Study design
We conducted prospective hospital-based active surveillance for Severe Acute Respiratory Illness (SARI) surveillance at 6 public hospitals in 4 surveillance sites situated across the country: Chris Hani Baragwanath Academic Hospital (CHBAH), Gauteng Province; Edendale Hospital, KwaZulu-Natal Province; the two hospitals of the Klerksdorp–Tshepong Hospital Complex (KTHC), Northwest Province; and Mapulaneng and Matikwana Hospitals, Mpumalanga Province [13]. At each sentinel hospital trained surveillance officers screened all admitted patients for surveillance inclusion criteria from Monday to Friday, except at CHBAH where adult patients were screened on 2 of every 5 days per week because of large patient numbers.
Case definitions
A case of SARI was defined as a hospitalised individual with onset of illness within seven days of admission meeting age-specific clinical inclusion criteria. We included children aged two days through <3 months with physician-diagnosed sepsis or acute lower respiratory tract infection (ALRI), children aged three months through <5 years with physician-diagnosed ALRI (including, for example bronchitis, bronchiolitis, pneumonia and pleural effusion) and patients aged ≥5 years meeting a modified World Health Organization (WHO) case definition for SARI including sudden onset of fever (>38°C) or reported fever, cough or sore throat, and shortness of breath or difficulty breathing [14]. In addition expanded case definitions were applied at three of the hospitals in two of the surveillance sites, (Edendale Hospital and KTHC) to include individuals meeting the above case definition but with symptoms for >7 days and individuals with suspected or confirmed tuberculosis; these patients were defined as having severe chronic respiratory illness (SCRI). For this analysis severe respiratory illness (SRI) was used to refer to all enrolled patients i.e. patients admitted with an acute or chronic respiratory illness.
The South African National Department of Health tuberculosis case definitions were modified to define suspected tuberculosis [15]. A suspected tuberculosis case in children <12 years was defined as: i) any child presenting with a history of exposure to an infectious tuberculosis case or with a positive tuberculin skin test and symptoms of tuberculosis (chronic cough, weight loss and fever) with or without an abnormal chest x-ray suggestive of tuberculosis, or ii) admitted with a physician diagnosis of tuberculosis or suspected tuberculosis, or iii) initiated on tuberculosis treatment on admission; or iv) was on tuberculosis treatment for <2 months at the time of admission. A suspected tuberculosis case in individuals ≥12 years of age was defined as: i) any patient presenting with night sweats, chronic cough, weight loss or haemoptysis lasting for >2 weeks, or ii) had a physician diagnosis of tuberculosis or suspected tuberculosis, or iii) was initiated on tuberculosis treatment on admission, or iv) was on tuberculosis treatment for <2 months at the time of admission.
Patients who were on tuberculosis treatment for ≥2 months at the time of admission were not enrolled into the surveillance programme as cases of suspected or confirmed tuberculosis (because we assumed that after two months of treatment tuberculosis was no longer the reason for their clinical presentation). These patients were, however, enrolled if they met the criteria for SARI. The tuberculosis status of the enrolled patients was based on results of tuberculosis testing at current admission.
For this analysis, a laboratory-confirmed tuberculosis case was defined as an individual with a positive result for Mycobacterium tuberculosis on microscopy, culture or polymerase chain reaction (PCR) by GeneXpert MTB/RIF test (Cepheid, Sunnyvale, California) from the current hospital admission or from a specimen taken within two weeks preceding or following the admission. An influenza case was defined as an individual with a positive PCR test for influenza. An influenza-tuberculosis co-infection case met criteria for both laboratory-confirmed tuberculosis and influenza during the same admission.
Data collection
A standardized questionnaire was used to collect demographic and clinical data, medical history of the patient and in-hospital outcome. Hospital and intensive care unit (ICU) admission and collection of specimens for bacterial culture, tuberculosis testing and CD4+ T-cell counts were performed according to attending-physician discretion.
Sample collection and processing
Respiratory specimens (oropharyngeal and nasopharyngeal swabs for patients ≥5 years of age or nasopharyngeal aspirates for children <5 years of age), were collected and placed in 4 ml virus transport medium. Whole blood samples were collected in EDTA-containing vacutainer tubes within 24 hours of hospital admission for the detection of Streptococcus pneumoniae and HIV infection.
After collection, respiratory and blood samples were kept at 4°C at the sentinel site, and transported on ice at least twice per week to the National Institute for Communicable Diseases of the National Health Laboratory Services (NICD-NHLS) for testing.
Detection of respiratory viruses and S. pneumoniae
Respiratory samples were tested by multiplex real-time reverse-transcription polymerase chain reaction (PCR) (rRT-PCR) for influenza type A and B, adenovirus, enterovirus, rhinovirus, human metapneumovirus, respiratory syncytial virus and parainfluenza virus types 1–3 [16]. The influenza A positive specimens were subtyped using the Centers for Disease Control and Prevention (CDC) rRT-PCR protocol for detection and characterization of influenza. DNA was extracted from whole blood specimens using the MagNA Pure LC 2.0 instrument and DNA Isolation kit III for bacteria (Roche, Mannheim, Germany) and were tested for the S. pneumoniae lytA gene by a quantitative real-time PCR [17].
