Adopting Business Intelligence Techniques in Healthcare Practice

Abstract

With the rapid development of information technology, digital health technologies have become increasingly prevalent in the field of healthcare. In this study, business intelligence (BI) techniques were combined with research-based prediction models to increase the efficiency and quality of healthcare practices. A data scenario involving 200 older adults with various measurements, including health beliefs, social support, self-efficacy, and disease duration, was used to establish a medication adherence prediction model in a BI system. A regression model, logistic regression model, tree model, and score-based prediction model were used to predict medication adherence among older adults. The developed BI-based prediction model has visualization, real-time feedback, and data updating functionality. These features enhanced the effectiveness of prediction models in clinical practice. Healthcare professionals can incorporate the proposed system into their care practice for health assessments and management, and patients can use the system to manage themselves. The developed BI-based care system can also be used to achieve effective communication and shared decision-making between care managers and patients. Further empirical studies integrating prediction models into the proposed BI system for assessment, management, and decision-making in healthcare practice are warranted.

1. Introduction

Predictive models are key resources for research in healthcare [1]. Predictive models are used by healthcare professionals, such as healthcare researchers, care managers, and nursing staff, to explore the risk factors for diseases and phenomena and to design healthcare interventions. Establishing accurate predictive models for disease prevention, incidence, and prognosis can improve decision-making and health outcomes. Healthcare situations are variable, complex, and uncertain; therefore, making predictions about health outcomes is challenging.

Rapid advancements have been made in information technology. Advances have been made in data communication, deep learning, artificial intelligence (AI), cloud computing, big data, and other machine learning techniques, accelerating the intellectualization of decision-making processes in the medical field [2,3]. Using information technology in healthcare services can improve patient outcomes, enhance communication and collaboration among healthcare professionals, and streamline clinical workflows [4,5]. Information technology can facilitate accurate predictions of disease analyses for classification purposes [6]. With advancements in the field of digital technology transformation, combining healthcare and information technology has become a major topic in the development of healthcare systems and services.

AI is a computer science technology that has exponentially increased computational power, big-data processing technology, and access to large clinical data sets through electronic health records. AI has the potential to transform healthcare services [7]. AI systems can help nurses use real-time information to prevent adverse events, predict treatment outcomes, and provide personalized and evidence-based care [8].

Business intelligence (BI) is a specialized AI-derived application for collecting, analyzing, and organizing data into meaningful and actionable information to support decision-making and organizational performance [9]. BI offers three innovative approaches for nursing professionals to apply research-derived predictive models in a smart manner.

First, rich visualization features enable nursing researchers to explore data from a fresh perspective. In addition, intuitive charting capabilities enable nursing researchers to conduct cross-filtering between charts, representing a major improvement to the static charts that are typically used in nursing research. This functionality enables nursing researchers and practitioners to explore the relationships between charts and to gain novel insights from data.

Second, real-time data synchronization features allow nursing researchers to specify time intervals for data updates and to enable automatic data updates upon detecting changes, ensuring that predictive information provided by BI is accurate and up to date.

Third, BI can be combined with predictive models to yield statistical models, including linear and logistic regression models, mediation models, decision tree models, and score-based models, enabling healthcare providers to move beyond simply reading tables. Healthcare providers can use tablets and electronic devices to perform real-time risk predictions and planning for each patient and communicate with patients in real time.

BI involves the combination of regression analyses and decision trees and the attainment of data-driven decisions and improved overall performance [10]. BI can be used to demonstrate real-time information, to perform data updates, and to satisfy user demands, thereby improving decision outcomes. Several studies have adopted BI to manage the consumption of iodinated contrast agents in computed tomography [11] and to establish telepharmacy activity indicators for the effective management of pharmacy services [12]. Given that healthcare is characterized by large and various health records, BI applications involving user-friendly formats for data management, data analysis, and real-time data updates are considered to be essential in the field of healthcare to support the decision-making processes of healthcare professionals.

Welton [13] argued that BI can be used to establish an information link between individual nurses and patients to identify optimal practices and monitor variations in nursing care. Integrating BI into nursing care and research may aid in the provision of more personalized, efficient, and effective care to patients. Combined BI platforms that focus on healthcare practice and research can be used to establish healthcare application interfaces that provide real-time feedback, visualization, and data updating functionalities. Such an interface can use metadata stored in repositories without requiring data warehouses, predefined data models, or rigid data flows, thereby enabling visualization, a deep understanding of working data sets, and the extraction of valuable information. Analytical tools may be used by individuals without programming skills. Other platforms (e.g., SPSS, Python, and R) typically require the user to have programming skills. BI platforms also offer intuitive web-based interfaces that enable the application of advanced data mining techniques, including data cleansing, filtering, and clustering functions. In addition, BI platforms feature extensible dashboards with customizable charts and reports. Few studies have explored the applications of BI systems, the strengths and weaknesses of BI systems, and the use of BI systems in health assessment and management scenarios. In this study, case examples of medication adherence to antihypertension drugs by older adults were used to demonstrate the application of a BI interface in the field of healthcare. By implementing a BI interface in the management of medications for older adults, healthcare professionals can promptly identify individuals not complying with treatment recommendations, thereby enabling the rapid implementation of appropriate strategies for this vulnerable population.

