This retrospective cohort study was conducted using linked administrative health data from public nonadmitted patient or outpatient oncology services in Victoria and Queensland, Australia. The study included individuals aged 18 years or older who received nonadmitted patient or outpatient oncology services in Victoria and Queensland, Australia, between January 1, 2018, and December 31, 2020, and who had a confirmed cancer diagnosis in the Victorian or Queensland Cancer registries, respectively. These services deliver specialist cancer care within hospital-based outpatient clinics across the 2 states. Participants were followed for 1 year from their index outpatient encounter or were censored at death.

Ethical approval for the study was obtained from the Victorian Department of Health and the Department of Families, Fairness and Housing Human Research Ethics Committee (HREC/97793/DOH-2023-383794(v3)) and was approved by Queensland Health under Section 282 of the Public Health Act 2005 (Qld) s 282 (PHA 97793). As this study involved a retrospective analysis of routinely collected administrative data, a waiver of informed consent was granted by the approving ethics committees. Data linkage was conducted by the Centre for Victorian Data Linkage and the Queensland Statistical Services Branch, and researchers accessed only deidentified datasets within the Secure Unified Research Environment to ensure participant privacy and confidentiality. No compensation was provided to participants.

Outpatient data from the Victorian Integrated Non-Admitted Health (VINAH) minimum dataset and the Queensland Health Non-Admitted Patient Data Collection (QHNAPDC) were linked to ED presentations, cancer registries, hospital admissions, and mortality datasets. VINAH and QHNAPDC capture all public outpatient services in Victoria and Queensland, including patient demographics, service dates, service types, contact streams, and delivery modes. Cancer-related data were obtained from the Victorian and Queensland Cancer Registries, which are population-based registries receiving notifications on cancer diagnoses, excluding nonmelanoma skin cancers. ED presentation data from the Victorian Emergency Minimum Dataset and Queensland Emergency Data Collection capture patient demographics, presentation details, and visit types. Comorbidity information was sourced from the Victorian Admitted Episodes Dataset and Queensland Hospital Admitted Patient Data Collection, which include admissions to both public and private hospitals classified using International Classification of Diseases, 10th Revision, Australian Modification (ICD-10-AM). Mortality data from the respective registries of births, deaths, and marriages provided the fact and date of death. Data linkage was performed by the Centre for Victorian Data Linkage and Queensland Statistical Services Branch.

The study cohort included adults aged 18 years or older residing in Victoria or Queensland who received outpatient oncology services between January 1, 2018, and December 31, 2020, and had a cancer diagnosis (ICD-10-AM codes: C00–C43, C45–C96, D45, D46, D47.1, D47.3–D47.5) recorded in a cancer registry prior to or within 7 days of their first outpatient oncology encounter. Oncology service use was identified using contact stream ID 110 in the VINAH and Tier 2 clinic codes (oncology, medical oncology consultation, chemotherapy, and radiation oncology—consultation, simulation, and planning) in the QHNAPDC. Virtual care was defined as consultations delivered via telephone or video, based on the mode-of-delivery variable. To reduce selection bias, inclusion criteria were applied uniformly across both jurisdictions.

Cancer type was identified using primary site ICD-10-AM codes from cancer registry records, supplemented by inpatient data. The 4 most common cancer types in the dataset, such as breast, digestive (including colorectal), lung, and prostate, were analyzed individually, with all others combined (Table S1 in Multimedia Appendix 1). Comorbidities were determined from admission diagnoses, based on the 17 Charlson comorbidities, with malignant neoplasm excluded to prevent collinearity with cancer diagnosis, and were classified as present or absent. In addition, inpatient records were used to identify mental health disorders (ICD-10-AM: F00–F99) and tobacco use (ICD-10-AM: F17.0–F17.9, P04.2, T65.2, Z58.7, Z71.6, Z72.0, Z81.2, Z86.43). To account for prior outpatient health service use, individuals were categorized based on their preindex outpatient activity during the year preceding the study period (January 1 to December 31, 2017) as “low users,” defined as 0 to 3 outpatient encounters, while “high users” had ≥4 encounters [15].

Patients’ residential location in the outpatient data was classified using Statistical Area Level 2 (SA2) for Queensland and using postcode for Victoria, based on the Australian Statistical Geography Standard Remoteness classification. The 5 remoteness categories were collapsed into urban (major cities) and rural (ie, inner regional, outer regional, remote, and very remote areas). Socioeconomic disadvantage was assessed using the Index of Relative Socioeconomic Disadvantage, derived from the patient’s Statistical Area Level 2 (SA2)/postcode of residence, and classified into quintiles (Q1=most disadvantaged; Q5=least disadvantaged). Country of birth was categorized as Australian—or overseas—born from inpatient and outpatient records. Patients were classified as receiving treatment during the COVID-19 lockdown period if any portion of their 12-month follow-up period intersected with the defined COVID-19 lockdown period in both states (March 1, 2020-July 31, 2021).

