Clinical Decision Support Systems for Diabetes Care: Evidence and Development between 2017 and Present

Abstract

The clinical decision support systems (CDSs) for diabetes have improved significantly over the years. Multiple factors serve as driving forces for the uptake of CDSs. Newer technologies, initiatives, government mandates, and a competitive environment collectively facilitate advancement in diabetes care. This book chapter summarizes global CDSs development in recent years. Our review of the past few years’ publications on CDSs for diabetes shows that the United States is leading the world in technology development and clinical evidence generation. Developing countries worldwide are catching up in CDSs development and standards of patient care. Though most CDSs and published studies are on diabetes diagnosis, treatment, and management, a small portion of the research is devoted to prediabetes and type I diabetes. Increased efforts worldwide have been devoted to artificial intelligence and machine learning in diabetes care.

1. Introduction

Globally, chronic care conditions burden society with high costs and diminished quality of life for affected individuals. According to the Center for Disease Control and Prevention (CDC), more than one in 10 Americans ha diabetes mellitus, commonly referred to as type-2 diabetes (T2DM), and approximately one in three has prediabetes. Diabetes was the seventh leading cause of death in the United States in 2017. People with diagnosed diabetes, on average, have medical expenditures 2.3 times higher than those without diabetes [1], and 25% of all medical costs in the United States are spent on caring for people with diabetes. Diabetes can result in disabling complications, comorbidities, and reduced life expectancy. Effective management of diabetes is important to improve the quality of life for diabetics as well as improve population health and control medical costs. Attention and interventions are needed to address the issue of rising costs. Clinical decision support systems (CDSs) may offer the solution to rising costs, quality of care, patient engagement, patient-centered care, personalized medicine, clinical inertia, and clinical outcomes.

According to KBVResearch, the global CDSs market will grow from 2.9 billion in 2017 to 8.9 billion in 2027 [2]. The adoption of Electronic Health Record (EHR) and CDSs has been on the rise across the globe. Developed countries lead the development and implementation of CDSs. Several factors contribute to the increased acceptance and adoption of CDSs: general acceptance of using technologies across the entire healthcare spectrum, including adherence to clinical guidelines, evidence of improved clinic outcomes, government incentives, compliance/regulatory requirements, and operational efficiency. In this book chapter, we provide an overview of recent evidence on CDSs for diabetes care by searching relevant publications in CINAL, PsychInfo, Web of Science, Scopus, Medline, and PubMed from 2017 to the present.

2. Clinical decision support systems for diabetes care

2.1 Diabetes care

Diabetes is a chronic disease. Adequate diabetes care requires attention to biomarkers such as blood pressure, cholesterol, blood sugar level, and lifestyle changes. Care typically involves management of blood pressure, lipids, smoking, glucose, weight, screening for eye, foot, renal and vascular complications, and immunizations. It is common that patients with diabetes also have one or more other comorbid conditions. Thus, caring for diabetics is a team effort, and many providers may be involved, including various types of physicians or nurse practitioners, pharmacists, case managers, dieticians, and specialty doctors such as cardiologists, dentists, ophthalmologists, others. The literature has consistently reported a gap between current diabetes care practice and recommended diabetes care standards. This includes the concept of clinical inertia or the failure to start or accelerate a current or new therapy when appropriate. Clinical inertia may be due to the clinician’s lack of knowledge or inexperience with new therapeutic interventions and drugs available to treat diabetes [3].

Many IT-based interventions have been developed to improve adherence to the quality of care standards for chronic illnesses such as diabetes. CDSs for diabetes have been developed to address prediabetes screening, type I, type II, and gestational diabetes diagnosis, treatment, and care. Figure 1 shows the publications related to CDSs in diabetes. Though CDSs predated EHR, it is well documented in the literature that the adoption of CDSs is low [4].

Figure 1.

CDSs research areas.

2.2 Clinical decision support (CDS)

A clinical decision support (CDS) is a computerized system that uses case-based reasoning to assist clinicians in various decision-making such as assessing disease status, diagnosis, selecting appropriate therapy, or making other clinical decisions [5]. CDSs are typically used at the point of care where clinicians can make treatment decisions either based on their own knowledge or by combining their knowledge with patient characteristics or recommendations provided by the CDS through a clinical disease-specific knowledge base. CDSs provide alerts, reminders, or feedback to a care team [6]. A CDS can improve healthcare delivery by improving medical decisions with targeted clinical knowledge, patient information, and other health information [7].

