A novel risk stratification model based on creatine kinase and direct bilirubin for identifying severe Chlamydia psittaci pneumonia: a double-center, hypothesis-generating retrospective cohort study

Abstract

Background

Chlamydia psittaci is a global zoonotic pathogen causing psittacosis, which can progress to severe pneumonia and multi-organ dysfunction. Identifying reliable predictors for severe disease is crucial for improving outcomes, yet current community-acquired pneumonia severity scores may not adequately capture psittacosis-specific systemic manifestations. However, the proposed model is intended to identify patients who already meet criteria for severe pneumonia at presentation, not to predict progression from non-severe to severe disease.

Methods

We conducted a retrospective cohort study of 36 hospitalized patients with confirmed C. psittaci pneumonia, classified into severe (n = 11) and non-severe (n = 25) groups. We employed Firth’s penalized-likelihood logistic regression with bootstrap validation to address small sample size limitations and potential model overfitting.

Results

Our analysis establishes a comprehensive framework organized around four key pillars: (1) Radiological evolution - serial computed tomography demonstrated characteristic temporal progression with peak severity at 2 weeks (median Total CT Severity Score: 7), followed by gradual resolution; (2) Biomarker validation - severe cases showed numerically higher creatine kinase (CK) (415.0 vs. 170.0 U/L) and direct bilirubin (DBil) (9.4 vs. 3.8 µmol/L), with exceptional bootstrap stability (relative bias: -8.9% and -7.5%, respectively); however, no variable reached statistical significance (P < 0.05) in univariable Firth regression, and confidence intervals were wide and included the null; (3) Clinical translation - we developed a risk stratification model categorizing patients into four distinct risk groups based on CK and DBil thresholds; (4) Pathophysiological basis - we propose the “lung-liver-muscle” axis as a hypothesis-generating conceptual framework for systemic disease progression.

Conclusion

This exploratory study identifies CK and DBil as stable, readily available biomarkers associated with psittacosis severity at presentation. The integrated four-pillar approach provides hypothesis-generating risk stratification tools and pathophysiological insights, potentially aiding early identification of high-risk patients. All findings require external validation in larger prospective cohorts before any clinical application.

Clinical trial number

Not applicable.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12879-026-13738-0.

Background

Chlamydia psittaci, the etiological agent of psittacosis (ornithosis), is an obligate intracellular bacterium transmitted to humans from infected birds [1, 2]. While bird exposure is the main risk factor, limited human-to-human transmission has been documented, which can pose an additional challenge in outbreak settings [3, 4]. It is an underdiagnosed cause of community-acquired pneumonia (CAP) that can lead to severe complications, including acute respiratory distress syndrome (ARDS), septic shock, and multi-organ dysfunction, with a reported mortality rate of up to 15% if untreated [5–8].

The clinical presentation of psittacosis is highly variable and non-specific, often mimicking other types of CAP, which frequently leads to delays in diagnosis and appropriate antibiotic therapy [9–12]. The diagnostic challenge is underscored by a high misdiagnosis rate of 50–80% [2]. While rapid diagnostic methods like metagenomic next-generation sequencing (mNGS) have improved detection, predicting which patients will develop severe illness remains a significant clinical challenge [13–16]. Current severity scores for CAP, such as CURB-65 and PSI, may not fully capture the unique systemic and extrapulmonary manifestations of psittacosis, which can include hepatitis and myositis [7, 17, 18]. Furthermore, comprehensive quantitative description of temporal radiographic evolution remains limited.

Several routinely available biomarkers have been identified as potential predictors of disease severity and prognosis in psittacosis. The profile includes decreased hemoglobin and elevated levels of absolute lymphocyte count, neutrophils, neutrophil-to-lymphocyte ratio, C-reactive protein, procalcitonin, lactate dehydrogenase, creatine kinase (CK), alanine aminotransferase, aspartate aminotransferase, blood urea nitrogen, creatinine, D-dimer, and interleukin-6 [3, 17, 19–21]. This pattern of biomarker elevation reflects the bacterium’s recognized capacity to directly affect multiple organ systems, including cardiac, hepatic, muscular, and central nervous system function [22]. Consequently, these markers could enhance risk stratification for the spectrum of extrapulmonary involvements. However, findings across existing studies remain inconsistent. A major confounding factor is the high prevalence (31.8%–73.3%) of suspected co-infections, which creates diagnostic ambiguity and makes it difficult to definitively attribute organ dysfunction solely to C. psittaci versus a concurrent infection [14, 23, 24]. A key methodological strength of the present study is that, upon enrollment, we systematically excluded all cases with any laboratory evidence of co-infection based on comprehensive mNGS testing. This design uniquely allows for the attribution of observed systemic manifestations—such as hepatic and muscular injury—specifically to C. psittaci infection, minimizing diagnostic ambiguity. Among these biomarkers, CK, a marker of muscle injury, and direct bilirubin (DBil), a marker of hepatocellular or cholestatic injury, are routinely available, but their combined prognostic value and stability for risk stratification in psittacosis require systematic evaluation.

