Associations between HIV infection and clinical spectrum of COVID-19: a population level analysis based on US National COVID Cohort Collaborative (N3C) data
Abstract
Background
Evidence of whether people living with HIV are at elevated risk of adverse COVID-19 outcomes is inconclusive. We aimed to investigate this association using the population-based National COVID Cohort Collaborative (N3C) data in the USA.
Methods
We included all adult (aged ≥18 years) COVID-19 cases with any health-care encounter from 54 clinical sites in the USA, with data being deposited into the N3C. The outcomes were COVID-19 disease severity, hospitalisation, and mortality. Encounters in the same health-care system beginning on or after January 1, 2018, were also included to provide information about pre-existing health conditions (eg, comorbidities). Logistic regression models were employed to estimate the association of HIV infection and HIV markers (CD4 cell count, viral load) with hospitalisation, mortality, and clinical severity of COVID-19 (multinomial). The models were initially adjusted for demographic characteristics, then subsequently adjusted for smoking, obesity, and a broad range of comorbidities. Interaction terms were added to assess moderation effects by demographic characteristics.
Findings
In the harmonised N3C data release set from Jan 1, 2020, to May 8, 2021, there were 1 436 622 adult COVID-19 cases, of these, 13 170 individuals had HIV infection. A total of 26 130 COVID-19 related deaths occurred, with 445 among people with HIV. After adjusting for all the covariates, people with HIV had higher odds of COVID-19 death (adjusted odds ratio 1·29, 95% CI 1·16–1·44) and hospitalisation (1·20, 1·15–1·26), but lower odds of mild or moderate COVID-19 (0·61, 0·59–0·64) than people without HIV. Interaction terms revealed that the elevated odds were higher among older age groups, male, Black, African American, Hispanic, or Latinx adults. A lower CD4 cell count (<200 cells per μL) was associated with all the adverse COVID-19 outcomes, while viral suppression was only associated with reduced hospitalisation.
Interpretation
Given the COVID-19 pandemic's exacerbating effects on health inequities, public health and clinical communities must strengthen services and support to prevent aggravated COVID-19 outcomes among people with HIV, particularly for those with pronounced immunodeficiency.
Funding
National Center for Advancing Translational Sciences, National Institute of Allergy and Infectious Diseases, National Institutes of Health, USA.
Introduction
As of Oct 7, 2021, SARS-CoV-2, which causes COVID-19, has been confirmed to have infected over 236 million people and has caused more than 4·8 million deaths worldwide.1 Since the first confirmed case of COVID-19 in the USA, the countrywide COVID-19 outbreak has surged quickly, making it one of the countries hardest hit by the pandemic.1 As the pandemic surges in the USA, it is important to identify patients at elevated risk of developing severe symptoms to inform clinical management decisions. Older age and presence of comorbidities are recognised as factors that increase the severity of COVID-19.2 Patients who have malignant disease or solid-organ transplants have overall poorer outcomes of COVID-19,3 but the evidence is less clear for people with other types of immunocompromising conditions, including people living with HIV.4
Existing evidence of the association between HIV infection and COVID-19 outcomes is mixed. Throughout the COVID-19 pandemic, data have been limited and have largely consisted of case reports or case series.5 According to a systematic review, COVID-19 prevalence among people with HIV was comparable to that in the general population although there were occasional reports of atypical, but no more severe, disease course relative to people without HIV.5 Later on, emerging data from observational cohort studies showed similar findings;6, 7, 8, 9, 10 however, most of these studies were restricted to hospitalised patients. By contrast, several large population-based studies have found conflicting results. Large-scale studies conducted in the UK and South Africa suggested that people with HIV had a higher risk (more than double) of COVID-19 mortality than people without HIV, although different factors were adjusted in the different studies.11, 12 A prospective study of patients hospitalised with COVID-19 showed an increased 28-day mortality in people with HIV after adjusting for age.13 One study in New York, USA, reported a standardised in-hospital mortality ratio of 1·23 for HIV patients.14 Prognosis, according to HIV immune status, is also difficult to evaluate because most studies from Europe and the USA reported on individuals with overall high CD4 cell counts.15 In the largest published cohorts, the potentially higher risk for poorer COVID-19 outcomes were observed in people with HIV with lower CD4 cell counts.8, 16 Investigating whether people with virologically controlled HIV who are clinically stable will have a greater risk for COVID-19 complications than people without HIV is of great clinical significance.
Nevertheless, the evidence linking HIV status and COVID-19 outcomes is still scarce and some knowledge gaps remain. Several studies were based on only a small number of cases; some either did not have direct comparative data for people without HIV and HIV markers,5 or focused only on hospitalised patients.7, 8, 9 A large, multicentre, representative clinical dataset is needed to provide timely and robust risk assessment and thereby inform prioritisation of critical therapies, vaccination, and targeted intervention. Using the US National COVID Cohort Collaborative (N3C) data, this study aims to understand the role of HIV infection and levels of immunity affecting the COVID-19 clinical outcomes (ie, disease severity, hospitalisation, and mortality).
Methods
Study design and population
The N3C Enclave, sponsored by multiple institutes of the US National Institutes of Health,17 is the largest cohort of US COVID-19 cases and representative controls to date. The N3C is a large, multicentre dataset updated on an ongoing basis that harmonises electronic health records data for all individuals with laboratory confirmed, suspected, or possible COVID-19 during any encounter after Jan 1, 2020.18 Control cases are those individuals who have tested negative for COVID-19, and are demographically matched on age group, sex, race, and ethnicity within the same submitting health-care system at a case to control ratio of 1 to 2.18 All patients in the N3C Enclave include historical data within the same health-care system as of Jan 1, 2018, which provides information about pre-existing health conditions (eg, comorbidities) and other medical history (look back data).19 We included all adult (aged ≥18 years) COVID-19 cases from 54 clinical sites across the USA with data being deposited into the N3C and harmonised into a data release set from Jan 1, 2020, to May 8, 2021. The data ingestion and harmonisation process are described in the appendix (p 2). We excluded people with missing age, race, and ethnicity data because absence of data on these key variables probably indicated that poor data quality for these records.