2. Methods

2.1. Empirical Data Sources

Case examples were derived from an empirical data set and used to demonstrate the functionalities of a BI system. The details of these empirical data are described elsewhere [14]. The data set comprised extensive information on 200 older adults. Data on demographic and disease characteristics, health beliefs, self-efficacy, social support, and medication adherence to antihypertensive drugs were analyzed. Health beliefs, disease duration, self-efficacy, and social support contribute to various pathways of low medication adherence. Therefore, these four variables were used in this study to establish a BI prediction model for medication adherence. Health beliefs were measured using a health belief scale comprising 21 items pertaining to participants’ health perceptions of and attitudes toward hypertensive medication adherence [15]. Each item was rated on a 4-point Likert scale with endpoints ranging from 1 to 4, with higher scores indicating more favorable attitudes toward hypertensive medication adherence. A total score of 63 points or less was considered to indicate unfavorable health beliefs, and a total score greater than 63 points was considered to indicate favorable health beliefs.

Patients’ confidence in hypertensive medication was evaluated using the Chinese version of the 13-item Medication Adherence Self-Efficacy Scale: Short Form. Each item was rated on a 4-point Likert scale, with an average score of 2 or less indicating low self-efficacy, and an average score of 3 or greater indicating high self-efficacy [16].

Social support was measured using the Chinese version of the Social Support Scale, which includes 13 items across four subscales, namely informational support, tangible support, emotional support, and appraisal support. Each item was rated on a 5-point Likert scale, with higher scores indicating greater social support. A total score of 39 or less indicated low social support, meaning that the participants rarely or never received support from their family and friends, and a total score of more than 39 indicated high social support, meaning that the participants received some or adequate support from their family and friends [14].

Adherence to hypertensive medication was evaluated using the Chinese version of the Medication Adherence Scale, which has eight items rated on a scale with endpoints ranging from 1 (never forget) to 0 (definitely forget). A total score of less than 6 indicated low adherence [17].

2.2. Data Analysis

BI systems such as Microsoft Power BI, Tableau, and Qlik Sense offer various services, apps, and connectors that transform disparate data sources into coherent, visually immersive, and interactive insights. Figure 1 depicts the differences between traditional and intelligent statistical workflows. Traditional statistical workflows begin with a centralized database, where all necessary data are stored. This database is connected to various statistical software tools, such as SAS, SPSS, STATA, Python, and R, that are used to access and analyze the data. Analyses are typically conducted from a single perspective (i.e., by an individual or a small team using the same set of tools), which can lead to a narrow analytical perspective. These workflows are usually linear, lacking flexibility and the ability to dynamically update. In addition, their sharing mechanism for analysis results may be limited, making multiangle comprehensive analyses and real-time applications challenging for different departments or personnel and thereby limiting the value of the data.

Unlike traditional statistical workflows, BI systems start with a centralized data platform or data center that securely stores and manages data. These data are accessed through secure gateways to ensure the continuous integration of new information. BI techniques are typically used in advanced and comprehensive data analyses to offer advanced sophisticated analytical capabilities and generate visualization reports. These systems allow different stakeholders from various departments (e.g., clinic, emergency department, intensive care unit or ward, or laboratory) to access and utilize data simultaneously. They also enable the continuous feeding of new data and the combination of various statistical models, thereby allowing for real-time updates and dynamic analyses for prediction and management. BI systems facilitate the establishment of interactive dashboards and visualizations, enabling stakeholders to understand and respond to data, thereby promoting effective communication and information retrieval for real-time management.

In this study, four statistical predictive models, namely a regression model, a logistic regression model, a decision tree model, and a score-based prediction model, were integrated into a BI system to identify risk factors and predict health outcomes for further assessment and management. In nursing research and clinical practice, events can be continuous variables (suitable for linear regression) or categorical variables (suitable for binary logistic regression or multinomial logistic regression). The BI interface proposed in this study allows for the quick selection of an appropriate model based on the type of event and subsequently presents the risk equation. Therefore, a linear regression model was used to examine the risk factors affecting total medication adherence scores, while a logistic regression model was applied to identify the risk factors associated with good or poor medication adherence.

