Digital Health Interventions and Virtual Care Models to Optimise Resource Allocation and Patient Outcomes in Post-Pandemic Healthcare Systems

2.5. Data Extraction

A structured Excel template was used to extract the data, including the author, year, country, study design, sample size, type of DHI, and outcome measures. Resource allocation was conceptualised as capacity, workforce, and cost deployment decisions in healthcare systems, with capacity outcomes including bed-days, length of stay, and hospital utilisation; workforce outcomes including clinician time and staffing workload; and financial outcomes including cost per episode and cost per bed-day. The predictive models were considered planning tools to assist decision-making, not mechanisms for direct savings. Two independent reviewers extracted data, and disagreements were resolved by consensus. Relative risks, odds ratios, mean differences, and 95 per cent confidence intervals were provided where appropriate; the direction of the effect was recorded narratively.

2.6. Conceptual Organising Framework

To plan synthesis at the health-systems level, the study’s findings were analysed using the input-process-output-outcome framework. The inputs included staffing, digital infrastructure, devices, and implementation resources; the processes included patient selection, monitoring intensity, escalation pathways, and integration into the clinical workflow; the outcomes included mortality, patient experience, readmissions, equity, and system resilience. There is an intersection between each type of intervention and this framework, allowing the organisation of an otherwise heterogeneous body of study through the systematic comparison of the narratives. Primary empirical studies informed the analytical synthesis, and contextual evidence sources were used to support understanding of policy and implementation contexts.

2.7. Data Analysis

The analysis method was a manual thematic synthesis based on the approach presented in the journal article [12]. Synthesis clusters are focused on five domains close to one another, including but not limited to the effects on resource allocation (e.g., length of stay, hospital capacity, cost-efficiency), effects on patient outcomes (e.g., mortality, hospitalisation readmissions, patient satisfaction, clinical improvements), and contextual enabling or hindering factors (e.g., digital infrastructure, equity issues, staff adoption). Codes were iteratively grouped into descriptive themes corresponding to the input–process–output–outcome framework. Themes were compared across studies and interpreted in relation to study design and methodological quality. Thematic clustering was able to integrate measure quantities and qualitative information.

2.8. Quality Assessment

The methodological quality of all studies, including [13–35] were evaluated using the Critical Appraisal Skills Programme (CASP), MMAT, and JBI checklists. It provides structured guidance for assessing validity, relevance, and rigour across qualitative and quantitative designs. Overall, the quality of the evidence was high or moderate and varied significantly due to study design, methodological transparency, and control of confounders. High-quality evaluations were most common in large systematic reviews of digital interventions and economic frameworks including Gentili et al. [21] Benedetto et al. [27]; Zakhiyah et al. [33]. Further, well-powered comparative and device-based RCT analyses showed strong internal validity [35]. Virtual ward and remote monitoring cohorts were featured with moderate-quality evidence from retrospective designs, selection bias, and inadequate adjustment for illness severity, which made it difficult to draw causal conclusions [8, 15, 26]. Although the external validity was well-founded in the context of real-world implementation studies, a number of them lacked standardised outcome measures. It relied on descriptive data, limiting opportunities for generalisation, as noted by Vo et al. [36] and Jalilian et al. [32]. The lack of risk-of-bias analysis restricted the use of policy commentary and scoping analysis by Oliver [26], Islam et al. [37], and Santos et al. [29] as suitable choices to their purposes (Appendix A). Systematic reviews and economic syntheses were appraised using the CASP Systematic Review Multiple appraisal tools (CASP, MMAT, and JBI) were used to ensure design-specific, methodologically appropriate quality assessment across heterogeneous study types. Checklist for evaluating search completeness, risk of bias assessment, data synthesis, and applicability to healthcare settings. High-quality reviews including Shi et al., Gentili et al. [21] Benedetto et al. [27] were considered robust when they demonstrated transparent methods, comprehensive searches, and appropriate synthesis. In contrast, protocols and scoping reviews were rated lower, as they lacked outcome synthesis or formal risk-of-bias assessment. They were therefore used only for contextual or mapping purposes rather than evidential weighting (Appendix A). Analogues were used to assess cohort and diagnostic studies using the Critical Appraisal Skills Program (CASP) tools, randomised controlled trials using the Joanna Briggs Institute (JBI) methodology, and the synthesis of evidence using the CASP systematic review checklist. Appraisal criteria were tailored to the study design, and systematic reviews were evaluated for search completeness, bias appraisal, and synthesis rigour, rather than for experimental control.

