Yidu Cloud CDSS: Empowering Physicians with Four Core Capabilities for a Seamless Clinical Decision Support Experience

The development and exploration of Clinical Decision Support Systems (CDSS) in China’s healthcare market have spanned approximately two decades, gradually evolving into a variety of CDSS products tailored to different clinical scenarios and designed to serve physicians at various levels.

Currently, there are approximately two categories of Clinical Decision Support Systems (CDSS) in the Chinese market: one is knowledge base-based query systems, and the other is knowledge rule-based recommendation and audit systems.

From the perspective of usage scenarios, knowledge base-based query Clinical Decision Support Systems (CDSS) are often employed to address fragmented situations where clinicians need to retrieve information when encountering unfamiliar clinical issues. Their drawback lies in the lack of deep integration with hospital information systems, which fails to significantly shorten the pathway for clinicians to resolve problems. Consequently, they serve merely as a supplement for addressing specific issues in clinical settings, having a limited impact on improving healthcare efficiency and quality.

Review-oriented CDSS is typically integrated with production systems to audit and recommend clinical actions, thereby improving healthcare quality, safety, and efficiency. However, its limitations are evident: it essentially relies on finite data rules to address the infinite variability of individualized clinical scenarios, which may yield minimal returns despite substantial investment.

Therefore, during clinical practice, physicians frequently encounter the following realities: ① there is a significant discrepancy between the machine-generated recommended review results and the physician’s clinical judgment; ② the system’s dimensional understanding of clinical issues and its reasoning logic are overly simplistic, failing to provide substantive assistance.

On April 28, 2018, the General Office of the State Council formally stated in the “Opinions of the General Office of the State Council on Promoting the Development of ‘Internet + Healthcare’ (Guo Ban Fa [2018] No. 26)” that clinical decision support systems based on artificial intelligence should be developed. This signifies that existing CDSS solutions no longer meet the current development needs of hospitals.

In the era of big data, what should a Clinical Decision Support System (CDSS) look like? Based on policy guidelines and the practical needs of hospitals, VCBeat’s research suggests that it should possess at least the following four characteristics:

1. A highly structured, computable, and authoritative knowledge base;

2. By integrating real-world data with artificial intelligence technologies, recommendations become more intelligent and precise;

3. Can be embedded into clinical systems to intelligently extract patient disease characteristics and provide recommendations, facilitating use by physicians;

4. Capable of evidence-based traceability, with recommended results linked to supporting evidence such as clinical guidelines, literature, and similar case records.

To achieve the four objectives outlined above, Clinical Decision Support Systems (CDSS) must leverage big data technologies to integrate vast amounts of medical literature and electronic health record data, thereby constructing medical knowledge graphs capable of computational reasoning. However, given the low level of structuring in raw medical data within Chinese hospitals, severe non-standardization of data, and the lack of integration among various types of healthcare information systems, the training of CDSS faces a series of challenges. Consequently, the realization of truly intelligent CDSS remains a long and arduous endeavor.

Abroad, CDSS is garnering increasing attention from major healthcare enterprises.

In January 2018, GE Healthcare and Roche Diagnostics announced a long-term strategic partnership to jointly develop and promote digital clinical decision support solutions, aiming to make the diagnostic process faster and more accurate.

In January 2019, EBSCO Health, a clinical information resources company, announced the acquisition of Health Decision. This acquisition aims to develop tools that assist clinicians and patients in making more rational clinical diagnosis and treatment decisions. Through this transaction, EBSCO Health will gain access to Health Decision’s relevant technological resources.

In February 2019, at the HIMSS Global Conference & Exhibition 2019, Royal Philips launched its IntelliSpace suite of healthcare solutions, including IntelliSpace Epidemiology and version 4.1 of IntelliSpace Cardiovascular, aiming to address infection challenges through medical technology and improve the treatment of pediatric cardiovascular and cerebrovascular diseases.

In China, the interest in Clinical Decision Support Systems (CDSS) is equally high. As a representative enterprise in China’s medical artificial intelligence and big data sector, Yidu Cloud has recently launched a new generation of intelligent clinical decision support systems capable of multi-dimensional evidence-based analysis.