This study aimed to comprehensively describe clinical features and radiographic progression using serial Total CT Severity Score (CTSS) measurements and identify robust predictors of severe disease [25]. We specifically focused on evaluating CK and DBil as potential biomarkers. Utilizing Firth’s regression with bootstrap validation to address small sample size limitations, we sought to develop a biomarker-based risk stratification tool and investigate the pathophysiological model underlying systemic progression.

Methods

Study design and participants

This retrospective cohort study was conducted at two tertiary care hospitals. The study was approved by the Institutional Review Board/Ethics Committee of the participating institutions, which waived the requirement for informed consent.

We screened patients aged ≥ 18 years hospitalized with pneumonia and positive mNGS for C. psittaci from bronchoalveolar lavage fluid or blood between April 2020 and July 2025. A total of 65 patients were initially identified. Pneumonia was defined by new pulmonary infiltrates on imaging plus relevant clinical symptoms.

Exclusion criteria included: (a) co-infection with other respiratory pathogens; (b) incomplete medical records; (c) hospital-acquired pneumonia. After applying these criteria, 36 patients were included in the final analysis, and 29 were excluded.

mNGS was performed on bronchoalveolar lavage fluid (n = 31) or blood (n = 5). Positivity for C. psittaci was defined as reads per million ≥ 3. Co-infection was defined as concurrent detection of other bacterial or viral pathogens meeting standard diagnostic criteria (e.g., read count ratio ≥ 10 times background reads or recognized pathogenic species with clinical correlation). The median turnaround time for mNGS results was 48 h (IQR 36–72 h). All patients also underwent routine conventional microbiological testing, including blood cultures, sputum cultures, and respiratory viral PCR panels (influenza, RSV, adenovirus, etc.); no additional respiratory pathogens were identified.

Data collection and definitions

Data were extracted using a standardized case report form, including:

• Demographics and history: age, sex, and poultry exposure. Poultry exposure was assessed through standardized patient interviews conducted by attending physicians at admission, using a structured questionnaire that inquired about direct contact with birds (including poultry, pet birds, or wild birds) or exposure to bird-contaminated environments within two weeks before symptom onset; data were subsequently extracted from electronic medical records.

• Clinical features: symptoms, maximum body temperature, lowest SpO₂, duration of chief complaint.

• Comorbidities and complications: including respiratory failure, septic shock, and hepatic/cardiac dysfunction.

• Laboratory investigations: complete blood count, coagulation profile, serum chemistry (liver enzymes, bilirubin, albumin, CK), and inflammatory biomarkers (C-reactive protein, procalcitonin, interleukin-6).

• Radiological findings: Chest computed tomography (CT) scans were independently reviewed by two blinded radiologists. The Total CTSS was calculated to quantify the extent of lung involvement, with a possible range of 0 (no involvement) to 25 (maximum involvement). Each of the five lung lobes was visually scored from 0 to 5 based on the percentage of involvement: 0 (0%), 1 (< 5%), 2 (5–25%), 3 (26–49%), 4 (50–75%), and 5 (> 75%). The scores from all lobes were summed to obtain the total CTSS [25]. Patients were categorized into four stages based on the time from symptom onset to CT scan: Stage-1 (0–7 days), Stage-2 (8–14 days), Stage-3 (15–28 days), and Stage-4 (≥ 29 days). CT features such as ground-glass opacities, consolidation, and fibrotic streaks were documented [13, 26].

• Treatment and outcomes: antibiotic regimens, need for the Intensive Care Unit (ICU) admission, hospital length of stay, and clinical outcome.

Severe pneumonia was defined as meeting ≥ 1 criterion according to the guidelines by the Chinese Thoracic Society (Version 2016): (1) need for invasive/non-invasive ventilation; (2) ICU admission for respiratory support/septic shock; (3) development of septic shock [27].

Statistical analysis

Analyses were performed using R software (version 4.5.1; R Foundation for Statistical Computing, Vienna, Austria), including Firth’s penalized-likelihood regression and bootstrap validation. Data management and preliminary analyses were conducted using SPSS (version 29.0; IBM Corp., Armonk, NY, USA). A two-sided P-value of less than 0.05 was considered statistically significant.

Normally distributed continuous variables are presented as mean ± standard deviation (SD) and compared using the independent samples t test, while non-normally distributed data are presented as median (interquartile range, IQR) and compared using the Mann-Whitney U test. Categorical variables, expressed as No. (%), were compared using Fisher’s exact test or the Fisher-Freeman-Halton exact test, as applicable (Supplementary Appendix 1).

Handling of missing data

Variables with > 20% missing data were excluded. For variables with < 20% missingness, multiple imputation was performed using the Fully Conditional Specification (FCS) method under the Multivariate Imputation by Chained Equations (MICE) framework to create five complete datasets [28]. The imputation model was run for 10 iterations and incorporated predictive mean matching for scale variables. The imputed variables included cardiac troponin I (cTnI), N-terminal pro-B-type natriuretic peptide (NT-proBNP), interferon-gamma (IFN-γ), and interleukin-6 (IL-6). Subsequent statistical analyses were performed on each imputed dataset, and the results were pooled using Rubin’s rules to obtain final estimates and standard errors that account for the uncertainty due to missing data (Supplementary Appendix 3).

