HIV‐associated morbidity and mortality in a setting of high ART coverage: prospective surveillance results from a district hospital in Botswana

Abstract Introduction Antiretroviral therapy (ART) has significantly improved survival in Africa in recent years. In Botswana, where adult HIV prevalence is 21.9%, AIDS‐related mortality is estimated to have declined by 58% between 2005 and 2013 following the initial wide‐spread introduction of ART, and ART coverage has steadily increased reaching 84% in 2016. However, there remains little data about the burden of HIV and its impact on mortality in the hospital setting where most deaths occur. We aimed to describe the burden of HIV and related morbidity and mortality among hospitalized medical patients in a district hospital in southern Botswana in the era of widespread ART coverage. Methods We prospectively reviewed medical admissions to Scottish Livingstone Hospital from December 2015 to November 2017 and recorded HIV status, demographics, clinical characteristics and final diagnoses at discharge, death or transfer. We ascertained outcomes and determined factors associated with mortality. Results were compared with similar surveillance data collected at the same facility in 2011 to 2012. Results A total of 2316 admissions occurred involving 1969 patients; 42.4% were of HIV‐positive patients, 46.9% of HIV‐negative patients and 10.7% of patients with unknown HIV status. Compared to HIV‐negative patients, HIV‐positive patients had younger age (mean 42 vs. 64 years, p < 0.0001) and higher mortality (22.2% vs. 18.0%, p = 0.03). Tuberculosis was the leading diagnosis among mortality cases in both groups but accounted for a higher proportion of deaths among HIV‐positive admissions (44.5%) compared with HIV‐negative admissions (19.4%, p < 0.0001). Compared with similar surveillance in 2011 to 2012, HIV prevalence was lower (42.4% vs. 47.6%, p < 0.01), and among HIV‐positive admissions: ART coverage was higher (72.2% vs. 56.2%, p < 0.0001), viral load suppression was similar (78.6% vs. 80.3%, p = 0.77), CD4 counts were higher (55.0% vs. 44.6% with CD4 ≥200 cells/mm3, p < 0.001), mortality was similar (22.2 vs. 22.7%, p = 0.93), tuberculosis diagnoses increased (27.5% vs. 20.1%, p < 0.01) and tuberculosis‐associated mortality was higher (35.9% vs. 24.7%, p = 0.05). Conclusions Despite high ART‐coverage in Botswana, HIV‐positive patients continue to be disproportionately represented among hospital admissions and deaths. Deaths from tuberculosis may be contributing to lack of reduction in inpatient mortality. Our findings suggest that control of HIV and tuberculosis remain top priorities for reducing inpatient mortality in Botswana.


| INTRODUCTION
The introduction and scale-up of antiretroviral therapy (ART) over the past decade has led to significant improvements in quality of life, life expectancy and mortality in sub-Saharan Africa, but the impact of expanded ART availability on the burden of HIV-related mortality in the hospital setting remains unknown [1,2]. Recent studies from countries in sub-Saharan Africa report high inpatient mortality rates despite increased access to free ART [3][4][5][6][7][8]. In-hospital mortality among HIV-positive patients may be influenced by factors such as rate and timing of ART uptake, toxicity of available ART regimens, co-morbidities and facility level and health system resources in different settings. The persistently high inpatient mortality reported across diverse settings in Africa suggests that some HIV-related conditions may not be addressed (or immediately addressed) following ART initiation, and that the scale-up of ART programmes has not in itself been sufficient to curb the impact of HIVrelated mortality on hospital systems.
Botswana, a middle-income country in southern Africa, has the world's third highest prevalence of HIV with a current adult HIV prevalence of 21.9% [9]. Botswana has long been considered a leader in HIV response amongst African nations and in 2001 became the first country in sub-Saharan Africa to provide free ART. This coverage is now offered regardless of CD4 cell count and reaches approximately 84% of citizens living with HIV [9].As a consequence, model-based trend estimations suggest that AIDS-related mortality in Botswana has declined by 59% between 2005 and 2013 [9,10] and by a further 31% between 2013 and 2018 [11]. However, there remains little ART-era data from Botswana evaluating the burden of HIV-related mortality in the hospital setting, where most deaths occur.
We established comprehensive inpatient data surveillance at Scottish Livingstone Hospital (SLH), a government district hospital in southern Botswana. Our objectives were to assess the burden of HIV among medical inpatients at SLH, to describe mortality by HIV status, and to identify factors independently associated with inpatient mortality.