The N3C data transfer to the National Center for Advancing Translational Sciences (NCATS) was done under a Johns Hopkins University Reliance Protocol (IRB00249128) or individual site agreements with the US National Institutes of Health (NIH). An institutional data use agreement was signed between the University of South Carolina and NCATS N3C Data Enclave. The N3C Data Enclave is managed under the authority of the NIH. The N3C Data Enclave is approved under the authority of the NIH Review Board. The analyses reported in this Article were approved separately by the institutional review board of University of South Carolina (Pro00107403) with data access. The NIH's N3C data access committee approved the data use request for this project (RP-E72986).
Procedures
The N3C phenotype18 is designed to be inclusive of any diagnosis codes, procedure codes, laboratory tests, or combination thereof that might be indicative of COVID-19 (eg, Centers for Disease Control and Prevention coding guidance20). N3C includes patients with any encounter after Jan 1, 2020, who have either one or more of a set of a priori-defined SARS-CoV-2 laboratory tests with a positive result; or one or more strong positive diagnostic codes from the International Classification of Diseases (ICD) 10 or SNOMED tables; or two or more weak positive diagnostic codes from the ICD-10 or SNOMED tables. The cohort definition is publicly available on GitHub.18
N3C harmonises data across four clinical data models (ACT Network, PCORnet, Observational Health Data Sciences and Informatics, and TriNetX) and provides a unified analytical platform in which data are encoded by use of the Observational Medical Outcomes Partnership (OMOP) version 5.3.1.21 The concept sets in OMOP 22, 23 are a list of concepts from the standardised vocabulary that, taken together, describe a topic of interest for a study, were used to identify each clinical concept (eg, laboratory measure, conditions, or medication). Data domains extracted by N3C include demographics, encounter details, medications, diagnoses, procedures, vital signs, laboratory results, procedures, and social history. Specific variables included in each domain are listed in each model's documentation (ie, tables). Both concept sets and tables were used to define variables of interest.
A total of 13 170 people with HIV were identified by use of N3C concept sets and codes in the phenotype template (appendix p 2), which mapped to various domain tables, including any HIV diagnosis code (ICD-10 codes, SNOMED codes) in the condition occurrence table, HIV laboratory tests (LOINC codes) in the measurement table, and HIV drug exposure in the drug exposure table. Patients who met at least one of these inclusion criteria were counted as people with HIV in our study. Within the population of people with HIV, the most recent value of CD4 cell count and viral load before initial COVID-19 diagnosis (but during the preceding 18 months) was retrieved for analysis from laboratory tests (LOINC codes) in the measurement table. The absolute CD4 count was categorised into less than 200, 200–500, and more than 500 cells per μL. HIV viral load was classified into less than 200 copies per mL (virally suppressed) and 200 or more copies per mL (unsuppressed).
COVID-19 hospitalisation in the current study was identified by case insensitive string matching “inpatient visit” or “inpatient critical care facility” or “emergency room and inpatient visit” in the Selected Critical Visit table. The Selected Critical Visit table was created by the N3C Consortium to document a COVID-19 related medical encounter. Specifically, N3C defined a single index encounter for each laboratory-confirmed COVID-19 positive patient by selecting encounters that start up to 30 days before or 7 days after the positive test result, or a positive test result during the visit.19 When multiple encounters met these criteria, the N3C Consortium broke ties by preferentially selecting the encounter in which the most severe outcome was observed, then the longest visit, and finally the most recent visit.19
Clinical severity was classified with the Clinical Progression Scale (CPS) established by WHO for COVID-19 clinical research.24 On the basis of WHO criteria, N3C placed patients into strata defined by the maximum clinical severity from selected critical visits:19 unaffected (ie, no laboratory test, laboratory test negative, or suspected COVID-19 with laboratory tests, but identified by other diagnosis codes or procedure codes), mild (outpatient, WHO severity 1–3); mild emergency department (outpatient with emergency department visit, WHO severity 3), moderate (hospitalised without invasive ventilation, WHO severity 4–6), severe (hospitalised with invasive ventilation or extracorporeal membrane oxygenation, WHO severity 7–9), and mortality or hospice (hospital mortality or discharge to hospice, WHO severity 10).19 Because of the small number in certain categories among people with HIV, we collapsed and regrouped WHO CPS categories into three categories: unaffected, mild (including mild emergency department) or moderate, and severe (including mortality or hospice). The binary death outcome was determined through the death table. The month of each patient's COVID-19 diagnosis was also retrieved from laboratory test and clinical conditions.
We included lifestyle factors such as smoking status and obesity (indicated by body-mass index [BMI]). Smoking status was defined by a concept set, whose member comprised of “ARIScience-Smoker-JA”, “smoker_NMH”, “UVA Former Smoker”, and “UVA Current Smoker” in the observation and condition tables. BMI information was retrieved from patient severity score tables. The comorbidities were defined based on the ICD codes in the updated Charlson Comorbidity Index (CCI) scoring instrument.25 A series of binary variables were used to indicate the presence or absence of each comorbidity, such as myocardial infarction, chronic pulmonary disease, and chronic kidney disease. The concept code sets we used to define each comorbidity was listed in the appendix (p 2). The adapted CCI score (subtracting the score assigned to HIV diagnosis) was also calculated for the analysis.25
Statistical analysis
Descriptive statistics were used to examine the sociodemographics of all the COVID-19 cases by HIV status. The variable distributions between COVID-19 patients with and without HIV infection were summarised and compared with the independent t test (for continuous variables) or χ2 test (for categorical variables). For all COVID-19 outcomes (ie, hospitalisation, death, and disease severity), we used hierarchical logistic regression analyses to estimate the association of HIV infection and HIV markers (CD4 cell count, viral load) with these COVID-19 outcomes. A subsample of people with HIV (n=1544) who had data for both CD4 cell count and viral load were available for analysing the association between HIV markers and COVID-19 outcomes. In step one we first adjusted for age (18–49, 50–64, and ≥65 years), sex (male, female), race (Black or African American; White; Asian or other or unknown) and ethnicity (Hispanic or Latinx; not Hispanic or Latinx; others or unknown). In step two we subsequently adjusted for smoking (non-smoker, current, or former smoker) and obesity (BMI >30 kg/m2, ≤30 kg/m2, or unknown). In step three we adjusted for an array of comorbidities (hemiplegia or paraplegia, dementia, liver disease, myocardial infarction, congestive heart failure, chronic pulmonary disease, cancer, diabetes, stroke, peripheral vascular disease, rheumatological disease, renal disease, and peptic ulcer disease) or CCI score, and month of COVID-19 diagnosis. For COVID-19 severity, multinomial logistic regressions were applied with unaffected COVID-19 individuals as a reference group. Specifically, the same aforementioned factors were included in multiple multinominal regressions (mild or moderate vs unaffected, severe vs unaffected). Because information from at least one health-care encounter is required to generate index encounter and define COVID-19 severity, we first conducted multinominal regressions among the full sample in the main analysis, with those cases without health-care encounter information grouped into the first category of CPS (ie, unaffected). Then we did a sensitivity analysis to exclude those without health-care encounter information in the multinomial analysis (ie, 1 396 795 [97·2%] of total sample). We did additional sensitivity analyses with people with and without HIV matched on age group, sex, and number of comorbidities using 1:2 and 1:4 ratios. Collinearity was checked by calculating variance inflation factors for each covariate listed in the adjusted models.