Regression modeling and bootstrap validation

Given the small sample size (N = 36) and complete separation, Firth’s penalized-likelihood logistic regression was employed (Supplementary Appendix 4). Due to the small sample size, multivariable regression was not performed; univariable Firth regression results are presented as hypothesis‑generating only. The absence of multivariable adjustment means we cannot exclude confounding by demographic factors, comorbidities, or other laboratory abnormalities. Internal bootstrap validation with 1000 resamples evaluated coefficient stability, rated by relative bias: <10% (Excellent), 10–20% (Acceptable), > 20% (Concerning) (Supplementary Appendix 5).

Power and sample size consideration

Because post-hoc power depends on the observed effect size, it does not provide additional information beyond the confidence interval and should not be used to conclude that the study was adequately powered for any specific effect. Post hoc power analysis indicated 59.9% power to detect large effect sizes (Cohen’s h = 0.8, OR = 3.51) (Supplementary Appendix 4). However, power to detect minimum clinically important effect sizes (Cohen’s h = 0.3, OR = 1.60) was limited, requiring approximately 371 patients for 80% power. These results indicate adequate power for identifying strong predictors while highlighting the exploratory nature for factors with more modest effects.

Radiological data analysis

The Friedman test was used to compare the Total CTSS across the four disease stages to evaluate the temporal evolution of radiographic severity. The cubic polynomial model was selected based on Akaike Information Criterion (AIC) comparison among linear, quadratic, cubic, and spline models; the cubic model provided the lowest AIC, indicating the best fit (Supplementary Appendix 2).

Risk stratification model development

CK and DBil were selected a priori based on prior literature and clinical relevance, and because they demonstrated the most stable bootstrap performance (relative bias < 10%) among all routinely available laboratory parameters examined. No hypothesis‑free discovery analysis (e.g., LASSO or stepwise selection) was performed due to the small sample size; therefore, the selection of CK and DBil should be considered hypothesis‑generating rather than data‑driven. Based on the distribution characteristics and bootstrap stability results of CK and DBil, clinically relevant cut-off points were selected to create a risk stratification matrix. The proposed thresholds are hypothesis‑generating, based on observed distributions and bootstrap stability; they require external validation before clinical application. This model categorizes patients into distinct risk groups to guide clinical decision‑making.

Results

Baseline characteristics and clinical outcomes

A total of 36 patients were included. Eleven patients (30.6%) developed severe pneumonia. Severe patients were older (67 ± 11 vs. 60 ± 12 years, P = 0.108) with higher prevalence of type 2 diabetes mellitus (36.4% vs. 0.0%, P = 0.006). All severe patients presented with cough (100% vs. 64%, P = 0.034), with significantly higher gastrointestinal symptoms (45.5% vs. 8.0%, P = 0.018).

Clinical outcomes revealed significant disparities. While overall improvement rate was 94.4%, all non-severe patients improved (100%) versus 81.8% in severe group (P = 0.087). ICU admission was exclusively required for severe cases (81.8% vs. 0%, P < 0.001) (as shown in Table 1).

Table 1.

Baseline characteristics and clinical features of patients with Chlamydia psittaci pneumonia by severity group

