Search strategy

The search was conducted to identify relevant studies published from January 1, 2010, to July 1, 2024. The databases and platforms were searched, including PubMed, Embase (Ovid), Cochrane Central Register of Controlled Trials (Wiley), CINAHL (EBSCO), Web of Science (Clarivate), PsycINFO (EBSCO), and Google Scholar. The search strategy was developed by a research librarian (AM). It included a combination of keywords, medical subject headings (MeSH), and free text terms related to concepts of telemedicine, antibiotic stewardship, and antibiotic use. The Full details of search strategies are provided in S1 Appendix. The references to the included studies and relevant reviews were also hand-searched. EndNote 21.0 citation management software organized references throughout the review process.

Study selection

Two reviewers independently conducted literature screening and selection using Covidence systematic review management software [48,49]. The screening was performed in 2 stages - titles/abstracts and then full texts. Studies were included if they met the following criteria: (1) assessed telemedicine interventions aimed at implementing or delivering antibiotic stewardship programs (telestewardship); (2) included a comparator receiving usual care without telemedicine; (3) reported quantitative outcome measures related to antibiotic use or resistance; (4) published between January 2010 to the first of July 2024 in English. Reviews, protocols, editorials, and qualitative studies were excluded. Based on these initial selection criteria, The current study further refined study selection using the following detailed inclusion and exclusion criteria to ensure a comprehensive and relevant analysis of the current evidence in telemedicine for antibiotic stewardship.

Inclusion criteria

1. 1). Population: Patients receiving care for infectious diseases or conditions that may require antibiotic treatment, with no restrictions on age, gender, or comorbidities.
2. 2). Interventions: Telemedicine programs (e.g., video consultations, remote monitoring, clinical decision support systems) aimed at improving antibiotic stewardship or reducing antibiotic prescribing, delivered by licensed healthcare professionals (e.g., physicians, nurse practitioners, physician assistants).
3. 3). Comparators: In-person care or standard antibiotic prescribing practices without telemedicine components.
4. 4). Outcomes: Quantitative measures related to antibiotic prescribing patterns, antibiotic appropriateness, guideline adherence, and resistance rates.
5. 5). Study Designs: Randomized controlled trials and observational studies, including quasi-experimental, before-after, cohort, case-control, and cross-sectional studies.
6. 6). Language: English.
7. 7). Years: January 1, 2010 to July 1, 2024.

Exclusion criteria

1. 1). Study Designs: Reviews, protocols, editorials, opinion pieces
2. 2). Outcomes: Qualitative outcomes only
3. 3). Language: Non-English publications

Definition of terms

1. 1). Inpatient Telestewardship: Telemedicine interventions for hospitalized patients, typically involving remote infectious disease consultations, virtual rounds, or decision-support tools aimed at optimizing antibiotic therapy in real-time.
2. 2). Outpatient Telemedicine Stewardship: Telemedicine-based programs implemented in ambulatory or primary care settings (e.g., clinics, telemedicine platforms) to guide antibiotic prescriptions and reduce inappropriate antibiotic use outside the hospital setting.
3. 3). Comparator Arm: For included studies, the comparator had no telemedicine component. This could be usual care with or without an existing (in-person) antimicrobial stewardship program, as reported by each study.
4. 4). Antibiotic Appropriateness: Defined as selecting the correct agent, dose, route, and duration in line with recognized local or international treatment guidelines.

Data extraction

A standardized data extraction process was undertaken by two independent reviewers to gather relevant details from the included studies in a consistent manner. Microsoft Excel created a predetermined form to capture key data points on study identification, population, interventions, comparators, measured outcomes, statistical analyses, and confounding factors. Specifically, extracted information included author, publication year, study design, location, sample size, participant demographics, healthcare setting characteristics, types of telemedicine modalities, duration of interventions, details on usual care comparators, antibiotic prescribing, and appropriateness parameters, effect sizes with statistical significance values, and factors adjusted for in the analysis. Any disagreements between the two primary reviewers were resolved through discussion and consensus; if consensus could not be reached, the disagreement was elevated to a third reviewer to make a final determination. The senior author thus did not unilaterally decide outcomes when the primary reviewers’ consensus was lacking, ensuring impartial resolution of conflicts.

Risk of bias assessment

Two independent reviewers assessed the risk of bias for each included study. For randomized controlled trials, the Cochrane Risk of Bias tool 2.0 [49] and non-randomized studies, the ROBINS-I [50] tool was applied to assess the risk of bias due to confounding, selection, classification, deviation from interventions, missing data, measurement of outcomes, and selection of the reported results. Any disagreements in risk of bias judgments were resolved through discussion between the two reviewers to reach a consensus. The overall risk of bias was categorized as low, moderate, or high for each study based on criteria and domains outlined in the Cochrane and ROBINS-I tools. The assessment process was tailored to judge the biases most applicable to the study design. Additionally, we evaluated the certainty of evidence for each outcome using the GRADE approach [51], considering factors such as risk of bias, inconsistency, indirectness, imprecision, and publication bias.

