Diagnosis of Bacterial Tracheostomy-Associated Respiratory Tract Infections in Pediatric Patients Free

We performed a retrospective cohort study of patients aged <18 years who are tracheostomy-dependent at a single, 155-bed academic children’s hospital with 35 000 pediatric emergency department (ED) visits per year in the Southeast United States. At our institution, these children are primarily cared for by the pediatric hospital medicine or PICU physicians in the inpatient setting and by general pediatricians, complex care pediatricians, pediatric pulmonology, and pediatric otolaryngology in the outpatient setting. Our institution does not have any existing pathways or clinical guidelines employed in caring for patients with tracheostomy dependence. We reviewed all encounters in calendar year 2018 related to fever and/or respiratory illness; children with at least 1 encounter were included. The local institutional review board approved this project.

We used a multistep approach to ensure accurate identification of our cohort. First, we identified all children with tracheostomy in the hospital electronic medical record using International Classification of Diseases, Ninth and 10th Edition, codes consistent with tracheostomy presence and care (Z43.0, Z93.0, J95.0, J95.03, Z98.890). We then reviewed patient lists maintained by specialty services that follow our institution’s cohort of tracheostomy patients (pediatric pulmonology and a complex care service) and reconciled any discrepancies. All children with a tracheostomy during calendar year 2018 identified by either of these methods (145 in total) were eligible for inclusion, including those who received a new tracheostomy or underwent decannulation during this period. Demographic and clinical data were collected from children in the identified cohort with no encounters during the study period for comparison with our final cohort.

For eligible patients, data were collected for all encounters (inpatient, outpatient, and ED) between January 1, 2018, and December 31, 2018, for fever and/or respiratory illness, which were identified by review of clinical documentation and identification of those encounters reporting fever >38°C or respiratory symptoms such as cough or increased work of breathing as a presenting symptom. Encounters at any affiliated hospitals (all occurred in adult EDs) and clinics within our electronic medical record were included. Telephone encounters documented within the electronic health record were also included and were primarily conducted by ED physicians and pediatric pulmonologists. The primary outcome for this study was a diagnosis of bTARTI, which was defined as a documented diagnosis of pneumonia or tracheitis and a prescription for a full course of antibiotic treatment by the physician responsible for the ultimate disposition of the patient. Instances where an initial diagnosis of bacterial infection was later changed to viral infection were excluded from this definition. We obtained data by chart review on demographic and clinical characteristics, including patient and family characteristics, the location of the encounter, and diagnostic and therapeutic interventions. A chest radiograph was considered to be consistent with bacterial pneumonia if the treating team’s documentation indicated this and the patient received treatment with antibiotics. Medical complexity was determined as the number of complex chronic conditions using the classification system outlined by Feudtner et al.¹⁶  Clinical outcomes obtained included LOS and 7- and 30-day readmission rates. Data were extracted by manual chart review into the Research Electronic Data Capture database.^17,18 

We used descriptive statistics to characterize demographic and clinical characteristics of patients with and without eligible encounters during the year, as well as those of encounters with and without a diagnosis of bTARTI. Patient clinical and laboratory findings were tested for association with a diagnosis of bTARTI using bivariate mixed-effects logistic regression to account for repeated encounters within the same patient.^19,20  The models included a fixed effect for the respective demographic or clinical variable and a random effect for patient. Variables found to be potentially significant in the bivariate analyses (with P value <.2) were incorporated as fixed effects into a multivariable mixed-effects logistic regression model for the primary outcome, which was diagnosis of bTARTI, with a random effect again added to account for repeated encounters within the same patient. A P value <.05 was considered to be statistically significant in the final model. Our secondary analyses aimed to quantify the relationships between bTARTI diagnosis and LOS, 7-day readmission, and 30-day readmission. We employed generalized linear mixed-effects models with fixed effects for bTARTI and the potential confounders of ventilator use and number of chronic conditions, and a random effect for participant. A log transformation was applied to LOS (because it was right-skewed) before modeling using a linear mixed-effects model. Seven-day and 30-day readmissions were modeled as binary outcomes using mixed-effects logistic regression. All statistical analyses were performed using R version 4.0.2.²¹ 

We identified 145 children with tracheostomy-dependence followed by our institution during the 2018 calendar year; of those, 79 children had at least 1 encounter during the study period. A comparison of the demographic and clinical characteristics of children included in our cohort and those with no encounters in 2018 is shown in Supplemental Table 5. Children with 0 encounters were more likely to have missing data; no substantial differences among the groups were noted otherwise.

