The effects of the attributable fraction and the duration of symptoms on burden estimates of influenza‐associated respiratory illnesses in a high HIV prevalence setting, South Africa, 2013‐2015

Background The attributable fraction of influenza virus detection to illness (INF‐AF) and the duration of symptoms as a surveillance inclusion criterion could potentially have substantial effects on influenza disease burden estimates. Methods We estimated rates of influenza‐associated influenza‐like illness (ILI) and severe acute (SARI‐10) or chronic (SCRI‐10) respiratory illness (using a symptom duration cutoff of ≤10 days) among HIV‐infected and HIV‐uninfected patients attending 3 hospitals and 2 affiliated clinics in South Africa during 2013‐2015. We calculated the unadjusted and INF‐AF‐adjusted rates and relative risk (RR) due to HIV infection. Rates were expressed per 100 000 population. Results The estimated mean annual unadjusted rates of influenza‐associated illness were 1467.7, 50.3, and 27.4 among patients with ILI, SARI‐10, and SCRI‐10, respectively. After adjusting for the INF‐AF, the percent reduction in the estimated rates was 8.9% (rate: 1336.9), 11.0% (rate: 44.8), and 16.3% (rate: 22.9) among patients with ILI, SARI‐10, and SCRI‐10, respectively. HIV‐infected compared to HIV‐uninfected individuals experienced a 2.3 (95% CI: 2.2‐2.4)‐, 9.7 (95% CI: 8.0‐11.8)‐, and 10.0 (95% CI: 7.9‐12.7)‐fold increased risk of influenza‐associated illness among patients with ILI, SARI‐10, and SCRI‐10, respectively. Overall 34% of the estimated influenza‐associated hospitalizations had symptom duration of >10 days; 8% and 44% among individuals aged <5 and ≥5 years, respectively. Conclusion The marginal differences between unadjusted and INF‐AF‐adjusted rates are unlikely to affect policies on prioritization of interventions. HIV‐infected individuals experienced an increased risk of influenza‐associated illness and may benefit more from annual influenza immunization. The use of a symptom duration cutoff of ≤10 days may underestimate influenza‐associated disease burden, especially in older individuals.


| INTRODUCTION
Influenza virus infections cause substantial morbidity and mortality globally, in particular among individuals aged <5 and ≥65 years and persons with underlying medical conditions, including HIV infection. [1][2][3][4][5][6] The World Health Organization (WHO) highlighted that there is a need for influenza disease burden estimates especially from low-and middle-income countries. 7 Such estimates would enable governments to make informed evidence-based decisions when allocating scarce resources and planning intervention strategies to limit the impact and spread of the disease. In addition, national estimates would contribute to the global understanding of the burden of influenza-associated severe illness and inform global public health priorities.
Whereas estimates of influenza-associated mortality have been generated mainly from ecological studies, 1,5,6,[8][9][10] the majority of estimates of influenza-associated hospitalization and outpatient consultation have been obtained from laboratory-confirmed influenza surveillance conducted at selected sentinel sites. [2][3][4]11 In recent years, such surveillance systems have relied on the use of polymerase chain reaction (PCR) assays for the detection of influenza viruses and the use of standard case definitions for the identification and inclusion of patients in the surveillance programs.
Although the wide use of PCR assays has increased diagnostic and surveillance capacity, establishing a clinical association between pathogen detection and illness remains challenging. A study conducted in South Africa suggested that the attributable fraction (AF) of influenza virus detection to mild and severe respiratory illness (referred to as INF-AF hereafter) varies by age group and HIV serostatus. 12 However, the effect of the INF-AF on influenza disease burden estimates has not been fully investigated.
In addition, the WHO recommends the implementation of influenza sentinel surveillance using standard case definitions among outpatients with influenza-like illness (ILI) and inpatients with severe acute respiratory illness (SARI). 13 The WHO recommended case definitions were updated in 2012 to include, among other changes, patients with symptoms duration of ≤10 days (compared to ≤7 days as previously recommended) with the aim to increase sensitivity of influenza virus detection. 13 Nonetheless, the effect of symptom duration on influenza disease burden estimates among individuals of different age groups and HIV serostatus is not fully understood.
A better understanding of the effects of the INF-AF and the duration of symptoms as a surveillance inclusion criterion on influenza disease burden estimates could provide insights on the vaccinepreventable fraction of illness and contextualize the use of recommended surveillance case definitions for the estimation of disease burden.
We aimed to assess the effects of the INF-AF and the duration of symptoms on influenza disease burden estimates among HIV-infected and HIV-uninfected patients of different age groups presenting with unadjusted and INF-AF-adjusted rates and relative risk (RR) due to HIV infection. Rates were expressed per 100 000 population.

