Validation of a Prediction Rule for Legionella Pneumonia in Emergency Department Patients.

We validated a clinical prediction rule for Legionella based on clinical parameters (dry cough, fever) and laboratory findings (C-reactive protein, lactate dehydrogenase, sodium, platelet counts) in 713 consecutive patients with community-acquired pneumonia. The Legionella Score performed well in estimating the likelihood for Legionella infection and thus may help to direct diagnostic and therapeutic decisions.

Legionella spp. cause community-acquired pneumonia (CAP) with a high burden of morbidity and mortality, particularly if adequate antibiotic therapy is delayed [1, 2]. Differentiating Legionella CAP from other types of CAP has important implications regarding empirical antibiotic therapy. Legionella spp. are intracellular bacteria that penetrate and proliferative within the phagosomes of alveolar macrophages and blood monocytes [3]. Hence, antimicrobials that cannot penetrate the host’s cellular membrane, such as beta-lactams, are ineffective [4]. Macrolides, quinolones, or tetracyclines are required for effective treatment of Legionella CAP [5]. Rapid identification of the causative microorganism in patients presenting with CAP to the emergency department is thus of high relevance.

Current testing options for Legionella infection have limited sensitivity and substantial time delays [6]. As a result, use of broad-spectrum antibiotics or combination treatment including coverage of typical and atypical pathogens is often advised, leading to antibiotic overuse, which in turn results in an increase in antimicrobial resistance [7, 8]. Accurately differentiating Legionella CAP from other types of pneumonia on hospital admission may help to reduce the unnecessary use of broad-spectrum antibiotics.

Recently, a clinical Legionella Score based on 6 clinical and laboratory parameters, namely dry cough, high fever, high C-reactive protein (CRP) and lactate dehydrogenase, and low sodium levels in the blood, as well as thrombocytopenia, has been proposed to predict Legionella in CAP patients [9]. In the initial study, this score showed a high diagnostic accuracy (area under the curve, 0.86) [9], which was confirmed in 2 validation samples in the United States and Japan [10, 11]. These studies, however, were limited by their retrospective design and did not investigate whether the score could be improved with addition of further clinical parameters.

Herein, we aimed to independently validate the performance of the Legionella Score in a sample of consecutive patients with CAP due to Legionella spp., Mycoplasma spp., Pneumococci, and viruses over a 4-year time period and to investigate whether additional clinical and laboratory parameters would further improve its accuracy.

METHODS

This is a secondary, retrospective analysis of patients who participated in the TRIAGE project, a prospective observational study that investigated the utility of different biomarkers in adult patients presenting to the emergency department with different medical diseases [12]. From this sample of patients, we evaluated all consecutive patients hospitalized for CAP with defined etiologies (ie, Legionella spp., Mycoplasma spp., Pneumococci, viruses) in a Swiss study center (Kantonsspital Aarau) between October 2013 and December 2017. More details about the methodology are presented in the Appendix.

For all patients, the Legionella Score was calculated by adding 1 point for each parameter meeting the predefined cutoffs within the first 24 hours of hospitalization (dry cough, fever >39.4°C, CRP ≥187 mg/L, lactate dehydrogenase ≥225 mmol/L, plasma sodium level <133 mmol/L, and platelet count <171 G/L), as previously suggested (Table 1) [9]. The overall aims of this analysis were to validate the Legionella Score, as proposed by Fiumefreddo et al. [9], and to identify new parameters that allow differentiation of Legionella from other CAP etiologies. First, we assessed the performance of each of the 6 clinical and laboratory parameters in univariate analysis. Second, we evaluated the performance of the original clinical score [9]. We used logistic regression analysis to study the association of the different parameters and Legionella CAP. Discrimination was assessed by calculation of receiver operating characteristic curves. The results of receiver operating characteristic curve statistics are reported as area under the curve (AUC). Furthermore, we investigated additional clinical and laboratory parameters for adequate predictive value for diagnosis using a forward step-up selection procedure.

Table 1.

