Study characteristics

The selected studies were mainly from Australia (18 percent, n = 11), the USA (14.75 percent, n = 9), England (13 percent, n = 8), Spain (11 percent, n = 7) and the Netherlands (11 percent, n = 7). The economic evaluations used in this review covered the selected time period, with a higher frequency in 2017 (21 percent, n = 13).

Remote patient monitoring was initially developed as a response to the need to monitor chronic pathologies. The initial analysis found that cardiology was the best-represented pathology (36 percent, n = 22) (Reference Dehmer, Maciosek and Trower12;Reference Greving, Kaasjager and Vernooij13), followed by mental health (Reference Barrett, Begg, O’Halloran and Kingsley14;Reference Romero-Sanchiz, Nogueira-Arjona and García-Ruiz15), pulmonary pathologies (Reference De San Miguel, Smith and Lewin16;Reference Schmier, Ong and Fonarow17), endocrinology (Reference Gordon, Bird and Oldenburg18;Reference Krishnan, Finkelstein and Levine19), dermatology (Reference Arora, Harvey and Glinsky20), neurology (Reference Comans, Mihala and Sakzewski21), gastroenterology (Reference Del Hoyo, Nos and Bastida22;Reference Heida, Dijkstra and Muller23), nephrology (Reference Kaier, Hils and Fetzer24), physiotherapy (Reference Fatoye, Gebrye and Fatoye25;Reference Suman, Schaafsma and Van Dongen26), oncology (Reference Wall, Kularatna and Ward27), and chronic non-target diseases (Reference Kaier, Hils and Fetzer28;Reference Upatising, Wood and Kremers29).

Telemonitoring system

Regarding technical solutions, the choice of digital technologies depends heavily on the pathology. The analysis identified the following general trends: a telemonitoring platform (39 percent, n = 24), connected objects (21 percent, n = 13), a telephone system (18 percent, n = 11), a web platform (21 percent, n = 13), a mobile application (n = 1), and a mobile telemonitoring platform (n = 2).

Three types of data were identified according to their mode of transmission: data transmitted by the patient; additional data supplied by the healthcare professional to the patient (e.g., advice or feedback on their condition, or actions that had been taken); and data generated automatically by the technical solution. Regarding data transmitted by the patient, in most cases, these concerned health status indicators: blood pressure, blood sugar, weight, etc. (59 percent, n = 36). Other examples involved performing exercises requested by the doctor (21 percent, n = 13), direct communication between the patient and the doctor about symptoms or adopted behaviors (15 percent, n = 9), responses to questionnaires (15 percent, n = 9), or perception scales/ scores (8 percent, n = 5).

In many cases, information transmitted to the patient by healthcare professionals concerned adjustments to the therapeutic treatment (31 percent, n = 19). In other cases, it was personalized feedback: an analysis of results, the supervision of actions taken by the patient (33 percent, n = 20), providing advice (24.5 percent, n = 15), incentives to change behavior, achieve goals, or increase motivation (6.5 percent, n = 4), or therapeutic education (6.5 percent, n = 4).

The results highlighted that the organization of remote monitoring was based on the transmission of automatic signals, in particular in the form of alarms (26 percent, n = 16), and automatic feedback or self-regulation (13 percent, n = 8). Six studies (10 percent) reported the use of algorithms to generate alerts. In one case, a connected object (a pill organizer) transmitted schedules directly to the patient, in order to better-monitor compliance with his drug treatment.

We also assessed the frequency of data transmission. In 61.5 percent of studies, data were transmitted by the patient on a daily basis. The remainder were distributed as follows: every week (15 percent, n = 9), several times a week (13 percent, n = 8), every month (6.5 percent, n = 4), several times a month (3 percent, n = 2), every 3 months (n = 1), and continuous or automatic (8 percent, n = 5).

The frequency of routine interventions by healthcare professionals varied as follows: daily (13 percent, n = 8), every week (15 percent, n = 9), twice a week (5 percent, n = 3), every 15 days (10 percent, n = 10), twice a month (n = 2), and once a month (n = 5). In addition, in the majority of cases, healthcare professionals intervened if there was an automatic alert, or if they identified abnormal values in the data transmitted by the patient (33 percent, n = 22).

Patient involvement

The selected studies were classified into three categories, according to the telemonitoring system, and the degree of patient involvement. In the first level of patient involvement, (weak), the medical device automatically transmits data (via an implanted cardioverter, a connected pacemaker, a connected blood pressure monitor, etc.). This was the case in 17 studies, including seven in cardiology, and seven in pulmonary disease. In the second level of involvement (medium), the patient has limited involvement, and is supported by healthcare professionals or paramedics. These interventions mainly concern telephone calls to retrieve data, or to reassure the patient that they are not alone, and reinforce the message that communication with medical staff is at the heart of the operation. Seventeen studies fall into this category, and they concern several medical specialties (cardiology, dermatology, endocrinology, gastroenterology, mental health, nephrology, pulmonary disease). In the third level of involvement (strong), the patient is deeply involved because he or she is asked to perform exercises, follow videos, consult websites, undertake physical activity, and so forth Therapeutic education is an example of this organization, and it was described in 27 studies. Cardiology (10 studies) and mental health (8 studies) were particularly well-represented.

