How do medical and social contexts affect telemedicine efficiency and quality? A propensity-score matching protocol in Canada’s primary care

Abstract

Introduction

Telemedicine use has risen significantly since the COVID-19 pandemic. Evidence suggests that the quality of care in telemedicine could be as good as in-person care, but this is likely context-dependent. Expert guidelines have declared the appropriate medical conditions, but often without empirical evidence that grapples with the fundamental information limitations facing telemedicine. We draw on the task-technology fit theory and empirical evidence around human communication to examine how the medical and social contexts affect the efficiency and clinical quality of primary care.

Methods and analysis

We will use a population-based dataset from the Canadian province of British Columbia (BC) to inform a quasi-experimental study using propensity score matching (PSM). The treatment group will consist of telemedicine visits from April to December 2022. We will use PSM to create a control group of matched, in-person visits in the same period. We will then use cluster-robust linear regression to identify how specific medical conditions and social contexts are associated with higher rates of prescription, follow-up with primary care providers, emergency department visits and acute care admissions. We plan for the study to take place from 1 August 2025 to 1 August 2026.

Ethics and dissemination

The Research Ethics BC has granted approval for this study (H21-02244-A006). Our findings will be shared with patients, healthcare providers and policymakers and disseminated through conference presentations and peer-reviewed publications.

STRENGTHS AND LIMITATIONS OF THIS STUDY.

• The matching approach used to identify the context-dependent effect of telemedicine can decrease the risk of confounding.

• The study will use multiple specifications for continuity of care and socioeconomic status.

• The study proxies individual socioeconomic status via neighbourhood-level income.

• The study cannot fully identify the extent to which subsequent visits from in-person services are duplicative of, or complementary to, the virtual visits.

Introduction

Telemedicine offers significant promise for improvement of healthcare service delivery. Because healthcare providers (HCPs) can help patients over the internet or telephone without being in the same physical location, telemedicine can minimise geographical barriers to access and travel-related costs, as well as potentially improve continuity of care.¹ Appropriate use can also improve patient satisfaction and efficiency of care.²

Existing literature suggests that telemedicine can provide similar, if not superior, service quality to in-person services.³,⁹ However, this may be a context-dependent phenomenon. HCPs may have lower quantity and quality of clinical information when addressing specific presentations (eg, patient symptoms and/or concerns) and conditions (ie, diagnosis), which can lower diagnostic accuracy and therapeutic quality. For example, HCPs cannot listen to patients’ lungs to check for pneumonia or conduct other physical exams that require direct physical contact with the patients.¹⁰ Some clinical scenarios require physicians to gather mainly visual or verbal information without physical interactions. Such scenarios may be more appropriate for telemedicine.^{11 12} Professional medical bodies have picked up on such heterogeneity. The American Medical Association and Canadian Medical Association have published guidelines outlining which medical presentations and conditions are appropriate or not for telemedicine settings.^{11 12} These guidelines are implicitly based on the intuition that a lack of access to necessary physical exam information in virtual settings could limit diagnostic accuracy.

However, such expert consensus often lacks evidence and in-depth explanation about what kinds of clinical information are relevant for varying presentations and conditions. The lack of meaningful rationale may limit generalisability, validity and uptake of the recommendations. More worrisome is the possibility of harm due to the uptake of non-evidence-based guidelines. Researchers have highlighted the need for physical exams as the key distinction between the appropriateness of virtual and in-person care,¹³ citing, for example, that pulmonary and cardiology specialists emphasise the importance of potentially picking up clinical signs during in-person visits. But this distinction is overly simplistic, as the range of physical examination techniques varies widely. In other words, there is general acceptance that the appropriateness of telemedicine as an alternative to in-person care should depend on the need for physical exams. However, clear descriptions of the types of physical exams that might affect telemedicine quality and the potential for social factors to support the medical information needs case are lacking.

This study, therefore, aims to examine how medical conditions and continuity of care affect the efficiency and quality of care for telemedicine. It will draw on task-technology fit theory to identify hypotheses, use propensity score matching (PSM) to construct a cohort with exchangeable telemedicine exposure and interact the explanatory variables with telemedicine exposure to examine the moderation effect.

