Abstract
Background
Digital health platforms are transforming primary care delivery in low- and middle-income countries. Babyl provided Rwanda’s first nationwide telemedicine service from 2019 to September 2023, integrating nurse-led triage with physician oversight, e-prescriptions, and national health insurance. Despite processing 3.9 million consultations, evidence on population-level impacts of scaled telemedicine services like Babyl in sub-Saharan Africa remains limited. This study aimed to quantify the association between national-scale telemedicine implementation and facility-based healthcare utilization for common primary care conditions in Rwanda, using interrupted time series analysis to estimate immediate and sustained effects across introduction and discontinuation periods.
Methods
We integrated deidentified administrative data from Babyl (n = 3,899,788 consultations), Rwanda’s Health Management Information System (2015–2024). We employed two analytical approaches: (1) descriptive analysis of user demographics, insurance coverage, and clinical characteristics; (2) segmented regression with interrupted time series modeling using ordinary least squares with Newey-West heteroskedasticity- and autocorrelation-consistent standard errors to quantify level and slope changes across pre-intervention, intervention, and post-discontinuation periods for gastritis, diarrhea, urinary tract infections, malaria, and respiratory infections.
Results
The platform recorded 3.90 million consultations (2019–2023), with 75.4% covered by community-based health insurance and 54.7% among female patients. Task-shifting was substantial: triage nurses managed 44.2% of consultations, senior nurses 25.6%, and general practitioners 30.2%. Interrupted time series analysis revealed immediate reductions in facility-based cases following Babyl’s introduction: respiratory infections decreased by 1055 cases (95% CI -1098 to -1011; P < .001), malaria by 246 cases (95% CI -258 to -234; P < .001), gastritis by 137 cases (95% CI -146 to -127; P < .001), and urinary tract infections by 114 cases (95% CI -124 to -105; P < .001). Post-discontinuation, monthly increases ranged from 1 case (gastritis, diarrhea, urinary tract infections) to 10 cases (respiratory infections), suggesting demand reversal to facilities.
Conclusions
National-scale telemedicine implementation was associated with substantial reductions in facility-based consultations for common conditions and successful task-shifting to nurses. The post-discontinuation reversal patterns underscore telemedicine’s role in healthcare access. Future digital health initiatives should prioritize sustainable financing, system interoperability, and regulatory frameworks to maintain service continuity.
Supplementary Information
The online version contains supplementary material available at 10.1186/s12875-026-03179-8.
Keywords: Telemedicine, Digital health, Interrupted time series, Primary health care, Task-shifting, Health systems, Rwanda, Low- and middle-income countries
Introduction
Primary healthcare systems in low- and middle-income countries face persistent challenges in ensuring timely, accessible care [1–3]. Geographic barriers, workforce shortages, and infrastructure limitations contribute to delayed diagnoses, treatment gaps, and preventable morbidity [4, 5]. The COVID-19 pandemic accelerated global adoption of telemedicine, with evidence from multiple settings demonstrating feasibility for first-contact care, though rigorous population-level evaluations in sub-Saharan Africa remain scarce [6–8].
Rwanda has pursued universal health coverage through community-based health insurance (Mutuelle de Santé), achieving 79% population coverage by 2019 [9, 10].The country’s digital infrastructure investments, including 4G network expansion and the Irembo e-governance platform, created favorable conditions for digital health implementation [11, 12]. Babyl was initially launched in 2016 and was implemented as Rwanda’s first payer-integrated telemedicine service, offering free consultations to insured citizens via mobile phones until September 2023 [13, 14]. The Babyl platform dataset available for this study covered June 2019 to September 2023, which defines the observation window used in our analyses.
The Babyl platform employed a structured clinical workflow operating from central call centers where doctors and nurses provided approximately 3,000 daily consultations through a three tiered system: triage nurses (44.2% of consultations) conducted initial assessments using standardized protocols, escalating complex cases to senior nurses (25.6%) or general practitioners (30.2%) [15]. All providers received training for Babyl service provision, covering clinical protocols, platform use, and telemedicine-specific competencies, with ongoing in-service training organized by Babyl. The clinical workflow included patient enrolment via USSD codes, telephone-based consultation, electronic prescription transmitted via SMS to patients and pharmacies, laboratory test ordering through the integrated HMIS platform, and follow-up consultations as needed [15, 16]. Quality assurance was maintained through standardized clinical guidelines aligned with Ministry of Health protocols, with ongoing monitoring and in-service training organized by Babyl. The system integrated e-prescriptions, laboratory referrals, and sick-note issuance with the national Health Management Information System, enabling continuity between virtual and facility-based care [16]. Early evaluations documented high user satisfaction and clinical appropriateness for common conditions but did not assess system-wide impacts [17, 18].
Task-shifting represents a critical mechanism for expanding healthcare capacity in resource-constrained settings [19–21]. Evidence from Uganda, Kenya, and South Africa demonstrates that nurse-led management of uncomplicated conditions maintains quality while reducing physician workload [22–24]. In virtual care models, structured triage protocols and decision support tools may further enable safe task delegation [25].
When randomization is infeasible for national-level interventions, interrupted time series designs provide robust causal inference by comparing outcomes before and after clearly defined implementation points [26]. This approach controls for underlying temporal trends while estimating both immediate level changes and sustained slope modifications [22, 27].
This study aimed to quantify the association between Rwanda’s babyl telemedicine service and facility-based utilization for common ambulatory conditions using interrupted time series analysis to model immediate and sustained effects across program introduction and discontinuation periods.
Methods
Study design and setting
We conducted a retrospective observational study analyzing the impact of Babyl’s telemedicine services on healthcare utilization in Rwanda. The study employed interrupted time series analysis with three distinct periods: pre-intervention (January 2015-May 2019), intervention (June 2019-September 2023), and post-intervention (October 2023-July 2024).