Economic evaluation methods

CUA were reported most frequently (67 percent, n = 41) (Reference Romero-Sanchiz, Nogueira-Arjona and García-Ruiz15;Reference De San Miguel, Smith and Lewin16;Reference Krishnan, Finkelstein and Levine19–Reference Henderson, Knapp and Fernández28;Reference Kaambwa, Bryan and Jowett30–Reference Little, Stuart and Hobbs52). On the contrary, only 13 percent (n = 8) of studies implemented a CEA (Reference Dehmer, Maciosek and Trower12;Reference Krishnan, Finkelstein and Levine19–Reference Wall, Kularatna and Ward27;Reference Kaambwa, Bryan and Jowett30;Reference Stoddart, Hanley and Wild31;Reference Warren, Carlisle, Mihala and Scuffham51–Reference Kloek, van Dongen and de Bakker54). CMA was the method of choice in 20 percent (n = 12) of studies (Reference Greving, Kaasjager and Vernooij13–Reference Upatising, Wood and Kremers29;Reference Calò, Gargaro, De Ruvo and Palozzi32;Reference Heidbuchel, Hindricks and Broadhurst34–Reference Little, Stuart and Hobbs52;Reference Kloek, van Dongen and de Bakker54–Reference Cowie, Simon, Klein and Thokala72).

With respect to CEA, health outcomes were assessed using various criteria: physical skills (n = 1), number of events (n = 1), arterial pressure (n = 3), quality of life (n = 1), HbA1c rate (n = 1), and weight loss (n = 1). However, years of life, measured as all-cause mortality, was not retained in any study. In the context of CUA, QALY (quality-adjusted life years) was the only health outcome criterion used. In the majority of cases, the EQ-5D measurement system was used to assess the utility score (44 percent, n = 27), followed by the SF-6D (n = 4), the SF-12 (n = 3), and the SF-36 (n = 2). Four articles reported the use of a specific scale.

Our analysis found that the scope of the evaluation was generally restricted to institutions responsible for financing the health system (69 percent of studies). Relatively few evaluations adopted a broader, collective perspective (19 percent, n = 13). Other studies considered the point of view of the patient (13 percent, n = 9), particularly with regard to transport costs or lost productivity.

Most studies examined an adult population with a chronic pathology, populations at risk of complications, or those whose pathology had been recently diagnosed. Only two studies considered children, and seven studies focused on older populations. A few American studies investigated military veterans.

Mean sample size was 235 patients, with a minimum of 42 and a maximum of 1,225 patients. In most cases, the timescale for the implemented evaluation was approximately 1 year. Costs and results were, therefore, not discounted.

Most studies (88.5 percent) compared remote monitoring to care-as-usual monitoring. In a few rare cases, a full telemonitoring intervention was compared to a less-complete or less-developed monitoring system, considered to be a degraded solution (Reference De San Miguel, Smith and Lewin16–Reference Suman, Schaafsma and Van Dongen26;Reference Hazenberg, Kerstjens and Prins44–Reference Little, Stuart and Hobbs52;Reference Kloek, van Dongen and de Bakker54;Reference Stoddart, van der Pol and Pinnock69–Reference Cowie, Simon, Klein and Thokala72). In some cases, the main difference in follow-up modalities concerned the frequency of contact between patients and the follow-up team, which was higher in the case of the intervention group (Reference Greving, Kaasjager and Vernooij13;Reference Barrett, Begg, O’Halloran and Kingsley14;Reference Sangster, Church, Haas, Furber and Bauman40–Reference Dixon, Hollinghurst and Ara43;Reference Brabyn, Araya and Barkham60–Reference Dear, Zou and Ali63).

The analysis identified the following costs. Medical costs include consultations with general practitioners or specialists, paramedical consultations, medical biology acts and other clinical and technical acts, drugs, hospitalization, etc. (87 percent). Costs related to the use of the remote monitoring technology (the device, the platform, connected objects, etc.) were explicitly considered in only 49 percent of studies, while the cost of the technology itself was not always considered, especially when using a web solution, software, or a simple data transmission device.

With respect to data, 36 percent of studies used public, medico-administrative data. Other sources of information were: hospital data (n = 7), retrospective studies of patient records (n = 5), data published in the literature (n = 3), data from insurers (n = 3), and data related to veterans (n = 1). Micro-costing and Activity Based Costing methods were used in six studies. In 28 percent of studies, patients were asked to keep a written record of all of their resource consumption.

The quality of economic analyses

The quality of the economic evaluation was initially categorized using Drummond’s checklist.

We made sure that the CHEERS checklist items were clearly reported. Figure 3 summarizes the items reported and recommended by the CHEERS checklist (excluding items concerning economic modeling). The items that least complied with the CHEERS were on discount rates, as most investigations had a time horizon of less than 1 year.

Figure 3.

Visual representation of CHEERS checklist.

The main methodological shortcomings were as follows:

• ▪ a failure to consider heterogeneity in the groups analyzed (79 percent);

• ▪ a failure to use advanced statistical techniques, for example, to compensate for missing data (43 percent);

• ▪ the lack of a sensitivity analysis (40 percent); and

• ▪ an insufficient time horizon (20 percent).

