Construction and analysis of a nomogram prediction model for post-infectious bronchiolitis obliterans in children with adenovirus pneumonia after invasive mechanical ventilation

From January 2019 to December 2019, 863 children who were hospitalized in our hospital, diagnosed with ADVP, and who underwent IMV were selected as the research subjects. The children were under 3 years of age. According to the 2-year follow-up, demographic and clinical variables were compared between the PIBO and non-PIBO groups. The Declaration of Helsinki was followed throughout the duration of the study. Informed consent was obtained from the subjects and their guardians who understood and participated in these experiments. The study was approved by the Ethics Committee of Hunan Provincial People's Hospital with judgment reference number 2020–07. All of the methods were performed in accordance with the relevant guidelines and regulations.

The diagnostic criteria for ADVP include acute lower respiratory symptoms, lung infiltration, and ADV infection [10]. The exclusion criteria included patients with chronic lung disease, wheezing or a history of asthma, and incomplete case information, among other criteria. The clinical diagnostic criteria of PIBO are as follows [11]. (i) Persistent wheezing, coughing, shortness of breath, dyspnea, or exercise intolerance. (ii) Extensive wheezing and crackles that can be heard in both lungs for > 6 weeks, as well as the lungs responding poorly to bronchodilators. (iii) High-resolution computed tomography (CT) showing PIBO changes (such as a mosaic sign, bronchiectasis, and bronchial wall thickening). (iv) Lung function showing small airway obstructive or mixed ventilatory dysfunction and bronchodilation tests mostly being negative. (v) Cystic fibrosis, congenital bronchopulmonary dysplasia, and other diseases caused by cough and asthma being excluded.

Clinical information (gender, age, significant comorbidities, signs and symptoms, laboratory characteristics, associated multiorgan dysfunction, and treatment) was collected. The experimental data mainly included routine blood tests, mixed infections, and T-cell differentiation, among other data.

Specialist nurses collected the specimens in our hospital. The length of the throat swab was extended from the earlobe to the tip of the nose. The specimens were collected in normal saline and stored at -80 °C. Frozen nasopharyngeal swab samples were taken and naturally thawed at room temperature. Two hundred microliters of the sample was extracted, and ADV-DNA replication was detected by using a RT‒PCR kit (Qiagen, USA). According to the kit's instructions, the bronchoalveolar lavage fluid (BALF) sample was used for the DNA extraction. Commercial primers and probes were utilized for PCR amplification. Furthermore, ADV DNA > 1.0 × 10³ copies/mL was considered to be a positive result.

SPSS25.0 software was applied for the data analysis. Normality was analyzed via the Shapiro‒Wilk method. Normally, quantitative data are expressed as the mean ± standard deviation (x ± s). The independent sample t test was utilized for comparisons between the two groups [12]. Non-normally distributed quantitative data are shown as the median (IQR, interquartile range). The rank-sum test was used for the comparisons between the groups. In addition, a percentage (%) was utilized to analyze enumeration data, and the χ2 test was performed. Logistic regression was also performed [13]. P < 0.05 and P > 0.1 were used as independent variable inclusion and exclusion criteria, respectively, and the test level was set at 0.05 (both sides). R3.6.3 was utilized to build a nomogram risk prediction model. Moreover, diagnostic performance was evaluated by using receiver operating characteristic (ROC) curves. For the calibration, a curve was drawn. To assess the model's fit, we used the Hosmer–Lemeshow (H–L) goodness of fit test. Furthermore, a P value < 0.05 was considered to be statistically significant.