Data sources
We integrated two administrative databases:
Babyl platform database from Irembo: Deidentified records of 3,899,788 teleconsultations including patient demographics (age, sex), insurance type, consultation timestamps, International Classification of Diseases 10th Revision diagnoses, clinician type, and prescription data. Although Babyl launched in 2016, only records from June 2019 to September 2023 were available for this study and were used to define the analytic observation window.
Health Management Information System: Monthly aggregated facility-based consultation counts by district, condition, and patient sex from all public health facilities (n = 502 health centers, 42 district hospitals).
Population and eligibility
The study population included all individuals seeking care for five selected conditions: respiratory infections, malaria, gastritis, urinary tract infections, and diarrhea. These conditions were selected based on disease burden (accounting for 42% of Babyl consultations) and data completeness in both databases.
Variables and outcomes
Primary outcome: Monthly facility-based consultation counts for each condition.
Exposure: Babyl service availability (binary: pre-intervention = 0, intervention/post-intervention = 1).
Covariates: Time (continuous months from January 2015), seasonality (monthly indicators), COVID-19 pandemic period (March 2020-December 2021).
Statistical analysis
Descriptive analysis
We summarized Babyl platform utilization by analyzing patient demographics, service delivery patterns, and clinical conditions managed. We grouped users by sex, age category, and insurance status (Mutuelle/CBHI, RSSB/RAMA, or other/unknown). For each group, we calculated total consultations and the proportion completed versus missed.
We classified clinicians into three cadres: triage nurses, senior nurses, and general practitioners and examined the distribution of consultations managed by each. We also identified the most frequent presenting complaints and diagnoses, with a focus on five outpatient conditions targeted in the time series models: respiratory infections, malaria, urinary tract infections, gastritis, and diarrhea.
Interrupted time series analysis
We applied segmented regression using the following specification:
 |
Where:
: The observed monthly outcome at time tt, such as the number of consultations for a specific disease (e.g., diarrhea or malaria).
: The intercept: baseline level of the outcome at the start of the time series (January 2015).
The underlying time trend before Babyl’s implementation: this captures whether outcomes were increasing or decreasing monthly before any intervention.The immediate level change at the time of Babyl’s introduction (June 2019). It captures any abrupt increase or decreases in the outcome following the start of the intervention.
The change in trend (slope)
after the Babyl launch. It estimates whether the rate of change in outcomes accelerated or decelerated during the intervention.
The level change at the point of Babyl suspension (September 2023). This models any sudden shift in the outcome once Babyl was stopped.
: The change in trend (slope) after Babyl’s suspension. It indicates whether trends increased or decreased after the intervention.
Dummy variables for calendar months (February through December), used to control for seasonality. January is the reference month.
: The error term, capturing random variation not explained by the model (assumed to be normally distributed, potentially autocorrelated).
We estimated models using ordinary least squares with Newey-West standard errors (lag = 3) to account for autocorrelation and heteroskedasticity. Sensitivity analyses included Prais-Winsten regression with first-order autocorrelation and models with alternative lag specifications. The entire framework of the analysis was implemented using the R version 4.3.1 (R Foundation for Statistical Computing).
Results
From 2019 to 2023, a total of 3,899,788 Babyl consultations were recorded, with pronounced variation across years. Consultation volumes rose from 627,430 in 2019 (16.09% of all consultations during the study period) to 743,851 in 2020 (19.07%), indicating rapid early uptake and expansion of service use. Utilization peaked in 2021, reaching 1,392,937 consultations (35.72%), representing the highest annual volume observed across the series. Thereafter, consultations declined to 809,936 in 2022 (20.77%) and fell further to 325,634 in 2023 (8.35%) (Fig. 1). The increase through 2021 suggests strong scale-up and sustained demand during the early years of implementation, while the reduction after 2021 indicates a contraction in utilization relative to the peak; the sharp drop in 2023 is consistent with reduced service availability and/or use during the year, including the discontinuation of Babyl operations in September 2023 (Fig. 1).
Fig. 1.
Distribution of consultations by year
Descriptive characteristics
Overall, consultation completion rates were consistently high, exceeding 94% across all categories, which reflects strong service utilization and effective appointment management. Triage nurses delivered the highest proportion of consultations (44.9%), yet they also had the highest no-show rate (3.6%) compared with general practitioners (1.0%) and senior nurses (1.1%). This suggests that triage-level appointments may be more susceptible to missed visits.
Mutuelle beneficiaries represented the majority of service users (74.9%) and accounted for the largest share of no-shows (4.1%). In contrast, clients insured under RSSB/RAMA showed greater attendance reliability, with a no-show rate of just 0.5%.Young adults aged 18–35 years comprised nearly half of all consultations (48.0%) and had the highest no-show rate (2.9%). No-show rates decreased progressively with age, reaching a low of 0.4% among clients aged ≥ 60 years. Females(54.7%) accessed services more frequently than males (45.3%) and exhibited a slightly higher no-show rate 2.9% and 2.5% respectively (Table 1).
Table 1.
