Pneumonia Prevention Strategies for Children With Neurologic Impairment

We conducted a comparative effectiveness study of a retrospective cohort of children with NI enrolled in California Children’s Services (CCS), the largest Title V program in the United States, a federal block grant program supporting state delivered maternal and child health services. CCS provides health care services for children with special health care needs (CSHCNs) in California on the basis of qualifying diagnosis and family income.^13,14  CCS data include all inpatient care and all outpatient episodes including primary care, specialty care, and pharmacy. We included children aged 0 to 21 years enrolled in CCS between July 1, 2009, and June 30, 2014, the period for which we have high-quality data. We chose a cutoff age of 21 years, the age at which CCS coverage ends for children. We included children with NI and ≥1 pneumonia hospitalization in our study population. We chose initial (index) pneumonia hospitalization as an indicator for a child with NI who is at risk for severe pneumonia and pneumonia hospitalization as a surrogate marker for severe pneumonia. We did not include outpatient pneumonia episodes because of the lack of evidence that pneumonia is preventable. NI was defined by International Classification of Diseases, Ninth Revision (ICD-9) and International Classification of Diseases, 10th Revision codes previously described in the literature.¹  Pneumonia hospitalization was identified as an inpatient admission with an International Classification of Diseases code of community-acquired pneumonia or aspiration pneumonia as validated in previous studies.^15,16  Children with cystic fibrosis were excluded given their different risk factors for pneumonia. Pneumonia instances during birth hospitalizations were also excluded because of the potentially iatrogenic nature and likely different risk factors from the birth course.

The primary outcome variable was subsequent severe pneumonia, defined as a pneumonia hospitalization that occurred after the index pneumonia hospitalization. We excluded pneumonia hospitalizations that occurred within 30 days of discharge from the index pneumonia hospitalization, which could be a readmission and reflective of the care quality or illness severity of the index hospitalization.

Children with NI are more susceptible to community-acquired and aspiration pneumonia because of 2 underlying mechanisms: neuromuscular control (poor control of oral secretions, weak respiratory muscles) and systems factors (delayed care, suboptimal home therapies).^1,6  Interventions recommended by expert consensus guidelines or that have previously been associated with changes in risk of pneumonia in children with NI are: dental care after index pneumonia admission, new gastrostomy tube placement, use of gastric acid suppression medication, access to a primary care provider, outpatient clinic visit within 30 days after discharge from index pneumonia hospitalization, receipt of outpatient chest physiotherapy, and oral secretion management.^17–26  We did not include influenza and pneumococcal vaccines as treatment variables despite strong evidence they prevent pneumonia because Current Procedural Terminology (CPT) codes for vaccine administration were not reliably captured in the data.^27–29  A pediatric dentist (YT) reviewed the Code on Dental Procedures and Nomenclature codes associated with each dental visit in the cohort to confirm that dental encounters were consistent with routine dental care. We used receipt of outpatient antibiotics before index pneumonia hospitalization and outpatient clinic visit before index pneumonia hospitalization as surrogate markers for access to a primary care provider because we were unable to distinguish between primary care providers and subspecialty care providers in the data. On the basis of previous studies, we hypothesized a priori that gastric acid suppression would be the only intervention associated with increased risk of subsequent severe pneumonia.²⁴  All variable definitions are detailed in Supplemental Information.

We reviewed and presented the literature to a modified Delphi panel to determine the time to efficacy for each intervention because each of the interventions may have different anticipated time to efficacy but no clearly defined time to efficacy exists in the literature except for dental care, which is recommended every 6 months.^30,31  The panel consisted of 5 clinicians: an outpatient pediatrician specializing in care for CSHCNs, a pediatric hospitalist practicing at a specialty children’s hospital, a pediatric neurologist, a pediatric pulmonologist, and a pediatric infectious disease specialist. The expert panel determined that interventions documented within 180 days after discharge could reasonably be attributed to the hospitalization. Patients were considered to have received an intervention if it was documented within the minimum time to efficacy and within 180 days of discharge or before subsequent pneumonia hospitalization. Our modified Delphi panel removed mucolytic therapies as an intervention because of the small number of patients who would be candidates for mucolytic therapies and limited evidence in the literature and in clinical practice.