The 79 children in the cohort had a total of 208 eligible encounters, of which 66 (31.7%) resulted in a diagnosis of bTARTI. Findings on history, physical exam, and diagnostic evaluation are compared in Tables 1 and 2. The majority (63 of 79, 79.7%) had tracheostomy sites established before 2018. No significant differences were seen in baseline characteristics of the patients, including age, sex, payer, use of home ventilation, or number of complex chronic conditions. Encounters resulting in a diagnosis of bTARTI more often had fever documented (62.1% vs 49.3%; P = .04), a chest radiograph (CXR) interpreted by the treating physician as bacterial pneumonia (39.4% vs 4.9%; P < .01), and a higher WBC count (16.4 vs 13.1 × 10³/µ; P = .03). We found no difference in the number of diagnostic tests ordered during encounters resulting in a diagnosis of bTARTI versus those without a bTARTI diagnosis, but did find that encounters with diagnosis of bTARTI had more total abnormal studies documented (1.5 vs 1; P < .01).

For our primary outcome of bTARTI diagnosis, we performed multivariable mixed-effects logistic regression modeling (Table 3). A diagnosis of bTARTI was significantly associated with a CXR interpreted as bacterial pneumonia (OR 1.77, 95% CI 1.50–2.08), a positive tracheal aspirate culture result (odds ratio [OR], 1.30; 95% confidence interval [CI], 1.05–1.61), a change in oxygen requirement from baseline (OR, 1.14; 95% CI, 1.00–1.31), higher WBC count (OR, 1.02; 95% CI, 1.00–1.03; P < .03), a patient living at home with their family (OR, 1.42; 95% CI, 1.06–1.92; versus another living situation), and an encounter performed via telephone (OR, 1.41; 95% CI, 1.09–1.81; versus ED or urgent care visit). No significant associations were seen between diagnosis of bTARTI and fever or change in ventilator settings.

Of all encounters reviewed, 76 of 208 (37%) were inpatient. In generalized linear mixed models adjusted for ventilator use and number of chronic conditions, admitted patients with a diagnosis of bTARTI experienced an LOS 2.19 times that of patients who did not receive a diagnosis of bTARTI (CI, 1.23–3.88) (Table 4). This finding is consistent with the median LOS in those with bTARTI (117.61 hours) being 2.46 times as large as the median LOS in those without bTARTI diagnosis (47.73 hours). There was no association between diagnosis of bTARTI and 7- or 30-day readmission rates.

To our knowledge, this is the first study to characterize the demographic and clinical findings associated with a physician diagnosis of bTARTI. In pediatric patients who are tracheostomy-dependent, several clinical variables were significantly associated with a final diagnosis of bTARTI. Unsurprisingly, the treating physician’s interpretation of the CXR as being consistent with bacterial pneumonia was associated with that diagnosis, as was positive tracheal aspirate culture, despite evidence that positive culture is a poor predictor of clinical tracheitis.^9,10  Higher WBC count was also associated with diagnosis of bTARTI, along with a higher number of total abnormal studies, indicating that physicians likely use these results in conjunction with each other to arrive at a diagnosis, rather than any factor in isolation.

Increased oxygen requirement was also associated with bTARTI diagnosis, although change in ventilator settings was not. Presence of fever was positively associated with diagnosis of bTARTI but did not reach clinical significance in multivariate modeling. Thus, clinical indicators that typically reflect worsening disease severity (such as increased oxygen requirement, ventilator setting changes, and fever), were not uniformly found to be associated with the diagnosis of bTARTI diagnosis in this study, likely reflecting a lack of standardized diagnostic criteria and practice variability. Physicians may have extrapolated previously proposed criteria for tracheitis, community-acquired pneumonia, and/or ventilator-associated pneumonia to arrive at a diagnosis of bTARTI.^11,22–25  However, previously described definitions of tracheitis are conflicting, and criteria for diagnosis of community-acquired pneumonia and ventilator-associated pneumonia are not specific to this unique population.

Telephone encounters were associated with bTARTI diagnosis. In our cohort, a small number of these telephone encounters occurred for patients discharged from the ED after a visit for fever or respiratory illness and whose tracheal aspirate cultures return positive for bacterial growth. A larger number of these encounters were for children whose caregivers called into a clinic when their child became ill. Families of children with medical complexity, especially those who are chronically ventilated, may call their clinician when they become ill, rather than bring their child to the clinic or hospital, because of the large transportation burden on these families.²⁶  This is an especially common occurrence in a pulmonology clinic, and often necessitates a diagnosis made by the clinician on the basis of the caregiver’s report of symptoms rather than on their own physical exam or other objective data. This phenomenon may contribute to increased diagnosis and treatment of bTARTI in the setting of telephone encounters.

