FormalPara Key Points for Decision Makers

In this paper, we critically review and describe studies that assess the impact of digital health interventions in type 2 diabetes mellitus on healthcare utilization and cost.

We found that the impact of digital health may differ according to the care sector, such as medications or different types of appointments.

Drawing on the findings of this review, we propose some areas for further work, including an assessment of the impact of digital health interventions across a broader range of care utilisation components and settings, such as social and mental healthcare services.

1 Introduction

The number of digital health interventions (DHIs) delivered through devices such as smartphones, tablets or computers has increased exponentially [1]. Such DHIs have been used to facilitate remote access to effective treatments, improve the management of chronic conditions and promote healthy behaviours. Previous systematic reviews have synthesised evidence about the impact of DHIs on health outcomes and patient experience [2, 3]. However, the economic impact of DHIs is less well understood. On the one hand, one might expect DHIs to be cost saving because they may help deliver healthcare more efficiently. On the other hand, the implementation of DHIs may imply large investments by the health system, for example, restructuring care pathways and health services, implementation costs and staff training and upskilling. While economic evidence is rapidly emerging, the overall direction and magnitude of the economic impact of DHIs is unclear.

One of the areas where DHIs have received considerable attention is the management of type 2 diabetes [4]. Digital health interventions are particularly valuable in this context because they can facilitate the management of the condition and encourage behavioural change (e.g. diet and exercise) [5]. Studying the economic impact of DHIs in type 2 diabetes is crucial because the burden of disease is large [6], and hence, there is ample scope to deliver more health gains from scarce health and social care resources. A recent study [7] reviewed cost-effectiveness analyses of DHIs for the prevention and control of type 2 diabetes but the focus was on DHIs delivered through smartphones and included decision-analytical modelling studies.

Our study aims to critically review the impact of DHIs for the management of type 2 diabetes across a broad range of health and social care services and their associated costs. The review focuses on DHIs that seek to promote or manage health and well-being through supporting behavioural change and decision making of the patient diagnosed with type 2 diabetes, sometimes in combination with a healthcare professional. In this population, DHIs may be used for the management of the long-term condition. Such interventions are typically interactive, personalised, based on user input and may be provided at the individual or population level. This review aims to generate a deeper understanding of the economic impacts (on individuals and health systems) of more complex types of DHIs. Simpler forms of DHIs, such as diagnostic and monitoring devices, are much closer to the nature of conventional medical devices, and their economic impact has been reviewed elsewhere [8]. While we focus on a single long-term condition, we anticipate a similar economic impact of DHIs that support behavioural change and decision making across other chronic conditions, such as arthritis or heart disease.

2 Methods

The review protocol was registered on PROSPERO (CRD42020172621) [9]. The review includes quantitative assessments of the economic impact of DHIs for type 2 diabetes.

Studies met the inclusion criteria for this review if:

  1. 1.

    they included a population of adults (aged 18 years and above) with type 2 diabetes who had received a DHI for information provision, self-management or behaviour management;

  2. 2.

    comparators were standard care, or information provision or group information sessions, no comparator or other DHIs;

  3. 3.

    the outcomes considered were health and social care utilisation and cost;

  4. 4.

    they were evaluations undertaken in a public or private clinical setting internationally; and

  5. 5.

    they were reported in journal articles published in English.

Studies were excluded if they were studying a patient population with type 1 diabetes, gestational or other rarer forms of diabetes, or the DHI was focussed on comorbidities or complications of type 2 diabetes. They were also excluded if clinicians alone used the interventions, for example decision-support systems, and if the study outcomes were only cost-effectiveness modelling (not including patient-level data) or patient experience or satisfaction alone, not including reporting of health and social care utilisation and costs.

Four databases, MEDLINE, EMBASE, PsycINFO and EconLit, were searched for articles published between January 2010 and March 2021. A search strategy was developed in MEDLINE with the help of an information specialist to ensure proper use of search terms. This strategy was adapted for use in the other databases. Appendix 1 shows the search strategy for all databases. It combined key terms about DHIs, type 2 diabetes and the economic impact in terms of health and social care utilisation and costs.