A total of 863 cases with ADVP were enrolled in the study, including 46 cases (5.3%) and 817 cases (94.7%) in the PIBO and non-PIBO groups, respectively. A statistically significant difference was found between the PIBO and non-PIBO groups, with 35 men (76.1%) and 525 men (64.3%) in each group, respectively (P = 0.1, Table 1). The median age at the time of diagnosis of PIBO was younger, but there was no significant (P = 0.55, Table 1). Moreover, the age of onset in the two groups was mainly concentrated between 6–24 months. There was a significantly longer hospital stay (P = 0.02) for patients with PIBO, as well as a higher duration of fever (P = 0.00, Table 1). Additionally, wheezing, lung consolidation, and complications were also more frequent among patients in the PIBO group (P < 0.05, Table 1). The percentage of neutrophils (N%, P = 0.000), lactate dehydrogenase (LDH, P = 0.000), D-dimer (DD, P = 0.000), ADV load (P = 0.026), alanine transaminase (ALT, P = 0.000) and aspartate aminotransferase (AST, P = 0.000) were significantly higher in the PIBO group compared to the non-PIBO group. However, platelet (PLT, P = 0.02) was the opposite of that. Except for these parameters, there were no significant differences in allergic constitution, mixed infection, leukocyte (WBC), hemoglobin (Hb), C-reactive protein (CRP), procalcitonin (PCT), fibrinogen (Fig), creatine kinase isoenzyme (CK-MB), total immunoglobulin (Ig) A, IgG, IgM, CD3 + , CD3 + CD4 + , and CD3 + CD8 + . In addition, the PIBO group was more likely to receive interventions such as IMV, corticosteroids, IV immunoglobulin, and tracheoscopy (P = 0.000, Table 1).

Due to the high proportion of PIBO in children with ADVP after IMV, we examined the causes of PIBO in cases of ADVP after IMV. There were 66 cases of ADVP after IMV, including PIBO (33 cases, 50.0%) and non-PIBO (33 cases, 50.0%) groups. There were 29 men (87.9%) and 21 men (63.6%) in the PIBO and non-PIBO groups, respectively. Moreover, the difference was statistically significant (P = 0.02, Table 2). There was no significant difference in age between the PIBO group and the non-PIBO group (P = 0.88, Table 2). Additionally, patients in the PIBO group had a longer duration of IMV treatment (P = 0.03) and duration of fever (P = 0.00) throughout the disease (Table 2). Significant increases in mixed fungi coinfection (P = 0.006) were observed in the PIBO group (Table 2). The median ADV load of the PIBO group was higher than that of the non-PIBO group (7.1 vs. 6.0, respectively, P = 0.002, Table 2). Furthermore, only one child with PIBO (8 months) died due to recurrent infection, while no child in the non-PIBO group died. Figure 1 shows the bronchoscopic and imaging features of a child with ADVP after IMV who developed PIBO 2 years later.

Univariate logistic regression revealed that duration of fever (odds ratio [OR]: 1.088; 95% CI: 1.031–1.148), IMV (OR: 7.597; 95% CI: 2.900–19.901), complications (OR: 3.921; 95% CI: 1.539–9.989), and N% (OR: 1.045; 95% CI: 1.016–1.074) were independent impact indicators for PIBO in pediatric patients with ADVP (P < 0.05, Table 3).

To examine the formation of PIBO in ADVP cases with IMV, we performed an impact factor analysis for PIBO. Male sex (OR: 0.186, 95% CI: 0.036–0.973), duration of fever (OR: 1.151; 95% CI: 1.017–1.302), ADV load (OR: 1.842; 95% CI: 1.129–3.006), and fungi coinfection (OR: 5.616; 95% CI: 1.369–23.041) were independent impact factors for the formation of PIBO (P < 0.05, Table 4).

As shown in Fig. 2A, based on the abovementioned four impact factors (gender, duration of fever, ADV load, and fungi coinfection), the PIBO formation nomogram prediction model of children with ADVP under IMV was established. The ROC curve for the risk of PIBO occurrence was used to assess the discrimination of the nomogram model. The calculated AUC by using the ROC model to predict impact factors for PIBO in children with ADVP under IMV was 0.857 (95% CI: 0.744–0.928), thus suggesting good discrimination from the nomogram model (Fig. 2B). Subsequently, a calibration curve was drawn, and the H–L goodness-of-fit test was performed to assess the accuracy of the nomogram model. The calibration curve was a straight line with a slope close to 1 (Fig. 2C), thus indicating that the model predicted that the risk of PIBO was in good agreement with the actual risk. The H–L goodness-of-fit was 68.75 (P < 0.01), thus indicating good calibration for the model.