Etiology and Risk Factors for Mortality in an Adult Community-acquired Pneumonia Cohort in Malawi

Rationale: In the context of rapid antiretroviral therapy rollout and an increasing burden of noncommunicable diseases, there are few contemporary data describing the etiology and outcome of community-acquired pneumonia (CAP) in sub-Saharan Africa. Objectives: To describe the current etiology of CAP in Malawi and identify risk factors for mortality. Methods: We conducted a prospective observational study of adults hospitalized with CAP to a teaching hospital in Blantyre, Malawi. Etiology was defined by blood culture, Streptococcus pneumoniae urinary antigen detection, sputum mycobacterial culture and Xpert MTB/RIF, and nasopharyngeal aspirate multiplex PCR. Measurements and Main Results: In 459 patients (285 [62.1%] males; median age, 34.7 [interquartile range, 29.4–41.9] yr), 30-day mortality was 14.6% (64/439) and associated with male sex (adjusted odds ratio, 2.60 [95% confidence interval, 1.17–5.78]), symptom duration greater than 7 days (2.78 [1.40–5.54]), tachycardia (2.99 [1.48–6.06]), hypoxemia (4.40 [2.03–9.51]), and inability to stand (3.59 [1.72–7.50]). HIV was common (355/453; 78.4%), frequently newly diagnosed (124/355; 34.9%), but not associated with mortality. S. pneumoniae (98/458; 21.4%) and Mycobacterium tuberculosis (75/326; 23.0%) were the most frequently identified pathogens. Viral infection occurred in 32.6% (148/454) with influenza (40/454; 8.8%) most common. Bacterial–viral coinfection occurred in 9.1% (28/307). Detection of M. tuberculosis was associated with mortality (adjusted odds ratio, 2.44 [1.19–5.01]). Conclusions: In the antiretroviral therapy era, CAP in Malawi remains predominantly HIV associated, with a large proportion attributable to potentially vaccine-preventable pathogens. Strategies to increase early detection and treatment of tuberculosis and improve supportive care, in particular the correction of hypoxemia, should be evaluated in clinical trials to address CAP-associated mortality.


Exclusion criteria
Patients with any of the following were excluded from participation: symptoms for greater than 14 days; suspected co-existent meningitis; pre-admission diagnosis of terminal illness (e.g. metastatic malignancy, terminal AIDS); current anti-tuberculous treatment; admission to hospital more than 24 hours previously; prior hospitalisation within preceding 4 weeks; or prior participation in the study.

Clinical assessment and follow-up
For each participant, the study team completed a standardised clinical assessment consisting of a comprehensive medical history, physical examination and measurement of physiological observations. Information was obtained via direct questioning of the study participant (or accompanying guardian) and by reference to medical notes and health passport. Physiological observations were measured by the study team on admission or abstracted from the medical notes for patients recruited from the medical wards. In lieu of a validated cognitive assessment tool that is not available in Malawi, the presence of confusion was determined using three standardised questions assessing orientation in time, place and person.
Study participants were reviewed on a daily basis each morning until discharge or 14 days post-admission. At the point of hospital discharge or inpatient death, the study team reviewed the clinical notes and recorded details of treatment administered. Patients surviving to discharge were contacted at 30-days and 90-days post admission to determine vital status and hospital readmission. Two attempts were made to obtain this information by phone call to the patient or their guardian; if unsuccessful field workers visited the patient's home.
Patients with influenza formed the case population of a linked case-control study E4 describing the impact of HIV on influenza severity (E2). Some patients with radiographic pneumonia who survived to discharge also participated in a further case-control study of the association of exposure to indoor air pollution with the occurrence of pneumonia (E3).

Radiographic assessment
During the period of the study, there was no functional portable radiograph machine available at QECH; consequently, radiographs were not performed in clinically unstable patients that could not be safely transferred to the radiology department. Plain chest radiograph films were photographed on a light-box using a standard light-reflex digital camera mounted on a tripod in a darkened room. All study radiograph reports were generated by review of the set of digital images.
Chest radiographs were reported independently by two study radiologists (EJ and SG) and the study Principal Investigator (SA, an Infectious Diseases Resident Physician). All reporters were blinded to demographic and clinical data at the time of reporting.
Radiographs were reported using a standardised form, that was piloted prior to use for study reporting to ensure consistent application. Definitions of radiological features were based on the Fleischner Society: Glossary of terms for thoracic imaging (E4).
Parenchymal abnormalities were categorised as consolidation, reticulonodular change, miliary appearance and cavitation. Consolidation was further characterised in terms of its quality (confluent or patchy), extent (segmental, lobar or multifocal) and distribution (lower, mid-and/or upper zone, or diffuse). Radiographic pneumonia was defined as the presence of consolidation or other parenchymal abnormality (including reticulonodular change, cavitation or miliary appearance) or pleural effusion (E5). Multilobar consolidation was not specifically reported on the form; consolidation was considered to E5 be multilobar if characterised as diffuse, present in two or more non-contiguous lung zones regardless of extent or present in two or more contiguous lung zones and characterised as multifocal. Multifocal consolidation within a single lung zone was not classified as multilobar. Study analyses were based on a consensus report for each specific feature. If there was disagreement between the reporters, the majority opinion was used. Variability of interpretation was assessed by calculating average agreement and kappa coefficient.

