The WHO clinical staging uses clinical presentation of patients to categorize HIV infection into 4 stages based on the severity of ADIs and prognosis. The stages reflect progression of HIV infection from asymptomatic to conditions where presumptive diagnoses can be made on the basis of clinical presentation or simple investigations [8]. Clinical staging is progressive, meaning that when a patient has recovered from a condition defined in a higher stage, they retain the higher clinical stage and can never be classified at a lower stage. Patients with stage I and II HIV infection present with (or have a history of) conditions such as minor skin diseases and upper respiratory tract infections, which are indicative of early HIV infection. WHO clinical stages III and IV, which are indicative of advanced HIV disease, may manifest as one or more illnesses such as severe bacterial infection, extra-pulmonary tuberculosis and non-Hodgkin lymphoma. WHO recommends initiation of ART for patients who have stage III or stage IV conditions [8].

The Kenyan Ministry of Health (MOH) HIV care and treatment guidelines [17], which are adapted from the WHO guidelines, recommend that all HIV-infected pregnant women receive ART, irrespective of their immunological or clinical indication of opportunistic infections associated with HIV [7].

SNOMED CT is an international logic-based, controlled vocabulary used in clinical care that amongst other areas includes terms for diagnostic concepts [18]. SNOMED CT enables meaning-based retrieval of clinical information from electronic health records (EHR) and is able to express finer-grained concepts than classification systems such as ICD-10. For this study, we used the January 2014 release of SNOMED CT which had 298,581 active concepts associated with 781,878 active descriptions. The concepts are interrelated by 896,942 active relationships which can be hierarchical (Is-A relationship) or non-hierarchical (e.g. “associated morphology” or “finding site”). A SNOMED CT reference set provides a mechanism to group concepts and/or descriptions for a common goal (such as diagnosis concepts for AIDS defining illnesses).

Rosenbloom et al. describe an interface terminology as a systematic collection of health-care phrases or terms that support clinicians’ entry of patient-related information into computer programs and decision tools [19]. Interface terminologies also support the display of computer-stored patient information to clinical users in simple human-readable format. The use of a SNOMED CT based terminology on the user interface of an EMR supports efficient selection of terms during data entry based on uniquely coded concepts and promotes data re-use in decision support, statistical analysis and reporting.

The Kenya Medical Research Institute (KEMRI), a collaborator in this study, applied for and obtained a free institutional SNOMED CT license from IHTSDO within the category of developing countries. Under this license, KEMRI can access the SNOMED CT database and use the concepts, descriptions and relationships in EMRs supporting clinical care and research. The license is renewed annually.

OpenMRS is a free, collaborative, open source software that supports the delivery of health care and is widely used in developing countries. It is based on a “concept dictionary” that describes data items such as clinical findings and laboratory results. It provides a platform which can be customized through addition of new data items, forms and reports without programming. It is web based but can be deployed on a single standard-alone desktop computer, a laptop or a large server and runs on Linux, Windows or Mac OS X.

The JOOTRH provides ambulatory HIV care and treatment services to over 7000 active patients, of whom about 4800 are receiving ART. The services are provided within the Maternal and Child Health (MCH), the HIV clinic/Patient Support Centre (PSC) and the Integrated Tuberculosis clinic. The three clinics in the hospital use a hybrid system of paper records and EMR to manage patient data; the EMR system is currently under staged deployment and is expected to replace the paper records by the end of 2014. The study was conducted at the HIV/PSC clinic and the MCH clinic which includes the Ante-Natal Care (ANC) clinic.

We extracted free-text data used to describe diagnoses, signs and symptoms from paper-based clinical notes for all HIV-positive patients receiving care and treatment services at the HIV/PSC as well as HIV-positive pregnant women presenting for antenatal care or at the labor and delivery room during the months of February and March 2014, and receiving Antiretroviral Therapy (ART). We also conducted a Focus Group Discussion (FGD) among clinicians and nurses who attend to patients visiting the MCH and HIV/PSC clinics.

A trained clinician (clinical officer) extracted terms for diagnoses, signs and symptoms recorded in the patient charts to determine the terms used to document ADIs at the ANC and HIV/PSC. Data were retrieved from the two clinics to ensure comprehensive coverage of terms used to describe ADIs. Although terms for diagnoses, signs and symptoms were retrieved from the charts, we only included in the analysis those that matched SNOMED CT “disorder” concepts as described later.