To investigate whether age, sex, race, and ethnicity could be potential effect modifiers of HIV status, we fitted interaction terms between age (18–49, 50–64, and ≥65 years), sex, race (White vs Black or African American, and White vs Asian, other, or unknown), ethnicity (not Hispanic or Latinx vs Hispanic or Latinx) and HIV status in all analyses (not all the interaction analysis results are shown due to space limitation but are available from the authors upon request). We fitted stratified models with HIV infection, demographics, lifestyle, and comorbidities on each selected subgroup (ie, male; female; aged 18–49 years; aged 50–64 years; aged ≥65 years; White; Black or African American; and Asian, other, or unknown). We implemented all analyses with SQL and R (version 3.5) and created reproducible pipelines in the Code workbook on N3C Data Enclave.
Role of the funding source
The National Center for Advancing Translational Science contributed to the design, maintenance, and security of the N3C Enclave. The funders of the study had no role in the study design, data analysis, data interpretation, or writing of the report.
Results
In this population level analysis, of the 5 830 841 COVID-19 cases and controls harmonised into the N3C data release set from Jan 1, 2020, to May 8, 2021, a total of 1 436 622 adult individuals who were positive for COVID-19 were included in this study (figure 1 ).
Study participant selection
*The numbers for missing race data and missing ethnicity data did not add up to 730 because some records missed both data.
Compared with people without HIV, those with HIV had a narrower age distribution overall (lower proportion aged ≥65 years) but the median age was 2 years older (49 vs 47 years); a greater proportion were males and people of Black or African American race. People with HIV had higher prevalence of all comorbidities, including diabetes, chronic pulmonary disease, and liver disease (table 1 ).
Table 1
Characteristics of adult COVID-19 cases by HIV status in National COVID Cohort Collaborative data, Jan 1, 2020, to May 8, 2021
| Overall (n=1 436 622) | People with HIV (n=13 170) | People without HIV (n=1 423 452) | p value | ||
|---|---|---|---|---|---|
| Social demographics | |||||
| Age, years | 47 (32–61) | 49 (36–60) | 47 (32–61) | .. | |
| 18–49 | 770 099 (53·60%) | 6703 (50·90%) | 763 396 (53·63%) | <0·0001 | |
| 50–64 | 371 489 (25·86%) | 4533 (34·42%) | 366 956 (25·78%) | .. | |
| ≥65 | 295 034 (20·54%) | 1934 (14·68%) | 293 100 (20·59%) | .. | |
| Sex* | |||||
| Male | 645 956 (44·96%) | 9641 (73·20%) | 636 315 (44·70%) | <0·0001 | |
| Female | 789 148 (54·93%) | 3521 (26·74%) | 785 627 (55·19%) | .. | |
| Race | |||||
| Black or African American | 202 947 (14·13%) | 4092 (31·07%) | 198 855 (13·97%) | <0·0001 | |
| White | 853 997 (59·44%) | 6013 (45·66%) | 847 984 (59·57%) | .. | |
| Asian, other, or unknown | 379 678 (26·43%) | 3065 (23·27%) | 376 613 (26·46%) | .. | |
| Ethnicity | |||||
| Hispanic or Latinx | 213 205 (14·84%) | 2227 (16·91%) | 210 978 (14·82%) | <0·0001 | |
| Not Hispanic or Latinx | 1 001 390 (69·70%) | 9479 (71·97%) | 991 911 (69·68%) | .. | |
| Other or unknown | 222 027 (15·45%) | 1464 (11·12%) | 220 563 (15·49%) | .. | |
| Comorbidities | |||||
| Diabetes | 233 838 (16·28%) | 3066 (23·28%) | 230 772 (16·21%) | <0·0001 | |
| Renal disease | 86 803 (6·04%) | 1758 (13·35%) | 85 045 (5·97%) | <0·0001 | |
| Congestive heart failure | 75 461 (5·25%) | 1026 (7·79%) | 74 435 (5·23%) | <0·0001 | |
| Chronic pulmonary disease | 200 998 (13·99%) | 2974 (22·58%) | 198 024 (13·91%) | <0·0001 | |
| Peripheral vascular disease | 72 808 (5·07%) | 979 (7·43%) | 71 829 (5·05%) | <0·0001 | |
| Stroke | 68 171 (4·75%) | 878 (6·67%) | 67 293 (4·73%) | <0·0001 | |
| Cancer | 78 909 (5·49%) | 1205 (9·15%) | 77 704 (5·46%) | <0·0001 | |
| Dementia | 24 622 (1·71%) | 251 (1·91%) | 24 371 (1·71%) | 0·095 | |
| Myocardial infarction | 42 165 (2·94%) | 693 (5·26%) | 41 472 (2·91%) | <0·0001 | |
| Liver disease | 72 701 (5·06%) | 2152 (16·34%) | 70 549 (4·96%) | <0·0001 | |
| Rheumatological disease | 46 861 (3·26%) | 501 (3·80%) | 46 360 (3·26%) | 0·0005 | |
| Hemiplegia or paraplegia | 11 656 (0·81%) | 222 (1·69%) | 11 434 (0·80%) | <0·0001 | |
| Peptic ulcer disease | 14 237 (0·99%) | 230 (1·75%) | 14 007 (0·98%) | <0·0001 | |
| Lifestyle factors | |||||
| Body-mass index, kg/m2 | |||||
| >30 | 218 159 (15·19%) | 2469 (18·75%) | 215 690 (15·15%) | <0·0001 | |
| ≤30 | 295 370 (20·56%) | 4560 (34·62%) | 290 810 (20·43%) | .. | |
| Unknown | 923 093 (64·25%) | 6141 (46·63%) | 916 952 (64·42%) | .. | |
| Smoking status | |||||
| Non-smoker | 1 194 746 (83·16%) | 9318 (70·75%) | 1 185 428 (83·28%) | <0·0001 | |
| Current or former smoker | 241 876 (16·84%) | 3852 (29·25%) | 238 024 (16·72%) | .. | |
| Clinical spectrum outcomes | |||||
| COVID-19 death | 26 130 (1·82%) | 445 (3·38%) | 25 685 (1·80%) | <0·0001 | |
| COVID-19 hospitalisation | 262 331 (18·26%) | 3724 (28·28%) | 258 607 (18·17%) | <0·0001 | |
| COVID-19 disease severity | |||||
| Unaffected | 476 250 (33·15%) | 6395 (48·56%) | 469 855 (33·01%) | <0·0001 | |
| Mild† or moderate | 895 491 (62·33%) | 6209 (47·15%) | 889 282 (62·47%) | .. | |
| Severe‡ | 25 054 (1·74%) | 475 (3·61%) | 24 579 (1·73%) | .. | |
| Unknown | 39 827 (2·77%) | 91 (0·69%) | 39 736 (2·79%) | .. | |
| HIV factors (n=1544) | |||||
| Most recent CD4 count, cells per μL§ | |||||
| >500 | 920 (59·59%) | 920 (59·59%) | .. | .. | |
| 200–500 | 445 (28·82%) | 445 (28·82%) | .. | .. | |
| <200 | 179 (11·59%) | 179 (11·59%) | .. | .. | |
| Most recent viral suppression, <200 copies per mL§ | 1265 (81·93%) | 1265 (81·93%) | .. | .. | |
Data are median (IQR) or n (%). NA=not applicable.
