Multicentre validation and update of a Legionella prediction score to guide testing and treatment in community-acquired pneumonia

This study externally validated and updated a prediction score for community-acquired Legionnaires' disease using data from 20 Swiss hospitals, resulting in a simplified "SwissLEGIO" tool that maintains high sensitivity while improving specificity to potentially reduce unnecessary microbiological testing in hospitalized pneumonia patients.

The Gist

Imagine you are a doctor standing in a busy emergency room. A patient walks in with pneumonia (a lung infection). You know that most pneumonia is caused by common bacteria that respond well to standard antibiotics. However, there is a sneaky, dangerous type of pneumonia called Legionnaires' disease caused by Legionella bacteria.

Here's the tricky part: Standard antibiotics don't work on Legionella. You need special, stronger drugs to kill it. If you wait too long to give the right medicine, the patient could get much sicker or even die.

So, the big question is: How do you know which patient has the sneaky Legionella without testing everyone?

The Problem: The "Fishing Net" is Too Big

Currently, doctors often have to test everyone with pneumonia to find the Legionella cases. But Legionella is rare (only about 4 out of 100 pneumonia cases). Testing everyone is like using a giant fishing net to catch one tiny goldfish in the ocean. It's expensive, time-consuming, and creates a lot of "false alarms" (catching regular fish when you're looking for goldfish).

The Old Tool: A Clunky, Heavy Net

A few years ago, doctors created a "prediction score" (a checklist) to help decide who to test. It looked at six things:

1. High fever
2. No phlegm (dry cough)
3. Low salt in the blood
4. High inflammation markers
5. High levels of a liver enzyme (LDH)
6. Low platelet count (blood cells)

The problem with the old checklist:

• It was too heavy: One of the items (LDH) isn't even checked in many hospitals because it's not routine.
• It was too sensitive: It caught almost all the Legionella cases (which is good), but it also flagged way too many healthy people as "suspicious." It was like a metal detector that beeps for every piece of metal, even a paperclip, making it hard to find the real treasure.

The New Solution: The "SwissLEGIO" Score

The researchers in this study (from 20 Swiss hospitals) decided to upgrade the checklist. They wanted a tool that was lighter, easier to use, and smarter.

They took the old checklist and made some changes, creating the SwissLEGIO Score. Think of it as upgrading from a heavy, rusty net to a high-tech, laser-guided net.

Here is the new, simplified checklist (The SwissLEGIO Score):

1. Fever: Is it over 38°C (100.4°F)? (They lowered the threshold because even a moderate fever is a warning sign.)
2. Cough: Is it dry (no phlegm)?
3. Salt: Is the sodium in the blood low?
4. Inflammation: Is the CRP (a marker of infection) very high?
5. Medicine History: Did the patient already take standard antibiotics (beta-lactams) before coming to the hospital? (If they took the "wrong" antibiotics and didn't get better, it's a big clue it might be Legionella.)

What they threw out:

• They removed the "LDH" test because it's rarely done anyway.
• They removed the "platelet count" because it wasn't a good predictor.

How It Works in Real Life

Imagine the score is a traffic light system:

• Score < 2 (Green Light): The patient is very unlikely to have Legionella. You can safely skip the expensive test and treat them with standard pneumonia medicine. This saves money and time.
• Score ≥ 2 (Red Light): The patient has a higher risk. You should immediately order the specific Legionella test and start the special antibiotics just in case.

The Results: A Smarter Net

The study found that this new "SwissLEGIO" score is fantastic:

• It still catches the bad guys: It is still very good at finding the real Legionella cases (high sensitivity).
• It stops the false alarms: It is much better at ignoring the regular pneumonia cases (higher specificity).
• It saves resources: By using this score, hospitals could reduce the number of unnecessary tests by 36% to 52%. That's like cutting your grocery bill in half because you stopped buying food you didn't need.

The Bottom Line

This paper is about giving doctors a better map. Instead of guessing or testing everyone blindly, the SwissLEGIO score gives them a simple, 5-point checklist to quickly identify who really needs the special treatment. It's a win for patients (faster, correct treatment) and a win for the healthcare system (less waste).

In short: They took a complicated, clunky rulebook and turned it into a simple, efficient guide that helps doctors catch the dangerous "sneaky" pneumonia without wasting time on the harmless stuff.

Technical Summary

1. Problem Statement

Community-acquired Legionnaires' disease (CALD) accounts for 4–7% of community-acquired pneumonia (CAP) cases in Europe. Early diagnosis is critical because effective treatment requires specific antibiotics (macrolides, quinolones, or tetracyclines) that achieve high intracellular concentrations, whereas standard $\beta$-lactams are ineffective.

