A national cohort study of COVID-19 in-hospital mortality in South Africa: the intersection of communicable and non-communicable chronic diseases in a high HIV prevalence setting

Study setting and data sources

South Africa is administratively divided into nine provinces. South Africa has a dual health system with a publicly funded district health system, that serves approximately 84% of the population, and a private health system largely funded by private health insurance schemes. (24)

In the absence of existing national COVID-19 hospital surveillance systems, the National Institute for Communicable Diseases (NICD) established DATCOV as a national surveillance system for COVID-19 hospitalizations on 1 April 2020. (25) DATCOV was adopted for national implementation following South African government endorsement on 15 July 2020, and by October 2021 full coverage of all hospitals had been achieved. By 27 March 2021, a total of 393 public-sector hospitals and 251 private-sector hospitals have reported COVID-19 hospitalisations on DATCOV. Further details about the DATCOV surveillance system including its implementation, data management and data quality assurance are described in supplementary methods (page 1, Supplementary Material).

Potential risk factors and covariates

Age, sex, race, and comorbidities (hypertension, diabetes, chronic cardiac disease, asthma, other chronic respiratory disease, chronic renal disease, malignancy in the past five years, HIV and past and current TB) were considered as potential risk factors for COVID-19 in-hospital mortality. Data on comorbidities including HIV, ART status, and CD4 and viral load (VL) done in the past 12 months, were submitted to DATCOV by the hospital, based on information contained in the patient’s written hospital record, and were not independently verified by the DATCOV team and neither were laboratory records accessed to obtain this data. Date of most recent CD4 or VL was not documented. The level of virological control or immunosuppression was assessed based on available VL result within the past year and categorised as virologically suppressed (HIV-RNA<1,000 copies/ml) or as viraemic (HIV-RNA≥1,000 copies/ml); and immune reconstituted (CD4 count≥200 cells/μl) or immunosuppressed (CD4 count<200 cells/μl). (26) CD4 count is no longer routinely performed on all patients, as per South African HIV care guidelines. (27)

In-hospital mortality was defined as a death related to COVID-19 that occurred during the hospital stay and excluded deaths that occurred due to other causes or after discharge from hospital. Because the main outcome of the study was in-hospital mortality, all analyses were implemented among SARS-Cov-2 hospitalised patients with a known in-hospital outcome (i.e. discharged alive or died) at the time of data extraction (27 March 2021).

Statistical analysis

We analysed DATCOV data from 5 March 2020 to 27 March 2021. For the main analysis, to account for incomplete or missing data on selected variables, we used multivariate imputation by chained equation (MICE) and generated ten complete imputed datasets that were used for subsequent analyses. Variables analysed using MICE included sex, race, month of admission and comorbidities including HIV and TB infection, hypertension, diabetes, asthma, malignancy, and chronic pulmonary, cardiac and renal diseases. ART, HIV viral load and CD4 counts were also incomplete and were conditionally imputed only among HIV-positive patients (either with observed or imputed HIV status). Complete variables included in the imputation process were age, province, health sector (i.e. public or private) and in-hospital outcome (i.e. discharged alive or died). Descriptive statistics such as frequencies and percentages were used for categorical variables, and continuous variables were expressed as mean with standard deviation (SD) or median and interquartile range (IQR) on the imputed datasets.

We implemented post-imputation random effect (on admission facility) multivariable logistic regression models to: (i) compare PLWH and individuals who were HIV-uninfected; (ii) determine the factors associated with in-hospital mortality, including ART, HIV viral load and CD4 count among HIV-positive individuals; and (iii) evaluate the combined effect of multiple non-HIV comorbidities on COVID-19 in hospital mortality. A combined comorbidity variable categorised absence, one, two, and three or more comorbidities among those that were individually significantly associated with in-hospital mortality in model (ii) above. We evaluated the effect of multiple comorbidities among any patient and among PLWH and HIV-uninfected patients separately through stratification by HIV infection status. In addition, we assessed potential differential effect (through the inclusion of an interaction term) of multiple or individual comorbidities on in-hospital mortality between PLWH and HIV-uninfected individuals.

A random effect on admission facility was included for all analyses to account for potential differences in the service population and the quality of care at each facility. For each multivariable model we assessed all variables that were significant at p<0.2 on univariate analysis (to evaluate lack of significance at the univariate analysis after adjusting for potential confounders) and dropped non-significant factors (p≥0.05) with manual backward elimination. Pairwise interactions were assessed by inclusion of product terms for all variables remaining in the final multivariable additive model. We implemented several sensitivity analyses separately analysing risk factors for mortality in the public and private sector using post-imputation random effect multivariable logistic regression models. We also reported the univariate and multivariate association of all covariates evaluated in the analyses described above to the main outcomes (i.e. HIV infection or in-hospital mortality) using non-imputed data. The statistical analysis was implemented using Stata 15 (Stata Corp®, College Station, Texas, USA). We followed STROBE guideline recommendations.