2.9. Ethical Consideration

The study relied exclusively on secondary analysis of published literature and did not involve human participants or identifiable patient data. As such, institutional ethical approval was not required. However, all efforts were made to ensure ethical integrity, including accurate citation, avoidance of plagiarism, and transparent reporting in accordance with PRISMA guidelines.

3. ANALYSIS

Effect sizes across heterogeneous study designs and outcome definitions are not directly comparable; therefore, findings are synthesised narratively by domain.

3.1. Virtual Wards and Hospital-at-Home Models

The rapid expansion of virtual wards (VWs) and hospital-at-home (HAH) models during the COVID-19 pandemic revealed mixed findings. Van Goor et al. [20] conducted an RCT that showed that remote hospital care reduced hospital stays by 1.6 days (Table 1A and 1B). However, there was no significant increase in hospital-free days within 30 days, indicating that overall care responsibility increased. Remote hospital care reduced index length of stay by a mean difference (MD) of −1.6 days (95% CI −2.4 to −0.8; p < 0.001), but did not increase hospital-free days at 30 days (MD = 1.7 days; 95% CI −0.5 to 4.2; p = 0.112), and was associated with a longer total duration of hospital responsibility (MD = +4.1 days; 95% CI 0.5 to 7.7; p = 0.028), demonstrating a shift rather than a reduction in care burden. Zychlinski et al. [17] reported more optimistic results, with telemedicine-controlled HAH reducing both length of stay (LOS) and mortality rates for COVID and non-COVID patients, along with no increase in readmissions. Median length of stay was 2 days shorter in the home model for both COVID-19 and non-COVID patients (median difference; 95% CI 1–3 days; p = 0.008 and p < 0.001, respectively), while 30-day readmission was not significantly different (COVID-19: OR = 1.00, 95% CI 0.49–2.04; non-COVID: OR = 0.70, 95% CI 0.39–1.28). In contrast, 30-day mortality was lower in hospital-at-home patients (COVID-19: OR = 0.34, 95% CI 0.11–0.86; non-COVID: OR = 0.38, 95% CI 0.14–0.90) Ward et al. [16] demonstrated that home oxygen therapy, within a virtual ward, resulted in low 30-day readmission (3%) and mortality (1%) rates. However, this study was limited by its single-centre design and patient selection bias. In contrast, Beaney et al. [13] found that the Oximetry@home program in England showed no mortality benefit and increased hospital and critical care admissions, with only 2.5% of eligible patients enrolled, severely diluting any potential benefits. These findings highlight significant variation in the effectiveness of virtual care, with infrastructure and patient engagement being key determinants of success.

Table 1. Summarises the included characteristics and key outcomes of the selected studies.

Table 1(A). Included primary studies.

Table 1(B). Contextual sources.

Recent review evidence and implementation studies indicate that the success of virtual wards (VWs) and hospital-at-home (HAH) still heavily relies on context, scalability, and the standardisation of care. Although large syntheses, including Shi et al. [23], documented decreased length of stay and similar or even lower readmission rates than in hospital care, in real-world assessments, these benefits are not always associated with a decrease in overall healthcare demand. As an illustration, van Goor et al. [20] showed a decrease in index length of stay but did not see a decrease in hospital-free days, indicating that it is not that the responsibility has been shifted, but that the load has been shifted. Implementation assessments, such as Vo et al. [36] also demonstrated that only a sufficient staffing of the services and a digital supporting environment will lead to the implementation of bed-day savings, whereas a critical look at the policy by Oliver [26] cautioned them about the risk of creating a promising naivety and creating clinical risk when the implementation of virtual wards is poorly resourced. Economic reviews like that by Jalilian et al. [32] support this warning and show that even though VWs may liberate capacity, the price per bed-day might be higher than the conventional inpatient treatment under unfavourable staffing or occupancy.

