Survival and Health Care Use After Feeding Tube Placement in Children With Neurologic Impairment

Study Design and Setting

In this population-based cohort study, we used health administrative data from Ontario, a province with a population of ∼13 million people. To quantify changes in acute health care use before and after tube placement, we used a self-matching methodology, the exposure-crossover design.¹⁹

Study Data

The health care system in Ontario provides government insurance that includes free access to inpatient, emergency department, and outpatient care for all residents. By using validated data sources linked by coded patient identifiers,²⁰ children can be managed over time. For this study, we accessed demographic information, vital statistics, hospitalization data, outpatient procedure records, emergency department visit data, physician billing for outpatient and emergency department visits, aggregated neighborhood census data, a database linking maternal and infant records, immigration data, and a physician specialty database (Supplemental Information). These data were analyzed at the Institute for Clinical Evaluative Sciences (ICES), which is legislatively permitted to use these data sources without individual consent for health system research, provided strict privacy guidelines are followed. We obtained research ethics approval from the Hospital for Sick Children and administrative approval from the University of Toronto.

Study Population

We identified all children in Ontario undergoing primary feeding tube placement (Supplemental Table 9) between April 1, 1993, and March 3, 2015, with data collection through March 31, 2016. To be included, the feeding tube had to be placed at least 1 year plus 4 weeks (56 weeks) after the newborn hospitalization discharge and 56 weeks before the child’s 18th birthday. Additionally, children had to have a diagnosis code consistent with NI on a hospital record before or at the time of tube placement. We defined NI as a neurologic diagnosis on an established list of pediatric complex chronic conditions (CCCs)^21,22 or a nonneurologic CCC that was also included on an extensive list of diagnoses associated with pediatric NI (Supplemental Table 10).¹ We excluded children born before April 1, 1992, children without a valid encoded identifier to allow data set linkage, and those not continuously eligible for the Ontario Health Insurance Plan for 56 weeks before and after tube placement. Because of clinical differences between GTs and GJTs, children with ambiguous codes for tube type were also excluded. We excluded children who died within 28 days of feeding tube placement from the main study cohort. However, we included their data in the survival analysis for completeness.

Exposure and Time Intervals

Time 0 was the date of feeding tube placement. Each child was managed forward for up to 5 years and backward for up to 2 years. For the exposure-crossover analysis, we limited comparison with the 2 years before (baseline interval) and after (subsequent interval) tube placement, excluding the 4 weeks immediately before and after tube placement (induction interval) because of fluctuating risk of hospitalization from unsafe preoperative oral feeding and short-term postoperative complications. For model stability, children, except children who died during the subsequent interval, required a minimum of 1 year plus 4 weeks of data before and after tube placement.

Outcomes

We calculated days of survival from tube placement until date of death. Data from surviving children with <5 years of follow-up were censored on the date of their last clinical encounter. Feeding tube revision was defined as a second tube placement procedure code after the initial placement. Antireflux surgeries occurring >1 week after the primary tube were considered subsequent. Procedure codes are listed in the Supplemental Information.

For health care use, we divided time into 1-week segments before and after time 0 and counted outcomes during each week. Outcomes included unplanned hospitalization days, unplanned hospital days with primary diagnosis codes related to reflux¹⁸ (Supplemental Information), emergency department visits without admission, and outpatient visits.

Covariates

Covariates were defined at the time of tube placement. Neighborhood income quintile, adjusted for community and household size, was determined for areas of 400 to 700 inhabitants by using postal code and census data and was used as a proxy for socioeconomic status. Rurality was determined with an Ontario-specific score by using postal code. Immigration status was determined through immigration records for children or mothers in the Immigration, Refugees, and Citizenship Canada Permanent Resident Database. We classified tube type using physician billing and procedure codes. Children with presurgical hospitalizations >2 days had “inpatient” procedures. With reference to a published algorithm,²³ 3 pediatricians (K.E.N., A.G., and E.C.) with relevant clinical expertise and experience validating International Classification of Diseases codes for research defined NI diagnosis codes as “progressive” if significant decline or death would be expected during childhood. All hospital records before tube placement were evaluated for CCC diagnosis codes. Presence of other medical technology (tracheostomy, cerebrospinal fluid shunt, evacuation tube, renal or cardiac support, infusion pump) was determined by inclusion of diagnosis or procedure codes on a hospital record without a subsequent removal code before tube placement. Antireflux procedures included both GJTs and antireflux surgery (Nissen fundoplication); procedure codes for antireflux surgery within 1 week of tube placement were considered concomitant. Provider type was determined by using an Ontario-specific algorithm. Additional covariate details are described in the Supplemental Information.

Descriptive Analyses

We fit a Kaplan-Meier survival curve for the cohort from the date of tube placement, stratifying by antireflux procedure status. We created bamboo plots,¹⁹ which are specific to the exposure-crossover design and depict rates over time, revealing weekly per-child rates of the health care use outcomes over 7 years (2 years before and 5 years after tube placement).

Primary Analysis

For the exposure-crossover analysis, we compared acute health care use during the 2 years before and after feeding tube placement, excluding the 8-week induction interval adjacent to the procedure. When underlying time trends were visible on the bamboo plot, we used a negative binomial generalized estimating equation (GEE) with a compound symmetry covariance structure to calculate a rate ratio (RR) comparing the mean rates in the subsequent interval to the baseline interval using the child’s proportion of weeks at risk during the interval as an offset. To account for time trends, we adjusted estimates for the number of weeks from the beginning of follow-up. We further adjusted for age at intervention.

To evaluate for survivor bias and era effect, we performed prespecified sensitivity analyses excluding children who died during the subsequent interval and stratifying by era (1993–2000, 2001–2005, 2006–2010, and 2011–2015). To assess robustness, we ran models in which outliers were excluded, and we tested longer and shorter induction intervals. Because an emergency department–specific database debuted in 2002, we performed a sensitivity analysis excluding pre-2002 data.

For the primary outcome of unplanned hospital days, we performed prespecified subgroup analyses for children older and younger than 4 years of age at tube placement,²⁴ with and without progressive NI, with and without antireflux procedures, and children who were and were not inpatients at the time of feeding tube placement. To evaluate for variation in patterns of secondary outcomes between children with and without antireflux procedures, we created stratified bamboo graphs.

Secondary Analysis

As an exploratory post hoc analysis to quantify the relative effects of membership in the subgroups (older and younger than 4 years of age,²⁴ with and without progressive NI, with and without antireflux procedures, and children with inpatient and outpatient tube placement), we fit a negative binomial multilevel growth curve model on days of hospitalization per week during the subsequent interval. The level 1 components of the model included random slopes and random intercepts for each child. The level 2 components included all covariates, which were defined at time 0 and treated as time invariant. For covariates with >5% of missing values, we assigned a category as missing; otherwise, we performed complete-case analysis. All analyses were performed in SAS version 7.1 (SAS Institute, Inc, Cary, NC); all testing was 2 sided, and P values <.05 were considered significant.