The study will offer three main contributions. First, robustly identifying how medical conditions and continuity of care influence the efficiency and quality of telemedicine will enable patients, HCPs, regulators and policymakers to make more informed decisions about when to use telemedicine in specific medical situations. Second, applying the task-technology fit theory provides a lens to understand the influence of specific types of information within a medical encounter. This theoretical application can inform future studies on what affects the uptake of medical technology. Third, exploring how social contexts, such as established patient-provider relationships, moderate the effectiveness and efficiency of telemedicine can help identify the potential role of complementary policies to minimise the digital divide.^{14 15}

Theoretical considerations and hypotheses

We first consider the medical context by drawing on the task-technology fit theory. In addition to its simplicity, this theory focuses squarely on the utility of the technology for meeting the needs of the task at hand, the deficiency of which could lead to problematic efficiency and quality of care. The theory suggests that novel technologies’ ability to meet a task’s needs determines both their usefulness and ability to improve the user’s performance,¹⁶ and the rest of this section highlights the types of information that are necessary for medical tasks but may not be consistently available in telemedicine. This then motivates the hypothesis development (online supplemental appendix A provides further explanation).

In medicine, physicians determine a potential diagnosis and care plan by integrating information from both patient interactions and diagnostic tests.¹⁷ The types of information a doctor can collect via history and physical exam include verbal, visual, auscultatory and tactile information.¹⁸ Doctors collect these types of information by conversing with and physically examining patients through observing, listening and/or palpating different parts of the patient’s body. Telemedicine would limit this information-gathering process,¹³ as current telemedicine technologies can generally provide only verbal and visual information but not tactile and auscultatory information.

We term medical presentations and conditions that only require telemedicine-supported information as telemedicine-compatible, and they could include general skin disorders, eczema, psoriasis, anxious states and depression. These predominantly rely on verbal and visual information,¹⁹,²¹ which can be transmitted readily via telemedicine. Telemedicine-incompatible visits could include headache, chest pain and abdominal pain, which generally require tactile and auscultatory information that is not available over telemedicine.²²,²⁷

The medical conditions were chosen based on expert recommendations and clinical knowledge. First, the Virtual Care Playbook from the Canadian Medical Association, the College of Family Physicians of Canada and the Royal College of Physicians and Surgeons of Canada suggested several conditions that would be appropriate and inappropriate for telemedicine.¹² They stated that doctors ‘can safely use virtual care to assess and treat mental health issues and skin problems’, but not ‘chest pain’. Second, building on these recommendations, the first author (ST), a practising family physician who delivered telemedicine during the COVID-19 pandemic for a year, selected the conditions that most strongly align with the theoretical considerations of information limitation in telemedicine. In terms of the telemedicine-compatible conditions, the skin and psychiatric conditions required mostly visual and verbal information. As for the telemedicine-incompatible conditions, the possible diagnoses are broad and require a significant history and physical examination. This is evidenced in UpToDate articles,²²,²⁴ which physicians consult regularly for diagnostic and therapeutic advice. It is also supported in medical textbooks, discussing how to work up patients with such presentations.²⁵,²⁷ Last, we recognise that this limits our sample size and potential power. We limited the focus on these conditions purposefully to make sure that we can accentuate the theoretical contrast.

When faced with telemedicine-incompatible situations, physicians may engage in behaviours that lower overall efficiency. To compensate for information insufficiency, physicians may order more laboratory tests, imaging, treatments and follow-up visits, which could undermine overall efficiency. Where such information and compensation behaviours are insufficient, the quality of care could be compromised. Increased rates of complications could reflect in higher rates of visiting emergency departments, follow-up visits and/or acute care admissions. Thus, our two hypotheses are as follows:

Hypothesis 1 (H1): telemedicine-incompatible presentations and conditions will be associated with demonstrated lower efficiency (ie, higher rates of laboratory tests, imaging, prescription and follow-up) and quality (ie, higher rates of emergency visits and acute care admission) when telemedicine is introduced.