Regarding demographic variables, living at home with family (as opposed to in a care facility, foster care, etc) was associated with increased likelihood of diagnosis of bTARTI. Previously, the living situation of patients with tracheostomy was demonstrated in a large retrospective database study to be associated with readmission risk, with those who were discharged from the hospital found to be at increased readmission risk because of bacterial respiratory tract infection when compared with patients discharged to a care facility.²⁷  The ready availability of health services in a care facility may account for this risk difference. Rebnord et al demonstrated in a Norwegian population that antibiotic prescriptions in the setting of viral illness are positively associated with the parents’ perception that the child has a bacterial illness.²⁸  We speculate that family members were more likely than nonfamily members to advocate for antibiotic prescriptions in the setting of illness. Interestingly, neither the number of complex chronic conditions nor home use of ventilation was associated with bTARTI diagnosis. This finding is in contrast to data published by Watters et al in 2016, who found that adverse event rates in children with tracheostomy were higher in children with complex chronic conditions.²  We postulate that the diagnostic criteria for bTARTI is sufficiently vague that patient characteristics, such as the presence of complex chronic conditions, or home ventilator use were not associated with bTARTI in our study. Additionally, our study did not characterize chronic conditions individually, and Tan et al have previously described an increased risk of respiratory infections in those patients with cerebral palsy or gastroesophageal reflux disease.²⁹ 

Notably, many laboratory tests frequently employed in evaluating for bTARTI were not associated with a diagnosis of bTARTI in our study. Specifically, neither blood culture nor respiratory viral panels appeared to reliably aid clinicians in making this diagnosis. Inflammatory markers were infrequently obtained and were also not associated with bTARTI. These findings call into question the utility of extensive laboratory testing in evaluating children with tracheostomy dependence presenting with fever and/or respiratory illness, and presents an opportunity for decreased utilization of laboratory testing. Tracheoscopy, arguably the gold standard for the diagnosis of bacterial tracheitis, was used in only 1 patient over the study period. In patients hospitalized for their illness, those with a diagnosis of bTARTI had an LOS over twice that of those without a diagnosis of bTARTI. A standardized approach to diagnosis could potentially reduce LOS by promoting accurate diagnosis of bTARTI.

A notable strength of this study was review of the use of telephone notes (rather than in-person evaluation or reevaluation) to help understand the eventual diagnosis of bTARTI when relying on symptom report rather than full provider assessment. This information is especially pertinent during the time of increasing use of telehealth during the coronavirus disease 2019 pandemic.

A significant limitation of this study is its single-center design. Our findings are likely affected by local practice patterns and thus may not be fully generalizable. Two affiliated community hospital sites were included, however, which helps partially mitigate this limitation. Additionally, we reviewed only clinic visits within our hospital system, and therefore did not capture the practice patterns of primary care physicians outside our system. Team members vary from hospital to clinic settings. Residents are often involved in patient encounters at the hospital, whereas residents may or may not be involved in the clinic setting, which may affect practice patterns, as well.

We included telephone encounters as a single-visit type, although the nature of those encounters varied within our cohort (eg, clinician calling with culture results, patient or family seeking care for sick symptoms, etc). Children with both new and established tracheotomy sites were included, although clinical decision-making may vary among these populations. Another limitation is the size of our cohort, which is relatively small as compared with similar studies using the Pediatric Health Information System database.³⁰  This may have limited our statistical power to assess associations with some clinical and demographic variables. The physicians reviewing the charts were not directly involved in each case and could not confirm a diagnosis of bTARTI, relying instead on the treating physician’s assessment, which may be variable given the lack of a clear definition. Thus, though our data contribute to an understanding of how clinicians make a diagnosis of bTARTI in practice, they do not address the accuracy of the final diagnosis. Future directions should include multicenter and prospective studies which contribute to development of a predictive model for diagnosis of bTARTI and ultimately to guideline development.

In our single-center study, we identified several clinical and nonclinical factors associated with a diagnosis of bTARTI. Despite widespread use, most laboratory tests were not associated with a diagnosis of bTARTI, suggesting that clinicians should use discretion when deciding whether to use laboratory testing to investigate for bTARTI. This study identifies a need for improved standardization of bTARTI diagnosis, allowing for decreased overutilization of diagnostic testing and improvement in the accuracy of diagnosis of bTARTI. More accurate diagnosis of bTARTI also has potential to improve antibiotic overuse and LOS.