In summary, the 417 procedures found in the 200 EPR entries were coded with 36 different KVÅ categories and 148 different SNOMED CT concepts. Of the procedures, 22.8% could not be coded with any KVÅ category and 4.3% could not be coded with any SNOMED CT concept. In SNOMED CT, 206 procedure-concept/category pairs were assessed as a complete match compared to 10 in KVÅ.
Procedures documented by GPs
Referrals to other caregivers were found in 13.3% of the encounters in general practice in Australia during the period 2009-2010, and these included referrals of reproductive health problems and excluded Radiology referrals . In a Nordic study this figure was 8-9% . In our study we found that 16.0% of the encounters had at least one referral to other caregivers, excluding referrals to Radiology departments. We also found that 48 procedures regarding recommendations to the patient, counselling and discussion were found in 200 encounters compared to 29.6 in 100 encounters in 2009-2010  and 25.5 in encounters in 1990 in Australia . Advice and support probably occur as an inherent part of most doctor-patient encounters, as they are an inherent part of general practice .
Description and comparison of content
We had reason to expect that SNOMED CT would cover the GPs' primary care domain well, as one of the terminology systems that was merged into SNOMED CT was originally developed for primary care [16, 17]. This expectation was confirmed, as SNOMED CT had much higher content coverage of procedures documented by GPs in primary care than KVÅ.
Low reliability between coders and coding inaccuracy have previously been documented [11, 33]. However, medium to high reliability, correctness and completeness have also been reported regarding primary care data [3, 34]. Our method was designed to reach a high level of correctness regarding finding all relevant procedures in the record entries, as well as to code the procedures on a high level of detail. As we expected, coding to SNOMED CT with this ambition was both difficult and time-consuming when the optional concepts found were not a complete match. This raised new questions about the procedures identified in the EPR. The coding and assessment tasks meant that analysis of the text had to proceed interactively until a decision was taken. Also, the differences between similar concepts in SNOMED CT could be difficult to understand and needed to be analysed.
We discovered early in the coding process that it was not possible to code the procedures found to KVÅ on a high level of detail, which was not expected. Coding to the general categories in KVÅ did not raise the same questions about the meaning of the procedures in the EPRs and the meaning of categories found. Instead, more time was spent in trying to find possible categories in order to reach a higher level of content coverage.
Description and comparison of the degree of concordance
The procedures coded with KVÅ were assessed as having much lower concordance than the coding to SNOMED CT, which was expected given the size of the terminology systems. Taking into account that 48.1% of the procedures coded with SNOMED CT were considered more specific than the concept found, combining concepts (post-coordination) could have improved the concordance with SNOMED CT. The reason for not using post-coordination more than we did was that the focus of our study was on what was actually done (the procedures). Post-coordination could possibly have been used in this study to specify the reasons for examinations, the body-parts examined, and the drugs prescribed, if such information was wanted. Post-coordination could also be used to indicate that procedures were not performed on the patient. The structure of a terminology system such as SNOMED CT, increased the possibilities for multipurpose data aggregation. The first generation terminology system structure of KVÅ required manual grouping of categories, which is a method that could lead to arbitrary, non-comparable groups and is prone to error . This is a disadvantage of every first generation terminology system such as, for example, ICD-10. The possibility of aggregating data through the SNOMED CT hierarchy means that concepts on the same general level as in KVÅ can also be found in SNOMED CT and used for statistical purposes. This implies that SNOMED CT can be used in primary care for the same purposes as KVÅ. However, certain areas need to be better covered such as information or instructions to relatives.
The number of record entries (200) is limited, and our figures therefore have to be considered as estimates. The record entries were randomly selected from 11 000 000 record entries in the database and therefore constituted a representative sample. The identification of GPs responsible for the record entries had to be done through types of encounters. However, measures were taken to ensure that the record entries used were in fact documented by GPs. The coding performed was secondary coding, based on the text that was written, and it was done by persons other than the GPs responsible for the documentation in the record entries. This may have resulted in minor differences compared to primary coding by GPs.
Implications for health care and research
In a review that included earlier versions of SNOMED over a 40-year period, it was reported that studies of SNOMED in clinical practice were scarce . Most uses of SNOMED CT remain basic, and do not capitalize on the rich semantics of the terminology . However, SNOMED CT was approved as a new information standard in the UK and may be used as the clinical terminology within all electronic patient-level communications within the healthcare environment , thereby further increasing the importance of studies focusing on the clinical use of SNOMED CT.
Coding of procedures can both clarify and reduce information. If coding is performed by health care professionals when documenting in the patient record, the process used by the documenting GP/health care professional in deciding about a procedure code could possibly lead to more exact expressions about what procedures are done or planned for the patient. Coding can be done as secondary coding by personnel other than the GP or other health care professionals. In our study it was difficult when coding to decide between "possible" procedures and procedures decided upon, and reasoning regarding alternative courses of action depending on a future health condition of the patient. The text sometimes included reasoning about treatment options that was difficult to analyse regarding what decisions were finally made. On the other hand, some information reduction cannot be captured by coding; for example, in the context of what procedures should be done under certain conditions.
Record entries written by GPs in primary care contain a broad range of procedures that are often coded with different terminology systems . What is to be coded and what constitutes a meaningful level for recording procedures depends on the purpose of the coding. Primary care in Sweden accounts for 17% of the net costs of health care, and yet there is little knowledge about the procedures performed by different professions. However, coding of procedures is time-consuming, and is an additional administrative task over and above practical care [23, 24]. Collection and secondary use of procedures as coded data should have clear clinical or administrative purposes such as information for other caregivers or professions about planned procedures, as a basis for reimbursement, or for studying certain areas of interest to evaluate procedures related to health problems.
Coding must be supported with suitable tools and integrated in the EPR design with appropriate terminology integration or binding to the information model, which includes support for documenting the dynamic states of procedures in the clinical process. If SNOMED CT is to be used in procedure coding, it is necessary for the end user to have support with the coding. This would include a subset of procedure concepts available for the domain, and/or tools for text reading with semiautomatic primary or secondary coding suggestions. There is also a lack of tools for aggregation and presentation of data based on SNOMED CT coding in order to take advantage of the SNOMED CT concept model. Our experience with the process of coding with post-coordination and reuse of post-coordinated concepts in SNOMED CT is that the procedure is complicated and requires supportive tools if done by end users.
In choosing a PCS, the degree of content coverage is an important factor . Previous attempts to introduce procedure coding in primary care in Sweden have not been successful. The reasons for this remain unknown, but the poor content coverage described in this study needs to be addressed in the future to potentially increase acceptance and usability of PCS. The reasons for using a PCS can vary and can include clinical, statistical or reimbursement purposes. This means that a PCS should have a flexible design and be functional in different settings. It is important for terminology systems to be able to be used in everyday multi-professional environments. Thus procedures documented by different health care professionals in primary care, such as nurses and physiotherapists, need to be explored in relation to KVÅ and SNOMED CT. SNOMED CT could be of benefit in the aggregation and interpretation of epidemiological statistics when analysing data from primary care, as well as in following up clinical data and in quality assurance.