| Study design
We conducted a prospective review of all patients discharged from the adult medical wards at SLH over a 24-month period from 1 December 2015 to 30 November 2017. All medical admissions to the adult female and male medical wards were included. As per hospital policy, adult admissions included females ≥12 years and males ≥13 years.

| Study site
SLH is a 350-bed public sector district hospital located in Molepolole (population 73,102), the largest village in Botswana and the main referral centre for the Kweneng East district, which has a population of approximately 190,000 and an overall HIV prevalence of 19% [12,13]. ART is offered in 20 public sector clinics across the district and, though exact ARTcoverage data from the district is unavailable, it is estimated to approximate the national average (UNAIDS, private communication).
Available diagnostic resources at SLH include basic haematology, chemistry and microbiology testing, plain x-rays and ultrasound. Stock-outs of reagents and machine breakdowns frequently interfere with regular availability of diagnostic test results. The hospital's medical wards are staffed by non-specialized medical officers, rotating medical interns and internal medicine specialists.

| Data collection
Collected data included patient demographics, co-morbidities, HIV status, HIV testing results, anti-retroviral treatment, CD4 count, HIV viral load, final diagnosis at discharge, death or transfer and mortality status at discharge or following transfer. Data were collected upon patient discharge, transfer or death and entered into an electronic data abstraction form. Clinical data were collected from patients' hand-written medical records. Laboratory data were collected from the Integrated Patient Management System (IPMS), an electronic medical record system used in most hospitals in Botswana. Post-transfer outcomes were ascertained from IPMS and, when not available electronically, by direct communication with the accepting facility. Data were collected by trained research assistants and medical ward doctors, and all final discharge diagnoses underwent secondary review and adjudication by an internal medicine specialist.

| Statistical methods
Final outcomes were classified as "discharged from hospital alive, " "died in hospital, " "left against medical advice" or "transferred to another ward of facility. " For the purposes of final outcome analysis, patients who left against medical advice were considered alive at final discharge and transferred patients were analysed per the hospitalization outcome at the location to which they were transferred.
All data were analysed using SAS 9.4 (SAS Institute, Cary, NC). We compared categorical data using chi-square and Fisher's exact tests and t-tests or Wilcoxon rank-sum tests were used to compare continuous variables. Risk factors for mortality were evaluated using modified Poisson regression to estimate risk ratios and 95% confidence intervals, adjusted for age, sex and presence or absence of known co-morbidities.
Approval to conduct data collection and report outcomes were granted by institutional review boards at Scottish Livingstone Hospital, Molepolole, Botswana and Beth Israel Deaconess Medical Center, Boston, US and by the Botswana Human Research and Development Committee, Gaborone, Botswana.

| Study population
Over the 24-month surveillance period, 2316 admissions occurred among 1969 patients; 1079 (46.6%) admissions were to the male medical ward, and 1237 (53.4%) were to the female medical ward. Median patient age at admission was 51 years (interquartile range (IQR): 34 to 71), and the median duration of admission was 6.0 days (IQR: 3.0 to 11.0). The majority of admissions were among patients who lived in Molepolole (66.7%), with a smaller number of patients admitted from rural villages in the Kweneng East district (25.3%), other rural areas (3.4%), or from the nearby urban centre of Gaborone (2.6%). (Table 1).

| HIV status and HIV testing
A total of 904 (39.0%) admissions were of patients known to be HIV-positive prior to presentation, 575 (24.8%) were of patients with a previously negative test and 837 (36.1%) were of patients with unknown HIV status. Among patient admissions with unknown HIV status on presentation, 591 (70.6%) were tested during the hospitalization and of those tested 69 (11.7%) were found to be HIV positive. Males were as likely as females to have an unknown HIV status prior to admission (35.8% of male admission vs. 36.5% of female admissions, p = 0.73) and among those who were tested during hospitalization the risk of HIV positivity did not differ significantly among male admissions compared with female admissions (13.7% vs. 9.7%, p = 0.13). Among patient admissions with prior documentation of a negative HIV test, 157 (27.3%) retests occurred during the hospitalization and 10 (6.4%) were found to be HIV positive.

