In this population-based cohort study, presence of a pulmonary infiltrate was associated with a doubled 30-day mortality from pneumococcal LRTI, whereas bacteraemia was associated with a four-fold increased mortality compared with patients with none of these clinical manifestations. Differences in age, comorbidity and lifestyle factors seemed to explain only a minor part of the associations. LOS, risk of ICU admission, and risk of pulmonary complications also rose gradually from patients with no infiltrate and no bacteraemia to those with an infiltrate without bacteraemia and to those with bacteraemia. Non-bacteraemia patients had more acute LRTI symptoms compared to bacteraemia patients, whereas other markers of disease severity (including CRP, leukocyte count, fever, and CURB65 score) gradually increased from patients with no infiltrate and no bacteraemia to those with an infiltrate without bacteraemia to those with detected bacteraemia.
Biological mechanisms underlying the poorer outcome among patients with X-ray changes and bacteraemia may include hypoxaemia due to thickening of the blood-gas barrier and loss of functional residual capacity in the lungs, and increased risk of severe sepsis with decreased perfusion of vital organs. Inflammatory exudate filling alveoli during pneumonia causes a volume loss that impact on lung capacity roughly proportional to the extent of the pulmonary infiltrate . A study of 48 pneumonia patients from Japan found a significant correlation between the size of X-ray pneumonic infiltrates and Pa02 both in young and elderly patients . Hagaman et al.  suggested the magnitude of the inflammatory response in the lung tissue to be the most important determinant whether a patient presents with a pulmonary infiltrate or not. Both Hagaman et al. and Basi et al. [19, 20] found lower leukocyte counts for patients with negative chest X-rays, corroborating our results. Our findings that additional inclusion of CRP, leukocyte count, MAP and SAT in the regression model reduced the MRRs for both infiltrate without bacteraemia and bacteraemia patients suggest that some of the increased mortality associated with these manifestations is mediated through increased inflammation, hypoxaemia and sepsis.
In the American PORT study based on 2,287 in- and out- patients with CAP of whom 158 had confirmed pneumococcal pneumonia, Brandenburg et al.  also noted a tendency towards greater clinical severity for bacteraemic compared with non-bacteraemic patients, as reflected by increased median LOS (7.5 days vs. 6.5 days), pneumonia-related 30-day-mortality (7.7% vs. 2.7%), and ICU admission (15.4% vs. 12.2%) (p-values between 0.4-0.8). In line with this, Musher et al.  found bacteraemia to be indicative of a higher 30-day mortality (21 vs. 13%; OR 1.88, p-value 0.25) and higher ICU admission risk (44 vs. 25%; OR 2.38, p-value 0.05) among 100 American veteran patients with pneumococcal pneumonia. Likewise, a Spanish study comprising 57 bacteraemic and 25 non-bacteraemic hospitalized pneumococcal CAP patients found a higher mortality and LOS among bacteraemic patients . In contrast, a South African ICU-study of 63 severe pneumococcal CAP patients found lower in-hospital mortality in bacteraemic compared to non-bacteraemic patients (15% vs. 28%) . In comparison, a sub-analysis of our ICU patients showed a 30-day mortality of 29, 33 and 42% for patients with no infiltrate and no bacteraemia, infiltrate without bacteraemia, and bacteraemia, respectively.
To our knowledge no previous studies have focused on outcomes of pneumococcal LRTI without bacteraemia or pulmonary infiltrate, but studies on unspecified “clinical pneumonia” or “infiltrate-negative all-cause CAP” have been conducted. In a population-based Canadian study of 2706 hospitalised patients Basi et al.  reported that approximately one-third of patients suspected of CAP on admission did not have an infiltrate on their admission chest X-ray. The in-hospital mortality was 10% in patients with X-ray confirmed pneumonia compared with 8% in patients without an infiltrate (p=0.09). In a smaller American study of CAP, Hagaman et al.  found that patients with presence of an infiltrate stayed one day longer and had slightly higher Pneumonia Severity Index scores and mortality than those with no infiltrate. Similar to our study, Hagaman et al. and Basi et al. did not find any clear association between the amount of pulmonary symptoms and chest X-ray findings.
Our study’s strengths include its relatively large size covering one-third of the Danish population for one year, its population-based design including all hospitals in a geographical region, complete follow-up for mortality, and detailed information on blood chemistry, lifestyle factors, and microbiology data for all patients, thus allowing for extensive adjustment for important confounders and investigation of mediators. Use of routinely recorded health care data, collected without knowledge of our research aim, reduced the risk of information bias.
Our study was limited by lack of information on the indication for collecting the blood or LRT specimen. The no infiltrate and no bacteraemia patient group may thus have included people with chronic LRTI or other pulmonary diseases who had pneumococci cultured e.g. as part of a diagnostic process in the absence of any acute infection, resulting in an overestimation of the relative risk of adverse outcomes in the other groups. However, all patients in our no infiltrate and no bacteraemia group had to have at least one new or increased pulmonary symptom together with CRP >50mg/L and/or leukocytes >8.8*109/L, and our sensitivity analysis using even stricter CRP and leukocyte cut points showed robust results.
The validity of our estimates depends on S. pneumoniae being the causative agent, as well as on accurate division into manifestation groups based of chest X-ray findings and blood cultures. For the purpose of this study we included patients with chest X-rays describing a “likely” or “probable” infiltrate in the infiltrate without bacteraemia or bacteraemia groups. Assuming that patients with a likely infiltrate are less ill than patients with a definite infiltrate, any misclassification is unlikely to create a false difference between the infiltrate and no infiltrate patient groups, and would not change our conclusions. Results of sputum culture may be falsely positive due to upper airway colonization with pneumococci [36–38]. Nevertheless, colonization is more common in young children and patients with chronic pulmonary disease, and excluding the latter group did not alter our risk estimates notably.
Because the study was done in a routine clinical setting, microbiology tests and chest X-rays were only obtained at the managing physician’s discretion. Thus, not all patients had all tests done. Some patients in the two non-bacteraemic groups might have had bacteraemia, and some patients in the no-infiltrate group might have had an infiltrate without our knowledge. However, this would likely lead to bias towards the null, and our sub-analysis excluding patients with no X-ray or blood culture taken did not change our findings markedly.