2.2.1 History of clinical decision support

The idea was generated in the 1950s. In the late 1960s, F. T. deDombal and his associates at the University of Leeds studied the diagnostic process. They developed the Leeds abdominal pain system, a computer-based decision aid using Bayesian probability theory to explain seven possible causes of acute abdominal pain. In the 1970s, Stanford University developed MYCIN, rule-based decision support using a reasonably simple inference engine and a knowledge base of 600 rules. Later, Help was developed, and both MYCIN and HELP could generate alerts when abnormal factors were observed. Earlier studies on CDSs report that the use of automated clinical guidelines for diabetes in general practice did not result in a clinically significant change in doctors’ behavior or in patient outcomes [8].

2.2.2 Components of CDSs

Figure 2 depicts the components of a CDS. Typically, a CDS consists of a knowledge base, inference engine, and communication mechanism. The knowledge base contains facts, best practices, clinical guidelines or protocols, drug interactions, drug allergies, and logical rules. The inference engine combines patient-specific data (demographic data, medical history, family history) with clinical knowledge and performs reasoning. The communication mechanism takes patient data as input and produces output including alerts, reminders, summaries, etc.

Figure 2.

Components of CDSs.

Different technologies are used to build CDSs. Some use open-source software. For example, Protégé and WebProtégé are free software programs for building ontology knowledge solutions, and Jena is the Java rule-based inference engine. WebProtégé builds drug knowledge, and Jena evaluates the antidiabetic medications reasoning module [9].

2.3 Benefits of clinical decision support systems in diabetes

Digital transformation involves fundamentally rethinking healthcare delivery processes, treatments, and services from a technology-enabled perspective. CDSs promote diabetes care by facilitating evidence-informed insulin use, improving blood glucose control, and quality indicators in caring for patients with diabetes. Given the complex undertaking for clinicians, CDSs may simplify and improve the care process and patient outcomes. CDSs could be valuable when delivering medical care to better match patients’ preferences and biological characteristics. Normally, CDSs automatically provide specific treatment recommendations.

Commercial developers typically promote CDSs to improve clinical decision-making, reduce medication errors and misdiagnoses, provide consistent and reliable information, enhance operational efficiency, increase patient satisfaction, improve quality of care, and lower costs. The literature echoes some of the claims made by these vendors. For example, a systematic review suggests that CDSs reduce unwarranted practice variation, improve healthcare quality, reduce waste in the healthcare system, and decrease the risk of overload and burnout among clinicians [10]. Some devoted efforts to developing a user-friendly, comprehensive, fully integrated web and mobile-based clinical decision support and monitoring system for the screening, diagnosis, treatment, and monitoring of diabetes [11].

2.3.1 Outcomes

Recent studies show positive outcomes in controlling glucose levels for patients with diabetes. A CDS was associated with improving the comprehensive control of blood pressure, LDLc, and HbA1c for diabetics in primary care [12]. Glucose Path, an AI-enabled CDSs for diabetes, effectively reduces the glucose level of patients with poorly controlled diabetes in the Medicaid population. These CDSs facilitate team-based care allowing a cost-effective solution to be produced for patients [13]. GlycASSIST, another diabetes CDS, facilitated treatment intensification and was acceptable to patients with diabetes and general practitioners [14]. A CDS tool on the management of diabetes in small- to medium-sized primary care practices participating in Delaware’s patient-centered medical home project finds the use of CDS is correlated with greater reductions from baseline in hemoglobin A1c and low-density lipoprotein cholesterol, and more patients achieving treatment goals, aiding physicians and staff in better clinical decision-making [15]. EHR CDS was successful in reducing hyperglycemic events among hospitalized patients with dysglycemia and diabetes and inappropriate insulin use in patients with type 1 diabetes [16].