Consultation completion status
| Variable | Consultation status | Total | |
|---|---|---|---|
| Completed | No show | ||
| Clinician Type | |||
| GP | 1,128,403 (28.9%) | 39,064 (1.0%) | 1,167,467 (29.9%) |
| Senior Nurse | 936,512 (24.0%) | 42,922 (1.1%) | 979,434 (25.1%) |
| Triage Nurse | 1,611,470 (41.3%) | 141,417 (3.6%) | 1,752,887 (44.9%) |
| Total | 3,676,385 (94.3%) | 223,403 (5.7%) | 3,899,788 (100.0%) |
| Insuarance | |||
| Mutuelle | 2,728,459 (70.8%) | 158,163 (4.1%) | 2,886,622 (74.9%) |
| Not identified | 517,369 (13.4%) | 36,112 (0.9%) | 553,481 (14.4%) |
| RSSB/RAMA | 392,640 (10.2%) | 18,740 (0.5%) | 411,380 (10.7%) |
| Total | 3,638,468 (94.5%) | 213,015 (5.5%) | 3,851,483 (100.0%) |
| Age group | |||
| 18–35 | 1,757,108 (45.1%) | 113,720 (2.9%) | 1,870,828 (48.0%) |
| 36–44 | 986,686 (25.3%) | 56,259 (1.4%) | 1,042,945 (26.7%) |
| 45–59 | 677,825 (17.4%) | 37,792 (1.0%) | 715,617 (18.4%) |
| 60+ | 254,314 (6.5%) | 15,570 (0.4%) | 269,884 (6.9%) |
| Unknown | 452 (0.0%) | 62 (0.0%) | 514 (0.0%) |
| Total | 3,676,385 (94.3%) | 223,403 (5.7%) | 3,899,788 (100.0%) |
| Gender | |||
| Female | 1,982,781 (51.8%) | 111,404 (2.9%) | 2,094,185 (54.7%) |
| Male | 1,640,144 (42.8%) | 96,578 (2.5%) | 1,736,722 (45.3%) |
| Total | 3,622,925 (94.6%) | 207,982 (5.4%) | 3,830,907 (100.0%) |
Interrupted time Series(ITS) analysis: infection trends by diseases
The Interrupted Time Series (ITS) analysis shows a clear reduction in disease cases across all categories over time, with noticeable changes between the pre-intervention, intervention, and post-intervention periods. Before the intervention (2015–May 2019), all diseases had high case numbers and large month-to-month variation. After the intervention started in June 2019, case numbers dropped quickly and remained lower and more stable throughout the implementation period (June 2019–August 2023). Respiratory infections showed the largest decrease, falling from peaks of around 5,000 cases before the intervention to much lower and steady levels during implementation. Malaria cases declined sharply after June 2019, reaching almost zero during 2020 and 2021, followed by a small increase after the intervention ended. Gastritis, urinary tract infections, and diarrhea also showed immediate drops at the start of the intervention, moving from highly variable pre-intervention patterns, often with extreme values, to more consistent and lower case levels during the intervention period. In the post-intervention period (from September 2023 onward), trends for some diseases, especially malaria and diarrhea, show a slight upward movement; however, case numbers for all diseases remained well below the levels observed before the intervention (Fig. 2).
Fig. 2.
Interrupted Time Series panel of monthly infection cases in Rwanda’s health facilities pre-Babyl (2015–2019), during Babyl operation (2019-September 2023), and post-discontinuation (October 2023–2024)
Interrupted time series (ITS) regression results by disease
The interrupted time series analysis of respiratory infections showed a baseline intercept of 1,396.81 cases (95% CI: 1,390.14–1,403.48; p < .001) and a declining pre-intervention trend of − 10.71 cases per month (95% CI: −10.84 to − 10.58; p < .001). Babyl implementation was associated with a substantial immediate level reduction of − 899.18 cases (95% CI: −908.32 to − 890.04; p < .001) and a slope increase of + 10.25 cases per month (95% CI: 10.10–10.40; p < .001). Seasonal peaks were observed in March (+ 107.67) and May (+ 82.55), with troughs in August (− 169.94) and September (− 149.02). The COVID-19 pandemic period was associated with a reduction of − 62.28 cases (95% CI: −64.59 to − 59.97; p < .001), while lockdowns were associated with an increase of + 51.90 cases (95% CI: 46.31–57.48; p < .001) (see Additional file 1).
Malaria cases demonstrated a pre-intervention decline of − 2.43 cases per month (95% CI: −2.46 to − 2.40; p < .001) from a baseline level of 285.72 cases (95% CI: 284.16–287.29; p < .001). The introduction of Babyl was associated with an immediate level decrease of − 256.37 cases (95% CI: −258.52 to − 254.23; p < .001) and a subsequent slope increase of + 2.57 cases per month (95% CI: 2.53–2.60; p < .001). Seasonal lows occurred in August (− 56.34) and July (− 48.34), with a smaller decline in December (− 16.51). The COVID-19 pandemic period was associated with an increase of + 4.66 cases (95% CI: 4.12–5.20; p < .001), whereas lockdowns corresponded to a reduction of − 1.78 cases (95% CI: −3.09 to − 0.47; p = .01) (see Additional file 2).
For gastritis, the baseline level was 234.73 cases (95% CI: 233.21–236.24; p < .001), with a declining pre-intervention trend of − 1.27 cases per month (95% CI: −1.30 to − 1.24; p < .001). Babyl implementation resulted in an immediate level decrease of − 123.15 cases (95% CI: −125.22 to − 121.07; p < .001), followed by a positive slope change of + 0.93 cases per month during the intervention period (95% CI: 0.89–0.96; p < .001). Seasonal peaks were observed in June (+ 13.34) and May (+ 3.57), with troughs in August (− 22.31) and July (− 18.82). The COVID-19 period was associated with a reduction of − 5.50 cases (95% CI: −6.03 to − 4.98; p < .001), while lockdowns were linked to an increase of + 1.40 cases (95% CI: 0.13–2.67; p = .03) (see Additional file 3).