3.2. Remote Monitoring and Resource Optimisation

The remote monitoring turned out to be an essential approach during the years of the COVID-19 pandemic, aimed at reducing the number of patients in hospitals and improving resource management. The evaluation of the COVID Oximetry@home program was performed by Beaney et al. [13] and was based on a national stepped-wedge design, which did not show any significant decrease in mortality, but did decrease the number of visits to the hospital. At population level, implementation of the COVID Oximetry@home programme was not associated with a reduction in mortality and showed higher service utilisation, with increased odds of emergency department attendance (OR = 1.12; 95% CI 1.06–1.18), emergency admission (OR = 1.12; 95% CI 1.05–1.20), and critical care admission among those admitted (OR = 1.24; 95% CI 1.05–1.47), reflecting increased downstream demand rather than resource substitution. Ward et al. [16], in their turn, focused on a single-centre virtual ward where COVID-19 patients are in need of oxygen support, and their readmission and mortality rates were low. In a real-world virtual ward providing home oxygen, 30-day outcomes showed 3.4% readmission (5/147) and 1.4% mortality (2/147), with no statistically significant difference between patients discharged with or without oxygen, indicating safety in carefully selected patients but without comparative effect estimates against standard inpatient care. Similarly, van Goor et al. [20] reported that telemedicine-supported remote hospital care for COVID-19 patients resulted in reduced hospital stays but no increase in “hospital-free days”. Zychlinski et al. [17] further demonstrated that hospital-at-home models lowered both length of stay and mortality, highlighting the benefits of remote care in well-resourced settings. Despite these positive outcomes, the evidence indicates that the success of remote monitoring is contingent upon patient selection, program scale, and consistent implementation.

Remote patient monitoring (RPM) has equally heterogeneous results. Meta-analyses indicate slight but significant decreases in hospitalisation and length of stay in the case of chronic and high-risk populations, especially where adherence and clinical monitoring are good [24, 34]. Future evidence by Po et al. [24] supports the idea that post-discharge RPM can have the most significant impact on high-risk populations with a well-organised escalation protocol, with wider systematic reviews also indicating a mixed impact on emergency department use [25]. In specialised populations studied by Rockey-Bartlett et al. [30], including oncology, the potential is demonstrated, but lacks large sample sizes, measures are not homogenous, and confounding is not fully adjusted. Greater adherence has been seen in device-based models of monitoring: Jansen et al. [35] found that 72 per cent of RCTs reported less hospital use with patient-activated or continuously supported monitoring, which suggested that engagement and responsiveness are factors of success. A combination of these studies highlights the idea that the outcomes of RPM are not determined by the used technology, but rather by patient selection, the intensity of supervision, and the quality of engagement.

Together, these results suggest that VWs do not necessarily help to save money, and they cannot be universal but must be applied in specific ways to digital infrastructure and sustainable workforce models.

3.3. Telehealth and Equity in Access to Care

During the COVID -19 pandemic, the blistering development of telehealth services revealed glaring disparities in access and the use of modalities. Chang et al. [18] showed that the rate of telephone encounters was higher in high-vulnerability communities (41.7 %), compared to video consultations, which were more suitable in areas with wealth (33.7 % versus 18.7 %), due to the presence of such barriers as access to devices, computer literacy, and broadband access. Conversely, Sankaranarayan et al. [14] found that heart failure virtual wards (HFVW), where telehealth was integrated, significantly lowered rehospitalisations and death rates at various follow-up stages. According to the authors, the clinical risk consistently was lower than those of standard care, the clinical risk odds ratio of 1-month readmission was 0.30 (95 CI 0.20 -0.50) and odds ratio mortality 0.26 (95 CI 0.14-0.48), that is, equity-neutral benefits have a higher possibility of success in the environments that are furnished with the infrastructure of specialists and intensive monitored environments. Ward et al. [16] reported low mortality and readmission rates in an oxygen therapy virtual ward in the UK, with limitations put on the study by its single-centre nature. On the contrary, Beaney et al. [13] found that the Oximetry @home programme in England did not provide any mortality benefit and had reduced overall enrolment (2.5 per cent), thus negatively impacting the outcome. In addition, van Goor et al. [20] discovered that hospital care using telemedicine reduced inpatient length of stay and did not result in more hospital-free days, casting doubt on the ability of telehealth to alleviate the impact of acute-care burdens. Yahya et al. [28] opined that digital interventions have the potential to contribute to inequalities, redundancy in services, and overtaxing community providers without lowering hospital demand. Such apprehensions resonate with those depicted by Chang et al. [18], who found that more reimbursable telehealth modalities were lower among high-income groups, which then consolidated access inequities. Together with the risk of overstating performance in policy discourse, these findings also point to the fact that digital health interventions cannot be treated as a neutral solution to problems; they can further increase inequity unless accompanied by digital literacy, inclusive enrolment policies, and infrastructure provision.