2.3.2 Clinician satisfaction

Additional studies have found clinician satisfaction with CDSs use in treating diabetes and facilitating treatment intensification by the general practitioners [14]. In a cluster-randomized trial, an EHR-linked, web-based CDSs significantly improved glucose and blood pressure control in diabetes patients. The CDS has high use rate and clinician satisfaction. As a result, users are willing to recommend the CDS to others [17]. Furthermore, recent evidence shows that the majority of physicians are satisfied with CDSs [18]. The CDS was feasible and acceptable to GPs [19].

2.3.3 Operational efficiency

CDS for diabetes can help with disease management, and its web-based system CDS provides on-time registration, reports of diabetic prevalence, uncontrolled diabetes, and diabetic complications and reduces the rate of mismanagement of diabetes [20]. In a qualitative evaluation of a standalone CDSs for medication reminders, CDSs were found to improve adherence to evidence-based guidelines and support a more efficient ordering process for providers; providers are satisfied with the CDS for diabetes [21]. CDSs improve healthcare professionals’ adherence to suggested insulin doses and workflow tasks. The decision support system facilitates safe and efficacious inpatient diabetes care by standardizing treatment workflow and providing decision support for basal-bolus insulin dosing [22]. The CDSs integrated with the Epic EHR at the University of Utah enable clinicians and patients to review relevant patient parameters, select treatment goals, and review alternate treatment strategies based on prediction results. The proposed analytical method outperformed previous machine-learning algorithms on prediction accuracy [23].

2.4 Barriers

Despite the benefits documented in the literature, there are barriers to using CDSs. Prior studies suggest time and reimbursement [15], interference with established workflow, unhelpful or irrelevant recommendations, and time pressures [24]. In practice, time constraints, patient overpopulation, and complex guidelines require alternative solutions for real-time patient monitoring. Physician guidelines use rates for diagnosis, treatment, and monitoring of diabetes are very low. To successfully implement a CDS, organizations must conduct adequate validation of programs, evidence and knowledge-based assimilation, users’ feedback, widespread implementation in collaboration with stakeholders, and consistent evaluation of programs’ impact [16]. In order for the CDSs to be effective, the CDS should be conceived as part of a broader, coherent, and department-wide quality improvement strategy, where a clinical quality gap between current patient outcomes or processes and the desired end state has been clearly identified and carefully measured.

3. Global overview of CDSs

This section covers the global development of CDSs; two subsections are created to highlight leading CDSs in industrialized countries and developing countries on technological infrastructure, practice habits, and patient expectations.

3.1 Industrialized countries

In Europe, C3-Cloud is a European Union’s initiative to implement digital health Europe; it is a multinational effort for integrated patient-centered care in the co-occurrence of chronic diseases. C3-Cloud has a group of 12 partners across seven countries in Europe. The care for patients with multiple chronic conditions is complex; it is common that patient data are located across multiple systems and in silos; it is difficult to get a complete, accurate, and reliable view of patients’ medical history. C3-Cloud project aims to build an integrated care platform, so clinicians have better and complete patient information to make clinical decisions; such systems address the increasing demand for improved health outcomes of patients with multiple chronic conditions.

In addition to C3-Cloud addressing multiple chronic conditions, the MOSAIC project in European Union particularly focuses on decision support for diabetes; this project takes a participatory development approach; it applies persuasive design techniques and business modeling to define three phases: (1) user needs, (2) system implementation, and (3) evaluation of the use of CDSs in diabetes management. Qualitative studies using focus groups were used to compile system requirements to gain new insights in the definition of effective Decision Support Systems to deal with the complexity of diabetes care [25].

Several countries (Turkey, Spain, the United Kingdom, Sweden, Finland, and France) collaborated and developed an ICT infrastructure with guidelines to enable personalized care plan management for addressing the needs of patients with multi-morbidity. The team designed 43 logical flowcharts of four disease guidelines (Type 2 Diabetes, Heart Failure, Renal Failure, and Depression) and implemented 181 CDS rules [26].

In Italy, a multidisciplinary research team consisting of doctors, clinicians, and IT engineers develop a fuzzy inference machine to improve the quality of the day-to-day clinical care of type-2 diabetic patients at the Anti-Diabetes Center. This CDS has the function of remote patient monitoring, which includes the ability to monitor a patient regularly from home. This may help to reduce hospitalizations or other acute events [27].