We included any measurable child characteristics previously shown to be associated with health care quality, medical fragility, and pneumonia severity as covariates. Sociodemographic variables included: age at admission for index pneumonia hospitalization, sex, insurance, household income, and race and/or ethnicity. Medical complexity was defined by using the pediatric medical complexity algorithm as nonchronic disease, noncomplex chronic disease, or complex chronic disease.³²  We defined type of NI as static, progressive, static and progressive, or neither static nor progressive. Technology dependence included individual variables for gastrostomy tube and for tracheostomy. Ventriculoperitoneal shunts were not included because they are not a risk factor for pneumonia. The CCS data did not include CPT codes, so technology dependence was defined by presence of keywords. We performed sensitivity analyses around these variable definitions to confirm accurate capture of technology dependence.

We also included characteristics of the index pneumonia hospitalization as covariates. We included time from enrollment in CCS as a potential indicator of access to resources preceding the index pneumonia hospitalization because recent or emergent enrollment in CCS could be associated with more severe symptoms due to suboptimal home therapies or delays to accessing care. We also included distance from admitting hospital and admission to a specialty children’s hospital, defined as any Children’s Hospital Association member in California. Markers for clinical severity of the index pneumonia hospitalization included variables associated with pneumonia severity: length of stay, ICU admission, intubation, presence of empyema, pleurisy, pleural effusion, abscess of the lung, and pleural drainage.

Sample characteristics were compared for children with and without a subsequent pneumonia hospitalization. Bivariate comparisons were assessed with Student’s t test for continuous data and with χ² test for categorical data.

For our primary analysis, we used 1:2 propensity score matching using 0.2-SD calipers and nearest-neighbor matching to address potential confounding by indication.^33,34  To generate our propensity score, we used all measurable sociodemographic variables, medical complexity, type of NI, technology dependence, and characteristics of the index pneumonia hospitalization. We defined strong propensity score matches as those with standardized mean differences of <0.2. Propensity score matching was overall strong for all exposure variables except for chest physiotherapy and clinic visit after discharge as shown in Supplemental Information. The members of the modified Delphi panel were unable to identify additional covariates to add to the chest physiotherapy propensity score. For clinic visit within 30 days after discharge, the high prevalence of exposed (77%) in this population made finding strong 1:2 propensity score matches difficult given the small number of controls to draw from, so we reduced our model to a 1:1 propensity score match. For new gastrostomy tube placement, we excluded those patients who had previously received a gastrostomy tube before index hospitalization because they would not be eligible for consideration for new gastrostomy tube placement.

We specified secondary analyses a priori of a traditional logistic regression to evaluate the differential effect sizes of each intervention and a sensitivity analysis of inverse probability treatment weighting for average treatment effect of the overall population.³⁵  Covariate selection for traditional logistic regression was based on bivariate logistic regressions of the primary outcome variable and covariate of interest that demonstrated a change of ≥10% in the odds ratio and clinical judgment.³⁶  We did not use inverse probability treatment weighting as our main analysis because of the greater potential for skewed results in the setting of a weak propensity score resulting from the high weights given to subjects with low probability of treatment receipt.³³  We performed additional post hoc sensitivity analyses to evaluate for residual confounding by indication using an instrumental variable analysis and prognostic score matched analysis.

All analyses were performed by using SAS version 9.4 (SAS Institute Inc, Cary, NC). The study was reviewed by the Stanford University Institutional Review Board and deemed exempt.

A total of 10 473 children were enrolled in CCS during the study period. Of those children, 5493 (52.4%) were children with NI. Of the children with NI, 3632 (66.1%) had ≥1 pneumonia hospitalization, and of these children, 1362 (37.5%) had a subsequent pneumonia hospitalization. The study population is further described in Fig 1. Most subsequent hospitalizations occurred within the same year as the index hospitalization. Most children with a subsequent hospitalization had multiple subsequent pneumonia hospitalizations (n = 691; 50.7%; range: 2–10). Temporal patterns between index and subsequent pneumonia hospitalizations are detailed in Supplemental Information.

