Chronic diseases are the most prevalent and costly health conditions worldwide [1]. Patient self-management combined with timely intervention from care providers are essential to control the progression of chronic diseases [2]. However, within traditional care settings, the disconnected and time-consuming management procedure is unable to meet the long-term care needs of patients [3,4]. Furthermore, several patients with chronic diseases live with more than one chronic condition (ie, multiple chronic conditions [MCC]), which create diverse, and sometimes contradictory, needs for health services [5,6].

Integrated care has been proposed as a means to meet the above challenges by transforming traditional health services [7]. In an integrated care setting, health services are provided by a coordinated multidisciplinary team of care providers. The core objective is to implement patient-centered health systems through comprehensive delivery of quality services across the life course [8]. Further, the advent of information technologies has promoted the delivery of integrated care services, which can be understood from different scales. From an individual scale, information technologies are crucial to facilitate the development of shared care plans [7], which clearly articulates the roles of care providers and patients in the care process to deliver more personalized and targeted care [9]. From a group scale, information technologies play a key role in achieving the goals of bidirectional communication within care provider teams and provision of continuous self-management support to patients [10].

As practical applications of information technologies in the health care domain, telehealth systems have demonstrated potential to improve the outcomes of chronic disease management [11-14]. Patient self-monitoring at home and remote guidance from care providers can be realized with the assistance of telehealth systems [15]. However, most of these systems focus on a single chronic disease, and few studies have attempted to integrate multiple types of chronic diseases into one system [16-19]. A telehealth system designed for managing multiple chronic illnesses simultaneously can not only support the management of patients with MCC but can also reduce the cost of developing multiple telehealth systems for managing different chronic diseases.

In China, the current health service system is in a three-tier form: community health service institutions at the bottom, secondary hospitals in the middle, and tertiary hospitals at the top [20]. General practitioners (GPs) are at the core of primary health care (ie, community level), providing basic treatment and long-term care for patients, especially in the management of chronic diseases [21]. In response to the government policy on promoting integrated care [22], specialists and case managers (CMs) are gradually involved in the management to form a coordinated multidisciplinary team called the “three-manager” mode [23]. The specialists are mainly from secondary or tertiary hospitals, providing more specialized and professional treatment [21]. CMs are mainly composed of nurses who work together with GPs to assist them in their daily work, similar to other countries [24,25].

The “three-manager” mode has been implemented in several provinces of China; however, there remain some practical challenges, which result in a significant gap between standards of care and medical practice [26-28]. First, current management guidelines [29-31] do not clearly specify the responsibility of each role in the “three-manager” mode. Second, the unbalanced allocation of medical resources in China leads to a difference in the abilities of primary care providers [32-34]. GPs and CMs in remote areas may rarely perform comprehensive and effective management following the guidelines.

In this paper, we present the design, development, and retrospective evaluation of a telehealth system that simultaneously supports the management of single or multiple chronic diseases. The proposed system aims to address the existing challenges of the “three-manager” mode in China through a tailored integrated care model, mainly focusing on specifying the responsibility of involved roles and providing a universal care pathway for common chronic diseases. Currently, the system supports three main chronic conditions in China [35]: hypertension (HTN), type 2 diabetes mellitus (T2DM), and chronic obstructive pulmonary disease (COPD).

Figure 1 illustrates the system architecture that consists of two components: (1) a service engine for data storage and decision support, and (2) clients for care providers and patients. Concretely, the service engine is a web service deployed on the cloud server, interacting with clients via several types of application programming interfaces (APIs). The core of the engine is a custom ontology called Universal Care Pathway Ontology (UCPO), which represents the knowledge contained in our pathway-driven integrated care model. Given specific patient data, the engine will generate a small-scale knowledge graph based on UCPO to provide personalized decision support.

The clients are represented in different forms for both care providers and patients. For care providers, the client is in the form of a website that can be accessed on their computers in the hospital or health center. Care providers can utilize the client to monitor patients and perform the intervention. For patients, the client is in the form of a mobile app that can be downloaded to their personal smartphones. Patients can use the client to check their self-management plans, perform self-monitoring, and receive health education. We provided three versions of our app: native apps, including Android and iOS versions, for patients who prefer a better user experience, and a WeChat mini program for patients who are more familiar with WeChat.

The implementation of the website for care providers followed a widely used agile development methodology [44]. We iteratively added functional modules to the platform according to the proposed pathway-driven model. Figure 6 shows an overview of the main functions in the web platform. The primary users of the website are GPs and CMs. Specialists can also log in to check the status of their patients. We provided four major functional modules via tab-based navigation: patient registration, patient warning, patient follow-up, and patient list. The patient registration module is responsible for including new patients in the management, which is mainly performed by CMs. As shown in Figure 6, patients could formally receive the management after providing several types of information, including basic information (eg, demographics, phone number, ID number), disease information (eg, main disease type, associated symptoms), and the corresponding care provider information. For patients who enter the management period, the timing of intervention is determined by the other three major functional modules: (1) the patient warning module displaying all of the untreated warnings of patients’ self-monitoring data, with different types of displayed information for different types of warnings; (2) the patient follow-up module presented as a patient list in order of the next follow-up date; and (3) the patient list module, demonstrating information of patients with different diseases also in the form of a list, mainly focusing on checking compliance and searching for a specific patient.

