Impact of Human Immunodeficiency Virus on the Burden and Severity of Influenza Illness in Malawian Adults: A Prospective Cohort and Parallel Case-Control Study

Human immunodeficiency virus (HIV)–related immunosuppression is a major risk factor for influenza illness and severity in Malawian adults. Household crowding, food insecurity, and poor sanitation are additional risk factors. Influenza preventive strategies should target HIV-infected adults in Africa.

between January and April [17]. There is no national influenza immunization program. We performed 2 prospective observational studies at the Queen Elizabeth Central Hospital (QECH), the only inpatient facility providing free healthcare to 1.3 million residents in Blantyre district, and a primary care center adjacent to QECH.
We conducted a cohort study of HIV-infected and HIVuninfected adults over 2 years; the primary end point was laboratory-confirmed influenza illness. We also performed a case-control study in adults presenting with mild and severe influenza, to establish risk factors for severe influenza (including HIV infection).

Cohort Study
We enrolled adults (aged ≥18 years) from the ART and voluntary counseling and testing clinics at QECH beginning 1 April 2013 (eligibility criteria shown in Figure 1). Active follow-up comprised bimonthly routine clinic reviews. Participants were also instructed to attend the study clinic when they experience an influenza-like illness (ILI), defined as reported or documented fever (≥38°C) and ≥2 of the following symptoms: cough, rhinorrhea, sore throat, myalgia, headache, and vomiting/diarrhea. The study clinician assessed ill participants and instituted appropriate management. Paired nasopharyngeal and oropharyngeal swab samples (FLOQswabs; Copan Diagnostics) were obtained at routine and ILI visits [18].
We compared the incidence of laboratory-confirmed influenza-associated ILI between HIV-infected and HIV-uninfected participants. The at-risk period was calculated from enrollment to the study end date (31 March 2015), death, or loss to follow-up. For participants lost to follow-up, follow-up was censored on the date of relocation, withdrawal of consent, or the last recorded clinic visit.

Case-Control Study
Between 15 May 2013 and 28 February 2015, we recruited adults admitted to QECH with acute lower respiratory tract infection (LRTI) (severe cases) and adults attending the primary care center with ILI (nonsevere disease) (eligibility criteria shown in Figure 2). Nasopharyngeal aspirate samples were obtained at enrollment. Participants with influenza-positive hospitalized LRTI and outpatient-managed ILI made up the case patients and controls, respectively.

Laboratory Procedures
Laboratory testing was performed at the Malawi-Liverpool-Wellcome Trust Clinical Research Programme laboratory. HIV status was established by means of sequential rapid HIV tests (Alere Determine and Uni-Gold, Trinity Biotech) [19]. CD4 + cell counts were performed using a FACScount flow cytometer (Becton Dickinson; BD Biosciences). Nasopharyngeal specimens were tested for influenza A and B viruses using the Centers for Disease Control and Prevention human influenza quantitative reverse-transcription polymerase chain reaction (PCR) diagnostic panel and influenza A subtyping kit [20].