In addition, the timing and value of using decision tree models and score-based models lie in their ability to provide an intuitive and clear standard score range for early warning management systems, rather than simply predicting probabilities or values. Decision tree models are subject-oriented analytical models that can be used to identify combinations of factor interactions that have the greatest effect on clinical events or outcomes and to determine the corresponding population ratio. Nguyen et al. [18] used a decision tree model to explore various risk interactions. The model in their study effectively identified risk groups for sarcopenia in older patients. Decision tree models can elucidate the interactions between risk factors and identify discriminative pathways for at-risk individuals. Score-based prediction models are used to establish simple prediction scores. The probability or risk of an event can be determined using a comparison table that presents various composite scores of outcome probabilities based on the predictors used. A study used a score-based prediction model to predict vitamin D deficiencies in patients with a critical illness in an attempt to mitigate the associated risks of mortality, infection, and other diseases [19]. The main advantage of score-based prediction models is their ability to enhance patient self-assessment and to offer an effective management and monitoring system for healthcare professionals. These approaches allow healthcare professionals to quickly identify high-risk patients based on a risk list sorted by score levels, and promptly implement interventions or improvement strategies. This not only enhances the practical application and value of risk prediction but also improves the characteristics and value of risk prediction.

Therefore, in this study, low medication adherence was used as a key feature in the dashboard of the proposed BI system. Predesignated variables, as mentioned earlier, were imported to this dashboard for data organization and interpretation. This dashboard was used to display the risk estimates of low medication adherence derived from the four statistical predictive models for various management purposes. Overall, the proposed BI system enabled users to immediately add more data for model validation and data monitoring.

3. Results

Figure 2 shows the dashboard of the proposed BI system with the medication adherence prediction model. The left side depicts the frequencies of outcome variables including mean medication adherence and frequencies of high and low medication adherence under the influence of predictors. The middle depicts the frequencies of high and low medication adherence among various predictors. The right side depicts the dynamic effects of various disease durations on medication adherence scores. Case 1 involved favorable health beliefs, high self-efficacy, and high social support, with a mean medication adherence score of 5.26, indicating high medication adherence (Figure 2). Nevertheless, medication adherence scores varied with different disease durations ranging between 0 and 6.57 years, indicating the variation of high medication adherence depending on disease duration. Case 2 involved unfavorable health beliefs, low self-efficacy, and low social support, with the lower-right plot demonstrating mean medication adherence scores close to 0 at short or long disease durations, indicating low medication adherence. Case 3 involved unfavorable health beliefs, low self-efficacy, and high social support, with the lower-right plot demonstrating mean medication adherence scores of ≤1 at short or long disease durations, indicating low medication adherence.

Figure 3A shows the regression model used in the proposed BI system. Regression coefficients were calculated for each predictor of medication adherence. Disease duration was positively correlated with medication adherence. When disease duration increased by 1 year, medication adherence scores increased by 0.22 points (β = 0.22, p = 0.01). By contrast, unfavorable health beliefs (β = −1.69, p < 0.001), low self-efficacy (β = −1.86, p < 0.001), and low social support (β = −1.59, p = 0.001) were associated with reductions in medication adherence scores.

Figure 3B depicts the logistic regression analysis results of the four predictors of medication adherence. These predictors, including favorable health beliefs, high social support, and high self-efficacy, contributed to a high probability of high medication adherence, with an odds ratio (OR) of less than 1. A negative correlation was observed between disease duration and high medication adherence, indicating that those with a shorter disease duration were more likely to have high medication adherence. The corresponding OR of medication adherence varied with changes in disease duration. Overall, the area under the receiver operating characteristic curve was 0.83, indicating a strong discriminatory ability of the four predictors for medication adherence.

Six discriminative pathways were revealed for medication adherence. Health beliefs were the predominant determinant. Participants with unfavorable health beliefs had lower medication adherence scores than those with favorable health beliefs (1.47 vs. 3.10). Among participants with favorable health beliefs, those with a disease duration of less than 10 years and low self-efficacy had the lowest medication adherence scores. Among participants with unfavorable health beliefs, those with a disease duration of less than 10 years and low social support had the lowest medication adherence scores (Figure 3C).