In Belgium, a cluster-randomized trial with before-and-after measurements of a CDS was conducted in Belgian Primary Care Practices over 1 year between May 2017 and May 2018. The majority of physicians were satisfied with the EBMeDS system. Clinicians report many benefits of using CDS, including rapid access to (patient-specific) drug interactions, problems, evidence-based links, etc. Clinicians do not need to perform extensive searching for guidelines. On the disadvantage side, clinicians mention the time required to use the system, the increased alertness by the system, and incorrect reminders. The clinical trial concluded that EBMeDS did not improve diabetes care in Belgian primary care despite the benefits. However, this trial has a significant drop-out rate of 43%. This high drop rate may weaken the conclusion drawn from this study. Further analysis shows the lack of improvement was mainly caused by inadequate software training, EHR data transfer issues, auto coding of lab results, and technical and reporting issues [18].

Another study on CDS for diabetes in Belgium tackles the inappropriate tests as they are a waste of healthcare resources with a pragmatic, cluster-randomized, open-label, controlled clinical trial. This CDS is integrated into a computerized physician order entry (CPOE) to examine the appropriateness and volume of laboratory test ordering and diagnostic errors in primary care. The results show that a CDS within the CPOE improves the appropriateness of lab tests and decreases the volume of laboratory test ordering without increasing diagnostic error [28].

In Saudi Arabia, an evaluation study of EHR integrated CDS reports no significant improvement in chronic disease outcomes [29]. In South Korea, a CDS for Diabetes was developed based on the innovative integration of ontology and fuzzy-ruled reasoning with real data sets. This CDS has an open architecture that is scalable, extensible and increases accuracy in diagnosing diabetes [30].

In Taiwan, a CDS with a focus on antidiabetic medication recommendations was developed based on the guidelines of the American Diabetes Association and the European Association for the study of diabetes. The CDS enables doctors’ clinical diagnosis and decision-making for specialty physicians, nonspecialty doctors, and young doctors with their drug prescriptions. The physician evaluation of the system shows that 87% think the system is useful, and 85% are satisfied with the CDS in their care of diabetes patient [9].

In Australia, a prototype (GlycASSIST) is integrated into an electronic medical record containing evidence-based guidelines. GlycASSIST helps general practice and patients during encounters for setting glycated hemoglobin (HbA1c) targets and intensifying treatment. Interviews and focus groups are conducted with clinicians, including four General Practices, five endocrinologists, three diabetes educators, and six patients with type 2 diabetes. Clinicians and people with diabetes believe that GlycASSIST is useful in individualized treatment intensification. They recommended that GlycASSIST enhances the visual appeal and allows clinicians to overwrite recommendations. In addition, clinicians requested CDS be easily navigated and have greater prescribing guidance [14].

In Turkey, a web and mobile-based application will be developed, which allows the physician to remotely monitor patient data through mobile applications in real time. This system will perform the function of screening, diagnosis, treatment, and monitoring of diabetes diseases. The developed CDS will be tested in two stages: first, the usability, understandability, and adequacy of the application will be determined. Second, a parallel, single-blind, randomized controlled trial will be implemented. Diabetes-diagnosed patients will be recruited for the CDS trial by their primary care physicians [11]. GlycASSIST was able to achieve its purpose of facilitating treatment intensification and was acceptable to people with T2D and GPs. The GlycASSIST prototype is being refined based on these findings to prepare for quantitative evaluation [14].

In Canada, a CDS assists primary care practitioners in applying standardized behavior change strategies and clinical practice guidelines-based recommendations to an individual patient and empowers the patient with the skills and knowledge required to self-manage their diabetes through planned, personalized, and pervasive behavior change strategies. A qualitative study was then conducted to evaluate usability, functionality, usefulness, and acceptance [31].

In summary, CDSs developed in industrialized countries typically incorporate evidence-based practice into the design and development. The commonly followed guidelines are either published by American Diabetes Association or the European Association. Recent findings report a more positive user experience with CDSs, user acceptance, operational efficiency, and clinical outcomes.