UTI cases had a baseline level of 148.57 cases (95% CI: 146.85–150.30; p < .001) and a near-flat pre-intervention trend of − 0.10 cases per month (95% CI: −0.14 to − 0.07; p < .001). The intervention was associated with an immediate decline of − 106.20 cases (95% CI: −108.56 to − 103.83; p < .001) and a slope increase of + 0.92 cases per month (95% CI: 0.88–0.96; p < .001). Seasonal effects included a modest peak in June (+ 3.28) and troughs in August (− 20.12) and July (− 18.80). Pandemic effects showed a reduction of − 1.40 cases (95% CI: −2.00 to − 0.80; p < .001), while lockdowns contributed an additional decrease of − 4.61 cases (95% CI: −6.06 to − 3.17; p < .001) (see Additional file 4).
Diarrhea consultations had a baseline level of 97.15 cases (95% CI: 96.39–97.91; p < .001) and a declining pre-intervention trend of − 0.64 cases per month (95% CI: −0.66 to − 0.63; p < .001). Babyl implementation was associated with an immediate level reduction of − 67.00 cases (95% CI: −68.04 to − 65.95; p < .001) and a slope increase of + 0.68 cases per month (95% CI: 0.67–0.70; p < .001). Seasonal variation was modest, with the largest reductions in August (− 11.49) and July (− 10.65). The COVID-19 period was associated with − 2.54 cases (95% CI: −2.80 to − 2.27; p < .001), and lockdowns added a further reduction of − 1.89 cases (95% CI: −2.52 to − 1.25; p < .001); the COVID-19 case variable itself was not statistically significant (p = .62) (see Additional file 5).
Service discontinuation effects
Following Babyl’s closure in September 2023, facility consultation volumes increased across all analyzed conditions. The most pronounced effects occurred for respiratory infections (+ 809 monthly cases) and malaria (+ 256 monthly cases) in the immediate post-discontinuation period. These increases exceeded pre-intervention baselines, suggesting patients who had adopted telemedicine were forced back to facilities, potentially creating additional burden on the conventional system.
Discussion
This population-level analysis demonstrates that babyl telemedicine implementation in Rwanda was associated with reductions in facility-based consultations for common primary care conditions [1, 2], with respiratory infections showing the most decrease of 1,055 monthly cases (95% CI − 1098 to − 1011; P < .001) immediately following Babyl’s introduction. The magnitude of this effect representing approximately a 75% reduction from baseline levels exceeds reductions reported in similar interventions in Kenya (32% reduction) and Uganda (45%) for comparable conditions [1, 8, 28]. The task-shifting to nurses, who managed 69.8% of all consultations, surpassed World Health Organization targets of 50% for nurse-led primary care management in resource-limited settings [20, 28]. The temporal dynamics revealed through our interrupted time series analysis provide insights into service adoption patterns. The pre-intervention declining trends for respiratory infections (− 10.71 cases/month) and malaria (− 2.43 cases/month) likely reflected ongoing health system strengthening efforts, yet the immediate level changes at Babyl’s introduction far exceeded these gradual improvements. The post-intervention slope changes, particularly for respiratory infections (+ 10.25 cases/month during operation), suggest a reversion toward in-person care pathways [21, 24].
Our findings extend previous evidence from digital health implementations in several important ways. While Raza et al. (2021) reported 23% consultation diversion from facilities through Pakistan’s telemedicine program (39), our observed reductions of 75% for respiratory infections and 90% for malaria represent substantially larger impacts [29–32]. This difference likely stems from Rwanda’s unique enabling environment: 75.4% insurance coverage through community-based schemes compared to Pakistan’s 12% coverage, and mobile phone penetration of 83% versus 65% [8, 28, 32–34].
The task-shifting outcomes align with but exceed those from India’s eSanjeevani platform, where nurses managed 45% of consultations compared to our observed 69.8% [29, 30]. Critical differences include Rwanda’s standardized triage protocols and real-time physician backup, which may have increased provider confidence in nurse-led management [20, 29, 30]. The high completion rates (94.3%) contrast sharply with dropout rates of 15–30% reported in Nigeria’s digital health pilots, suggesting that integration with insurance systems and e-prescriptions enhanced service continuity [3].
The post-discontinuation period provides rare empirical evidence of digital health service withdrawal impacts. The immediate monthly increases 809 cases for respiratory infections and 256 for malaria exceeded pre-intervention baselines by 15% and 22% respectively, suggesting a “rebound effect” where suppressed demand returned to facilities with additional accumulated need [21, 24]. This pattern contrasts with gradual service transitions observed when Ghana’s mobile health program ended with six-month phase-out period, which showed only 5% increases in facility visits [34, 35].
The differential rebound magnitudes across conditions offer insights into care-seeking elasticity. Conditions with larger immediate drops at introduction (respiratory infections: −1,055; malaria: −246) showed proportionally larger rebounds, while conditions with modest initial impacts (diarrhea: −69) demonstrated minimal post-discontinuation changes. This suggests that patients may have developed condition-specific preferences for care modalities, with respiratory symptoms being particularly amenable to virtual consultation [1, 2].
Our findings generate three evidence-based recommendations for digital health policy:
First, sustainable financing architectures must be established before scale-up. The abrupt service discontinuation and subsequent 15–22% increase in facility visits above baseline demonstrates the risks of donor-dependent models [28].
Second, nurse-led triage protocols require formalization and continuous quality assurance. The 44.2% of consultations managed entirely by triage nurses without escalation suggests substantial efficiency gains, potentially freeing 8,750 physician hours monthly (assuming 15-minute consultations). However, this necessitates competency-based training programs, algorithmic decision support, and liability frameworks that protect nurses operating within defined scopes [20].
Third, integration depth determines impact magnitude. Babyl’s connections to insurance databases, laboratory networks, and pharmacy systems enabled the observed high completion rates and prescription fulfillment. Countries should prioritize interoperability standards and data governance frameworks before launching digital health services. Our analysis suggests that each additional system integration (insurance, labs, pharmacy) was associated with 12–15% increases in consultation completion rates [1, 28, 31].