Economic results

The majority of studies (44 of 61; 72 percent) conclude that telemonitoring is reported to be cost-effective compared to care-as-usual monitoring. This result is broken down as follows: 8 of 8 for CEA, 11 of 12 for CMA, and 28 of 41 for CUA. Only two studies report poorer health outcomes with telemonitoring (Reference Cui, Doupe, Katz, Nyhof and Forget38;Reference Warren, Carlisle, Mihala and Scuffham51).

Fourteen studies report no significant differences in terms of costs, and 21 find no differences in terms of effectiveness (including 19 who report results in terms of QALY).

Table 1 presents the results of the economic evaluation according to the technology used, and the degree of patient involvement (Authors/Country/Intervention description/Comparison and number/Design/Perspective/Time horizon/Health outcome/Economic results/Sensitivity analysis/Findings). Supplementary material S3 presents all costs reported in each study.

Table 1.

Economic results of the scoping review

The first level of patient involvement (weak, automated monitoring): the economic results are very positive. A single study reports higher costs of telemonitoring, established by a large-scale European project (Reference Henderson, Knapp and Fernández28). Only four papers reported lower (or equivalent) costs for the care-as-usual scenario. This system favors data transmission through technology, the mobilization of algorithms, the creation of alerts, and so forth Although only eight out of 17 studies considered the cost of the technology itself, in most cases, medical costs were considered. Transport costs received little attention (6 of 17 studies). CMA was the most popular evaluation method in this type of remote monitoring. Only one study showed a gain in QALY (Reference Zanaboni, Landolina and Marzegalli35). This type of remote monitoring supports experiments with a significant number of patients, the average sample size was 262 patients, while the average study duration was 11.66 months.

The second level of involvement (medium, support from professionals): when patient participation is limited to responding to telephone calls or filling in questionnaires, the evidence suggests that telemonitoring is still reported to be cost-effective (13 of 17 cases). In this system, the patient is supported by health or medico-social personnel, and related costs are significant. The latter were considered in 16 of 17 studies, notably the cost of consultations, but also the cost of staff time. Details of communication costs were only given in four studies. Four papers reported the cost of training, and one reported the cost of coordination. Finally, seven (out of 17) studies reported that telemonitoring costs were higher than care-as-usual. The average sample size was a little smaller than the first level of organization of remote monitoring: 187 patients and 10 out of the 17 studies carried out a CUA. In seven studies, telemonitoring was more effective than comparative practices, while the remainder reported no significant differences. Half of the sampled studies reported a gain in QALY.

The third level of patient involvement (strong, active remote participation): here, patients are significantly involved in their follow-up. Our analysis found that in 20 of 27 studies, telemonitoring was reported to be cost-effective; however, this was not the case in four studies. In three of the latter investigations, there were fewer than 100 patients per arm. The fourth study evaluated 1,225 patients and found that equipment costs were high (a large set of connected objects made available to the patient). Nine studies reported higher costs. Extensive patient involvement is reflected, above all, in access to information about their pathology, therapeutic education sessions, and undertaking exercises. This investment could result in a loss of patient productivity and 8 of 27 studies took this indirect cost into account.

Another cost relates to the staff who provide medical follow-up. Although most (22 of 27) studies considered personnel costs, either via the time spent, or by rating a consultation, they did not account for time spent upstream (e.g., preparing health information, or exercises) or downstream (e.g., monitoring behavior change, coaching, providing advice). While 10 studies considered the cost of the intervention, only five took training costs into account. Quality of life is another important concern at this level, as the patient takes charge of his or her pathology, and is expected to be proactive. Hence, most (22 of 27) studies implemented a CUA. Of the sample, 10 reported a gain in quality of life, and only one found a decrease. Despite the need for extensive involvement by all stakeholders, the average sample size was 237 patients. On the contrary, the average time horizon was 8 months.

Five studies developed economic models to estimate the long-term cost effectiveness of the telemonitoring intervention (Reference Schmier, Ong and Fonarow17;Reference Kaambwa, Bryan and Jowett30;Reference Dixon, Hollinghurst and Ara43;Reference Cowie, Simon, Klein and Thokala72). All demonstrated that telemonitoring remains cost-effective in the long term (from 5 to 30 years). Analyses relied on a Markov (Reference Schmier, Ong and Fonarow17;Reference Gordon, Bird and Oldenburg18;Reference Kaambwa, Bryan and Jowett30;Reference Cowie, Simon, Klein and Thokala72) or Dixon (Reference Dixon, Hollinghurst and Ara43) model, based on a cohort simulation. Extrapolation was based on trial data, with the exception of Cowie (Reference Cowie, Simon, Klein and Thokala72) who used a hypothetical population of 20,000 patients suffering from the studied disease. No study reported an effect on mortality or sustainability. Only Cowie (Reference Cowie, Simon, Klein and Thokala72) extended the model used to other countries (the Netherlands, Belgium, Italy and Germany) and showed that telemonitoring remained cost-effective in the selected countries.