Study design, setting, and data sources

We conducted a population-based cohort study in Ontario, Canada using linked clinical and health administrative databases. These datasets were linked using unique encoded identifiers and analyzed at ICES (formerly the Institute for Clinical Evaluative Sciences). These datasets are commonly used to study end-of-life care [11,15](S2 Table).

Ontario has over 10 million adults, making it Canada’s largest province that accounts for nearly 40% of its national population. Ontario residents have access to medically necessary services via the Ontario Health Insurance Plan (OHIP), and those aged ≥65 years receive prescription drug insurance coverage via the Ontario Drug Benefit program [21].

ICES is an independent, non-profit research institute whose legal status under Ontario’s health information privacy law allows it to collect and analyze health care and demographic data, without consent, for health system evaluation and improvement. The use of the data in this project is authorized under section 45 of Ontario’s Personal Health Information Protection Act (PHIPA) and does not require review by a Research Ethics Board.

Study cohort

Our study cohort included all Ontario adults (aged ≥18 years) in their last 90 days of life who died between January 25, 2018, and December 31, 2021. People were excluded from the study if on the 90^th day prior to death they were missing data on age or sex; aged <18 or >105 years; non-residents of Ontario; not eligible for OHIP for a period of ≥3 months in the previous year; did not have at least one healthcare encounter in the prior 10 years; resided in a nursing home; or were hospitalized during the entire period of follow-up between index and death date as they would not be eligible to receive virtual end-of-life care. Decedent cohorts are frequently used to study end-of-life care practices; by setting a standard lookback period in relation to death they do not rely on physician prognostication of survival [15,22,23].

Use of Virtual End-of-Life Care (Exposure)

We measured exposure to virtual end-of-life care during the 90 days prior to death using a distinct set of specialized virtual care physician fee codes (S1 Table). People who received at least 1 virtual end-of-life care visit were assigned to the exposed group and their first virtual end-of-life care visit was the index study date. People who did not receive any virtual care visits in the last 90 days of life were assigned to the unexposed group and were randomly assigned an index study date based on the distribution of index dates from the exposed group. Prior to March 14, 2020, healthcare providers were reimbursed by OHIP for delivering virtual end-of-life care via a single telephone palliative care fee code. Providers could also be reimbursed through Ontario Telemedicine Network’s (OTN) distinct set of video-based fee codes, which required both the physician and patient to attend an authorized virtual care centre in person (S1 Table). On March 14, 2020, the Ontario Government introduced an additional set of specialized telephone and video-based provider fee codes to enable broader delivery of virtual care during the COVID-19 pandemic due to the restrictions on in-person clinical care. These codes removed the previous requirement for both parties to attend the appointment from an authorized virtual care centre and permitted the freedom to attend these appointments in any location using either the telephone or by video (S1 Table).

People were assigned to the “pre-March 14” or “post-March 14” intervention group based on their index date and date of death in relation to March 14, 2020 that allowed a focus on the environment in which someone received the majority of their care during the last 90 days of life. A person whose index and death dates both fell before March 14, 2020 was assigned to the pre-March 14 group, while a person whose index and death dates both fell after March 14, 2020 was assigned to the post-March 14 group. People whose index date came before, and death date came after March 14, 2020 were assigned to the pre-March 14 intervention group if their death occurred <45 days after March 14, 2020, and the post-March 14 intervention group if their death occurred ≥45 days after March 14, 2020.

Patient characteristics

We measured multiple validated demographic and clinical variables at the index date including age, sex, socioeconomic status, surname-based ethnicity [24], rurality of residence, multiple chronic conditions [25–34], and Local Health Integrated Network (LHIN, now called the Home and Community Care Support Services [35]) organization based on residency. There are 14 LHINs/Home and Community Care Support Services organizations in Ontario that are responsible to plan and distribute provincial funding for regional public healthcare services. We used a 5-year look back period from index date to measure prevalent comorbidities and the hospital frailty risk score [36]. We recorded each person’s year and location of death, use of acute healthcare services and engagement with palliative care (using distinct provider fee codes[11,15,37,38]) in the one year prior to study index date, as well as the receipt of homecare in the two years prior to study index date. Homecare services are allocated based on needs identified by a homecare case manager, and may include longitudinal care with visits from allied healthcare providers including personal support workers, nurses, physiotherapists, occupational and speech-language therapists. These patient characteristics were measured because they were potential confounders between the association of virtual end-of-life care receipt and other acute health service use and were available in the ICES datasets.

