We screened in total 1,309 publications of which twenty-seven studies meet our inclusion criteria for this economic review.[5,13-38] The process of our literature search as well as the reasons for excluding a number of studies is provided within the PRISMA flow-diagram in Figure 1. Furthermore, an overview of the characteristics of included studies is listed in Table 2.
Table 2: Characteristics of included studies (n=27)
Generally, twenty-two studies (81%) [5,13-16,18,20-25,28-31,32,34-38] out of the included twenty-seven studies report cost savings after implementing an EHR based CDS intervention. Four studies (15%) [17,26,27,33] report a rise in cost expenditure. The remaining study (4%)  did not detect significant differences in cost outcomes. Furthermore, in the majority of included studies the main cost outcome measures were laboratory test cost.[15-17,20,21,25,28,29,31,32,38]
Exploitation of different front-end CDS intervention categories
According to the taxonomy by Wright et al., we identified ten (37%) studies [5,13,15,20,22,23,26,36-38] which explored EHR based CDS interventions based on point-of-care alerts or reminders (category 3). In addition, three interventions (11%) [17,27,34] were order facilitators (category 2), while medication dosing support, relevant information display as well as expert systems (categories 1, 4 and 5) were each reported only once from an economic perspective (11%).[18,19,24] In eight studies (30%), [14,16,25,28,30,31,33,35] interventions from two different categories were explored in combination. Moreover, we found three studies (11%) [21,29,32] in which the option to place a certain order or test, e.g. a laboratory test, was removed from the EHR CPOE system or the clinician’s laboratory ordering preference list. These restrictive frond-end CDS intervention types were not yet mentioned in the pre-defined categories by Wright et al.. Thus, for this study we extend their taxonomy by a new category
- Restriction of choice
The removal of an order option ultimately resulted in less laboratory tests ordered, and therefore in a reduction of healthcare expenditure in all studies it was implemented [14,28,31]. Finally, we identified two different types of implemented hard-stops. An interruptive intervention, which requires a clicking response from the physician before being able to move forward, and a restrictive hard-stop, which prevents the physician from ordering a test, e.g. by directing them to call the laboratory director in case they persist on the order. We grouped studies regarding the interruptive alert to category 3 and the hard stop restrictive intervention to category 7.
Economic impact for prevalent application areas
In Table 3, we summarized our findings and created an overview of application areas and cost outcome measures in relation to the applied CDS intervention types. Due to the heterogeneity of included studies with regard to different types of cost outcomes reported and different intervention duration, it was not possible to conduct a subgroup analysis considering the economic impact of each CDS-front end category. A detailed evidence synthesis of all included twenty-seven studies as well as a brief description of their intervention types, their application area and the resulting economic impact is provided in an additional file [see Additional file 3].
Table 3: Application areas and cost outcome measures in relation to CDS intervention categories 1.-7.
Application areas for cost-savings: Thereupon, we identified four main application areas based on their investigated prevalence that resulted in cost-savings after EHR based CDS implementation. Firstly, two studies report on essentially reducing unnecessary Vitamin D routine testing that led to a decrease of laboratory test cost of $300,000 and $1,4mill. per year.
Secondly, two studies addressed the economic outcomes of the reduction of waste in transfusion practice and red blood cell usage.[36,37] Acquisition product cost of red cell units were decreased with the help of EHR based CDS and resulted in cost savings of in total $4,821,000 within three years and about $62,715 within one year after implementation, respectively.
Thirdly, two cost-effectiveness-analyses modeled the cost outcome of reducing antibiotic prescriptions for acute respiratory infection as well as for acute bronchitis.[5,33] Gong et al. include a full accounting of costs into their Markov model and explore that the implemented CDS intervention, called “suggested alternatives”, yielded more quality adjusted life years (QALYs) at a lower cost of $500,000 per 100.000 individuals over thirty years of implementation. Michaelidis et al. on the other hand report a small increase in costs compared to a printed decision support system, i.e. posters. However, the outcome of the latter mainly results from a cost difference between the direct costs of poster printing and the computer programming cost.
Lastly, five studies[20,29,31,32,38] report on the potential for cost savings through reducing duplicate orders or laboratory tests by using hard-stops or applying order frequency rules to prevent ordering the same test within a certain timeframe. Reducing laboratory duplicate tests resulted in savings of $3,395 in three months for a small patient size cohort and up to $315,565 within twenty-four month for a large patient size cohort.