Hypothesis 2 (H2): telemedicine-compatible presentations and conditions will demonstrate no significant change in the efficiency (ie, higher rates of laboratory tests, imaging, prescription and follow-up) and quality (ie, higher rates of emergency visits and acute care admission) when telemedicine is introduced.

In addition to the clinical context, the social context may also be important. Empirical evidence suggests that longer relationships between patients and care providers help develop a shared understanding and establish more effective communication with one another.²⁸ As patients and physicians trust each other, the quality and quantity of information between the two may improve.²⁹ Physicians may also have a better grasp of the patients’ background history. These potential benefits for improved efficiency and quality of care are reflected in empirical literature,³⁰,³² which leads to hypothesis 3.

Hypothesis 3 (H3): telemedicine visits between patients and providers with higher continuity of care experience a lower decrease in efficiency (ie, higher rates of laboratory tests, imaging, prescription and follow-up) and quality (ie, higher rates of emergency visit and acute care admission) for telemedicine-incompatible presentations and conditions.

Table 1 and figure 1 summarises the theoretical considerations and hypotheses.

Table 1. Summary of hypotheses.

	Medical context				Social context
Corresponding hypotheses	1			2	3
Types of information	Auscultatory	Tactile	Verbal	Visual
Available via telemedicine	No	No	Yes	Yes
Situations requiring corresponding types of information	Vague complaints (eg, headache, chest pain and abdominal pain)		Psychiatric presentation and conditions (eg, depression and anxiety)	Dermatologic presentations and conditions (eg, psoriasis, eczema and acne)
Telemedicine-compatible	No		Yes	Potentially*
Types of social context					Longitudinal
Expected effect on efficiency† and clinical quality‡	Inferior		Non-inferior	Non-inferior*	Non-inferior
Hypothesis restricted to telemedicine-incompatible presentations and conditions	No		No	No	Yes

^*The telemedicine services delivered in British Columbia, Canada, mostly relied on telephone without video capacity, so the compatibility with telemedicine may be lower than expected.

^†Efficiency includes the rate of prescriptions.

^‡Clinical quality is operationalised as the rate of complications.

Figure 1. Directed acyclic graph summarising theoretical considerations and hypotheses. H1, hypothesis 1; H2, hypothesis 2; H3, hypothesis 3.

Methods and analysis

Setting

Canadian healthcare systems have a single payer, publicly financed model to cover hospital and physician services. The administration of financing and service delivery is decentralised to provincial and territorial governments.³³ Clinical service delivery depends significantly on primary care. In British Columbia (BC), a province with a population of more than five million people, the bulk of primary care rests on family doctors and nurse practitioners, who provide the first point of care and, if necessary, subsequent referrals to specialists. They generally work in physician-owned clinics, and during the period of study, they were mostly paid through public provincial insurance plans via fees-for-service.^{34 35} As such, fee codes can meaningfully capture many of these clinical activities.

Before the COVID-19 pandemic, although telemedicine had been introduced as a potential solution to improve geographical access, Canada lagged behind other countries in its implementation. Although 41% of Canadians reportedly wanted the option of telemedicine,³⁶ only 3% of service providers frequently offered telemedicine services before the pandemic.³⁷ In contrast, other high-income countries had higher rates of telemedicine service provision. For example, England delivered 19% of their primary care appointments via telemedicine,³⁸ and many of the leading healthcare organisations in the USA offered primary care telemedicine services.³⁹,⁴¹ The delay in telemedicine adoption in Canada stemmed in part from the absence of or insufficiently lucrative telemedicine fee codes, HCPs’ unfamiliarity with the relevant technology and inadequate digital infrastructure.⁴²

As the COVID-19 pandemic hit North America in March 2020, the Canadian provincial and territorial governments implemented social distancing restrictions.⁴³ To complement such policies, provinces rapidly rolled out payment and regulatory reforms to encourage telemedicine uptake.⁴⁴ BC was one of the first provinces in Canada to allow family physicians to bill for telemedicine services before the pandemic,² and the fees were comparable to usual levels of reimbursement for in-person consultations. In March 2020, to facilitate more widespread access to telemedicine services, the BC government broadened the scope of services eligible for telemedicine fee codes,⁴⁵ which have remained in use ever since. By April 2020, about 70%–80% of BC physicians were providing telemedicine services.^{46 47} Partially in response to the rapid rise in telemedicine services, the College of Physicians and Surgeons of BC issued a mandate in June 2020 that telemedicine providers must have complementary in-person care capacity for necessary follow-up care⁴⁸; however, enforcement has been questionable.