Patient sociodemographic and clinical characteristics are shown in Table 1. Compared with children without a subsequent pneumonia hospitalization, children with a subsequent pneumonia hospitalization were less likely to be insured by Medicaid managed care, more likely to identify as Hispanic, and more likely to be categorized as medically complex, technology dependent, and with static NI. Children with a subsequent pneumonia hospitalization had longer lengths of stay and higher prevalence of intubation and admission to a specialty children’s hospital during the index hospitalization. All other characteristics of the index hospitalization were similar between the groups.

Documented receipt of pneumonia prevention strategies varied broadly, as shown in Table 2. The most commonly documented strategy was a clinic visit within 30 days of discharge from index hospitalization (n = 2798; 77%), and the least common was chest physiotherapy (n = 191; 5%). Children with a subsequent pneumonia hospitalization had higher rates of documented receipt of most strategies. Temporal patterns between documented receipt of a strategy and index hospitalization are seen in Supplemental Information. All strategies except dental care occurred in close proximity to the index hospitalization.

The results of our primary analysis using propensity score matching are shown in Table 3. A decreased adjusted odds ratio (aOR) of subsequent pneumonia hospitalization was associated with documented receipt of dental care (aOR: 0.64; 95% confidence interval [CI]: 0.49–0.85). Increased aORs of subsequent pneumonia hospitalization were associated with documented receipt of the following exposures before the index hospitalization: antibiotics within 7 days (aOR: 1.42; 95% CI: 1.06–1.92) and clinic visit within 7 days (aOR: 1.30; 95% CI: 1.11–1.52). Increased adjusted odds of subsequent pneumonia hospitalization were also associated with documented receipt of the following exposures after the index hospitalization: new gastrostomy tube placement (aOR: 2.15; 95% CI: 1.63–2.85), chest physiotherapy (aOR: 2.03; 95% CI: 1.29–3.20), and clinic visit within 30 days after discharge (aOR: 1.72; 95% CI: 1.35–2.20). These patterns persisted in the traditional logistic regression and the inverse probability treatment weighted analyses.

For our sensitivity analyses detailed in Supplemental Information, we used gastrostomy tube placement as the exposure variable given its strongest association with increased risk of subsequent pneumonia hospitalization. We were unable to perform an instrumental variable analysis because of weak potential instruments in our first-stage analysis with P values of .13 to .73. Thus, we performed a prognostic score matched analysis for gastrostomy tube placement, which is an established balance measure also known as a disease risk score that is used in an attempt to ensure the groups being compared in a propensity score matched analysis are similar in their baseline risk for the outcome. This is further described in Supplemental Information.³⁷  The association between increased risk of subsequent pneumonia hospitalization and gastrostomy tube placement persisted (aOR: 3.13; 95% CI: 2.26–4.35).

With the results of our study, we raise important issues regarding recommended strategies for promoting respiratory health in children with NI. Of the 8 strategies we evaluated, only dental care was associated with decreased risk of subsequent pneumonia hospitalization. In fact, our results suggest strong associations between certain interventions and increased risk of subsequent pneumonia hospitalization. These results also suggest additional research is needed to improve the evidence base that is used to guide preventive care decisions for children with NI, especially with regard to reducing unnecessary caregiver burden for the often lifelong caregivers of this population.³⁸ 

Our results suggest that more attention should be paid to dental health for children with NI. Consistent with statements by the National Academy of Medicine, dental care likely reduces the risk of severe pneumonia by reducing oral flora that may be pathogenic in lower respiratory tract infections and enter the lung through aspiration or mechanical ventilation.^39,40  Similarly, routine dental care in elderly adults is associated with reductions in pneumonia.²³  However, dental care remains the most common unmet health care need for CSHCNs.⁴¹  CSHCNs on Medicaid have even lower rates of preventive dental care visits, likely because of the impact of reimbursement structures for dental care.^41,42  Our results suggest that expanding dental health access for children with NI may reduce associated Medicaid spending on some preventable hospitalizations.

