For this section we used results from the domain search (including domain searches from EFF and ORG; search strategy and selection criteria are described in the general methodology description) as well as results from additional hand searching, e.g. within study references. Also results from the TEC domain regarding description of the interventions were used.

Two studies were identified which include a cost effectiveness analysis on structured telephone support (STS) vs. usual care for heart failure (HF) patients and also include intervention costs. From the same research project, both Pandor et al. (2013) and Thokala et al. (2013) ({4} and {5}) used the perspective of the National Health System of the United Kingdom and included estimates of resource use associated directly with the interventions themselves as well as the estimates of resource use associated with hospitalization. Miller et al. (2009) {3} also took a healthcare system perspective and included resource utilisation based on a clinical trial {8}.

As characterized by Pandor et al. (2013) resource consumption directly associated with the intervention can be divided into three parts:

1. Resources used at the patient’s home
2. Resources used at the support centre
3. Other health care resources used to deal with events or alerts (e.g., office visits or home visits at, or by, family practitioners or specialists, emergency room visits)

It should be noted that interventions in this field are heterogeneous and rarely described in detail (see also TEC domain). The same applies to the comparator (usual care). Pandor et al. (2013) used data from a randomised study conducted across 16 hospitals in Germany, the Netherlands and the UK (in the years of 2000 to 2002) comparing home telemonitoring, nurse telephone support, and usual care {9} to estimate the amounts of health care resources used to deal with events/alerts. Miller et al. (2009) include ‘additional costs for the administration of the disease management program’ without specifying individual cost components.

None of the identified cost effectiveness analysis studies includes potential indirect resource consumption due to any negative effects or adverse events associated with STS. This may be partly explained by the absence of significant evidence of adverse events but, alternatively, could just be a simplifying assumption.

The following table lists resource items identified by the authors as potentially relevant, together with the units by which they can be measured, and indicates if they have been included in Pandor et al. {4}. This is done – here and further below - using the study by Pandor et al. (2013) as the other study (Miller et al. (2013)) does not provide sufficient methodological and other details.

Table 1: Resource consumption for STS and usual care: resource items identified as potentially relevant

In addition to these resource items, a cost-effectiveness analysis usually attempts to take account of other changes in the consumption of health care resources which are relevant to the chosen perspective. In the case of STS, one of the most important effects or outcomes is estimated to be on the number of hospitalizations, since heart failure is an important cause of hospitalizations. However, hospitalizations are only an intermediate outcome indicator and although each hospitalization utilises health care resources, hospitalizations do not necessarily provide a robust measure of either costs or effectiveness. Three of the identified studies included the effect on hospitalizations into their analysis {1, 3, 4}. Klersy et al. 2011 {1} took the perspective of a third-party payer and (only) included healthcare resource consumption caused by HF related hospitalization (this was based on the assumption that expenditures for included patients are dominated by HF hospitalization costs).

Within a model analysis that takes lifetime perspective, other costs may also be relevant. Pandor et al. (2013), for example, modelling a 30-year time horizon, include routine clinical assessments as well as laboratory tests into their analysis with regard to long-term health care costs.

The above-mentioned resource items refer to resource consumption from a payer’s perspective. It can be argued that telemedicine interventions generate cost savings outside the healthcare system (e.g., through reduced patient travel costs) {10}, which would be highlighted, e.g., when including a patients’ perspective. Another related question is whether or not to include indirect costs in terms of productivity losses through, e.g., taking a societal perspective. The average age of the included study populations however usually lies between 60 and 75 years. Although a societal perspective is sometimes recommended, it may be argued that the proportion of people either retired or being off work because of their HF may be very high (thus estimates of, e.g., production losses would likely be rather negligible, even though, ideally, sensitivity analysis would be able to be performed surrounding such ‘productivity costs’) {10}, {3}.

The following table lists the identified resource items according to the cost and resource categories above.

Table 2: Resource items according to different cost (and resource) categories

*perspective: UK NHS, time horizon: 30 years, ** perspective: third-party payer, time horizon: 1 year, *** healthcare system perspective, time horizon: lifetime ^†HF=heart failure

Further information concerning the types of costs associated with ‘STS’ and ‘usual care’ can be found in Appendix ECO-4, this appendix attempts to summarise the information from the TEC domain, from the viewpoint of the ECO domain.