Characteristic	Overall (N = 36)	Non-severe Pneumonia (n = 25)	Severe pneumonia (n = 11)	P-value
Demographics
Age, years, mean ± SD	60 ± 13	60 ± 12	67 ± 11	0.108a
Male gender, No. (%)	25 (69.4%)	16 (64.0%)	9 (81.8%)	0.439b
Poultry exposure, No. (%)	8 (22.2%)	7 (28.0%)	1 (9.1%)	0.388b
SOFA score, mean ± SD	3.4 ± 2.1	2.1 ± 1.4	6.3 ± 2.1	< 0.001a
Vital Signs & Symptoms
Maximum body temperature, °C, mean ± SD	39.5 ± 0.7	39.4 ± 0.6	39.2 ± 1.2	0.435a
Lowest SpO₂, %, mean ± SD	93.7 ± 2.4	93.8 ± 2.4	91.7 ± 4.6	0.085a
Fever, No. (%)	35 (97.2%)	25 (100.0%)	10 (90.9%)	0.306b
Cough, No. (%)	27 (75.0%)	16 (64.0%)	11 (100.0%)	0.034b
Sputum, No. (%)	6 (16.7%)	4 (16.0%)	2 (18.2%)	1.000b
Chest discomfort, No. (%)	13 (36.1%)	9 (36.0%)	4 (36.4%)	1.000b
Dyspnea, No. (%)	14 (38.9%)	8 (32.0%)	6 (54.5%)	0.273b
Dizziness, No. (%)	12 (33.3%)	8 (32.0%)	4 (36.4%)	1.000b
Fatigue, No. (%)	18 (50.0%)	10 (40.0%)	8 (72.7%)	0.146b
Nasal congestion, No. (%)	1 (2.8%)	0 (0.0%)	1 (9.1%)	0.306b
Sore throat, No. (%)	5 (13.9%)	5 (20.0%)	0 (0.0%)	0.295b
Headache, No. (%)	9 (25.0%)	6 (24.0%)	3 (27.3%)	1.000b
Myalgias, No. (%)	12 (33.3%)	11 (44.0%)	1 (9.1%)	0.059b
Altered mental status, No. (%)	2 (5.6%)	0 (0.0%)	2 (18.2%)	0.087b
Hemoptysis, No. (%)	4 (11.1%)	2 (8.0%)	2 (18.2%)	0.057b
Gastrointestinal symptoms, No. (%)	7 (19.4%)	2 (8.0%)	5 (45.5%)	0.018b
Comorbidities, No. (%)
Hypertension	10 (27.8%)	5 (20.0%)	5 (45.5%)	0.224b
Type 2 diabetes mellitus	4 (11.1%)	0 (0.0%)	4 (36.4%)	0.006b
Complications, No. (%)
Rhabdomyolysis	1 (2.8%)	0 (0.0%)	1 (9.1%)	0.306b
Gastrointestinal bleeding	3 (8.3%)	0 (0.0%)	3 (27.3%)	0.023b
Stress ulcer	3 (8.3%)	0 (0.0%)	3 (27.3%)	0.023b
Pleural effusion	19 (52.8%)	10 (40.0%)	9 (81.8%)	0.031b
Respiratory failure	10 (27.8%)	1 (4.0%)	9 (81.8%)	< 0.001b
Type I respiratory failure	10 (27.8%)	1 (4.0%)	9 (81.8%)	< 0.001b
Type II respiratory failure	0 (0.0%)	0 (0.0%)	0 (0.0%)
Multiple organ dysfunction syndrome	1 (2.8%)	0 (0.0%)	1 (9.1%)	0.306b
Coagulopathy	2 (5.6%)	0 (0.0%)	2 (18.2%)	0.087b
Cardiac dysfunction	6 (16.7%)	1 (4.0%)	5 (45.5%)	0.006b
Hepatic dysfunction	24 (66.7%)	15 (60.0%)	9 (81.8%)	0.268b
Renal dysfunction	6 (16.7%)	2 (8.0%)	4 (36.4%)	0.057b
Hypoalbuminemia	26 (72.2%)	15 (60.0%)	11 (100.0%)	0.016b
Laboratory investigations, median (IQR)
Neutrophil count, ×10⁹/L	9.1 (5.0)	6.8 (3.9)	9.1 (5.0)	0.220c
Neutrophil percentage, %	89.2 (6.8)	83.8 (34.0)	89.2 (6.8)	0.022c
Lymphocyte count, ×10⁹/L	0.6 (0.4)	0.9 (0.8)	0.6 (0.4)	0.063c
Lymphocyte percentage, %	6.6 (3.7)	8.6 (7.4)	6.6 (3.7)	0.080c
D-dimer, mg/L FEU	0.69 (1.33)	0.35 (0.73)	0.69 (1.33)	0.045c
International normalized ratio	1.32 (0.26)	1.16 (0.13)	1.32 (0.26)	0.049c
Albumin, g/L	34.7 (5.5)	35.2 (5.1)	32.6 (5.2)	0.029c
Alanine aminotransferase, U/L	45.0 (59.8)	44.0 (38.0)	108.4 (116.0)	0.024c
Aspartate aminotransferase, U/L	58.5 (91.0)	51.0 (75.5)	104.0 (146.0)	0.032c
Creatine kinase, U/L	185.0 (361.7)	170.0 (274.0)	415.0 (3230.0)	0.058c
Creatine kinase-MB, U/L	13.0 (13.5)	13.0 (9.0)	15.0 (77.0)	0.068c
Total bilirubin, µmol/L	16.4 (14.8)	14.9 (11.6)	21.4 (19.8)	0.100c
Direct bilirubin, µmol/L	5.4 (7.3)	3.8 (6.2)	9.4 (11.4)	0.049c
Indirect bilirubin, µmol/L	10.8 (8.2)	10.1 (5.5)	13.1 (8.3)	0.233c
C-reactive protein, mg/L	164.1 (99.5)	152.0 (99.1)	184.9 (106.2)	0.261c
Procalcitonin, µg/L	0.45 (2.24)	0.34 (0.88)	1.46 (8.13)	0.026c
Cardiac troponin I, µg/L	0.007 (0.027)	0.006 (0.012)	0.025 (0.104)	< 0.001c
N-terminal pro-B-type natriuretic peptide, pg/mL	631 (979)	519 (873)	1741 (2401)	0.017c
Interferon gamma, pg/mL	44.76 (81.87)	38.95 (73.06)	65.26 (326.78)	0.044c
Interleukin-6, pg/mL	136.71 (219.28)	115.78 (164.70)	282.1 (464.79)	< 0.001c
Clinical outcome, No. (%)
Improved	34 (94.4)	25 (100.0)	9 (81.8)	0.087d
Discharge against medical advice	1 (2.8)	0 (0.0)	1 (9.1)
Death	1 (2.8)	0 (0.0)	1 (9.1)
ICU admission, No. (%)	9 (25.0)	0 (0.0)	9 (81.8)	< 0.001b