Care providers could enter the interface of “patient information and intervention” by clicking on the corresponding buttons in the above three major functional modules. In this interface, care providers could check various types of patient information, including demographics, management plans, self-monitoring records, warning history, follow-up records, and assessment records. The platform supports three types of interventions for care providers: complete follow-up via telephone or a clinic visit, short message service text reminders, and message push via the app. Various types of templates and options are provided to simplify and normalize the intervention procedure. Moreover, care providers could edit and push educational materials to patients in another separate interface. The detailed screenshots of the web platform can be found in Multimedia Appendix 3.

We utilized a goal-directed design [45] to develop the mobile app for patients. Patients were engaged in the design process to identify their needs. The concrete design process has been described in our previous study [46]; Figure 7 shows an overview of the main functions in the mobile app. Owing to the distinction between the HTN and T2DM care pathway and the COPD care pathway, we applied different user interface designs to the app for HTN/T2DM and the app for COPD. The two apps share the same underlying framework and provide similar functional modules. As shown in Figure 7, the app includes 5 major functional modules that can be accessed from the main interface: management plan, health checkup, health report, reminder service, and health education. The management plan module is the core function of the app that can be checked directly in the main interface. Currently, the initial self-management plan for patients is generated by the engine based on the management level of patients, and is then manually adjusted by care providers according to patients’ specific conditions. The management plan on the patient app is shown in the form of daily tasks along with control targets. Each task is required to be accomplished at a designated time during each day. Patients could click on the corresponding task and input the required data in a new interface. The submitted data would then be uploaded to the engine for further analysis.

The other four major functional modules were designed for patients with different needs of self-management, aiming to further improve their compliance. Concretely, the health checkup module would analyze patients’ self-monitoring data, and provide immediate and understandable feedback with the aid of the engine; the health report module would summarize the recent completion status of provided tasks and change trends in health data; the reminder service module would set reminders for the execution of daily tasks; and the health education module would display various types of educational materials selected by care providers. The detailed screenshots of the app can be found in Multimedia Appendix 4.

The service engine was developed based on the Spring Boot Framework, an open-source micro service framework for the Java platform. The connected database uses MySQL 8.0, an open-source relational database management system. The clients were developed by a team of experienced programmers collaboratively. All patients, specialists, GPs, and CMs have unique IDs, and their login passwords are encrypted and kept anonymous to the database administrator.

UCPO was constructed using the Protégé 5.5.0 open-source ontology editor in W3C Web Ontology Language (OWL) standard format (second edition). We integrated UCPO into the service engine through the OWL API, a Java API implementation for manipulating OWL ontologies. For the rule-based reasoning, the SWRL Rule Engine Bridge in the SWRL API [47] was used to invoke the execution of SWRL rules through a third-party rule engine. The official implementation currently adopts the Drools rule engine owing to its good execution speed and compatibility with Java programs.

We selected three primary health care facilities that registered in our system at the same time (in the first half of 2019) from different districts of Ningxia Hui Autonomous Region. All three institutions were staffed with one GP and one CM to participate in the management. Moreover, several specialists from the nearest secondary or tertiary hospitals aided with patient diagnosis and recruitment. Figure 12 shows an overview of the selected three facilities. Patients mainly comprised those with HTN, with a small proportion of patients with diabetes. Patients with COPD were not included in these three institutions.

GPs and CMs worked collaboratively to perform pathway-driven management through the system. Table 8 presents the descriptive statistics of the records of interventions performed by GPs and CMs. Further, we calculated the work efficiency of GPs based on their regular follow-up records, as shown in Figure 13. The pie chart demonstrates the percentage of different numbers of follow-up records in a day, whereas the box plot presents the distribution of response days to a follow-up request per month. From the pie charts, all three GPs performed less than 20 follow-ups in over 80% of the follow-up days, and the average number of follow-ups in a day was 16.4, 10, and 5.3, respectively. From the box plots, almost all of the medians of monthly response days were maintained within 5 days, with a couple of outliers in several months. As time progressed, the trends of response days differed for different GPs. Due to the pandemic outbreak of COVID-19 in China in early 2020 [60], for this case study, we only evaluated the intervention records in 2019 to ensure credibility of the results.

