year period, reducing variability due to short‐term changes in health system demands or health service delivery. Limitations of the study include differences in the measurement time frame of the administrative case definitions (5 years, which would have increased detection compared to shorter intervals) and the RAI (diagnoses at assessment). Although diagnoses from the RAI are supposed to be derived from clinical assessments and medical records, their accuracy as compared to gold standard semi‐structured psychiatric interviews is unknown and reliability is acceptable but imperfect. The prevalence of mental illness is high in HC and LTC settings across Canada. RAI‐diagnoses provide lower prevalence estimates than administrative data. Since administrative data reflect treatment of conditions, the lower prevalence estimates based on RAI‐diagnoses, and the variation in the degree of disparity between data sources by demographic and clinical characteristics indicate further evaluation of the RAI as a stand‐alone means of mental health surveillance is needed including concurrent comparison of the RAI‐diagnoses and administrative definitions to a gold standard clinical interview, and evaluation of the stability of prevalence estimates over time. Author Contributions Ruth Ann Marrie: conceptualization, funding acquisition, methodology, investigation, supervision, visualization, writing – original draft, writing – review and editing. James M. Bolton: conceptualization, funding acquisition, supervision, writing – review and editing. Xueyi Chen: data curation, writing – review and editing, formal analysis, investigation. Erind Dvorani: methodology, writing – review and editing. Okechukwu Ekuma: data curation, formal analysis, writing – review and editing. Zahra Goodarzi: conceptualization, methodology, funding acquisition, writing – review and editing. Andrea Gruneir: conceptualization, funding acquisition, writing – review and editing. Jin Luo: data curation, formal analysis, writing – review and editing. Nathan Herrmann: funding acquisition, writing – review and editing. Matthias Hoben: funding acquisition, writing – review and editing. David B. Hogan: funding acquisition, writing – review and editing. Andrea Iaboni: funding acquisition, writing – review and editing. Dallas Seitz: funding acquisition, writing – review and editing. Randy Walld: writing – review and editing, data curation, formal analysis. Walter P. Wodchis: funding acquisition, writing – review and editing. Colleen J. Maxwell: conceptualization, methodology, supervision, funding acquisition, investigation, project administration, writing – review and editing, validation. Funding This study was funded by the Public Health Agency of Canada (2425‐HQ‐000095). Ethics Statement We obtained approval from the University of Alberta Health Research Ethics Board (pSite00000181), University of Calgary Conjoint Health Research Ethics Board (#REB24‐0793), University of Manitoba Health Research Ethics Board (HS26489) and University of Waterloo Human Research Ethics Board (#46451). We obtained approval for data access from Health Shared Services (Alberta), Provincial Health Research Privacy Committee (Manitoba), and ICES (Ontario). Data access (via ICES) was authorized under Section 45 of Ontario's Personal Health Information Protection Act. Consent The authors have nothing to report. Conflicts of Interest R.A. Marrie receives research funding from: CIHR, MS Canada, National Multiple Sclerosis Society, CMSC, The Arthritis Society, US Department of Defence, UK MS Society, Public Health Agency of Canada, Brain Canada, MMSF, CHRIM, Pfizer Foundation and is a co‐investigator on studies receiving funding from Biogen Idec, Roche Canada (no funds to her/her institution). She is supported by the Gillian's Hope Multiple Sclerosis Clinical Research Chair (Dalhousie University). J. Bolton has research funding from the Public Health Agency of Canada. Z. Goodzhari has research funding from the Public Health Agency of Canada. A. Gruneir receives research funding from the Public Health Agency of Canada. D.B. Hogan receives research funding from the Public Health Agency of Canada. C.J. Maxwell receives research funding from CIHR, MS Canada, National Multiple Sclerosis Society, CMSC and the Public Health Agency of Canada. All other authors declare no conflicts of interest. Supporting information Supporting Information S1 Acknowledgements The views expressed herein do not necessarily represent the views of the Public Health Agency of Canada. In Ontario, this study was supported by ICES, which is funded by an annual grant from the Ontario Ministry of Health (MOH) and the Ministry of Long‐Term Care (MLTC). We thank IQVIA Solutions Canada Inc. for use of their Drug Information File. This document used data adapted from the Statistics Canada Postal CodeOM Conversion File, which is based on data licensed from Canada Post Corporation, and/or