Most family physicians initially delivered telemedicine services in their existing clinics or at home via telephone because of the cost of upgrading digital infrastructure to include videoconferencing capacity.⁴⁴ In April 2020, the proportion of telephone-based services across Canada for non-COVID-related reasons outweighed video-based services three to one, but this difference largely dissipated by August 2020.⁴⁹ Many health information technology corporations also entered the telemedicine market by directly recruiting family physicians into video-equipped telemedicine platforms with complementary brick-and-mortar clinics.^{50 51}

Despite the rapid and significant uptake, the rate of telemedicine utilisation has not been constant. In Canada, telemedicine quickly declined following its peak in April 2020, with non-COVID-related telemedicine visits comprising only 25% of care by August 2020.⁴⁹ Over the following 20 months, the rate of telemedicine visits for all physician care fluctuated from 23% to 37%.⁴⁹ Explanations from the supply side include physicians’ growing awareness of patient appropriateness for telemedicine, the rising availability of personal protective equipment and the relaxation of social restrictions.⁴⁴ Demand-side considerations include patient preference for in-person visits, in addition to a virtual care option.⁴⁹ Seasonal patterns of illness may be another factor. Between 2020 and 2022, the summer months generally favoured in-person visits, potentially because of the relatively low prevalence rate of upper respiratory infections.

The last major wave of COVID-related infections stemmed from the Omicron variant. The related incidence and mortality significantly diminished by the end of March 2022,⁵² and the public health messaging also changed significantly after this time period.

Overall study design

We will conduct a retrospective study using routinely collected visit-level data, using a health administrative database. We will use PSM as the primary method to improve the balance between in-person and telemedicine visits, controlling for patients’ and providers’ demographic characteristics. The primary study period will be April through December 2022. Our study period is intentionally late in the COVID-19 pandemic to ensure that providers and patients were both freed from most of the pandemic-related health policy changes. Doctors would also have had time to learn about, adapt to and settle into their telemedicine practice. The analysis will use cluster-robust linear regression to identify the association between telemedicine exposure and changes in efficiency and quality within specific medical and social contexts.

Datasets

We will access provincial administrative databases (for the calendar years 2021–2022) through Population Data BC, linked through unique and anonymised patient and provider identifiers. Table 2 summarises the datasets used. They include the Medical Services Plan (MSP) Consolidation File, which includes patient demographic details, such as their age, sex and region of residence. The BC College of Physicians and Surgeons Registry File captures the physicians’ demographic information, including their birth date, sex, years in practice and location of training. The MSP payments dataset captures visit date, fee code and primary diagnostic code (International Classification of Diseases-9 (ICD-9)). The Discharge Abstracts Database provides acute care admission records, including the date of admission. The National Ambulatory Care Reporting System database provides records of the date of emergency room visits. The PharmaNet database contains records of prescriptions dispensed by community-based pharmacies regardless of payor, including the prescriptions filled and the date of dispensation.

Table 2. Datasets used from Population Data BC.

Dataset name	Description of dataset	Level	Key variables used
MSP Consolidation File	Demographic information of patients	Patient	Patients’ age, sex and region of residence
BC College of Physicians and Surgeons Registry File	Demographic information of physicians	Provider	Doctors’ age, sex, years in practice and location of training
MSP payments	Payment information for each visit	Visit	Doctors’ claims, visit dates, fee codes and most responsible diagnostic code (ICD-9)
Discharge Abstracts Database	Hospital admission records for acute inpatient services and day surgery	Admission	Date of admission
NACRS database	Records of emergency room visits	Visit	Date of visit to emergency room
PharmaNet	Records of prescriptions dispensed by community-based pharmacies	Prescription	Prescriptions filled and date of prescription

BC, British Columbia; ICD-9, International Classification of Diseases-9; MSP, Medical Services Plan; NACRS, National Ambulatory Care Reporting System.