Outcomes were levels of virtual outpatient service use and unplanned ED visits during the 12-month follow-up. Based on the number of virtual consultations, patients were categorized into 3 groups: no virtual care use, low (1‐3 visits), and high (≥4 visits) use. Unplanned ED visits were identified using the visit type code classified as an “emergency presentation” and categorized as no, low (1‐3 visits), or high (≥4 visits) use [15].

All analyses were performed using SAS 9.4 (SAS Institute Inc) within the Secure Unified Research Environment. Cohort characteristics and outcomes were summarized using descriptive statistics and Pearson chi-square tests. Modified Poisson regression models [16,17] with robust variance estimation were used to examine the association between patient characteristics and virtual care use and between virtual care use and unplanned ED visits within the same 12-month follow-up period. Among patients with both events, only 6% (4699/78,323) had an ED visit prior to their first virtual care encounter, indicating that, in the majority of cases, virtual care preceded ED presentation.

Separate fully adjusted models were fitted for each outcome and for each level, comparing low and high levels of each outcome with no use as the reference category [16]. Covariates were selected for their known association with service use and availability [18-20]. These included age group, gender, country of birth, Charlson comorbidities, remoteness, socioeconomic status, mental disorders, cancer type and morphology, COVID-19 lockdown period, and prior outpatient use.

Collinearity between covariates was assessed using pairwise correlation coefficients. Tobacco use was excluded due to strong collinearity with mental health disorders (r=0.90). Backward stepwise regression sequentially removed nonsignificant variables using a removal criterion of P>.05, and relevant 2-way interactions were tested. Analyses were conducted using complete cases and the proportion of missing data reported in the tables. A sensitivity analysis excluded unplanned ED visits with a COVID-19 diagnosis (ICD-10-AM code U07 as the principal or among the first 20 additional diagnoses). The results were reported as adjusted relative risks (RRs) with 95% CIs, with P≤.05 denoting significance.

The cohort included: 39,099 (49.9%) adults from Queensland and 39,224 (50.1%) from Victoria who accessed nonadmitted patient oncology services. Of these, 37,706 (48.1%) did not use virtual care during the 12-month follow-up, 24,196 (30.9%) had low virtual care use, and 16,421 had high (20.9%) virtual care use. High virtual care users had the highest proportion of unplanned ED visits, with 38.9% (n=4030) experiencing high ED use compared to 31.4% (n=3252) of low virtual care users and 29.8% (n=3091) of nonusers (Table 1, Table S2 in Multimedia Appendix 1).

High users of virtual care had a slightly higher proportion of individuals aged 18 to 44 years (ie, n=1627, 9.9%) compared to low users (n=1864, 7.7%) and nonusers (n=1864, 7.7%) and a higher proportion residing in regional/rural areas (n=7163, 43.6% vs n=9827, 40.6% and n=12,514, 33.2% for low users and nonusers, respectively). Mental health disorders were slightly more common among high users (n=3090, 22.9%) than among low users (n=3087, 18.4%) and nonusers (n=4972, 17.4%). Among high users, there was a higher proportion of digestive cancers (n=3919, 23.9%) and a lower proportion of prostate cancer (n=840, 5.1%). A higher proportion of high virtual care users (n=9929, 60.5%) received care during the COVID-19 period (Table S2 in Multimedia Appendix 1).

Older individuals had a lower likelihood of high virtual care use compared to adults aged 18 to 44 years (eg, ≥75 y: RR 0.78; 95% CI 0.74‐0.83). Overseas-born individuals also had a lower likelihood of high virtual care use compared to adults born in Australia (RR 0.83; 95% CI 0.80‐0.86). Individuals from socioeconomically advantaged areas had a lower likelihood of virtual care use; for example, adults in the least disadvantaged areas had a 23% lower likelihood of high virtual care use compared with patients in the most disadvantaged areas (RR 0.77; 95% CI 0.74‐0.81).

Mental health disorders were associated with higher virtual care use (RR 1.28; 95% CI 1.24‐1.33), as were Charlson comorbidities (RR 1.12; 95% CI 1.09‐1.16), compared to individuals without these conditions. Compared to individuals with breast cancer, adults with digestive cancers (RR 1.24; 95% CI 1.19‐1.30) or lung cancer (RR 1.14; 95% CI 1.07‐1.20) had a higher rate of high virtual care use, while males with prostate cancer had a lower likelihood (RR 0.53; 95% CI 0.49‐0.57).