Components, Optimal Cutoffs, and Predictive Performance of the Different Predictors of the Originally Proposed Legionella Score

Components	Optimal Cutoff	Sensitivity (95% CI)	Specificity (95% CI)
Dry cough		48.5 (30.8–66.5)	78.8 (75.6–81.8)
Temperature	>39.4°C	36.4 (20.4–54.9)	87.9 (85.3–90.3)
C-reactive protein	≥187 mg/L	69.7 (51.3–84.4)	79.4 (76.2–82.4)
Lactate dehydrogenase	≥225 mmol/L	54.5 (36.4–71.9)	66.0 (62.3–69.6)
Sodium	<133 mmol/L	45.5 (28.1–63.6)	84.3 (81.3–86.9)
Platelet counts	<171 G/L	39.4 (22.9–57.9)	66.8 (63.1–70.3)

RESULTS

A total of 713 CAP patients were eligible and included in this analysis. Legionella spp. were the causative pathogen in 33 (5%) of the cases, Mycoplasma spp. in 56 (8%), Pneumococci in 164 (23%), and viral etiology including influenza in 460 (64%).

First, we investigated the association of different clinical and laboratory parameters with Legionella etiology. Although several of the parameters showed a significant difference according to type of CAP, the parameters with the highest discriminatory ability were inflammatory and infection biomarkers (procalcitonin, CRP, neutrophils, neutrophil-lymphocyte ratio), body temperature, and albumin (all AUCs > 0.70). The results from univariate analysis and AUCs are presented in Table 2.

Table 2.

Association of Demographic, Clinical, and Laboratory Parameters and Legionella Cause