Conclusion:
The marginal differences between unadjusted and INF-AF-adjusted rates are unlikely to affect policies on prioritization of interventions. HIV-infected individuals experienced an increased risk of influenza-associated illness and may benefit more from annual influenza immunization. The use of a symptom duration cutoff of ≤10 days may underestimate influenza-associated disease burden, especially in older individuals.

K E Y W O R D S
attributable fraction, HIV, influenza, influenza-like illness, rates, severe respiratory illness, South Africa, symptom duration ILI, SARI, and severe chronic respiratory illness (SCRI) in South Africa A case in individuals aged ≥5 years included any hospitalized patient presenting with manifestation of acute lower respiratory tract infection with temperature ≥38°C or history of fever and cough. 13 A case of SCRI was defined as a hospitalized person who had illness onset >10 days (referred to as SCRI-10 hereafter) and who met the above age-specific clinical inclusion criteria.

| Influenza-like illness surveillance
We conducted prospective surveillance for cases presenting with ILI at two outpatient clinics (Edendale Gateway Clinic, KwaZulu-Natal Province, and Jouberton Clinic, North West Province) located in the same catchment area to the above-mentioned hospitals over the same study period.
An ILI case was defined as an outpatient of any age presenting with either temperature ≥38°C or history of fever and cough of duration of ≤10 days.

| Study procedures
The procedures of these surveillance programs have been previously described. 4,12,[14][15][16] In brief, study staff completed case report forms for all enrolled ILI, SARI-10, and SCRI-10 cases. Referral to hospital was recorded for all enrolled ILI cases. ILI cases that were referred to hospital were excluded from the analysis. Numbers of patients meeting the ILI, SARI-10, and SCRI-10 case definitions and numbers enrolled were collected throughout the study period. Outpatient care prior to hospitalization was also recorded for enrolled SARI-10 and SCRI-10 cases.
Age-and year-specific population denominators were obtained from projections of 2011 census data, 17 while age-and year-specific HIV prevalence in the study population was obtained from the projections of the THEMBISA Model. 18

| Laboratory procedures
Respiratory specimens (ie, nasopharyngeal aspirates for children aged <5 years and nasopharyngeal and oropharyngeal swabs from persons aged ≥5 years) were collected from all enrolled patients (ILI, SARI-10, and SCRI-10 cases), placed in universal transport me-

| Determination of HIV status
HIV results were obtained from a combination of two sources: (i) patient clinical records when available and (ii) for consenting patients, a dried blood spot was tested at NICD. Testing included HIV enzyme-linked immunosorbent assay (ELISA) for patients aged ≥18 months and PCR for children aged <18 months if the ELISA was reactive.

| Unadjusted and attributable fractionadjusted rates of influenza-associated respiratory hospitalizations and outpatient consultations
The details of the calculations of rates are provided in Supplementary Material. In brief, we estimated the overall and age-specific rates of influenza-associated SARI-10 hospitalizations per 100 000 population using the number of SARI-10 hospitalizations, adjusting for non-enrollment (refusal to participate and non-enrollment during weekends), healthcare-seeking behavior in the surveyed community, 20,21 and the INF-AF during 2013-2015. The INF-AF was obtained from published estimates derived from the same sentinel sites and study period. 12 The same approach was used to estimate the rates of influenzaassociated SCRI-10 hospitalizations.
We estimated the rates of influenza-associated ILI consultations per 100 000 population from the estimated SARI-10 rates using the proportion of SARI-10 cases that sought care at primary healthcare facilities prior to hospitalization and the proportion of ILI cases referred to hospital from ILI surveillance data.
For all calculations, we assumed that the influenza detection rate among individuals tested and not tested was the same within syndromes and age groups.
For all estimates the unadjusted and AF-adjusted rates, rate difference and percent reduction (PR) between unadjusted and AF-adjusted rates were reported.
Because our SARI-10 case definition among children aged <5 years differed from those recommended by WHO (ie, any hospitalized patient presenting with manifestation of acute lower respiratory tract infection with temperature ≥38°C or history of fever and cough), we also calculated the proportion of any influenza-positive respiratory hospitalizations that met the WHO case definition in this age group.