The database searches were run in April 2021. Duplicates were removed using the Rayyan web application [10] and Endnote referencing software [11]. The first reviewer screened all the records using Microsoft Excel. A second reviewer screened 20% of the included abstracts and full texts for eligibility. Any disputes were discussed between the first and second reviewers. If the conflicts remained, they were referred to a third reviewer.

Templates for data extraction were piloted and adapted to enable key data to be extracted from the included papers. Information was extracted on the characteristics of the studies (i.e. number of participants and participant characteristics, country, study type, length of follow-up, key outcome measures reported and their results), the impact on health and social care utilisation (e.g. self-reported or administrative data source changes in several types of health and social care utilisation) and the resulting impact on costs (e.g. how cost had changed in the reported follow-up period). The data extraction was reviewed by the third reviewer for quality assurance. The review results were analysed using a narrative synthesis approach, in which themes were identified within the study results that related to the aims of the study [12]. Because of the heterogeneity of the study designs and outcomes measures, it was not possible to perform a meta-analysis.

The ROBINS-I tool [13] was used to appraise all papers by the first reviewer and all studies were also appraised by a second reviewer. It is a quality assessment checklist designed to assess the overall risk of biases in the study design and analysis, including aspects related to: (1) confounding; (2) sample selection; (3) classification and measurement error; (4) deviations from study protocol; (4) missing data; and (5) reporting. Within each of these components, studies can be rated as ‘low risk of bias’ (comparable to a well-conducted randomised controlled trial [RCT]), ‘moderate risk of bias’ (sound for a non-randomised trial, but not comparable to a RCT), ‘serious risk of bias’ (the study has some significant issues that may affect the validity of the results), ‘critical risk of bias’ (too problematic to provide any useful evidence) and ‘no information’ (no information on which to base a judgement) if the design or analysis component has not been reported.

3 Results

The results of the searches are shown in the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) flowchart (Fig. 1). The database searches identified 617 papers after removing duplicates. These papers were then screened and assessed for eligibility. The reasons for papers to be excluded are outlined in the PRISMA flowchart. During abstract screening, many studies were excluded because they focused on a non-relevant DHI (n = 157). For example, they did not consider a DHI at all, or focused on medical devices for type 2 diabetes and interventions that were targeted at clinicians such as decision support systems. Others were excluded because they did not consider impacts on healthcare utilisation and costs (n = 159), focused on other types of diabetes or targeted complications (e.g. foot ulcers) of type 2 diabetes (n = 53). Systematic reviews synthesising the impact of DHIs on health outcomes and patient experience were also excluded (n = 51).

Fig. 1
figure 1

Preferred reporting items for systematic reviews and meta-analyses (PRISMA) flowchart

Full size image

At full-text screening, studies were excluded mainly because they did not evaluate the impact on healthcare utilisation and costs (n = 25), did not focus on type 2 diabetes (n = 9) or were not an impact study (n = 7). Twenty-two studies were included in the review.

3.1 Study Characteristics

The key study characteristics are shown in Table 1. Most studies (n = 16, 73%) had over 100 participants and were conducted in North America (n = 8, 36%) or Europe (n = 9, 41%). Over half of the studies (n = 13, 59%) were conducted alongside RCTs. Almost half of the studies (n = 10, 45%) involved a comparison of DHI with treatment as usual, although this meant a variety of face-to-face services across a range of settings (e.g. primary and secondary care). Most studies considered relatively short follow-up periods: five (23%) had a follow-up of up to 6 months and 14 studies (64%) had a follow-up of 6–12 months.

Table 1 Summary of study characteristics
Full size table

Most studies considered the impact of DHIs on a single component of healthcare utilisation or costs. For example, eight studies reported the impact on medication utilisation (e.g. metformin or insulin) [14,15,16,17,18,19,20,21]. A measurement of HbA1c over 6.5% corresponds to a diagnosis of type 2 diabetes, and management of the condition includes targeting a level of HbA1c below 7%. Five of these were studies in which most patients had HbA1c levels between 7 and 9% [14,15,16,17,18] and the rest had populations where the mean HbA1c level was higher. In the studies where the time since onset was recorded, patients varied from ‘newly diagnosed’ to later stages of the disease requiring insulin or treatment intensification.