Statistical Analysis
Analysis was performed with Stata software (version 12.1). We tested differences in categorical variables using χ 2 or Fisher exact test, and differences in continuous variables using t or Wilcoxon rank sum test, as appropriate. The cohort study was powered to detect an incidence rate ratio (IRR) of ≥3.0 (α = .05; 2-tailed β = .2), for influenza-associated ILI by HIV status, requiring 608 recruits and allowing for 20% loss-to follow-up, Incidence rates of influenza-associated ILI were calculated by dividing the number of events by the number of person-years of follow-up. Poisson regression models were used to estimate IRRs and 95% confidence intervals (CIs) for the effect of HIV and other risk factors on influenza. Age, sex, and HIV infection were included as potential confounders in the multivariable models. Population average Poisson regression models using generalized estimating equations were constructed for recurrent events.
In both studies, stepwise backward elimination of covariates with P values < .20 was used to rationalize the multivariable models. We limited the number of covariates in a multivariable model to maintain a limit of 10 events per variable [21]. Twoway interactions were evaluated in all final models. All available case information was used in each univariable analysis. In the multivariable models, we excluded patients with missing data for included variables (data were >95% complete for all variables). The impact of ART and CD4 + cell count was assessed in subgroup analyses of HIV-infected individuals.
For the case-control study, a sample size of 57 case patients and 114 unmatched controls provided 80% power to detect an odds ratio (OR) of ≥2.5. Based on observed influenza prevalences of 11% and 16.4% in adults presenting to QECH with severe or mild acute respiratory illness, respectively, in our sentinel surveillance (unpublished data), we estimated recruitment of approximately 518 adults with hospitalized LRTI and 695 adults with ILI.
We estimated the OR of having HIV infection and other potential risk factors for severe influenza in case patients and controls, controlling for confounders, using unconditional logistic regression models. HIV status and recruitment season were included a priori in the multivariable model. Population-attributable fractions of modifiable risk factors for severe influenza were estimated from the prevalence of exposure in case patients and adjusted OR from Reported fever OR recorded fever (≥38°C)  the multivariable logistic regression model [22]. Ethical approval was provided by the University of Malawi College of Medicine Research Ethics Committee (Study No. P.11/12/1310) and the Research Ethics Committee of the University of Liverpool (Study No. 12.43).
HIV-uninfected participants had larger households, but crowding was similar in the 2 groups. HIV-uninfected participants had better sanitation facilities, water supply, and food security. No significant differences were observed in education level, employment, or asset ownership.
Eleven participants died during the study (HIV infected, n = 10); none had reported respiratory symptoms before death. Sixty-one participants (10%) migrated out of Blantyre, 24 (3.9%) withdrew consent, and 25 (4.1%) were lost to follow-up ( Figure 3). There was no differential loss to follow-up by HIV status. The total person-time follow-up was 520 and 348 PY in the HIV-infected and HIV-uninfected cohorts, respectively.
In the univariable analysis, history of previous pneumonia, household crowding, lower education level, unemployment, and food insecurity were associated with influenza-related ILI ( Table 2). After adjustment for age, sex, household crowding and food security, HIV-infected adults had an Reasons for ineligibility included intention to relocate out of Blantyre (n = 10), inability to attend regular study visits (n = 6), inability to give written informed consent (n = 1), evidence of active acute respiratory disease at enrollment (n = 17), and enrollment of another household member in the study (n = 2). Reasons for declining consent included not having time (n = 3), not interested (n = 2), fear of participation (n = 4), and wished to seek spouse's consent (n = 2). Abbreviations: HIV, human immunodeficiency virus; ILI, influenzalike illness; PCR, polymerase chain reaction; PYFU, person-years follow-up. Among HIV-infected participants, individuals with enrollment CD4 + cell counts <200/µL had a higher incidence of influenza-associated ILI than those with counts >200/µL, but the association was nonsignificant (79.1 vs 40.5 per 1000 PY; IRR, 1.95; 95% CI, .78-4.92). The effect of HIV on influenza did not differ by ART status.

Other Risk Factors for Influenza Infection
A household crowding index (number of persons in the household divided by the number of sleeping rooms) of 1.5-2.4 was associated with a 3-fold increased risk of influenza-associated ILI, compared with households with <1.5 persons per sleeping room (aIRR, 3.41; 95% CI, 1.12-10.36). The increased risk was not observed in participants who lived in household with a crowding index >2.5 (2.06; .62-6.83) ( Table 2). Participants who reported difficulties accessing food more than twice per month had a 3-fold increased risk of influenza-associated ILI, compared with those with no food access difficulties (aIRR, 3.09; 95% CI, 1.30-7.36). Food insecurity was also a possible effect modifier of the impact of HIV infection on influenza (P = .01). Among those with frequent difficulty obtaining food, HIV infection was associated with 10-fold increased incidence of influenza-associated ILI (IRR, 9.50; 95% CI, 1.27-70.99), but no association was evident in those with little or no food insecurity (IRR, 1.18; 95% CI, . 32-4.39). No interaction was demonstrated with the other covariates.