We will restrict the physician payment data to visits (virtual and in-person) that are considered compatible or not compatible with telemedicine, using the ICD-9 codes in online supplemental appendix B. As outlined above, incompatible presentations include headache, chest pain and abdominal pain, while compatible conditions will include general skin conditions, eczema, psoriasis, generalised anxiety disorder and depression.

The analysis will include only visits to family doctors from 1 April through 31 December 2022, excluding specialist visits. We also exclude those preceded by a hospitalisation or encounter with other HCPs within 7 days prior to the visit to ensure that each visit represents a new episode of care. We will identify telemedicine visits using the BC-specific fee codes in appendix B.

Consent

This analysis draws on secondary analysis of administrative data. Privacy legislation in BC allows access to these data for approved research purposes without consent.⁵³

Variables

Explanatory variables: we will construct an indicator variable to capture the telemedicine incompatibility of each visit. The variable will equal 1 for any visit with an ICD-9 code that corresponds with a theoretically telemedicine-incompatible condition and 0 otherwise. For the continuity of care variable, we will identify from billing records the proportion of total patient visits over the study period that were with a specific family doctor. We will then define the most responsible physician as the one who provided the highest proportion of care over the previous year.⁵⁴ The socioeconomic status (SES) will be captured through the MSP Consolidation File’s neighbourhood income quintile variable, based on an individual’s postal code in the study year.⁵⁵

Outcome variables: we will measure efficiency through the number of laboratory tests and imaging completed and medications dispensed within 7 days, and follow up with any primary care provider on the same day and within 7 days after each visit. The billing codes used to identify the relevant tests and imaging are provided in online supplemental appendix B. We will measure quality via the number of emergency department visits within 7 days and acute care admissions within 7 and 30 days of the visit.⁵⁶

Covariates: the vector of variables that may affect the probability of a visit conducted via telemedicine and affect the efficiency and quality of care includes both patient and provider variables.⁵⁷ We will use these variables in both the construction of PSM and logistical regressions. Patient characteristics will include age, sex, morbidity burden, SES, region of residence (Health Service Delivery Area (HSDA), n=16 in BC), and past healthcare utilisation intensity. Morbidity will be measured using the Charlson Comorbidity Index constructed from diagnostic codes in physician and hospital data from 2000 to 2001, the two calendar years previous to the calendar year’s index visit. SES and continuity of care variables are constructed as described above. Past healthcare utilisation intensity includes the number of outpatient visits in the previous year. Physician covariates include their age, sex, training location, practice location and years of practice.

Propensity score matching (PSM)

We will use the PSM to construct a comparison group. This quasi-experimental method can improve balance compared with conventional regression analysis.⁵⁸ We will construct the propensity score via estimating equation (1).

$${\displaystyle {\displaystyle \begin{array}{l}\text{logit}(Pr({\text{Telemedicine}}_{i,p,d,t,a}=1))={\beta }_0+{\beta }_1{\text{Patient}}_{p,a}+{\beta }_2{\text{MD}}_{d,a}+{\beta }_3{\text{Month}}_t+{\beta }_4{\text{HSDA}}_a\end{array}}}$$ (1)

where the log odds of visit i for patient p with doctor d at time t and HSDA_a, within the same ICD code category, would be conducted through telemedicine, is determined by the vector of patient-level and physician-level covariates, respectively, indicated through the terms ‘Patient’ and ‘MD’. We also include month, HSDA and provider fixed effects to control for the season-specific, geography-specific and provider-specific uptake of telemedicine. We will start by setting the callipers at a width of 0.2 SD of the logit of the propensity score and adjust to optimise the best balance in observed covariates in the samples.⁵⁹ We will match without replacement.⁶⁰

Analysis

We will first perform descriptive analysis to explore the extent to which the treatment and control groups differ along key provider and patient covariates and outcomes. Statistical tests will use linear regression analysis. The SE will be clustered at the level of HSDA, which may affect the availability of and patients’ interest in using telemedicine.

