HIS modelling and simulation based cost–benefit analysis of a telemedical system for closed-loop diabetes therapy

1. Introduction 

Telemedical systems are often characterised by high investments and high regular costs. The decision for introducing such systems into clinical routine is usually based on a positive cost–benefit relation from the perspective of payors for health services, i.e. health insurance companies. Thus, new telemedical approaches need to show an adequate level of benefit in relation to their costs compared with current standards of diagnosis or treatment [1].

Current economic evaluation methods that focus on the comparison of decision alternatives are cost-minimisation analysis (CMA), cost-effectiveness analysis (CEA), respectively, cost-utility analysis (CUA), cost–benefit analysis (CBA), return of investment, and functional economic analysis [2]. Several health economic evaluation guidelines have been developed by government and industry organisations that usually recommend CME, CEA/CUA and CBA [3]. CMA compares the cost of each alternative usually assuming equal effectiveness of each alternative. CEA/CUA compares costs and effectiveness of each alternative through defining a non monetary effectiveness parameter like, e.g. gained quality adjusted life years (QALY [4]) and calculating cost-effectiveness ratios. The CBA additionally transforms all costs and benefits, respectively, the health outcomes of the alternatives into monetary terms. A prominent approach for CBA is assigning QALYs gained a monetary value [5]. Besides societal perspectives for economic evaluation that takes into account costs and benefits of all parties including patient, health care professionals and payors, an economic evaluation may focus on a certain narrower perspective of a decision maker, e.g. of an insurance company [6].

A thorough economic evaluation is recommended to be part of each evaluation of a telemedical approach [6], [7]. Several reviews focused on the incorporation of economic evaluation in telemedicine interventions and commonly conclude that evidence regarding the effectiveness or cost-effectiveness of telemedicine is limited and good-quality studies are still scarce [8], [9], [10], [11], [12]. Especially studies addressing the longer-term economic impact of telemedicine are relatively rare [13]. The evidence for the effectiveness of telemedical solutions for diabetes care in reducing costs is also not strong [14].

The telemedicine project INCA1 (Intelligent Control Assistant for Diabetes [15], [20], [21]) focuses on improvements in insulin pump therapy of diabetes type 1 (abbreviation CSII for continuous subcutaneous insulin infusion). INCA is a follow-up project of the European projects ADICOL (Advanced Insulin Infusion using a Control Loop [18]), and M2DM (Multi-Access Services for telematic Management of Diabetes Mellitus [16]) that itself bases on the DIABTel-Project [17].

Diabetes mellitus is a metabolic disorder characterised by hyperglycemia (high blood glucose level) and thus diabetes therapy focuses mainly on glycemic control, i.e. to keep the blood glucose level in a normal range. Besides preventing hyper- and hypoglycaemias that are direct effects of an insufficient regulation, also long-term complications of diabetes, like microangiopathy (retinopathy, nephropathy, etc.), macroangiopathy (stroke, heart failure, etc.), neuropathy, etc. can be slowed or prevented by intensive glycemic control [19]. There are several different types of diabetes. Diabetes mellitus type 1 is characterised by a loss of the beta cells of the pancreas leading to a deficiency of insulin, the hormone that normally regulates the blood glucose level.

The idea of INCA is that a Smart Assistant (a personal digital assistant with smart phone capabilities) controls the insulin pump using an algorithm that besides other parameters depends on continuous blood glucose measurement—thus realising a closed-loop. This so called Personal Loop is embedded in a telemedical supervision of the therapy called Remote Loop. Within this Remote Loop a physician monitors the glucose metabolism of each patient and gives therapy recommendations when needed. Vice versa the patient can get into contact with the physician using a sophisticated messaging system. Besides analyses of clinical effects of the INCA system through clinical studies that are subject of other publications (e.g. ref. [20]) the question of suitable procedures for the health economic evaluation of this or similar telemedicine approach arises.

