Effect of HIV on mortality among hospitalised patients in South Africa

Background HIV and AIDS continues to impose substantial healthcare challenges in sub-Saharan Africa, but there are limited local data comparing inpatient outcomes between people with HIV (PLWH) and those uninfected. Objectives To compare cause-specific mortality among hospitalised adolescents and adults, stratified by HIV-serostatus. Method A cross-sectional analysis was performed, analysing cause-specific inpatient mortality data and total admissions, from 01 January 2017 to 30 June 2020, at Tshepong Hospital, North West province, South Africa. Results The overall inpatient mortality rate decreased from 14.5% (95% confidence interval [CI]: 13.4–16.0) in 2017, to 11.3% (95% CI: 10.6–11.9) in 2020; P < 0.001. People living with HIV accounted for 53.9% (n = 2342) of inpatient deaths, 22.6% (n = 984) were HIV-seronegative patients and 23.5% (n = 1020) patients with unknown HIV-serostatus. People with HIV died at younger ages (median: 44 years, interquartile range [IQR]: 35.8–54.2) compared to HIV-seronegative inpatients (median: 64.4 years, IQR: 55.5–73.9); P < 0.001. Leading causes of death were pneumonia (19.9%, n = 863), then pulmonary and extrapulmonary tuberculosis (15.0%, n = 654). People with HIV who had CD4+ counts < 350 cells/mL or viral load ≥ 1000 copies/mL had increased risk of death from tuberculosis compared to virally suppressed patients (adjusted relative risk: 2.10 [95% CI: 1.44–3.04, P < 0.009] and 1.56 [95% CI: 1.22–2.00, P < 0.001]). Conclusion Our study, conducted in a regional hospital in South Africa, showed PLWH had higher mortality rates and died at younger ages compared to HIV-seronegative patients.

PLWH and HIV-seronegative inpatients. This study describes inpatient mortality rates, at a regional healthcare facility in South Africa, over a period of 42 months. We compared the differences in cause-specific mortality between adolescent and adult PLWH and HIV-seronegative inpatients.

Methods
A cross-sectional analysis was conducted at a regional hospital in the North West province of South Africa. We analysed routinely collected data, describing mortality among hospitalised adolescents and adults between 01 January 2017 and 30 June 2020. Patients ≥ 15 years of age admitted to adult wards at Tshepong Hospital were considered eligible for inclusion in the study.
Two routinely collected databases were used in these analyses. The first included the number of admissions to medical wards each calendar month, stratified by year of admission, gender, and HIV-serostatus. There were no individual admission or discharge diagnoses, nor outcome data available to discriminate between patients who survived or died.
The second database was generated by weekly, standardised mortality meetings. The meetings were held throughout the study period and headed by at least two senior physicians. The same senior physicians supervised the process, ensuring a standardised approach of investigation and classification. To accurately complete statutory death notification forms, medical records and laboratory and histological results were discussed. The immediate, underlying and contributory causes of death were determined, and the outcome data for each case was entered into this second database (paper-based and electronic).
Additional variables entered into this mortality database included: demographic information, dates of admission and death, HIV-serostatus, ART history (including ART non-adherence), tuberculosis co-infection, haemoglobin concentration, CD4+ counts, and viral load (VL). Laboratory results obtained during admission were rechecked at the mortality meetings, and afterwards rechecked using the National Health Laboratory Service electronic database before being recorded in the local database. The database was deidentified prior to analysis; names were replaced by initials, and dates of birth were converted to ages at death, prior to analysis. The causes of death were categorised according to the International Statistical Classification of Diseases, Tenth Edition (ICD-10), mortality coding system. 18 The initial paper-based and electronic database a were compared and rechecked to ensure imputation errors were not made and that missing or incorrect data were kept at a minimum. If records were incomplete and causes of death not reliably determined, the specific cases were used in the calculation of overall mortality rates, but not included in the evaluation of cause-specific mortality.
Overall mortality rates were determined by using the total admission numbers as the denominator. Similarly, mortality rates in PLWH were determined by using the number of HIV-seropositive admissions as the denominator. Due to the unclear distribution of denominator data for patients without HIV (HIV-seronegative and unknown HIV-serostatus), these two groups were combined to calculate mortality rates.