To test both H1 and H2, we will use estimation equations2 3, respectively, where equation 3 builds on equation 2 through the addition of covariates.

$${\displaystyle {\displaystyle \begin{array}{ll}\text{logit}(Pr(Y_{i,p,d,t,a}=1))& ={\beta }_0+{\beta }_1{\text{Incompatible}}_{i,p,d,t,a}+{\beta }_2{\text{Telemedicine}}_{i,p,d,t,a}\\ +{\beta }_3({\text{Telemedicine}}_{i,p,d,t,a}\times {\text{Incompatible}}_{i,p,d,t,a})\end{array}}}$$ (2)

$${\displaystyle {\displaystyle \begin{array}{ll}\text{logit}(Pr(Y_{i,p,d,t,a}=1))& ={\beta }_0+{\beta }_1{\text{Patient}}_{p,a}+{\beta }_2{\text{Doctor}}_{d,a}+{\beta }_3{\text{Month}}_t+{\beta }_4{\text{HSDA}}_a\\ +{\beta }_5{\text{Incompatible}}_{i,p,d,t,a}+{\beta }_6{\text{Telemedicine}}_{i,p,d,t,a}\\ +{\beta }_7({\text{Telemedicine}}_{i,p,d,t,a}\times {\text{Incompatible}}_{i,p,d,t,a})\end{array}}}$$ (3)

In the above equations, the term Y indicates the vector of outcome variables (ie, same-day and 7 day in-person visits with the same provider, 7 day medication dispensation, 7 day emergency room visit and 30-day acute care admission) and ‘Incompatible’ indicates whether the visit was for a telemedicine-incompatible presentation and condition. The key coefficients of interest will be β3 and β7 in equations2 3, which, respectively, denote the extent to which telemedicine-incompatible or compatible visits changed the probability of the outcome variables. We expect the coefficient to be statistically significant for H1 (ie, telemedicine is inferior for telemedicine-incompatible visits) and non-significant for H2 (ie, telemedicine is non-inferior for telemedicine-compatible visits).

We will also use equations2 3 to qualitatively explore H3 (ie, continuity of care mitigates the detrimental impact on efficiency and quality of telemedicine-incompatible visits) and the effect of SES. We will run these regressions separately for most responsible physicians (MRP) and non-MRP visits and for each income quintile of patients. MRPs are defined as the physician with responsible for the most number of visits for a specified patient. β3 and β7 in equations2 3, respectively, remain the key coefficients.

Formal statistical testing will use equations4 5, which test for the significance of social context in modifying the effect of telemedicine incompatibility. The key coefficients of interest are β1 and β5.

$${\displaystyle {\displaystyle \begin{array}{ll}\text{logit}(Pr(Y_{i,p,d,t,a}=1))& ={\beta }_0+{\beta }_1{X}_{p,a}+{\beta }_2{\text{Doctor}}_{d,a}+{\beta }_3{\text{Month}}_t+{\beta }_4{\text{HSDA}}_a\\ +{\beta }_5{\text{Telemedicine}}_{i,p,d,t,a}+{\beta }_6{\text{Incompatible}}_{i,p,d,t,a}\\ +{\beta }_7{\text{SocialContext}}_{i,p,d,a}\\ +{\beta }_8({\text{Telemedicine}}_{i,p,d,t,a}\times {\text{Incompatible}}_{i,p,d,t,a})\\ +{\beta }_9({\text{Incompatible}}_{i,p,d,t,a}\times {\text{SocialContext}}_{i,p,d,a})\\ +{\beta }_{10}({\text{Telemedicine}}_{i,p,d,t,a}\times {\text{SocialContext}}_{i,p,d,a})\\ +{\beta }_{11}({\text{Telemedicine}}_{i,p,d,t,a}\times {\text{Incompatible}}_{i,p,d,t,a}\times {\text{SocialContext}}_{i,p,d,a})\end{array}}}$$ (4)

