Risk Assessment for Postoperative Pneumonia in Children Living With Neurologic Impairments

We used the Participant User Files for the National Surgical Quality Improvement Program-Pediatric (NSQIP-P) for the years between 2012 and 2018. Briefly, the National Surgical Quality Improvement Program (NSQIP) is an ongoing data-driven, participatory, quality improvement initiative to produce risk-adjusted outcomes of major surgical procedures from >100 US participating hospitals.⁸  Trained surgical clinical reviewers abstract patient-level information from electronic health records and medical charts pertaining to preoperative and 30-day postoperative data by abiding to standardized definitions. Data are collected by using a systematic sampling protocol to reduce selection bias, thereby capturing the first 35 consecutive surgical cases meeting inclusion criteria within an 8-day cycle. The NSQIP Participant User Files is one of the most reliable and complete surgical databases, with an interrater reliability audit and overall disagreement rate of 2% among participating hospitals.⁹  Further details about the NSQIP have been published elsewhere.⁹  We included in our analytical sample all children (<18 years) who underwent inpatient surgical procedures between 2012 and 2018. This study was approved by the institutional review board of our institution.

Our primary outcome was the time to incident pneumonia within the 30 days after surgery. Pneumonia diagnosis was based on a standardized definition that requires patients to meet 2 criteria, radiologic and clinical (or laboratory),¹⁰  as defined by the NSQIP and adopted in previous studies.^11–14  We included in the Supplemental Information the fully elaborated NSQIP definition of pneumonia. Our secondary outcome was the length of postoperative hospital stay within the 30 days after surgery.

Neurologic diagnoses were elicited from the medical records of patients and were grouped into 4 binary variables, including structural central nervous system abnormalities, cerebral palsy, seizure disorder, and neuromuscular disorder. (1) Structural central nervous system abnormality was defined in the study database as the presence of structural anomaly of the central nervous system, including, myelomeningocele (spina bifida), microcephaly, macrocephaly, hydrocephalus, hypertelorism, trigonocephaly, Dandy-Walker malformation, Arnold-Chiari malformation, syrinx, tethered cord, gray or white matter changes, absent corpus callosum, fused ventricles, aqueduct stenosis, other neural tube defects, or other cystic or degenerative lesions of the central nervous system. (2) Patients were reported as having a neuromuscular disorder if they had a congenital or acquired degenerative neuromuscular disorder that result in a slow, progressive deterioration in motor function. Examples of diagnosis that fit the definition of neuromuscular disorder included degenerative disorders of gray and white matter (dystrophies), demyelinating disorders, and peripheral neuromuscular disorders and neuromuscular scoliosis. (3) Seizure disorder was defined as chronic seizure symptoms that require medical and/or dietary management; the definition does not include febrile seizures. (4) The diagnosis of cerebral palsy was elicited in the medical records of patients with documented motor and/or cognitive deficits due to known or unknown etiology.

Categorical variables were presented as frequency and column percentage. Noncontinuous variables were summarized as median (interquartile range [IQR]). We used Cox proportional-hazards models to estimate the hazard ratio (HR) and 95% confidence intervals (CIs), as a measure of the magnitude of the association between neurologic diagnoses and postoperative pneumonia. The Cox proportional hazard model would account for the possibility that not all patients included in the data set developed pneumonia during the follow-up period (right censoring). Our choice of covariate adjustment was based on the intent to reduce the impact of confounding and produce risk-adjusted estimates that reflect the impact of neurologic impairments and not the presence of comorbid conditions associated with them.