| Baseline characteristics by final HIV status
Among all admissions with a known final HIV status, differences in the number of male and female admissions by HIV status were not observed (p = 0.38). In addition, geographic distribution was similar between HIV-positive and HIV-negative admissions. However, HIV-positive admissions were younger, with a median age at admission of 42 years (IQR 33 to 52) compared with 64 years (IQR 40 to 78) (p < 0.0001) for HIV-negative admissions and had fewer recorded chronic co-morbidities compared with HIV-negative admissions (35.9% vs. 75.0%, p < 0.0001). Baseline characteristics by HIV status are outlined in Table 1.

| ART, CD4, HIV viral load
Among admissions of patients who were known to be HIV positive prior to admission and whose pre-admission ART status was known, 653(72.5%) were on ART. Males were more likely not to have been on ART prior to admission compared with females (32.4% vs. 23.0%, p = 0.002). Among admissions of patients on ART who had a known viral load, 493 (89.6%) had undetectable viral load (<400 cp/mL) at their last test (either prior to or during the admission). Detectable viral loads (≥400 cp/mL), despite being on ART, were more likely to occur among younger patients, with a median age of 37 years (IQR: 31 to 49) compared with a median age of 43 years (IQR: 36 to 55) in patients who had undetectable viral loads (p = 0.02). The occurrence of detectable viral loads in patients on ART did not differ significantly between males and females (7.9 vs. 9.4%; p = 0.48). Being on ART prior to admission was associated with higher CD4 cell count at last test (either prior to or during admission), with a median of 310 cells/mL (IQR, 136 to 522) for those on ART versus 103 cells/mL (IQR, 40 to 252) for those not on ART, p < 0.0001.

| Admission outcomes and effect of HIVpositive status on mortality
In total, 420 (18.1%) admissions resulted in death in-hospital, 1749 (75.5%) resulted in live discharge, 17 (0.7%) resulted in patient leaving against medical advice, and 130 (5.6%) resulted in patient transfer to another ward or facility. Among patient transfers, 42 (32.3%) resulted in patient death, 87 (66.9%) resulted in live discharge from the ward or facility the patient was transferred to, and in 1 (0.8%) the post-transfer outcome could not be ascertained. Overall, 462 (20%) admissions resulted in patient death either on the SLH medical wards or following transfer to another ward or facility. In total, while HIV-positive patients accounted for 42.4% of admissions, they accounted for 47.2% of deaths. Admissions of HIV-positive patients were 1.2 (95% CI: 1.0 to 1.5) times more likely to result in death compared with admissions of HIV-negative patients. After adjusting for age, sex and co-morbidities, HIV-positive admissions were 1.6 (95% CI: 1.4 to 2.0) times more likely to result in death compared with HIV-negative admissions. Of note, among admission of patients aged ≤60 years, HIV-positive patients accounted for 79.7% of inhospital deaths and were 2.2 (95% CI: 1.6 to 3.0) times more likely to die compared to HIV-negative patients in fully adjusted analysis. Median time-to-death during hospitalization was similar among HIV-positive (four days, IQR 2 to 12) and HIV-negative (five days, IQR 2 to 13) admissions. Hospitalization outcomes by HIV status are shown in Table 2.
Clinical or laboratory diagnosis of tuberculosis accounted for a higher proportion of admissions (27.5% vs. 8.2%, p < 0.0001) and a higher proportion of deaths (44.5% vs. 19.4%, p < 0.0001) among HIV-positive admissions compared with HIV-negative admissions respectively. Of note, among all patients, the majority of diagnoses of both pulmonary and extra pulmonary tuberculosis were made based on clinical criteria. Of the 283 patients diagnosed with pulmonary tuberculosis (with or without extra pulmonary disease), sputum AFB smear results where available for only 75 patients, 13 of which (17.3%) were sputum positive and sputum GeneXpert results were available for 39 patients, 15 of which (38.5%) were sputum positive. TB culture results were not available for any patient. In most cases unavailability of sputum AFB smear, GeneXpert and TB culture results was due to stockouts of lab reagents, equipment breakdown or patient inability to produce sputum. Table 3 lists the 10 most common diagnoses among admissions with known final HIV status, the observed associated mortality by HIV status and the corresponding crude and adjusted risk ratios for mortality among HIV-positive versus HIV-negative admissions. Admissions of HIV-positive patients for gastroenteritis (aRR 11.8, 95% CI 1.5 to 96.3) and diabetes (aRR 6.6, 95% 1.2 to 36.6) were more likely to result in death when compared to admissions of HIV-negative patients for the same conditions, after adjustment for age and sex (however these estimates are based on small numbers). Among patients admitted for any of the remaining eight most common diagnoses no significant differences in the likelihood of death were observed between HIVpositive and HIV-negative patients.  Table 4, and included older age, male sex, low CD4 count, detectable HIV RNA and not being on ART. Among HIV positive admissions, after adjustment for age, sex and co-morbidities, male sex was associated with1.4 (95% CI: 1.0 to 1.7) times increased risk of death; admission with a CD4 count <200 cells/mm 3 was 1.5 (95% CI: 1.2 to 2.0) times more likely to result in death compared to admission with a CD4 count ≥200 cells/mm 3 ; and admission with a detectable HIV RNA viral load of >400 copies/mL was 1.5 (95% CI: 1.1 to 2.1) times more likely to result in death compared to admission with an undetectable HIV RNA viral load (<400 copies/mL).The risk of mortality during admissions of HIV-positive patients who were not on ART was significantly higher compared with admissions of HIV-positive patients who were on ART prior to admission (29.8% vs. 19.5%, p < 0.001).
After adjusting for age, sex and co-morbidities being on ART prior to admission conferred a 31% (95% CI: 12.0 to 47.0%) reduced risk of death compared to not being on ART. Overall mortality among HIV-positive admissions who were on ART prior to admission did not differ significantly from that of HIV-negative admissions (19.5% vs. 18.0%, p = 0.44). However, after adjusting for age, the risk of mortality among HIVpositive patients on ART was significantly higher when compared to HIV-negative patients (aRR 1.6, CI: 1.3 to 1.9).

