This was a cross-sectional analysis nested within the Registrar Clinical Encounters in Training (ReCEnT) study.

ReCEnT is a multicenter inception cohort study of Australian GP registrars’ in-consultation clinical and educational experience. The detailed methodology is published elsewhere [28]. ReCEnT is a routine part of the educational program of registrars in participating training regions. Registrars may also elect to provide informed consent for the data to be used for research purposes. During the period reported here, all registrars training in New South Wales, Tasmania, and eastern Victoria (approximately 43% of Australian GP registrars) [29] were eligible for participation.

In ReCEnT, each registrar electronically records details of 60 consecutive consultations, once in each of their first 3 general practice training terms [28]. At the beginning of each collection period, registrars provide demographics and practice data via a questionnaire. In each of the 60 individual consecutive consultations, patient demographics, diagnosis formulations, and clinical and educational actions are recorded. Data are recorded via a dedicated internet-based portal.

For the analyses reported here, data from 7 six-monthly data collection rounds between 2020 and 2023 (following the onset of COVID-19 pandemic with consequent expanded telehealth remuneration for GPs) were used.

The outcome factor was “antibiotics prescribed” (yes or no). This was determined using the Anatomical Therapeutic Chemical Classification J01 [30]. The study factor (independent variable of interest) was “consultation modality” (F2F or telehealth). “Telehealth” included both telephone and videoconference consultations.

Analyses were adjusted for a large range of independent covariates. These included registrar, patient, practice, and consultation variables. The registrar variables were age, gender, training term, full-time, or part-time training status, if qualified as doctor in Australia, and if the registrar had worked at the practice previously.

Patient variables were age group specific to the ARTI problem type (refer to Table S1 in Multimedia Appendix 1), gender, Aboriginal and/or Torres Strait Islander status, and whether the patient was new to the practice or to the registrar.

Practice variables were training region, rurality determined by the Australian Standard Geographical Classification [31], practice size (practices with <6 full-time equivalent doctors are considered “small practices”), socioeconomic status of the practice location (Socioeconomic Index for Areas-Index of Relative Social Disadvantage) [32], and billing policy (whether the practice routinely bulk bills all patients, ie, accepts the government-provided rebate as payment in full).

Consultation content variables were consultation duration, number of diagnoses managed, and whether registrars sought assistance or information during the consultation (no information source used, supervisor called, and another source used).

Consultation action variables included whether pathology was ordered, whether imaging was ordered, whether referrals were made, whether follow-ups were arranged, and whether learning goals were generated.

Analyses were conducted at the level of individual diagnosis (rather than at the registrar or consultation level). The diagnoses were restricted to initial presentations of several self-limited and nonpneumonia ARTIs: acute bronchitis, URTI, acute otitis media, acute sore throat, and acute sinusitis. Refer to Table S1 in Multimedia Appendix 1 for the International Classification of Primary Care 2 plus codes used for these diagnoses [33].

Ethics approval was by the University of Newcastle Human Research Ethics Committee (Reference H-2009-0323) and the RACGP National Research and Evaluation Ethics Committee (NREEC-23-0000000161). Registrars provided written informed consent for data routinely collected as part of the educational program to also be used for research purposes. All data were deidentified. No compensation was provided to participants as ReCEnT is a routine part of training.

A total of 2392 registrars (response rate: 93.4%) contributed data from 301,403 consultations, entailing in 425,059 diagnoses. Table 1 provides details of the characteristics of participating registrars.

Overall, there were 21,384 new diagnoses of our 5 nominated conditions available for analysis. These included 3327, 12,773, 1772, 1754, and 1758 new diagnoses of sore throat, URTI, bronchitis, sinusitis, and otitis media representing 0.7%, 3%, 0.4%, 0.4%, and 0.4%, of all diagnoses, respectively.

The proportions of diagnoses managed via telehealth were as follows: ARTI, 25% (5283/21384); acute sore throat, 19% (641/3327); URTI, 29% (3733/12773); acute bronchitis, 21% (364/1772); acute otitis media; 4.1% (72/1758) and acute sinusitis; and 27% (473/1754). Of the total ARTI consultations managed via telehealth (n=5283), most were managed over the phone (video consult rate: 3.8% [201/5283]). Patient and consultation factors are given in Table 2 for telehealth versus F2F consultations.

