Risk Factors for Coronavirus Disease 2019 (COVID-19) Death in a Population Cohort Study from the Western Cape Province, South Africa

Received 12 July 2020; editorial decision 7 August 2020; published online 29 August 2020 All contributing authors are listed at the end of the paper. Correspondence: M.-A. Davies, Health Impact Assessment Directorate, Western Cape Government: Health and Centre for Infectious Disease Epidemiology and Research, School of Public Health and Family Medicine, University of Cape Town, Faculty of Health Sciences, Anzio Road, Observatory, 7925, Cape Town, South Africa (mary-ann.davies@uct.ac.za). Clinical Infectious Diseases 2021;XX(XX):0–0 © The Author(s) 2020. Published by Oxford University Press for the Infectious Diseases Society of America. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted reuse, distribution, and reproduction in any medium, provided the original work is properly cited. DOI: 10.1093/cid/ciaa1198 Risk Factors for Coronavirus Disease 2019 (COVID-19) Death in a Population Cohort Study from the Western Cape Province, South Africa

We conducted a cohort study using deidentified data from the Western Cape Provincial Health Data Centre (WCPHDC) of public-sector patients aged ≥20 years with documented sex and not known to have died before 1 March 2020 (before the first diagnosed COVID-19 case in South Africa, and several weeks before the first documented COVID-19 death) and included all follow-up through 9 June 2020. The outcome was COVID-19-associated death. Our main analysis examined the risk of COVID-19 death in the general population, so all patients were included irrespective of severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) testing. The study was approved by the University of Cape Town and Stellenbosch University Health Research Ethics Committees and the Western Cape Province Department of Health. Individual informed consent requirement was waived for this secondary analysis of deidentified data.

Study Population and Data Sources
The Western Cape has nearly 7 million inhabitants, of whom ~520 000 are PLWH, with >90% of them dependent on publicsector health services. The WCPHDC has been described in detail [18]. Briefly, WCPHDC consolidates administrative, laboratory, and pharmacy data from routine electronic clinical information systems used in all public-sector health facilities, with linkage through a unique identifier. Multiple data sources are triangulated to enumerate health conditions such as diabetes mellitus, hypertension, tuberculosis, and HIV, with a label of high-or moderate-certainty evidence assigned for each inferred condition (Supplementary Table 1). High-certainty evidence of HIV comprises a positive HIV diagnostic test, HIV RNA test, triple antiretroviral therapy (ART), and/or registration in the HIV disease management system; moderate certainty is assigned for those with only a CD4 count measure, 2 antiretroviral drugs prescribed (previously used for vertical HIV transmission prevention), and/or an International Classification of Disease, Tenth Revision, diagnosis code of HIV. HIV testing coverage is high, as >90% of PLWH know their HIV diagnosis [19]. High-certainty evidence of tuberculosis comprised laboratory evidence of Mycobacterium tuberculosis infection (any anatomical site, using Xpert RIF/MTB, microscopy, and culture), registration on the electronic tuberculosis registers, combination tuberculosis treatment, and/or admission to a tuberculosis hospital. Comorbidities were based on high-or moderatecertainty evidence but were restricted to high-certainty evidence in sensitivity analyses. The virologic, immunologic, and ART statuses of PLWH on 1 March 2020 were categorized, based on the most recent measures, as "confirmed virologically suppressed on ART" (HIV RNA < 1000 copies/ml in the last 15 months and ART dispensed in the last 6 months), "likely virologically suppressed on ART" (HIV RNA < 1000 copies/ml 15-24 months previously or HIV RNA < 1000 copies/ml > 24 months previously if ART dispensed in the last 6 months), "viremic or immunosuppressed" (HIV RNA > 1000 copies/ml in the last 15 months or CD4 count < 200 cells/µl within 18 months before March 2020), or "unknown. " Until January 2020, adult first-line ART was TDF, emtricitabine/lamivudine, and efavirenz, with abacavir replacing TDF for patients with kidney disease; zidovudine plus an emtricitabine/protease inhibitor were used for second-line ART for most patients. Dolutegravir was introduced in first-and second-line ART in January 2020. Diabetic control was categorized according to glycosylated hemoglobin (HbA1c) measurements within the last 2 years, with <7% indicating diabetes was controlled, 7-8.9% indicating diabetes was poorly controlled, and ≥9% indicating diabetes was uncontrolled.