Consequently, the following covariates were included in our model: age, sex, preoperative ventilator dependency, history of prematurity, congenital malformation, hematologic disorder, malignancy, cardiac risk factor, gastrointestinal disease, and chronic lung disease. We also adjusted the analyses for surgical profile: emergency case status, previous cardiac surgery, year of operation, surgical specialty, and surgical complexity index. The attending anesthesiologist or surgeon reports information on urgent or emergent status. Urgent surgeries were usually performed within 24 hours of surgical evaluation, and emergent cases were usually performed within 12 hours of surgical intervention.¹⁰  The surgical complexity index was developed from a principal component analysis of operating time and work relative value units. The index was derived from the first component, which accounted for 75.4% of the variability in the sample and had an eigenvalue of 1.51. We included race in the models to account for adjust for potential patient selection due a racial disparity in the prevalence of neurologic comorbidities.¹⁵  Previous studies have also shown a higher incidence postoperative pulmonary complications, including pneumonia, in children of Black race.^16,17  We did not include Hispanic ethnicity status in the models because we did not find empirical evidence suggesting ethnic variation in both the prevalence of neurologic comorbidities and the incidence of postoperative pneumonia.

For the analysis of hospital length of stay, we used the Poisson regression with robust standard errors for the parameter estimates. To disentangle the effect of pneumonia from the effects of neurologic comorbidities, we divided the study population into 4 mutually exclusive groups: (1) children without pneumonia or neurologic comorbidity, (2) children with neurologic comorbidity only, (3) children with pneumonia only, and (4) children with both pneumonia and neurologic comorbidity.

We estimated the E value to evaluate the robustness of our findings to the potential impact of unmeasured confounding. Briefly, the E value is a measure of the magnitude of the association between an unmeasured confounder and both the exposure and the outcomes variables required to reduce the confidence limits to 1.0 (ie, explain away the observed associations). We performed all data analyses and management using Stata, version 16 (Stata Corp, College Station, TX).

Table 1 show the descriptive statistics of the study cohort, which includes 349 163 children who underwent surgical procedures during the study period. The median age at the time of surgery was 7 years (IQR: 1–13). Most children were of white race (67.1%), and more than half were male (54.5%). Thirty-four percent of children had an emergent or urgent surgery, and general surgery was the most represented surgical specialty, accounting for 47.7% of the study cohort. Children with any neurologic comorbidities were more likely to have a history of prematurity, congenital malformation, hematologic disorder, childhood malignancy, cardiac risk factor, gastrointestinal disease, and/or previous cardiac surgery.

Figure 1 shows the findings from the Cox proportional-hazards models to estimate the magnitude of association between neurologic diagnoses and pneumonia. Of the 349 163 children in the study cohort, 2191 developed pneumonia within the 30 days after surgery (Cumulative incidence: 0.6%). The presence of a preoperative neurologic comorbidity was associated with a twofold higher risk of postoperative pneumonia (adjusted HR: 1.91, 95% CI: 1.73–2.11). We further explored the association of the different components of the composite variable (neurologic disorder) with postoperative pneumonia. Children with cerebral palsy were at a markedly higher risk compared with their peers without any neurologic comorbidity for postoperative pneumonia (adjusted HR: 3.92, 95% CI: 3.38–4.56). The risk of postoperative pneumonia was 2.63 times higher among children with a neuromuscular disorder (adjusted HR: 2.63, 95% CI: 2.32–2.99). Finally, seizure disorder was associated with an almost threefold increase in pneumonia risk after surgery (adjusted HR: 2.93, 95% CI: 2.60–3.30).

The presence of a preoperative neurologic comorbidity was associated with a longer length of hospital stay (median [IQR]: 3 [2–6] vs 2 [1–5], P < .001). The adjusted-incidence rate ratio comparing children with and without neurologic comorbidity was 1.26 (95% CI: 1.25–1.28). In addition, the occurrence of postoperative pneumonia was associated with markedly prolonged hospital length of stay among all children, although the effect was most pronounced in children with preoperative neurologic comorbidity (Fig 2).

The E value analysis showed that our findings were robust to unmeasured confounding. Ranging from 2.85 to 6.22, the E values implied that an unmeasured confounder must be associated with both neurologic diagnoses and postoperative pneumonia by at least an HR of 2.85-fold to reduce the confidence limits to 1.0 (ie, explain away the observed associations).