Contribution of Influenza to Mild and Severe Respiratory Infection
Influenza was identified in 56 (10.8%) patients with hospitalized LRTI (case patients) and 88 (13.7%) patients with ILI (controls) ( Table 4). Influenza A virus was detected in 30 (53.6%) of the case patients and 35 (39.8%) of the controls; all had influenza A(H3N2) except 1 control (unsubtyped). Two controls were positive for both influenza A(H3N2) and B. Influenza A(H1N1) pdm09 was not detected during the study period.

Risk Factors for Severe Influenza Presentation
Among the 56 case patients and 88 controls, no difference in age and sex was observed (Table 4). Thirty nine (69.6%) of the case patients and 26 (29.6%) of the controls were HIV-infected (P < .001). Compared with HIV-infected controls, HIV-infected case patients had more advanced immunosuppression (median CD4 + cell count, 140/µL vs 265/µL; P = .03) and were more likely to be receiving ART (35.7 vs 9.1%; P < .001). Case patients were more likely to have a low body mass index and report a history of tuberculosis and pneumonia. Controls had better sanitation, education, and food security. However, household exposure to children <5 years old or crowding did not differ according to case-control status.
In the univariable analysis, HIV infection, reported history of tuberculosis, pneumonia within past 5 years, low body mass index (<18.5 kg/m 2 ), month of recruitment, type of water supply, sanitation facility, lower education level, and food insecurity were associated with being a case patient ( Table 4). Month of recruitment was included in the multivariable model a priori owing to seasonal discrepancy in the recruitment of patients with hospitalized LRTI or ILI. In the multivariable model, HIV infection was strongly associated with severe influenza (adjusted OR, 4.98; 95% CI, 2.09-11.88; P < .001). In addition, reported pneumonia   In a subgroup analysis of HIV-infected case patients and controls, there was a trend toward lower CD4 + cell counts in case patients (OR, 2.72; 95% CI, .97-7.60; CD4 + cell count, <200/µL vs >200/µL). No association was found between ART status and influenza severity.