Characteristics	Legionella CAP	Non-Legionella CAP	P-Value	OR (95% CI)	AUC (95% CI)
No.	33	680
Patient characteristics
Age, median (IQR), y	64 (57–70)	70 (58–80)	.243	0.99 (0.97–1.01)	0.63 (0.54–0.71)
Male gender, No. (%)	22 (66.7)	367 (54.0)	.157	1.71 (0.81–3.57)	0.56 (0.48–0.65)
Clinical findings and symptoms
Confusion, No. (%)	7 (21.2)	81 (11.9)	.119	1.99 (0.84–4.73)	0.55 (0.47–0.62)
Body temperature, median (IQR), °C	39.3 (38.3–39.8)	38.2 (37.3–39.0)	<.001	2.25 (1.55–3.27)	0.73 (0.64–0.81)
Fever ≥38.3°C, No. (%)	26 (78.8)	319 (48.9)	.002	3.88 (1.66–9.06)	0.65 (0.58–0.72)
Oxygen saturation, median (IQR), 02	93 (91–95)	93 (89–96)	.890	1.00 (0.93–1.06)	0.54 (0.45–0.62)
Heart rate, median (IQR), beats/min	93 (88–101)	94 (78–108)	.697	1.00 (0.99–1.02)	0.53 (0.44–0.62)
Respiratory rate, median (IQR), breaths/min	21 (18–23)	21 (17–26)	.731	0.99 (0.92–1.06)	0.55 (0.39–0.72)
Systolic blood pressure, median (IQR), mmHg	139 (127–156)	138 (122–154)	.743	1.00 (0.99–1.02)	0.52 (0.41–0.63)
Diastolic blood pressure, median (IQR), mmHg	82 (66–92)	78 (67–89)	.072	1.02 (1.00–1.04)	0.55 (0.43–0.68)
Clinical history
Cough, No. (%)	23 (69.7)	538 (79.1)	.201	0.61 (0.28–1.30)	0.55 (0.47–0.63)
Dry cough, No. (%)	16 (48.5)	144 (21.2)	.001	3.50 (1.73–7.10)	0.64 (0.55–0.72)
Dyspnea, No. (%)	13 (39.4)	347 (51.0)	.195	0.62 (0.31–1.27)	0.56 (0.47–0.64)
Myalgia, No. (%)	7 (21.2)	104 (15.3)	.363	1.49 (0.63–3.52)	0.53 (0.46–0.60)
Nausea, vomiting, No. (%)	13 (39.4)	96 (14.1)	<.001	3.95 (1.90–8.21)	0.63 (0.54–0.71)
Diarrhea, No. (%)	12 (36.4)	81 (11.9)	<.001	4.23 (2.00–8.91)	0.62 (0.54–0.71)
Headache, No. (%)	7 (21.2)	102 (15.0)	.336	1.53 (0.65–3.61)	0.53 (0.46–0.60)
Breath-dependent thoracic pain, No. (%)	2 (6.1)	78 (11.5)	.346	0.50 (0.12–2.12)	0.53 (0.48–0.57)
Duration of symptoms, median (IQR), d	16 (9–21)	16 (7–20)	.781	1.01 (0.96–1.06)	0.52 (0.41–0.62)
Laboratory values, median (IQR)
CRP, mg/Lb	269 (180–340)	82 (33–160)	<.001	2.29 (1.78–2.95)	0.86 (0.82–0.90)
PCT, µg/La	1.73 (0.78–5.19)	0.26 (0.15–0.98)	.005	1.17 (1.05–1.31)	0.78 (0.73–0.83)
WBC, ×10^9/L	12.92 (9.25–15.64)	8.81 (6.17–12.87)	.137	1.02 (0.99–1.04)	0.67 (0.58–0.76)
Neutrophils, abs.	11.44 (8.94–14.33)	6.65 (4.13–10.18)	.003	1.06 (1.02–1.11)	0.74 (0.65–0.84)
Lymphocytes, abs.	0.76 (0.51–1.01)	0.91 (0.56–1.38)	.162	0.62 (0.32–1.21)	0.58 (0.48–0.68)
NLR	14.30 (9.41–19.40)	7.50 (4.22–13.03)	.003	1.02 (1.01–1.04)	0.72 (0.63–0.81)
Platelet count, ×10^9/La	200 (114–230)	198 (155–256)	.224	0.97 (0.93–1.02)	0.56 (0.45–0.66)
Hemoglobin, g/L	139 (125–145)	131 (116–143)	.205	1.02 (0.99–1.04)	0.60 (0.48–0.72)
Plasma sodium, mmol/L	133 (130–137)	137 (134–139)	.074	0.97 (0.94–1.00)	0.67 (0.56–0.78)
Plasma potassium, mmol/L	3.6 (3.4–3.9)	3.8 (3.5–4.2)	.076	0.40 (0.15–1.10)	0.64 (0.53–0.76)
Creatinine, µmol/L a	104 (88–140)	98 (78–133)	.960	1.00 (0.93–1.07)	0.57 (0.45–0.68)
Urea, mmol/L	7.1 (5.5–11.0)	6.9 (5.0–10.1)	.606	0.98 (0.89–1.07)	0.50 (0.37–0.63)
Lactate dehydrogenase, mmol/L b	290 (228–418)	229 (199–279)	.048	1.12 (1.00–1.26)	0.69 (0.57–0.81)
Glucose, mmol/L	7.2 (6.3–7.8)	6.8 (5.8–8.5)	.990	1.00 (0.86–1.17)	0.57 (0.47–0.67)
ALT, U/La	51 (24–91)	26 (19–38)	.212	1.02 (0.99–1.06)	0.68 (0.53–0.83)
AST, U/La	45 (25–97)	26 (18–38)	.071	1.03 (1.00–1.05)	0.69 (0.52–0.85)
Calcium, mmol/L	2.36 (2.30–2.43)	2.33 (2.25–2.42)	.215	7.61 (0.31–188.70)	0.58 (0.44–0.72)
Albumin, g/L	28 (25–32)	33 (29–36)	.003	0.88 (0.81–0.96)	0.73 (0.62–0.84)
Coexisting illnesses, No. (%)
Diabetes type 2	5 (15.2)	107 (15.7)	.928	0.96 (0.36–2.53)	0.50 (0.44–0.57)
Anemia	4 (12.1)	75 (11.0)	.845	1.11 (0.38–3.25)	0.51 (0.45–0.56)
Coronary heart disease	1 (3.0)	31 (4.6)	.681	0.65 (0.09–4.95)	0.51 (0.48–0.54)
Hypertension	12 (36.4)	328 (48.2)	.186	0.61 (0.30–1.27)	0.56 (0.47–0.64)
Congestive heart failure	2 (6.1)	75 (11.0)	.377	0.52 (0.12–2.22)	0.52 (0.48–0.57
Stroke	1 (3.0)	13 (1.9)	.654	1.60 (0.20–12.64)	0.51 (0.48–0.54)
COPD	3 (9.1)	97 (14.3)	.408	0.60 (0.18–2.01)	0.53 (0.47–0.58)
Chronic renal failure	13 (39.4)	204 (30.0)	.255	1.52 (0.74–3.11)	0.55 (0.46–0.63)
Neoplastic disease	3 (9.1)	107 (15.7)	.310	0.54 (0.16–1.79)	0.53 (0.48–0.58)
Immunosuppressionc	6 (18.2)	49 (7.2)	.027	2.86 (1.13–7.26)	0.55 (0.49–0.62)
Solid organ transplantation	1 (3.0)	3 (0.4)	.095	7.05 (0.71–69.69)	0.51 (0.48–0.54)
Dementia	1 (3.0)	31 (4.6)	.681	0.65 (0.09–4.95)	0.51 (0.48–0.54)
Risk assessment, No. (%)
PSI Class I	1 (4.0)	63 (16.2)	.914	1.02 (0.72–1.45)	0.50 (0.40–0.60)
PSI Class II	8 (32.0)	57 (14.6)
PSI Class III	6 (24.0)	117 (30.0)
PSI Class IV	10 (40.0)	133 (34.1)
PSI Class V	0 (0.0)	20 (5.1)

Continuous values are presented as median and IQR, categorical/binary values as absolute number and percentage.

