Current Version - Under Revision
Selective retrieval is an essential function for a health record system. There are two main types of requirements:
- Retrieval of selected records from the records of members of a population of patients
for one or more purposes, including the following:
- Aggregations and analysis of data to support:
- Epidemiological monitoring;
- Clinical research;
- Audit of care delivery;
- Service planning.
- Identification of patients with specific risk factors or other characteristics:
- To allow specific preventative, investigative or therapeutic measures to be appropriately focused;
- To allow further selective retrieval and analysis of the records of a subpopulation of patients;
- To enable selection of patients for entry in a clinical trial.
- Aggregations and analysis of data to support:
- Retrieval of selected records from an individual patient record to enable:
- Display of summary views and/or pre-completed template screens containing appropriate selected information.
Example: Active problems/diagnoses, current medication, recent investigation results or blood pressure readings.
- Automating responses to questions posed by a decision support protocol.
Example: To check the record for specified symptoms, findings, investigations, procedures or diagnoses.
The following subsections address issues and requirements that are common to all types of retrieval.
The following sections identify factors that may influence performance when undertaking selective data retrieval. There are no fixed rules for optimization of retrieval performance. Application developers should interpret the issues outlined in the guide in the light of their experience with the operating systems and data management tools that they use.
An evaluation of different approaches to retrieval was undertaken for the NHS, in connection with work on Clinical Terms Version 3 implementation. This showed that the "best" approach was not the same for all relational databases. Some software environments favor one approach while a different approach may be more effective in another environment. Therefore, it is likely that some of the factors discussed will have a significant impact on some developers, while being less relevant in others.
The quality of selective retrieval is measured in terms of two factors:
- Completeness: Retrieval should select all records that meet the selection criteria;
- Specificity: Retrieval should not select any records that do not meet the selection criteria.
The semantic structures of SNOMED CT assist application developers to achieve these goals by allowing different expressions that represent the same or similar information to be recognized and compared (see 11 Supporting Selective Data Retrieval ).
The meaning of a SNOMED CT expression may be modified by the context in which it is used. Aspects of this context are represented by:
- The record structure in which the expression is stored.
- Data directly associated with the expression in the record structure (e.g. dates and times, numeric values and units, the identity and role of the originator of the information or the performer of a procedure).
- Explicit or temporal associations with other information in the same record (e.g. co-existent conditions, likely causes of an abnormal observation, reasons for an investigation or therapeutic intervention, etc).
Storing similar information in differing structures complicates retrieval since each query must take account of alternative ways in which the required information may be stored. As a result the semantic strength of SNOMED CT may be obscured by the varied approach to structure. Therefore, realization of the full potential benefit of SNOMED CT, requires an information model that accommodates SNOMED CT expressions and ensure consistent storage of similar information.
Another limiting factor for retrieval is the consistency and completeness of recording. The extent of use of SNOMED CT depends in part of policy and guidance at national or organizational levels, which in turn depends on requirements and priorities for data retrieval and reuse. From a technical implementation perspective a key factor in delivering consistent retrieval is a user-interface that facilitates, simplifies and encourages consistent data entry which uses SNOMED CT expressions to the extent need to meet relevant requirements.
Retrieval criteria involving record structure
Before addressing the specifics of SNOMED CT related retrieval criteria it is important to recognize that these only form one part of the picture. Most selective retrieval criteria will include a mixture of predicates, some of which apply to SNOMED CT encoded data and some of which apply to other data in the patient record. This non - SNOMED CT encoded data includes:
- Data directly related to coded clinical statements. This includes:
- Dates and times (e.g. time of an event of finding).
- Organizations , people or devices involved in a recorded activity or finding.
- Temporal or causal relationships with other clinical activities or findings.
- Quantitative values associated with SNOMED CT encoded statements.
- Associated status and contextual information.
- Data related to the patient:
- Age and sex;
- Organizations and people responsible for care;
- Occupation, pre-existing disorders or other known risk factors.
The interplay of these factors with SNOMED CT encoded data may affect the optimum approach for data retrieval. Some non - SNOMED CT encoded retrieval criteria may significantly reduce the potential set of patients or in patient record entries that qualify for retrieval. In these cases, it may be useful to apply these criteria before testing SNOMED CT specific criteria.
- A retrieval request for the rubella vaccination status of eight-year-old girls in a family practice with average population distribution requires the review of less than 1% of the population of records.
- A retrieval request for patients who have undergone a particular procedure in the last month only needs to review record entries made in the last month.
- A retrieval request for the most recent investigation results and current medication might be more processed by initially identifying a set of recent records. Checking these records for relevant SNOMED CT values may be more efficient than applying individual queries to the entire record for each of the required items of recent information.
- A retrieval request for people with a rare clinical condition, who also have a relatively common disorder, may be more efficient if the few people with the rare condition are selected first, limiting the scope of the query for the more common condition.
These examples illustrate a general point rather than to offer guidance on the specific searches. It is important to bear in mind that the performance, completeness and specificity of retrieval are dependent on the structure of the record as well as the semantics of SNOMED CT .