Community acquired pneumonia (CAP) is a leading cause of morbidity, with an annual incidence estimated at 5–11 per thousand adult population (higher in the young and the elderly). The diagnosis of CAP in hospital is made with a chest radiograph. In the community, the diagnosis relies upon a combination of symptoms suggestive of an acute lower respiratory tract infection (LRTI), presence of signs on chest examination, and systemic features (e.g. temperature >38.0 deg C). Diagnosing CAP in the community setting is therefore more problematic.1 In all settings the primary aim when treating CAP is to prevent death. Whilst mortality due to CAP in adults managed in the community in the UK is low (<1%), this is higher in patients admitted to hospital (5.7–14%) and higher still at 5-year follow-up (35.8–39.1%).

In order to identify patients at both high and low risk of mortality, CAP severity scores have been developed to complement clinical judgement. The CURB-65 score,2 endorsed by the British Thoracic Society (BTS), consists of five elements – Confusion; Urea [>7mmol/l]; Respiratory Rate [≥30 breaths/min]; Blood pressure [systolic<90mmHg and/or diastolic ≤60mmHg]; age ≥65 years — and has been shown in prospective studies to predict 30-day mortality in hospitalised patients with CAP.3–5 Recent systematic reviews and meta-analyses showed that CRB-65 (CURB-65 minus serum urea) is a useful indicator of severity in hospitalised patients,6 and in this group, equivalent to CURB-65 in predicting 30-day mortality.7

Currently, clinical judgement in combination with CRB-65 is advocated in assessing the severity of CAP in community settings in the UK.8 Studies in support of this recommendation have included few patients in the community,9,10 and there is little data validating severity assessment tools in LRTI (without localising chest signs) in patients in primary care. The question therefore remains as to whether CRB-65 alone in this context accurately predicts patient outcomes.

The study by Francis et al.11 in this issue of the PCRJ set out to determine a) whether primary care clinicians routinely recorded elements of the CRB-65 score and b) whether the score accurately predicted prognosis in adults with LRTI in the community. Primary care clinicians in 13 European countries, including 14 primary care networks, completed a case record form (CRF) on initial patient assessment. Patients completed a daily diary from presentation until symptom resolution (or up to 28 days, whichever came first). Patients ≥18 years old were included with an illness of ≤28 days duration with cough (acute or worsened) or features suggestive of a LRTI. Each element of the CRB-65 was scored to assess compliance of clinicians with the score, and a Cox proportional hazards model was used to determine the association between an elevated CRB-65 score (≥1) and prolonged illness.

Of the 3,368 subjects included (n=524 [15.6%] ≥65 years old), 2,690 (79.9%) had diary data available with 2,468 (73.3%) having data at recovery. The authors were able to calculate CRB-65 scores for only 339 (12.6%) study subjects, but by imputing zero for missing components of the CRB-65 score, created an additional dataset of 2,690 subjects. A clinical diagnosis of pneumonia was made in 111 subjects; however, a complete CRB-65 score was only available for 12 in this group. No subject had a CRB-65 score higher than 2 (n=9 [2.7%]) with the majority scoring 0 (n=235 [69.3%]). Interestingly, the authors report wide variation in scores between primary care networks. Regression analysis of the complete dataset found that a CRB-65 score ≥1 was associated neither with a prolonged moderately severe illness (OR 0.42 [95% CI 0.04–4.19]), nor with hospitalisation (OR 3.12 [95% CI 0.16–60.24]).

This study has major strengths. Firstly a large, multinational primary care population was studied. Secondly, each primary care network contributed large numbers of subjects (>100 in all but one centre) suggesting that this population is reflective of primary care across Europe. Thirdly, standardised methodology and data collection were used. There is, however, a large amount of missing data, confirming that clinicians do not regularly record each element required to calculate a CRB-65 score. The authors are probably correct in assuming that clinicians are less likely to measure and record normal variables; the attribution of a zero score for missing variables may be correct, but it does add another unnecessary variable to the study. The authors may therefore have missed a real effect, resulting in a type 2 error. Additionally, the subjects studied were young (n=524 [15.6%] ≥ 65 years old) and therefore the findings of this study may not be applicable to an older population. Moreover, the endpoints were not the same as in previous CRB-65 studies. The validity of the scores in patients admitted to hospital is based on their ability to predict mortality. The causes of death may not be the same as the causes of prolonged illness nor of admission to hospital, so the premise that CRB-65 might work using a different endpoint may be flawed. Variation in scores between networks also questions the similarity of patient recruitment between networks.

Most studies using CRB-65 as a predictor of outcome are in either secondary care or in both primary and secondary care. Additionally, most studies include chest radiograph infiltrates as one of the subject entry criteria. One study, carried out in primary care in the Netherlands,12 used a logistic regression model to examine predictors of a combined endpoint of hospitalisation and death. The authors studied 3,166 episodes of LRTI (defined as pneumonia [with chest radiograph infiltrates], acute bronchitis and exacerbations of COPD) in an older population (mean age 75.5 years). They found that a combination of age ≥80 years, presence of congestive cardiac failure or diabetes, use of oral glucocorticoids, hospitalisations in the previous year and antibiotic use the previous month, were the best predictors. The same group then studied the CRB-65 score in CAP in an older population (mean age 77.3 years) in the community. They demonstrated in this study that the CRB-65 score was equally as good as CURB-65 at predicting 30-day mortality in this subject group.10

More research is needed to confirm whether the CRB-65 score genuinely has value in managing CAP in the community as other studies have suggested.9,10 The main value for a general practitioner (GP) might be in predicting who should be referred to hospital. This does not mean that the CRB-65 should necessarily be used in all adults where CAP is suspected, but it might be especially useful in those where there is uncertainty about whether hospital referral is in the patient's best interests.

The study by Francis et al.11 importantly implies that CRB-65 may not be useful in predicting prolonged illness in adults with LRTI in primary care. The question of how to predict patient outcomes in LRTI in the community therefore remains unanswered. We hope that the authors of this study go on to use their data to help address this important unmet need.