Outcomes

The primary outcome was emergency department visits or hospitalization for any cause in the last 90 days of life and after the index date. The secondary outcome was the location of death, categorized as occurring in- or out-of-hospital. Location of death was determined based on the database in which the death was registered (S2 Table).

Statistical analysis

Multivariable modified Poisson regression measured the association of use of virtual end-of-life care with number of emergency department visits and hospitalizations in the period after the index date to death, where the logarithm of the follow-up period was used as an offset. It was also used to measure the association between virtual end-of-life care and out-of-hospital death. Use of modified Poisson regression allows estimation of relative rate which is more interpretable than odds ratios as well as including a robust standard error to account for overdispersion [39,40]. All models were adjusted for deciles of age, sex, neighbourhood income quintile, rural residence, LHIN, prevalent comorbidities, hospital frailty score, acute healthcare system use in the year prior to index date, and the receipt of homecare in the two years prior to index date. All analyses were performed using SAS version 9.4 (SAS Institute, Cary, North Carolina).

Patient and public involvement

Patient and caregiver partners assisted in the development of the research questions and interpretation of the main findings.

Characteristics of the study cohort

The final study cohort included 323,995 adults; 176,687 people died before March 14, 2020, and 147,308 people died after (Fig 1). Prior to March 14, 2020, 13,974 people (7.9%) received virtual end-of-life care and 104,165 people (71%) received virtual end-of-life care after. The characteristics of people in the pre- and post-March 14 period are presented in Tables 1 and 2.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Creation of the study cohort.

https://doi.org/10.1371/journal.pdig.0000463.g001

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Baseline characteristics of patients during their last 90 days of life, pre-March 14, 2020.

https://doi.org/10.1371/journal.pdig.0000463.t001

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Baseline characteristics of patients during their last 90 days of life, post-March 14, 2020.

https://doi.org/10.1371/journal.pdig.0000463.t002

The characteristics of people receiving virtual end-of-life care changed in the pre- to post-March 14, 2020 period. Compared to people who did not receive virtual end-of-life care, people who received virtual end-of-life care and died prior to March 14, 2020 were generally younger, had more emergency department visits and hospitalizations in the last year. A higher proportion lived in a rural area, lived in the North East and North West LHINs, had cancer and alcohol/substance use disorders, and received palliative care in the prior year compared to people who did not receive virtual end-of-life care. A lower proportion of people who received virtual end-of-life care had heart failure, dementia, and hypertension, and lived in the Central and Toronto Central LHINs.

In contrast, after March 14, 2020, people who received virtual end-of-life care were generally older and had more hospitalizations in the last year than those who did not receive virtual end-of-life care. A lower proportion lived in a rural area, lived in the North East or North West LHINs, lived in a neighbourhood in the bottom income quintile, or had alcohol/substance use disorders, compared to people who did not receive virtual end-of-life care. A higher proportion of people who received virtual end-of-life care received palliative care in the prior year, homecare in the prior two years, and had heart failure, diabetes, cancer, end-stage renal disease, and hypertension, and lived in the Central, Central West, and Mississauga Halton LHINs (Tables 1 and 2).

