To estimate the causative pathogens of CAP, we constructed a decision tree that focused on the characteristics of the top ten known causative pathogens, focusing mainly on the distribution of lesions, centrilobular nodules, and bronchial wall thickening (findings that have previously been reported on thin-section CT) [6,7,8,9,10,11,12,13,14,15,16].

Pneumonia can be dichotomized into lobar pneumonia and bronchopneumonia [4, 5]; the former has a non-segmental distribution caused by the rapid spread of large amounts of exudate through inter-alveolar passages such as the pores of Kohn, while in bronchopneumonia, lesions form mainly from inflammatory cell infiltrates with little exudate, resulting in a segmental distribution. The causative pathogens of lobar pneumonia are almost exclusively S. pneumoniae, C. pneumoniae, K. pneumoniae, and L. pneumophila [4, 5, 7,8,9], whereas other organisms are known to cause bronchopneumonia. The frequency of bronchial wall thickening ranges between 16.7 and 35% in lobar pneumonia caused by the above four pathogens, whereas this frequency ranges from 75.9 to 88.1% in bronchopneumonia caused by pathogens such as H. influenzae, M. pneumoniae, and M. catarrhalis [6,7,8, 10,11,12, 27]. The frequencies of centrilobular nodules are 4.0–19.8% in lobar pneumonia and 56.3–90.5% in bronchopneumonia [6,7,8, 10,11,12]. The CT findings of centrilobular nodules and bronchial wall thickening are significantly more frequent with the causative organisms of bronchopneumonia than with those causing lobar pneumonia (Table 2). The initial lesions of lobar pneumonia may present with a segmental distribution, as seen in bronchopneumonia. However, even in the case of segmental lesions, the presence of centrilobular nodules and bronchial wall thickening is less frequent in lobar pneumonia than in bronchopneumonia, which can help distinguish the causative organisms. Segmental lesions with fewer centrilobular nodules and bronchial wall thickening should raise suspicion for initial lesions of lobar pneumonia.

The following findings were also added to the decision tree: the CT finding of an acinar pattern is suggestive of C. pneumoniae pneumonia [7, 16], and S. milleri is the most common pathogen causing pulmonary abscess and empyema [13, 28, 29]. Clinical findings that may be useful for estimating the causative pathogens were also reflected in the decision tree. According to previous reports, patients with M. pneumoniae pneumonia are generally younger than patients with pneumonia caused by other pathogens [6,7,8,9,10,11,12,13,14,15]. Based on previous reports and the Japanese Respiratory Society guidelines, the statistical frequency of the causative organisms of CAP and the age of onset were also reflected in the decision tree [6,7,8,9,10,11,12,13,14,15, 17,18,19,20,21,22]. For each group of endpoints in the decision tree, the candidate pathogens were ordered by statistical frequency. The following facts were also added to the decision tree as additional findings; S. pneumoniae, H. influenzae, and M. catarrhalis are frequently detected in CAP in emphysema patients [12, 23, 24], and S. pneumoniae and S. aureus account for the majority of infections following influenza virus infection [25, 26].

The decision tree reflects that S. pneumoniae is the most common pathogen in concurrent pneumonia in CAP, with concurrent pneumonia accounting for 30–66.4% of all S. pneumoniae pneumonia cases [6, 30,31,32]. Additionally, in a study of 962 K. pneumoniae pneumonia cases, 79.4% were reported to be mixed infections with other pathogens such as S. aureus [8]. Most cases of pneumonia caused by mixed infections are mixes of pathogens causing lobar pneumonia and those causing bronchopneumonia; mixed infections between multiple pathogens causing lobar pneumonia or bronchopneumonia are rare [6, 8]. Okada et al. compared the pulmonary thin-section CT findings of patients with S. pneumoniae pneumonia with and without concurrent infection [6]. Centrilobular nodules and bronchial wall thickening were significantly more frequent in patients with pneumonia caused by concurrent infection than in those infected with S. pneumoniae alone. Additionally, in a previous report comparing thin-section CT findings between patients with K. pneumoniae pneumonia alone and those with concurrent infection, findings of centrilobular nodules, bronchial wall thickening, and cavity were significantly more frequent in patients with concurrent pneumonia [15]. In contrast, mixed infections are less common in C. pneumoniae pneumonia [7]. These findings on concurrent infections were reflected in the decision tree.

The usefulness of the decision tree for estimating the organisms responsible for CAP was examined using two student doctors, six residents, and eight radiologists. The decision tree increased the percentage of correct answers for all examiners, especially the radiologists; estimation of the first three most likely candidates using the decision tree resulted in more than 80% correct answers for everyone. When the usefulness of the decision tree was examined according to the causative organisms, it was most useful for M. pneumoniae, followed by H. influenzae and C. pneumoniae.

Europe and the United States recommend de-escalation therapy for the treatment of CAP, whereas Japan recommends escalation therapy. In Japan, S. pneumoniae, H. influenzae, M. pneumoniae, and C. pneumoniae are the main organisms that cause CAP. Because most S. pneumoniae in Japan are macrolide-resistant, unlike in Europe and the United States, macrolides are not used as the first-line treatment for S. pneumoniae; instead, penicillins combined with β-lactamase inhibitors are recommended. An exception is M. pneumoniae and C. pneumoniae, which are treated as atypical pneumonia because penicillins in combination with β-lactamase inhibitors are ineffective; for these organisms, macrolides are the first-line therapy. Using the decision tree, atypical pneumonia cases caused by M. pneumoniae and C. pneumoniae could be predicted with high accuracy (residents, student doctors, and radiologists were all p < 0.001). The decision tree developed here may allow for early suspicion of atypical pneumonia and its early treatment.

There were several limitations to our study. This was a retrospective study, and only the top ten CAP causative bacterial pathogens were included in the decision tree. Additionally, cases of viral pneumonia were excluded from the present study; however, the CT findings of viral pneumonia are described in many reports and are quite different from the findings of bacterial pneumonia in otherwise healthy individuals, meaning that radiologists should not have any difficulty in differentiating between them [33]. The number of cases studied was also small. Additionally, the decision tree was unable to provide any insight into the prediction of L. pneumophila in either group. When the microorganisms responsible for lobar pneumonia were ordered from the highest to the lowest statistical frequency of the causative pathogens, L. pneumophila was the least frequent, ranking fourth (Fig. 1). It can be reasonably assumed that readers would be unable to select L. pneumophila as a causative organism if estimating causative organisms up to the third candidate; consequently, they would be unable to provide the correct answer. However, it has recently become possible to detect antigens for all types (type 1–15) of L. pneumophila in urine, and high diagnostic accuracy for this organism could have been obtained if the results of the urinary antigens were reflected in the present decision tree. In Japan, CT scans are usually performed before the results of urine antigen tests (S. pneumoniae and L. pneumophila) or blood tests, or at least before the results of these tests are available. To reflect this clinical reality in this study, the reading experiments were carried out without the results of urinary antigen tests or clinical findings being known.

In conclusion, we developed a decision tree to estimate the causative pathogens for CAP based on thin-section CT findings, patient characteristics, the statistical frequency of causative pathogens, the age of onset, and the Japanese Respiratory Society guidelines. The results of validation show that the decision tree is useful for estimating the causative pathogens in CAP. The addition of further patient information to the decision tree, such as urinary antigen results and clinical manifestations, may allow for an even higher accuracy rate.