Ethical considerations

The Human Research Ethics Committee (Medical), University of the Witwatersrand, approved the project protocol as part of a national surveillance program (M160667). In South Africa, surveillance for notifiable medical conditions such as COVID-19 requires health facilities to submit data on all cases of emerging pathogens to national authorities. As such, individual consent for inclusion of their data to DATCOV is waivered. All personal identifying information was de-linked for our analysis and stored in a secure server.

Role of the funding source

DATCOV as a national surveillance system, is funded by the NICD and the South African National Government. No additional funding was obtained towards the completion of this analysis and the development of this manuscript. The funders of the study had no role in study design, data collection, data analysis, data interpretation, or writing of the report. The corresponding author had full access to all the data in the study and had final responsibility for the decision to submit for publication.

Study population

From 5 March 2020 to 27 March 2021, 229,154 individuals were hospitalised with laboratory-confirmed SARS-Cov-2 and reported to DATCOV. Of these, 219,265 met the inclusion criteria (Figure S1, page 4 Supplementary Materials). The percent of missing data for the imputed variables varied and was highest for race (33.5%), and for HIV-specific variables including HIV diagnosis (30.8%), being on ART (41.4%), viral load (87.6%) and CD4 count (79.9%) (Table S1, page 5 Supplementary Materials).

South Africa experienced a first wave of SARS-Cov-2 infection, peaking at 10,086 admissions per week in July 2020, and a second wave peaking at 17,976 admissions per week in January 2021 (Figure S2, page 7 Supplementary Materials).

The median age of SARS-Cov-2 hospitalised patients was 54 years (interquartile range [IQR] 40 – 66) and 55.7% of patients (121,937/218,827) were female (Table S1, page 5 Supplementary Materials). Race information was available for 145,876 (66.5%) patients, of whom 114,571 (78.5%) were black. The public sector accounted for 51.9% (113,856/219,265) of hospitalisations reported to DATCOV. Most admissions (176,272/219,265 [80.4%]) were recorded by hospitals in four provinces, namely Western Cape, Gauteng, KwaZulu-Natal and Eastern Cape.

Following multiple imputation, the estimated proportion of individuals reporting at least one comorbidity was 60.5% (95%CI: 59.8%-61.2%). The most prevalent comorbidities were hypertension (38.6%), diabetes (29.0%), and HIV (11.7%). The prevalence of non-communicable comorbidities increased with age, while HIV and TB were most prevalent in individuals aged 20–59 years (Figure 3).

Factors associated with HIV infection

Of 151,779 (69.2%) hospitalised patients with available data on HIV, 13,793 (9.1%) were PLWH (Table 1). The HIV prevalence was 20.4% in the public sector and was 2.2% in the private sector (age-specific HIV prevalence for public and private sector presented in Table S2, page 8 Supplementary Materials). A subset of the 13,793 PLWH had ART (8078 [58.6%]), viral load (1,716 [12.4%]) and CD4 (2,770 [20.1%]) information available. Of these, 92.7% (7,484/8,078) were receiving ART, 25.8% (443/1,716) were viraemic (HIV-RNA≥1000 copies/ml) and 39.0% (1,080/2,770) were immunosuppressed (CD4 count<200 cells/μl) based on last available test.

Compared to HIV-uninfected SARS-CoV-2 hospitalised patients, PLWH were more likely to be aged 20-39 years, 40-59 years, and 60-69 years, compared to <20 years; black or mixed race compared to white race; admitted in the public health sector; to have comorbid chronic cardiac disease, renal disease, malignancy, current TB, past TB and previous and current TB, and to die in hospital. PLWH were less likely to be aged 70-79 years and ≥80 years; male; of Indian ancestry; and to have comorbid hypertension, diabetes, and chronic pulmonary disease or asthma. HIV prevalence of COVID-19 patients varied with lower prevalence in Eastern Cape and Gauteng province, and higher prevalence in KwaZulu-Natal and Mpumalanga provinces (Table 1 and Figure 2).

In-hospital COVID-19-associated mortality

Of the 219,265 hospitalised patients with recorded outcomes, 168,228 (76.7%) were discharged alive and 51,037 (23.3%) died. The unadjusted in-hospital case-fatality ratio (CFR) for PLWH was (3,407/13,793 [24.7%]) compared to (30,697/137,986 [22.3%]) for HIV-uninfected individuals.

Factors statistically associated with in-hospital COVID-19 mortality were increasing age 20-39 years, 40-59 years, 60-69 years, 70-79 years and ≥80 years, compared to <20 years; male sex; black, mixed race, and Indian ancestry compared to white race; having comorbid hypertension, diabetes, chronic cardiac disease, chronic renal disease, malignancy, HIV, past TB, current TB or both past and current TB, and being admitted in the public health sector. In-hospital mortality increased each month of the epidemic to the peak of the first wave in July then decreased between waves and increased again to the peak of the second wave in January. In-hospital mortality was significantly higher in five provinces, Eastern Cape, Free State, KwaZulu-Natal, Limpopo and Mpumalanga, compared to the Western Cape (Table 2 and Figure 3). The un-imputed multivariate analysis showed similar findings (Table 2).

