Because Electronic Laboratory Reporting (ELR) has been seen to be more accurate, timely, and cost-effective than reporting by other conventional means (e.g., mail, fax, etc.), ELR adoption has been systematically promoted as a public health priority [1-7]. A major deterrent to laboratory data being used in public health research and biosurveillance programs seems to be the lack of interoperability of automated laboratory information management systems [8-12]. Standardized, universal coding that records laboratory test order and result information in a structured and systematic fashion is an essential component of interoperable ELR systems [9,13,14]. The use of local codes or terminology and unstructured text fields to describe laboratory test orders and results varies widely among laboratories [15,16]. Several coding strategies are available to make electronic laboratory data more computational and thus readily interchangeable electronically. Long established coding systems such as International Classification of Diseases (ICD) and Current Procedural Terminology (CPT) are used for insurance reimbursement and other administrative purposes. ICD codes are primarily designed for classifying diseases and other health conditions while CPT is designed to record a medical service or a procedure. Therefore, coding laboratory information is out of scope for ICD and CPT coding systems. Several coding strategies are available to make electronic laboratory data more computational and thus readily interchangeable electronically; Logical Observation Identifiers Names and Codes (LOINC) and Systemized Nomenclature of Medicine Clinical Terms (SNOMED CT) are the two most comprehensive coding systems representing lab test type and result information.

Therefore, these two information coding systems were specifically recommended for use in coding laboratory information in electronic health records by the U.S. Department of Health and Human Services Office of National Coordinator for Health Information Technology

Established in 1994, LOINC is a coding system designed to identify medical laboratory observations and procedures. LOINC covers both laboratory tests and clinical observations enabling coding of test orders and test results. The system has been endorsed by the American Clinical Laboratory Association and the College of American Pathologists. A LOINC code is composed of six attributes that characterize the details of laboratory test orders and results, namely: component (e.g., Escherichia coli, potassium), property (e.g., arbitrary concentration, mass concentration), timing (e.g., point in time, over a span of time), system (e.g., stool, blood), scale (e.g., ordinal, nominal), and method (e.g., culture, microscopy). The most recent version of the LOINC database contains more than 70,000 codes [17-22]. LOINC codes are updated twice a year to reflect changes in diagnostic practices over time.

Established in 1965, SNOMED CT is a coding system designed to identify anatomic and clinical pathology information and laboratory results. The system is a collection of medical terms, codes, findings and procedures. A SNOMED CT code is composed of the following three attributes; unique concepts (e.g., E. coli), concept descriptions (e.g., is a), and relationships between concepts (e.g., bacteria present in stool). The system is updated twice a year and currently includes over 300,000 unique concepts [23]. SNOMED CT is the ontological basis of the upcoming International Classification of Diseases 11^th Edition (ICD-11) revision spearheaded by the World Health Organization [24,25].

We examined the use of LOINC and SNOMED CT codes for coding laboratory test orders and results in laboratory reports transmitted to BioSense program from 63 non-federal hospitals in the calendar year 2011. In this report, we present the first national level description of the use of LOINC and SNOMED CT codes from biosurveillance data to characterize laboratory test orders and results reported by hospital-based laboratories.

BioSense is a syndromic surveillance system supported by the U.S. Centers for Disease Control and Prevention (CDC), and it receives daily medical encounter data from participating hospitals. During 2011, 11% of participating non-federal hospitals also shared laboratory data with BioSense [26,27]. Details of the data transmission from hospitals to CDC’s BioSense program have been previously described [28-30]. In brief, data reporting standards follow guidance from the Public Health Information Network Messaging System’s (PHINMS) syndromic surveillance message guide and are transmitted securely via digital certificates and data encryption [31]. The laboratory reports included in this analysis were transmitted using version 2.3.1 of Health Level 7 (HL7) message formats. All laboratory reports from non-federal hospitals reporting to BioSense from January 1, 2011 through December 31, 2011 were extracted from HL7 messages and converted to SAS analytic data files. Reports included laboratory test orders only, test orders with results, and results only.

Data elements in each laboratory message can be broadly categorized into three sections: 1) administrative data elements, 2) laboratory test order related data elements, and 3) laboratory result related data elements. Administrative data elements include date of visit, type of service facility, testing laboratory ID, and date of HL7 message creation. Laboratory test order data elements include order number, order test name, order test codes (local and “standard”), order test coding system, and segments of coding structure. Laboratory result data elements are grouped into two categories: observation identifier (OBR) and observation value (OBX). Elements of OBR include observation result code, observation result coding system, local observation result code, observation result text (local and “standard”), and observation result type. Elements of OBX include result coding system, result test code (local and “standard”), result status, result notes, result test date, result test name (local and “standard”), result unit, test interpretation, and details of the sample to be tested (component, property, timing, system, scale, method). Additionally, BioSense laboratory message data include site and type of the specimen, diagnostic criteria, and test sequence number.