For this section we used results from the domain search (including domain searches from EFF and ORG; search strategy and selection criteria are described in the general methodology description) as well as results from an additional hand search, e.g. within study references. Also results from the TEC domain regarding description of the interventions were used

Different approaches have been used for the quantification of resource consumption. Miller et al. (2009) generally followed a micro-costing approach as they used data from a clinical trial. They give resource expenditures in U.S. Dollars but do not provide detailed data on resource consumption in units. Klersy et al. (2011) only regard the (average) number of HF hospitalizations per patient per year (calculated using a meta-analytical approach) {1}. Pandor et al. (2013) rely on different sources for their cost effectiveness analysis and include data from a randomized trial for the frequency of medical care visits {4}. For resource consumption related to the interventions themselves a bottom-up approach is used relying on advice from clinical experts as well as literature {11}. Long term costs are determined in reference to NICE clinical guidelines.

The following table lists resource items for intervention and comparator as well as long-term costs included in Pandor et al. (2013) together with quantification per 6 months of treatment as given within the study.

Table 3: Resource consumption for STS and usual care: quantification of resources

^{* Pandor et al. give yearly costs and include this only in a high cost scenario and obviously partly in the baseline scenario ** 3-year depreciation, centre has a monitoring capacity of 250 patients}

For this section we used results from the domain search (including domain searches from EFF and ORG; search strategy and selection criteria are described in the general methodology description) as well as results from an additional hand search, e.g. within study references. Also results from the TEC domain regarding description of the interventions were used.

The following table lists the resource items included in the three studies {1, 3, 4} that were selected as exemplifying STS cost items, together with measured and/or estimated costs as given in those studies. Concerning costs, in the modelling study of Miller et al. (2009) it should be noted that only patients with systolic heart failure were included. Miller et al. also use the assumption that both the intervention and the subsequent effects of the intervention last for the modelled patients’ lifetimes. In contrast, Pandor et al. (2013) {4} assume 1) the intervention lasts for six months, and 2) there are benefits due to reduced hospitalisations (producing HRQoL benefits) as well as 3) reduced mortality in the first 6 months. A lifetime perspective on health effects and costs is modelled by there being more people alive in the intervention group, after the first six months, than in the ‘usual care’ group, and this results in relatively more life years, and associated costs, in the STS groups.

Table 4: Resource consumption for STS, usual care, long term care and hospitalization: measured and/or estimated costs

* If not indicated otherwise.

** Range, depending on New York Heart Association (NYHA) Functional Classification –status. *** Only the average values of Table 3 are reported here.

The above table gives an idea about the rough dimensions of costs. Cost values however cannot directly be compared – not only because of different currencies and cost years, but also because of differing interventions, different modelling assumptions, and different populations between the studies.

All the studies reviewed here use charges or fees for estimating costs of the health care sector. Although these prices may not reflect the true opportunity costs of resource use, they seem to be justified for pragmatic reasons. Perhaps because of the assumptions in their model, Pandor et al. (2013) conclude that intervention costs only constitute a small part of the overall costs, hospitalization costs being the main contributor to those overall costs {4}. However, in all studies the question of the duration of treatment effects (whether of six to 18 months or lifetime -duration) and the implied costs is important.

The results from ORG8 provide the information that more than 70% of the studies reviewed by Grustam et al. (2014) did not take into account some cost items, or any costs, in at least one of the following categories:  healthcare sector; other sectors; costs to patients or family; and productivity losses for the patient or family {15}. None of the studies broadly analysed the shift of cost, for instance, from specialist HF nurses to general practitioners. In 80% of those studies the perspective, the source and the methods of the evaluations were not clear {15}. Authors mostly focused on direct costs and did not include indirect costs (e.g., productivity gains or losses) or ‘intangible costs’ (such as relief from pain, lost leisure time for patients or families). Of course, depending on the chosen analytical perspective, such approaches can be justified, however, some costs were not clearly included across majority of the studies, such as those costs related to the intervention’s overheads, costs associated with the training of personnel, and patient-related costs. It is also possible to value participant time,  in terms of labour, using either wages or a value for unpaid work. The valuation of participant time can be considered to be particularly relevant if the intervention has a long duration. Despite such considerations, the quality of evidence in much of the available scientific literature is poor, therefore, more studies on all aspects of costs related to STS would be needed to reach an unbiased conclusion {15}.