Data synthesis

Due to the range of different telemedicine interventions and outcome measures reported across studies, a structured narrative synthesis was performed. Where sufficient clinical and methodological homogeneity existed, summary estimates were pooled using random-effects meta-analysis. All analyses were conducted using Review Manager software (RevMan) [52].

Study characteristics

Of the 21 included studies [53–73], observational studies constituted 47.6% (n = 10) [53,55,56,58,59,62,63,67–69], followed by quasi-experimental studies 38.1% (n = 8) [54,57,60,61,70–73], randomized controlled trials 9.5% (n = 2) [64,65], and mixed-methods studies 4.8% (n = 1) [66]. Studies were predominantly conducted in the United States 57.0% (n = 12) [54,55,57,61,64–71], with representation from the United Kingdom 14.3% (n = 3) [59,63,72], Brazil 9.5% (n = 2) [53,58], and individual contributions 4.8% (n = 1) each from France [56], Australia [60], Kuwait [62], and Canada [73]. Hospital-based interventions dominated at 71.4% (n = 15) [53–56,60–63,65–67,70–73], while primary care settings represented 23.8% (n = 5) [57,59,64,68,69] and telemedicine centers 4.8% (n = 1) [58]. Regarding intervention comparisons, 71.4% (n = 15) evaluated telemedicine against no formal stewardship service [53–58,61–63,65,66,70–73], while 28.6% (n = 6) compared telemedicine with in-person care [59,60,64,67–69]. Study duration ranged from 3 months [62] to 84 months [72,73], with a median of 18 months. Studies were published between 2015 and 2024, with most conducted during 2019–2024 [54,55,58–68,70,71]. Population sizes showed considerable variation, ranging from 20 beds [62] to 221,128 patients [58], precluding standardized comparison. A summary of study characteristics is presented in Table 1.

Download:

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

Table 1. Summary of study characteristics.

https://doi.org/10.1371/journal.pone.0320840.t001

Quality assessment and risk of bias

The quality evaluation encompassed all studies, utilizing the ROBINS-I tool for 90.5% (n = 19) studies [53–63,66–73] and the Cochrane Risk of Bias 2.0 tool for 9.5% (n = 2) studies [64,65] (Fig 2). Overall risk assessment revealed 66.7% (n = 14) studies demonstrating low risk [53–59,62,64–67,70,71], 28.6% (n = 6) studies showing moderate risk [60,61,68,69,72,73], and 4.8% (n = 1) study exhibiting high risk [63]. The predominant factors contributing to elevated risk included potential confounding and selection bias, while the randomized controlled trials [64,65] demonstrated robust methodological quality. The GRADE assessment revealed 9.5% (n = 2) studies of high quality [64,65], 76.2% (n = 16) studies of moderate quality [55–63,66,68–73], and 14.3% (n = 3) studies of low quality [53,54,67]. A detailed risk of bias assessment for individual studies is presented in S4 Appendix. Table 2 presents the GRADE Summary of Findings for quality assessment of outcomes.

Download:

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

Table 2. GRADE summary of findings.

https://doi.org/10.1371/journal.pone.0320840.t002

Download:

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

Fig 2. Risk of bias assessment.

https://doi.org/10.1371/journal.pone.0320840.g002

Synthesis of results

Meta-analysis was not conducted across the 21 included studies [53–73] due to substantial methodological heterogeneity in intervention modalities, comparator groups, and outcome metrics (Cochrane Handbook, Section 10.10.2) [47]. We implemented a structured descriptive synthesis using an outcomes-first approach, stratifying evidence by primary outcomes (antibiotic utilization, guideline adherence, prescribing rates), comparator types (telemedicine vs. no service; telemedicine vs. in-person care), and clinical settings (inpatient/outpatient). Quality assessment was integrated at each analytical level, with findings weighted by risk of bias assessments and maintaining a minimum threshold of three studies per subgroup to ensure reliable conclusions. The complete dataset for all 21 included studies is available in S5 Appendix.