Haematology and biochemistry
Full blood count was performed on whole blood specimens using a Beckman Coulter HmX Haematology Analyser (Beckman Coulter, California, USA). Whole blood specimens for biochemical assays were centrifuged within 24 hours of collection. Urea and creatinine concentrations were measured in serum specimens using a Beckman Coulter AU480 Chemistry Analyser (Beckman Coulter, California, USA). CD4 cell counts were measured using a Becton Dickinson FACSCount (Becton Dickinson, California, USA) or a Partec CyFlow CD4 Analyser (Sysmex Partec, Görlitz, Germany).
Urine was collected in standard containers and refrigerated. The BinaxNOW ® urine antigen test for Streptococcus pneumoniae (Alere, Massachusetts, USA) was performed in E6 accordance with the manufacturer's instructions (E8). The same procedure was applied for pleural fluid specimens. The BinaxNOW ® test for Legionella pneumophila urinary antigen was performed on urine specimens that had been stored at -80°C following collection.

Respiratory pathogen multiplex PCR
The nasopharyngeal aspirate (NPA) specimens were refrigerated immediately following collection and then divided into aliquots before being stored at -80°C in Universal Transport Medium (Copan, Brescia, Italy) for later batch-testing. Total nucleic acids were extracted from 300μl aliquots of each specimen with the Qiagen BioRobotÒ Universal System using the QIAamp One-For-All nucleic acid kit (Qiagen, Manchester, UK). Influenza A and B viruses were detected by real-time reverse transcription polymerase chain reaction (qRT-PCR) using the CDC Human Influenza RT-PCR diagnostic panel (CDC Influenza Division) (E9). Adenovirus, bocavirus, Chlamydophila pneumoniae, coronaviruses 229E, HKU1, OC43 and NL63, enterovirus, human metapneumovirus (hMPV), Mycoplasma pneumoniae, parainfluenza virus types 1, 2, 3 and 4, parechovirus, respiratory syncytial viruses (RSV) and rhinovirus were detected using the FTD Respiratory Pathogens 33 kit (Fast-track Diagnostics, Luxembourg) (E10). The results for the other organisms detected by the FTD 33 kit were disregarded. 5 or 10 μL of nucleic acid extract was used in each qRT-PCR reaction in combination with the AgPath one-step RT-PCR reagents (Applied Biosystems, Foster City, California, USA). For both kits, PCR conditions were according to the manufacturer's instructions. For the CDC Human Influenza RT-PCR diagnostic panel, samples with cycle threshold (Ct)-value <40 were recorded as positive; for the FTD 33 kit a cycle threshold (Ct)-value <33 was regarded as positive. Appropriate negative and positive control specimens were run alongside each reaction.
Sputum and pleural fluid microscopy were performed according to standard procedures.
Briefly, the most viscous part of the sputum specimen was aspirated and expelled onto a dry glass slide, spread to make a thin smear and stained using the Auramine O method.
Pleural fluid specimens were concentrated by centrifugation prior to smear preparation.
Smear specimens were each examined twice by two independent readers using LED fluorescence microscopy and reported according to standard criteria; discordant results prompted review by a third reader. All sample smears graded as scanty, 1+, 2+, and 3+ were defined as acid-fast bacilli (AFB) smear-positive.
Mycobacterial culture was performed according to standard procedures using the BACTEC MGIT 960 Mycobacterial Detection System (Becton Dickinson Diagnostic Systems, Sparks, Maryland, USA) as previously described (E11, E12). Briefly, following decontamination in sodium hydroxide and concentration by centrifugation, sputum and pleural fluid specimens were inoculated into mycobacterial growth indicator tubes (MGIT) tubes and incubated at 37°C in the BACTEC MGIT automated liquid culture system for up to 44 days. Cultured isolates identified as AFB on ZN microscopy were positively confirmed as Mycobacterium tuberculosis complex by microscopic cording and MPT-64 lateral flow assay (Capilia; TAUNS Laboratories, Numazu, Japan). AFB isolates that were negative on either confirmatory test were inoculated onto plain Löwenstein-Jensen (LJ) media and incubated at 25°C, 37°C and 45°C and onto paranitrobenzoic acid (PNB) and incubated at 37°C. Isolates that grew on LJ media at 37°C only were classified as M.
tuberculosis. Those that grew at 25°C or 45°C or on PNB were classified as nontuberculous mycobacteria and not speciated further. Positive cultures that did not reveal AFB on microscopy were re-cultured using a stored aliquot of the primary E8 specimen; if the same result was obtained, they were classified as contaminants or false positives.
The Cepheid Xpert MTB/RIF assay (Cepheid, Sunnyvale, California, USA) is a selfcontained, fully integrated, automated rapid diagnostic system that uses nested real-time PCR to detect M. tuberculosis genomic DNA in sputum and other clinical specimens (E13).
The assay was performed in accordance with the manufacturer's instructions (E14).