3.4. Economic Sustainability & Evaluation Frameworks for DHIs

The interpretation of economic findings was done narratively based on a predefined health-system approach that targeted direct service delivery costs. Since no formal economic extraction grid or meta-analytic cost synthesis was performed, these results ought to be viewed as organised narrative economical observations as opposed to an economic synthesis. Even though secondary systematic reviews have shown that numerous digital health interventions emerge to be cost-effective or cost-saving [21, 33] substantial methodological inconsistencies and limited reporting of implementation and operating costs restrict the transferability of findings across healthcare systems. In this review, economic evidence was interpreted using a predefined and consistent evaluation framework. The primary analytical perspective adopted was that of the health-care system, focusing on resource use and expenditure borne by providers rather than wider societal costs. A short-to-medium time horizon was applied, corresponding to the follow-up periods most frequently reported by the included studies, typically 30-day and 6–12-month post-implementation windows.

Cost components were operationalised as direct service delivery and implementation inputs, including staff time for monitoring and virtual consultations, digital devices and software infrastructure, training and technical support, hospital admissions avoided, emergency department attendances, and bed-days saved. In line with methodological critiques of relying solely on QALYs, outcome valuation in this review did not privilege health-utility measures alone [27]. Instead, economic outcomes were extracted and synthesised primarily as cost per admission avoided, cost per readmission prevented, and cost per bed-day saved, reflecting resource allocation and operational efficiency more directly.

The need for multidimensional valuation of digital interventions, including workforce effects, continuity of care, and equity considerations, is further emphasised by scoping reviews and protocols [29, 31, 37]. Consequently, policy-relevant interpretations in this review are grounded in structured cost components and clearly defined time horizons, rather than in simplified or naïve cost-saving assumptions, thereby supporting context-specific economic decision-making in post-pandemic healthcare systems.

5. DISCUSSION

The resulting discussion is an interpretation of findings based on the input process outcome and output framework to reduce the overlap of themes. The contrasts in heterogeneity among studies were organised patternically and not by chance. The effects of the intervention were different based on the intensity of implementation, structure, and health-system context. Low-technology methods, like telephone follow-up and pulse-oximetry monitoring, were more scalable but yielded mixed results in the presence of ineffective enrolment and escalation pathways. Greater-intensity models, such as specialist-devoted virtual wards and hospital-at-home programmes, fared better in well-resourced systems. In general, digital care models seemed the most successful when three conditions were combined: a deep level of patient engagement, formalised escalation routes, sufficient staffing and digital infrastructure, and generalisability to lower-resource contexts were unclear.

The discussion of the results highlights a massive difference between the effectiveness of remote monitoring patterns and virtual wards. Even though Beaney et al. [13] and Ward et al. [16] proved the possibility of taking a load off hospitals by introducing national programmes and single-centre trials, they also show serious limitations. According to Beaney et al. [13], there was no statistically significant mortality reduction and an increased admission rate in a population-level assessment, which was mainly due to low enrolment and unbalanced programme delivery by region. The low involvement and inefficient enrolment in national rollouts hindered the ability of the programme to achieve quantifiable benefits, considering the dilution effect of a 2.5 enrolment rate. Policies that are equity-conscious and that broaden telehealth service coverage are thus necessary to realise digital health potential to enhance care delivery. These factors remind the reader of the significance of implementation fidelity, a key success factor of digital health programmes. Ward et al. [16], on the contrary, had a more specific virtual ward model, which gained promising outcomes such as low mortality and readmission rates. These favourable outcomes were almost inevitably due to the targeted, clinically selected patient cohort, high involvement, and standardised intervention protocols. This comparison thus highlights the need for a patient selection and engagement strategy to achieve the required results. In addition, findings were reported in highly structured models by van Goor et al. [20] and Zychlinski et al. [33].