Sensitivity analysis of risk factors for mortality in public and private sector (Table S3, page 9 Supplementary Materials), confirm increased association of mortality for age, male sex, race and comorbidities. Divergent findings in the health sector analysis included (a) hypertension was associated with mortality in the private but not the public sector; (b) a stronger effect of age, sex, race and comorbidities in the private sector; and (c) a stronger effect of province and month of admission in the public sector, with a significant difference in mortality observed only at the peak of the second wave in the private sector.

Association of in-hospital COVID-19 mortality with immune suppression among PLWH

When adjusting for age, sex, race, health sector, province, month of admission, NCDs, and past or active TB, compared to HIV-negative individuals, PLWH not on ART [aOR 1.45, 95% CI: 1.22-1.72] were more likely to die in-hospital compared to PLWH on ART; PLWH with history of immune suppression (CD4<200 cells/μl) [aOR 2.31, 95% CI: 1.82-2.93] were more likely to die in-hospital compared to PLWH with CD4≥200; and PLWH with history of VL≥1,000 [aOR 1.55, 95% CI: 1.20-2.01) were more likely to die in-hospital compared to PLWH with VL<1,000. (Table 3).

Effect of multiple comorbidities on COVID-19 mortality

Among PLWH, the prevalence of other comorbidities was 29.2% compared to 30.8% among HIV-uninfected individuals. In a multivariable model adjusting for age, sex, race, HIV (for the non-stratified model on HIV status only), health sector, province, and month of admission, there was an increasing odds of in-hospital mortality for individuals with multiple non-HIV comorbidities, irrespective of HIV status among PLWH and HIV-uninfected individuals (Table 4). There was no statistical evidence of interaction between the presence of other multiple comorbidities and HIV infection status on in-hospital mortality (Table 4). No interaction was found for individual co-morbidities and HIV infection status on in-hospital mortality (Table S4, page 11 Supplementary Materials).

Strengths and Limitations

The main strengths of this analysis are that the surveillance system is nationally representative across all provinces, has 100% coverage of hospitals in the public and private health sector in South Africa and contains close to 220,000 SARS-Cov-2 hospital admissions.

There were several limitations related to the study population and data completeness.

Patients with SARS-CoV-2 reported to DATCOV either had COVID-19 symptoms, were admitted for isolation, developed nosocomial COVID-19 or tested positive SARS-Cov-2 incidentally during admission for unrelated reasons. We were not able to exclude patients admitted for isolation and those with incidental findings due to incomplete data on reason for admission to DATCOV, and this may introduce bias, although likely due to high pressure for hospital beds, such individuals would have made up a small percent of included cases.

The hospitals reported a small number of 219 deaths due to cause other than COVID-19, however we have little information on the process used by the hospital for classification of cause of death other than that it is done by the attending clinician. It is possible that other deaths reported to DATCOV have not undergone similar review and may have been misclassified as COVID-19 deaths. It is also possible that some COVID-19 admissions have not been reported to the surveillance system.

Patients who were transferred to other hospitals and had no further records of admission, and those still in hospital at the time of the analysis were excluded. We do not know whether the CFR among excluded patients would differ than that of included patients possibly introducing biases. However, given the small number of the excluded patients (1.8%) the effect of biases would be limited.

There is an over-representation of private sector hospitalisations due to a lower threshold for hospitalisation in private hospitals, resulting in similar proportions of SARS-CoV-2 cases hospitalised in the public and private sector even though they serve 84% and 16% of the population respectively. This may explain why the HIV prevalence in the overall cohort (11.7%) was lower than the population prevalence of HIV in South Africa (14%). (11) However, HIV prevalence among hospitalised individuals in the public sector aged 20-59 years was similar to population HIV prevalence in this age group (Table S2, page 6 Supplementary Materials).

Data quality in a surveillance system is dependent on the information submitted by healthcare institutions. Data on comorbidities were submitted to DATCOV by the hospital based on information contained in the patient’s written or electronic hospital record, and were not independently verified as there are no systematic information systems that would verify pre-existing comorbidities. We used multiple imputation to address missing data; however, the validity of the imputed data relies on the assumption that data were missing at random. Fields with the highest proportion of incomplete data included, race (33.5%) and comorbidities (25.5% - 32.2%). The level of control of NCDs such as diabetes, using objective measures like HbA1C, was not consistently reported. Other data have shown that individuals with poorly controlled NCD were at greater risk for mortality. (39)

The most recent CD4 and VL results in the past year were submitted by the hospital based on information contained in the patient’s hospital record and not obtained from laboratory information systems. These fields had a high degree of missing data, which is an important limitation given that this analysis focuses on outcomes in PLWH. Also, the hospital did not record the date of the CD4 and VL test, so we are not able to calculate the median time between the test and COVID-19 admission, or to be certain that the status of immunosuppression or viraemia had not changed, which may also introduce measurement bias. It must be noted that DATCOV is a new surveillance system that has not yet been developed to link to other data sources and allow, for example, linkage of laboratory records to the hospitalisation record.