The STROBE (Strengthening the Reporting of Observational Studies in Epidemiology) guidelines for reporting observational cohort studies were followed when reporting this study (22).

This study used the Surviving Pneumonia research platform. Surviving pneumonia is an observational, prospective cohort study including hospitalized patients with CAP at Copenhagen University Hospital - North Zealand, Denmark. Patients were enrolled at the Emergency Department and medical wards between January 2019 and April 2021 and followed for 90 days after admission. The exposure was adiponectin measured at baseline.

In Surviving Pneumonia, all patients hospitalized with a suspected lower respiratory tract infection were screened for eligibility. The eligibility criteria were age ≥ 18 years, new infiltrate on chest x-ray or chest computed tomography scan and at least one symptom or clinical sign consistent with pneumonia, e.g., cough, sputum production, fever (≥38.0°C), hypothermia (<35.0°C), chest pain, breathlessness, and abnormal chest auscultation. The exclusion criterion for the present study was a positive polymerase chain reaction test for severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2). Patients were enrolled within 24 h of admission.

The study was approved by the Scientific Ethics Committee at the Capital Region of Denmark (H-18024256), registered on ClinicalTrials.gov (NCT03795662), and conducted in accordance with the Declaration of Helsinki (23). Oral and written informed consent was obtained from all patients before enrolment.

Data were collected prospectively with standardized forms and entered into a REDCap database by a few experienced nurses to reduce information bias (24). Information about demography, past medical history, comorbidities, radiology reports, vital signs at admission, blood test results, microbiological test results, and clinical outcomes were collected during a structured interview at study enrolment and from the electronic medical record during the follow-up period. The comorbidity burden was assessed with the Charlson comorbidity index (CCI) (2). Low, medium and high CCI was defined as an index score of 0, 1–2, and ≥3, respectively (2). CAP severity was assessed with the CURB-65 score based on level of confusion, respiratory rate, systolic or diastolic blood pressure and age (25). Clinical stability was defined according to the modified Halm’s criteria when temperature ≤ 37.2°C, heart rate ≤ 100 beats/min, respiratory rate ≤ 24 breaths/min, systolic blood pressure ≥ 90 mm Hg and oxygen saturation ≥ 90% without supplemental oxygen therapy (or habitual oxygen) and quantified in hours since admission (26).

Anthropometric measurements and body composition estimation were performed 24 h after study enrolment. BMI was calculated from self-reported height in meters and weight measured to the nearest 0.1 kg on an electric scale (Seca, Hamburg, Germany). Waist circumference was measured midway between the lowest portion of the rib cage and the uppermost portion of the iliac crest at the umbilical level. Underweight, normal weight, overweight and obesity were defined as BMI < 18.5, 18.5–24.9, 25.0–29.9 and >30 kg/m², respectively. Body fat percentage (%) was estimated with bioimpedance analysis (BioScan touch i8 STD, Maltron International, United Kingdom) using multi-frequency (5, 50, 100, and 200 kHz) hand-to-foot measurements of resistance and reactance with the patient in a supine position.

Blood samples were collected in serum tubes and centrifuged at 3,000 x g for 15 min at 4°C. The supernatant was stored at −80°C. Serum adiponectin was measured with a commercially available immunoassay (R-PLEX^® Human Adiponectin Antibody Set, Meso Scale Diagnostics, Rockville, Maryland, United States) on an electrochemiluminescent platform (Meso Scale Diagnostics, Rockville, Maryland, United States) according to the manufacturer’s instructions. All samples were run in duplicate. The inter-and intra-assay coefficient of variation was <8%. Standard blood tests such as c-reactive protein (CRP) and plasma glucose were performed daily at the attending physician’s discretion.

The primary outcome was 90-day mortality. The secondary outcomes were associations of adiponectin with baseline characteristics, in-hospital mortality, the need for non-invasive respiratory support (high flow oxygen therapy, continuous positive airway pressure therapy), intensive care unit (ICU) transferral, time to clinical stability, length of stay (LOS), and 90-day readmission. Age, sex, and adiposity measures were included as confounders in all analyses, with adiponectin as a predictor according to the directed acyclic graphs of the expected causal relationships (Supplementary Figure 1).

The continuous variables were skewed and reported as medians [interquartile range (IQR)]. Categorical variables were reported as counts and percentages. As BMI influences adiponectin levels (27), baseline characteristics of patients with and without a BMI measurement were compared. Comparisons of 2 independent samples were performed with Mann–Whitney U tests. Kruskal–Walli’s test was used for more than two independent samples, followed by Dunns’s post hoc test for multiple comparisons. Differences in categorical variables were estimated with the chi-squared test.

The study participants were divided into four groups based on the sex-stratified quartile distribution of baseline adiponectin because adiponectin levels are higher in females than males (9). Cumulative mortality during the follow-up period across the adiponectin quartiles was assessed with Kaplan–Meier curves and log-rank tests.

Associations between baseline characteristics and adiponectin were modeled with univariate and multivariate linear regression adjusted for age, sex, BMI, select comorbidities (diabetes mellitus, COPD, ischemic heart disease, heart failure, chronic kidney disease, cancer, rheumatic disease), CURB-65 score, and admission CRP level.