Data are presented as No. (%), mean ± SD, or median (IQR). P-values were derived from ^aIndependent samples t test, ^bFisher’s exact test, ^cMann-Whitney U test, ^dFisher-Freeman-Halton exact test

Abbreviations: ICU, intensive care unit; IQR, interquartile range; SD, standard deviation; SOFA, Sequential Organ Failure Assessment; SpO₂, peripheral oxygen saturation

Complications and organ dysfunctions

As expected, complications were more frequent in the severe group. Respiratory failure was the most defining complication, occurring in 9 (81.8%) severe patients compared to 1 (4.0%) in the non-severe group (P < 0.001). Hypoalbuminemia was universal in the severe group (100% vs. 60%, P = 0.016). Hepatic dysfunction was common in both groups but more so in severe cases (81.8% vs. 60.0%, P = 0.268). Cardiac dysfunction was significantly more prevalent in the severe group (45.5% vs. 4.0%, P = 0.006). Sequential Organ Failure Assessment (SOFA) scores were significantly higher in the severe group (6.3 ± 2.1 vs. 2.1 ± 1.4, P < 0.001).

Laboratory findings and predictors

Key laboratory differences are shown in Table 1. Neutrophil percentage was significantly higher in the severe group (89.2% vs. 83.8%, P = 0.022), while lymphocyte counts were lower (0.6 vs. 0.9 × 10⁹/L, P = 0.063). Coagulation parameters like international normalized ratio were higher in severe patients (1.32 vs. 1.16, P = 0.049). Markers of inflammation and organ damage, including procalcitonin, cTnI, NT-proBNP, IFN-γ, and IL-6, were markedly elevated in the severe group.

The most notable findings were the differences in CK and DBil. The median CK level was more than twice as high in the severe group (415.0 U/L vs. 170.0 U/L, P = 0.058). Similarly, the median DBil level was 2.5 times higher in the severe group (9.4 µmol/L vs. 3.8 µmol/L, P = 0.049).

Firth regression and bootstrap validation

Given the sample size limitations, Firth’s penalized-likelihood logistic regression was used to assess the association between variables and disease severity. Consistent with the exploratory nature of this study, no variable reached the traditional threshold of statistical significance (P < 0.05) in univariable analysis. Numerically elevated odds ratios were observed for several variables, including CK (aOR 1.00, 95% CI 1.00–1.00) and DBil (aOR 1.07, 95% CI 0.45–2.59); however, confidence intervals were wide and included the null, indicating that these findings may be consistent with no true effect (Table 2).

Table 2.

Identification of stable biomarkers for severe C. psittaci pneumonia: regression and bootstrap stability analysis

A. Primary regression and stability analysis of key variables
Variable	aOR	95% CI	P-value	Relative Bias (%)	Stability Rating
Stable Biomarkers
CK	1.00	1.00–1.00	0.231	-8.9	Excellent
DBil	1.07	0.45–2.59	0.832	-7.5	Excellent
Other Variables
CK-MB	0.97	0.89–1.06	0.434	19.4	Acceptable
Age	0.93	0.77–1.12	0.342	77.5	Concerning
Albumin	0.85	0.42–1.72	0.546	82.2	Concerning
ALT	1.03	0.95–1.11	0.426	-94.8	Concerning
Other 22 variables (see Supplementary Appendix 5)	…	…	…	…	Concerning

B. Distribution and Secondary Analysis of Stable Biomarkers
Category	CK (U/L)	DBil (µmol/L)
Distribution by Outcome
Severe Group (n = 11)	415 (72–29353)	9.4 (2.5–28.8)
Non-Severe Group (n = 25)	170 (29–1034)	3.8 (1.4–56.2)
Secondary Tests
Mann-Whitney U Test	P = 0.059	P = 0.050
Fisher’s Exact Test	P = 1.000	P = 0.146

Data are presented as median (range) for the “Distribution by Outcome” section. Between-group comparisons used the Mann-Whitney U test. For Fisher’s exact test, variables were dichotomized (CK: 200 U/L; DBil: 5 µmol/L). Firth’s penalized-likelihood logistic regression was used to address small sample size (N = 36) and complete separation. Variable stability was assessed via bootstrap validation (1000 samples) and rated by relative bias: <10% (Excellent), 10–20% (Acceptable), > 20% (Concerning). Wide confidence intervals reflect Firth’s correction for finite sample bias

Abbreviations: aOR, adjusted odds ratio; ALT, alanine aminotransferase; CI, confidence interval; CK, creatine kinase; CK-MB, creatine kinase myocardial band; DBil, direct bilirubin

Bootstrap validation identified CK and DBil as the most stable variables (relative bias: -8.9% and − 7.5%, respectively), classified as “Excellent” stability. In contrast, most other variables exhibited low stability (relative bias > 20%), confirming the challenge of identifying reliable predictors in this small dataset and highlighting the relative robustness of coefficient estimation for CK and DBil.

