Data sources study design

We conducted a cross-sectional study using data collected in 2022 from participants enrolled in the OCS [22]. The OCS is an ongoing, multisite clinical cohort of people living with HIV who are receiving HIV care in Ontario, Canada. Detailed descriptions of the cohort have been published previously [23]. Briefly, the source population of the OCS includes individuals aged 16 years and older living with HIV who receive medical care within Ontario’s publicly funded healthcare system. Participants are recruited from 15 HIV care sites across the province, including hospital-based outpatient clinics and community-based practices. The OCS integrates longitudinal clinical data with patient-reported information using standardized data collection procedures.

Clinical data are obtained through manual and electronic abstraction of medical records from routine follow-up visits and are supplemented by linkage with Public Health Ontario Laboratories, the sole provincial provider of HIV-related laboratory testing, including viral load measurements. These clinical data are complemented by standardized questionnaires administered at enrollment and annually thereafter, which collect detailed sociodemographic, socioeconomic, psychosocial, and behavioural information, including employment, income, mental health, substance use, and healthcare utilization [24].

For the present analysis, we used cross-sectional data from the 2022 OCS questionnaire, which included detailed measures of HIV care delivery modalities, including in-person and virtual care via telephone or video, as well as relevant demographic, socioeconomic, and clinical covariates.

Study population

We used cohort data available as of 2022 to define our study population. Participants were eligible for this analysis if they were 16 years or older, had at least one appointment with their HIV physician in 2022, and had completed the OCS questionnaire. Individuals with incomplete information regarding the type of care received were excluded from the analysis.

Measures/outcomes

HIV care was classified into three mutually exclusive categories: (i) in-person clinic visits; (ii) virtual care via telephone or video call; and (iii) a combination of virtual and in-person services (hybrid). Virtual care was defined as consultations with an HIV care physician by telephone or video call, as defined in the 2022 OCS Questionnaire.

Baseline demographic and clinical covariate data were extracted from the 2022 OCS questionnaire to examine potential predictors of retention in 2022. Age was analyzed both as a continuous and categorical variable, divided into four groups: ≤ 30, 31–40, 41–50, and > 50 years. Sex and sexual orientation were combined into a single variable with three categories: females, MSM, and heterosexual men [7]. Other covariates are defined as follows: race or ethnicity (African, Caribbean, or Black versus White versus others); language fluency, derived by combining two variables: Canadian-born and immigrants living ≥ 10 years, defined as participants with proficient English-speaking skills, versus immigrants living for < 10 years, defined as having below-average English-speaking skills; relationship status, categorized as stable (married, living common-law, or in a committed relationship) versus unstable (widowed, separated/divorced, or single); regions in Ontario categorized into four areas (Greater Toronto Area, Eastern Ontario, Northern Ontario, Southwestern Ontario); education level (elementary versus high school versus college versus higher education); employment status (employed versus unemployed); and annual personal income (≤ $50,000, $51,000-$70,000, $71,000-$100,000, and> $100,000).

Barriers to virtual care were defined as privacy or lack of privacy, depending on the availability of a private space for attending the virtual visit. The housing situation was categorized as stable (condo, housing facility, shelter, or room in a motel or hotel) or unstable (correctional facility, couch surfing, or living on the street).

Clinical covariates extracted include adherence to ART (self-reported), categorized as optimal adherence (≥ 95%: never skipped or skipped more than three months ago) and suboptimal adherence (< 95%, defined as missed doses within the past week, one to two weeks, two to four weeks ago, or one to three months ago) [26]; viral load was defined as suppressed (≤ 40 copies/mL) or unsuppressed (> 40 copies/mL) [27]; alcohol abuse, measured using the Alcohol Use Disorder Identification Test (AUDIT-10), with harmful alcohol use defined as a score of ≥ 8 regardless of gender/sex [28]; depression, measured by the Patient Health Questionnaire (PHQ) scale, with nine items, with depression defined as normal (0–2), mild (3–5), moderate (6–8), and severe depression (9–12) [29]; health-related QoL, assessed using the Short Form 36 Health Survey with two components: Mental Component Summary Score (MCS) and Physical Component Summary Score (PCS) [30]; smoking, classified as heavy smokers (at least 20 cigarettes daily) vs. moderate smokers [31]; diagnosis of mental health comorbidities, based on self-report in the OCS questionnaire; and stigma, assessed using a 10-item HIV-related stigma scale, categorized into four major components: personalized stigma, worries about disclosure of status, negative self-image, and sensitivity to public reactions regarding HIV status. Individuals who responded with “agree” or “strongly agree” were identified as experiencing stigma in at least one of the four components [32].