The condition-specific patterns observed have important clinical implications. Respiratory infections, showing the largest absolute reduction, appear ideally suited for telemedicine management given their symptom-based diagnosis and standardized treatment protocols. The 94.3% completion rate for respiratory consultations, compared to 87% for malaria cases requiring laboratory confirmation, supports this interpretation. This suggests that digital health programs should initially focus on conditions amenable to remote diagnosis while developing hybrid models for conditions requiring physical examination or diagnostics [2, 24, 33].
The September 2023 service discontinuation raises important considerations regarding beneficence and non-maleficence in digital health implementation. The platform provided 3.9 million consultations over six years, with 75.4% of users covered by community-based health insurance, indicating substantial population-level dependence on the service. Post-discontinuation analysis revealed facility consultations increased 15–22% above pre-intervention baselines, suggesting patients who had adopted telemedicine were forced to revert to facility-based care, potentially experiencing care disruption and increased access barriers. When digital health services demonstrate clear benefits, evidenced by high utilization rates and 94.3% completion rates, sustainable financing mechanisms must be established before national-scale deployment [15, 16]. Service withdrawal may disadvantage populations who integrated these platforms into their care-seeking patterns. Future implementations should incorporate planned transitions, advance notification systems, and sustainable financing to minimize harm to service-dependent populations. These considerations extend beyond Rwanda to inform global digital health policy, particularly as governments increasingly adopt telemedicine as a core component of universal health coverage strategies.
This study’s methodological strengths include the use of population-level data capturing 3.9 million consultations and a quasi-experimental design. The 114-month observation period provided sufficient power to detect modest effect sizes while controlling for seasonal variations and pandemic disruptions. However, several limitations constrain causal interpretation. The ecological design precludes individual-level inference about care-seeking decisions or clinical outcomes. We could not assess whether reduced facility visits represented appropriate substitution or potentially delayed care for serious conditions. The absence of mortality and morbidity data prevents evaluation of health impacts beyond utilization metrics. Additionally, concurrent interventions including community health worker programs (expanded 2018), facility renovations (2019–2020), and COVID-19 responses may confound observed associations despite our statistical adjustments. Data quality limitations include potential undercounting of private sector utilization, which comprises approximately 20% of outpatient care but lacks systematic reporting [28, 31]. We were unable to assess the economic costs of Babyl’s service provision or perform cost-effectiveness comparisons with facility-based care due to unavailability of comprehensive financial data from the platform’s operations. Available evidence from previous evaluations indicates that operational costs exceeded those of conventional care, with healthcare provider salaries comprising 70–80% of expenditure [15, 16]. The sustainability challenges that led to service discontinuation in September 2023 were attributed to dependency on external donor funding without establishment of sustainable government financing mechanisms. These findings underscore that Future evaluations of digital health interventions should prioritize incorporate comprehensive economic analyses, including cost-effectiveness and budget impact assessments, to inform policy decisions regarding resource allocation and sustainable financing architectures .
Conclusions
This comprehensive evaluation of Rwanda’s national telemedicine program provides robust evidence that digital health services can achieve substantial impacts on healthcare delivery patterns in low-resource settings [1, 2]. The immediate reduction of over 1,000 monthly respiratory infection consultations at facilities, successful management of 69.8% of cases by nurses, and high service completion rates (94.3%) demonstrate that well-designed telemedicine programs can effectively decompress overburdened health systems while maintaining care access [20].
The post-discontinuation analysis offers crucial lessons for sustainability. The rapid reversal of utilization gains with facility visits rebounding 15–22% above pre-intervention baselines underscores that digital health services fulfill genuine healthcare needs rather than creating artificial demand. This “rebound effect” should motivate policymakers to prioritize service continuity through diversified financing mechanisms and regulatory frameworks that ensure long-term viability [28, 34, 35].
Our spatial analysis revealing three-fold variations in district-level impacts highlights that digital health is not uniformly accessible or effective across populations. Future implementations must address these disparities through targeted infrastructure investments, community engagement, and locally adapted service models. The minimal pediatric utilization (0.01%) identifies gap requiring platform redesign and provider training to meet children’s healthcare needs [1]. As countries pursue universal health coverage amid health workforce shortages and growing disease burdens, Rwanda’s experience demonstrates both the transformative potential and implementation prerequisites for digital health. Success requires more than technological deployment; it demands systematic integration with existing health infrastructure, sustainable financing models, appropriate regulatory frameworks, and continuous adaptation based on utilization patterns and health outcomes. The evidence presented here should guide other nations in designing, implementing, and sustaining digital health interventions that meaningfully improve population health while strengthening health systems for long-term resilience.
Supplementary Information
Additional file 1. Interrupted time series regression for respiratory infections: monthly case counts in Rwanda’s health facilities (2015–2024). Segmented regression coefficients supporting the respiratory infections ITS model, including baseline level and pre-intervention trend, immediate level change at Babyl implementation, post-intervention slope change, month (seasonality) indicators, and COVID-19 period and lockdown covariates (with 95% CIs and p-values).
Additional file 2. Interrupted time series regression coefficients for malaria (2015–2024). Segmented regression coefficients supporting the malaria ITS model, including baseline level and pre-intervention trend, immediate level change at Babyl implementation, post-intervention slope change, month (seasonality) indicators, and COVID-19 period and lockdown covariates (with 95% CIs and p-values).
Additional file 3. Regression analysis results for gastritis: interrupted time series model with seasonal and pandemic controls (2015–2024). Segmented regression coefficients supporting the gastritis ITS model, including baseline level and pre-intervention trend, immediate level change at Babyl implementation, post-intervention slope change, month (seasonality) indicators, and COVID-19 period and lockdown covariates (with 95% CIs and p-values).
