In patients with community acquired pneumonia (CAP), clinical and radiological features at the time of presentation do not predict the microbiological aetiology with any certainty.¹ Initial treatment is therefore usually empirical and directed by the severity of the illness at the time of presentation. A large number of studies have been conducted over the last 10 years to determine prognostic factors in CAP. In turn, clinical prediction rules based on a number of key prognostic factors have been developed, such as the pneumonia severity index (PSI) and the CURB-65 score, and incorporated into CAP management guidelines.^2–6 Most of these CAP severity studies, and the resulting prediction rules, use mortality as the main outcome measure. However, mortality is not the only clinically important outcome. In this issue of Thorax, Menéndez and colleagues report on a large observational study of the risk factors related to failure of initial empirical treatment for CAP.⁷

The definition of treatment failure adopted was complex and based on (a) the time from admission (less than or more than 72 hours corresponding to “early” and “late” treatment failure) and (b) the occurrence of clinical features such as “haemodynamic instability”, “the appearance or impairment of respiratory failure”, and “radiographic progression”. While pragmatic, these features were not rigorously defined, thus making it difficult to compare the results of this study with other research. Also, the ordering of repeat chest radiographs, a key element in the definition, was left to the discretion of the attending physician, so introducing a potential bias into the detection of treatment failure. Accepting these limitations, 15% of a cohort of 1424 hospitalised patients experienced treatment failure. These patients had a longer mean length of hospital stay (18.5 days v 9.4 days) and increased mortality (25% v 2%). Most of the treatment failures occurred in the first 72 hours.

Initial treatment with fluoroquinolones was found to be associated with a lower risk of treatment failure but not with in-hospital mortality. The researchers offer a good discussion on the possible explanations for this finding. The high prevalence of penicillin resistant Streptococcus pneumoniae (∼30% of isolates) in the study country (Spain)^8,9 compared with the prevalence of fluoroquinolone resistant strains (<1%) may indeed be relevant.¹⁰ The excellent coverage of atypical pathogens by the fluoroquinolones may also be important. Patients in the study were treated according to prevailing Spanish guidelines. This means that hospitalised non-ICU patients could be treated with either a third generation cephalosporin or co-amoxiclav with or without a macrolide, or monotherapy with a fluoroquinolone. Certainly those patients treated with only a third generation cephalosporin or co-amoxiclav (29% of cases) would not have had coverage for infection by an atypical pathogen compared with those treated with a fluoroquinolone

Retrospective studies have suggested that treatment with β-lactam drugs in combination with a macrolide or a quinolone is associated with lower mortality in CAP compared with other antibiotic regimens.¹¹ Unfortunately, the difficulty with these observational studies is the inability to correct adequately for confounding factors that might have influenced the initial choice of antibiotic. Conversely, the emergence of fluoroquinolone resistant pathogens in areas with a high consumption of fluoroquinolones is a very real problem and cautions against their overenthusiastic use.¹² More work is needed to clarify the advantages of the fluoroquinolones in comparison with other antibiotic regimens in the empirical treatment of CAP.

The following risk factors were found by Menéndez and colleagues to be independently associated with treatment failure: the PSI prediction rule for risk of mortality (the PSI categorises patients into risk classes I–V corresponding to ascending risk of mortality), leucopenia (<4000 cell/mm³), liver failure, and the presence of adverse chest radiographic features on admission (specifically, the presence of pleural effusions (OR 2.6), multilobar involvement (OR 2.2), or cavitation (OR 5.2)). Each of these risk factors (except lung cavitation) has previously been reported to be independently associated with mortality in CAP.^13,14 Indeed, the presence of liver disease and a pleural effusion are two of the 20 variables included in the PSI prediction rule. This is not altogether surprising since treatment failure was itself associated with mortality, and no clinical prediction rule can be expected to fully account for all the recognised features of disease severity.

Impact on clinical management

How then might these findings enhance our current management of CAP, if at all? Their main contribution is going to be in the management of patients at low risk of mortality. With regard to the decision to admit to hospital, current recommendations are based on an assessment of mortality risk, social circumstances, and the stability of co-morbid illnesses. In recognition of the limitations of assessing disease severity solely according to risk of mortality, all guidelines underline the importance of clinical judgement.^4,15 The study by Menéndez and colleagues is helpful in highlighting the additional risk factors associated with failure of empirical treatment (such as adverse chest radiographic features) that should be taken into account in patients identified as being at low risk of mortality according to the PSI. However, how these patients should best be managed is not known. Patients at risk of treatment failure may still be suitable for ambulatory care provided adequate early outpatient follow up is arranged. Alternatively, they may require hospital admission.¹⁶

Awareness of the expected time course of clinical resolution allows timely detection of treatment failure. Halm and colleagues have shown that the median time to clinical stability is 2 days for heart rate (⩽100/min) and systolic blood pressure (⩾90 mm Hg) and 3 days for respiratory rate (⩽24/min), oxygen saturation (⩾90%), and temperature (⩽37.2°C).¹⁷ Measurement of the C-reactive protein (CRP) level is also helpful because a CRP level that does not fall by 50% within 4 days of admission is suggestive of treatment failure.^18,19

What to do once treatment failure is recognised is not well studied. Repeat and supplementary microbiological investigations are generally recommended in order to detect new, resistant, or nosocomial infections. Bronchoscopy yields a diagnosis in up to 41% of patients.²⁰ One study found it to be beneficial mainly in non-smoking patients aged less than 55 years with multilobar infiltrates.²¹

Where do we go from here? Further work using robust and reproducible definitions for treatment failure is required to confirm the findings of Menéndez and colleagues. The use of a different prediction rule to adjust for risk of mortality—for example, CURB-65 instead of PSI—may result in the identification of different risk factors for treatment failure. Most importantly, the optimal management of patients at risk of treatment failure and how it might differ from usual management needs to be determined, ideally through intervention studies with clinically relevant end points.