Dataset: 11.1K articles from the COVID-19 Open Research Dataset (PMC Open Access subset)
All articles are made available under a Creative Commons or similar license. Specific licensing information for individual articles can be found in the PMC source and CORD-19 metadata
More datasets: Wikipedia | CORD-19

Logo Beuth University of Applied Sciences Berlin

Made by DATEXIS (Data Science and Text-based Information Systems) at Beuth University of Applied Sciences Berlin

Deep Learning Technology: Sebastian Arnold, Betty van Aken, Paul Grundmann, Felix A. Gers and Alexander Löser. Learning Contextualized Document Representations for Healthcare Answer Retrieval. The Web Conference 2020 (WWW'20)

Funded by The Federal Ministry for Economic Affairs and Energy; Grant: 01MD19013D, Smart-MD Project, Digital Technologies

Imprint / Contact

Highlight for Query ‹2019 novel coronavirus respiratory syndrome screening

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

Study Setting and Design

We conducted a prospective observational study of adults hospitalized with CAP at Queen Elizabeth Central Hospital, a 1,200-bed teaching hospital that provides free healthcare to the 1.3 million residents of Blantyre district in Southern Malawi. Ethical approval was provided by the Research Ethics Committees of University of Malawi College of Medicine (P.11/12/1309) and Liverpool School of Tropical Medicine (13.02). Some patients participated in linked case-control studies describing the impact of HIV on influenza severity (21) and the association of indoor air pollution exposure with pneumonia (22).

Study Procedures

We recruited adults (≥18 yr) hospitalized with clinically diagnosed CAP defined as: reported or recorded fever (≥38°C), at least one relevant symptom (cough, chest pain, breathlessness, hemoptysis), and at least one focal chest sign (crepitations, pleural rub, bronchial breathing, percussive dullness, or diminished breath sounds) (adapted from Reference 23). Patients with symptoms for more than 14 days, current antituberculous treatment, or prior admission within the last month were excluded. Further details of the eligibility criteria are provided in the online supplement. Written informed consent was provided by the patient, or in the case of incapacity, by their accompanying guardian.

Patients underwent a standardized clinical assessment on admission and were reviewed daily during hospitalization and then followed at 30 and 90 days after admission to determine vital status. Blood, urine, sputum, and nasopharyngeal aspirate (NPA) were collected at enrolment. Diagnostic pleural aspiration was performed in patients with radiologically confirmed pleural effusion. Clinical care was provided in accordance with local guidelines and directed by the patients’ clinical team.

Radiographic Assessment

Chest radiographs were performed as early as possible after admission and independently reported by two radiologists and a study clinician using a standardized form. A consensus report of the three assessors was used for analysis. Radiographic pneumonia was defined as the presence of consolidation or other parenchymal abnormality (including reticulonodular change, cavitation, or miliary appearance) or pleural effusion (4). Additional details are provided in the online supplement.

Laboratory Testing

Hematologic and biochemical analyses were performed by standard methods. HIV status was established by sequential rapid tests (Determine, Alere; and Uni-Gold, Trinity Biotech) (24) and CD4 cell count was performed on FACSCount flow cytometer (Becton Dickinson). Blood cultures were performed using aerobic bottles in the BacT/ALERT 3D automated system (bioMérieux) and isolates identified using standard procedures (25). Urine antigen testing was performed for the detection of Streptococcus pneumoniae (BinaxNOW, Alere) and, in a subset of patients, Legionella pneumophila (BinaxNOW, Alere). A PCR assay was performed on NPA specimens for the detection of adenovirus; bocavirus; Chlamydophila pneumoniae; coronaviruses 229E, HKU1, OC43, and NL63; enterovirus; human metapneumovirus; influenza A and B viruses; Mycoplasma pneumoniae; parainfluenza virus types 1–4; parechovirus; respiratory syncytial viruses, and rhinovirus. Xpert MTB/RIF assay (Cepheid) and mycobacterial microscopy and culture (BACTEC MGIT 960 Mycobacterial Detection System, Becton Dickinson) were performed on noninduced sputum specimens. When obtained, pleural fluid samples were sent for Gram stain, aerobic culture, mycobacterial microscopy and culture, and pneumococcal antigen testing. Further details of laboratory methods are provided in the online supplement.