2. Objectives 

Objective of our study is to analyse the costs and benefits of a new telemedical concept (i.e. INCA) from the perspective of the payor for health services (i.e. a health insurance company) in comparison with the costs for the conventional therapeutic concept (i.e. manually controlled CSII). This raises the following questions:

3. Materials 

To answer Q1 and Q2 our cost–benefit analysis is divided into a cost and a benefit analysis. The cost analysis (Section 3.1, question Q1) determines the yearly costs of diabetes management and treatment, but without considering diabetes related secondary diseases. A better therapy of diabetes results in a decrease of the probability to develop diabetes related complications. Thus, the monetary benefit of a new therapy for the chosen perspective corresponds to a reduction of the costs for treating complications. Therefore, the benefit analysis (Section 3.2, question Q2) focuses on the potential of the INCA system to slow onset and progression of diabetes related complications in the long term and quantifies the reduction of costs for treating these complications. The clinical outcome of utilising INCA concepts, that is the basis of the benefit analysis presented here, has been analysed in a clinical study conducted in Munich [20]. For both, cost and benefit analysis, different modelling and simulation approaches were used: MOSAIK-M for cost analysis and Archimedes for benefit analysis.

3.1. Cost analysis 

MOSAIK-M (an acronym for modelling, simulation and animation of information and communication systems in medicine) is a methodical framework and tool environment for health information systems (HIS) analysis and design [21]. The methodical framework consists of a generic process scheme and an HIS meta model. The MOSAIK-M generic process scheme has been developed to guide modelling projects in producing HIS models of high quality in terms of correctness, completeness and validity corresponding to the unnecessary objectives of the project. It has to be adapted for specific modelling projects, but generally consists of the phases definition of the problem domain, analysis of the problem domain, modelling, evaluation (including verification, respectively, validation by experts or future users) and utilisation of the resulting model.

The meta model of MOSAIK-M defines the modelling language for creating HIS models and is based on the IS definition similar to [22]: A HIS is the partial system of a health care organisation that comprises all information processes and all of the human and technical resources that are involved in these processes in their information processing roles. The concepts of human and technical resources in their information processing roles are formalised by the meta model components actor and actor role (e.g. a health care professional, a medical device or a piece of software). Actors, respectively, actor roles are assigned to functional units that can hierarchically contain other functional units.

The behaviour of an actor or a functional unit is described through processes. Processes are modelled hierarchically using a specific, formal Petri Net syntax that uses boxes for basic activities or sub processes and circles for conditions or object flows [23]. The MOSAIK-M tool environment simulates an HIS model through stepwise execution of processes by actors. A process execution is initiated by corresponding events. While executing a single activity of a process, the manipulation of objects or applications by an actor is also simulated, eventually triggering further events. All relevant real world objects (e.g. forms, records, devices) are modelled using the Unified Modelling Language (UML [24]) that is partly integrated into MOSAIK-M.

To describe the software application system of a HIS, the meta model is divided into application objects and applications, which work with them. Both are modelled using UML. Applications are described implementation independent as application models usually exemplified by user interface prototypes.

MOSAIK-M HIS models can be used for HIS analysis, HIS design, business process (re)engineering or process oriented quality assurance. In the INCA project MOSAIK-M was used as an integrated approach for system analysis, design and cost-analysis (see Fig. 1, ref. [21]). Therefore, the generic process scheme of MOSAIK-M was instantiated four times. The first iteration of this process scheme took place in work package “Identification of user needs” and led to a model of the problem domain reflecting the current conventional CSII treatment. This CCSII-model comprises 8 health care institutions (e.g. diabetes clinic), 25 actor roles (e.g. diabetologist, patient), 156 processes (e.g. opthalmological examination) and 81 real world object classes (e.g. insulin pump).

View Large Image Download to PowerPoint

Fig. 1. Use of modelling approaches in the INCA-project for system analysis, design and cost-/benefit analysis.

The CCSII-model was the starting point for generating a model of the INCA telemedical system including the personal and the remote loop. This second instantiation of the generic process model in the INCA project phase “Analysis and Design” led to the INCA-model that includes the system specification useful for the implementation (12 institutions, 37 roles, 341 processes, 81 object classes, 358 application models, 112 application object classes; see Fig. 2). Both models have been evaluated by domain experts and future users of the system [21].