Additional file 4. Interrupted time series regression model for urinary tract infections (UTI): coefficients and significance tests (2015–2024). Segmented regression coefficients supporting the UTI ITS model, including baseline level and pre-intervention trend, immediate level change at Babyl implementation, post-intervention slope change, month (seasonality) indicators, and COVID-19 period and lockdown covariates (with 95% CIs and p-values).
Additional file 5. Regression analysis results for diarrhea: interrupted time series model with seasonal and pandemic controls (2015–2024). Segmented regression coefficients supporting the diarrhea ITS model, including baseline level and pre-intervention trend, immediate level change at Babyl implementation, post-intervention slope change, month (seasonality) indicators, and COVID-19 period and lockdown covariates (with 95% CIs and p-values), including the non-significant COVID-19 case term.
Acknowledgements
We thank the Rwanda Ministry of Health for granting access to the Health Management Information System (HMIS) data and for their support throughout this research. We acknowledge Babyl Rwanda for providing access to the telemedicine platform database that made this analysis possible. We are grateful to the University of Rwanda School of Public Health and CIIC-HIN for their institutional support in conducting this research. We also acknowledge the Bill & Melinda Gates Foundation for funding the broader evaluation project of which this quantitative study was a component. Finally, we thank all healthcare providers and patients whose de-identified consultation data contributed to this analysis.
Abbreviations
- AI
Artificial Intelligence
- BMGF
Bill & Melinda Gates Foundation
- CBHI
Community-Based Health Insurance
- CI
Confidence Interval
- COVID
19-Coronavirus Disease 2019
- HC
Health Center
- HMIS
Health Management Information System
- LMIC
Low-and Middle-Income Country
- MOH
Ministry of Health
- OPD
Outpatient Department
- RAMA
Régime d’Assurance Maladie de l’Armée Rwandaise
- RSSB
Rwanda Social Security Board
- SMS
Short Message Service
- WHA
World Health Assembly
- WHO
World Health Organization
Authors’ contributions
The original draft was written by F.R.K., J.H., J.C., and G.A. The methodology and formal analysis were developed and performed by F.R.K., J.H., J.C., T.C.U., J.D.H., E.N.C., I.P., and G.A. The manuscript underwent review and editing by F.R.K., G.A., Y.D.N., J.C., M.S., E.R., and J.H. All authors contributed to the conceptualization and design of the study. Data collection and management were supervised by J.H. and J.C. Critical revisions and intellectual content were provided by all co-authors. All authors reviewed and approved the final manuscript for publication.
Funding
This research was conducted as part of a larger evaluation project funded by the Bill & Melinda Gates Foundation (BMGF). The funders had no role in the study design, data collection, analysis, interpretation, or manuscript preparation.
Data availability
The aggregated data that support the findings of this study are included in the manuscript in the form of summary statistics, tables, and figures. Individual-level de-identified data from the Babyl platform and HMIS cannot be shared publicly due to data sharing agreements with the data custodians and ethical considerations regarding patient confidentiality. However, the statistical analysis code and study protocol are available from the corresponding author upon reasonable request, subject to approval by the ethical review board and data custodians.
Declarations
Ethics approval and consent to participate
Ethical approval was obtained from the Rwanda National Ethics Committee (approval number #00001973) prior to data collection commencement. This study was conducted in accordance with the principles of the Declaration of Helsinki. The study utilized de-identified secondary data from the Babyl platform database and the Health Management Information System (HMIS), which included aggregated facility-based consultation records. As this was a retrospective analysis of de-identified administrative health data, individual informed consent was not required per institutional ethical guidelines. The study adhered to international ethical standards for health data research, ensuring all data were anonymized with no personally identifiable information accessed. Data security protocols were maintained with access restricted to authorized research team members. The study was designed to generate evidence for digital health service improvement without causing harm, adhering to principles of beneficence and non-maleficence.
Consent for publication
Not applicable.
Competing interests
The authors declare no competing interests.