Based on a framework developed by Bakhshi-Raiez et al. [16], we derived the reference set of ADI concepts through a multi-step approach. The first three steps describe how the ADI reference set was derived as shown in Fig. 1. In the first step, concept-based subsets of ADIs from SNOMED CT and Accenture/CDC (described below) were merged. The CliniClue Xplore browser [20] and NICTIZ terminology explorer [21] were used to explore the concepts. The second step entailed extraction of term-based concepts from patient charts and Focus Group Discussions. In step 3, the concept-based and term-based concepts were merged.

We matched each of the term-based concepts to SNOMED CT concepts in order to assess content coverage in the latter. Due to the small number of concepts, we did a manual matching and recorded the concepts that could not be found in SNOMED CT.

The resulting concept subset and description subset from the above steps were represented in the OpenMRS concept dictionary and implemented as the interface terminology on OpenMRS data entry forms to improve ease and accuracy of recording of ADIs. A separate set of non-ADI concepts representing common co-infections or signs, symptoms among patients seeking HIV treatment at JOOTRH was also created from the chart review and FGD for the purpose of presentation on the user interface. Subsetting the concepts into ADI and non-ADI disorders for the purpose of presentation on the user interface provides clinicians quick access to shorter lists of relevant diagnosis concepts to choose from. The ADI concepts were presented on the user interface as a dropdown menu with autocomplete/autosuggest textbox feature derived using a list builder. A ’more details’ button is displayed for each of the ADIs selected for a particular patient. Clicking on this button allows the selection of qualifiers of a particular diagnosis.

The patients’ WHO staging is automatically inferred based on the recorded ADI concept. To accomplish this, the ADI concept is matched with the corresponding WHO stage as defined in the HIV treatment guidelines [8,17]. The WHO clinical staging list showing the ADI concepts and corresponding WHO clinical stage is in Appendix B [8].

The study was approved by the Associate Director for Science at the Center for Global Health of the US Centers for Disease Control and Prevention (CDC) and the Kenya Medical Research Institute (KEMRI) institutional review board. The women whose data were included in the study consented to the use of their data for clinical care and research studies taking place at the ANC clinic. As the study subjects were considered a vulnerable population (HIV-positive pregnant women), the study team ensured that the individual patient data were de-identified and their privacy and confidentiality maintained in line with the KEMRI guidelines on research ethics.

The CDC/Accenture ADI subset had 907 unique disorder concepts described by 1528 terms. For the SNOMED CT subset, the first query (concepts described by terms including “assoc” and (”AIDS” or “HIV”)) yielded 102 concepts while the second query (concepts linked by the relationship “associated with” (ID = 47429007) to the concept “HIV infection” (ID = 86406008)) resulted in 105 unique concepts. Merging the two subsets described above and removing duplicates resulted in 109 unique concepts. Finally, the subtype children of the resulting concepts were derived using the fourth query. Where relevant, these concepts were replaced by their supertype concept for which the description did not contain “associated with AIDS”. For the resulting 110 concepts, 254 terms together with their description ID were retrieved from SNOMED CT.

A total of 87 ADI concepts were derived from the term-based subset i.e. those abstracted from the paper records of patients attending the MCH clinic and PSC, and from the FGD. The concepts were described using 157 terms.

The overall content coverage was 99.9% with pre-coordinated concepts and 100% with post-coordination. All but two term-based concepts from the chart abstraction at the MCH and PSC, as well as the FGDs matched the SNOMED CT concepts. The two concepts that required post-coordination were: (i) Oropharyngeal candidiasis, and (ii) genital ulcer disease. Oropharyngeal candidiasis was post-coordinated as: Candidiasis (78048006): finding site (363698007) = Mouth and/or pharynx structures (312533001). Genital ulcer disease was post-coordinated as: Ulcer (429040005): finding site (363698007) = Structure of genital organ (700037000).

The implementation of the interface terminology includes 1860 terms describing 1054 unique concepts. The data entry forms, developed using the HTML Form Entry module of OpenMRS, display ADI concepts from the reference set and include qualifiers such as severity (e.g. “mild”, “moderate” or “severe”), onset (e.g. “sudden” or “gradual”) and laterality which describes the side of the body that is infected (e.g. right middle zone pneumonia indicative that the infection is in the right middle lobe of the lung). Defining characteristics which include causative agent were implicit to the interface terminology, allowing more specific types of diagnosis to be selected (e.g. for “pneumococcal pneumonia” where “pneumococcus” is the causative agent). Additional context information that can be selected is the type of diagnosis, i.e. presumptive (working), differential or final diagnosis and status (“active” or “inactive”) (see Fig. 2). Once ADI diagnoses are captured, using the data entry form, the concept ID is stored and can be re-used for decision support, e.g. inferring ART eligibility and to generate statistical summaries. Statistical summaries are reported to the Kenyan Ministry of Health on a monthly basis.