Virtual care uptake was higher during the COVID-19 period, with individuals more likely to have low use (RR 1.47; 95% CI 1.43‐1.51) and over 3 times more likely to have high use (RR 3.03; 95% CI 2.92‐3.15) compared to the non–COVID-19 period. Additionally, individuals with high preindex outpatient use were more likely to have high virtual care use (RR 1.38; 95% CI 1.31‐1.45) (Table S3 in Multimedia Appendix 1).

The higher proportion of virtual care use was observed among adults with higher unplanned ED use. Among individuals with no ED visits, 57.4% (n=22,680) had not used virtual care, whereas among adults with high ED use, only 29.8% (n=3091) had not used virtual care and 38.9% (n=4030) had high virtual care use.

Among individuals with high unplanned ED use, 46.7% (n=4077) had Charlson comorbidities, 33.1% (n=2892) had mental health disorders, 60.6% (n=5265) reported tobacco use, 24.8% (n=2575) had high preindex outpatient use, and 45.9% (n=4760) resided in rural areas. Socioeconomic disadvantage was more common among high unplanned ED users, with 27.2% (n=2825) in the most disadvantaged quintile, compared with 23.6% (n=6718) among those with low ED use and 18.9% (n=7463) among those with no ED use. Lung (17.7%, n=1836) and digestive cancers (25.1%, n=2602) were more prevalent among high unplanned ED users, whereas breast (12.0%, n=1248) and prostate cancers (5.4%, n=559) were less common (Table 1).

Compared to nonvirtual care users, individuals with low virtual care use had 1.27 times the risk (95% CI 1.1.24‐1.31) and adults with high virtual care use had 2.64 times the risk (95% CI 2.52‐2.79) of high unplanned ED use. Older age was associated with lower risk, with individuals aged 75 years or older having a 35% lower risk of high unplanned ED use compared to adults aged 18 to 44 years (RR 0.65, 95% CI 0.60‐0.70). Females were less likely than males to have high unplanned ED use (RR 0.89, 95% CI 0.85‐0.93) (Figure 1 and Table S4 in Multimedia Appendix 1).

Compared to referent groups, individuals born outside Australia had a slightly higher risk of high ED use (RR 1.08; 95% CI 1.03‐1.13), as did rural residents (RR 1.15; 95% CI 1.10‐1.20), adults with Charlson comorbidities (RR 1.38; 95% CI 1.33‐1.44), mental health disorders (RR 1.64; 95% CI 1.57‐1.71), and high preindex outpatient users (RR 1.18; 95% CI 1.12‐1.24). Individuals from the least disadvantaged socioeconomic quintiles were less likely to experience unplanned ED use (RR 0.66; 95% CI 0.62‐0.70).

Receiving care during the COVID-19 period was associated with a lower risk of high unplanned ED use (RR 0.88; 95% CI 0.84‐0.92). Individuals with lung (RR 1.82; 95% CI 1.67‐1.98) or digestive (RR 1.74; 95% CI 1.61‐1.88) cancers had a higher risk of high unplanned ED use compared to adults with breast cancer (Figure 1 and Table S4 in Multimedia Appendix 1). No interaction effects were observed. Supplementary analysis excluding COVID-19-related encounters yielded similar results (Table S5 in Multimedia Appendix 1).

This study investigated factors associated with virtual care use and its association with unplanned ED visits among patients with cancer receiving outpatient care in Victoria and Queensland. Virtual care was more frequently used by younger adults, individuals residing in rural areas, and those with mental health conditions or a greater burden of Charlson comorbidities. Use was also higher during the COVID-19 lockdown period. In contrast, older adults and individuals born overseas were less likely to use virtual outpatient services. Notably, higher levels of virtual care use were associated with increased unplanned ED presentations.

The observed disparities in virtual care use among older adults and individuals born overseas align with previous research [21,22] and could reflect barriers, such as digital literacy, limited access to devices or reliable internet, and lower confidence in using technology for health purposes [23]. Similarly, individuals born overseas may encounter language barriers, cultural preferences for in-person care, and reduced access to digital services [24]. Adults with a higher comorbidity burden, including mental health disorders, were more frequent users of virtual care, potentially due to an increased need for regular monitoring, symptom management, and multidisciplinary input [25]. Additionally, virtual care uptake was higher during the COVID-19 lockdown period, as health services shifted online to maintain continuity and minimize infection risk. Cancer type was also associated with virtual care use, with higher use observed among individuals with digestive or lung cancers compared to adults with breast cancer. This may reflect greater clinical complexity and a higher need for ongoing monitoring and management in these cancers [26,27].