Among the 1 436 622 COVID-19 cases, 262 331 (18·26%) were hospitalised and 26 130 (1·82%) died. People with HIV disproportionately required more COVID-19 related hospitalisation than those without (28·28% vs 18·17% [table 1]). Crude odds ratios (ORs) of COVID-19 hospitalisation and death were both higher in people with HIV (table 2 ). The associations were both attenuated, but remained significant, after sequentially adjusting for demographics, lifestyle factors, comorbidities, and month of COVID-19 diagnosis (hospitalisation adjusted OR [aOR] 1·20, 95% CI 1·15–1·26; mortality 1·29, 1·16–1·44; table 2; figure 2 ; appendix pp 3–6).
Table 2
Association between HIV status and COVID-19 clinical spectrum outcomes based on hierarchical logistic regression models
| Death, OR (95% CI)* | Hospitalisation, OR*(95% CI) | Mild†or moderate COVID-19 vs unaffected‡, OR (95% CI)§ | Severe¶COVID-19 vs unaffected‡, OR (95% CI)§ | |||
|---|---|---|---|---|---|---|
| Unadjusted model | 1·90 (1·73–2·09) | 1·78 (1·71–1·84) | 0·55 (0·53–0·57) | 1·52 (1·38–1·67) | ||
| Adjusted models | ||||||
| Adjusted for age + sex + race + ethnicity | 1·85 (1·67–2·04) | 1·62 (1·56–1·69) | 0·53 (0·51–0·55) | 1·34 (1·21–1·48) | ||
| Adjusted for age + sex + race + ethnicity + smoking + BMI | 1·76 (1·59–1·94) | 1·48 (1·42–1·54) | 0·59 (0·57–0·61) | 1·34 (1·21–1·47) | ||
| Adjusted for age + sex + race + ethnicity + smoking + BMI + comorbidities‖ + month of diagnosis | 1·29 (1·16–1·44) | 1·20 (1·15–1·26) | 0·61 (0·59–0·64) | 1·04 (0·94–1·16) | ||
| Adjusted for age + sex + race + ethnicity + smoking + BMI + CCI +month of diagnosis | 1·43 (1·29– 1·59) | 1·28 (1·23–1·34) | 0·61 (0·58–0·63) | 1·15 (1·04–1·27) | ||
| Interaction models** | ||||||
| Age and HIV status | ||||||
| With HIV and aged 50–64 years vs aged 18–49 years | 7·86 (5·86–10·53) | 2·17 (1·93–2·44) | 0·48 (0·49–0·61) | 3·08 (2·30–4·13) | ||
| With HIV and aged ≥65 years vs aged 18–49 years | 22·20 (16·91–29·12) | 3·42 (2·98–3·93) | 0·57 (0·50–0·64) | 7·57 (5·73–9·99) | ||
| Without HIV and aged 50–64 years vs aged 18–49 years | 4·23 (4·00–4·48) | 1·43 (1·41–1·45) | 0·81 (0·81–0·82) | 2·32 (2·21–2·43) | ||
| Without HIV and aged ≥65 years vs aged 18–49 years | 12·46 (11·82–13·13) | 2·39 (2·36–2·42) | 0·71 (0·7–0·72) | 4·66 (4·45–4·87) | ||
| Sex and HIV status | ||||||
| With HIV and male vs female | 3·13 (2·29–4·28) | 1·32 (1·20–1·44) | 0·50 (0·47–0·53) | 1·89 (1·44–2·48) | ||
| Without HIV and male vs female | 1·54 (1·50–1·59) | 1·16 (1·15–1·17) | 1·13 (1·12–1·14) | 1·75 (1·70–1·8) | ||
| Race and HIV status | ||||||
| With HIV and Black or African American vs White | 3·64 (2·59–5·11) | 3·16 (2·85–3·49) | 1·10 (1·001–1·2) | 3·25 (2·4–4·41) | ||
| With HIV and Asian, other, or unknown vs White | 4·32 (2·88–6·46) | 2·85 (2·48–3·27) | 0·79 (0·70–0·89) | 3·63 (2·52–5·23) | ||
| Without HIV and Black or African American vs White | 1·16 (1·12–1·20) | 1·66 (1·64–1·68) | 0·92 (0·91–0·93) | 1·25 (1·21–1·3) | ||
| Without HIV and Asian, other, or unknown vs White | 1·16 (1·11–1·20) | 1·35 (1·33–1·37) | 1·01 (1–1·02) | 1·18 (1·13–1·22) | ||
| Ethnicity and HIV status | ||||||
| With HIV and Hispanic or Latinx vs not Hispanic or Latinx | 3·48 (2·25–5·37) | 1·88 (1·62–2·19) | 0·86 (0·76–0·97) | 2·33 (1·55–3·50) | ||
| With HIV and other or unknown vs not Hispanic or Latinx | 3·50 (2·20–5·59) | 1·70 (1·42–2·03) | 0·49 (0·42–0·57) | 1·98 (1·28–3·05) | ||
| Without HIV and Hispanic or Latino vs not Hispanic or Latinx | 1·07 (1·02–1·12) | 1·26 (1·24–1·28) | 1·09 (1·08–1·10) | 1·38 (1·32–1·44) | ||