Impact on antibiotic use

Of the twenty-one included studies, 52.4% (n = 11) reported antibiotic use outcomes [53–55,61–63,65,66,70,71,73], all conducted in inpatient settings comparing telemedicine against no formal stewardship service. Significant reductions in antibiotic use were reported in 90.9% (n = 10) studies [53,54,61–63,65,66,70,71,73], with magnitudes ranging from 5.3% [54] to 62.7% [71]. The highest-quality evidence [65] demonstrated a 28% reduction (95% CI: 22-34%) in total antibiotic consumption. Notable findings included decreased broad-spectrum antibiotic use [53,54], with specific reductions in carbapenems (59%) [54] and fluoroquinolones (35%) [54]. One study reported mixed results, showing decreases in some antibiotics while others increased [55]. The evidence quality distribution showed 9.1% (n = 1) high-GRADE [65], 18.2% (n = 2) low-GRADE [53,54], and 72.7% (n = 8) moderate-GRADE studies [55,61–63,66,70,71,73]. Risk of bias was predominantly low in 72.7% (n = 8) studies [53–55,62,65,66,70,71], with 18.2% (n = 2) showing moderate risk [61,73], and 9.1% (n = 1) exhibiting Serious risk [63].

Impact on guideline adherence

Among the twenty-one included studies, 57.1% (n = 12) reported guideline adherence outcomes [54–56,60,62,63,66–68,71–73]. Distinct patterns emerged across settings and comparators. For telemedicine versus no service, 42.9% (n = 9) studies [54–56,62,63,66,71–73], all in inpatient settings, demonstrated consistent improvements. Acceptance rates ranged from 67.7% [71] to 98% [55,63], with comprehensive adherence metrics showing significant enhancements. Full compliance rates varied from 79% (95% CI 76.4-81.6%) for diagnostic recommendations [56] to 89.7% for overall interventions [62]. Policy adherence improved by 6.48-6.63% across medical and surgical services [72], with documentation compliance increasing by 15.20-35.97% [72]. For telemedicine versus in-person consultation, 14.3% (n = 3) studies provided evidence [60,67,68]. In the inpatient setting, two studies revealed comparable adherence rates between modalities (79.1% vs 80.4%, p = 0.36 [60]; 89% vs 86%, p = 0.33 [67]). The single outpatient study demonstrated marginally superior guideline concordance with telemedicine (92.5% vs 90.7%, p = 0.004) [68]. Evidence quality assessment revealed 75.0% (n = 9) moderate-GRADE studies [55,56,60,62,63,66,68,71,72] and 25.0% (n = 3) low-GRADE studies [54,67,73]. Risk of bias was predominantly low in 66.7% (n = 8) studies [54–56,62,66,67,71,73], with 25.0% (n = 3) showing moderate risk [60,68,72], and 8.3% (n = 1) exhibiting Serious risk [63].

Impact on prescribing rates

Among the twenty-one included studies, 38.1% (n = 8) reported prescribing rate outcomes [57–59,61,64,65,68,69]. Analysis revealed distinct patterns across settings and comparators. For telemedicine versus no service, 19.0% (n = 4) studies provided evidence [57,58,61,65], equally distributed between inpatient and outpatient settings. In the inpatient setting, both studies demonstrated significant reductions: a 28% increase in patients receiving no antibiotics (p < 0.0001) [61] and a 32% reduction in infant antibiotic prescriptions (RR 0.68; 95% CI 0.63-0.75) [65]. The outpatient studies showed modest improvements, with one reporting a 3.9% absolute reduction in sinusitis prescribing (p < 0.001) [57] and another maintaining consistently low prescribing rates for COVID-19 cases [58]. For telemedicine versus in-person consultation, 19.0% (n = 4) studies, all in outpatient settings [59,64,68,69], revealed mixed outcomes. Adult telemedicine prescribing showed higher rates compared to face-to-face care (52% vs 42%; OR 1.23; 95% CI 1.18-1.29) [59], while three studies demonstrated favorable reductions in prescribing rates [64,68,69], particularly for respiratory conditions. Evidence quality assessment revealed 25.0% (n = 2) high-GRADE studies [64,65] and 75.0% (n = 6) moderate-GRADE studies [57–59,61,68,69]. Risk of bias was predominantly low in 62.5% (n = 5) studies [57–59,64,65], with 37.5% (n = 3) showing moderate risk [61,68,69].