Statistical Analysis

Statistical analyses were performed with Stata version 12.1 (StataCorp). We tested for differences in continuous variables using Wilcoxon rank sum test, and categorical variables by chi-square test or Fisher exact test as appropriate. Differences in microbial etiology by radiographic status controlling for the effect of HIV were examined using the Mantel-Haenszel method.

Candidate risk factors for 30-day mortality were selected a priori based on literature review (26–29). Continuous variables, with the exception of age, were dichotomized at standard cutoff points (29–32). Univariable analyses were performed using logistic regression to explore associations with 30-day mortality. Multivariable models were generated using age, sex, HIV status, and all variables with a P value less than 0.2 on univariable analysis, prevalence of at least 5% and for which data were available in more than 95% of patients, and then rationalized by stepwise backward elimination with removal of variables with a P value greater than 0.05. The number of covariates in the multivariable model were limited to maintain at least 10 events per variable (33). Subgroup analyses of HIV-infected patients and those with radiographic pneumonia were performed. The association of etiology with mortality was also assessed (see online supplement). All available case information was used in each univariable analysis. In multivariable models, we excluded patients with missing data for included variables.

The prognostic performance of several CAP severity-assessment tools, such as CURB65 (29), CRB65 (29), SMRT-CO (simplified version of the SMART-COP tool) (34), modified Infectious Disease Society of America/American Thoracic Society 2007 minor criteria (35), and SWAT-Bp (28), for predicting 30-day mortality was described by calculating sensitivity, specificity, positive and negative predictive values, positive and negative likelihood ratios, and area under the receiver-operating characteristic curve with 95% confidence intervals (CI). Details of how scores were calculated for each severity-assessment tool are provided in the online supplement.

Baseline Characteristics

Between May 15, 2013, and January 31, 2015, we screened 1,711 adult patients, of whom 489 fulfilled the eligibility criteria, 472 were enrolled, and 459 included in the analysis (Figure 1). The median age was 34.7 years (interquartile range, 29.4–41.9) and 285 (62.1%) were male (Table 1; see Table E1 in the online supplement). HIV infection was common (355/453; 78.4%; HIV status missing in six individuals) and often newly diagnosed (124/355; 34.9%), whereas other comorbid illnesses were reported infrequently (31/451; 6.9%). A total of 83.3% (189/227) of those known to be HIV-infected reported current ART use. Prior pneumonia was common (108/457; 23.6%) and associated with HIV (odds ratio [OR], 2.91; 95% CI, 1.46–6.30). The median length of admission was 7 days (interquartile range, 4–10).

Radiographic pneumonia was identified by consensus report in 317 (76.0%) of 417 patients with available interpretable chest radiographs with moderate interobserver agreement (Kappa, 0.75; 95% CI, 0.69–0.81) (Table 1; see Table E2). Pleural effusion was identified in 118/416 (28.4%), being the sole basis of defining radiographic pneumonia in 38/315 (12.1%). The proportion of patients with radiographic pneumonia did not vary with HIV status (HIV-infected 247/320 [77.2%] vs. HIV-uninfected 66/91 [72.5%]; P = 0.36) or with CD4 count (CD4 < 200 cells/mm3 176/222 [79.3%] vs. CD4 ≥ 200 cells/mm3 49/64 [76.6%], P = 0.64; CD4 count available in 315/355 [88.7%] HIV-infected patients).


Blood for aerobic culture was obtained from 450 of 459 (98.0%) of the participants, NPA from 455/459 (99.1%), and urine specimen for S. pneumoniae antigen detection from 433/459 (94.3%) with stored specimens available for Legionella antigen testing in 193 (see Table E5). At least one sputum specimen for a mycobacterial diagnostic test was obtained from 322/459 (70.2%) participants: smear microscopy, mycobacterial culture, and Xpert MTB/RIF assay were completed in 305/459 (66.4%), 273/459 (59.5%), and 308/459 (67.1%), respectively. A pleural fluid specimen was obtained in 35/459 (7.6%) patients: aerobic bacterial culture, S. pneumoniae antigen detection, and mycobacterial smear microscopy and culture were completed in 31, 31, and 35 patients, respectively.