In terms of data authority, Yidu Cloud’s CDSS knowledge base covers thousands of diseases, 32 medical specialties, thousands of clinical guidelines, and a vast amount of medical literature. The knowledge base is built upon authoritative guideline consensuses, clinical pathways, standard textbooks, and pharmacopoeias, and is continuously updated with the latest domestic and international guideline consensuses, high-quality medical literature, drug package inserts, and other data sources.

Meanwhile, Yidu Cloud’s CDSS has collaborated extensively with specialists across various disciplines in China to refine and optimize its knowledge base for dozens of clinically complex diseases, thereby developing specialty-specific knowledge bases that more closely align with clinical practice and cover diagnosis, treatment, and prevention.

Yidu Cloud’s disease knowledge graph is jointly constructed based on evidence-based evidence and real-world data. It uses evidence-based knowledge to build the nodes and relationships of the knowledge graph, while real-world data are used to train the strength of these relationships. The resulting knowledge graph combines both accuracy and computability.

For knowledge discoveries extracted from medical records that are not defined in evidence-based knowledge, Yidu Cloud will have specialized clinical medical experts review them. Those approved can be incorporated into the knowledge graph and applied to the CDSS system. Some of the valuable medical findings will undergo more rigorous scientific experiments for validation and publication in collaboration with experts, further enriching evidence-based knowledge.

Clinical Decision Support Systems (CDSS) are primarily designed to provide appropriate recommendations, audits, and risk alerts across various stages of physician diagnosis and treatment. If implemented as a standalone system, CDSS would be inconvenient for physicians to use, resulting in low adoption rates. To ensure a positive user experience for physicians, Yidu Cloud integrates CDSS directly into the physician workstation rather than deploying it as an independent system.

After the CDSS is embedded into the physician workstation, the system automatically performs background calculations. It provides automated recommendations when physicians need to make diagnostic and treatment decisions, monitors and alerts on patient abnormalities and disease risks, and automatically audits decisions as they are made. Additionally, it supports writing recommendation results back into the business system. This approach not only serves as a real-time alert and warning mechanism but also enhances physicians' work efficiency.

To avoid excessive CDSS alerts that could interfere with physicians’ normal workflow, Yidu Cloud has improved its product from two aspects.

On the one hand, in terms of recommendation principles, the system incorporates alerts and notifications based on statistical analysis of historical in-hospital diagnosis and treatment behaviors, enabling it to automatically learn physicians’ clinical practice patterns. The Clinical Decision Support System (CDSS) triggers alerts only for actions that deviate from those frequently observed in historical medical records. This approach helps avoid the frequent false positives and erroneous alerts that often result from discrepancies between a pure knowledge-base-driven reminder system and actual hospital practices.

On the other hand, in terms of product experience, the CDSS interface can be set to collapsed by default, and hospitals are allowed to configure alert rules and severity levels, thereby ensuring that the product does not interfere with physicians’ normal workflow.

In terms of product categories, Yidu Cloud's CDSS is mainly divided into two types: one is general practice CDSS, and the other is single-disease CDSS.

Yidu Cloud’s General Practice CDSS is positioned to provide “broad yet shallow” decision support, suitable for physicians across the entire hospital. Its core functionalities include knowledge base queries, suspected diagnosis assistance, symptom alerts, order recommendations, interpretation of tests and examinations, and intelligent auditing. It offers knowledge graph-based recommendations and alerts during diagnosis and treatment, while also enabling physicians to proactively search and query the knowledge base and medical literature.

In addition to recommendations from authoritative knowledge bases, the General Practice Clinical Decision Support System (CDSS) leverages a big data platform to process and learn from massive volumes of electronic medical records, training quantitative relationships among various disease entities. This enhances the accuracy of recommendations while better aligning with physicians’ diagnostic and treatment practices.

Yidu Cloud’s single-disease CDSS is positioned to provide “specialized and in-depth” decision support for specialists. At different stages of disease treatment (such as preoperative chemotherapy, surgery, postoperative chemotherapy, and follow-up for cancer), it recommends the most suitable regimen based on the patient’s specific condition, standardizing disease treatment while keeping physicians informed of the latest advancements in the field.