We sought to empirically evaluate the excess risk of postoperative pneumonia in a large cohort of children with neurologic disorders who underwent a wide variety of inpatient pediatric surgical procedures. We found that the risk of postoperative pneumonia was markedly increased in children with preoperative neurologic disorders relative to their counterparts without preoperative neurologic comorbidities. We further found that the presence of a preoperative neurologic comorbidity combined with the development of postoperative pneumonia was associated with a significantly longer postsurgical hospital length of stay. Our work builds on previous findings indicating that children with cerebral palsy who underwent orthopedic procedures were more likely to develop postoperative pulmonary complications.¹⁸  This study included a relatively small sample of children with cerebral palsy and did not report on other neurologic disorders or other surgical procedures.

Altogether, our observations have major clinical importance for several reasons. First, patients who develop postoperative pneumonia face the dual challenge of rehabilitation from surgery and recovery from pneumonia and its complications. Given that children with preoperative neurologic impairment are especially vulnerable, it is important to identify areas of research and interventions to reduce the burden of postoperative pneumonia in this group of children. Second, the number of children living with neurologic impairment continues to increase, with current estimates suggesting that care for these children is responsible for approximately one-third of all pediatric inpatient charges.²  To address the surgical need of this emerging population of neurologically impaired patients, it is essential to develop targeted and coordinated perioperative care that will specifically include postoperative pneumonia prevention strategies. More research is also needed to establish clinical and biological profiling of children with neurologic impairment to preoperatively identify the subset of patients with a disproportionate risk of pneumonia. Although researchers in several studies have developed multivariable models to predict the occurrence of postoperative pneumonia,^19–23  no study has expanded the prediction to the subset of children who are living with neurologic impairment. Such empirical data would help clinicians on which preoperative factors to look for in children presenting with neurologic impairment, not only to reduce their risk but also to improve the rescue of those who develop postoperative pneumonia. Third, our findings are of clinical importance because they represent empirical data to inform the preoperative evaluation of children with neurologic disorders and the counseling of their parents. The magnitude of the observed associations also underlines the need to reevaluate the perioperative management of children with neurologic impairments to reduce their burden of postoperative pneumonia.

The pathogenesis of pneumonia inherent to neurologic comorbidities is not well addressed in the surgical literature, and there is a lack of consensus on the perioperative management of affected patients. In the medical literature, several studies with findings consistent with ours have evaluated the role of neurologic impairments in the excess risk of pneumonia. Children with neurologic impairments have a disturbed oropharyngeal motor system, resulting in recurrent aspiration of solid and liquid materials.^3,24  Aspirations tend to be insidious in children²⁴  and may occur during and between feeds, with thin liquids having the greatest predilection of being aspirated.²⁵  Oropharyngeal secretions (usually contaminated) may slowly track down the tracheobronchial tree because of inadequate cough and clearance mechanisms in children with neurologic impairment.³  Furthermore, a common feature of neuromuscular disorders is respiratory muscle weakness, which may have several consequences, including respiratory failure and upper respiratory obstruction.³  Additionally, there is a tendency toward a weaker immune system in neurologically complex patients or an enhanced effect of anesthetic mediated immune system downregulation. Alternatively, patients with neurologic comorbidities may use medications (eg antiepileptic, antispasmodic agents) that may have synergistic effects with anesthetic agents leading to increased sedation or delayed awakening. This altered sensorium could put patients at risk for inability to clear secretions, potentially leading to overt or silent aspiration.