DISCUSSION
In this urban African adult population, HIV infection is an important risk factor for both symptomatic influenza and severe illness. Compared with HIV-uninfected adults, HIV-infected adults had a 3 times higher incidence of influenza-associated ILI, and a 5-fold greater odds of severe influenza disease. Furthermore, nearly 60% of influenza-related hospitalized LRTI cases were attributable to HIV. Although neither study was powered to examine the effect of HIV at different levels of immunosuppression, our data suggest higher incidence and greater disease severity among those with CD4 + cell counts <200/µL. Previous population-level surveillance and retrospective studies in South Africa and Kenya that have examined the association between HIV and influenza found a higher disease burden [8][9][10] and an increased risk of influenza-associated hospitalization in HIV-infected persons [11][12][13], particular among those with severe immunosuppression [13]. The current studies provide robust evidence to support these findings having prospectively ascertained HIV status, CD4 + cell count, and information on ART, as well as other exposures, including household and socioeconomic characteristics that may confound the association between HIV and influenza.
We found a higher incidence of influenza-associated ILI among HIV-infected than among HIV-uninfected individuals. Although this could indicate an increased susceptibility to infection, it may instead, or also, reflect a greater propensity of HIV-infected individuals to develop symptomatic illness after influenza infection, which has been described elsewhere [23]. This is a pertinent finding in high-HIV prevalence settings, because HIV-infected individuals may play an important role in the community transmission of influenza.
Taken together, data from the 2 studies present a persuasive argument for a strong association between HIV and influenza. The finding that more than half of hospitalized influenza presentations in Malawian adults were attributable to HIV further emphasizes its critical role in severe influenza in this population. Because pneumonia is the most common cause of adult medical admissions at QECH [24], effective influenza preventive strategies could substantially reduce the burden of acute respiratory infections in Malawi and other similar resource-limited settings.
Inactivated influenza vaccines have demonstrated safety and efficacy in HIV-infected persons [25]. Clinical efficacy of 75.5% has been reported in South African HIV-infected adults, but   the trial excluded patients with comorbid conditions and ARTnaive patients with CD4 + cell counts <100/μL [26]. Influenza vaccines are not widely deployed in most African countries [2,27]. Before consideration of HIV-infected persons as a target group for immunization, policy makers will require evidence of vaccine efficacy in the context of advanced immunosuppression and/or comorbid conditions, as well as the potential public health impact and cost-effectiveness of vaccinating HIVinfected individuals, compared with other target groups (eg, pregnant women, young children). The acceptability of annual vaccination, feasibility of vaccine administration at ART clinics, and optimal timing of vaccination in the absence of clear seasonality will require elucidation. Numerous regulatory, logistical, and financial obstacles will need to be overcome if targeted influenza vaccination policies are to be successfully and sustainably implemented in the region. The introduction of ART was associated with a dramatic decline in influenza-attributable hospitalizations in the United States [28], and improved survival after pandemic influenza A(H1N1) infection in Mexico [29]. However, this beneficial effect has not been observed in Africa. Cohen et al [30] found no difference in case fatality ratio by ART status among HIVinfected individuals with influenza-positive severe acute respiratory illness [30]. A Malawian study demonstrated poor reconstitution of influenza-specific CD4 + T-cell response in HIV-infected adults after 12 months of highly active ART, despite a rise in CD4 + cell count [31]. Hence the impact of ART on the relationship between HIV and influenza severity requires further evaluation.
Identification of household crowding, poor sanitation, and food insecurity as risk factors for influenza highlight the importance of current public health interventions to alleviate hunger and poverty and improve access to clean water and sanitation [32]. Food insecurity emerged as a previously unrecognized risk factor for both influenza illness and severity. More in-depth evaluation of food insecurity and its association with influenza and other respiratory infections is warranted [33].
Our results, along with other data from the region, contrast with those from developed settings [6]. This highlights the limitations of extrapolating findings from developed settings to inform influenza control policies in Africa. Differences in the observed impact of HIV on influenza burden and severity may be due to more advanced immunosuppression, poorer access to ART, higher prevalence of comorbid conditions, poverty-related factors identified in this study, compared with other regions.
Our study has several limitations. First, it was conducted in a single urban center, which may limit the generalizability of our findings to rural populations. Second, this study was not powered to assess the impact of CD4 + cell count or ART on influenza incidence and severity. CD4 + cell counts were measured during acute illness in the case-control study; the degree of immunosuppression may not be accurately represented because CD4 + cell count depression can occur during acute illness. Third, passive surveillance was used for case ascertainment in both studies; underascertainment of ILI episodes, and therefore influenza cases, is conceivable. Biased estimates away from the null could have resulted if HIV-infected cohort participants were more likely than uninfected individuals to present with ILI . Similar bias could have arisen if HIV-infected individuals had a higher propensity to attend hospitals with severe respiratory symptoms in the case-control study. However, there was no significant difference in the incidence of ILI between HIV-infected and uninfected cohorts, and a greater proportion of HIV-related ILI cases had severe clinical signs (Supplementary Table S1). Furthermore, the majority of case-control study participants were unaware of their HIV status at enrollment. Finally, we were unable to control for bacterial coinfection in the case-control study, because diagnostic tests for bacteria were undertaken in case patients but not controls.
We have comprehensively evaluated the association between HIV and influenza, identifying HIV-infected persons at particular risk of symptomatic influenza and severe disease. Influenza-preventive strategies should be an important aspect of the management of HIV-infected adults in sub-Saharan Africa. Further studies are needed in Malawi and other high-HIV prevalence settings to determine influenza vaccine efficacy in persons with advanced immunosuppression and evaluate the potential public health impact ahead of operational research addressing the logistical barriers to implementing large-scale vaccination programs.

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.