| HIV-stratified rates and risk of HIV infection on influenza-associated respiratory illness
We obtained the HIV-stratified rates of influenza-associated ILI, SARI-10, and SCRI-10 by multiplying the total number of influenza cases by the age-specific HIV prevalence among influenza-positive cases divided by the HIV serostatus-specific population at risk.
Subsequently, we estimated the HIV-stratified AF-adjusted rates of influenza-associated ILI, SARI-10, and SCRI-10 by multiplying the observed rates by the estimated age-and HIV serostatus-specific INF-AF. 12 Age-specific and overall age-adjusted relative risk (RR) for influenza-associated ILI, SARI-10, and SCRI-10 among HIV-infected and HIV-uninfected persons was estimated using log-binomial regression. This was done for both unadjusted and AF-adjusted rates.

| Sensitivity analysis of the impact of duration of symptoms on rates of influenza-associated hospitalizations
We implemented the above analysis after reclassifying the SARI and SCRI cases using a duration of symptoms cutoff of 7 days (referred to as SARI-7 and SCRI-7 hereafter). This classification was in accordance with practices previously recommended by WHO for global influenza surveillance. This analysis was implemented to assess the changes in rates of influenza-associated hospitalization among SARI cases as defined in previously published literature in Africa, 22

| Ethical approval
The SARI and SCRI protocols were approved by the University of the Witwatersrand Human Research Ethics Committee (HREC) and the University of KwaZulu-Natal Human Biomedical Research Ethics Committee (BREC) protocol numbers M081042 and BF157/08, respectively. The ILI protocol was approved by BREC protocol number (BREC BF 080/12). This surveillance was deemed non-research by the US Centers for Disease Control and Prevention (non-research determination number: 2012-6197). The HIV prevalence by age group and syndrome is provided in Figure   S1.

| Effects of the INF-AF on estimated rates
During the study period, the overall unadjusted mean annual influenzaassociated illness rates per 100 000 population were 1467.7 among ILI cases, 50.3 among SARI-10 cases, and 27.4 among SCRI-10 cases (  Figure 2). After adjusting for the INF-AF, the percent reduction (PR) from the overall unadjusted rates was 8.9% (AF-adjusted rate: 1336.9), 11.0% (AF-adjusted rate: 44.8) and 16.3% (AF-adjusted rate: 22.9) among ILI, SARI-10, and SCRI-10 cases, respectively. The 95% CIs overlapped between the unadjusted and AF-adjusted rates across syndromes, age groups, and HIV serostatus.
Overall, the effects of the INF-AF on the estimated rates were higher among HIV-uninfected (Table 2) than HIV-infected (Table 3)  The effects of the INF-AF on the estimated rates among SARI-7 and SCRI-7 cases (Table 4) were similar to those observed among SARI-10 and SCRI-10 cases (Table 1-3). The unadjusted and AFadjusted influenza-associated rates among any hospitalized patient are provided in Table S1 and Figure S2.
The effects of the AF of influenza virus infection to illness on the estimated RR among SARI-7 and SCRI-7 cases were similar to those observed among SARI-10 and SCRI-10 cases ( Table 5). The unadjusted RR and AF-adjusted RR of influenza-associated illness due to HIV infection among any hospitalized patient are provided in Table S1.