The association of continuous outcome variables with adiponectin was estimated with linear regression. Continuous outcome variables, including adiponectin, were skewed and log-transformed to satisfy linear regression model assumptions. The reported exponentiated β-coefficients with 95% CI represent the fold change. Logistic regression was used to estimate the association of categorical outcome variables with adiponectin and reported as odds ratios (OR) with 95% CI. The regression models for all outcomes except where adiponectin was the outcome variable were adjusted for age and sex (minimally adjusted models), followed by subsequent adjustments for BMI (fully adjusted models). Adiponectin was evaluated as a continuous variable and quartiles in these models.

Patients who died in-hospital were excluded from analyses where the outcome variable was 90-day readmission, LOS, or time clinical stability. In addition, patients with a do-not-resuscitate order or limitations on life-sustaining treatment were excluded from analyses where ICU admission was the outcome variable.

Complete case analysis was performed in case of missing independent variables in the regression analyses. Sensitivity analyses were performed by entering waist circumference or body fat % in the regression models instead of BMI as more precise measures of adiposity (8). BMI was the primary measure of adiposity because more patients had an available BMI measurement than waist circumference or fat percentage. The statistical comparisons were 2-sided, with p < 0.05 as the significance level. The statistical analyses were conducted with R software (R version 4.0.3).

Among the 3,100 patients screened for eligibility, 502 fulfilled the inclusion criteria and were included (Figure 1). Patient characteristics are shown in Table 1. The median age was 74 years (IQR, 65–81), 50% were female, 6.2% were underweight, 58% were overweight or obese, and 80% had a CCI score ≥ 3. Furthermore, 38, 20, and 19% of the patients had been diagnosed with COPD, cancer, or diabetes, respectively. Patients with no BMI measurement (20%) were older and had higher CCI scores, disease severity, and mortality rate than those with a BMI (Table 2). Adiponectin levels were similar in patients with and without a BMI measurement [median 19 (IQR, 13–27) μg/mL vs. 22 (IQR, 13–30) μg/mL, p = 0.2].

Patients presenting with moderate or severe disease, according to the CURB-65 score, had higher adiponectin levels than patients with mild disease (Figure 2). In multivariate linear regression analyses, increasing age, female sex, having COPD, and rheumatic disease were associated with higher adiponectin levels, whereas increasing BMI, having diabetes mellitus, and higher admission CRP were associated with lower adiponectin levels (Table 3).

Adiponectin levels were higher in non-survivors compared to survivors (Figure 3). The survival curves showed a decreasing survival probability with increasing adiponectin quartiles (p < 0.003) (Figure 4). The 90-day survival probability was 94% (95% CI: 91–98%) for patients with adiponectin levels in the 1st quartile, 87% (95% CI: 82–94%) for those in the 2nd quartile, 80% (95% CI: 73–88%) for the 3rd quartile, and 76% (95% CI: 69–84%) for the 4th quartile (Figure 3). Pairwise log-rank tests showed that the survival probability for patients with adiponectin levels in the 4th quartile was significantly lower than for those in the 1st (p < 0.001) and 2nd quartiles (p = 0.048), and for patients in the 3rd quartile compared to the 1st (p = 0.001).

In minimally adjusted models, the odds ratio (OR) for 90-day mortality increased by 2% for each 1 μg/mL increase in adiponectin and was 199 and 234% higher for adiponectin levels in the 3rd and 4th quartiles, respectively, than in the 1st quartile (Table 4). When BMI was added to the analysis (the fully adjusted model), the association between adiponectin and 90-day mortality disappeared (Table 4). Likewise, the association disappeared when adjusted for waist circumference (n = 296) or body fat % (n = 311) (Table 5).

Overall, 6.4% (n = 32) of the patients died in hospital. Their adiponectin levels were higher than those of patients who survived until discharge (Figure 5). The OR for in-hospital mortality was 482 and 365% higher for adiponectin levels in the 3rd and 4th quartiles, respectively, than in the 1st quartile in minimally adjusted models (Table 4). In the fully adjusted model, the association between adiponectin and in-hospital mortality disappeared (Table 4).

There were no associations between adiponectin levels and risk of ICU transferral, non-invasive respiratory support (Table 6), LOS or time to clinical stability in the minimally or fully adjusted models (Table 7).

33.5% (n = 168) of the patients were readmitted within 90 days. Their adiponectin levels were higher than those of patients who were not readmitted (Figure 3). Exacerbation of chronic diseases and infections were the most common causes of readmission (Table 8). The OR for 90-day readmission increased by 2% for each μg/mL increase in adiponectin and was 87 and 133% higher for adiponectin levels in the 3rd and 4th quartiles, respectively, than the 1st quartile in minimally adjusted models. This association remained in the fully adjusted model when adiponectin was entered as a continuous variable and for adiponectin levels in the 4th quartile (Table 4). The results were similar when BMI was substituted for waist circumference or body fat % (Table 9).

The prospective design, standardized data collection and detailed characterization of body composition are strengths of our study. Furthermore, we consider our cohort representative of patients hospitalized with CAP in a high-income, European country since we only had one exclusion criterion.

Nevertheless, our single-center study design limits the generalisability of the results. In addition, we cannot rule out consent bias since the study participants had to give active consent, thus excluding incapacitated patients. Another study limitation was the lack of a BMI measurement in 20% of patients. The patients without a BMI measurement had more severe CAP and thus could not fully participate in the study. Even though patients without a BMI had a higher mortality rate than those with a BMI, we do not expect this to have affected the overall associations because adiponectin levels were similar in both groups but increased the chances of type II error.