Study design and population
This study took place at the Komfo-Anokye Teaching Hospital (KATH), a 1200-bed tertiary hospital located in Kumasi, the capital city of the Ashanti Region of Ghana. Patients are referred to KATH from each of the ten regions of Ghana. Adult patients greater than 18 years old admitted to the internal medicine service at KATH with documented HIV infection were included in the study. The study included patients who were admitted to the hospital with a previous diagnosis of HIV, as well as those who were newly diagnosed during their current admission. Patients without valid dates of admission, discharge or death recorded in the patient chart were excluded from analysis.
Procedures and measurements
This retrospective, cross-sectional, single-center study took place over a six-month period from January to July 2016. The charts of patients who had been discharged or died during this study period of interest were reviewed for HIV status. In order to find relevant cases, physicians and research assistants involved in the study examined the daily discharge and death folders on each internal medicine ward in order to capture all eligible patient records.
For patients found to be HIV-positive, researchers completed a standardized data collection form from the patient chart. The form included demographic information, such as age, sex, and date of admission; presenting symptoms and physical exam findings; information about HIV diagnosis, including known or new diagnosis, HAART history, and HIV clinical staging; laboratory data, when available; and final diagnoses and outcomes (death or discharge). Only prerecorded data available in the patient charts were used for data collection, and no patients or physicians were interviewed for data collection.
Definition of HIV positive status and HAART experience
Patients were considered eligible for review if they were diagnosed with serologically-proven HIV upon current admission to the hospital or if a known prior diagnosis had been documented in the patient chart. A diagnosis was documented as “new” if the patient was diagnosed on current admission to KATH, while a diagnosis was deemed “known” if the patient had ever been previously diagnosed with HIV, even if that diagnosis occurred at a referral hospital just before arriving at KATH. Patients were considered to have HAART experience if the patient’s chart documented that the patient had ever previously been started on HAART or was currently receiving HAART therapy.
Diagnostic methods
Pulmonary tuberculosis was diagnosed when the patient exhibited (1) consistent clinical findings, (2) suggestive chest x-ray, and (3) placement on standard TB treatment. Patients who presented with focal or generalized neurological symptoms suggestive of an intracranial space-occupying lesion unable to obtain head CT were classified as “undifferentiated intracranial space occupying legion (ICSOL)” with a wide differential diagnosis including stroke, toxoplasmosis, brain abscess, tuberculoma or intracranial malignancy. Patients able to undergo head CT were categorized as CT-suggested toxoplasmosis when radiographic evidence revealed multiple, ring-enhancing lesions. Anemia, thrombocytopenia, and pancytopenia were confirmed via laboratory results in the patient chart. A diagnosis of cryptococcal disease was based on a record of serum or CSF cryptococcal antigen result or Indian ink stain result. Other, less common diagnoses listed in patient charts utilized a variety of clinical and laboratory findings to diagnose the patient.
Statistical analysis
All patient records used for the study were assigned a random alpha-numeric identifier in order to protect patient confidentiality. Patient records were entered into a FileMaker 12 Pro (FileMaker, Inc., Santa Clara, California) database. Subsequently, data were extracted from the database and analyzed using SAS version 9.4 (Cary, NC).
Summary frequencies and proportions were used to describe the sample for all nominal characteristics including patients’ sex, transfer from another hospital to KATH, symptoms, and comorbidities. Medians with interquartile range were used to describe all other demographics including patients’ age, months since HIV diagnosis, and laboratory values. Univariable Cox proportional hazards models were used to estimate the risk of mortality following date of admission as a function of patients’ comorbidities and demographic characteristics. In these models, elapsed time was measured in days from date of admission to date of death (if deceased) and living patients were censored on their discharge date. The proportional hazard assumption for each predictor was assessed graphically using Martingale residuals as described by Lin, Wei, and Ying [23].
A multivariable Cox proportional hazards model was used to estimate the adjusted risk of mortality as a function of HAART while controlling for patients’ age, sex, presence of neurological symptoms, clinical stage, ICSOL diagnosis, and pneumonia. These covariates were selected because of their importance on univariable analysis and improvement in multivariable model fit statistics, including Akaike information criterion (AIC).
Ethics clearance
Ethics approval was granted by the Kwame Nkrumah University of Science and Technology and the Komfo Anokye Teaching Hospital ethics review committee (CHRPE/347/15). Consent from individual patients was waived, as this project involved only a retrospective review of patient charts without any acquisition of identifying patient information.