Sensitivity analysis and quality assessment impact

A detailed comparison of main and sensitivity analyses, including separate exclusion criteria for severe-ROB and low-GRADE studies, is provided in Table 3, demonstrating that our core findings remained largely unchanged across all analytical scenarios. Our methodological quality assessment revealed 4.8% (n = 1) Serious risk of bias [63] and 14.3% (n = 3) low-GRADE evidence [53,54,67] among twenty-one studies. Quality-adjusted sensitivity analysis demonstrated robust findings across outcomes. For antibiotic use, excluding lower-quality evidence [53,54,63] maintained consistent effect sizes among 72.7% (n = 8) studies, with reductions ranging from 28% [65] to 62.7% [71] (p < 0.05). Prescribing rates outcomes, after excluding one low-GRADE study [67], retained statistical significance (p < 0.001) across 87.5% (n = 7) studies [57,59,64,68,69]. Similarly, guideline adherence outcomes remained stable after removing three lower-quality studies [54,63,67], with acceptance rates among 75.0% (n = 9) studies ranging from 67.7% [71] to 89.7% [62] (p < 0.05). This sensitivity analysis, employing a standardized quality-adjustment framework (Initial Studies - (Serious ROB +  Low GRADE) =  Adjusted Sample), confirms the resilience of our primary findings to methodological limitations across all outcomes, particularly supported by high-GRADE evidence [64,65].

Download:

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

Table 3. Main analysis versus sensitivity analysis results.

https://doi.org/10.1371/journal.pone.0320840.t003

Comparison with existing literature

Our systematic review extends the evidence base established by previous analyses. Nathwani et al. [78] demonstrated comparable effectiveness in traditional stewardship programs, with our inpatient antibiotic reduction rates (5.3%-62.7%) aligning with their reported outcomes (0.06%-80.1%). Bakhit et al.’s [79] analysis of 13 studies revealed condition-specific variations in outpatient prescribing patterns, contrasting with our inpatient findings where high-quality RCT evidence demonstrated consistent improvements (RR 0.68; 95% CI 0.63-0.75) [65]. While Dyar et al.’s [80] framework established comprehensive prescribing competencies through expert consensus (98% agreement, 24 countries), our review provides specific evidence for implementation effectiveness through guideline adherence rates (67.7%-98%) [55,63,71]. The comparable effectiveness between telemedicine and in-person consultation (79.1% vs 80.4%, p = 0.36) [60] suggests successful adaptation of stewardship practices to remote delivery models.

Strengths and limitations

The review’s strength lies in its comprehensive examination of telemedicine-based interventions during healthcare digitalization (2015-2024), with robust methodology and quality assessment. The predominantly low risk of bias [53–59,62,64–67,70,71] and substantial proportion of moderate GRADE quality studies enhance reliability. However, outcome heterogeneity precluded meta-analysis, and limited outpatient evidence restricts generalizability. Variable study durations may not fully capture long-term intervention sustainability.

Implications for practice and policy

The evidence supports specific considerations for telemedicine-based antibiotic stewardship implementation. In inpatient settings, our analysis demonstrates both clinical and economic benefits. Six studies comparing telemedicine versus no service [53–55,61,62,73] reported consistent cost reductions, ranging from USD 4,690 to USD 350,000 annually, with one study showing 20.5% reduction in antimicrobial budget [73]. Implementation effectiveness is supported by high guideline adherence rates (67.7-98%) [55,63,71] and equivalent outcomes between telemedicine and in-person consultation (79.1% vs 80.4%, p = 0.36) [60]. Operational improvements include reduced antibiotic days [61] and streamlined workflows [55], suggesting feasible integration into existing healthcare systems. The evidence particularly supports inpatient applications, where systematic monitoring demonstrated significant reductions in antibiotic consumption (range: 5.3%-62.7%) [53,54,61–63,65,66,70,71,73].

Implications for practice and policy

In the context of global antimicrobial resistance challenges and increasing healthcare digitalization, this systematic review provides evidence-based support for telemedicine as a viable solution for expanding antibiotic stewardship programs. The demonstrated effectiveness in inpatient settings, combining both clinical improvements [53,54,61–63,65,66,70,71,73] and economic benefits [53,54,73], suggests telemedicine offers a scalable model for healthcare systems struggling with antimicrobial resistance. The comparable effectiveness between telemedicine and traditional consultation methods [60] indicates potential for widespread implementation, particularly valuable for resource-limited settings lacking on-site infectious disease specialists. However, the variable effectiveness in outpatient settings highlights the need for context-specific implementation strategies. These findings provide timely evidence for healthcare policymakers and administrators considering digital solutions to enhance antimicrobial stewardship programs, especially given the growing global emphasis on combating antibiotic resistance.

Future research directions

Future research should address three critical methodological gaps identified in this systematic review. First, standardization of outcome measures for antibiotic stewardship interventions is needed to enable meta-analyses and strengthen the evidence base, as current heterogeneity in metrics precluded quantitative synthesis. Second, while our review found promising evidence for telemedicine-based stewardship, the limited number of randomized controlled trials [64,65] indicates the need for more robust experimental designs, particularly comparative effectiveness studies between telemedicine and traditional in-person services. Third, to inform implementation strategies, future trials should incorporate standardized economic analyses, as current evidence of cost-effectiveness is limited and inconsistently reported across studies [53–55,61,62,73].