In ECO4 we report findings from the literature and from other domains, such as SAF and EFF on the effectiveness of ‘STS’ versus ‘usual care’ as they relate to health-related outcomes which can be considered important in the ECO domain. Given the weaknesses of the available evidence, we do not report extensive numerical results from the included studies but, instead, briefly describe some of the findings from those studies. This description includes information from the included economic evaluation studies in general, and Pandor et al (2013) {4} and Thokala et al (2013) {5} in particular.

In the most recent cost-effectiveness analysis (Pandor et al (2013) {4}; Thokala et al (2013) {5}) the main outcomes of interest all-cause mortality and hospitalisations. In these studies a Markov model was developed to estimate the prognosis for each HF patient using the monthly probability of death and monthly risks for hospitalisation from HF-related and other causes. Effectiveness parameters during the treatment period were the hazard ratios (HR) for all-cause mortality, all-cause hospitalisations and HF-related hospitalisations. Cost parameters, either estimated or based on clinical opinion, included both the costs of the intervention and costs related to hospitalisation.

The study of Miller et al (2009) estimated the long-term impact of telephonic disease management (TDM) in systolic heart failure patients from the results of an 18-month South Texas trial with a Markov model (Galbreath et al. 2004 {8}). Effectiveness was expressed as discounted QALYs saved with the DM compared to control group without TDM. The utility-adjustment weights were developed by NYHA class from the baseline results of the Medical Outcomes Study 36-Item Short-Form Health Survey (SF-36) collected from all trial participants, and the estimated mean utility-adjustment weights were 0.75 for NYHA I, 0.64 for NYHA II and 0.58 for NYHA III and IV.

It was noted in the work by Pandor et al (2013) {4} and Thokala et al (2013) {5} presenting results from the same cost-effectiveness modelling study that clear descriptions of the interventions and usual care were not provided in many of the studies and that this had implications for the robustness of analyses of effectiveness. In addition, results from the EFF domain show that at least half the studies included in Feltner et al. 2014 {12} are classified as having a high risk of bias. Although "all-cause death" appears to be reduced in some studies (in a statistically significant manner), a similar reduction is not reported for "disease-specific" or "disease-related" death. This could be due to small sample sizes in the trials, but although Krum et al. (2013) {13} report all-cause hospitalisation to occur more frequently in the control group, their clinical indicator, the Packer clinical composite score, does not show a statistically significant difference between the arms. Moreover, it is not clear to what extent the combination of the endpoints of “all-cause death” and “HF hospitalisation” which are used in a number of studies can be seen as valid, composite, health-related outcomes. Further, the intermediate health-related outcome “HF hospitalisation rate” alone does not tell us to what extent there could be “over-treatment” in a ‘Usual Care’ -group or “under-treatment” in a ‘STS’ –group, nor does it necessarily provide information related to any associated changes in costs.

Two of the articles identified through the systematic literature review present results from the same cost-effectiveness study (Pandor et al (2013) {4}, which is a National Institute for Health Research (NIHR) Health Technology Assessment -report, and Thokala et al (2013) {5}, which is a journal article). Information from both studies are presented when it is possible that it would be useful. In those two pieces of research STS is divided into two different approaches: STS via human to machine (STS HM) interface and STS via human to human (STS HH) contact. The estimations and analyses in this study also include a third intervention, home telemonitoring (TM), in which transmitted data is reviewed by medical staff or medical support is provided during office hours. The cost-effectiveness of each of these three interventions is estimated in relation to usual care, which is defined as usual care for patients discharged in the past 28 days with a heart failure (HF)-related hospitalisation as per the NICE Clinical Guidelines for the Management of Adults with Chronic Heart Failure. It is mentioned in the articles that the actual impact of each intervention is for the first six months after an initial discharge, since after 6 months it is assumed that all the patients receive usual care according to the aforementioned clinical guidelines. Pandor et al. (2013) {4} and Thokala et al. (2013) {5} thus both present results of a study employing a Markov model for a hypothetical cohort of 250 heart-failure patients, using a 30-year (or patient lifetime) horizon, with an annual discount rate of 3.5% for both costs and outcomes. The perspective is that of the NHS in England and Wales. The base-case cost-effectiveness results of each strategy compared to usual care and to the next most effective alternative are presented from this study. The probabilities presented below, of each strategy being the ‘most cost-effective’ at varying levels of willingness to pay per QALY gained, are from Pandor et al. (2013) {4}, since the results of Thokala et al. (2013) {5} are from the same study and quite similar. In the probabilistic sensitivity analysis (PSA) the percentage of model runs in which an intervention was the ‘most cost-effective’ strategy (at a £20 000 per QALY threshold) was 44% for TM (during office hours), 36% for human-to-human STS (STS HH), 18% for human-to-machine STS (STS HM) and 2% for ‘usual care’. When one study, the Home-HF study {14}, was excluded, the probability for TM increased to 73%. In the base-case analyses (including or excluding the Home-HF study) TM was found to be the ‘most cost-effective’ strategy, then STS HH. Results for the different scenarios were also presented: higher usual care cost scenario, lower TM cost scenario, higher TM cost scenario, higher STS HH cost scenario and lower STS HH cost scenario.  The conclusions regarding the relative cost-effectiveness of TM did not substantially change in the analyses using higher usual care cost, lower TM cost and higher STS HH cost. In the scenario with higher TM cost TM was dominated by STS HH, since the difference in expected QALYs between these interventions was small (0.0006), the change in the difference between the costs leads to a sizeable change in the ICER. However, in the same scenario, after exclusion of the Home-HF study, TM was still the most cost-effective strategy. It is stated in both articles that there is substantial uncertainty as to which strategy is the optimal in terms of net benefit, since the CEAC (cost-effectiveness acceptability curve) suggests that the best strategy is cost-effective in less than half of the PSA runs (with base-case costs and Home-HF study included). It was reported that this uncertainty was lower in the analyses that excluded the Home-HF study.