The core logic of the decision engine for single-disease CDSS includes several components:

1. Deconstruct the latest clinical guidelines or top-tier expert consensus to establish foundational treatment recommendations; 2. Evidence-based correlation and scoring. Assign scores to different regimens based on evidence such as the latest clinical guidelines, literature, clinical trials, drug contraindications, and adverse reactions, tailored to the patient’s disease characteristics; 3. Real-world evidence correlation and scoring. Analyze the efficacy of different treatment regimens in similar patients from historical medical records to predict prognosis and the incidence of adverse events, thereby scoring each regimen. Ultimately, the system provides a personalized treatment plan best suited for the patient, supported by “multi-dimensional” evidence-based analysis.

Since original medical record data are distributed across different clinical systems (such as HIS, EMR, and PACS) and exist largely as lengthy text entries, it is particularly critical to ensure that Clinical Decision Support Systems (CDSS) access standardized data. Yidu Cloud’s big data platform, deployed within hospitals, integrates multi-source heterogeneous hospital systems during its foundational data processing stage, thereby creating a comprehensive 360-degree view of data at both the patient level and the encounter level.

Furthermore, based on international standards (such as ICD-9, ICD-10, ATC, and LOINC), Yidu Cloud has developed its own terminology system. By leveraging natural language processing technology to perform structured extraction and normalization from extensive text descriptions, the construction of Yidu Cloud’s disease knowledge graph and clinical decision support systems becomes a seamless and logical progression.

Compared to disease knowledge graphs constructed solely from knowledge bases, those built by integrating massive volumes of electronic medical record data can quantify the relationships among various medical entities, including patient populations, symptoms, test results, diseases, and medications. This makes them more suitable for reasoning and decision-making based on statistical machine learning. In single-disease Clinical Decision Support Systems (CDSS), machine learning models trained on data can better characterize the relationships between disease features, feature combinations, and decision targets, thereby achieving higher accuracy than traditional methods.

At hospitals that have deployed the Yidu Cloud DPAP big data platform, the Clinical Decision Support System (CDSS) not only provides recommendations linked to evidence-based medicine and detailed views but also suggests similar medical records of patients with conditions comparable to the current patient’s. It supports navigation to the big data platform for viewing detailed medical records, Patient 360 (Note: Patient Panoramic View), and other features, thereby offering reference support for junior physicians.

In addition, Yidu Cloud's CDSS can also enable personalized utilization of medical data.

For example, the incidence of diseases exhibits regional characteristics; due to factors such as local climate or dietary habits, the probability of a specific disease occurring in a certain region may be significantly higher than in others. Furthermore, owing to differences in etiology, patient demographics, and institutional capabilities, medical orders (including examinations, laboratory tests, and medication) issued for the same diagnosis may vary across different hospitals.

CDSS based on general knowledge bases fails to account for personalized factors such as regional characteristics, often resulting in recommendations for medications not available at the hospital or diagnostic tests with names that do not match the hospital’s nomenclature. In contrast, Yidu Cloud’s CDSS, integrated with real-world electronic medical record data, effectively learns regional disease patterns and the hospital’s individualized diagnosis and treatment practices.

As demonstrated by the case of Yidu Cloud’s CDSS, a new-generation Clinical Decision Support System (CDSS) that meets hospital requirements must not only possess a computable, authoritative knowledge base but also leverage medical big data and artificial intelligence technologies to deepen disease knowledge graphs and decision models. It should feature more intelligent and user-friendly operational logic to better align with clinical practice, while enabling evidence-based traceability through one-stop integration of multidimensional evidence supporting its recommendations. By organically combining “knowledge,” “data,” and “technology,” it forms a new generation of intelligent clinical decision support systems.

It is foreseeable that next-generation Clinical Decision Support Systems (CDSS), such as those developed by Yidu Cloud, will become powerful assistants for physicians in clinical practice and learning, helping to improve the efficiency and quality of diagnosis and treatment, and ultimately enabling patients to receive higher-quality medical services.