Virtual End-of Life care and acute healthcare use

Prior to March 14, 2020, the use of virtual end-of-life care was associated with a 16% higher rate of emergency department use (Adjusted Relative Rate (aRR) 1.16, 95%CI 1.12 to 1.20) and a 17% higher rate of hospitalization (aRR 1.17, 95%CI 1.15 to 1.20) in the last 90 days of life, compared with people who did not receive virtual end-of-life care (Fig 2 and Table 3). After March 14, 2020, its use was associated with a 58% higher rate of emergency department use (aRR 1.58, 95%CI 1.54 to 1.62) and a 45% higher rate of hospitalization (aRR 1.45, 95%CI 1.42 to 1.47) in the last 90 days of life, compared with people who did not receive virtual end-of-life care (Fig 2 and Table 3).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Forest plot of the adjusted relative risk of the association between the receipt of virtual end-of-life care and health service use (emergency department visits and hospitalizations) and out-of-hospital death for the pre-March 14, 2020 (blue squares) and post-March 14, 2020 (red circles) groups.

https://doi.org/10.1371/journal.pdig.0000463.g002

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 3. Association between the receipt of virtual end-of-life care and health service use (emergency department visits and hospitalizations) and out-of-hospital death §.

https://doi.org/10.1371/journal.pdig.0000463.t003

Location of death

Receiving virtual end-of-life care before March 14, 2020 was associated with a 34% higher risk of an out-of-hospital death (aRR 1.34, 95%CI 1.31 to 1.37) (Fig 1 and Table 3). Receiving virtual end-of-life care after March 14, 2020 was associated with a 65% higher risk of an out-of-hospital death (aRR 1.65, 95%CI 1.61 to 1.69) (Fig 2 and Table 3).

Limitations

The use of administrative data limits our ability to measure patient preferences for location of care, modes of care delivery including virtual end-of-life care and location of death, which are essential components of high-quality end-of-life care. However, prior evidence reported that most people generally prefer to die at home [10]. Additionally, it is important to reiterate that these fee codes are not exclusively palliative in nature, and our definition of virtual end-of-life care is based on care provided in the last 90 days of life. The use of a decedent study design may introduce selection bias given that one may not know whether a potential recipient of virtual care would be alive after 90 days. However, decedent cohorts are frequently used to study end-of-life care practices and delivery [15,45]. They offer advantages such as not relying on physician prognostication of survival and instead setting a standard lookback period in relation to death [23]. Additionally, this study did not exclude specific causes of death and therefore should be representative of broad disease categories with varying levels of healthcare use required in the last 90 days of life. Death also serves as a common denominator that marks the seriousness of disease and avoids decision making about stopping follow up of a prospective cohort study [23]. Second, this data does not include information on the cause of death (including COVID-19) or the specific care needs of the person. The use of acute healthcare services in our study may reflect an underlying need for these care settings. Conversely, it may also reflect a lack of access to home-based care or differences in individual preferences for care setting. Third, virtual care is a broad term that includes delivery by telephone, videoconference, secure email, and remote monitoring, though none of the fee codes specifically indicate secure email or remote monitoring so it is unlikely those modalities were used in the context of this study (S1 Table) [3]. We were not able to distinguish whether specific types of virtual end-of-life care (e.g., telephone vs. video) were differentially associated with acute healthcare use, which may help inform which modality may be optimal for specific types of care delivery. Additionally while it may represent a small number of people overall, the magnitude of virtual end-of-life care use pre-March 14, 2020 may be overestimated if people did not attend their appointments as there are fee codes for missed appointments or those abandoned for technical difficulties (S1 Table). Fourth, we were not able to categorize whether people were accessing care through their family physician or through a walk-in clinic. It has been suggested that patients of family physicians who provided a high proportion of care virtually during the COVID-19 pandemic did not have higher levels of emergency department visits than those who provided the lowest levels of virtual care [7]. Conversely, people attending virtual walk-in clinics during the COVID-19 pandemic were more likely to visit the emergency department than those who saw their family physician virtually [6]. Fifth, the COVID-19 pandemic resulted in major shifts in healthcare delivery. Thus, the generalizability of our findings to future states is uncertain [47]. However, the observed consistency of our findings in the pre- and post-March 14 periods strengthens our confidence in these associations. By accounting for potential COVID-19 pandemic-related changes (comparing pre- and post-March 14), our results reflect changes during the COVID-19 pandemic and indicate associations, not causation. Finally, Ontario exists within a high-income nation that provides publicly funded healthcare. The generalizability of these findings to other regions without publicly funded healthcare, or to middle- and low-income nations where access to healthcare may be different is unknown.