$${\displaystyle {\displaystyle \begin{array}{ll}\text{logit}(Pr(Y_{i,p,d,t,a}=1))& ={\beta }_0+{\beta }_1{\text{Telemedicine}}_{i,p,d,t,a}+{\beta }_2{\text{Incompatible}}_{i,p,d,t,a}\\ +{\beta }_3{\text{SocialContext}}_{i,p,d,a}\\ +{\beta }_4({\text{Telemedicine}}_{i,p,d,t,a}\times {\text{Incompatible}}_{i,p,d,t,a})\\ +{\beta }_5({\text{Incompatible}}_{i,p,d,t,a}\times {\text{SocialContext}}_{i,p,d,a})\\ +{\beta }_6({\text{Telemedicine}}_{i,p,d,t,a}\times {\text{SocialContext}}_{i,p,d,a})\\ +{\beta }_7({\text{Telemedicine}}_{i,p,d,t,a}\times {\text{Incompatible}}_{i,p,d,t,a}\times {\text{SocialContext}}_{i,p,d,a})\end{array}}}$$ (5)

The novel term ‘Social_context’ indicates whether the visit is with a patient of higher continuity of care with their family doctor (vs not) and the patient’s income quintile. We expect that β7 and β11 in equations(4) (5), respectively, will be statistically significantly negative for H3 (ie, higher continuity of care is better for telemedicine care).

All analyses will be conducted using R, and the relevant packages will be determined at the time of analysis.

We expect that there will be minimal (<1%) missing data, based on the extensive experience of the authors (LH, ML and KM) using these same datasets. The datasets include information on registration for health insurance coverage, and this requires all the relevant demographic data that are part of this study. Similarly, provider demographic data are required as part of registration that enables billing the BC MSP, and the relevant billing and diagnostic codes are necessary to receive payment. Given this, we will opt for the complete case analysis.

Robustness checks and sensitivity analysis

First, we will conduct a traditional regression without matching, which has the advantage of improving precision.⁶¹ Second, we will adjust the variable specifications. Continuity of care can be measured as the dispersion of visits across different providers and the presence of previous dyadic interactions between the provider and patient.⁶²Third, low-income individuals may also be registered for Plan C and/or Plan I, so we will measure low-income status through Plan C and Plan I registration. Plan C provides drug and limited medical supply/device coverage for those receiving income assistance through the government,⁶³ and Plan I registers BC families in an insurance for medication with an indicator regarding their income status.⁶⁴ Fourth, we will calculate the E-value in cases of significant findings. In essence, not all confounding variables have sufficient power to explain adequately or nullify the relationship between the explanatory and outcome variables, and the E-value shows ‘the minimum strength … an unmeasured (confounding relationship needs to be) to fully explain away a specific treatment-outcome association, conditional on the measured covariates’,⁶⁵ which qualitatively demonstrates the strength of the relationship identified. Fifth, since the providers’ practice patterns may affect telemedicine practice patterns, we will also cluster the SEs with providers.

Ethics and dissemination

We developed this project through conversations with health system policymakers and physicians. We aim to leverage our relationships with the government, physician organisations and academic institutions to disseminate the final results to the relevant policymakers, physician members, researchers, conferences and journals.

The Research Ethics BC approved this study (H21-02244-A006). Population Data BC will house the data for quantitative analysis within its secure research environment and provide only deidentified datasets. Only summary statistics will be used in the outputs.

Limitations

First, although this study considered follow-up visits and antibiotic prescriptions as an indication of inefficiency, this outcome variable can also indicate higher or lower quality, depending on the patient’s needs during the visit. Second, the study could not account for initial triaging considerations, whether from the provider or the patient side. ICD codes may also not perfectly capture the purpose of the visits. More granular studies, including patient records, may better contextualise these considerations. Third, reliable measurement of referrals as an indication of potential information compensation behaviour may help clarify the study’s theoretical assumptions. Fourth, practice sites’ idiosyncrasies may affect the practice patterns, but the data limitations precluded direct control of this dimension.

Discussion

This study aims to examine how medical conditions and continuity of care affect the efficiency and quality of care for telemedicine by drawing on task-technology fit theory to identify relevant hypotheses. We expect the study to enable patients, HCPs, regulators and policymakers to make more informed decisions about when to use telemedicine for specific medical situations while contributing to a deeper understanding of the implications around the types of information exchanged within a medical encounter and the importance of patient-provider relationships.