View Large Image Download to PowerPoint

Fig. 2. Components of the MOSAIK-M INCA-model.

The third and fourth iterations of the generic process scheme in the INCA project phase “Deployment and exploitation” enhanced the CCSII- and the INCA-model by cost structures to prepare a simulation based cost-analysis of the INCA system compared to conventional CSII treatment. Prior to these iterations and the simulation MOSAIK-M's meta model and tool environment have been extended to allow for modelling of resource classes (e.g. an infusion set), resource consumption (e.g. one glucose sensor in three days), process costs (e.g. costs of an ophthalmologic examination) and patient profiles (e.g. age, secondary diseases). Additionally the simulation environment has been equipped with capabilities to store resource consumption and process costs related to a patient during a simulation run. Thus, the health care related processes of a patient can be simulated on a day to day basis for a given time frame to determine the corresponding health care costs for the payor of health services, i.e. the costs of medical treatment, telemedical support, medical devices, other adjuvants and consumables. Accompanying the extensions of MOSAIK-M's meta model and tool environment, its generic process model has been specified towards conducting a cost-analysis based on MOSAIK-M (see Fig. 3, left side).

View Large Image Download to PowerPoint

Fig. 3. Process model of INCA cost–benefit analysis.

Based on these general enhancements of MOSAIK-M both the CCSII- and the INCA-model have been parameterised with costs of diabetes treatment and management. The cost parameters were acquired in September 2005 from price lists of German suppliers of medical devices, adjuvants or consumables and from official price lists for medical services in Germany [25]. To estimate patient related provider costs of using INCA telemedical services, cost structures of comparable commercial services have been adopted (€59 p.a.). This assumes that the payor (i.e. the insurance company) does not operate the service itself but pays for it as well as for patient related devices needed to access the service (i.e. the Smart Assistant including communication costs: €305 p.a.).

The simulation with MOSAIK-M comprises 1 year for both, the CCSII- and the INCA-model. The simulation runs focussed on CSII based diabetes treatment and management of a diabetic who does not suffer from diabetes related secondary diseases. A profile with no secondary diseases was chosen because the cost–benefit studies focus on the capability of the INCA system to prevent secondary diseases. The model assumes that patient and health professionals behave consistent with the evidence based guideline of the German diabetes association [26]. The simulation of the INCA-model additionally includes a Smart Assistant, a continuous blood glucose measurement device, and online connections to the Remote Loop system (via GPRS and the Internet). Because the original INCA system design is different to the way the INCA system was configured and used within the clinical studies of the INCA project two simulation runs of the INCA model were conducted to assess the implications of both configurations for the costs of system utilisation: one based on 24h seven days a week continuous glucose measurement (original design) and one based on the using continuous glucose measurement for 72h during 1 month (configuration of the clinical study in Munich [20]). For each 1-year simulation, the direct costs of diabetes treatment and management were accumulated.

4. Results 

For the selected patient profiles (diabetes type 1, no secondary diseases, guideline based diabetes treatment and management) a simulation was conducted for the MOSAIK-M CCSII-model and for the INCA-model to determine the costs of daily routine of diabetes management (i.e. preparative, singular processes like training have been omitted):

•Simulation of conventional CSII treatment (CCSII-model): yearly costs of conventional treatment and diabetes management are €5908 distributed between consumables (81.2%), devices and other adjuvants (15.4%) and medical services (3.4%).

•Simulation of telemedically supported treatment (INCA-model):

-24h, 7 days a week continuous glucose measurement (original INCA concept): yearly costs of INCA related treatment and diabetes management are €13,265 (consumables 80.8%, devices and other adjuvants 17.6%, medical services 1.6%).

-72h every 4 weeks continuous glucose measurement (Munich study configuration): yearly costs are €8140 (consumables 67.7%, devices and other adjuvants 29.6%, medical services 2.7%).

The costs of the INCA based treatment are sensitive to changes of the costs for continuous glucose measurement. Twenty-four percent of the costs for the INCA based treatment with 72h every 4 week continuous glucose measurement is caused by the measurement device and the consumables needed. The latter accounts for 10% of the costs. Five percent of the INCA costs are caused by IT (smart assistant and service costs). In the 24h 7 days a week scenario 64% of the costs are resulting from continuous glucose measurement (55% by the measurement consumables), 3% from IT.