Figure 3D depicts the score-based prediction model for medication adherence explained by the four predictors, referred to as the total literacy score. After the dynamic score and probability of a single or various predictors were examined for individual-oriented calculations, the prediction score and corresponding probability of high medication adherence were calculated using various combinations of these predictors. For example, for cases with favorable health beliefs, high self-efficacy, and high social support, the total literacy score was 88.24, indicating a greater than 70% likelihood of high medication adherence.

4. Discussion

Overall, this study offers an innovative approach for combining a BI system with research-based prediction models for research and clinical application purposes in healthcare. In the developed BI model, four statistical models are used for data management and analysis. The BI model enables researchers and participants to evaluate self-perceived health and to manage health outcomes. The proposed system facilitates model updates by enabling the integration of new data into its models, which in turn enhances model validity for real-world scenarios. Given the recent advancements in the fields of big data and AI, integrating a BI model into a healthcare system has the potential to enhance healthcare and research, leading to improved health outcomes.

Integrating the proposed BI system into healthcare practice has several advantages. First, the proposed BI system incorporates four statistical methods, including variable-centered approaches (i.e., regression models) and patient-centered approaches (i.e., classification and regression tree and score-based models), to collectively visualize care models. In these models, the outcomes of data frequencies and various combinations of data analyses can be visually represented to facilitate further interpretations. Using technological products such as software equipped with information communication capabilities can enhance the self-management of cardiovascular disease, osteoporosis, and diabetes mellitus, ameliorating the symptoms associated with these conditions [20,21,22]. Typically, the outcomes of regression models and classification and regression tree models can help care managers conduct risk factor assessments, identify high-risk groups, and design appropriate strategies for health management. In addition, the data generated by score-based prediction models can aid in self-management. The developed BI model can help healthcare professionals make effective shared decisions with their clients, improving health outcomes. The BI system is easy to interpret. The system facilitates improved communication, and its visualization feature can help care managers, professionals, and researchers develop customized care plans for patients in accordance with individual requirements.

Second, using real-time feedback in the proposed BI system facilitates the integration of research models into nursing practice. Implementing model-based data to practically interpret nursing research may contribute to overfitting, resulting in a disparity between research findings and their translation into clinical practice [23]. In overfitted models, inadequate conclusions foster improper clinical decision-making, leading to potentially negative outcomes [24]. Integrating machine learning algorithms with dynamic data into the proposed BI system facilitates the extraction of valuable insights and enables health assessments and management [25]. Therefore, implementing real-time feedback in BI systems in healthcare settings can increase the efficiency and quality of nursing care by offering timely and actionable insights for informed decision-making.

Third, the ability to update data in the BI model can increase the accuracy of the included prediction models. Typically, the validation and practical application of prediction models are limited to a single data analysis [26]. In the field of healthcare, the proposed BI system, with its continual data input capability, can enhance the effectiveness of prediction models. This system can be used as an AI-based expert system by clinical physicians to make more informed decisions in treatment and progression predictions. This AI-based expert system can also help care staff manage and analyze large amounts of daily care-related data, ultimately leading to improved health outcomes.

Overall, the proposed BI system is an open-source, cloud-based service. Data security is a key limitation of cloud-based services. Data storage and analysis require access to BI systems with cloud services. Patient privacy derived from medical records is regarded as a major ethical concern in medicine. Therefore, an advanced system with high-grade security is required. To address the data security problems associated with the proposed BI system, additional system designs with advanced database structures, such as data warehouses that offer robust security features and efficient data management, are required. A data warehouse is a centralized repository designed to consolidate large volumes of structured and unstructured data from various sources, optimizing it for query and analysis to facilitate reporting, decision-making, and the creation of analytical reports [27,28]. Thus, data warehouses serve as the backbone of BI systems by consolidating and securing large volumes of data from various sources, enabling comprehensive analysis, supporting advanced analytics, ensuring data governance, and facilitating scalable, reliable access to information for strategic decision-making.

5. Conclusions

In this study, an innovative approach involving a combination of a BI system and research-based predictive models was proposed. The system offers visualization, timely feedback, and data updating functionalities, and its effectiveness in clinical practice was demonstrated. Data analysis and mining can be performed on big data pertaining to institutional environments and nursing care. The system can improve the data management capabilities of healthcare professionals. Given the subscription and shareable nature of the proposed BI system, data visualization can enhance patients’ understanding of their health conditions and facilitate effective communication with healthcare professionals, thereby fostering shared decision-making and facilitating health management. Since multilevel models and machine learning can handle complex data structures, are used for predictive analysis, provide flexibility in model specification, and generate data-driven insights applicable across various domains [29,30], further empirical studies involving the implementation of machine learning or advanced statistical models, such as multilevel models, to the proposed BI system for assessment, management, and decision-making in healthcare practice are warranted.