Assignment of etiology
A bacterial pathogen was determined to be present if: i) detected on blood or pleural fluid culture; ii) S. pneumoniae antigen was detected in urine or pleural fluid; iii) L.
pneumophila antigen was detected in urine; or iv) C. pneumoniae or M. pneumoniae was detected by PCR in NPA. Certain bacterial blood culture isolates were considered to be contaminants. Mycobacterial infection was determined to be present if detected in sputum or pleural fluid by mycobacterial culture; or if Mycobacterium tuberculosis was detected by Xpert MTB/RIF assay. Positive AFB smear microscopy of sputum or pleural fluid in the absence of positive culture or Xpert MTB/RIF assay was assumed to be M. tuberculosis infection. Respiratory viral diagnosis was based on detection by PCR in NPA.

Data management
The clinical data were collected in paper-based CRFs and subsequently converted to electronic form using Intelligent Character Recognition scanning software (TeleForm; Cardiff Software Developers, Cardiff, UK). Prior to scanning the CRFs were checked by the clinical team and subsequently by study investigators (SA, AH, HJ and JH) and any discrepancies were queried. Automated validation checks were built in to the character recognition scanning process to identify outliers and erroneous values. Once all data were entered into the study database, further systemic data checks were performed to detect E9 potential errors, which were checked and corrected to create a final dataset.

Statistical methods
Statistical analyses were performed with Stata version 12.1 (StataCorp; Texas, US). The statistical methods for comparing microbial etiology by radiographic and HIV status and for identifying independent risk factors for 30-day mortality are detailed in the main manuscript. The association of microbial etiology with 30-day mortality for the most commonly identified pathogens (i.e. S. pneumoniae, M. tuberculosis and influenza) was estimated by multivariable logistic regression analysis with age, sex and HIV status included as covariates.

CAP severity assessment tools
The following CAP severity-assessment tools were calculated as published and their prognostic performance to predict 30-day mortality was assessed.
 SWAT-Bp: One point awarded for each of the following components present: male sex; wasting; non-ambulatory status; temperature <35C or >38C; systolic blood pressure <100 mmHg or diastolic blood pressure <60 mmHg (E20).
To calculate the prognostic performance characteristics of the severity assessment tools, the recommended cutoffs used to define "severe pneumonia" were used: CURB65 3; CRB65 2; SMRT-CO 2; modified IDSA/ATS minor criteria 3; and SWAT-Bp 3. For each severity score, sensitivity, specificity, positive and negative predictive values, positive and negative likelihood ratios and area under the receiver-operating characteristic curve with 95% confidence intervals were calculated using standard methods. Patients with incomplete data were excluded from the analysis. Definition of abbreviations: ART = antiretroviral therapy; BP = blood pressure; CPT = co-trimoxazole preventative therapy; IQR = interquartile range. Data are n/N (%) unless indicated by * where median and interquartile range are shown. Variation in denominator reflects missing data unless further specified. † HIV status missing in 6 patients; CD4 count missing in 40 of all with HIV; CPT and ART usage missing in 4 and 43, respectively, with known HIV; ART duration missing in 32 reporting ART use. ‡ Chronic lung disease includes asthma, COPD and bronchiectasis. § Chronic heart disease includes congestive cardiac failure, cor pulmonale and dilated cardiomyopathy. ll Any previous episode of treated tuberculosis regardless of site and confirmation. ** Any prior episode within the last 5 years of a syndrome compatible with lower respiratory tract infection reviewed in a healthcare facility and treated with antibiotics. † † Included attendance to other hospital, health centre, private clinic, traditional healer or pharmacy. ‡ ‡ Excluded co-trimoxazole prophylaxis in HIV-infected individuals. § § Chest radiographs available in 421 patients; reports based on consensus grading of assessors and denominator may vary when consensus not obtained. llll Status at hospital discharge, day 30 and day 90 missing in 2, 20 and 41, respectively. Definition of abbreviations: CI = confidence interval. Data are n/N (%) unless otherwise specified. Chest radiographs were available in 421 of 459 patients; chest radiographs unavailable because of death prior to radiograph (n=14), hospital discharge without radiograph or prior to digital capture (n=18) and unclear reason (n=6). * Radiologist 1, radiologist 2 and clinician considered 2, 14 and 4 radiographs, respectively, as uninterpretable. † Consensus values calculated by majority rating of all assessors; denominator may vary when consensus not obtained, most notably for specific features of consolidation which were reported only when assessor regarded consolidation as present. ‡ Average agreement and three-way kappa (estimated 95% confidence intervals) calculated for all three assessors.