Very few studies (n = 3, 17%) reported an impact across several primary and secondary care utilisation components [22,23,24], which included a mix of hospital, general practitioner (GP) and community diabetes-related services. Eight (36%) of the studies were based on self-management (patients monitoring their lifestyle and condition alone) [14, 16, 22, 24,25,26,27,28] and eight more on behaviour management (patients and professionals managing patients’ lifestyle and condition) [15, 17, 20, 21, 29,30,31,32]. Four (18%) were based on self-management and information provision together [18, 19, 33, 34], one (5%) on behaviour and self-management [23], and another on behaviour management and information provision [35].

3.2 Summary of Study Results

The study results are summarised in Table 2. Overall, the direction of the impact of DHIs on health and social care utilisation was mixed, irrespective of the mode of delivery of the DHIs. Four (18%) studies reported an increase in healthcare utilisation [20,21,22, 33], seven (32%) studies suggested a reduction in care utilisation [16, 18, 24, 27, 28, 30, 34], and five (23%) studies reported both positive and negative economic impacts [17, 25, 26, 31, 32]. Of those studies that reported costs (n = 7, 31%), six suggested a decrease in costs [15, 19, 24, 28, 29, 35], and one reported increased costs [23]. The magnitude of the cost impact in these studies was relatively small, usually less than $500 saving per patient over a time horizon of 6–12 months.

Table 2 Summary of study results
Full size table

The reviewed studies suggest that the impact of DHIs might differ according to the component of healthcare utilisation. For example, DHIs were mostly associated with lower medication use [14,15,16,17,18,19,20,21, 28, 31, 36] and fewer outpatient visits [25, 28,29,30,31,32, 34, 35]. In the studies where the time since onset was recorded, patients varied from ‘newly diagnosed’ to later stages of the disease requiring insulin or treatment intensification. Whereas, the impact of DHIs on GP visits [25, 26, 28, 30, 32, 35], hospital accident and emergency [25,26,27, 32] and inpatient admission [25, 27, 30, 35] was mixed. In studies that reported impacts across multiple healthcare components and settings of care, the evidence was also mixed [22,23,24, 35]. None of the studies reported impacts on social care or mental healthcare utilisation.

3.3 Quality Assessment

Results of the quality assessment are shown in Table 3. Overall, the risk of bias was relatively low across the reviewed studies [13]. Across the studies included, the most common issues were a lack of appropriate analysis methods for confounding, and missing data (lack of follow-up data for much of the original sample because of a high dropout, or a lack of information about methods for adjusting for missing data).

Table 3 Risk of bias
Full size table

Twelve (55%) RCTs were included in the analysis [15,16,17, 19, 23,24,25,26, 29, 32, 34, 35] with the vast majority being judged ‘low risk’ in all categories. The remaining ten (45%) [14, 18, 20,21,22, 27, 28, 30, 31, 33] studies were non-randomised, with three of them being rated ‘high risk’ because of potential issues with confounding [22, 27, 30]. Two studies were judged as having ‘moderate risk’ [20, 30] because of sample selection issues. Two of the observational studies provided insufficient information on missing data [21, 31] and were classified as ‘no information’.

4 Discussion

4.1 Summary of Findings

This study critically assessed published evidence on the economic impact of DHIs for type 2 diabetes on health and social care utilisation and costs. Overall, the review suggests that the direction of the impact of DHIs on healthcare utilisation is mixed and might differ according to the healthcare utilisation component. In particular, DHIs seem to be associated with a reduction in medication and fewer diabetes-related outpatient appointments. The vast majority of studies evaluating the impact of DHIs on healthcare costs reported a decrease in costs. The overall risk of biases in the reviewed studies is relatively low; the vast majority of the non-randomised studies was rated ‘low’ or ‘moderate’ risk across most bias components.

4.2 Contributions

This is the first study to comprehensively review the economic impact of DHIs for type 2 diabetes. Overall, our review suggests that the case for an economic benefit is not as clear as the clinical benefit shown in previous systematic reviews [3, 37]. The study also shows that the existing evidence base is limited because of the short follow-ups of many studies evaluating interventions for type 2 diabetes, and the difficulty in comparing heterogeneous results for the impact on service utilisation and costs. We have also shown that studies report a limited number of cost dimensions and do not show a benefit as clearly in the costliest types of care (e.g. hospital admissions). However, the studies have included participants at a range of various stages of the disease in terms of the time since diagnosis and the duration of type 2 diabetes symptoms (which may or may not be linked).