Determination of HIV infection
HIV status data were obtained from two data sources. Some patients had HIV testing requested by admitting physicians as part of clinical care. This included HIV enzyme-linked immunosorbent assay (ELISA) testing with confirmation by ELISA on a second specimen for patients ≥18 months of age and qualitative HIV PCR testing for confirmation of HIV-infection status in children <18 months of age. In addition, for consenting patients, linked anonymous HIV PCR testing for children <18 months of age or ELISA for patients ≥18 months of age was performed using a dried blood spot or whole blood specimen at the NICD-NHLS laboratory.
Determination of tuberculosis infection
Testing for M. tuberculosis, including microscopy, culture, GeneXpert MTB/RIF test, or a combination of these, was undertaken at the discretion of the attending-physician and performed at the laboratory serving the hospital where the patient presented. Specimens for tuberculosis testing were examined by light microscopy for the presence of acid fast bacilli. Tuberculosis culture was performed using the BACTEC MGIT automated culture system (Becton Dickinson, Franklin Lakes, New Jersey). Tuberculosis PCR was performed using the Xpert MTB/RIF system (Cepheid, Sunnyvale, California). The GeneXpert MTB/RIF is an automated sample processing and nucleic acid amplification test for detection of M. tuberculosis and is able to detect resistance to rifampicin. The National Department of Health tuberculosis treatment guidelines recommended treatment for patients with positive smears and that patients suspected of tuberculosis should have two specimens tested on microscopy and if these were negative then a 3rd sample should be tested for smear and culture [15].
Data analysis
To identify factors associated with tuberculosis testing, tuberculosis positivity and tuberculosis-influenza co-infection we included both potential determinants for, as well as outcomes or characteristics of the primary endpoints of the analysis. Univariate comparisons were performed using logistic or multinomial regression. In addition, we implemented three multivariable models to identify factors associated with: (i) tuberculosis testing among enrolled patients; (ii) tuberculosis positivity among enrolled patients tested for tuberculosis; and (iii) tuberculosis single infection compared to influenza single infection or tuberculosis-influenza co-infection. The tuberculosis testing and positivity models were implemented using stepwise forward selection logistic regression. Multinomial regression was used for the comparison of tuberculosis only, influenza only and tuberculosis-influenza co-infected groups and this analysis was performed separately in individuals with symptoms for <7 days and ≥7 days. Multinomial regression allows modelling of outcome variables with >2 categories and relates the probability of being in category j to the probability of being in a baseline category. A complete set of coefficients are estimated for each of the j levels being compared with the baseline and the effect of each predictor in the model is measured as relative risk ratio (RRR). For this analysis, we used the tuberculosis single infection group as the baseline category and compared it with the influenza single infection and tuberculosis-influenza co-infection groups and we restricted the analysis to laboratory confirmed cases for both influenza and tuberculosis. The general form of the multivariable multinomial model with dichotomous predictors was as follows:
$$ \ln \frac{p\left({Y}_j\right)}{p\left({Y}_1\right)}={\beta}_0^{(j)}+{\beta}_1^{(j)}{X}_1+\dots +{\beta}_n^{(j)}{X}_n+\varepsilon $$
(1)
Where p(Y
1
) is the probability of the outcome variable in the base category (tuberculosis single infection) and p(Y
j
) is the probability of the outcome variable in the other categories; j represents the two other categories (j = 2: influenza single infection; and j = 3: tuberculosis-influenza coinfection) that are compared to the base category; \( {\beta}_0^{(j)} \) is the model constant for category j; \( {\beta}_1^{(j)} \) to \( {\beta}_n^{(j)} \) are coefficients associated with predictors X
1
to X
n
in category j; and ε is the error term.
The model in equation 1 was fitted to any enrolled case with available tuberculosis and influenza results irrespective of duration of symptoms as well as separately among individuals with acute (duration of symptoms <7 days) and chronic (duration of symptoms ≥7 days) clinical presentation to specifically assess the clinical characteristics of individuals within these two groups. In order to allow for direct comparison of the final multivariable models for acute and chronic patients we elected to retain any variable which was significant at 0.05 in either one of these models in the final multivariable model. Covariates with a p-value <0.2 at the univariate analysis were assessed for significance at the multivariable analysis and statistical significance was assessed at p < 0.05 for all models. Statistically significant variables were retained in the multivariable models. HIV status was retained in the multinomial models a priori as this is a potential confounder of the interaction between influenza and tuberculosis. Two-way interactions were assessed by inclusion of product terms for all variables remaining in the final additive models. The statistical analysis was implemented using STATA® version 12 (StataCorp, Texas, USA).
Ethics
Ethical approval for the study was obtained from the University of the Witwatersrand, Human Research Ethics committee (reference M081042) and the University of KwaZulu-Natal Biomedical Research Ethics committee (reference BF157/08). The United States Centers for Disease Control and Prevention deemed this surveillance not to be research and therefore this project did not need human subjects review.