| Comparison with 2011 to 2012 surveillance
We compared our results with those of similar surveillance conducted at SLH over a 6-month period between November 2011 and April 2012 [14].

| DISCUSSION
This is the largest study to date examining the impact of HIV on medical inpatient morbidity and mortality in the ART era in Botswana. Our findings demonstrate an ongoing high burden of HIV-related inpatient mortality, particularly from tuberculosis. HIV-positive patients were disproportionately represented among hospital admissions and deaths, and admissions of HIVpositive patients were more likely to result in death compared with those of HIV-negative patients. Mortality was particularly high among HIV-positive males and among HIV-positive patients who were not on ART, and, of importance, HIV was also associated with a significantly younger age of both admission and death. Tuberculosis accounted for 44.5% of all mortality among HIV-positive patients, and a significantly higher proportion of mortality compared with HIV-negative admissions. The introduction of highly active ART in developed countries in the mid 1990's led to significant decline in hospital admissions for AIDS-defining conditions and opportunistic infections [15][16][17][18]. However, a recent systematic review and meta-analysis of hospital admissions among people living with HIV globally showed that AIDS-related illnesses and bacterial infections continue to be the most common causes of hospital admissions and mortality among HIV-positive children and adults in all geographical regions [19]. Few ART-era studies from the African region have previously reported on rates of ART-coverage and mortality among HIV-positive adult medical admissions. The HIV-related medical inpatient mortality rate of 22% on our wards was similar to that from hospital settings in Malawi (24%, 25%) where a comparable ART-coverage rate (72%) was reported [3,4] and lower compared to a multi-centre study from West Africa which reported 38% mortality in the setting of lower (48%) ART-coverage [5].
Widely divergent HIV-related inpatient mortality rates of 17% and 30% have been reported from settings in Kenya and DRC respectively, where a relatively high proportion of patients were on ART at the time of admission [20]. At the same time another study, from a district hospital in South Africa, reported lower rates of mortality (13.3%) despite lower ARTcoverage (45%) and higher rates of tuberculosis (33.5%) [21]. As in our study, tuberculosis was reported to be the most common cause of hospitalization and death among HIV-positive patients in all these settings. These findings suggest that beyond ART-coverage, hospitalization outcomes among HIV-positive patients are likely influenced by other factors including morbid and co-morbid illnesses, the quality of prehospitalization HIV-care, early identification and management of HIV treatment failure and tuberculosis, available healthcare resources and the quality of inpatient care. We compared our findings with similar surveillance conducted at SLH in 2011 to 2012. These comparisons revealed an improvement in HIV testing both pre-and post-admission, and a significant decline in HIV prevalence, but a surprising stability in mortality among hospitalized HIV-positive patients despite increased ART use. Strikingly, tuberculosis-associated mortality among HIV-positive patients increased by 11% between the two surveillance periods. Our finding of stable inpatient mortality by study era despite national populationlevel estimates of an overall decline in AIDS-related mortality [11] and HIV/TB-related deaths [22] during the same period, suggests that ART availability has in itself been insufficient to curb the impact of HIV on in-hospital mortality, especially from tuberculosis. Further study is needed to understand whether regional TB-trends, the coverage of ART in specific regions and populations, or the timing of a patient's HIV diagnosis and ART initiation in relation to admission, may impact these findings.
