Background

A major Quebec mental health reform was implemented between 2005 and 2015, with the primary aim of consolidating primary care services and developing within each network one-stop points of entry to services offered in community healthcare centres and specialised care.¹⁴ In Quebec, general practitioners (GPs) are the main primary care mental health providers.¹⁵ Some 65% work in family medicine groups, where patient registration is required and care is delivered in collaboration with nurses or social workers. Family medicine groups also offer extended office hours. Mental health teams providing individual or group psychosocial interventions were also created within the reform of community healthcare centres. Specialised mental healthcare is delivered mainly in hospital settings (e.g. emergency department and in-patient services). However, to receive specialised out-patient care at hospitals, patients had to be referred from a one-stop point of entry in the primary care network. Mental healthcare is also complemented by psychologists working in private clinics or by community-based organisations offering crisis, respite and peer support services. Services for substance-related disorders are available from specialised addiction treatment centres outside the mental health system.¹⁶ The Quebec healthcare system is divided into four types of healthcare regions: university regions that include university and psychiatric hospitals offering ultra-specialised care, teaching and research facilities; remote regions with little specialised care; and peripheral and intermediary regions that offer some mental health coverage, as they are located relatively close to the university regions.¹⁶

Study population and design

In total, 9514 patients diagnosed with a mental disorder or substance-related disorder who used one of six Quebec emergency department in 2014–2015 (index (first) emergency department visit; fiscal year: April 1–March 31) were identified through clinical administrative databases for this 3-year study. The cohort excluded patients who were hospitalised at the time of their index emergency department visit (n = 613, 6%) or during the ≤30-day follow-up period. Participants had to be aged ≥12 years and eligible for the Quebec Health Insurance Plan (Régie de l'assurance maladie du Québec (RAMQ)). They had to receive a diagnosis of a mental disorder or substance-related disorder at least once in the 2 years (2012–2013 to 2013–2014) before their index emergency department visit or at the index visit in 2014–2015. Figure 1 presents the conceptual framework for the study, which links each variable to the databases used. As stated earlier, predictors of 30-day out-patient physician follow-up after emergency department use were measured in terms of patient clinical characteristics 2 years before index emergency department use; patient sociodemographic characteristics at the index emergency department visit in 2014–2015 and service use in the 12 months before the index emergency department visit. However, some clinical and service use variables were measured at the index emergency department visit itself (e.g. reason for emergency department use, illness acuity and referral after emergency department discharge).

Fig. 1 Conceptual framework of predictors for patients with mental disorders with or without out-patient physician follow-up within 30 days after emergency department discharge in 2014–2015.

The authors assert that all procedures contributing to this work comply with the ethical standards of the relevant national and institutional committees on human experimentation and with the Helsinki Declaration of 1975, as revised in 2008 (approval number: IUSMD-20-26). All procedures involving human patients were approved by the ethics committee of a mental health university institute. As this study was based on clinical administrative data, individual patient consent was not required, but the study was authorised by the Quebec Access to Information Commission.

Data sources

Clinical administrative data, including demographic and socioeconomic information, data on hospital admission and emergency department use, and public psychosocial interventions delivered from community healthcare centres, were collected from the RAMQ databases. RAMQ integrates billing systems for most physician services in Quebec. Only 6% of billing occurred outside the public system in 2016–2017.^{Reference Simard, Sirois and Candas17} Data were merged annually for each patient, and all databases were integrated with a unique RAMQ identifier. Diagnostic codes are shown in Table 1, and were based on the ICD-9 and the Canadian version of the ICD-10 (ICD-10-CA). Patient diagnoses were identified by selecting those with a mental disorder diagnosed more than once, those diagnosed during a hospital admission or those with a diagnosis by the patient's physician (particularly a psychiatrist). Chronic physical illnesses are registered in the RAMQ database twice yearly, or once in hospital databases.^{Reference Blais, Jean, Sirois, Rochette, Plante and Larocque18}

Table 1 Codes for mental disorders including substance-related disorders and chronic physical illnesses, according to the ICD-9 and ICD-10-CA

a. All diagnoses identified in the Régie de l'Assurance Maladie du Québec (Quebec Health Insurance Plan or database) were based on the ICD-9, which included a four-digit code. The ICD-10-CA was used for the Maintenance et Exploitation des Données pour l'étude de la Clientèle Hospitalière (MED-ECHO; hospitalisation database) and Banque de Données Commune des Urgences (emergency department use database). Diagnoses related to all the above databases were considered, and all data were integrated each year for each patient. The MED-ECHO is the only database including several diagnoses, both primary and secondary diagnoses. In this database, mental disorders were considered only when listed as the primary diagnosis, but substance-related disorders as primary or secondary diagnoses were included, as these are often underdiagnosed.

b. The list of chronic physical illnesses is based on an adapted and validated version of the Elixhauser Comorbidity Index, integrating the Charlson Index, which consists of 32 major categories of physical illnesses (see ‘Method’ section). In this list of chronic physical illnesses, three categories of mental disorders and two of substance-related disorders (identified with *) were also included in the list of co-occurring mental and substance-related disorders, thus appearing twice.