Overall at least one potential pathogen was identified in 278/459 (60.6%) patients including at least one bacterium in 125/452 (27.7%) and at least one virus in 148/459 (32.6%) (Table 2). S. pneumoniae was the most commonly identified organism, present in 98 (21.4%) of 458 for whom results of blood culture and/or urinary antigen assay were available, of whom 92 (93.9%) were identified by detection of urinary antigen alone (see Table E5). Pneumococcal etiology was more common in HIV-uninfected patients (30/97 [30.9%] vs. 68/355 [19.2%]; P = 0.01) and those with radiographic pneumonia (83/316 [26.3%] vs. 8/100 [8.0%]; P < 0.001) (see Table E6), in particular confluent consolidation (OR, 2.51; 95% CI, 1.26–4.98) (see Table E7). Overall, 26 (5.8%) of 450 patients for whom blood was sent for culture were bacteremic: Salmonella enterica serovar Typhi (n = 9), nontyphoidal Salmonella (n = 7), and S. pneumoniae (n = 5) were most commonly isolated. S. Typhi was identified more frequently in HIV-uninfected patients (5/94 [5%] vs. 3/350 [1%]; P = 0.02), whereas all cases of nontyphoidal Salmonella infection occurred in HIV-infected patients, and both were more common among patients lacking radiographic pneumonia (see Table E6). Infection with M. pneumoniae (6/455; 1.3%) and C. pneumoniae (2/455; 0.4%) were uncommon. None of the 193 patients tested had a positive Legionella urinary antigen assay result.

Overall, M. tuberculosis was identified in 75 (23.0%) of 326 patients who submitted a sputum and/or pleural fluid specimen (Table 2). TB was confirmed by culture in 65/75 (86.7%) and diagnosed on the basis of Xpert MTB/RIF alone, sputum microscopy alone, or pleural fluid microscopy alone in 8/75 (10.7%), 1/75 (1.3%), and 1/75 (1.3%) patients, respectively (see Table E5). Nontuberculous mycobacteria were isolated in 8/273 (2.9%) for whom sputum for culture was obtained, of whom one was smear positive. The frequency of TB did not differ significantly between the groups with and without radiographic pneumonia (58/232 [25%] vs. 13/68 [19.1%]) (see Table E6), although in the latter, TB displaced S. pneumoniae as the most commonly identified pathogen, with all cases occurring in HIV-infected patients.

Influenza (40/454; 8.8%), adenovirus (35/455; 7.7%), and coronavirus (31/455; 6.8%) were the most commonly detected viruses. Detection of these respiratory viruses was not associated with HIV status or the presence of radiographic pneumonia (Table 2; see Table E6).

Coinfection relationships were investigated in 307 patients (66.9% of cohort) for whom results for blood culture, pneumococcal urine antigen, sputum mycobacterial culture and/or Xpert MTB/RIF, and NPA PCR were available. At least one organism was detected in 206/307 (67.1%) and coinfection with two or more organisms was present in 67/307 (21.8%) (Figure 2); the combination of bacterial–viral coinfection was most frequent (28/307; 9.1%). S. pneumoniae was codetected in one-fifth of patients with influenza (6/31; 19.4%). M. tuberculosis was isolated in 9.6% (7/73) of those with S. pneumoniae (see Table E8). Detection of multiple organisms did not vary with HIV status (54/243 [22.2%] vs. 12/62 [19.4%]).