Given the preceding biological mechanisms, it may be argued that the association of comorbid neurologic conditions with pneumonia is intuitive. We contend, however, that the magnitude of the observed associations rather than their existence is a critical factor to consider. Concerted efforts should be directed at potential perioperative interventions that would narrow the excess risk of postoperative pneumonia in children with preoperative neurologic disorders. Such risk mitigation efforts can target the sterility of aspirated materials through the reduction of oral bacteria load.²⁶  Perioperative teeth brushing and the use of chlorhexidine are typical approaches to prevent the colonization of oropharyngeal secretions, although findings are inconclusive regarding their efficacy. Preoperative dental brushing performed by pediatric dentists has been shown to reduce the risk of postoperative pneumonia.²⁷  In another setting, it was demonstrated that the combination of tooth brushing and chlorhexidine had a greater impact on the risk of pneumonia, compared with the use of chlorhexidine without brushing.²⁸  Researchers in future interventional studies could evaluate the differential impact of various perioperative oral care modalities on the risk of pneumonia in children with neurologic impairments.

Our study was limited by several factors, mainly inherent to the characteristics of the study database. Some of the categories in the NSQIP-P data set that fall under the heading of neurologic comorbidities were excluded, including stroke or traumatic or acquired brain injury with resulting neurologic deficit, tumor involving central nervous system, and coma >24 hours. The spectrum of the included neurologic comorbidities is limited to only those disorders described in the NSQIP-P data set. It is also crucial to recognize that overlap exists between the comorbidities: that is, some patients had >1 of the studied neurologic diagnoses (a patient with cerebral palsy may also have a history of seizures). Similarly, the adjusted HR calculations may be affected by the presence or absence of certain patient characteristics. Variables that independently contribute to an increased incidence of pneumonia may disproportionately increase adjusted HRs by having an added influence on the prevalence of postoperative pneumonia. We acknowledge that our analysis did not account for the specific type of anesthesia (example: inhalation versus intravenous). We, however, do not think of a reason that our findings will be explained away by variation in anesthesia technique for two main reasons. First, the E value analysis revealed that our findings were robust to unmeasured confounding, in that an unmeasured confounder must be associated with both neurologic diagnoses and postoperative pneumonia by an effect size of at least 2.85-fold to explain away the observed associations. We have not found an empirical evidence suggesting that specific type of anesthesia is independently associated with pneumonia by such effect size. Second, our analysis accounted for several indicators of surgical profile that include operating time, surgical specialty, work relative value unit, and emergent or urgent status. Another limitation related to the use of the NSQIP-P database is the possibility of variations in data reporting. Several institutions contribute to the database, and, despite efforts to the contrary, there may not be total uniformity in the reporting of pneumonia occurrences, neurologic comorbidities, or other patient characteristics. In a retrospective analysis such as this, there is inevitably no uniformity in surgical and perioperative care practice between participating institutions. Available resources, types of procedures, institutional practices, and staff experience may all play a part in the rates of pneumonia and contribute to differences in length of stay. Because there is a financial requirement to participate in the NSQIP-P program, there may be a disproportionate representation of large, teaching institutions contributing to the data. Finally, we must acknowledge that the NSQIP-P database does not include information on participating institutions. This limitation precluded us from accounting for potential within hospital clustering of the outcome. However, even if clustering were present, our effect estimates would not have been biased because clustering only affects the standard errors and does not lead to biased estimates of effect. In addition, it is unlikely that the statistical significance of our effect estimates would be the result of a substantial clustering because higher-performing hospitals do not systematically have lower mortality rates.²⁹ 

In this large-scale analysis of a noncardiac inpatient pediatric surgical cohort, we documented a markedly increased risk of postoperative pneumonia among children with preoperative neurologic comorbidities relative to their peers without neurologic disorders. Although children with cerebral palsy appear to have the greatest risk of postoperative pneumonia, other neurologic disorders were also associated with elevated risk. Therefore, when caring for children with neurologic comorbidities, perioperative care should focus on measures to reduce the occurrence of postsurgical pneumonia.

CI

confidence interval

HR

hazard ratio

IQR

interquartile range

NSQIP

National Surgical Quality Improvement Program

NSQIP-P

National Surgical Quality Improvement Program–Pediatrics