Abbreviations: AST, aspartate aminotransferase; AUC, area under the curve; CAP, community-acquired pneumonia; CI, confidence interval; COPD, chronic obstructive pulmonary disease; CRP, C-reactive protein; IQR, interquartile range; NLR, neutrophil-lymphocyte ratio; OR, odds ratio; PCT, procalcitonin; PSI, pneumonity severity index; WBC, white blood cell count.

aIncrease of 10.

bIncrease of 100.

cPrednisone equivalent >10 mg for >2 weeks.

In a second step, we investigated the diagnostic performance of the Legionella Score and each of the 6 included clinical and laboratory variables for discrimination of Legionella CAP from other pneumonia causes. All variables except platelet count and low sodium level were significantly associated with Legionella, with AUCs ranging between 0.64 and 0.86 as can be seen in Table 3. In a logistic regression model including all 6 nondichotomized variables (ie, without a defined cutoff), the discrimination was excellent, with an AUC of 0.91 (95% confidence interval [CI], 0.86–0.96). The Legionella Score based on the predefined cutoffs provided an odds ratio of 2.72 (95% CI, 2.03–3.65; P < .001) per point increase in the score and an AUC of 0.83 (95% CI, 0.76–0.90).

Table 3.

 Association of the Clinical and Laboratory Parameters of the

	AUC (95% CI)	OR (95% CI)	P-Value
Nondichotomized variables
Dry cough	0.64 (0.55–0.72)	3.50 (1.73–7.10)	.001
Temperature	0.73 (0.64–0.81)	2.25 (1.55–3.27)	<.001
C-reactive protein	0.86 (0.82–0.90)	2.29 (1.78–2.95)	<.001
Lactate dehydrogenase	0.69 (0.57–0.81)	1.12 (1.00–1.26)	.048
Sodium	0.67 (0.56–0.78)	0.97 (0.94–1.00)	.074
Platelet counts	0.56 (0.45–0.66)	0.97 (0.93–1.02)	.224
Combined model	0.91 (0.86–0.96)

Predictive performance of each variable in univariate analysis and combined in multivariate logistic regression analysis.

Abbreviations: AUC, area under the curve; CI, confidence interval; OR, odds ratio.

We also assessed the diagnostic performance of the score at the different cutoffs as well as at the optimal cutoffs as originally proposed with regard to sensitivity, specificity, and positive and negative predictive value (Figure 1; Table 1 and Table 4). The positive predictive value to diagnose Legionella increased stepwise from 0% with a score of 0%–40% (95% CI, 12.2%–73.8%) with at least 5 points and 50% (95% CI, 1.3%–98.7%) if all 6 points were present. At a score of ≥4 points, the positive predictive value of the Legionella Score was 40%, which represents an 8-fold increase compared with a prevalence of 5%. A score of ≥5 had a specificity of 99.1% (95% CI, 98.1%–99.7%). Conversely, a score of <2 points had a sensitivity of 97% (95% CI, 84.2%–99.9%), with a negative predictive value of 99.4% (95% CI, 96.4%–100.0%) to rule out Legionella in CAP. With the implementation of a cutoff <2 points, Legionella could have been ruled out correctly in 153 patients (22.5%), while missing 1 patient (3.0%) with Legionella pneumonia.

Figure 1.

 Sensitivity compared with specificity of the Legionella Score at each scoring cut-point. Abbreviation: OR, odds ratio.

Table 4.