For all ARTIs, overall, antibiotics were prescribed in 20% (1035/5283) of telehealth diagnoses compared with 31% (5015/16101) of F2F diagnoses. Refer to Table S7 in Multimedia Appendix 3 for the most commonly prescribed antibiotics for each condition. Refer to Tables S8-S12 in Multimedia Appendix 4 for the characteristics associated with prescribing antibiotics for each ARTI condition.

Examining each of the 5 individual ARTI conditions: the proportions for which antibiotics were prescribed, and the unadjusted and adjusted ORs (from the univariable and multivariable logistic models) for prescribing antibiotics in telehealth, as compared with F2F consultations, are given in Table 3 (refer to Tables S13-S17 in Multimedia Appendix 5 for the full models). The univariable analyses showed reduced odds of prescribing antibiotics via telehealth (compared with F2F) for sore throat (OR 0.62, 95% CI 0.52-0.74; P<.001), and URTI (OR 0.62, 95% CI 0.52-0.75; P<.001) but greater odds of prescribing for acute bronchitis (OR 1.40, 95% CI 1.11-1.76; P=.005), whilst in multivariable analyses, there were statistically significant differences for sore throat (adjusted OR 0.69, 95% CI 0.55-0.86; P=.001), URTI (adjusted OR 0.64, 95% CI 0.51-0.81; P<.001), and otitis media (adjusted OR 0.47, 95% CI 0.26-0.84; P=.01).

The percentages of registrars’ new ARTI diagnoses with follow-up GP appointments organized are presented in Table 4. There were no statistically significant changes in follow-up, but there was some evidence for more follow-ups following F2F for new otitis media problems (P=.06) with clinically significant effect size n/N (11.5%).

On multivariable analysis, antibiotics were (statistically and clinically) significantly more likely to be prescribed in F2F consultations than in telehealth consultations for sore throat, URTI, and otitis media problems. For sinusitis and bronchitis, there were no significant differences in prescribing. There were no significant differences in rates of follow-up consultations organized for ARTIs between the initial consultation modalities, though there was some evidence (P=.06) for follow-up appointments to be made more frequently following F2F appointments when the diagnosis was acute otitis media.

A context for our findings (which are specific to antibiotic prescribing for ARTIs) is that previous studies have reported significantly higher rates of prescribing of medicines in general in F2F consultations compared with telehealth consultations, both by Australian GP registrars [34] and in the wider Australian general practice context [35]. In line with this trend, a recent international publication which compared antibiotic prescribing rates for infections (including but not limited to ARTIs) reported higher prescribing rates in F2F consultations [36].

Our results similarly suggest that antibiotics were more likely to be prescribed in F2F consultations for URTI, otitis media, and sore throat. However, these findings differ from results from 2 systematic reviews and meta-analyses [22,23]. Bakhit et al [23] reported greater antibiotic prescribing for otitis media in telehealth consultations, but no difference between consultation modalities in antibiotic prescribing for pharyngitis. Suzuki et al [22] similarly reported that antibiotics were more frequently prescribed via telehealth for otitis media and pharyngitis. This may be partially due to differences in disease classifications. For example, our classification of sore throat includes, but is not limited to, pharyngitis, whereas both of these systematic reviews examine pharyngitis alone, or group all ARTIs together [22,23]. Furthermore, our study was restricted to a GP registrar population, compared with a wider GP population in these systematic reviews.

Regarding sinusitis and bronchitis, our results are consistent with previous studies that reported no significant difference in antibiotic prescribing rates between consultation modalities for sinusitis [22] or bronchitis [23]. In contrast, some studies have reported that fewer antibiotics were prescribed in telehealth vs F2F consultations for sinusitis [23,37,38]. Previous studies reported that the overall antibiotic prescription rates were lower in telehealth consultations for respiratory tract infections [27], especially in early stages of the COVID-19 pandemic (between 2020 2021), both in Australia [26] and internationally [39,40]. Therefore, it is possible that some of these discrepancies may relate to changes in antibiotic prescribing by modality over time, given that our data are from 2020 to 2023.

One possible explanation for our finding of fewer antibiotic prescribing in telehealth consultations is that registrars organized prompt review appointments (eg, to be able to adequately examine the patient) and deferred antibiotic prescription to the F2F review. In a previous qualitative study, some Australian GPs indicated they were not likely to prescribe antibiotics for acute infections without physically examining the patient (in a follow-up consultation) [21]. However, in our post hoc univariate analyses, we found no significant evidence to support this contention. Similarly, a study looking at the management of sinusitis reported no difference in the follow-up rate between e-visits and F2F consultations [38].