Radiological evolution

Among the 36 patients, a total of 62 CT scans were performed. The median number of CT scans per patient was 2 (IQR 1–3); patients with severe pneumonia underwent a median of 3 scans (IQR 2–4), while non-severe patients underwent a median of 1 scan (IQR 1–2). The temporal evolution of chest CT findings is summarized in Supplementary Appendix 2, with the continuous change in radiographic severity over time detailed in Fig. 1. The Total CTSS demonstrated characteristic temporal evolution, peaking in early stages (Stage-1: median 7; Stage-2: median 7) and decreasing thereafter (Stage-3: median 4; Stage-4: median 1); however, the Friedman test across the four stages was not statistically significant (P = 0.392). Therefore, the observed trend should be interpreted as descriptive and hypothesis‑generating, not as definitive evidence of temporal change. The cubic polynomial fit (y = 0.0003x³ − 0.023x² + 0.300x + 6.266) is provided for visualization only and does not imply distinct biological phases. The apparent decline at later time points may be influenced by selective imaging practices (e.g., fewer scans in recovered patients) rather than true disease evolution. This progression pattern was further elucidated by continuous modeling, which demonstrated that the dynamic change in the Total CTSS over the first 60 days was best described by a third-order polynomial equation (y = 0.0003x³ − 0.023x² + 0.300x + 6.266). The cubic term is primarily descriptive; it may reflect the complex interplay of acute inflammation, resolution, and fibrotic remodeling phases, but this model is hypothesis‑generating rather than biologically prescriptive (Supplementary Appendix 6).

Fig. 1.

Temporal evolution of the total computed tomography severity score (CTSS) in patients with Chlamydia psittaci pneumonia. (A) Scatter plot with connected lines shows the individual Total CT Severity Score (CTSS) trajectories over 60 days from symptom onset, illustrating peak radiographic involvement within the first two weeks followed by gradual resolution. Lines were made semi-transparent to reduce overlap, and individual trajectories are color-coded by severity group. (B) Nonlinear regression analysis of the Total CTSS over time. The apparent decline at later time points is based on fewer observations and may reflect clinical practice patterns rather than true disease evolution. One outlier (the Total CTSS = 23 on day 11) was excluded using the ROUT method (Q = 1%). Abbreviations: CTSS, CT Severity Score

The right lower lobe was consistently the most involved lobe across all stages. Imaging features demonstrated a clear temporal pattern: acute inflammatory manifestations such as ground-glass opacity (100% in Stage-1 and Stage-2) and consolidation (96.4% in Stage-1) predominated in the early phases, while fibrotic streaks emerged and increased in frequency during the later stages (35.3% in Stage-3 and 28.6% in Stage-4). Correspondingly, multiple lobe involvement was common in the early disease course (71.4% in Stage-1, 76.9% in Stage-2) but became substantially less frequent during convalescence (21.4% in Stage-4).

To integrate radiologic and biomarker analyses, Spearman correlation analysis showed a weak positive correlation between peak CK and peak Total CTSS (ρ = 0.22, 95% CI: − 0.12 to 0.51, P = 0.199) and a moderate positive correlation between peak DBil and peak Total CTSS (ρ = 0.33, 95% CI: 0.00 to 0.60, P = 0.047), supporting the association between biomarker elevation and radiographic disease burden, though neither reached statistical significance. Detailed correlation matrices, including scatterplots and stratified analyses by severity group, are presented in Supplementary Appendix 9.

Proposed clinical risk stratification model

We propose a practical risk stratification model based on admission levels of CK and DBil. This model is intended to identify patients who already meet the criteria for severe pneumonia at presentation, rather than to predict future progression from mild to severe disease. Given the retrospective design, the model serves as a risk‑stratification tool for early recognition of high‑risk patients upon hospital admission, facilitating timely escalation of monitoring and treatment. The model categorizes patients into four groups based on CK and DBil thresholds: (1) Low Risk: DBil < 5 µmol/L and CK < 200 U/L; (2) Intermediate Risk: DBil 5–7.5 µmol/L or CK 200–400 U/L; (3) High Risk: DBil 7.5–15 µmol/L or CK 400–1000 U/L; (4) Very High Risk: DBil > 15 µmol/L or CK > 1000 U/L (Table 3).

Table 3.