Each laboratory test result with result code was categorized into one of six result status categories, namely: “final result”, “preliminary result”, “specimen in the laboratory”, “correction”, “deletes OBX record”, and “result can’t be obtained for this observation”. Since results identified as “preliminary” constituted almost one-third of all laboratory results reported we analyzed all reported results regardless of their status. Later we compared our analysis with findings from a separate review limited to “final” results.

We analyzed the frequency distribution of laboratory test orders and results to determine characteristics of the test orders and results reported, and to examine the 25 most common test orders and results. Unique LOINC or SNOMED CT codes used in the reports were identified, along with the percentage of reports that were missing a standardized code. For each of the SNOMED CT codes reported without corresponding LOINC codes, we used the Public Health Information Network Vocabulary Access and Distribution System (https://phinvads.cdc.gov) web sites to obtain the SNOMED CT concept or descriptive text. Then each concept was searched in Regenestrief LOINC Mapping Assistant (RELMA®) software version 5.8, to determine if there was a LOINC code that might corresponded to the SNOMED CT concept. Since the value “OTH” in HL7 messages does not clearly indicate a specific laboratory test order or results, rather it indicates concepts not represented by the code system; we treated reports coded with “OTH” as non-informative data. This analysis was determined to be a public health surveillance activity rather than human subject research requiring institutional board review.

Out of 569 non-federal hospitals reporting data to BioSense in 2011, 63 (11.1%) hospitals in 14 states submitted 14,028,774 laboratory reports (test orders and/ or results) from 821,108 unique laboratory visits. On average, 38,000 laboratory reports were reported daily. In the reports with at least one result code, the hospital diagnostic service ordering the test was categorized into five groups: microbiology (94.7%), serology (2.8%), outside laboratory (1.7%), virology (0.5%), and immunology (0.4%).The reports consisted of final (68.7%), preliminary (31.0%), specimen in the lab (0.2%) and corrected (0.1%) result status (Table 1).

Among the 14,028,774 laboratory test reports, 76.8% had LOINC order codes, while for the rest either the codes were missing (16.8%) or had the HL7 null code, “OTH” (6.4%). No test orders were reported using a SNOMED CT code. Out of 10,776,494 laboratory test orders with LOINC codes, 9,347,179 (86.7%) also had either LOINC or SNOMED CT or both codes for the test result.

Of all laboratory reports, 9,347,179 (66.6%) had at least one result code (LOINC or SNOMED CT or both); of these, 91.8% had only LOINC codes, 0.7% had only SNOMED CT codes while 7.4% had both codes (Table 1). Of the remaining 4,681,595 laboratory reports, 63.4% were missing a SNOMED or LOINC result code with “OTH” reported as a result value while 36.6% had neither LOINC nor SNOMED codes for the results. Laboratory test results with “final” status comprised the majority (68.7%) of all laboratory reports with result codes (Table 1) Findings from the analysis comparing results with “final” status to all results were comparable except that the number of laboratory test results with only LOINC codes differed (91.8% in all messages versus 61.4% in only “final” result messages). This suggests that results reported as “preliminary” were not updated when result status changed to “final”.

For test orders, 805 unique LOINC codes were used. After excluding “OTH” and missing, values, LOINC codes to identify bacteria in blood (35.0%), urine (15.2%), respiratory specimens (4.8%), or wounds (4.2%), and to test antibiotic susceptibilities by agar diffusion (3.0%), were the five most frequent laboratory orders reported (Table 2). Fourteen hundred twenty eight unique LOINC and 608 unique SNOMED CT codes were used to describe laboratory test results. For results with LOINC codes, the five most commonly reported tests were: bacteria identified by blood culture (12.2%), microscopic observation of unspecified specimen by Gram stain (8.9%), appearance of unspecified specimen (7.6%), microorganism identified in unspecified specimen by culture (7.5%), and bacteria identified by urine culture (6.6%)(Table 3a). For results with SNOMED CT codes, a qualifier for antimicrobial susceptibility (10.4%), Escherichia coli (10.4%), a qualifier for bacterial sensitivity (9.9%), Staphylococcus aureus (9.7%), and a qualifier for non-reactive status (6.5%) were the five most frequent codes (Table 3b).

Among the 608 unique SNOMED CT codes used to report laboratory test results, 498 (81.9%) had corresponding LOINC codes reported. For the 108 SNOMED codes that did not have corresponding LOINC codes in the reported laboratory results, we found that 49 (45.4%) of the SNOMED CT concepts could be associated with at least one LOINC code. The majority of SNOMED CT concepts matched to LOINC (47 of 49) were microorganism related, while one was an anatomical structure (urethra) and one was a system concept (blood). Among 49 SNOMED CT concepts that matched to LOINC codes, 27 concepts matched to more than one LOINC code, depending on other LOINC components including property, timing, system, scale, and method. Among 59 SNOMED CT concepts not identified in mapping to LOINC codes, 47 were for microorganisms, seven were for qualifiers, three were for anatomical structures, and two were for systems.