| Without HIV and other or unknown vs not Hispanic or Latinx | 0·87 (0·84–0·91) | 0·62 (0·61–0·63) | 0·72 (0·71–0·73) | 0·77 (0·74–0·81) | ||
| Stratified models, people with HIV vs people without HIV at each subgroup | ||||||
| Age, years | ||||||
| 18–49 (n=770 099) | 1·24 (0·93–1·67) | 1·25 (1·17–1·34) | 0·62 (0·59–0·66) | 0·96 (0·75–1·22) | ||
| 50–64 (n=371 489) | 1·05 (0·88–1·26) | 1·04 (0·97–1·12) | 0·59 (0·55–0·62) | 0·81 (0·68–0·96) | ||
| ≥65 (n=295 034) | 1·26 (1·06–1·50) | 1·27 (1·18–1·36) | 0·67 (0·61–0·74) | 1·23 (1·06–1·43) | ||
| Sex | ||||||
| Female (n=645 956) | 1·90 (1·57–2·30) | 1·77 (1·64–1·92) | 0·93 (0·86–0·997) | 1·80 (1·49–2·18) | ||
| Male (n=789 148) | 1·14 (0·98–1·26) | 1·01 (0·96–1·06) | 0·52 (0·50–0·54) | 0·84 (0·75–0·95) | ||
| Race | ||||||
| White (n=853 997) | 1·31 (1·11–1·54) | 1·11 (1·03–1·18) | 0·50 (0·47–0·53) | 0·90 (0·76–1·07) | ||
| Black or African American (n=202 947) | 1·26 (1·06–1·50) | 1·27 (1·18–1·36) | 0·84 (0·79–0·90) | 1·18 (0·995–1·40) | ||
| Asian, other, or unknown (n=379 678) | 1·27 (1·02–1·57) | 1·19 (1·09–1·30) | 0·59 (0·55–0·64) | 1·15 (0·94–1·40) | ||
BMI=body-mass index. CCI= Charlson Comorbidity Index. OR=odds ratio.
Estimates for the associations between HIV status and COVID-19 clinical spectrum outcomes
(A) Mild or moderate disease. (B) Severe disease. (C) Hospitalisation. (D) Death. All stratified models (by age, sex, and race) were adjusted for age, sex, race, ethnicity, smoking, BMI, and comorbidities, including hemiplegia or paraplegia, dementia, liver disease, myocardial infarction, congestive heart failure, chronic pulmonary disease, cancer, diabetes, stroke, peripheral vascular disease, rheumatologic disease, renal disease, and peptic ulcer disease. Mild COVID-19 includes both the mild (outpatient, WHO severity 1–3) and mild emergency department (outpatient with emergency department visit, WHO severity ∼3) categories. Moderate COVID-19 includes patients who were hospitalised but without invasive ventilation (WHO severity 4–6). Severe COVID-19 includes both severe (hospitalised with invasive ventilation or extracorporeal membrane oxygenation, WHO severity 7–9) and mortality or hospice (hospital mortality or discharge to hospice, WHO severity 10) categories based on WHO criterion. BMI=body-mass index.
Compared with people without HIV, those with HIV had a higher proportion of severe illness (3·61% vs 1·73%), but a lower proportion of mild or moderate illness (47·15% vs 62·47%; table 1). Using unaffected COVID-19 individuals as a reference group in multinomial regression, people with HIV had lower odds of presenting with mild or moderate illness than people without HIV even after adjusting for all the covariates (aOR 0·61, 95% CI 0·59–0·64); by contrast, the odds of severe COVID-19 were comparable after sequential adjustments for all the covariates (1·04, 0·94–1·16; table 2; figure 2; appendix pp 7–8). In the sensitivity analysis (excluding individuals without health-care encounter information), the results were similar to the findings in the models with the full sample (appendix pp 9–10). The results from additional sensitivity analyses among the subsample of 1:2 and 1:4 matched people with and people without HIV were similar to the findings in the primary analyses (appendix p 11). Among 1544 people with HIV with both CD4 cell count and viral load data, a lower CD4 cell count (<200 cells per μL) was positively associated with all the adverse COVID-19 outcomes (ie, disease severity, hospitalisation, mortality) after adjusting for all the covariates, while viral suppression was only negatively associated with hospitalisation (table 3 ).