COVID-19 Diagnosis
All COVID-19 diagnoses were based on a positive SARS-CoV-2 polymerase chain reaction test. Testing was available for all patients with COVID-19 symptoms until 1 June 2020; thereafter, public-sector laboratory testing was restricted to patients requiring admission, aged >55 years, or with comorbidities, due to the temporary limited testing capacity. Hospital admissions and all deaths in SARS-CoV-2-positive cases are recorded and reviewed daily.

Statistical Analysis
We used Cox proportional hazards models, adjusted for age, sex, and other comorbidities, to examine the associations between HIV, tuberculosis, and COVID-19 death among (1) all publicsector patients with ≥1 health visit in the 3 years before 1 March 2020 (considered "active patients"); (2) laboratory-diagnosed COVID-19 cases; and (3) hospitalized COVID-19 cases. We adjusted for location within Cape Town versus the rest of the province, and by the subdistrict of residence within Cape Town, to account for geographical variation in infection rates and as a proxy for socioeconomic status. Patients were censored on their date of death if deceased without a COVID-19 diagnosis, or on 9 June 2020, whichever was earliest. The database closure was 7 days later to allow for death reporting delays. For the analysis of COVID-19 deaths in laboratory-diagnosed cases, we included cases diagnosed before 1 June 2020, when testing was available for all patients with COVID-19 symptoms, but included all patients diagnosed by 9 June 2020 in the sensitivity analysis. The proportional-hazard assumption was assessed with Schoenfeld residuals [20]. All analyses were conducted using Stata 15.1.
We also calculated the standardized mortality ratio (SMR) of the actual number of COVID-19 deaths in PLWH versus the number that would be expected if PLWH had the same risk of COVID-19 death as people living without HIV of the same age and sex. We used data on the age, sex, and HIV status of all COVID-19 deaths (public and private sector) and the Thembisa Western Cape HIV model to estimate the Western Cape population size and HIV prevalence, by age and sex, in 2020 [21]. We calculated 95% confidence intervals (CI) for the SMR using 1000 bootstrap replications (see Supplementary Appendix 2).
Since individual socioeconomic status and some comorbidities are not recorded in WCPHDC, we calculated E-values to determine the minimum strength of association that an unmeasured confounder (eg, raised body mass index [BMI] or socioeconomic status) would need to have with HIV/ tuberculosis and COVID-19 death to fully account for any association between HIV/tuberculosis and COVID-19 death [22]. We conducted a quantitative bias analysis to assess the impact of potential confounding by obesity on an association between HIV and COVID-19 death.

Patients With and Without HIV
Although the proportion of PLWH was similar among surviving and deceased COVID-19 cases, a greater proportion of COVID-19 deaths were in patients aged <50 years in those living with versus without HIV (39% vs 13%, respectively; Table 2). A substantial proportion of PLWH who died from COVID-19 had diabetes (50%) and hypertension (42%); however, these conditions were more common in deceased people living without HIV (62% for each condition). Current and previous tuberculosis were more frequent in PLWH, irrespective of COVID-19 survival status, with 14% and 37% of COVID-19 deceased cases in PLWH having current and/or previous tuberculosis, respectively.