Variables

The dependent variable was dichotomous, referring to patients with or without 30-day out-patient physician follow-up after index emergency department use (2014–2015). Prompt emergency department follow-up could be provided by any physician (either a GP or psychiatrist) working at the hospital, or from other out-patient services in the community. Regarding independent variables, patient clinical characteristics included types of mental disorder or substance-related disorder; having chronic physical illnesses or co-occurring disorders (mental disorder, substance-related disorder, chronic physical illness); reasons for index emergency department use, including suicidal ideation, suicide attempt, violence or social issues; and illness acuity (triage priority levels related to index emergency department use). Mental disorders comprised common mental disorders (anxiety, depressive or adjustment disorders), serious mental disorders (schizophrenia spectrum and other psychotic disorders, or bipolar disorders) and personality disorders. Substance-related disorders referred to alcohol- and drug-related disorders (use or induced disorders, intoxication, withdrawal). Based on an adapted version of the Elixhauser Comorbidity Index,¹⁹ severity levels of chronic physical illnesses (scored 0–3) were recorded. Illness acuity, or triage priority level at index emergency department use, was ranked as levels 1–2 (immediate or very urgent care), 3 (urgent care) or 4–5 (less urgent or non-urgent care), based the Canadian Triage Acuity Scale.¹⁹ Emergency department triage nurses well trained in identifying reasons for emergency department use and in determining triage priority recorded this information, providing reliable data.^{Reference Rahme, Low, Lamarre, Daneau, Habel and Turecki20}

Sociodemographic characteristics included age group, gender, Material and Social Deprivation Indices, and healthcare region. Based on the smallest dissemination areas established for the 2011 Canadian census, the Material Deprivation Index was derived from the ratio of population employment, average income and number of individuals without a high school diploma; the Social Deprivation Index measured proportions of individuals living alone, single-parent families and those without a spouse.^{Reference Pampalon, Hamel, Gamache and Raymond21} Both indices were classified in five quintiles, not including unassigned areas (e.g. homelessness), with the fifth quintile representing highest level of deprivation. Healthcare regions for the study included university and peripheral regions.

Service use variables for the 12 months before index emergency department use included: having a family physician; usual physician (GP only, psychiatrist only, both GP and psychiatrist, or none); number of consultations with usual GP or usual psychiatrist; continuity of physician care; number of psychosocial interventions in community healthcare centres, excluding GP consultations; and prior emergency department use or hospital admissions for mental health reasons. Referrals at discharge from index emergency department visit were also assessed, whether to a GP, community healthcare centres or community-based services, other health organisations or none. Patients with a usual GP needed to have at least two consultations with the same GP or with at least two GPs working in the same family medicine group. Usual GP was considered a proxy for a family physician.^{Reference Rodriguez, Marshall, Rogers and Safran22} To qualify as having a usual psychiatrist, patients needed to have at least two consultations with this provider, or only one consultation for those who had a usual GP, which was considered a proxy for collaborative care.^{Reference Dreiher, Comaneshter, Rosenbluth, Battat, Bitterman and Cohen23} Continuity of physician care was measured with the Usual Provider Continuity Index, which describes the proportion of visits to the usual GP and psychiatrist of all visits made to GPs and psychiatrists in out-patient care, including walk-in clinics.^{Reference Breslau and Reeb24} A score of <0.50 was considered low continuity of physician care, 0.51–0.79 was considered moderate and ≥0.80 was considered high.^{Reference Ionescu-Ittu, McCusker, Ciampi, Vadeboncoeur, Roberge and Larouche25}

Statistical analyses

Descriptive and bivariate analyses were performed on the sample and included two-way frequency tables. Since the clustering effect (106 hospital units) was small (intraclass correlation coefficient: 0.018), a multilevel model was not needed. Collinearity statistics were tested with variance inflation factors (VIFs) and tolerance tests, with 5 as the maximum level of VIF. Independent variables without collinearity were entered into a logistic hierarchical regression model, with alpha set at <0.10. The reference category for the dependent variable was patients without prompt emergency department follow-up (i.e., no out-patient physician follow-up, whether GP or psychiatrist, within 30 days of index emergency department use). Independent variables were selected based on previous studies and entered by blocks as follows: patient clinical characteristics first, then patient sociodemographic characteristics, followed by service use variables. All possible combinations of blocks of variables were entered and tested, with no impact on the final model or in the identification of significant variables. A forward model selection was also used to enter variables into the hierarchical logistic regression model for the estimation of parameters. Akaike's Information Criterion (AIC) was used for selection of the model, with choice of the final multivariate model based on the smallest AIC. Odds ratios were calculated, with 95% confidence intervals. All analyses were performed with software SPSS version 24.0 for Windows.