Managing patient expectations may be easier in a telehealth consultation compared with a F2F consultation. A recent qualitative study on the management of acute infections via telehealth reported that some GPs found it easier to deny antibiotics during a telehealth consultation, which could be justified by the lack of physical examination [21]. Conversely, other GPs found themselves prescribing more antibiotics in telehealth consultations as a way to mitigate diagnostic uncertainty [21]. Therefore, while it may be easier to manage patient demand via telehealth, the evidence remains mixed.

The ReCEnT study has a large sample size, a singularly high response rate for studies of GPs [41], and collects are large amount of registrar, patient and practice variables, which allows for fine-grained adjustment for potential confounding in relationships of factors (eg, consultation mode) with registrars’ prescribing. A particular strength is the tight linkage of prescribed medicine with the diagnosis for which it was prescribed.

A further strength is that, in contrast to a number of other studies of this topic, we examined individual clinical presentations (sore throat, acute bronchitis, acute otitis media, URTI, and acute sinusitis) rather than a heterogeneous combination of conditions, such as “acute respiratory infections” [27].

The generalizability of our registrar population results to the wider, established GP population is potentially limited as vocational training is a singular period in GPs’ professional lives. However, within the apprenticeship-like model of GP training in Australia, registrar prescribing behavior (including antibiotic prescribing) is strongly influenced by, and reflects, the prescribing of their supervisors and senior GP colleagues [20,42]. This is reflected in the temporal pattern of Australian registrars’ antibiotic prescribing mirroring that of established GPs, though at slightly lower rates [15-17]. The generalizability of our findings to vocational training settings in countries beyond Australia is strong in countries with similar apprenticeship-like training structures (eg, the United Kingdom, New Zealand, Ireland, and several European countries). However, generalizability to countries with “residency” training programs, such as in North America [43], is less certain.

This study has several limitations. First, patients’ past medical history and the severity of the index ARTI, which could affect patients and practices electing to schedule telehealth or F2F consultations, were not documented. However, antibiotics are not recommended by authoritative Australian evidence-based guidelines for URTI and acute bronchitis [6], irrespective of severity. For conditions such as acute otitis media, sore throat, and acute sinusitis, where antibiotics are indicated in some circumstances and symptom severity may influence both decision-making and the decisions to consult by F2F or telehealth, symptom severity is an unmeasured potential confounder in our analyses. However, it should be noted that symptom severity is not reliably recorded in primary care databases and is infrequently adjusted for in analyses of antibiotic prescribing [8,23,27,36,44].

Second, ReCEnT does not collect data on whether an antibiotic prescription was filled by the patient. However, this limitation is not of significant importance as the aim of the study was to establish registrars’ prescribing behavior rather than patients’ adherence.

Third, the cross-sectional study design does not allow for implications of causality. However, this study design was appropriate for the aim of estimating the prevalence of antibiotic prescribing by GP registrars for ARTIs during telehealth compared with F2F consultations.

It is possible that there is more clinical uncertainty in telehealth consultations due to the lack of physical examination, communication barriers, and potential miscommunication from poor internet connection or slow speeds [45]. However, concerns that diagnostic uncertainty (especially due to limitations to physical examination) may drive inappropriate antibiotic prescribing for ARTIs and compromise antimicrobial stewardship have not been supported by our findings, which may, in fact, suggest the contrary. Noting the heterogeneity in the literature on this topic, further research is indicated to establish if our findings are particular to registrars, during the COVID-19 and immediately after the COVID-19 period, or if there is unmeasured confounding in our findings.

Our findings also demonstrate the importance of considering individual respiratory tract infections rather than relying solely on analyses of grouped heterogeneous respiratory infections.

Registrars’ antibiotic prescribing for sore throat, URTI, and otitis media diagnoses was significantly higher in F2F consultation than in telehealth consultations. However, there were no significant differences in antibiotic prescribing for bronchitis or sinusitis between the 2 consultation modalities. Therefore, we have no evidence from our work to discourage registrars from conducting telehealth consultations for ARTIs on the basis of concerns regarding antibiotic stewardship. Further research is needed to explore the reasons why registrars prescribe antibiotics more highly in F2F consultations than telehealth consultations for sore throat, URTI, and otitis media consultations.