Data analysis
The data set was imported to STATA ® 14.2 (College Station, Texas, United States for analysis. Descriptive statistics were used to describe the inpatients who died during the period of interest. Percentages and frequencies were used to describe the categorical data. For the description of continuous variables, medians and interquartile ranges (IQR) were used. Mortality rates were determined as the total number of inpatient deaths divided by total admissions for the same period and expressed as percentages with a 95% confidence interval (CI). The chi-squared test was employed, to test for differences in mortality rate trends, between groups stratified by HIV-serostatus.
The leading causes of death overall were explored, as well as for groups stratified by HIV-serostatus, age categories, most recent CD4+ counts, and VL results within 6 months of death. We defined patients on ART with VL ≤ 50 copies/mL (within 6 months preceding death) as being virally suppressed. 19 Proportions of the leading (immediate, underlying and contributory) causes of death were determined, and compared across categories.
Univariable and multivariable Poisson regressions with robust error variance were used to determine the factors associated with the leading causes of death: pneumonia and tuberculosis (pulmonary and extrapulmonary). Viral, bacterial (excluding mycobacterium tuberculosis) and fungal infections of the lower respiratory tract were classified as pneumonia. Age, gender, year of death, HIV-serostatus, as well as most recent CD4+ counts and VL results, were included in a priori multivariable models, and we reported adjusted relative risks (aRR). P-values of less than 0.05 were determined to represent statistically significant differences.