In this study, we designed and implemented a coordinated telehealth system that supports the management of multiple chronic diseases based on a tailored integrated care model for the emerging “three-manager” mode in China. The system could provide pathway-driven decision support throughout the management process via an ontology-based approach. According to the retrospective analyses on long-term system usage data, our system was able to link patients’ self-management to care providers’ interventions through semiautomatic decision support following the predefined care pathway. Furthermore, patient outcomes showed a certain degree of improvement after receiving management through the system.

Several interesting aspects can be found from the system evaluation results. First, in terms of patients’ self-monitoring data, the different emphasis of records for patients with different diseases reflects the focus of out-of-hospital management regimens on different chronic conditions. The management regimen of HTN focuses on lowering BP through medication and lifestyle changes simultaneously, whereas the management regimen of T2DM tends to focus first on lifestyle intervention instead of medication. Moreover, the management regimen of COPD requires persistent medication and close attention to acute exacerbation and mental condition of patients.

Second, in terms of comparison tests, the different levels of statistical significance among patient outcomes could be mainly attributed to the difference in cardinality among these three types of records. Compared to patients with diabetes or COPD, patients with HTN constituted the majority of the registered patients. Moreover, the self-measurement of BG at home was slightly more complicated than that for BP due to its invasiveness [61], which would potentially lower patient compliance. The self-monitoring of PEF relied on having a peak flow meter that has not been massively promoted for patients with COPD in China. Furthermore, from a clinical viewpoint, both the SBP and FBG values showed an expected decrease over most time spans, whereas the PEF value only increased over short time periods (30 days and 60 days). We considered that the relatively large SDs for the PEF value and the irreversibility of pulmonary function for patients with COPD might account for the absence of differences.

Third, in terms of causal effect estimation, the estimated ATE for SBP and FBG was consistent with the comparison tests (decreased after intervention), and the results derived by PSS and PSM were similar. The SBP value showed a relatively greater change than the FBG value under causality assumptions. By contrast, the estimated effect of PEF by the two causal inference methods showed a great degree of dissimilarity (a part of refutation tests for PEF also showed sensitivity). A possible explanation for this is that the distribution of outcomes and selected confounders among patients with COPD had a large discrepancy, which led to different intermediate results when performing stratification and matching. Moreover, certain bias might exist in the extraction strategy itself for causal data. Therefore, the true effect of our intervention for patients with COPD remains a question for further investigation. In another prospective study, we found an improvement in COPD-specific quality of life and mental health status of patients after 6-month pathway-driven management, with no significant difference in the mean PEF value [62].

Fourth, according to the regional case study, the system was able to generate different intervention tasks based on patients’ health status, and then assigned them to the corresponding care providers. In this case study, the intervention tasks referred to regular follow-up and abnormal condition intervention performed by GPs, as well as compliance intervention performed by CMs. GPs and CMs were able to work with each other to provide comprehensive management support for patients. In terms of work efficiency, intuitively, the more follow-ups care providers perform in one day and the less handling time a follow-up request costs, the more effective their work will be. However, in practice, considering the workload and work arrangement of care providers, we believe that for one care provider, approximately 10 to 20 follow-ups in a single day and response within one work week to a follow-up request are reasonable, and can guarantee the quality and timeliness of follow-up. According to these criteria, the GP in Hongsibu district did a good job from both aspects, whereas the GP in Xingqing district had a relatively long duration of response to follow-up requests in the last few months of 2019. The GP in Yuanzhou district had a small average number of follow-ups in a single day, which might be attributed to the small cardinality of patients compared with the other two regions.

To better delineate the contribution of this study, we compared our work with prior research from multiple aspects. In terms of model construction, the integrated care model proposed in this study can be considered as an individual-level customization of the well-known chronic care model (CCM) [63-66]. Interventions that incorporated one or more elements of the CCM have shown benefits for primary care outcomes, with large effect sizes for self-management support, delivery system design, and decision support [2]. The core of our model is the management plan combined with pathway-driven coordinated intervention, which adequately represent the above three elements. Further, the system itself is an implementation of the clinical information system in the CCM. In addition, several elements of other models can be found in our system, such as a complete eHealth-based feedback loop between patients and care providers mentioned in the eHealth Enhanced Chronic Care Model [10], effective use of health care personnel mentioned in the Innovative Care for Chronic Conditions model [67], and utilization of remote patient monitoring mentioned in the Transitional Care model [68,69].

We then compared our research with 4 prior studies that explored the comanagement of multiple chronic diseases using information technologies. The results are shown in Table 9. Among these studies, three ([16,17] and our study) utilized ontology to provide decision support abilities during the care process. Four of the studies ([16-18] and ours) supported the management of MCC through different mechanisms. Three studies ([18,19] and ours) designed an individual platform for both care providers and patients, respectively. In terms of evaluation, Riaño et al [16], Lasierra et al [17], and Laleci et al [18] only performed a technical evaluation or pilot application on their solutions, whereas Omboni et al [19] and our study deployed the system in a real-world setting for a relatively long period to test the effectiveness. In addition, due to the limitations of labor and time costs, our system currently only supports three types of chronic conditions and one type of MCC.