In the study by Miller et al. (2009) {3}, a Markov model was developed to estimate the cost-effectiveness of a telephonic disease management (TDM) program compared to control group without TDM. Both costs and effects were discounted at a rate of 3% per year. Costs are expressed as the difference in total discounted lifetime costs with and without TDM and effectiveness as discounted QALYs saved with TDM. The discounted effect in terms of QALYs was 0.111 and the discounted net TDM cost was $4 850 per patient; costs per QALY saved were estimated to be $43 650.

In the study by Pandor et al (2013) {4} and Thokala et al (2013) {5} uncertainties remain about the assumptions made in the estimation of both costs and effectiveness.

In the study by Miller et al. (2009) {3} the authors concluded that model results indicated both that TDM could be thought of as ‘cost-effective’ in the long term and that short-term results from a clinical trial alone might not reveal long-term cost-effectiveness.

As shown in ECO1, Table 2, only hospitalisation costs were included and other costs such as costs related to the remote monitoring of patients, outpatient visits and drug costs were not considered in the study by Klersy et al. 2011 {1}. In the study by Miller et al. (2009) {3}, the authors provided only an average cost of the STS programme and did not provide the breakdown of the individual cost components {4}. In summary, as noted in ORG8, it is useful to note that the methods of cost calculation vary widely across the studies related to STS {15}.

From the EFF domain and from Pandor et al. (2013) {4} we see that the varied estimates of effectiveness are also based on a variety of methods of calculation ranging from estimates based on a single trial, to those based on network meta-analysis (NMA). As Pandor et al. (2013) note, it is important to consider different approaches to STS separately as they are likely to have different clinical effectiveness and costs associated with them. Therefore, specific approaches to STS would ideally be fully described before estimating the cost-effectiveness of the interventions {4}. In the study by Pandor et al (2013) {4} and Thokala et al (2013) {5} individual patient-level data was not used and no adjustment was made for potential biases arising from study quality of the studies included in the NMA. The study by Miller et al. (2009) {3} mainly uses information from Galbreath et al. 2004 {8}, which is classed in EFF1 as having a high risk of bias, and the potential extent of the effect of structural uncertainty on results is not described. For these reasons, it is not fully possible to assess the effect of parameter and structural uncertainty on the results presented in the three cost-effectiveness studies reviewed here ({1}, {4} and {5}).

The available primary and secondary studies were not sufficient to provide an answer to this question.

The lack of information concerning subgroups was noted, for example, by Pandor et al. (2013) {4}. They suggested that future studies should publish data in such a way as to identify which patient subgroups benefited most from the intervention.

Because of the results presented in ECO1, ECO2, ECO3, ECO4, ECO5 and ECO6, serious doubts are raised about the extent to which the estimates of costs, the estimates relating to health-related outcomes, and economic evaluations per se can be considered as providing valid descriptions of Structured telephone support (STS) for adult patients with chronic heart failure compared with its comparator, 'usual care' without STS.