Footnotes
Publisher’s note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
References
- 1.Dodoo JE, Al-Samarraie H, Alsswey A. The development of telemedicine programs in Sub-Saharan africa: progress and associated challenges. Health Technol (Berl) Springer Berlin Heidelberg. 2022;12(1):33–46. 10.1007/s12553-021-00626-7. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 2.Mahmoud K, Jaramillo C, Barteit S. Telemedicine in Low- and Middle-Income countries during the COVID-19 pandemic: A scoping review. Front Public Heal. 2022;10(June):1–15. 10.3389/fpubh.2022.914423. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 3.Agbeyangi AO, Lukose JM. Telemedicine adoption and prospects in Sub-Sahara africa: A systematic review with a focus on South Africa, Kenya, and Nigeria. Healthc. 2025;13(7):1–25. 10.3390/healthcare13070762. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4.Campbell K, Greenfield G, Li E, O’Brien N, Hayhoe B, Beaney T, Majeed A, Neves AL. The impact of virtual consultations on the quality of primary care: systematic review. J Med Internet Res. 2023;25:1–17. PMID:37647117. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5.Kruse CS, Williams K, Bohls J, Shamsi W. Telemedicine and health policy: A systematic review. Heal Policy Technol Elsevier Ltd. 2021;10(1):209–29. 10.1016/j.hlpt.2020.10.006. [Google Scholar]
- 6.Kemp M, Rising KL, Laynor G, Miao J, Worster B, Chang AM, Monick AJ, Guth A, Esteves Camacho T, McIntosh K, Amadio G, Shughart L, Hsiao TA, Leader AE. Barriers to telehealth uptake and use: a scoping review. JAMIA Open. 2025;8(2). 10.1093/jamiaopen/ooaf019. [DOI] [PMC free article] [PubMed]
- 7.Gijsbers H, Feenstra TM, Eminovic N, Van Dam D, Nurmohamed SA, Van De Belt T, Schijven MP. Enablers and barriers in upscaling telemonitoring across geographic boundaries: A scoping review. BMJ Open 2022;12(4). PMID:35443957. 10.1136/bmjopen-2021-057494. [DOI] [PMC free article] [PubMed]
- 8.Chemouni B. The political path to universal health coverage: Power, ideas and community-based health insurance in Rwanda. World Dev Elsevier Ltd. 2018;106:87–98. 10.1016/j.worlddev.2018.01.023. [Google Scholar]
- 9.Nyandekwe M, Nzayirambaho M, Kakoma JB. Universal health coverage in rwanda: dream or reality. Pan Afr Med J 2014;17. PMID:25170376. 10.11604/pamj.2014.17.232.3148. [DOI] [PMC free article] [PubMed]
- 10.Nyandekwe M, Nzayirambaho M, Kakoma JB. Universal health insurance in Rwanda: major challenges and solutions for financial sustainability case study of Rwanda community-based health insurance part i. Pan Afr Med J. 2020;37(55):1–12. PMID:33209182. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11.Lu C, Chin B, Lewandowski JL, Basinga P, Hirschhorn LR, Hill K, Murray M, Binagwaho A. Towards universal health coverage: an evaluation of Rwanda mutuelles in its first eight years. PLoS ONE 2012;7(6). PMID:22723985. 10.1371/journal.pone.0039282. [DOI] [PMC free article] [PubMed]
- 12.Eze P, Ilechukwu S, Lawani LO. Impact of community-based health insurance in low- and middle-income countries: A systematic review and meta-analysis. PLoS ONE. 2023. PMID:37368882ISBN:1111111111. 10.1371/journal.pone.0287600. [DOI] [PMC free article] [PubMed]
- 13.Nyakabau M-R. Digital-first integrated care: Rwanda’s innovative digital health care service. 2021. https://transformhealthcoalition.org/wp-content/uploads/2021/11/Rwanda.pdf.
- 14.Woldemichael A, Gurara D, Shimeles A, Mansoor A. The impact of community based health insurance schemes on Out-of-Pocket healthcare spending: evidence from Rwanda the impact of community based health insurance schemes on Out-of-Pocket healthcare spending: evidence from Rwanda. 2019. https://go.stlpartners.com/Telehealth-in-emerging-markets-Babyl-closes-the-gap-in-Rwandan-healthcare-inequality-PDF.
- 15.STL Partners. Telehealth in emerging markets: Babyl closes the gap in Rwandan healthcare inequality. 2020;(September). https://percept.co.za/wp-content/uploads/2021/05/Brief-3-babyl.pdf.
- 16.Consultants PA. and. Telehealth Innovation in Rwanda: Lessons and Insights from the Babyl Experience. 2021;(April):14. https://percept.co.za/wp-content/uploads/2021/05/Brief-3-babyl.pdf.
- 17.Müller A, Cau A, Muhammed S, Abdullahi O, Hayward A, Nsanzimana S, Lester R. Digital mHealth and virtual care use during COVID-19 in 4 countries: rapid landscape review. JMIR Form Res. 2022;6(11). 10.2196/26041. [DOI] [PMC free article] [PubMed]
- 18.Getler P, Alex W, Ada K, Condo J. The Effect of Telemedicine on Quality of Care in Rwanda : Comprehensive Report of the Analysis of SP and Provider Survey Data. 2024. https://ciichin.org/wp-content/uploads/2025/01/The-Effect-of-Telemedicine-on-Quality-of-Care-in-Rwanda.pdf.
- 19.Leong SL, Teoh SL, Fun WH, Lee SWH. Task shifting in primary care to tackle healthcare worker shortages: an umbrella review. Eur J Gen Pract Taylor Francis. 2021;27(1):198–210. PMID:34334095. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20.Paier-Abuzahra M, Posch N, Jeitler K, Semlitsch T, Radl-Karimi C, Spary-Kainz U, Horvath K, Siebenhofer A. Effects of task-shifting from primary care physicians to nurses: an overview of systematic reviews. Hum Resour Health BioMed Cent; 2024;22(1). PMID:39529012. 10.1186/s12960-024-00956-3. [DOI] [PMC free article] [PubMed]
- 21.Bernal JL, Cummins S, Gasparrini A. Interrupted time series regression for the evaluation of public health interventions: a tutorial. Int J Epidemiol Engl. 2017;46(1):348–55. PMID:27283160. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 22.Taljaard M, McKenzie JE, Ramsay CR, Grimshaw JM. The use of segmented regression in analysing interrupted time series studies: an example in pre-hospital ambulance care. Implement Sci BioMed Cent Ltd. 2014;9(1):1–4. PMID:24943919. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 23.Wagner AK, Soumerai SB, Zhang F, Ross-Degnan D. Segmented regression analysis of interrupted time series studies in medication use research. J Clin Pharm Ther J Clin Pharm Ther. 2002;27(4):299–309. PMID:12174032. [DOI] [PubMed] [Google Scholar]
- 24.Penfold RB, Zhang F. Use of interrupted time series analysis in evaluating health care quality improvements. Acad Pediatr Elsevier. 2013;13(6 SUPPL):S38–44. PMID:24268083. [DOI] [PubMed] [Google Scholar]
- 25.Zhang Y, Ren Y, Huang Y, Yao M, Jia Y, Wang Y, Mei F, Zou K, Tan J, Sun X. Design and statistical analysis reporting among interrupted time series studies in drug utilization research: a cross-sectional survey. BMC Med Res Methodol BioMed Cent Ltd. 2024;24(1):62. PMID:38461257. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 26.He Y, Kouabenan YR, Assoa PH, Puttkammer N, Wagenaar BH, Xiao H, Gloyd S, Hoffman NG, Komena P, Kamelan NPF, Iiams-Hauser C, Pongathie AS, Kouakou A, Flowers J, Abiola N, Kohemun N, Amani JB, Adje-Toure C, Perrone LA. Laboratory data timeliness and completeness improves following implementation of an electronic laboratory information system in Côte d’ivoire: Quasi-Experimental study on 21 clinical laboratories from 2014 to 2020. JMIR Public Heal Surveill. 2024;10:1–12. PMID:38506899. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 27.Smith DC, Thumm EB, Anderson J, Kissler K, Reed SM, Centi SM, Staley AW, Hernandez TL, Barton AJ. Sudden shift to telehealth in COVID-19: A retrospective cohort study of disparities in use of telehealth for prenatal care in a large midwifery service. J Midwifery Women’s Heal BMC Med Res Methodol. 2024;69(4):522–30. PMID:38111228. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 28.WHO. Global strategy on digital health 2020–2025. Geneva: World Health Organization. 2021. Licence: CC BY-NC-SA 3.0 IGO. Who. 2021. ISBN:9789240020924.