3.2 CDSs in developing countries

Developing countries face far greater challenges and barriers than industrialized countries managing chronic diseases. Economic backgrounds, lack of resources, and the absence of some laboratory tests may make clinical guidelines published by international associations not applicable to developing countries. In Sri Lanka, about 11% of its total population has diabetes [32]. A CDS for diabetes was developed through two stages: first, mapping the diabetes-related clinical guidelines using the business process model and notation 2.0 for type 1 and type 2 diabetes and gestational diabetes; second, treatment plans were developed with guidelines using flowcharting. Domain experts were consulted to design and evaluate the ontology. Several real-life diabetic scenarios are used to validate and evaluate the ontology [33].

In Egypt, data mining techniques were used to develop classifiers for the early diagnosis of diabetes. An ensemble algorithm significantly outperforms all other classifiers. Such an effort is essential in building a personalized decision support system, aiding physicians in their daily clinical practice [34].

In Iran, a web-based CDS for diabetes diagnosis and management was developed using ASP.Net MVC server technology, Razor engine, SQL Server database, HTML 5, CSS 3 world standard, and Ajax technology. The diabetes CDS is built following the American Diabetes Association and American Association of Clinical Endocrinologists (AACE) guidelines and physical activity 2017 guidelines recommended by the Netherland. Its interface is user-friendliness and easy to use. The interface displays demographic data, past medical history, laboratory tests, lifestyle, and family history. The web-based system allows for on-time registration, better reporting on diabetic status (uncontrolled diabetes, diabetic complications), and reducing the rate of mismanagement of diabetes. It helps the physicians in managing the patients more effectively [20].

In India, the cost of early diagnosis of diabetes is a barrier for many people to get the laboratory testing done. Various machine learning algorithms are integrated with a CDS to assess diabetes [35].

In summary, developing countries have improved their technological development in patient care. However, evidence-based guidelines are not consistently incorporated into the design of CDSs. Interestingly, developing countries explore data mining and machine learning in an innovative way. Algorithms and predictive models are developed to predict prediabetes and diabetes without any lab tests.

Predictive models could not be 100% accurate. Clinicians and data scientists need to work together to determine the acceptable level for model performance. There will be some false positive or false negative. Data scientists need to work with clinicians to determine the pros and cons of false positive and false negative. In the area of prediabetes, false positive may not produce detrimental effect than false positive. Then the models that produce false positive may be more acceptable than false.

4. Machine learning and artificial intelligence negative

Artificial intelligence (AI) allows computers to describe, understand, learn, reason, and integrate information to solve problems. AI simulates human intelligence so that better, quicker decisions can be made. AI is a fast-growing field utilized by many medical areas, enabling computers to gain human-like intelligence. For example, its applications to diabetes, a global pandemic, can change and improve the approach to diagnosis and management of diabetes. AI is useful in specialized CDSs for detecting diabetic retinopathy [36]. AI revolutionizes remote patient monitoring, continuously monitors the patient’s symptoms and biomarkers, and adjusts to medicine and treatment in real-time, resulting in better clinical outcomes, including glycemic control with reductions in fasting and postprandial glucose levels, glucose excursions, and glycosylated hemoglobin. AI will reform conventional diabetes care by using a targeted data-driven approach and personalized care [37]. However, in regard to user attitudes, a survey study finds that negative perceptions of AI-based CDS tools may reduce staff excitement about AI technology [36]. Thus, it is important to have hands-on experience with AI so that users can gain more realistic expectations about the technology’s capabilities.

Machine learning (ML) is a subset of AI. Machine learning features that machines can learn over time without being explicitly programmed. The ML algorithms include decision trees, random forests, artificial neural networks, genetic algorithms, and support vector machines. The ML algorithms have been used in building predictive risk models for diabetes or its consequent complications. For example, a web-based CDS can predict the early-stage risk of diabetes by classifying results using the patient’s questionnaire without a testing kit. This CDS applies a deep learning approach resulting in better prediction accuracy than supervised machine learning [38]. Another study finds that fuzzy inference machines improve the quality of the day-by-day clinical care of diabetic patients and allow the remote monitoring of patients’ clinical conditions, which helps to reduce hospitalizations [27].

Though AI seems to have unlimited possibilities, there are challenges to the adoption of diabetes AI devices, apps, and systems. Factors such as costs, user acceptance, physician cooperation, and interoperability between systems may affect how an innovation is adopted [39].