| Effects of the duration of symptoms on burden estimates of influenza-associated severe respiratory illness
Overall 39% of the AF-adjusted influenza-associated severe respiratory illness hospitalizations had symptom duration of ≤7 days prior to admission (Figure 3

| DISCUSSION
We described the effects of the INF-AF and duration of symptoms on influenza disease burden estimates among HIV-infected and HIVuninfected patients of different age groups with mild or severe respiratory illness. A minimal difference was observed between unadjusted and AF-adjusted rates or RR associated with HIV infection across syndromes, age groups, and HIV serostatus. Whereas shifting the symptom duration cutoff for inclusion in SARI surveillance from ≤7 to ≤10 days resulted in an overall 27% increase in estimated influenzaassociated hospitalization rates, 34% of patients with influenzaassociated severe respiratory illness of any duration had symptom The minimal difference between unadjusted and AF-adjusted rates observed in this study is likely a reflection of the high INF-AF used for adjustment in this study (ILI: 92.6%; SARI-10: 87.4%; SCRI-10: 86.2%). 12 HIV-infected patients were at increased risk of influenzaassociated mild or severe illness compared to HIV-uninfected patients. HIV infection has been reported to be associated with increased risk of influenza-associated SARI hospitalization in other studies. 4,23,24 The RR of HIV infection to illness among influenzapositive patients with ILI or SCRI has not been described to our knowledge.
The magnitude of the INF-AF used for adjustment in this study was highest among HIV-uninfected persons aged <1 and ≥65 years and T A B L E 1 Estimated mean annual rates of influenza-associated influenza-likeillness, severe acute respiratory illness (symptom duration ≤10 days) and severe chronic respiratory illness (symptom duration >10 days) among any patient (irrespective of HIV serostatus), Klerksdorp and Pietermaritzburg, South Africa, 2013-2015

Age group (in years)
Influenza-associated respiratory illness rates a (95% CI) Influenza-associated respiratory illness rates a (95% CI)

Rate difference d (% reduction)
Influenza-like illness Estimated rates without adjustment for the attributable fraction. c Estimated rates adjusted by the attributable fraction. d Unadjusted minus attributable fraction-adjusted rates. and cough as inclusion criteria for SARI surveillance as recommended by WHO may play more of an important role in the sensitivity of the case definition compared to symptom duration; however, this was not explicitly explored in this analysis.
In our study, individuals aged 5-24 years experienced the highest rates of influenza-associated ILI. This was similar to a study conducted in the United States whereby individuals aged 2-17 years experienced the highest rates of influenza-associated medically attended ILI, 25 suggesting the importance of this age group in the transmission of influenza viruses. 26 We observed a U-shaped trend of the magnitude of the influenza-associated SARI and SCRI hospitalization rates across age groups. Young children and older individuals are known to be at increased risk of influenza-associated severe illness. 4,23 In our study, the highest rates of influenza-associated SARI hospitalization were among children aged <1 year, whereas the highest rates of influenzaassociated SCRI hospitalization were among individuals aged ≥65 years. This suggests that using the recommended WHO SARI case definition may underestimate the burden of influenza-associated severe illness particularly in older individuals in our setting.
Our study has limitations that warrant discussion. First, we did not enroll ILI cases with symptoms duration >10 days hindering our ability to estimate the burden of influenza-associated medically attended T A B L E 3 Estimated mean annual rates of influenza-associated influenza-likeillness, severe acute respiratory illness (symptom duration ≤10 days) and severe chronic respiratory illness (symptom duration >10 days) among HIV-infected patients, Klerksdorp and Pietermaritzburg, South Africa, 2013-2015

Age group (in years)
Influenza-associated respiratory illness rates a (95% CI) developed influenza-associated severe illness, but did not seek care would have been missed in our study. Hence, our estimates should be considered minimum estimates focusing only on influenza-associated medically attended respiratory illness.
In conclusion, the AF-adjusted rates and RR due to HIV infection reported in this study reflect a more accurate description of the burden of influenza-associated mild and severe respiratory illness among HIV-infected and HIV-uninfected South African patients.
T A B L E 5 Estimated relative risk (RR) associated with HIV infection for influenza-associated influenza-like illness, severe acute respiratory illness (symptom duration ≤10 or ≤7 days), and severe chronic respiratory illness (symptom duration >10 or >7 days),