data adapted from the Ontario Ministry of Health Postal Code Conversion File, which contains data copied under license from Canada Post Corporation and Statistics Canada. Parts of this material are based on data adapted from Statistics Canada, Census, 2016, 2021. Parts of this material are based on data and/or information compiled and provided by CIHI and the Ontario MOH. The analyses, conclusions, opinions and statements expressed herein are solely those of the authors and do not reflect those of the Public Health Agency of Canada, the Ontario MOH, MLTC or Statistics Canada; no endorsement is intended or should be inferred. In Alberta, this study was supported by Provincial Research Data Services (PRDS) which is funded by the Alberta SPOR SUPPORT Unit (AbSPORU), the Alberta VITAL project and Health Shared Services. AbSPORU is funded by CIHR, Alberta Innovates, University Hospital Foundation, University of Alberta and University of Calgary. VITAL is funded by Innovation, Science and Economic Development (ISED) Canada. Data were extracted from the Health Shared Services Enterprise Data Warehouse. This study is based in part on data provided by Preventative and Primary Health Services and Health Shared Services. The interpretation and conclusions contained herein are those of the researchers and neither the Government of Alberta nor any provincial health agencies/corporations expressed any opinion in relation to this study. In Manitoba, data used in this study are from the Manitoba Population Research Data Repository housed at the Manitoba Centre for Health Policy, University of Manitoba and were derived from data provided by Manitoba Health and Shared Health. The authors acknowledge the Manitoba Centre for Health Policy for use of data contained in the Manitoba Population Research Data Repository under project #2024–056 (PHRPC# P2024‐46). The results and conclusions are those of the authors and no official endorsement by the Manitoba Centre for Health Policy, Manitoba Health, or other data providers is intended or should be inferred. The authors also acknowledge other members of the EMHLTH‐CC Research Team: Karl Everett MSc, ICES, Toronto, Canada; Liisa Jaakkimainen MD, MSc, Family & Community Medicine, University of Toronto & ICES, Toronto, Canada; Kimberley Wilson PhD, Department of Family Relations & Applied Nutrition, University of Guelph, Guelph, Canada; and Erik Youngson MMath, Provincial Research Data Services, Health Shared Services and Data and Research Services, Alberta SPOR SUPPORT Unit, Edmonton, Canada. Data Availability Statement The study datasets are held securely in coded form at the respective provincial data centres (ICES in ON; Health Shared Services Enterprise Data Warehouse in AB, and the Manitoba Centre for Health Policy in MB). Legal data sharing agreements prohibit each data centre from making the dataset publicly available. The source data for MB are not owned by the researchers or MCHP and as such cannot be provided to a public repository. With the appropriate approvals, the MB data can be accessed through the Manitoba Centre for Health Policy. The ON dataset from this study is held securely in coded form at ICES. While data sharing agreements prohibit ICES from making the data set publicly available, access may be granted to those who meet pre‐specified criteria for confidential access, available at www.ices.on.ca/DAS. The full dataset creation plan and underlying analytic code are available from the authors upon request, understanding that the computer programs may rely upon coding templates or macros that are unique to each provincial data centre and are therefore either inaccessible or may require modification. References Associated Data Supplementary Materials Supporting Information S1 Data Availability Statement The study datasets are held securely in coded form at the respective provincial data centres (ICES in ON; Health Shared Services Enterprise Data Warehouse in AB, and the Manitoba Centre for Health Policy in MB). Legal data sharing agreements prohibit each data centre from making the dataset publicly available. The source data for MB are not owned by the researchers or MCHP and as such cannot be provided to a public repository. With the appropriate approvals, the MB data can be accessed through the Manitoba Centre for Health Policy. The ON dataset from this study is held securely in coded form at ICES. While data sharing agreements prohibit ICES from making the data set publicly available, access may be granted to those who meet pre‐specified criteria for confidential access, available at www.ices.on.ca/DAS. The full dataset creation plan and underlying analytic code are available from the authors upon request, understanding that the computer programs may rely upon coding templates or macros that are unique to each provincial data centre and are therefore either inaccessible or may require modification.