COVID-19 Death in All Public-sector Patients
Among all public-sector patients, the probability of COVID-19 death by 100 days after 1 March 2020 was 180/million (95% CI, 167-196). COVID-19 death was associated with male sex, increasing age, diabetes (with a higher risk with elevated HbA1c), hypertension, and chronic kidney disease (Table 3). Current tuberculosis was associated with an increased hazard of COVID-19 death (adjusted hazard ratio [aHR], 2.70; 95% CI, 1.81-4.04), with a smaller increase for having previous tuberculosis (aHR, 1.51; 95% CI, 1.18-1.93). The increased hazard of COVID-19 death associated with current tuberculosis was present for both microbiologically confirmed and unconfirmed tuberculosis and for rifampicin-sensitive and -resistant disease during intensive phase treatment (Supplementary Table 3).
After adjusting for age, sex, and other comorbidities, HIV was associated with increased COVID-19 mortality (aHR, 2.14; 95% CI, 1.70-2.70), and this association was similar irrespective of viremia or immunosuppression prior to the COVID-19 episode. However, few patients were viremic or immunosuppressed, as reflected in the wide CIs for the hazard ratios in different groups. Associations with most comorbidities increased when restricting to those with high-certainty comorbidity evidence (Supplementary Table 4). The associations of most comorbidities with COVID-19 death were attenuated when restricting to patients with ≥1 medical visit/year in the last 3 years, as these patients were more likely to have comorbidities warranting regular visits; however, HIV remained significantly associated with COVID-19 death (Supplementary Table 4). Among all public-sector adults, 9.8% of COVID-19 deaths were attributable to HIV (95% CI, 6.2-13.3), 2.6% (95% CI, 1.0-4.2) to current tuberculosis, and 4.7% (95% CI, 1.5-7.8) to previous tuberculosis.

Potential Bias from Unmeasured Confounding
To assess whether the association between HIV or tuberculosis and COVID-19 mortality could be due to residual unmeasured confounding-for example, by socioeconomic status or unrecorded comorbidities-we calculated the E-value for an unmeasured confounder. For HIV, the E-value for the analysis among all public-sector patients was 3.70 (and 2.79 for the lower bound of the CI), suggesting that only a strong association between HIV and a confounder (eg, socioeconomic status) and between the confounder and COVID-19 death would account for all of the observed association between HIV and COVID-19 death. The effect of HIV on COVID-19 death was similar when restricted to the poorest subdistrict, with the highest HIV prevalence, in Cape Town [19]. Corresponding E-values for current and previous tuberculosis were 4.84 (3.02 for the lower bound of CI) and 2.39 (1.64 for the lower bound of CI), respectively. A quantitative bias analysis showed that the HIV-associated increased risk of COVID-19 death was unlikely due to confounding by raised BMI (Supplementary Appendix 4).