The results of the Archimedes simulation concerning development and progression of diabetes and diabetes related complications dependent on five HbA1c values are presented in Table 1. The table lists for each study group of the Munich clinical trial the statistical probability and probable age of developing a diabetes related disease—the latter represents the age where 50% of the patients that would develop this disease in the 30-year time frame at all statistically would have developed this disease in fact. To analyse the sensitivity of the results concerning the HbA1c level reached the last row presents the results of the simulation for the case of good glycemic control with an HbA1c value of 6.5% that is near normal. All of the disease probability values are equal or lower in each case of decreasing the HbA1c value.

Table 1. Results of the Archimedes simulation of the selected patient profile: probability and probable age of developing diabetes related secondary diseases within the next 30 years starting at age 34

	Study group	HbA1c (%)	Secondary disease (probability and probable age)
			Heart attack	Stroke	Kidney failure	Eye problems	Foot problems
	(1) INCA start	7.9	22.3%, 56.3	8.5%, 56.4	2.2%, 49.0	1.6%, 49.0	8.0%, 56.7
	(1) INCA end	7.5	22.1%, 56.3	8.5%, 56.6	1.7%, 49.0	1.6%, 49.0	4.1%, 59.2
	(2) INCA start	7.6	22.2%, 56.3	8.5%, 56.5	1.8%, 49.0	1.6%, 49.0	4.8%, 59.1
	(2) INCA end	7.3	22.0%, 56.2	8.4%, 56.6	1.5%, 49.0	1.6%, 49.0	2.9%, 59.2
	Near normal	6.5	17.6%, 56.0	7.1%, 57.4	1.3%, 49.0	1.5%, 49.1	0.9%, 61.0

Table 2 presents the addition of the yearly costs for treating and managing diabetes in a conventional CSII or an INCA conformant way resulting from the MOSAIK-M analysis with the average costs of treating diabetes related secondary diseases resulting from the cost adjusted Archimedes results presented in Table 1. The development of the total costs and of the included costs of treating diabetes related complications only are presented for two case scenarios of an HbA1c value reduction from 7.9 to 7.5% and from 7.6 to 7.3%, which corresponds to the results of the Munich clinical study. To analyse the sensitivity of the results concerning achieving a much better HbA1c level, again a third scenario with an HbA1c of 6.5% representing a very good metabolic control is added.

Table 2. Results of the cost–benefit analysis for two HbA1c decrease scenarios based on the results of the Munich clinical study and a virtual best-case scenario

		Munich study group
		1	2	va
	HbA1c
	HbA1c (conventional CSII) (%)	7.9	7.6	7.9
	HbA1c (INCA) (%)	7.5	7.3	6.5
	Costs (€)
	Complete p.a. (conventional CSII)	6608	6504	6608
	Complete p.a. (INCA)	8710	8666	8584
	Difference	2102	2162	1976
	Complications only p.a. (conventional CSII)	700	596	700
	Complications only p.a. (INCA)	570	526	444
	Difference	130	70	256

p.a.: Per annum.

a Virtual study group assuming the INCA system would result in a HbA1c value near normal.

The additional mean costs for an INCA conformant telemedical approach would be €2102 (HbA1c of 7.5%), respectively, €2162 (HbA1c of 7.3%) higher than the yearly costs of conventional treatment, when the latter results in a HbA1c value of 7.9%, respectively, 7.6%. The effect on the yearly costs for treating complications only are −€130 for a HbA1c decrease of 0.4% starting at 7.9%, respectively, −€70 for a 0.3% decrease starting at 7.6%. The savings of (virtually) reaching a HbA1c value of 6.5% are €256. These savings are the direct benefit of reducing the HbA1c value, i.e. any therapy approach that results in a corresponding HbA1c decrease and produces yearly additional costs below this threshold would produce equal or lower direct costs for treating diabetes and its complications compared with a therapy that results in a HbA1c value of 7.9%, respectively, 7.6%.