Clinical decision matrix for risk stratification and patient management

Risk Category	DBil (µmol/L)	CK (U/L)	Monitoring Level	Re-assessment	Treatment (hypothetical) *
Low Risk	< 5	< 200	General Ward	Weekly	Standard Antimicrobial Therapy
Intermediate Risk	5–7.5	200–400	High-Dependency Unit	Every 72 h	Standard Therapy + Basic Organ Support
High Risk	7.5–15	400–1000	ICU Pre-admission	Every 48 h	Enhanced Therapy + Active Organ Support
Very High Risk	> 15	> 1000	Intensive Care Unit	Every 24 h	Multidisciplinary Comprehensive Care

* The proposed thresholds are hypothesis‑generating, based on observed distributions and bootstrap stability; they require external validation before clinical application

The recommended clinical actions (monitoring level, reassessment frequency, and treatment intensity) are illustrative and derived from expert consensus rather than empirical outcome data from this cohort. They should not be used directly for clinical decision-making without prospective validation

The proposed thresholds are hypothesis-generating, based on observed distributions and bootstrap stability; they require external validation before clinical application. This stratification directly correlates with recommended clinical actions, including the appropriate level of monitoring (from weekly on a general ward to every 24 h in the ICU), the intensity of care setting, and the escalation of treatment (from standard antimicrobial therapy to multidisciplinary comprehensive care).

Discussion

This exploratory study identifies CK and DBil as promising biomarkers associated with severity in psittacosis pneumonia, synthesizing findings into four key pillars: radiological evolution, biomarker association, risk stratification, and pathophysiological framework. The proposed model is designed to identify patients who already have severe pneumonia at presentation, not to predict future progression from mild to severe disease.

Radiological evolution and pathophysiological basis

Serial CT revealed a characteristic temporal pattern peaking within two weeks, followed by a phase of gradual resolution and occasional fibrotic residue [26]. This evolutionary trajectory can be attributed to the underlying pathophysiological processes. The initial peak in severity scores corresponds to the acute phase of intense host immune-inflammatory response to C. psittaci infection. Following inhalation, this obligate intracellular bacterium replicates within alveolar macrophages and epithelial cells, triggering a robust release of pro-inflammatory cytokines and a massive cellular infiltration into the lung interstitium and alveoli. This results in widespread ground-glass opacities and consolidation, manifesting as the highest Total CTSS. Subsequent resolution parallels the control of bacterial replication with appropriate antibiotic therapy and the subsequent dampening of the inflammatory response, leading to the gradual absorption of exudates. The occasional development of fibrotic streaks in the convalescent phase represents aberrant tissue repair following severe alveolar injury. In cases of extensive initial damage, the normal healing process is superseded by a fibroproliferative response, which manifests as persistent linear opacities on CT. Although the overall Friedman test across the four disease stages did not reach statistical significance (P = 0.392), the descriptive trend and the fitted third-order polynomial curve provide a valuable representation of the disease’s natural history, aiding clinicians in contextualizing imaging findings throughout the disease course.

CK and DBil as stable biomarkers

Elevations in CK and DBil specifically signal extrapulmonary organ injury—myocyte damage and hepatic/cholestatic dysfunction, respectively. This pattern pathophysiologically aligns with C. psittaci’s tropism for the mononuclear phagocyte system (MPS) [29, 30]. The liver, with its abundant Kupffer cells, and skeletal muscle, with its resident macrophage network, become primary targets after hematogenous dissemination from the lung. Their infection establishes the “lung–liver–muscle” axis, directly linking bacterial spread to end-organ damage and systemic inflammation. Methodologically, despite limited sample size, both biomarkers showed numerically elevated odds ratios per unit increase (CK: aOR 1.00 per 1 U/L [1.00–1.00]; DBil: aOR 1.07 per 1 µmol/L [0.45–2.59]).

Clinical risk stratification model

It is important to recognize a key limitation regarding potential exploratory value. Because the proposed model identifies patients who already meet criteria for severe pneumonia at presentation (e.g., need for ventilation, ICU admission, septic shock), its incremental value beyond routine clinical assessment remains unclear. Clinicians can typically identify such patients based on readily apparent features such as hypoxemia, tachypnea, hemodynamic instability, or altered mental status. Therefore, whether CK and DBil measurement provides meaningful additional information for early triage decisions requires prospective comparison against clinical judgment alone.

We operationalized CK and DBil levels into a practical, four-tier risk stratification model (Low, Intermediate, High, Very High). This tool translates biomarker levels into specific clinical actions regarding monitoring frequency, care level, and treatment intensity. Using routinely available lab parameters, it may potentially assist in the identification of high-risk patients, suggesting a possible role in timely intervention and resource allocation, but this requires prospective validation. In an exploratory comparison, the proposed CK/DBil two-marker model (using thresholds CK ≥ 400 U/L or DBil ≥ 7.5 µmol/L) showed a numerically higher area under the ROC curve (AUROC, 0.76) than CURB-65 (0.68) and PSI (0.71) for identifying severe pneumonia, although confidence intervals overlapped substantially, indicating no statistically significant improvement in this small cohort (Supplementary Appendix 8).