Our study's strengths were its complete accounting for all patient outcomes during a 2-year period, including outcomes following transfer. Limitations were primarily related to our reliance on inpatient chart extraction, rather than patient interviews and outpatient clinical records. Results of pre-admission HIV testing, CD4 cell counts and viral loads were extracted from the inpatient chart and from the national electronic medical record which captures some but not all viral and immunologic testing occurring in the outpatient setting. The number of patients classified as having unknown HIV status on admission may have been overestimated if results of community-based HIV testing were not recorded in the inpatient chart or the electronic medical record. Many HIV-positive patients had CD4 counts and viral loads checked during their admission but in the absence of inpatient results, patients' most recent pre-admission results were used in our ART, anti-retroviral therapy. a Data presented as n (%) or mean AE standard deviation; b adjusted for age, sex and co-morbidities; c Patients were considered to have a known comorbidity if the medical recorded indicated a history of hypertension, diabetes, heart disease, lung disease, previous episodes of tuberculosis or any other chronic disease (see Table 1 for listed co-morbidities stratified by HIV status).
analysis and may not have reflected CD4 and viral load levels at the time of admission. We were also unable to ascertain prophylactic use of sulfamethoxazole-trimethoprim prior to hospitalization which may impact morbidity and mortality among hospitalized HIV-positive patients. Limitations in available diagnostics and documentation and unavailability of autopsy data may have also led to inaccuracies in classification of diagnoses. While tuberculosis was found to be a leading cause of admission and mortality, most diagnoses of tuberculosis were made based on clinical criteria and microbiological confirmation was available for only a minority of patients. Given the non-specific nature of the clinical signs and symptoms of tuberculosis, some patients who were classified as tuberculosis cases, especially among those who suffered mortality (and therefore may have not responded to anti-tuberculosis therapy) may have had different underlying diagnoses which could have been missed. Limitations in the accuracy and availability of tuberculosis diagnostic modalities present challenges to the clinical management of suspected TB cases in both high-resource and resource-limited settings. Previous studies in similar settings in Africa reported low rates of microbiological confirmation of suspected tuberculosis even when modalities such as TB culture are available [6,8].
Reported incidence risks of tuberculosis in this study nevertheless are comparable to those reported in other settings with similar burdens of HIV and tuberculosis. Finally, this study was designed to describe mortality of patients after admission to the medical department in our hospital and did not capture deaths which may have occurred prior to hospital presentation and in the emergency department prior to admission to the medical wards, nor can our results be used to infer mortality rates at a population level.

| CONCLUSIONS
Notwithstanding significant gains in the fight against HIV/ AIDS and the high ART-coverage rates achieved in Botswana, we found that HIV-positive patients continue to be disproportionately represented among hospital admissions and deaths. Tuberculosis continues to be the leading presumed cause of morbidity and mortality among HIV-positive patients. Our results are consistent with findings from other settings in sub-Saharan Africa demonstrating ongoing high rates of HIV-positive admissions and associated mortality (especially from tuberculosis) despite increased availability of ART [3][4][5][6][7][8]20,21], and contrast with experience from Europe and North America where HIV has become a condition largely managed in the outpatient setting [15][16][17][18]. These differences are likely explained by health system resources, differences in tuberculosis prevalence, different spectrum of opportunistic infections and socio-economic factors which disadvantage patients in developing-country settings. The high proportion of patients in our study who had not been previously tested for HIV, and the concentration of excess mortality among HIV-positive patients who are not on ART suggest that concentrated efforts to identify HIV-positive patients and initiate ART early in the outpatient setting may be essential to reduce inpatient mortality and improve patient outcomes in our setting. In addition, expansion of case-finding efforts, targeting suspected tuberculosis cases for early diagnosis and treatment and addressing the socio-economic factors that drive the tuberculosis epidemic may help curb the devastating impact this infection continues to have on morbidity and mortality in Botswana and similar developing-country settings throughout the world.