Table 3
COVID-19 outcomes among people living with HIV by HIV CD4 counts and viral load level (n=1544)
| Death, OR (95% CI)* | Hospitalisation, OR (95% CI)* | Mild†or moderate vs unaffected, OR (95% CI)* | Severe‡vsunaffected, OR (95% CI)* | ||
|---|---|---|---|---|---|
| HIV factors | |||||
| Most recent CD4 count§ | |||||
| >500 cells per μL | 1·00 | 1·00 | 1·00 | 1·00 | |
| 200–500 cells per μL | 1·49 (0·55–4·03) | 1·28 (0·94–1·75) | 1·15 (0·89–1·48) | 1·62 (0·59–4·44) | |
| <200 cells per μL | 3·10 (1·06–9·13) | 2·73 (1·80–4·14) | 1·51 (1·04–2·21) | 3·91 (1·31–11·62) | |
| Most recent viral suppression, <200 copies per mL§ | 0·71 (0·27–1·89) | 0·69 (0·49–0·97) | 0·87 (0·64–1·17) | 0·62 (0·24–1·57) | |
| Social demographics | |||||
| Age, years | |||||
| 18–49 | 1·00 | 1·00 | 1·00 | 1·00 | |
| 50–64 | 1·51 (0·52–4·39) | 0·88 (0·65–1·20) | 0·60 (0·47–0·77) | 0·62 (0·23–1·69) | |
| ≥65 | 3·39 (1·07–10·8) | 0·79 (0·50–1·25) | 0·37 (0·25–0·55) | 0·58 (0·17–1·96) | |
| Male vs female | 1·12 (0·42–2·94) | 0·88 (0·63–1·23) | 0·69 (0·52–0·91) | 1·29 (0·47–3·54) | |
| Race | |||||
| White | 1·00 | 1·00 | 1·00 | 1·00 | |
| Black or African American | 2·33 (0·82–6·67) | 1·77 (1·26–2·47) | 1·56 (1·18–2·04) | 2·08 (0·74–5·83) | |
| Asian, other, or unknown | 1·12 (0·30–4·18) | 1·66 (1·09–2·54) | 1·32 (0·93–1·86) | 1·36 (0·36–5·06) | |
| Ethnicity | |||||
| Not Hispanic or Latinx | 1·00 | 1·00 | 1·00 | 1·00 | |
| Hispanic or Latinx | 1·33 (0·33–5·35) | 0·83 (0·55–1·25) | 0·93 (0·68–1·27) | 0·89 (0·23–3·50) | |
| Other or unknown | 2·67 (0·69–10·35) | 0·82 (0·45–1·50) | 0·48 (0·29–0·80) | 1·16 (0·27–5·01) | |
| Lifestyle factors | |||||
| Body-mass index, kg/m2 | |||||
| ≤30 | 1·00 | 1·00 | 1·00 | 1·00 | |
| >30 | 3·30 (1·14–9·53) | 0·67 (0·47–0·96) | 0·77 (0·55–1·08) | 2·70 (1·00–7·29) | |
| Unknown | 1·71 (0·58–5·04) | 0·22 (0·16–0·31) | 0·45 (0·34–0·60) | 0·52 (0·16–1·68) | |
| Smoking status | |||||
| Non-smoker | 1·00 | 1·00 | 1·00 | 1·00 | |
| Current or former smoker | 2·57 (1·03–6·43) | 1·09 (0·80–1·47) | 0·46 (0·35–0·60) | 1·41 (0·57–3·53) | |
| Comorbidities | |||||
| Hemiplegia or paraplegia | 4·73 (0·79–28·17) | 5·55 (2·08–14·78) | 2·17 (0·90–5·24) | 4·37 (0·54–35·37) | |
| Dementia | 2·85 (0·55–14·91) | 0·90 (0·31–2·61) | 0·85 (0·32–2·28) | 5·94 (1·09–32·38) | |
| Liver disease | 1·46 (0·61–3·49) | 1·15 (0·83–1·59) | 1·12 (0·85–1·48) | 1·54 (0·64–3·71) | |
| Myocardial infarction | 0·42 (0·09–2·05) | 2·12 (1·13–3·98) | 0·90 (0·50–1·61) | 0·22 (0·03–1·49) | |
| Congestive heart failure | 2·37 (0·77–7·25) | 2·45 (1·49–4·04) | 0·88 (0·54–1·43) | 1·86 (0·59–5·79) | |
| Chronic pulmonary disease | 0·76 (0·30–1·90) | 1·21 (0·89–1·65) | 1·08 (0·83–1·39) | 1·61 (0·68–3·78) | |
| Cancer | 4·52 (1·85–11·03) | 1·46 (0·98–2·19) | 1·07 (0·75–1·54) | 3·56 (1·40–9·07) | |
| Diabetes | 0·93 (0·35–2·49) | 1·41 (1·01–1·96) | 1·20 (0·92–1·58) | 1·57 (0·60–4·14) | |
| Stroke | 1·03 (0·30–3·53) | 1·28 (0·76–2·14) | 0·75 (0·46–1·21) | 0·42 (0·09–1·96) | |
| Peripheral vascular disease | 0·79 (0·24–2·61) | 1·32 (0·81–2·16) | 2·21 (1·41–3·48) | 1·57 (0·47–5·27) | |
| Rheumatologic disease | 0¶ | 1·16 (0·56–2·39) | 1·53 (0·86–2·72) | 0 (0·00–0·00) | |
| Renal disease | 2·58 (1·02–6·49) | 1·55 (1·06–2·27) | 1·06 (0·75–1·49) | 3·06 (1·19–7·87) | |
| Peptic ulcer disease | 0·67 (0·06–7·21) | 1·26 (0·52–3·02) | 1·06 (0·48–2·33) | 0·69 (0·06–7·81) | |
| Month of COVID-19 diagnosis | 1·00 (0·92–1·08) | 1·00 (0·97–1·03) | 1·04 (1·01–1·06) | 1·03 (0·95–1·12) | |
The interaction effect of age and HIV status suggested that the ageing process in people with HIV exacerbated all the adverse outcomes of COVID-19. Those with HIV in the older age groups had much higher odds of death and hospitalisation than those without HIV in the same age range. As another potential modifier, male sex could also interact with HIV infection in increasing the odds of severe clinical outcomes of COVID-19, yet with a smaller magnitude. Similar results were found in the interaction of race or ethnicity and HIV status, by which Black or African American race and Hispanic or Latinx ethnicity interacted with HIV infection in developing higher odds of adverse COVID-19 outcomes. Stratified models revealed that the elevated odds were higher in the similar subgroups (eg, older age, male sex, and Black or African American race; table 2; appendix pp 12–13).
To adjust the role of cumulative burden of comorbidities in the model development, CCI was considered in all adjusted models. However, a high collinearity was detected between CCI and the other covariates (variance inflation factor=7·97). Therefore, additional models were developed to include CCI (replacing individual comorbid conditions) in the analyses. Findings from the two sets of adjusted models (adjusting individual comorbid condition vs adjusting CCI) were similar (appendix pp 14–16).
Discussion
Our population-level analysis from N3C data found that people with HIV might not be disproportionately vulnerable to SARS-CoV-2 infection but are more likely to be hospitalised and die from COVID-19, although such risk might be attenuated when other confounding factors are taken into consideration. The associations between HIV and these outcomes seem particularly pronounced among older people, males, Black or African American adults, and Hispanic or Latinx adults. Among people with HIV, we find that the risks for poor COVID-19 outcomes are much higher among those with lower CD4 cell counts (<200 cells per μL) and an association between viral suppression and the COVID-19 outcome of hospitalisation.