• The Clinical Dilemma: Current guidelines differ globally. European/Swiss guidelines recommend routine Legionella testing (e.g., Urinary Antigen Test [UAT]) for all hospitalized CAP patients, while US guidelines limit testing to severe cases or those with specific risk factors.
• The Issue: Routine testing yields low positivity rates (1.5–3%), raising concerns about cost-effectiveness. Conversely, selective testing based on severity or epidemiology often fails to identify CALD patients.
• The Gap: Existing clinical prediction scores, such as the one developed by Fiumefreddo et al., have not been validated in settings where routine testing is recommended, and their practical applicability is hindered by missing data (e.g., LDH levels) and suboptimal predictive performance in real-world scenarios.

2. Methodology

The study employed a multicentre, case-control design involving 20 Swiss hospitals.

• Study Population:
  • Cases: 196 patients with confirmed CALD (August 2022–March 2024).
  • Controls: 196 Legionella-negative CAP patients, matched by diagnosis date ($\pm$14 days).
  • Inclusion: Patients $\ge$18 years with clinical pneumonia (new infiltrate + symptoms) discharged alive.
  • Exclusion: Hospital-acquired pneumonia, travel-associated pneumonia, confirmed COVID-19.
• Data Collection: Electronic medical records were reviewed for demographics, comorbidities, clinical symptoms, and laboratory values (CRP, LDH, platelets, sodium, etc.).
• Statistical Approach:
  • Validation: The original Fiumefreddo score (6 predictors) was validated. Missing data (particularly LDH) were handled via multiple imputation (40 datasets).
  • Performance Metrics: Sensitivity, specificity, AUC, PPV, and NPV were calculated at various cut-offs.
  • Score Update: The dataset was split into a development set ($n=236$) and a validation set ($n=156$). New predictors were identified via univariable/multivariable logistic regression and classification tree analysis.
  • Goal: To create a simplified score that maintains high sensitivity (to rule out disease) while improving specificity (to reduce unnecessary testing).

3. Key Contributions

The study makes three primary contributions:

1. External Validation: It provides the first large-scale external validation of the Fiumefreddo Legionella prediction score in a setting where routine testing is standard practice.
2. Identification of Practical Limitations: It highlights that the original score's reliance on LDH is a major barrier to implementation, as LDH was missing in 57.7% of CALD patients in routine care. It also identifies platelet count as a poor predictor.
3. Development of the SwissLEGIO Score: It proposes a simplified, updated prediction model that removes weak predictors, adjusts fever thresholds, and adds clinically relevant variables (prior $\beta$-lactam therapy) to optimize diagnostic accuracy.

4. Results

Validation of the Original Score:

• Performance: At a cut-off $\ge$2, the original score showed high sensitivity (91.2%) but low specificity (35.3%).
• Missing Data: LDH was frequently unavailable. Removing LDH from the score improved specificity (61.0%) but reduced sensitivity (83.2%).
• Predictor Strength: Elevated CRP, hyponatremia, and dry cough were strong predictors. Platelet count and LDH were weak.

The Updated SwissLEGIO Score:
The new score consists of 5 predictors (1 point each):

1. Fever >38.0°C (Lowered from original >39.4°C).
2. Sodium <133 mmol/L (Hyponatremia).
3. CRP >180 mg/L.
4. No cough or dry cough (Absence of sputum).
5. Prior $\beta$-lactam therapy (New addition).

Performance of SwissLEGIO Score (Cut-off $\ge$2):

• Sensitivity: Maintained high levels (88–92% in validation).
• Specificity: Significantly improved compared to the original score (46–58% vs. 29–35%).
• AUC: Improved from 0.70 (original) to 0.77–0.78 (SwissLEGIO) in the validation set.
• Negative Predictive Value (NPV): At a 4% prevalence, the NPV for a score <2 was 98.9%, effectively ruling out CALD.

Impact on Testing:

• Using the SwissLEGIO score to screen patients could reduce the number of required microbiological tests by 36–52% compared to universal testing, while maintaining a high detection rate for CALD cases.

5. Significance and Conclusion

• Clinical Utility: The SwissLEGIO score serves as an effective "rule-out" tool. Patients with a score <2 can likely be spared from Legionella testing, reducing costs and laboratory workload without significantly increasing the risk of missed diagnoses.
• Guidance for Empiric Therapy: The score helps clinicians decide on empiric antibiotic coverage. A high score suggests the need for intracellular-active antibiotics (e.g., macrolides/quinolones) alongside $\beta$-lactams.
• Global Relevance: While developed in Switzerland, the score addresses the tension between European (routine testing) and US (selective testing) guidelines. It offers a pragmatic middle ground, particularly useful in settings where routine testing is not recommended or resources are limited.
• Limitations: The study used a case-control design (preventing direct prevalence estimation) and relied heavily on UAT (which detects L. pneumophila serogroup 1 best). External validation in other geographic regions is recommended before widespread guideline adoption.

Final Recommendation: The authors propose the SwissLEGIO score as a simplified, easy-to-apply screening tool to guide testing and treatment decisions in hospitalized CAP patients, potentially reducing unnecessary testing by nearly half while ensuring high sensitivity for CALD detection.