Higher virtual care use was associated with a greater likelihood of unplanned ED presentations in this study. Similar associations have been reported in prior studies [28,29], whereas others have reported no difference or reduced use [30,31]. It is important to note that these associations may reflect underlying differences in patient characteristics, and a causal relationship with the mode of care cannot be inferred. A higher likelihood of unplanned ED visits may reflect greater clinical complexity among virtual care users, who were more likely to have comorbidities and high prior service use, all markers of greater care needs. In oncology, where rapid deterioration is common, virtual consultations may prompt early symptom detection and precautionary ED referrals for evaluation when in-person assessment is not feasible (eg, directing a patient reporting chest pain or fever straight to the ED). Notably, only 2.6% of ED visits in our sample were coded under ICD-10-AM Z00–Z13 (examination and investigation), suggesting that most ED presentations were not investigation-only encounters; however, these codes are assigned after clinical assessment and do not capture referral rationale. Alternatively, this observed association may indicate a more proactive care-seeking behavior among virtual care users or limitations in virtual care’s ability to manage complex needs.

Older adults with cancer had a lower likelihood of unplanned ED visits, consistent with prior research among cancer survivors suggesting decreased ED use with age [18], possibly due to preferences for conservative care or greater engagement with other services or death [32]. Women had lower odds of unplanned ED visits, aligning with higher acute care use reported among male patients with cancer [19]. Patients in the socioeconomically least disadvantaged areas were less likely to present to the ED for an unplanned visit, potentially indicating better access to timely, coordinated care [33,34]. In contrast, individuals born overseas and adults living in rural areas had higher odds of unplanned ED use, which may be associated with systemic barriers, including limited access to primary or specialist care [35], language barriers [36], and longer travel times [37]. These findings align with literature on greater ED reliance in rural and underserved populations [35,38]. The COVID-19 lockdown period was associated with reduced unplanned ED presentations, reflecting international trends during the pandemic [39]. This decrease may be related to public health messaging, infection fears, and system-level measures discouraging hospital attendance [40].

Comorbidities, including mental health disorders, were associated with unplanned ED use, consistent with their known impact on clinical deterioration and crisis-driven health care utilization [25]. Patients with digestive and lung cancers had the highest likelihood of unplanned ED presentations, likely due to acute complications and rapid disease progression [26,27,41]. Increased preindex engagement with outpatient services was also associated with higher ED use, suggesting that prior care intensity may signal greater underlying need or a tendency toward more proactive care-seeking.

While conducted in a health system with relatively higher digital capacity, these findings are relevant to other settings, particularly those with larger digital divides or constrained oncology resources, where inequities in virtual care access may be amplified. Targeted interventions, such as digital literacy support for older adults and multilingual telehealth services for migrant populations, may help promote equitable access [42,43]. Further research is needed to better understand the content of virtual encounters, subsequent acute care pathways, and the extent to which observed associations reflect causal effects versus underlying patient complexity.

A key strength of this study is the use of large, population-based linked administrative datasets from both Victoria and Queensland, providing valuable insights into virtual oncology service use and unplanned ED presentations among patients living with cancer. However, its cross-sectional design limits interpretation to associations and precludes causal inference. Virtual care use and outcomes were measured within the same follow-up window, and despite virtual care preceding ED presentation for most patients, residual confounding is possible. Information on cancer stages and treatment types was not available, limiting the ability to account for treatment intensity, eligibility for virtual care use, and disease severity. Comorbidities were limited to hospital-recorded diagnoses, likely capturing only the most severe cases and resulting in underenumeration. However, the 1-year lookback period likely reduced but did not eliminate this limitation. Residual confounding may persist, as adults with more severe symptoms could preferentially choose virtual care, partly explaining its association with unplanned ED use. Virtual care use was classified using absolute visit counts, which may mask heterogeneity in clinical patterns across categories, particularly among higher-use patients. The use of a complete case approach may have introduced selection bias if missing data were not random. Preferred language was not included in the analysis due to substantial missing data (47%, n=36,812) and strong collinearity with country of birth. The provision of treatment during the COVID-19 pandemic further complicated interpretation, given service disruptions, rapid telehealth expansion, and varying digital readiness. However, a COVID-19 lockdown time period was included as a covariate in the analysis to distinguish nonpandemic and pandemic phases of care delivery. A stratified prepandemic and postpandemic analysis was not feasible due to low virtual care provision in the pre-COVID period. Finally, the content, quality, the modality of virtual encounters, or administrative data were not able to be validated.

This study highlighted disparities in virtual oncology use, with uptake highest among younger patients, rural residents, and adults with comorbidities, and lowest among older adults and individuals born overseas. To promote equity, policymakers and service planners should work to identify the factors driving these disparities and ensure appropriate use of virtual care. Although virtual care enhances timely access, its association with higher unplanned ED presentations likely reflects greater clinical complexity and differences in care pathways. Further research, including studies using causal analytic approaches, is needed to better understand how virtual consultations are conducted, how they are integrated into broader care pathways, and how these factors influence acute care use.