A total of 439/459 (95.6%) patients were successfully followed to 30 days, of whom 64 (14.6%) had died, including 15 and 31 within the first 3 and 7 days following admission, respectively (Table 1). A total of 418 were followed to 90 days, of whom an additional 23 had died. Male sex (adjusted OR [aOR], 2.60; 95% CI, 1.17–5.78), pre-presentation symptom duration greater than 7 days (aOR, 2.78; 95% CI, 1.40–5.54), heart rate greater than or equal to 125 per minute (aOR, 2.99; 95% CI, 1.48–6.06), oxygen saturations less than 90% (aOR, 4.40; 95% CI, 2.03–9.51), and inability to stand (aOR, 3.59; 95% CI, 1.72–7.50) were independently associated with 30-day mortality (Table 3). Neither age nor any underlying comorbid illness, including HIV, or initial antimicrobial treatment was found to be significantly associated with 30-day mortality. Detection of S. pneumoniae was associated with reduced mortality (aOR, 0.40; 95% CI, 0.17–0.91) (see Table E9), whereas TB was associated with increased mortality (aOR, 2.44; 95% CI, 1.19–5.01). Mortality did not vary with detection of multiple organisms compared with a single organism (5/65 [7.7%] vs. 13/135 [9.6%]; P = 0.80).

In a subgroup analysis of the 342 HIV-infected patients followed to Day 30, symptom duration greater than 7 days (aOR, 3.56; 95% CI, 1.60–7.93), heart rate greater than or equal to 125 per minute (aOR, 3.06; 95% CI, 1.40–6.71), oxygen saturations less than 90% (aOR, 2.97; 95% CI, 1.28–6.88), inability to stand (aOR, 4.25; 95% CI, 1.84–9.79), and CD4 count less than 50 cells/mm3 (aOR, 2.30; 95% CI, 1.07–4.92) were associated with mortality in multivariable analysis (see Table E10). CD4 count less than 200 cells/mm3 was not associated with mortality (OR, 1.63; 95% CI, 0.69–3.83). In the subgroup of 305 patients with radiographic pneumonia followed to Day 30, symptom duration greater than 7 days (aOR, 3.30; 95% CI, 1.37–7.96), heart rate greater than or equal to 125 per minute (aOR, 2.66; 95% CI, 1.08–6.57), and multilobar consolidation (aOR, 2.75; 95% CI, 1.17–6.47) were significantly associated with mortality, with a trend toward an association for inability to stand (aOR, 2.63; 95% CI, 1.00–6.89) (see Table E11). Thirty-day mortality was, however, significantly higher in the small group of patients for whom a chest radiograph was unavailable (17/36 [47.2%] vs. 47/403 [11.7%]; P < 0.0001) (see Table E3).

Prognostic Performance of Severity-Assessment Tools

Of the CAP severity-assessment tools examined, SMRT-CO had the highest sensitivity (89.7%; 95% CI, 72.6–97.8), negative predictive value (96.8%; CI, 91.0–99.3%), and overall discriminatory capability (area under receiver-operating characteristic curve, 0.66; 95% CI, 0.57–0.75) for 30-day mortality (Table 4). CURB65 had poor sensitivity and discriminatory capability. Only 38 (8.8%) patients had “severe CAP” by CURB65 (i.e., score ≥ 3) and mortality among those with “mild CAP” (i.e., CURB65 ≤ 1) was 10.8% (29/268). The prognostic performance of each tool was similarly poor in the subgroup with radiographic pneumonia (see Table E12).


In this prospective study of acute CAP in adults that is among the largest conducted in sub-Saharan Africa, we have shown that: the burden of hospitalized pneumonia in adults predominantly remains in young, HIV-infected patients; TB and vaccine-preventable pathogens, such as S. pneumoniae and influenza, are common; and mortality is substantial and associated with potentially modifiable risk factors but poorly predicted by widely used CAP severity-assessment tools.

Although completed within the era of ART scale-up, 35% of HIV-infected patients were newly diagnosed, 17% of those with known HIV were not receiving ART, and more than three-quarters had advanced immunosuppression (i.e., CD4 < 200 cells/mm3). Further substantial reductions in CAP incidence may therefore be achieved by improved HIV testing and linkage to early initiation of ART (36).

At first glance, the overall crude mortality rate of 14.6% is comparable with that seen in well-resourced settings (26, 29, 37). However, a recent U.K. cohort reported 30-day mortality of 1.5% in patients younger than 50 (38), suggesting a substantially higher age-adjusted mortality rate in our cohort. The identified associations of TB and uncorrected hypoxemia with death, may explain some of this apparent large discrepancy in mortality rate. In keeping with other studies of acute CAP (7, 27, 28), we did not find evidence that HIV overall influenced outcome; increased mortality was evident in only the most profoundly immunocompromised with CD4 cell count less than 50 cells/ml.

Given the observed strong association of hypoxemia with mortality, expansion of supplemental oxygen provision, which across much of sub-Saharan Africa is currently severely limited (39), represents an obvious strategy to be evaluated to improve CAP outcomes. Programmatic interventions in children in low-resource settings based on improved oxygen supply using oxygen concentrators have been associated with a 35% reduction in mortality (40). The lack of association between respiratory rate and mortality underlines the importance of expanding the availability of pulse oximetry in tandem. Similarly, the association of pre-presentation symptom duration with mortality suggests that interventions to improve patient education to alter health care–seeking behavior or referral mechanisms within community healthcare centers should be considered. The continued substantial mortality rate beyond Day 30 also warrants further study to devise targeted interventions to enhance care following discharge from hospital.

The use of objective CAP severity-assessment tools has been shown to improve the recognition of patients with severe CAP likely to require intensive care support (41) and those with mild disease who can be safely managed as outpatients (42). Current South African CAP guidelines recommend that CURB65 is used to inform site of care and initial antimicrobial treatment (43). However, we found that CURB65 showed poor prognostic performance in predicting 30-day mortality in this cohort compared with its performance in CAP cohorts from well-resourced settings (44). This variation likely reflects differences in microbial etiology, demographics, and comorbidity profiles between this cohort and the patient populations in which CURB65 was derived (29), and underlines the importance of prospective validation of prognostic tools in relevant settings (29, 44). The similarly poor performance of the SWAT-Bp score that was derived in a lower respiratory tract infection cohort from Malawi (28), however, highlights the challenges of developing a simple and accurate locally adapted tool. Rather than attempting to derive a single tool capable of accurately predicting mortality across the whole spectrum of disease severity, a tool with high negative predictive performance aimed at identifying those patients with a low risk of adverse outcome suitable for outpatient management may be more feasible and equally useful. In this respect, the SMRT-CO tool performed well with a negative predictive value of 96.8%, but the inclusion of a radiographic parameter (i.e., multilobar involvement) (34) limits its use as a triage tool in settings where chest radiography might not be readily available.

Recent CAP studies corroborate our observation that S. pneumoniae remains the commonest identified cause of adult CAP in sub-Saharan Africa (6, 10, 27). Universal infant vaccination with pneumococcal conjugate vaccination was introduced in Malawi in 2011, approximately 2.5 years before the start of recruitment. Following the introduction of pneumococcal conjugate vaccination in the United States and South Africa, rates of adult pneumococcal disease (including pneumonia hospitalizations) fell rapidly as a result of indirect protection caused by reduced pneumococcal transmission from vaccine recipients (45, 46). Further studies are needed to describe the serotype distribution of pneumococcal pneumonia in Malawi, to determine whether targeted vaccination programs of at-risk groups may be beneficial to tackle the persistent burden of adult disease. Recently developed urinary assays that can detect serotype-specific antigens in noninvasive pneumococcal pneumonia may provide a more comprehensive description (47).

When adequate diagnostic tests are performed, TB is identified in a substantial proportion of patients presenting with acute CAP in sub-Saharan Africa with the major burden of disease and associated mortality occurring in HIV-infected individuals (7–9; 27). At 23% of those tested and 16% of the overall cohort, the frequency of TB is similar to that previously reported (7–9; 27), but probably underestimates the true burden given the reliance on spontaneously expectorated sputum specimens and a single specimen submitted for culture. In most sub-Saharan African settings, laboratory diagnosis of TB has until recently relied on sputum smear microscopy that has poor sensitivity, particularly for HIV-associated TB. The World Health Organization has advocated an empirical “step-up” approach, whereby antituberculous treatment is initiated for HIV-infected adults with features of severe respiratory infection and negative sputum smears who fail to improve after 3–5 days of antibacterial treatment (48). Although this approach may limit inappropriate antituberculous treatment, it may miss up to 20% of patients with culture-positive TB and risks delaying treatment in a group of patients with a very high risk of early death (49). The increasingly available Xpert MTB/RIF assay used in this study offers much improved sensitivity and is now recommended as the first-line diagnostic for HIV-associated TB (50). Rapid urine-based diagnostics (e.g., urine Xpert MTB/RIF, mycobacterial glycopeptide lipoarabinomannan assay) may have a complementary role to guide initiation of antituberculous treatment, particularly for acutely ill patients from whom sputum is frequently unobtainable (51, 52).

In keeping with recent CAP studies from well-resourced settings that have used molecular diagnostics (4, 53), we identified high rates of coinfection with multiple organisms, most commonly bacterial–viral coisolation. Defining the clinical significance of each organism detected is challenging, particularly among immunocompromised individuals. Although coisolation may reflect genuine copathogenicity or simply detection of bystander colonizing organisms, more complex relationships in which one organism facilitates the acquisition or pathogenesis of another (e.g., influenza and S. pneumoniae) (54) are possible. Bacterial–viral coinfection has been associated with altered levels of innate immune factors (e.g., CRP and IP10) (55) and a more severe clinical course (56). Further work is needed to define the implications for CAP therapy and prevention.

Radiographic facilities are limited in most healthcare facilities in sub-Saharan Africa where the majority of CAP patients are initially managed. Although the clinical case definition used in this study may have missed some patients with radiographic pneumonia who lacked focal signs at the time of recruitment it nonetheless reflects routine clinical practice and usefully provides data to show how the etiologic spectrum of disease varies with chest radiograph findings (57). The association of radiographic changes compatible with infection with S. pneumoniae is unsurprising. Although still relatively uncommon, the greater prevalence of invasive Salmonella infection among those lacking radiographic pneumonia has important implications for antimicrobial management. Although specific radiologic features were statistically associated with certain pathogens, the marked interobserver variability in interpretation of these features suggests that the clinical utility of this approach is limited.

In recruiting patients exclusively from a single referral hospital, this study shares the limitations common to many CAP observational cohorts. Most patients were seen at another facility before attendance; how they were selected for referral and, correspondingly, the characteristics and outcome of those treated at home is unknown. Hence, the applicability of the findings to community-level settings is potentially limited. Extrapolation of these findings to other countries in sub-Saharan Africa must also be done cautiously because of the potential impact that variations in such factors as TB incidence, HIV prevalence, and climate may have on CAP epidemiology and etiology. Given limitations of the healthcare infrastructure in Malawi, chronic comorbid illnesses may be underdiagnosed. Inconsistencies in drug supply, equipment provision, and staff experience may result in variations in CAP treatment that may impact patient outcome but are not fully accounted for in analyses.

The description of CAP etiology was limited by the lack of sputum microscopy and culture, which may have increased the detection of bacterial pathogens, such as Haemophilus influenzae as has been described in previous studies from the region (27, 58). Similarly, we were unable to support invasive respiratory sampling for diagnostic testing for Pneumocystis jirovecii and other HIV-associated opportunistic pathogens. The low rate of bacteremia observed may relate to both prehospital antibiotic use and antibiotic initiation before blood couture collection in hospital, which occurred in up to one-third of patients. Interpretation of the significance of respiratory pathogen multiplex PCR results, particularly for respiratory viruses, is hampered by the lack of data from an appropriate control population. The relative contribution of pathogens with seasonal transmission patterns (e.g., influenza) may also have been skewed by a recruitment period that spanned 2 incomplete years.

In conclusion, more than a decade after the introduction of ART, the major burden of CAP in Malawi remains in young patients with HIV rather than the elderly patients with chronic noncommunicable comorbidities that predominate in well-resourced settings. Accordingly, context-specific approaches to severity assessment and defining empirical antimicrobial treatment are needed. The predominance of S. pneumoniae and influenza suggests significant opportunities for CAP prevention through targeted vaccination programs. Strategies to encourage prompt patient presentation, to increase early detection and treatment of TB, and to improve supportive care, in particular the correction of hypoxemia, should be evaluated in clinical trials to address the high burden of CAP-related mortality.