Application areas resulting in cost increase: Furthermore, we also identified risk areas, which possibly lead to a further increase in healthcare expenditure. One study found that after implementing a CPOE system with default settings, specialized HIV laboratory test cost increased by $14,000-$96,000 within six months. Another study reports that an unplanned change of a pre-selected default order for ‘complete blood count’ to ‘complete blood count with differential’ lead to an average cost increase of $293.11 per day. Finally, the implementation of order sets as decision facilitators possibly entail negative economic effects. One study found that only after the uncoupling of Vitamin B12 and serum folate joint orders within predefined order sets, laboratory test cost decreased by about $26,719 per year. Similarly, another study removed the option to order daily routine tests from automated admission order sets and found savings of $26,416 after two months.
Table 4 encompasses an overview of studies which conducted a cost-effectiveness-analysis (CEA) of EHR based CDS interventions considering various cost data as well as economic outcome measures, such as the incremental cost-effectiveness ratio (ICER), which depicts the incremental change in costs divided by the incremental change in health outcome or effect.
Cost-effectiveness-analyses aim to reveal the trade-offs in resource-allocation decisions. In this context, it is essential to investigate when and to what extend upfront and maintenance cost for an EHR based CDS system will be amortized by its benefits, which again can be measured either in health outcomes, such as quality adjusted life years (QALYs) saved or in reduction of unnecessary healthcare utilization.
Table 4: Overview of cost data and cost outcome of model-based studies (n=4)
Generally, two studies report an increase in healthcare expenditure from a societal perspective, [26,33] while the other two report cost savings from a societal perspective as well as the medical group’s perspective.[5,35] Notably, the measurement of effectiveness was single study-based estimates in all four studies.
Regarding the consideration of upfront implementation cost, Gong et al. include only base case consolidated cost data of $1.91 for a cohort of 100.000 individuals based on expert opinions. Sharifi et al. include intervention start-up cost for EHR modification of $2.7mill. as well as other direct cost, such as professional care provider training. Michaelidis et al. report implementation and maintenance cost data, which is physician education per hour and medical record and CDS programming per patient of $18 in the base case. Lastly, Forrester et al. report CPOE CDS system cost as hardware, software and maintenance cost starting from $373,000 in year one to $92,000 after five years, as well as personnel, $555,000 in year one, and indirect cost as 3% of the total cost. Interestingly, the latter also include the HITECH Meaningful Use incentives in their model in order to simulate the financial incentives by the Centers for Medicare & Medicaid Services in the US.[35,42]
Lack of considering all cost components
Despite revealing major potentials for cost-savings, we could not asses the quality of included studies, because of the lack of cost information provided, or predominantly the lack of considering all relevant cost components. According to the Consolidated Health Economic Evaluation Reporting Standards (CHEERS) statement, most of the reported recommendations were not satisfied. All twenty-three non-model studies (85%) only calculate the economic outcome based on financial data reported before and after intervention implementation, which for instance, ultimately results from the computation of price per healthcare resource utilization times the quantity of used healthcare resources or services. Thus, even though it was not intended in those studies, it is necessary to mention that only four of them adhered to sound economic evaluations as recommended by CHEERS.[5,35]
The challenge of heterogeneity for the CEA is also aggravated by considering different cost outcomes. Two studies do not directly report an incremental cost effectiveness ratio (ICER) for a predefined threshold, nor include comparative metrics.[33,35] Other standardized metrics, such as the return on investment or net present value , were also not examined in the included studies. Only one study reported the net monetary benefit (NMB) of the intervention in relation to a predefined threshold.[5,44]
Additional studies worth mentioning
Notably, five more studies [45-49] meet most of our inclusion criteria, but were excluded due to various, although little, deviations. Three studies [45-47] report cost-savings after a bundle of information technology was implemented simultaneously, but the economic benefit could not solely be attributed to the EHR based CDS intervention. The fourth publication is a NHS health technology assessment (HTA) report. In this HTA, a RCT was conducted in 79 general practices in the UK in which a multicomponent intervention was installed using electronic health records in order to reduce antibiotic prescribing for respiratory infections. The authors perform a basic cost-analysis on whether the cost of healthcare utilization, that is the number of provider consultations, will increase during the time of the trial under the CDS intervention arm, and if patients more often re-consult the physician when not given a prescription. However, the authors explored no difference in cost outcome between the intervention and control period.
The last study worth mentioning compared retrospectively generated alerts by an advanced machine learning CDS system with alerts triggered through the home-grown EHR based CDS system. The authors calculated the healthcare costs of potentially prevented adverse drug events and medication errors, and found that by using the advanced machine learning CDS system 68,2% of alerts were only fired by that new system resulting in cost savings of $60.67 per alert. After extrapolating these results to an local patient population of 747,985 over five years they estimated savings of $1,294,457.