5. Discussion 

The presented approach demonstrates that reusing HIS models originally produced for the development of an HIS can be an efficient way to conduct costs analyses. A prerequisite is that HIS models that can be simulated already exist, have a sufficient granularity, and that cost attributes can be integrated easily. Sufficient quality of the models in form of correctness and completeness is also a prerequisite that can be assured by corresponding model verification and validation steps being part of the modelling process. MOSAIK-M reflects this by systematically integrating an evaluation phase in its modelling process scheme.

Other integrated and commercially available IS modelling approaches like the ARIS approach [32] also offer functionalities to model process-cost related aspects. Essential features of the MOSAIK-M approach are its sophisticated process modelling and simulation capabilities combined with UML-based object oriented modelling and user interface prototype integration useful for user-centred system design. Further simulation based techniques for cost evaluation especially in telemedicine have also been used successfully (e.g. ref. [33] and more recently [34]) but did not made use of models produced originally for systems analysis and design. Simulation based approaches like MOSAIK-M have the advantage that they are flexible concerning analysis of changes in process descriptions (like e.g. alternative treatment or disease management processes), and event probabilities (e.g. dependent on patient characteristics like the amount and severity of secondary diseases) with regard to their impacts on costs. Analytical calculations can be utilised instead but they necessarily copy the complexity of the scenarios modelled and thus produce additional effort when simulatable models are already available.

The cost–benefit analysis was conducted from the perspective of the payor for health services and thus does not claim to consider a holistic societal perspective. Indirect costs (e.g. loss of working hours) as well as intangible costs (e.g. reduction of quality of life) have not been taken into account. Additionally potential losses of interests have also not been considered. Further analyses are needed to consider these aspects.

A huge share of the yearly costs of operating the INCA concept is resulting from continuous glucose measurement (24%, respectively, 64%). Especially the costs of consumables are immense when operating this device 24h 7 days per week. Significantly reduced prices for the glucose monitoring device and its consumables would considerable improve the cost/benefit relation of the INCA system.

The results of Archimedes are limited to a time period of 30 years and on a subset of the most relevant complications. Significant costs of treating diabetes complications arise in time periods beyond the 30-year time period considered here [35]. Due to the chronic character of diabetes complications, the calculated savings are therefore a lower bound of the real savings.

The benefit analysis presented bases on changes of the HbA1c level observed in the Munich clinical study. The study used an INCA system configuration that differs from the original concept of remotely controlled 24h/7 days closed-loop therapy. Thus, it can be assumed, that further improvements of the glycemic control can be achieved. But even if an HbA1c level of 6.5% is realised, the costs of INCA remain higher than the costs of conventional CSII. This is due to relative small mean savings concerning the treatment of diabetes related secondary diseases within the 30-year time frame considered. The relative small effect on the costs for treating secondary diseases is on a first glance surprising, but other computer simulation based studies observed similar effects [36]. Concerning our results it has to be considered, that the Archimedes simulation bases on a population similar to the study population that represents a subset of diabetes type 1 CSII patients. Thus, the results are valid only for this population. Other populations might profit from an early and durable improvement of the HbA1c level to a greater extent.

Other validated simulation based approaches like the CORE model [37] are an alternative to the Archimedes model. Simulation results of the CORE model also describe the long-term incidence and progression of diabetes-related complications. Roze et al. used the CORE model to conduct a methodically similar cost–benefit analysis concerning the comparison of CSII versus multiple daily injections for the treatment of type 1 diabetes in the UK [38].

6. Conclusion 

Efficient and integrated methods and tools are needed to ease cost evaluation of new telemedical systems. The presented approach of using MOSAIK-M and Archimedes for cost–benefit analyses produces information useful to discuss the introduction of telemedical systems. Its main advantage is that only few modifications of the model or its parameters are needed to simulate cost/benefit effects of new scenarios. The approach produces a conservative estimation of a lower bound for cost savings. Further work is needed to improve the approximation. To improve cost efficiency of INCA from the chosen perspective, the costs of system operation and especially the continuous glucose measurement need to be notably lowered.