Predictive Performance of the Legionella Score at Each Scoring Cut-Point

	Sensitivity (95% CI)	Specificity (95% CI)	PPV (95% CI)	NPV (95% CI)
Score ≥ 1	97.0 (84.2–99.9)	22.5 (19.4–25.8)	5.7 (3.9–8.0)	99.4 (96.4–100.0)
Score ≥ 2	90.9 (75.7–98.1)	59.3 (55.5–63.0)	9.8 (6.7–13.7)	99.3 (6.7–13.7)
Score ≥ 3	63.6 (45.1–79.6)	85.7 (82.9–88.3)	17.8 (11.4–25.9)	98.0 (96.5–99.0)
Score ≥ 4	27.3 (13.3–45.5)	96.8 (95.1–98.0)	29.0 (14.2–48.0)	96.5 (94.8–97.7)
Score ≥ 5	12.1 (3.4–28.2)	99.1 (98.1–99.7)	40.0 (12.2–73.8)	95.9 (94.1–97.2)
Score = 6	3.0 (0.1–15.8)	99.9 (99.2–100.0)	50.0 (1.3–98.7)	95.5 (93.7–96.9)

Sensitivity, specificity, PPV, and NPV of the Legionella Score at each scoring cut-point.

Abbreviations: CI, confidence interval; NPV, negative predictive value; PPV, positive predictive value.

Finally, we investigated whether addition of other clinical and laboratory parameters would further increase the performance of the Legionella Score. Nausea and diarrhea were significant predictors of Legionella pneumonia, with odds ratios of 3.95 (95% CI, 1.90–8.21; P < .001) and 4.23 (95% CI, 2.00–8.91; P < .001), but added no significant benefit when added to the score.

DISCUSSION

Our analysis validates the previously suggested Legionella Score and shows high validity of the score to differentiate between Legionella and other etiologies of CAP. With a negative predictive value of 99.4% (95% CI, 96.4%–100.0%) for a score <2 points, it was reliable to rule out Legionella and thus, in conjunction with previous similar results, may support clinical decision-making regarding further diagnostic testing and therapeutic management [9-11].

Interestingly, the performance of the score was not improved by the addition of other parameters including nausea or diarrhea, although both parameters were significant predictors of Legionella in univariate analysis. Despite the lack of statistical significance in the overall cohort, specific clinical clues from the patient history (eg, use of air condition, gardening, close contact with patients suffering from Legionella) may help to increase the likelihood for Legionella in individual patients. Thus, this score should not be a substitute for an in-depth risk assessment for Legionella but provides an objective estimate on the probability in an individual patient.

Current testing options for Legionella infection such as urinary antigen tests (covering Legionella pneumophila serotype 1, which accounts for approximately 80% of community-acquired cases), polymerase chain reaction (PCR; covering a manufacturer-defined set of Legionella pneumophila serotypes), and serology (covering Legionella pneumophila) all have limited sensitivity, whereas culture of respiratory samples, although providing the best sensitivity, is limited by important time delays [6]. Herein, the use of a systematic score, such as the Legionella Score, is an important additional tool in the work-up of patients with pneumonia. Use of this test may also help to reduce unnecessary antibiotics in some patients and provide more targeted treatment in other patients, which may translate into reducing the risk of side effects and antimicrobial resistance, which is a major public health issue of global interest [7, 8]. Taking the diagnostic limitations of urinary antigen testing into account, empirical antibiotic therapy for atypical pathogens in patients with negative urinary antigen testing but high diagnostic probability (diagnostic score ≥4 corresponding to a specificity of 96.8%) seems reasonable. On the other hand, in low-risk patients with low diagnostic probability (diagnostic score <2 corresponding to a sensitivity of 97%), Legionella seems highly unlikely, which again may influence diagnostic and therapeutic choices. The reliability of the Legionella Score in this analysis in a real-life cohort emphasizes the robustness of these parameters for clinical use. The introduction of the Legionella Score in the diagnostic armamentarium of an emergency department may thus improve the detection of Legionella pneumonias and thus improve the therapeutic management of patients hospitalized with CAP, enabling more targeted antibiotic use.

The limitations of this report include the lack of pathogen information in some of the patients with presumed viral pneumonia, possible missed Legionella cases due to low sensitivity of the urine antigen test, and possible false-positive results regarding the association of clinical parameters and Legionella due to multiple testing.

CONCLUSIONS

In this independent sample of patients with CAP, the previously proposed Legionella Score performed well in estimating the likelihood of Legionella infection and thus may help to direct diagnostic and therapeutic decisions.

Supplementary Data

Supplementary materials are available at Open Forum Infectious Diseases online. Consisting of data provided by the authors to benefit the reader, the posted materials are not copyedited and are the sole responsibility of the authors, so questions or comments should be addressed to the corresponding author.