Integrative pathophysiological model

We propose the “lung–liver–muscle” axis as a hypothesis-generating conceptual framework to guide future mechanistic studies. This integrative paradigm posits the pulmonary infection as the inciting event, with subsequent bacteremic seeding of organs enriched in phagocytic cells. The liver and skeletal muscle are selectively vulnerable due to their dense and functionally distinct macrophage populations. Hepatic infection, centered on Kupffer cells, can incite bystander hepatocellular injury or disrupt biliary transport, manifesting as elevated DBil. In skeletal muscle, the resident macrophage network, typically dedicated to tissue homeostasis and repair, is subverted by the infectious process. An initial pro-inflammatory macrophage response likely mediates myocyte damage and CK release, mirroring a sterile injury response. Consequently, the rising levels of DBil and CK serve not merely as biomarkers but as direct reflections of active extrapulmonary organ involvement. This injury, in turn, fuels a systemic inflammatory response, establishing a vicious, self-perpetuating cycle of immune dysregulation and progressive organ dysfunction that culminates in multi-organ failure. The tropism of C. psittaci for the mononuclear phagocyte system has been well described [29–31], with Kupffer cell infection documented in animal models and skeletal muscle macrophage involvement implicated in myositis pathogenesis. By framing CK and DBil as specific indicators of targeted extra-pulmonary organ damage, this model transcends a purely pneumocentric view of psittacosis and illuminates novel potential targets for future therapeutic strategies aimed at interrupting this deleterious cycle. A detailed schematic illustration of this “lung–liver–muscle” axis pathophysiological framework is provided in Supplementary Appendix 7.

It is critical to emphasize that the proposed risk stratification model is designed to identify patients who already have severe pneumonia at presentation, not to predict progression from mild to severe disease. Its potential exploratory value lies in early recognition, not risk forecasting.

Limitations and future directions

This study’s retrospective design, small sample size, and lack of a stable multivariable model limit generalizability. Specifically, the absence of multivariable adjustment means we cannot exclude confounding by age, comorbidities, coinfections, or other laboratory parameters. The observed associations between CK/DBil and severity may be confounded by unmeasured variables. The four-tier risk stratification thresholds lack empirical validation for discrimination and should be interpreted as illustrative pending confirmation in larger cohorts. DNA-based mNGS may not reliably detect RNA viruses; however, concurrent PCR testing mitigated this limitation. While Firth’s regression mitigated small-sample bias, external validation of the proposed CK and DBil cut-offs in larger, prospective cohorts is essential before clinical adoption. We recommend incorporating these biomarkers into initial patient assessment, particularly for high-risk individuals. Future work should focus on multi-center validation and model refinement.

Conclusions

This study integrates four key pillars of severe psittacosis: a characterized radiological evolution, validated CK and DBil as stable biomarkers, a practical risk stratification model, and the “lung-liver-muscle” axis pathophysiological framework. The consistent performance of CK and DBil supports further investigation of their potential role in identifying high-risk patients, but immediate clinical use is not recommended pending future validation studies in larger, prospective cohorts.

Supplementary Information

Below is the link to the electronic supplementary material.

Abbreviations

ALT

Alanine Aminotransferase

aOR

Adjusted Odds Ratio

ARDS

Acute Respiratory Distress Syndrome

CAP

Community-Acquired Pneumonia

CI

Confidence Interval

CK

Creatine Kinase

CK-MB

Creatine Kinase Myocardial Band

CT

Computed Tomography

cTnI

Cardiac Troponin I

CTSS

Computed Tomography Severity Score

CURB-65

Confusion, Urea, Respiratory rate, Blood pressure, age ≥ 65

DBil

Direct Bilirubin

FCS

Fully Conditional Specification

ICU

Intensive Care Unit

IFN-γ

Interferon Gamma

IL-6

Interleukin-6

IQR

Interquartile Range

MICE

Multivariate Imputation by Chained Equations

mNGS

Metagenomic Next-Generation Sequencing

NT-proBNP

N-terminal pro-B-type Natriuretic Peptide

PSI

Pneumonia Severity Index

SD

Standard Deviation

SOFA

Sequential Organ Failure Assessment

SpO₂

Peripheral Oxygen Saturation

Author contributions

Y.M.Huang. and Q.C.Lin. contributed equally to this work. Y.M.Huang. had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis, including and especially any adverse effects. Q.C.Lin. contributed to study design, data analysis, interpretation, and manuscript writing. Both authors approved the final version.

Funding

No external funding was received.

Data availability

The datasets used and/or analyzed during the current study are available from the corresponding author on reasonable request.

Declarations

Ethics approval and consent to participate

This study was conducted in accordance with the Declaration of Helsinki (2013 revision). This study was approved by the Institutional Review Board/Ethics Committee of Longyan First Affiliated Hospital of Fujian Medical University (Approval No: LYREC2025-k196-01) and The Second Hospital of Longyan (Approval No: LYEY-KY-2025-048). Patient consent was waived due to the retrospective nature of the study.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.