The 8 most common clinical documents in China's general hospitals are: 1) the outpatient medical record summary; 2) the emergency medical record summary; 3) the inpatient medical record summary; 4) the basic medical synopsis (a brief summary of medical activities concerning the evolvement of illness, including examining, diagnosing, and treating); 5) the inpatient outline (summary information during a hospital stay, which is usually as the first page of a paper-based medical record after discharge); 6) the discharge summary; 7) the referral summary; and 8) the labor and deliver record summary.

The Chinese National Health Data Dictionary (CNHDD) is a metadata repository that must comply with the standards for the construction of databases and health information systems. The metadata in the CNHDD was generalized and abstracted from various health information systems and legacy systems, with each metadata describing attributes of data identification, definitions, collection, usage guides, references and administration. At present, more than 1,500 metadata are available, and these metadata can be browsed by visiting the website described in [14]. In this research, we acquired standardized contents of each data item in the DGs by matching each data item with the metadata in the CNHDD.

1) Step 1: Development of the DGs' architecture and contents First, 1,487 original clinical record sheets from 14 representative general hospitals, including 4 hospitals with more than 2,000 beds, 6 hospitals with 1,000-2,000 beds and 4 hospitals with 500-1,000 beds across the country were collected. After merging the original sheets and removing redundant elements, 145 clinical record sheets were formed [15]. This dramatic reduction in the number of elements resulted from similar clinical procedures in most hospitals. Second, the framework of Diagnostics [16] knowledge and the approaches of the HL7 CCD, the IHE PCC and the HITSP for assembling templates and modules were considered together to propose the DGs' architecture. Third, the proposed DGs were used to construct 145 clinical record sheets as a pilot study to test their integrality. If the DGs could not completely build these sheets, the DGs were returned to a redefinition process. Lastly, data items within the DGs were identified by combining data items from the original sheets with the DG's architecture. Data items from the sheets were categorized and arranged in their related DGs, and data items having similar properties were abstracted. For example, data items B-mode ultrasonography examination ID, X-ray examination ID, CT examination ID and other examinations IDs were abstracted to two data items of examination type and examination ID. The B-mode ultrasonography, X-ray, CT, and other examinations became the codes in the value set for examination type after the abstraction.

2) Step 2: Definition of the structure, content and semantics of each DG

In this study, the HL7 CDA was chosen as our standard to represent the semantics of DGs and clinical documents for two reasons: 1) the HL7 CDA is a document markup standard that specifies the structure and semantics of a clinical document for the purpose of exchange [13], which suits our needs, and 2) the HL7 CDA has been chosen as the data exchange standard by MOH in the Technology Solution of Establishing Hospital Information Platform for Electronic Health Record (EHR) in China; thus, our standards should comply with the MOH standards [17].

By mapping each data item of a DG to the corresponding data element in the HL7 CDA, the structure of the DG was acquired. By matching each data items of the DG with the metadata in the CNHDD, a standardized description of the DG's items were obtained. Based on both above results, the DGs' semantics were defined. All the data items in the DGs have corresponding data elements in the CDA, and 90% of the data items were standardized directly by matching them with the CNHDD.

3) Step 3: Construction of each clinical document with the DGs If one or more data item in each clinical document was found in a DG, they were replaced by the DG. Thus, the contents of the clinical document were changed from being comprised of data items to being comprised of DGs, upon which the structure and semantics of the clinical document were finally produced.

During the formulation process, 4 discussion meetings were held to discuss the integrity and rationality of the developed DGs and the accuracy and significance of the clinical documents structured by the DGs. Altogether, 25 people participated in the consultations, including MOH leaders, health information experts, senior physicians, surgeons and software development engineers. The formulation did not proceed to the next step unless results of the current step were approved by 95% of those consulted.

The formulation process of the DGs and clinical documents are shown in Figure 1.

Altogether, 5 header DGs and 65 body DGs, including 48 section DGs and 17 entry DGs, were proposed. The section DGs consisted of 17 section DGs and 31 sub-section DGs. Of the section DGs, Health Histories, Diagnosis, Procedure and Intervention, Medications, Assessment, Process of Clinical Care and Health Guidance all contained sub-section DGs (Figure 2).

Each DG conveys specific information. A header DG conveys identification information for documents, patients and involved providers. A body DG comprised of relevant section DGs conveys clinical report information. A section DG contains a single narrative block and possible (zero or more) entry DGs representing narrative content by structured data items (Figure 2). Thus far, narrative blocks of most of the section DGs can be represented by entry DGs, except for the Referral, Medical Equipment Use, System Review, Marital History, Menstrual History, Childbearing History and Progress Note narrative blocks. More entry DGs will be developed to represent these unstructured section DGs in future studies.

One or more data item in each clinical document can be mapped to corresponding data items in the DG and then replaced with that DG. For example, the data item provider's hospital name and provider's department name were replaced with Healthcare Providers (EHR.HRD.04). Type of laboratory test, name of laboratory test, value of laboratory test and measurement unit were replaced with the section DG Laboratory Test (EHR.SEC.06). Finally 8 clinical documents were all structured by a number of DGs (Table 2), based on which standardized structures and semantics of the documents for semantic interoperability were produced in the HL7 XML format. For example, Figure 4 shows the detailed structure and semantics of outpatient medical record summary document structured by 3 header DGs and 12 section DGs in XML Schema.

The methodology of building shareable clinical documents using the HL7 CDA is a recognized solution [21,22], yet we do not use it completely because the attributes of the data elements and codes of the value sets in the HL7 do not completely suit our needs. Therefore, we customized the HL7 CDA in two ways in our research. According to our business needs and on the condition that the architecture of the HL7 CDA remains unchanged, one way to customize the HL7 CDA was to adjust the attributes (e.g., cardinality, data type) of data items in the DGs, and the other method was to redefine codes of parts of value sets. Therefore, a contribution of our research is to promote the use of the HL7 standards in China.

Based on DGs, clinical documents have certain characteristics. First, clinical documents constructed by DGs will be structured, enabling embedded information to be more complete and accurate [23,24]. Second, more than just these eight clinical documents can be built by flexibly reusing the DGs. The architecture of the DGs can stay stable merely by adding codes to value sets and adjusting the data items' attributes in the DGs when more documents need to be built. Third, DGs and data items that are irrelevant to clinical documents will be excluded by defining their attributes of optionality and cardinality, which can keep clinical documents clear and concise.

Almost all the sections of the CCD, the PCC and the HITSP can be matched to corresponding section DGs except for two: the section describing medical care expenses and the section representing hospital management information. The CCD, the PCC and the HITSP use Payers section to specify organizations or individuals who may pay for a patient's healthcare, while we use Medical Expense section to describe actual expenditures that have been paid. We use Administrative Use section to describe the information used for hospital management (e.g., cure rate, death rate), whereas this section is absent in the sections of the CCD, the PCC, and the HITSP. Furthermore, entries also differ as a result of the differing sections. In conclusion, these differences come from business variations among different cultural and language backgrounds.