Standardized Mortality Ratio
Among all laboratory-diagnosed COVID-19 cases, there were 135 deaths among an estimated ~520 000 PLWH in the province (260 deaths/million) and 786 deaths among 6.36 million people without HIV (124 deaths/million). The SMR for COVID-19    with suppressed VLs on ART [23,24]. Comparisons by HIV status of hospitalized COVID-19 cases in New York and London have not shown differences in mortality risks [25][26][27]. However, the absence of an increased mortality risk in hospitalized patients with comorbidities may be explained by selection bias: risk factors for COVID-19 death may be attenuated by restricting data to the subset of hospitalized patients already at high mortality risk [28]. It is therefore expected that the increased risk of death associated with all comorbidities in our analysis was progressively attenuated when restricting to cases (people with sufficiently severe symptoms to be tested) and hospitalized patients. Similar to our findings, several studies have reported a high prevalence of comorbidities among PLWH with severe COVID-19 [3,6,7]. The high prevalence of comorbidities in deceased PLWH suggests that the effect of HIV may at least partly be due to an increased risk of comorbidities at younger ages [2,7], including those not recorded in WCPHDC, such as cardiovascular disease. Persistent immune dysfunction may also be important in severe COVID-19 despite viral suppression; the hazard ratio point estimates for association with COVID-19 death were greater in immunosuppressed or viremic PLWH, although the numbers of these patients with COVID-19 were small, with wide CIs. Further, a CD4 count < 200 cells/µl during admission was associated with COVID-19 death. While this may partly be due to the well-described lymphopenia in severe COVID-19, which is prognostic of poor outcomes, about half of patients with low CD4 counts during admission were either new HIV diagnoses or had previous immunosuppression, viremia, or no recent ART [10]. Among COVID-19 cases in PLWH on ART, the receipt of TDF (vs other therapies) was associated with reduced COVID-19 mortality. However, this association is likely to be overestimated; in South Africa, only patients on second-line ART or with poor renal function would not be on TDF, and both of these factors may themselves increase mortality. Nonetheless, the association remained when adjusting for kidney disease, viral suppression, and ART duration, and concurs with results from a recently published cohort of PLWH on ART from Spain [13]. We found both current and previous tuberculosis to be associated with COVID-19 death, but since current tuberculosis itself causes death, in the absence of autopsy evidence it is difficult to disentangle the effects of COVID-19 versus tuberculosis disease on mortality [17].
In our study, the overall high prevalence of diabetes in people with and without HIV, high proportion with poor glycemic control, and very elevated risks for COVID-19 death for diabetics, compared to data reported from other countries, are concerning [8]. Diabetes is often diagnosed late and/or untreated or poorly controlled in resource-limited settings, and the resulting microvascular disease, even in people with good current diabetic control, may increase COVID-19 mortality [29].
To our knowledge, this is the largest report on SARS-CoV-2 from Africa, the largest report on HIV and tuberculosis coinfected patients, and the first comparison of COVID-19 outcomes in patients with and without tuberculosis. Strengths include the study size using population-level data, laboratoryconfirmed SARS-CoV-2 diagnoses in all COVID-19 cases, and the inclusion of hospitalized and nonhospitalized cases and deaths, as well as modeling the independent associations of HIV and tuberculosis with COVID-19 death. While the population analysis approach is robust to selection bias associated with cases and hospitalized patients only, it may overestimate associations between comorbidities and COVID-19 death if those with comorbidities live in areas with higher transmission or have closer follow-up and are more likely to be diagnosed with COVID-19. Nonetheless the coherence of associations found when analyzing the population cohort and SMR, diagnosed cases and hospitalized patients suggest that the population findings are unlikelyto be solely due to different probabilities of encountering SARS-CoV2 or being diagnosed once infected. Furthermore, an adjustment for subdistrict of residence should address geographic differences in transmission probability. Being an observational study, limitations include the underascertainment of comorbidities in routine administrative data; a lack of data on other potential risk factors, including socioeconomic status, smoking, and BMI; possible underascertainment of all COVID-19 cases and deaths; and potential misclassification of some incidental deaths in patients positive for SARS-CoV-2 as related to COVID-19, although almost all deaths occurred in hospitalized patients with clinical COVID-19. Further, we were unable to systematically exclude other opportunistic infections as contributors to COVID-19 mortality, as investigation for these causes varied by facility, clinical presentation, and time in hospital. Relatively large numbers of PLWH had no recent VL or CD4 count results, limiting our ability to distinguish outcomes for different strata of these measures. In particular, patients with no recent information on disease control (eg, HIV RNA or HBA1c) may have less contact with health services and not reside permanently in the province, with underascertainment of outcomes.

CONCLUSION
While our findings of increased COVID-19 mortality risks in those living with HIV or tuberculosis may overestimate associations of these conditions with COVID-19 death due to residual confounding, PLWH and/or those with tuberculosis should nonetheless be considered high-risk groups for COVID-19 management, especially if they have other comorbidities.

Supplementary Data
Supplementary materials are available at Clinical Infectious Diseases online. Consisting of data provided by the authors to benefit the reader, the posted materials are not copyedited and are the sole responsibility of the authors, so questions or comments should be addressed to the corresponding author.