5.1. Inputs

The results of the study point to one key issue: even though the telehealth modalities hold significant potential to enhance patient outcomes and benefits, they also reveal deep-rooted inequities that determine patient access to care. Chang et al. [18] found significant differences in telehealth adoption, with low-income and vulnerable groups overrepresented among those who rely on telephone consultations, which are less effective than video-based consultations. This digital divide contributes to the existing disparities in healthcare, particularly among marginalised groups, which typically lack the necessary infrastructure and resources to receive higher-quality telemedicine. Ward et al. [16] and Sankaranarayanan et al. [14] showed that well-planned, specialist-led interventions can improve patient care without overwhelming hospitals, thereby motivating serious clinical outcomes associated with telehealth interventions. However, these programmes were applied in resource-heavy environments, which makes their applicability to low- and middle-income countries (LMICs) questionable as the digital infrastructure is not currently widely available. The success of hospital-at-home models had also been reported by Zychlinski et al. [17]. However, they emphasised that these results were possible only in well-resourced settings, thus challenging their applicability in low-resource healthcare systems. The efficacy of the digital health interventions is also critical for enrolment and standardisation, as Beaney et al. [13] further emphasised. The Oximetry @home programme failed to achieve its desired results due to the lack of engagement of the participants. This is a severe constraint in large-scale implementations; engagement policies should be robust enough to implement desired interventions and achieve their potential. Moreover, van Goor et al. [20] established that inpatient length of stay can indeed be reduced by remote hospital care. However, these interventions do not necessarily lead to a net decrease in overall healthcare demand, which makes the effectiveness aspect of telehealth complicated. All these results point to the fact that digital health implementation is facing a set of structural challenges that include digital literacy, broadband connectivity, and device affordability, among others. There is therefore need to have equity-based policies which increase the range of telehealth services to realise the potential of digital health of enhancing care delivery.

It has been shown that predictive modelling and artificial intelligence in healthcare can be useful in terms of allocating resources. However, its shortcomings shed some light on the intricacy of the real-world implementation. Tariq et al. [15] have demonstrated that predictive accuracy is better with full datasets, i.e., medical histories, demographics, and other variables of interest of the patients. However, the application of the model in a single healthcare system in the U.S. cities prompts the issue of the model applicability to low-resource or rural locations where the number of patients differs, and the ability to retrieve all clinical records is restricted. Besides, the study notes the value of big data that are heterogeneous to enhance generalisability, which is often impossible with limited digital infrastructure and ill-conceived data-collection systems in most healthcare systems, particularly in low- and middle-income countries (LMICs). Moris et al. [19] used a technologically superior model for multimodal information fusion, including clinical history and imaging data, which improved predictive performance. However, high-tech imaging capabilities are highly demanding to rely on, particularly in resource-deprived contexts where imaging facilities are scarce. The inherently opaque so-called black-box character of deep-learning models also brings about the question of transparency; clinicians will be challenged in understanding the logic underlying the algorithmic predictions, thus undermining trust and hampering the smooth incorporation of such technologies into clinical decision-making procedures. All of these studies combined highlight the importance of using AI, which can potentially improve resource utilisation and patient outcomes; however, special attention should be paid to the issues of data accessibility, model transparency, and generalisability. Policy makers should ensure fair availability of the necessary data to train AI-based models and ensure that they develop interpretable AI models that can reduce bias and build trust.

5.2. Processes

One of the current trends in digital health interventions is that implementation fidelity mediates the efficacy. In England, the national COVID-19 Oximetry at home programme, for example, did not reduce mortality because of poor enrolment rates and inconsistent implementation practices [7]. On the other hand, positive results were recorded in a single-centre virtual ward where structured engagement protocols were used by Ward et al. [16]. However, even the most strictly developed models cannot scale without a strong systemic infrastructure that makes uptake possible. In comparison, the models with specialist leadership, like the Heart Failure Virtual Ward, achieved higher results Sankaranarayan et al. [14]; however, only in the resource-abundant settings. Van Goor et al. [20] and Zychlinski et al. [17] evidence supported the view that results are dependent on the contextual preparedness rather than on the type of intervention. The artificial-intelligence applications by [15, 19] demonstrated the predictive accuracy but were not without concern about bias and generalisability.