Compared with these prior studies, our study was innovative from several aspects. To the best of our knowledge, this study is the first to construct a theoretical model for care delivery under the “three-manager” mode in China. The proposed model utilizes the concept to address the challenges of the limited ability of GPs and CMs in the primary care setting. Through refinement of a universal care pathway and specification on different chronic conditions, care providers from primary health care facilities were able to perform effective management following the practice of evidence-based medicine. Further, our model fully embodies the characteristics of coordination and a “closed loop.” Conclusively, the coordination was mainly reflected in the cooperation of different management roles and inherent associations among different pathway tasks (eg, the frequency of regular follow-up is determined by the results of hierarchical management). The “closed-loop” feature was reflected in the feedback mechanism between patients and care providers, which was implemented via dynamically adjustable management plans with the aid of information technologies.

Based on the constructed model, we implemented a telehealth system that is highly applicable for practical deployment in Chinese rural areas. The system has been carefully designed with comprehensive functions and a user-friendly interface. Care providers and patients can easily grasp the operational methods of the system after brief training. Moreover, we evaluated our system through a multidimensional retrospective study. Long-term observational data from the real world were utilized to investigate the effect of our system from several aspects, including system usability, clinical validity, and quality of care.

Our study has several strengths. First, in terms of system implementation, we provided two forms of mobile apps for patients: the native app and the WeChat mini program. In practical use, we found that compared with the native version, the WeChat mini program did not require installation and was easy to access from WeChat, which is one of the most frequently used apps in China. A majority of enrolled patients (especially elderly patients) tended to use the mini program, which we believe might potentially improve their compliance. Second, benefitting from the design of the universal care pathway, the system can be readily generalizable to other chronic diseases through modification of concrete task contents and definition of the corresponding rule set. The backend service and user interface also need to be updated to complete the full extension of the system. Third, we explored a new approach for evaluating patients’ long-term management effect based on causal inference methods. Although the methods and assumptions adopted in this study were preliminary, we believe that compared with simple comparison tests, the evaluation methods from a causal perspective might be more appropriate for long-term observational data directly extracted from the real world instead of clinical trials. Fourth, to evaluate the quality of care under computer-based management, we proposed a simple assessment method based on the timing and numbers of follow-ups according to our previous study [70]. The proposed method is a type of process measure for care providers from a system usage perspective. The evaluation result was relatively objective and could present a quick understanding of care providers’ work efficiency.

Several potential weaknesses of this study also need to be acknowledged. First, although the ontology-based implementation of the pathway can represent the knowledge in a shareable and elegant way, the adjustment and expansion of ontology need to be completed by knowledge engineers. Care providers encountered some degree of difficulty in understanding the logical rules contained in the ontology. Second, the number of current patients with MCC in our system is relatively small, and we only provide support for one kind of MCC (HTN with T2DM) in the current service engine. The effect of our system on a large scale of patients with diverse MCC requires further exploration. Moreover, several parts of medical knowledge in the guidelines were not integrated in the present management plan, such as proper handling methods for severe acute complications and detailed drug dosage guidance. Third, the long-term compliance remains low in patients. More effective strategies need to be considered to enhance patients’ intrinsic motivation. The long-term work efficiency of care providers also needs to be improved through further medical education. Finally, the evaluation on work efficiency of care providers only considered the regular follow-up task of GPs, and the assessment method did not involve analysis on concrete intervention content.

In future work, we will keep optimizing the usability of the system and support other common chronic conditions such as asthma, stroke, and chronic kidney disease. More elements concerned with health behavior theory (eg, behavior change technologies [71]) could be incorporated into the system to further improve patient compliance. We also plan to deploy our system in more regions of China and perform the evaluation at a larger scale. The evaluation methods will also be refined to provide more comprehensive and credible evidence, such as cost-effectiveness analysis. Another direction for future work is to explore a more personalized care pathway for a specific patient through advanced artificial intelligence technologies such as using reinforcement learning techniques to schedule the follow-up for patients and generate more precise self-management suggestions based on their self-monitoring data.

This study revealed the commonality in the management of different chronic diseases and explored the feasibility of integrating multiple chronic conditions into a single telehealth system. Management models could be customized for specific policy and challenges in different areas to maximize effectiveness. The tailored closed-loop care pathway proved to be feasible and effective under the “three-manager” mode in China. A part of patient outcomes improved after receiving management through the system, whereas the work efficiency of care providers differed individually. Further research might investigate the effect of such systems in a higher evidence level or introduce state-of-the-art machine learning techniques for a more individualized care pathway.