Our results build on growing evidence that gastrostomy tube placement is not an obvious best option for all children with NI. Gastrostomy tube placement for children with NI may improve weight and height trajectory but has not definitively been shown to prevent aspiration pneumonia and an associated high risk of morbidity, with >80% of children with gastrostomy tubes experiencing ≥1 complication.^43–46  Several phenomena may explain our results. Gastrostomy tube placement may be a marker of rapid clinical decline, although strong propensity scores make this less likely. Other interventions that frequently accompany gastrostomy tube placement (changes in nutrition, feeding schedule, gastric motility, and gut microbiome) may increase risk of pneumonia.^47,48  General anesthesia for tube placement may temporarily decrease respiratory muscle control, potentiating pneumonia. Alternatively, for a high-risk population, gastrostomy tube placement may be a necessary but insufficient preventive measure. Among adults with amyotrophic lateral sclerosis, similar associations exist between gastrostomy tube placement and increased mortality.⁴⁹ 

We also found clinic visits were associated with increased risk of subsequent pneumonia hospitalization. These results align with studies describing more medically complex children as more likely to have an outpatient visit before an ambulatory care sensitive hospitalization.⁵⁰  In our study, receipt of outpatient care around index hospitalization may indicate underlying medical complexity.

Among children with NI, most recommended prevention strategies are not associated with risk reduction for subsequent severe pneumonia. The burden of each recommended treatment on the system and on patients and caregivers may not be justified given our results. The uncertainty around the preventability of severe pneumonia suggests that greater attention should be paid to patient and family preferences and values when initiating these interventions to provide care concordant with family-centered goals of care.^51,52  Current medical practice for patients with NI is geared toward acute care rather than long-term care planning.⁵³  However, our study adds to growing evidence that care decisions should consider a family’s long-term goals.^51,54  Integration of shared decision-making into care discussions may help shift care toward contextualizing care decisions with long-term care plans.^54,55  On the basis of our results, initiation of some pneumonia prevention strategies has uncertain clinical benefit and should warrant greater shared decision-making.

Our results should be interpreted in the context of a few limitations. The biggest limitation of our analysis is the potential for residual confounding by indication wherein more clinically severe patients are more likely to receive the intervention. We were limited to variables captured in Medicaid paid claims, which did not capture more granular clinical factors that could have improved clinical comparability of matched individuals. We were also unable to find a strong instrumental variable to simulate a randomized controlled trial. However, the addition of the prognostic score did improve the similarity between treatment and control groups without altering our results. Another limitation is misclassification bias resulting from incomplete identification of missing data. Specifically, gastrostomy tube, tracheostomy, and chest physiotherapy were identified primarily through ICD-9 codes and keyword search. We tested iterations of these definitions and chose the most restrictive definitions, which would bias the results toward the null.

A strength of this analysis is the longitudinal sample comparing 2 groups over time. Still, our results warrant additional investigation, including validation in other populations and with more granular clinical data (eg, from e-health records). Future studies should include prospective assessment of respiratory health in children with NI undergoing gastrostomy tube placement, including clinical trajectory, decision-making, and diet and feeding schedule changes.

Of the many pneumonia prevention strategies recommended by expert guidelines, only dental care appears to have potentially protective effects against severe pneumonia among children with NI. Notably, gastrostomy tube placement is associated with increased risk of severe pneumonia, but our results should be interpreted with caution because of possible residual confounding by indication. Our findings support a clinical trial of dental care for prevention of severe pneumonia in children with NI and do not support the widespread use of gastrostomy tubes for that purpose.

We thank Tamara Simon, MD, MSPH for advice on the use of the pediatric medical complexity algorithm, other clinical indicators of medical complexity for the study population, and comments on the results of the article. We also thank Doug Owens, MD, MS for comments on an earlier version of the article.

aOR

adjusted odds ratio

CCS

California Children’s Services

CI

confidence interval

CPT

Current Procedural Terminology

CSHCN

child with special health care needs

ICD-9

International Classification of Diseases, Ninth Revision

NI

neurologic impairment