Patient satisfaction survey

The OCS questionnaire includes a satisfaction survey on virtual and in-person HIV care, divided into three sections:

Data analysis

We calculated descriptive statistics, reporting proportions for categorical variables and medians with interquartile ranges (IQR) for continuous variables. We compared continuous covariates using the Kruskal-Wallis test and categorical variables using chi-square tests. We considered a p-value of < 0.05 to indicate statistical significance.

Prior to fitting the multinomial logistic regression models, diagnostic and preparatory analyses were conducted to assess model stability and specification. Multicollinearity among independent variables was evaluated using variance inflation factors (VIFs) derived from a linear proxy model that included the same set of covariates; all VIFs were below 2, indicating minimal multicollinearity among predictors [33]. Influential observations were assessed using Cook’s distance; although a small number of observations exhibited elevated influence values, these reflected plausible participant characteristics and did not materially affect model estimates [34].

Theoretical and clinical considerations guided model specification. Variables such as age, depression severity, and viral load were a priori categorized to capture potentially nonlinear relationships supported by the existing literature [27,29]. The only covariates modelled as continuous variables were the Physical Component Summary (PCS-12) and Mental Component Summary (MCS-12) scores, which are validated continuous measures of health-related quality of life [30,35]. Plausibility and potential departures from linearity of continuous variables were evaluated using graphical methods, including inspection of distributions, residual diagnostics from proxy regression models, and smoothed plots. These assessments did not reveal meaningful departures from linearity or evidence that inclusion of interaction terms or higher-order nonlinear effects would improve model fit. Accordingly, a parsimonious main-effects multinomial logistic regression model was retained.

The multinomial logistic regression model relies on the Independence of Irrelevant Alternatives (IIA) assumption [36]. We assessed the robustness of this assumption using both formal testing and sensitivity analyses. The Hausman–McFadden specification test did not indicate statistically significant violations of the IIA assumption (p > 0.05) [37]. Given the known limitations of this test in finite samples and in the presence of correlated outcome categories, we additionally conducted a sensitivity analysis excluding the hybrid care category and fitted a binary logistic regression comparing virtual versus in-person care [38]. Findings from the sensitivity analysis were consistent with those from the primary multinomial model, indicating that the substantive findings were not materially sensitive to potential violations of the IIA assumption. The outcome categories: in-person care, virtual care, and hybrid care, were defined a priori and reflect distinct modes of healthcare delivery. Covariates with sparse cell counts (housing and stigma) across outcome categories were excluded from the final models, as they resulted in numerical instability during model estimation.

Following model diagnostics and robustness assessments, we fit a three-category multinomial logistic regression model, with in-person care as the reference category, to identify independent correlates of care modality use [39]. Given the limited prior research on predictors of virtual care use, we included all covariates in the model irrespective of statistical significance, thereby creating an explanatory model. The model results are reported as odds ratios (OR) with 95% confidence intervals (CI). A 95% CI excluding the null value was considered statistically significant.

Since the OCS questionnaire had data missing at random, we performed 10 imputations for the model and combined the results using Rubin’s rule [40]. However, the dataset comprised two components, and to preserve unbiased responses and accurately represent participants’ preferences, we did not impute the participant survey data [41]. The survey responses and the assessment of barriers to care were summarized using numerical and percentage data. Statistical analysis was performed using R software version 4.4.1.

Socio-demographic factors associated with the type of care

Socio-demographic factors.

Compared to individuals with a university degree, those with a high school education had lower odds of attending virtual visits (AOR 0.67, 95% CI: 0.46, 0.98). Participants with an annual gross income between $70,000 and $100,000 were less likely to attend virtual visits than those earning over $100,000 (AOR 0.59, 95% CI: 0.38, 0.91).