5. Future care for diabetes

Medical futurists predict there will be a cure for diabetes. A recent study on stem cells also concludes that beta cell replacement holds a promising cure for diabetes [36]. Biological and medical breakthroughs like the artificial pancreas, and glucose-responsive insulin, provide the correct insulin and the right time to patients. Regarding patient care in diabetes, virtual doctors, big data, data analytics, and social media, all these will become intertwined in the entire patient care ecosystem. Virtual doctors, a proof-of-concept CDS powered by an AI speech recognition system, are able to interact with patients and predict diabetes based on noninvasive sensors and deep neural networks [40]. Wearable technologies enable individualized monitoring of physiological variables in real time. The real-time data collected from multiple devices combined are fed into an artificial intelligence model using adaptive-neuro fuzzy interference to detect prediabetes and diabetes [41].

There is no doubt that digital transformation in healthcare will continue. Big data, machine learning, artificial intelligence, EHR-integrated, web-based, and mobile apps will improve, enhance, and adopt diabetes care. Medical and consumer devices collect a vast amount and variety of data, including continuous glucose monitoring data, insulin pump data, heart rate, hours of sleep, the number of steps walked, movement captured by wristbands or watches, hydration, geolocation, and barometric pressure. Next-generation developments of CDS will leverage big data and prioritize clinical actions based on data analysis, delivering maximum benefits to a given patient at the point of care. In the meantime, innovative care models and delivery methods will emerge. Personalized medication recommendations offered by CDSs fit each patient’s insurance coverage, budget, lifestyle, and medicines. Outcomes can be analyzed constantly and regularly so that adjustments to medicines can be targeted based on the most recent patients’ biological data.

Early diagnosis of diabetes and treatment will reduce the risk of developing comorbidity, delay the development of comorbidity, and improve quality of life for patients. CDSs facilitate doctors in clinical diagnosis and overcome clinical inertia in terms of prescribing habits. In addition, patient-centered care should consider patients’ preferences in care decisions and identify effective methods to communicate CDS information to patients. Doctors need to be more tech-savvy in learning the latest technologies on patient care; patients want more empowerment by participating in self-care and care decision-making. Furthermore, increased number of diabetes journals publish AI-related technologies in diabetes care. Now, doctors must learn new skills and knowledge on AI tools, which have become part of diabetes health care [42].

A path forward may be computerized virtual coaches replacing human counseling; virtual doctors will be able to fully engage in the diagnosis, treatment, and continuous monitoring of chronic diseases. CDSs can be as good or superior to human doctors when prescribing diabetes medicines and may be more effective in overcoming clinical inertia as CDSs can remove human biases and habits. Considering the fact that the physician shortage is growing and 10.5% of the population has diabetes, CDSs play an important role in treating diabetes and more efficiently using clinical resources [43].

6. Conclusion

The CDSs for diabetes have improved significantly over the years. Multiple factors serve as driving forces for the uptake of CDSs. Newer technologies, initiatives, government mandates, and a competitive environment collectively facilitate advancement in diabetes care. This book chapter summarizes global CDSs development in recent years. Our review of the past few years’ publications on CDSs for diabetes shows that the United States is leading the world in technology development and clinical evidence generation. Developing countries around the world are catching up in CDSs development and standards of patient care. The literature has consistently documented evidence of operational efficiency delivered by CDSs (e.g., reduced medical errors and reduced duplicate tests). The current evidence shows that both developing and industrialized countries have put more effort into AI and ML and will use artificial intelligence to their own advantage and innovative ways to develop more sophisticated diabetes CDS tools.

Though studies conducted 5 years prior commonly reported a low adoption rate of CDSs [4], recent publications show an increase in the adoption of CDSs, especially if CDSs are integrated into workflow and EHR. Our recent study of a quality improvement project using Glucose Pathway confirms this trend. In our project, the vendor has been working on integration with EHR. With the increased integration of CDSs with EHR, CDS adoption and utilization will significantly increase. CDS’ true and long-term impact on outcomes, safety, and cost savings can be better measured and validated.

Advancements in technologies will continue to transform patient care, including doctors, processes, and patients. All entities in the patient engagement systems must learn, adapt, and adopt new developments to achieve better self-care, patient care, and clinical decisions. The future is bright but demands more learning on technologies.

Conflict of interest

The authors declare no conflict of interest.