Discussion
This study showed a decline in overall mortality, at a regional hospital, both for PLWH and those HIV-seronegative or of unknown serostatus. Higher mortality rates persisted in PLWH. Pneumonia and tuberculosis (pulmonary and extrapulmonary) were the leading causes of death, regardless of HIV-serostatus. The risk of dying from pneumonia and tuberculosis was independently associated with age, HIV-seropositivity and, in PLWH, higher VL and lower CD4+ counts.
We postulate that the improved outcomes were likely due to both improved patient care and improved access to ART. 2,3 Over the study period, Tshepong Hospital improved inpatient HIV testing and counselling, likely leading to earlier HIV diagnosis and ART initiation, improved adherence, and possibly improved overall survival among PLWH. Departmental mortality meetings were used as a learning opportunity to identify shortcomings in patient management, and might have contributed to the improved secular trend in mortality we report. 20,21 However, our data highlighted the remaining gaps between PLWH and their HIV-seronegative peers. Although PLWH had improved outcomes over time, higher inpatient mortality rates were observed compared to HIV-seronegative inpatients and those with unknown HIV-serostatus. 15,22,23,24,25 Moreover, deaths in PLWH occurred at younger ages, while HIV-seronegative patients died at ages comparable to those predicted by current life expectancy estimates in South Africa. 8 The effects of chronic inflammation and premature aging seen in PLWH may represent a possible reason for this discrepancy. 26 Screening for and treatment of comorbidities at younger ages may be needed for PLWH. 26,27,28 Initiation of ART closer to the time of HIV infection, improved adherence, and earlier switching of failing regimens may further reduce mortality. 29 Consistent with other studies in sub-Saharan Africa, the leading cause of death among PLWH was tuberculosis. 15,23,30,31,32,33,34 Regardless of CD4+ count or VL, PLWH were at an increased risk of dying from tuberculosis compared to HIV-seronegative inpatients. However, risks were reduced in virally suppressed PLWH and those with higher CD4+ counts. The highest risk of death resulting from pneumonia and tuberculosis was observed in men, inpatients with VL ≥ 1000 copies/mL, and CD4+ counts < 350 cells/mL. A concerted effort to improve determinants on a healthcare level, patient education, and to decrease the community transmission of tuberculosis, are needed to address this problem. 30 The improved mortality outcomes observed in virally suppressed PLWH demonstrates the efficacy of ART in decreasing mortality and improving survival.
In our study, up to 9.3% of reported deaths among PLWH were caused by neoplastic diseases. Similarly, a Zimbabwean  http://www.sajhivmed.org.za Open Access study reported increased rates of death resulting from malignancies and non-communicable diseases in PLWH. 23 South Africa is one of the countries with the highest numbers of HIV/AIDS-related deaths. 35 We reported HIV and AIDS as the leading specific underlying cause of death by a large margin.
A high proportion of inpatients who died reported previous tuberculosis, with disrupted lung architecture frequently seen as a sequela of pulmonary tuberculosis. This may be a possible reason for the high burden of respiratory tract infections observed in this study. When considering all pathological processes, diseases of the respiratory system were involved in 41.5% (n = 1801) of all deaths. Collectively, chronic obstructive pulmonary disease and fibro-cavitary lung diseases represented 12.5% of the reported underlying causes of death. Strategies promoting smoking cessation, vaccination, and earlier tuberculosis diagnosis may yield further improved outcomes. In South Africa, the first wave of the coronavirus disease 2019 peaked in July 2020. 36 This coincides with an increased admission and mortality rate from March to June 2020, as some of the admissions and deaths may have been due to coronavirus disease 2019. Anaemia was the leading contributory cause of death in our study, reflecting the negative impact of low haemoglobin concentrations on inpatients outcomes. 37,38,39,40 Furthermore, this study highlights the need to improve inpatient management of communicable diseases. Possible areas that need improvement include earlier recognition and aggressive management of pneumonia and septicaemia, with timely escalation of treatment. Better pneumonia scoring systems, that include HIV infection and CD4+ counts, may also ensure earlier recognition and treatment of high-risk patients. 41 Further improvement, with linked HIV and tuberculosis care, is needed, together with close communication between inpatient and outpatient care. 30 Efforts and research into the improving cooperation between the different levels of healthcare, earlier detection of diseases like tuberculosis, as well as education of hospitalised patients may yield further survival benefits. Specific emphasis on using the admission as opportunity to provide healthcare education and counselling on treatment adherence may impact readmissions and patient outcomes.
We evaluated a large data set, lending strength and better internal validity to the findings. However, our study has limitations. The study was conducted at a single centre, limiting external validity. To minimise misclassification, a systematic process was undertaken to review all the deaths that occurred during the period of interest. Misclassification of death is a substantial problem in South African mortality statistics, especially overreporting of tuberculosis as the underlying cause of death instead of using HIV infection. 42,43,44 Extensive cross-checking of patient records, clinical information, and investigations was performed to classify the causes of death, but the causes of death were not determined by autopsy, which often identifies incorrect premortem diagnoses. 45 Limited data were available regarding the characteristics of those who survived admission (including final discharge diagnoses), which made it impossible to calculate mortality rates for subgroups stratified by age, VL, and CD4+ count and diagnostic category. Moreover, the duration since HIV diagnosis, the time from diagnosis until treatment initiation, nadir CD4+ counts and other characteristics of patients who died (or survived) their admission, were not available.

Conclusion
Despite the high burden of communicable diseases, inpatient outcomes have improved. However, PLWH, including those who were virally suppressed, died at younger ages, and experienced a higher burden of communicable diseases compared to HIV-seronegative patients. Higher risk factors for death were seen in male patients, as well as in patients with CD4+ counts < 350 cells/mL and VL > 1000 copies/mL in the 6 months prior to death. As ART coverage increases, and continued improvement is observed in the survival of PLWH, mortality and burden of disease need to be monitored to ensure that trends continue to improve. Future research should be conducted to better understand the causes and trends of inpatient mortality and longevity gap that exists between PLWH and their HIVseronegative counterparts, and to develop interventions to improve outcomes.
supplementary materials. (Figures 1, 2a, 2b, 2c; Online Appendix 1 Figure 2d, 2e, 2f, and Tables 1, 2a, 2b, 3 have associated raw data). The data set, and raw data are not publicly available, but access can be provided on reasonable request to the corresponding author, D.J.L.

Disclaimer
The views and opinions in this article are those of the authors and do not necessarily reflect the official policy or position of any affiliated agency of the authors.