A recent study [7] reviewed the cost-effectiveness evidence on mHealth interventions for type 2 diabetes and suggested that, overall, these tend to provide good value for money. This does not necessary mean that the mHealth interventions considered in that study led to lower costs as they can be cost effective even if they are more costly than the comparator. However, this review [7] focused on (modelling-based) cost-effectiveness analyses, and interventions delivered through smartphones, whereas our study included a broader range of DHIs and all types of economic impact evaluations. By reviewing a more comprehensive range of impact studies, we find that the economic impact of DHIs may not be as well understood as suggested by the previous review.

Our review also found that the economic impact of DHIs might differ across the different types of healthcare utilisation. In fact, it suggests that potential economic benefits are most likely to materialise in certain care components such as medication and insulin use, or outpatient attendances [15, 16, 18, 19, 29, 34]. Unlike previous reviews [7], our study considered the impact of DHIs on both health and social care utilisation and costs, but found no evidence on the impact on social, community and informal care services.

4.3 Limitations

One of the limitations of this review is that it did not consider grey literature or studies published in languages other than English. It was deemed important to include high-quality evidence and full evaluations such as RCTs and large observational studies on the impacts of digital health, which are less likely to be reported outside journal articles.

Our study was not able to disentangle potential drivers of the impact of DHIs on health and social care utilisation and costs. Characteristics such as age, duration of disease, ethnicity or gender may have an impact on clinical and economic outcomes, but the reviewed studies have not considered whether the economic impact of DHIs varies across these factors. One notable exception is the evaluation of the HeLP-Diabetes programme, which adjusted for age and disease duration in the economic analysis [24]. Poduval et al.’s qualitative analysis as part of the same study finds that the shorter the duration of diabetes since diagnosis, the greater the likelihood of completing an intervention [38], and hence the higher the costs incurred.

The mix of results found in this review may have been partially explained by differences in care pathways and health financing systems across the different countries included in the review. However, this review was not able to disentangle this, and determine whether the findings are generalisable across jurisdictions. In addition, this review was not able to dissect whether the economic impact of DHIs differs according to whether the DHI complements or substitutes face-to-face health services. The heterogeneity of DHI types in terms of their function (self-management or behaviour management as defined previously) did not appear to have a clear effect on the impact of DHIs. However, the proportion of DHIs reducing service use or cost was slightly higher in DHIS for self-management (seven out of 12) [16, 18, 19, 24, 27, 28, 34] than when service users used the DHIs with healthcare professionals [15, 29, 30, 35].

4.4 Implications for Future Research

This review has identified some gaps in the literature about the impact of digital health for type 2 diabetes on healthcare utilisation. First, DHIs are likely lead to a broad range of health and non-health cost impacts, including those outside the healthcare sector [39]. Hence, studies should account for the potential economic impact across a broader range of sectors, such as social and community care settings, informal care and productivity impacts. Second, DHI software updates and iterations are an important consideration that has received little attention in the reviewed studies. Because digital health is a rapidly moving field, DHIs are likely to be updated during an evaluation. As a result, the interventions themselves may change during an evaluation and it will be important to assess the extent to which the evolving nature of DHIs impacts on health and social care utilisation. Third, further work is warranted to explore which type of patients benefit most from these DHIs and which elements of the DHIs drive this (e.g. therapeutic elements or dose), and how this heterogeneity may impact health and social care utilisation. This is important so that adaptations and iterations of DHIs target the populations that are most likely to benefit from the DHIs and lead to greater savings to the health system.

5 Conclusions

The review found mixed evidence on the economic impact of DHIs. Reviewed studies suggested that this impact might differ according to the care component, with DHIs being associated with a reduction in medication and diabetes-related outpatient appointments, but not GP visits or hospital admissions. Future work should consider exploring the economic impact of DHIs across a broader range of care components and settings, and whether this impact is mediated by patient or contextual factors.