To the best of our knowledge, this is the largest population-level analysis to investigate the role of HIV infection in COVID-19 clinical spectrum across the USA. Our results show a smaller but consistent effect of HIV infection on COVID-19 related mortality with large population-based cohort studies from South Africa11 and the UK.12 The differences of effect size between these three studies could possibly be explained by the different sample characteristics. Results from the interaction effects illustrate that the adverse COVID-19 outcomes among people with HIV might be explained by the overlapping demographic (eg, male and African American) and comorbidity characteristics (eg, a significant interaction effect of HIV and CCI, and data not shown but available upon request) that are highly prevalent in this population. Our study shows people with HIV require more COVID-19 hospitalisation, at a level of risk similar to a recent New York study14 and other USA studies using TriNETX network data, which controlled for BMI and various comorbidities.6, 26
Regarding the clinical severity of COVID-19, people with HIV are less likely to have mild illness, but more likely to have severe outcomes when only adjusting for demographics and lifestyle factors. The adjustment for comorbidities obviates the estimated risk of severe outcomes among people with HIV. This finding suggests that people with HIV might show less symptoms at the earlier stage of SARS-CoV-2 infection. Such protection from the most serious sequelae of COVID-19 might be attributable to the possible anti-SARS-CoV-2 activity of tenofovir disoproxil fumarate plus emtricitabine, as suggested in both observational and randomised closed trials studies.27, 28 Another hypothesis is that people with HIV with mild illness might be underrepresented (47·15% vs 62·33% in the overall group; table 1) because of higher stigma, increased fear of hospitalisation, higher social deprivation, and lower medical coverage when compared with people who do not have HIV. A consequence of such late linkage to care could be a higher risk of severe COVID-19.
As declining CD4 cell counts are associated with COVID-19 severity in general,29 people with HIV and low CD4 cell counts might have a raised risk of severe COVID-19.30 Our study supported this hypothesis and found that a lower CD4 cell count is associated with a higher risk of adverse COVID-19 outcomes, which is also in agreement with another multicentre study conducted by Dandachi and colleagues.16 No association was observed between viral suppression and COVID-19 disease severity or mortality. Although our study observed the protective effect of viral suppression in reducing hospitalisation, the multicentre study did not.16 Our larger sample size was possibly the reason for detecting such a difference, because the multicentre study had a smaller sample and most of the study participants were receiving antiretroviral therapy.16
In this study, the age, sex, and race or ethnicity disparities in COVID-19 severe outcomes are pronounced among people with HIV. The study sample characteristics mirror the demographics of this population in the USA, with higher proportions of males, Hispanic or Latinx adults, and Black or African Americans adults. Hispanic or Latinx individuals, as well as those of an older age, have higher mortality and hospitalisation rates among people with HIV than in people without HIV, which were not reported in the UK study.12 However, both our study and the UK study showed similar findings of the larger association between HIV and adverse COVID-19 outcomes among Black or African American adults. Understanding the reasons for the disproportionately large association between HIV and adverse COVID-19 outcomes in these subgroups will be a priority if effective policies are to be developed to mitigate any increased risks among these groups.
Our study had several limitations. First, although we included around 6 million individuals from the N3C dataset, the majority are based in the southeast, mid-Atlantic, and mid-west, and therefore might not be representative of the entire COVID-19 population or HIV population in the USA. Second, the algorithm for HIV case identification is not validated. Potential misclassification of HIV status might occur because of data availability and missing data on the N3C concept sets (HIV condition codes, antiretroviral therapy exposure, and laboratory results) that were available to the investigation team at the time of this study. Such identifications might change as the phenotype template of HIV patients changes as a result of continuous data updates from different contributing sites. Additionally, the release of other available concept sets might yield different classifications as well. However, previous studies have shown acceptable sensitivity and specificity of a similar approach.31 Therefore, these potential misclassifications are likely to be non-differential throughout the cohort and unlikely to change our conclusions. Third, some key exposure variables (eg, CD4 cell count, viral load, BMI, and smoking status) are not uniformly available or measured accurately across all the study sites; for example, a large proportion of patients have missing CD4 cell count and HIV viral load data. Furthermore, the effect of obesity on COVID-19 outcomes might be underestimated because of the large proportion of unknown responses and the uneven distribution of unknown responses between the two comparison groups. Moreover, the inability to separate the former smokers from current smokers in the dataset did not allow us to examine the effect of different smoking status between people living with HIV and people without HIV on adverse COVID-19 outcomes. Fourth, the adverse COVID-19 outcomes might vary when stratifying by other vulnerable statuses of people with HIV, such as transgender individuals or injection drug users. However, codes for identifying these statuses were unavailable in this dataset.
In conclusion, using data from the largest COVID-19 population level analysis with a heterogeneous population in the USA, our study could identify people with HIV with mild or asymptomatic COVID-19 and examine the different risks for SARS-CoV-2 acquisition versus progression to severe disease or death once infected. In this large study, people with HIV have an elevated risk of adverse COVID-19 outcomes. The attenuated risk after controlling for comorbidities, which are more prevalent and typically occur at a younger age among people with HIV, indicates that certain underlying medical conditions had a greater influence on COVID-19 outcomes of this population. Our observation that people with lower CD4 cell counts are at a higher risk of poor outcomes suggests that people with a history of advanced immunosuppression might warrant closer observation and monitoring. The robust risk assessment of this study could inform prioritisation of prevention messaging, disease monitoring and therapies, and vaccination for people with HIV, especially those with more pronounced immunodeficiency. Given the pandemic's exacerbating effects on health inequities, public health and clinical communities must strengthen services and support to prevent aggravated COVID-19 outcomes among people with HIV, particularly for those with pronounced immunodeficiency.
Data sharing
The National Institute of Health's (NIH) N3C data used in this study is available upon application at https://ncats.nih.gov/n3c.
Declaration of interests
We declare no competing interests.