5.3. Outputs

The digital models of care, such as virtual wards, hospital at home, and remote patient monitoring, have shown potential but highly contextual effects on hospital utilisation. Though in large comparative analyses, shorter length of stay and no or decreased readmission rates are observed in well-resourced environments [22–26, 36]. These models have not always been shown to be effective, and their implementation has been proposed to redistribute as well as to lower aggregate demand in case of insufficient staffing or escalation routes [20, 26]. Similarly, post-discharge RPM has the highest DEN, high-risk, actively monitored cohorts. Po et al. [24] and Tan et al. [22] have the highest certainty, although heterogeneity, small-sample studies, and weak control of confounders limit such certainty in specialised populations such as oncology [30]. Stable increases are achieved with the implementation of continuous or patient-triggered device monitoring, a consistent decrease in hospital utilisation observed across RCTs by Jansen et al. [35], which is important to stress that the level of engagement, but not the adoption of technology per se, is what leads to improvements. Economic results also point to the fact that cost-effectiveness cannot be presumed, since capacity gains might go along with increased operational costs when there are suboptimal occupancy, staffing, and digital infrastructure, and the methodological inconsistency prevents generalisability [21, 29, 31]. New models say that there is a need to have multidimensional, equity-sensitive metrics other than QALY, which include workforce impact, digital access, and system duplication [27, 37].

5.4. Outcomes and System Resilience

The recent national and international guidelines support the fact that virtual wards should be introduced with caution and well-defined patient selection criteria. Policy briefs by the UK point to the fact that the experiences of patients and carers are still positive, but the expectations are usually higher than what clinical potentials can achieve [38–40]. Frail elderly assessment demonstrates that it may be beneficial, but with great differences in safety and staff preparedness [41–44]. Greater studies warn that there must not be assumptions of universal cost-saving that lack standardised economic data [27, 42]. The results of large-scale RPM studies also indicate fewer admissions and caution that the results are reliant on the resources and digital literacy, as well as monitoring capacity [39, 40, 45]. Digital health projects must be transformed to go beyond the response-based, ad hoc, and crisis-driven initiatives to systematic inclusion in existing care pathways. Remote-monitoring, virtual-ward, and artificial-intelligence devices are to be implemented as supplements to traditional methods of care, and they need clear instructions concerning the route of escalation, inclusion criteria, and the individuals who will work with the patients, as well as multidisciplinary team coordination. Future digital health implementations are to be patient-centred and standardised. Policymakers must set minimum requirements on the quality of digital tracking, the equivalence of telehealth reimbursement, and ongoing clinician training to ensure successful digital monitoring. Transparency should maximise trust in AI-driven decision support. New models must prefigure an explainable AI (XAI) system, constantly testing in various groups, and strictly controlled. The ethical regulation of privacy, consent, and accountability must be enforced to avoid harm and reduce the disparities caused by deployment. Lastly, digital-health interventions are to be included in the strategies to improve system resiliency. Research has shown that hospital-at-home and telehealth programmes can potentially reduce acute care demand in times of crisis [14, 17]. The tools cannot be applied only during the pandemic but should be used to be ready to the shocks in the future like ageing, a load of chronic diseases, and health crisis in a climate. The post-pandemic world teaches that despite its potentially powerful nature, digital-health intervention is not transformative: its effectiveness lies on equity, fidelity, context, and ethical integration. It is believed that future healthcare will be hybrid systems where digital and traditional elements are bound together to guarantee efficiency, capacity, equity, transparency, and resilience. Because the included studies employed heterogeneous designs, outcome definitions, and analytical approaches, direct quantitative comparison of effect sizes was not appropriate and may limit the interpretability of cross-study differences. Cumulatively, this review suggests that digital care models can produce a capacity substitution in conditions of implementation fidelity and workforce elasticity.