- 29.{India Ministry of Health and Family Welfare}. Utilization of telemedicine /e-sanjeevani in the public health facilities of India. 2025.
- 30.Dastidar BG, Jani AR, Suri S, Nagaraja VH. Reimagining india’s National telemedicine service to improve access to care. Lancet Reg Heal - Southeast Asia Elsevier Ltd. 2024;30:100480. 10.1016/j.lansea.2024.100480. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 31.Waiyaiya E, Nyaga N, Nyangena J, Nyotu R, Wamicwe J, Langat B, Wanyee S. Digital health convergence workshop: the Kenyan experience. Oxf Open Digit Heal. 2024;(April)10.1093/oodh/oqae014. [DOI] [PMC free article] [PubMed]
- 32.Wasi Abbas M, Nawaz Tahir H, Jaffar N, Raza G, Aman Ali S, Jabeen R. Facilitators and barriers in acceptance of telemedicine among healthcare providers in pakistan: A cross-sectional survey. J Med Access. 2024;8. 10.1177/27550834241266413. [DOI] [PMC free article] [PubMed]
- 33.Ye J, He L, Beestrum M. Implications for implementation and adoption of telehealth in developing countries: a systematic review of china’s practices and experiences. Npj Digit Med Nat Res. 2023;6(1):1–14. 10.1038/S41746-023-00908-6;SUBJMETA. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 34.Kyei KA, Onajah GN, Daniels J. The emergence of telemedicine in a low-middle-income country: challenges and opportunities. Ecancermedicalscience Ecancer Global Foundation. 2024;18. 10.3332/ECANCER.2024.1679. [DOI] [PMC free article] [PubMed]
- 35.Dzando G, Akpeke H, Kumah A, Agada E, Lartey A, Nortu J, Nutakor H, Donyi A, Dordunu R. Telemedicine in ghana: insight into the past and present, a narrative review of literature amidst the coronavirus pandemic. J Public Health Afr. 2022;13(1):31–5. 10.4081/jphia.2022.2024. [DOI] [PMC free article] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Additional file 1. Interrupted time series regression for respiratory infections: monthly case counts in Rwanda’s health facilities (2015–2024). Segmented regression coefficients supporting the respiratory infections ITS model, including baseline level and pre-intervention trend, immediate level change at Babyl implementation, post-intervention slope change, month (seasonality) indicators, and COVID-19 period and lockdown covariates (with 95% CIs and p-values).
Additional file 2. Interrupted time series regression coefficients for malaria (2015–2024). Segmented regression coefficients supporting the malaria ITS model, including baseline level and pre-intervention trend, immediate level change at Babyl implementation, post-intervention slope change, month (seasonality) indicators, and COVID-19 period and lockdown covariates (with 95% CIs and p-values).
Additional file 3. Regression analysis results for gastritis: interrupted time series model with seasonal and pandemic controls (2015–2024). Segmented regression coefficients supporting the gastritis ITS model, including baseline level and pre-intervention trend, immediate level change at Babyl implementation, post-intervention slope change, month (seasonality) indicators, and COVID-19 period and lockdown covariates (with 95% CIs and p-values).
Additional file 4. Interrupted time series regression model for urinary tract infections (UTI): coefficients and significance tests (2015–2024). Segmented regression coefficients supporting the UTI ITS model, including baseline level and pre-intervention trend, immediate level change at Babyl implementation, post-intervention slope change, month (seasonality) indicators, and COVID-19 period and lockdown covariates (with 95% CIs and p-values).
Additional file 5. Regression analysis results for diarrhea: interrupted time series model with seasonal and pandemic controls (2015–2024). Segmented regression coefficients supporting the diarrhea ITS model, including baseline level and pre-intervention trend, immediate level change at Babyl implementation, post-intervention slope change, month (seasonality) indicators, and COVID-19 period and lockdown covariates (with 95% CIs and p-values), including the non-significant COVID-19 case term.
Data Availability Statement
The aggregated data that support the findings of this study are included in the manuscript in the form of summary statistics, tables, and figures. Individual-level de-identified data from the Babyl platform and HMIS cannot be shared publicly due to data sharing agreements with the data custodians and ethical considerations regarding patient confidentiality. However, the statistical analysis code and study protocol are available from the corresponding author upon reasonable request, subject to approval by the ethical review board and data custodians.