Acknowledgments
The analyses described in this publication were conducted with data or tools accessed through the National Center for Advancing Translational Sciences (NCATS) N3C Data Enclave and supported by NCATS U24 TR002306 and the National Institute of Allergy and Infectious Diseases (NIAID) of the NIH under Award Number R01AI127203-4S1. This study was also supported by the following grants from Stony Brook University (U24TR002306); University of Oklahoma Health Sciences Center (U54GM104938), Oklahoma Clinical and Translational Science Institute; West Virginia University (U54GM104942), West Virginia Clinical and Translational Science Institute; University of Mississippi Medical Center (U54GM115428), Mississippi Center for Clinical and Translational Research; University of Nebraska Medical Center (U54GM115458), Great Plains IDeA-Clinical & Translational Research; Maine Medical Center (U54GM115516), Northern New England Clinical & Translational Research Network; Wake Forest University Health Sciences (UL1TR001420), Wake Forest Clinical and Translational Science Institute; Northwestern University at Chicago (UL1TR001422), Northwestern University Clinical and Translational Science Institute; University of Cincinnati (UL1TR001425), Center for Clinical and Translational Science and Training; The University of Texas Medical Branch at Galveston (UL1TR001439), The Institute for Translational Sciences; Medical University of South Carolina (UL1TR001450), South Carolina Clinical & Translational Research Institute; University of Massachusetts Medical School Worcester (UL1TR001453), The UMass Center for Clinical and Translational Science; University of Southern California (UL1TR001855), The Southern California Clinical and Translational Science Institute; Columbia University Irving Medical Center (UL1TR001873), Irving Institute for Clinical and Translational Research; George Washington Children's Research Institute (UL1TR001876), Clinical and Translational Science Institute at Children's National; University of Kentucky (UL1TR001998), UK Center for Clinical and Translational Science; University of Rochester (UL1TR002001), University of Rochester Clinical & Translational Science Institute; University of Illinois at Chicago (UL1TR002003), University of Illinois at Chicago Center for Clinical and Translational Science; Penn State Health Milton S Hershey Medical Center (UL1TR002014), Penn State Clinical and Translational Science Institute; The University of Michigan at Ann Arbor (UL1TR002240), Michigan Institute for Clinical and Health Research; Vanderbilt University Medical Center (UL1TR002243), Vanderbilt Institute for Clinical and Translational Research; University of Washington (UL1TR002319), Institute of Translational Health Sciences; Washington University in St Louis (UL1TR002345), Institute of Clinical and Translational Sciences; Oregon Health & Science University (UL1TR002369), Oregon Clinical and Translational Research Institute; University of Wisconsin-Madison (UL1TR002373), University of Wisconsin Institute for Clinical and Translational Research; Rush University Medical Center (UL1TR002389), The Institute for Translational Medicine (ITM); The University of Chicago (UL1TR002389), ITM; University of North Carolina at Chapel Hill (UL1TR002489), North Carolina Translational and Clinical Science Institute; University of Minnesota (UL1TR002494), Clinical and Translational Science Institute; Children's Hospital Colorado (UL1TR002535), Colorado Clinical and Translational Sciences Institute; The University of Iowa (UL1TR002537), Institute for Clinical and Translational Science; The University of Utah (UL1TR002538), Uhealth Center for Clinical and Translational Science; Tufts Medical Center (UL1TR002544), Tufts Clinical and Translational Science Institute; Duke University (UL1TR002553), Duke Clinical and Translational Science Institute; Virginia Commonwealth University (UL1TR002649), C Kenneth and Dianne Wright Center for Clinical and Translational Research; The Ohio State University (UL1TR002733), Center for Clinical and Translational Science; The University of Miami Leonard M Miller School of Medicine (UL1TR002736), University of Miami Clinical and Translational Science Institute; University of Virginia (UL1TR003015), Integrated Translational health Research Institute of Virginia (iTHRIV); Carilion Clinic (UL1TR003015), iTHRIV; University of Alabama at Birmingham (UL1TR003096), Center for Clinical and Translational Science; Johns Hopkins University (UL1TR003098), Johns Hopkins Institute for Clinical and Translational Research; University of Arkansas for Medical Sciences (UL1TR003107), UAMS Translational Research Institute; Nemours (U54GM104941), Delaware CTR ACCEL Program; University Medical Center New Orleans (U54GM104940), Louisiana Clinical and Translational Science Center; University of Colorado Denver, Anschutz Medical Campus (UL1TR002535), Colorado Clinical and Translational Sciences Institute; Mayo Clinic Rochester (UL1TR002377), Mayo Clinic Center for Clinical and Translational Science; Tulane University (UL1TR003096), Center for Clinical and Translational Science; Loyola University Medical Center (UL1TR002389), ITM; Advocate Health Care Network (UL1TR002389), ITM; OCHIN (INV-018455), Bill & Melinda Gates Foundation grant to Sage Bionetworks. The analyses described in this publication were conducted with data or tools accessed through the NCATS N3C Data Enclave and supported by NCATS U24 TR002306 and the NIAID of the NIH under Award Number R01AI127203-4S1. RCP's effort was supported by NIAID of the NIH (K23AI120855). This research was possible because of the patients whose information is included within the data and the organisations (https://ncats.nih.gov/n3c/resources/data-contribution/data-transfer-agreement-signatories) and scientists who have contributed to the ongoing development of this community resource (https://doi.org/10.1093/jamia/ocaa196). The content of this publication and the opinions expressed do not necessarily reflect the views or policies of the NIH nor does mention of trade names, commercial products, or organisations imply endorsement by the US Government. We gratefully acknowledge contributions from the N3C consortium authors: Richard Moffitt, Hana Akelsrod, Keith A Crandall, Nora Francheschini, Evan French, G Caleb-Alexander, Kathleen M Andersen, Amanda J Vinson, Todd T Brown, Roslyn B Mannon. We also acknowledge support from the N3C Publication Committee and Miranda Cole-Nixon in the preparation of this manuscript.
Contributors
XY conceptualised and wrote the first draft and critically revised of the manuscript. JS led efforts on National COVID Cohort Collaborative (N3C) HIV markers harmonisation, as well as critically reviewed the manuscript. JZ set up the statistical test design. SG wrote data preparation code and SQL R code for data analysis, which was reviewed and verified by JZ. XY prepared tables and figures with input from SG. SBW provided clinical input and patient severity predictor considerations and use thereof. BO, RCP, JYI, GDK, and XL reviewed and edited the manuscript. RCP, JS, ALO, and QZ built N3C HIV definition, phenotype verification, and statistical analyses. ALO performed data preparation and reviewed and edited the manuscript. MH and CGC reviewed and edited the manuscript, and did the project administration. XY, JZ, and SG have accessed and verified the data. The corresponding author (and XY, JS, RCP, JZ, SG, QZ, and ALO) had full access to all the data in the study. All authors had final responsibility for the decision to submit for publication.
Contributor Information
National COVID Cohort Collaborative Consortium: