Clinical features for diagnosis of pneumonia among adults in primary care setting: A systematic and meta-review

Pneumonia results in significant morbidity and mortality worldwide. However, chest radiography may not be accessible in primary care setting. We aimed to evaluate clinical features and its diagnostic value to identify pneumonia among adults in primary care settings. Three academic databases were searched and included studies that assessed clinical predictors of pneumonia, adults without serious illness, have CXR and have conducted in primary care settings. We calculated sensitivity, specificity, positive and negative likelihood ratios, diagnostic odds ratio of each index test and the pool estimates for index tests. We identified 2,397 articles, of which 13 articles were included. In our meta-analysis, clinical features with the best pooled positive likelihood ratios were respiratory rate ≥20 min−1 (3.47; 1.46–7.23), temperature ≥38 °C (3.21; 2.36–4.23), pulse rate >100 min−1 (2.79; 1.71–4.33), and crackles (2.42; 1.19–4.69). Laboratory testing showed highest pooled positive likelihood ratios with PCT >0.25 ng/ml (7.61; 3.28–15.1) and CRP > 20 mg/l (3.76; 2.3–5.91). Cough, pyrexia, tachycardia, tachypnea, and crackles are limited as a single predictor for diagnosis of radiographic pneumonia among adults. Development of clinical decision rule that combine these clinical features together with molecular biomarkers may further increase overall accuracy for diagnosis of radiographic pneumonia among adults in primary care setting.


Materials and Methods
Search strategy and selection criteria. The study was performed in accordance with the recommendations of the Preferred Reporting Items for Systematic Review and Meta-Analyses (PRISMA, Appendix S1) 15 . The meta-review does not involve human participants or experiments on live vertebrates and/or higher invertebrates, only published aggregate data from the selected studies is used in the meta-analysis. The study protocol is available as Supplementary material (Appendix S2). The search strategy consists of two phases. The first phase was an extensive search using the identified index terms and keywords in three databases: PubMed, EMBASE and the Cochrane Library. The second phase was an additional search of the references of retrieved articles to find any articles that did not appear in the databases search. The keywords that were used in the search are 'pneumonia' or 'community acquired pneumonia' or 'community-acquired pneumonia' or 'respiratory tract infection' or 'respiratory tract infections' and 'predictive value of tests' or 'sensitivity and specificity' or 'diagnostic test' or 'diagnostic tests' or 'medical history taking' or 'medical history' or 'physical examination' or 'physical examinations' or 'clinical laboratory techniques' or 'laboratory diagnoses' or 'laboratory examinations' or 'laboratory testing' and 'ambulatory care' or 'primary care' or 'outpatient care' or 'general practitioner' or 'emergency clinic' . The bibliographical software package, EndNote version X7 (Thomas Reuters, New York, NY, USA), was used to import references and to remove duplicates references. The remaining studies were checked against the inclusion and exclusion criteria. Two reviewers (HTP and CHL) independently screened eligibility based on title, abstracts and assessed full reports, resolving discrepancies by consensus.
Studies were selected if they were published studies that assessed clinical predictors of community-acquired pneumonia without date restrictions to maximize the search. The first search was employed on Dec 4, 2017, with an update on Mar 5, 2018. Narrative review, letters to editors, case reports and case series were excluded. Studies were included if participants aged ≥18 years without serious illness (e.g. mechanical ventilation) and pre-existing immune suppression (HIV, malnutrition, and immunosuppressant medication). To be eligible, studies had to have reference standard of CXR for diagnosing pneumonia, and have conducted in ambulatory care or primary care settings. Index tests assessed were patient's socio-demographic, clinical signs and symptoms and laboratory tests.

Quality assessment.
After identifying studies that fulfilled the selection criteria and verifying their eligibility by reading the full articles, the quality assessment of the studies were done by using QUADAS-2 16 as recommended by the Cochrane collaboration. Studies were assessed for selection of patient, index test, reference standard, and flow and timing. Signalling questions were made to facilitate the rating of risk of bias into low, unclear or high. Data extraction. The following variables were extracted from each study using pre-designed forms: study characteristics (study design, year of publication, country and setting), study population (age, number of participants recruited, prevalence of pneumonia, inclusion and exclusion criteria), reference standard (number of readers, masking and interpretation criteria), and index tests. Index tests were classified as related to socio-demographic, clinical symptoms or signs, and laboratory tests to diagnose pneumonia. Outcome data were extracted and compiled in a table by one author (HTP). After which, all extracted data were cross-checked by another author (SY) by comparing them to the original data from the selected articles. Data analysis. Data analysis was based on a published methodological review -a systematic review of evaluations of diagnostic and screening tests 17 . We constructed 2 × 2 tables for each study included in the review to calculate sensitivity, specificity, positive and negative likelihood ratios and diagnostic odds ratio with 95% confidence intervals (CI). The likelihood ratio indicates the value of the test for increasing certainty about a positive diagnosis. The positive likelihood ratio (LR+) is the probability of a positive result in patients with the disease, compared to the probability in patients not having the disease; while a negative likelihood ratio (LR−) is the probability of obtaining a negative test result in patients without the disease, compared to the probability in patients with the disease 18 . Clinical signs and symptoms and laboratory tests with an LR+ greater than 2.0 and an LR− less than 0.5 are clinically useful for diagnosis of pneumonia 19 . Diagnostic odds ratio is defined as the ratio of odds of the test being positive for a patient with the disease in relation to odds of the test being positive for a patient without the disease. Considering the correlation between sensitivity and specificity within and across studies, we performed bivariate model to calculate the pooled estimates of sensitivity, specificity, positive likelihood ratio, negative likelihood ratio and diagnostic odds ratio with 95% CIs. To avoid the large variances between studies, we conducted random effect meta-analysis 20 approach as the final model. The final bivariate model was computed using the mada package in R version 3 3 4 21 . Finally, we performed the pool estimates of meta-analyses for index tests with at least four or more studies. Pool estimates for less than four studies have limited validity and hence, was excluded 22 . The index tests assessed at different thresholds were pooled together and analysed. The Summary Receiver Operating Characteristics (SROC) curve for index tests (at least four included studies) were computed using the Reitsma SROC model to obtain the summary point estimates of sensitivity and specificity as well as 95% predicting region and 95% confidence region for the summary operating point 23 .

Results
The selection process (PRISMA flow-diagram) is showed in Fig. 1. A total of 2,428 records were identified from the initial database search and additional records from other sources. The articles were curated using EndNote and 39 duplicates were removed. Following this, 2,397 were included for initial screening and 2,355 articles were excluded based on relevance of titles and abstracts. 42 full text articles were retrieved, reviewed and selected based on relevance and quality for eligibility. A further 29 articles were excluded because of irrelevant design (i.e. irrelevant content, unmet inclusion criteria), wrong target disease (i.e. diseases other than pneumonia e.g. influenza), wrong populations (i.e. performed in age group less than 18 years) and insufficient data. This brings the total number of included articles for this review to 13 13,24-35 .
The methodological quality of included studies are summarised in Figs 2 and 3. In risk of bias, eleven studies had low risk in patient selection and one study had high risk for enrolling participants with confirmed diagnosis and control group without the condition. Seven studies had low risk in index test and six studies were unclear for pre-specifying of threshold of index test. Four studies had high risk in reference standard due to lack of blinding in interpretation of radiograph and extraction of data from medical records. Seven studies had low risk in flow and timing. Five studies had high risk bias due to attrition of some participants and selectively receiving the  www.nature.com/scientificreports www.nature.com/scientificreports/ reference standards. In applicability concerns, eleven studies had low risk concerns in patient selection, whereas two studies had high risk concerns. All studies addressed low risk concerns for index test and reference standard.
The summary characteristics of the 13 included studies are shown in Table 1. A total of 11,144 participants were obtained from the studies and they are from varying locations: Iran (n = 1), USA (n = 3), Denmark (n = 2), Netherlands (n = 2), Norway (n = 2), Sweden (n = 1), Switzerland (n = 1) and Europe (n = 1). The studies were done in outpatient clinics (n = 6), emergency clinics (n = 6), primary care centres (n = 2) and GP clinic (n = 1). All the participants were adults aged ≥18 years and the sample size varied from 95 to 4,464. The study designs were prospective cohort (n = 8), case-control (n = 1), cross sectional (n = 3) and retrospective chart review (n = 1). The studies included consecutive patients with history of respiratory tract infection. In these studies, the inclusion criterion was cough alone (n = 3) and clinically suspected pneumonia (n = 10). The proportion of radiographic pneumonia in the studied populations varied from 5% to 50%.

Discussion
Clinicians have traditionally used certain clinical signs and symptoms to diagnose pneumonia in the community. We aimed to assess the clinical predictors for diagnosis of pneumonia in adults to complement the clinical judgement for the need of CXR in a primary care setting, where CXR may not be readily available. The results of the pooled diagnostic odds ratio for clinical signs and laboratory tests were promising in our findings. However, the pooled diagnostic odds ratio for socio-demographic and symptoms were not ideal as predictors except for cough.
Our meta-analysis showed that individual clinical history and symptoms do not have adequate discriminatory power except cough to diagnose pneumonia among adults in primary care setting. This is consistent with previous study showing that no clinical symptoms is sufficient on its own for diagnosis of radiographic pneumonia among children under five years old 14 . Consistent with the previous analyses, cough was a poorly specific indicator of pneumonia, assuming that patients visiting to clinic with symptoms of cough would unlikely to have pneumonia 14,36 . However, there is likely an overestimation of cough because it was part of the inclusion criteria for most of the studies. Thus, likely resulting in cough having a good pooled negative likelihood ratio and high diagnostic odds ratio in our study, Respiratory rate (one of the criteria to classify pneumonia) was one of the two most useful predictors among the clinical signs, beside temperature ≥38 °C based on diagnostic odds ratio. Fast breathing had highest specificity, therefore it might be useful clinically to identify patients without fast breathing would be unlikely to have pneumonia. There was evidence that an adult with a respiratory rate of over 20 per minute is probably unwell and an adult with a respiratory rate of over 24 breaths per minute is likely to be critically ill 37 . Pyrexia was the next most useful predictor, and followed by tachycardia. These findings are similar to the clinical decision rule of a published study 9 , that ordered CXR only for patients with at least one abnormal vital signs (i.e. temperature greater than 37.8 °C, respiratory rate greater than 20 breaths per minute, or pulse rate greater than 100 beats per minute). Consistent with other studies 7,11 , auscultation sounds such as crackles was shown as predictor of pneumonia in our study. Moreover, the predictors in our findings were also found in Heckerling clinical decision rules for pulmonary infiltrates. The rule identified five key predictors for pneumonia: temperature greater than 37.8 °C, pulse rate greater than 100 beats per minute, crackles, decreased breath sounds, and absence of asthma 11 . In addition, fever, tachycardia and crackles were observed to be useful as part of the predictions models externally validated for pneumonia in primary care 38 .
In our results, biomarker such as PCT and CRP were the strongest predictors among all variables tested and had significant discriminating power than clinical signs and symptoms for pneumonia. PCT, a marker of sepsis, strongly correlated with bacteria load 39 and the severity of infection 40 . In addition, elevated PCT levels point towards bacterial infection rather than viral infection 41 . There is some evidence that PCT >0.25 ng/ml reflects a typical bacterial aetiology 42 . This evidence is in line with our result demonstrating that PCT > 0.25 ng/ml was able to predict pneumonia. On the other hand, PCT has been most frequently studied with regard to its prognostic value and correlation with disease severity 42,43, in patients with pneumonia. Notably, PCT has been regarded as a prognostic rather than diagnostic factor in adult patients with community-acquired pneumonia 42 . CRP is a widely used point of care test in ambulatory care. CRP has been studied as a screening device for inflammation,  www.nature.com/scientificreports www.nature.com/scientificreports/ and as a marker of bacterial infection 44,45 . Our result revealed that the diagnostic role of CRP > 20 mg/l has value in ruling in pneumonia. This finding is similar to previous systematic reviews showing that pneumonia is ruled out if CRP below 20 mg/l 44,46 . Moreover, it also seems that CRP test cannot be used as a stand-alone diagnostic test for pneumonia. Current evidences have shown that adding CRP value to basic signs and symptoms models in diagnosing pneumonia improved diagnostic discrimination of adult patients in primary care 10,47 .
There are a number of limitations in this review. Firstly, there was heterogeneity among the selected studies in terms of inclusion criteria, chest radiograph (interpretation criteria and lack of blinding), inconsistencies in the reporting of clinical features across different studies and the prevalence of pneumonia. Sensitivity and specificity values are highly dependent on the prevalence of the pnemonia in the respective population of different studies. Moreover, the time between potential exposure to infection and the point when the test gives an accurate result was not clearly reported in the studies. Bivariate random effects model was used to account for heterogeneity between the studies. Secondly, a small number of variables did not allow for meta-analysis to be conducted to investigate the tests' accuracies. Thirdly, it is possible that there are some relevant studies which were not published, resulting in potential publication bias. In addition, only studies published in English were included in our review which may have resulted in limited generalizability. Finally, our findings may also have limited applicability in low-or middle-income countries, since all the selected studies except one study 24 were conducted in high income countries. Moreover, the study only focused on the predictive nature of the variables singly and potentially, performance of the variables may be improved but likely to a limited extent with more than one clinical signs and symptoms as covariables in the model.
The findings of this review suggest that individual clinical symptom (cough) and clinical signs (pyrexia, tachycardia, tachypnea, and crackles) are associated to pneumonia but limited as a single predictor for diagnosis of radiographic pneumonia. The combination of these clinical features in decision rule might indeed enhance the overall diagnostic performance of individual symptoms and signs. Future high quality and large-scale case-control studies using the clinical data relevant to the population of interest is necessary to assess the combination with the clinical features identified in this review, and to propose a practical scoring system to aid clinical judgement for ordering of CXR to confirm pneumonia. Moreover, the combination of these clinical features together with molecular biomarkers is likely to further add value to the overall diagnostic accuracy.