The hospitals included in our study used LOINC as the coding system to record laboratory test orders. Laboratory test results were coded using both LOINC and SNOMED CT coding systems, though, the use of LOINC was much more common. Our findings indicate that participating hospital laboratories undertook reporting of laboratory test orders and results as recommended in ELR HL7 messaging guidance [32]. However, missing or non-informative codes comprised almost a quarter of laboratory test orders and a third of test results from hospitals reporting to BioSense in 2011.

By design, the primary objectives of BioSense program were to monitor clinical syndromes related to infectious diseases for early outbreak detection and ongoing situational awareness [27-29]. Therefore, Biosense recommended that participating hospitals preferentially report laboratory information related to infectious diseases. As a result, the majority of the laboratory reports in this analysis were microbiology related. We examined both test orders and preliminary and final test results. Monitoring laboratory test orders could provide early warning signals of suspicion of infectious disease, while monitoring results could contribute to biosurveillance by providing increased diagnostic specificity to automated syndromic case definition algorithms based on chief complaint text fields [26,33-35]. Laboratory data could also be used to evaluate interventions, and to monitor disease trends and progression, which might result in timely more effective outbreak response and management [1,13].

In ELR reporting, the debate over the definition of “questions” (orders) and “answers” (results) is far from over. It has been suggested that LOINC is used for coding laboratory “questions” while SNOMED CT is used for coding the “answers”. [22] Laboratory reports from hospitals in our analysis indicate that LOINC codes are used to code laboratory orders as well as results, illustrating LOINC’s potential to provide codes for “questions” as well as “answers”. On the other hand, there may be unique situations where a SNOMED CT code is required in combination with a LOINC code to fully represent a lab test result. For example, the SNOMED CT system allows coding for certain clinical structures (cervix, urethra), and for certain conditions (e.g., hyperbilirubinemia), that are currently not accounted for in the LOINC system.

Several earlier reports have suggested different mapping strategies to associate laboratory codes to diseases or health conditions [20,36-41]. Our findings suggest that the BioSense surveillance program might be able to focus on a small subset of LOINC and SNOMED CT codes related to diseases or syndromes of interest, as evidenced by the use of a relatively small number of unique LOINC (n= 1,428) and SNOMED CT (n= 608) codes in the data we analyzed compared to the number of codes available. To determine if laboratory data improves syndromic surveillance performance for enhanced outbreak detection and improved situational awareness, an evaluation of syndromic case definitions that incorporate laboratory test order or result information is required.

Some limitations in interpreting our findings should be noted. Non-federal hospitals who participated in BioSense are not a representative sample of all the U.S. hospitals. Use of the convenience sample of facilities that are able and willing to share electronic laboratory data with CDC may limit the generalizability of our findings. Furthermore, laboratory data in this analysis were primarily collected for infectious disease syndromic surveillance purposes. Therefore our findings may not be representative of the use of LOINC and SNOMED CT codes for ELR for all hospital laboratory testing. Completeness of ELR data in BioSense has not been validated prior to this study, which shows that one-third of electronic laboratory records are missing standardized codes or are coded as HL7 null “OTH” for result information that would be necessary for interpretation for surveillance. We are unable to report specific accuracy metrics such as specificity or sensitivity related to LOINC and SNOMED CT due to lack of a reference “gold standard”.

The current analysis has several important strengths, beginning with computability of the coded laboratory test order and result information in the HL7 messaging format that supports automated categorization for surveillance. Consequently, we were able to merge data files from different hospitals to create one large laboratory data repository and apply standard criteria that could potentially augment information from other surveillance data sources. Laboratory reports in this analysis derived from local hospitals illustrate the potential value of standardizing laboratory data to provide timely information to hospital based infection preventionists and local health departments for situation awareness and early response, as well as contributing to national biosurveillance. Most importantly this study adds value to existing scientific knowledge by describing the use of LOINC and SNOMED CT codes in ELR contributed by select U.S. hospitals as healthcare organizations move towards implementing Meaningful Use and health information exchange [42,43].

We analyzed more than 14 million laboratory reports from non-federal hospital to assess the use of two structured coding systems, SNOMED CT and LOINC. The LOINC system was used more commonly than SNOMED system in the hospitals studied. Missing data and differences in representing laboratory test orders and results may inhibit effective analysis of electronic laboratory data. Increased completeness of coded lab data, data management tools that translate locally coded laboratory test information into LOINC or SNOMED CT codes, and increased participation of hospitals laboratories are needed to fully realize the value of laboratory data in public health practice and syndromic surveillance.