Participants diagnosed with HIV for less than 10 years were more likely to attend virtual visits compared to those diagnosed for more than 10 years. Additionally, participants with mild (AOR 2.20, 95% CI: 1.54, 3.13), moderate (AOR 2.06, 95% CI: 1.28, 3.31), moderate severe (AOR 2.80, 95% CI: 1.49, 5.26), and severe depression (AOR 2.46, 95% CI: 1.13, 5.32) preferred hybrid care services (Table 2).

Download:

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

Table 2. Factors associated with virtual care and hybrid visits in HIV care (reference category: in-person visit) in 2022(n = 1930).

https://doi.org/10.1371/journal.pone.0326989.t002

Barriers to virtual visits.

The OCS questionnaire identified several barriers to virtual HIV care. Among participants who preferred in-person visits, 69.3% (n = 624/900) preferred face-to-face consultations, and 40% (n = 355/900) reported not being offered virtual care. A lack of private space for virtual visits was reported by 33.7% of participants (652/1,930), including 17.1% (112/652) of those in the virtual care group.

Significant differences were observed in the proportion of participants seeking care from providers other than their usual HIV care provider: 41.2% of hybrid care users (n = 454/630), 39.2% of in-person care users (n = 600/900), and 54.9% of virtual care users (n = 220/367) sought care elsewhere (p < 0.05). Most additional appointments involved HIV-related care, including visits to family doctors, HIV specialists, pharmacists, and nurses. Table 3 summarizes the factors influencing virtual care assess.

Download:

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

Table 3. Factors influencing access to care by the mode of healthcare delivery.

https://doi.org/10.1371/journal.pone.0326989.t003

Implications for Practice and Public Health

These findings support the continued use of hybrid HIV care models that integrate both virtual and in-person services. The absence of statistically significant differences in care modality use across age groups and in most access-related barriers suggests that virtual care did not systematically disadvantage older adults or individuals facing technological constraints within this publicly funded healthcare setting. The higher use of hybrid care among individuals with depression highlights the importance of flexible care delivery models that can accommodate fluctuating mental health needs while maintaining continuity of care. From a public health perspective, maintaining multiple modes of care delivery may help promote equitable access, support patient-centred care, and strengthen engagement in HIV services without exacerbating existing disparities.

Limitations

This study’s cross-sectional design limits the ability to draw causal conclusions regarding the relationship between sociodemographic variables and care modality use. Data were also collected during the early stages of virtual care adoption, when standards, infrastructure, and provider training were still developing, which may reduce the applicability of the findings to current practices [46,52].

Although the patient satisfaction survey provides insights into care preferences, it primarily focuses on patient experience and access without capturing provider perspectives, which could introduce bias. Additionally, the study does not examine the usability and functionality of virtual care platforms. Future research should incorporate qualitative interviews and provider surveys to deepen understanding of healthcare professionals’ experiences, preferences, and perceived limitations across HIV care modalities.

Strengths

This study’s strengths lie in its comprehensive examination of a broad range of sociodemographic and health-related factors influencing access to care, combined with an in-depth exploration of patient experiences, barriers, and acceptance of virtual care. By integrating these perspectives, the study provides a detailed overview of virtual care utilization patterns and identifies important areas for future research, as well as actionable opportunities for health system improvements to enhance equitable access and patient-centred care.

Importantly, the robustness of these findings is supported by methodological considerations. Although non-response bias is a common challenge in survey research, our study achieved a 22% response rate (427/1,930), which is sufficient to represent a population of approximately 10,000 with a 5% margin of error [53]. The use of a complete and accurate sampling frame, along with a validated survey instrument, further strengthens the reliability of the results and supports population-level inferences while minimizing random sampling errors [54]. Together, these factors underscore the study’s ability to provide meaningful and generalizable insights into virtual care utilization.

Drawing on a large sample and real-time data, the study highlights the importance of long-term research in fully assessing the potential of virtual care. Future research should use longitudinal and administrative data to examine visit patterns, patient and provider preferences, and the cost implications of virtual care to better understand its long-term role in HIV service delivery.