Recurrent Intensive Care Episodes and Mortality Among Children With Severe Neurologic Impairment

Key Points Question Are the number and severity of recent critical illness episodes associated with increased short-term mortality among children with severe neurologic impairment (SNI) admitted to the pediatric intensive care unit (PICU)? Findings In this cohort study comprising 4774 children with SNI admitted to the PICU, survival at 15 years after their initial episode was 79%. One-year survival after high-risk episodes decreased from 90% for the first episode to 81% after the fourth episode. Meaning This study suggests that there is a modest dose-dependent association between number of recent high-risk critical illness episodes and mortality among children with SNI.


Introduction
[5] A subset of children with CCCs have severe neurologic impairment (SNI) with underlying neurologic or genetic diagnoses that are associated with significant functional impairments and medical complexity. 6Although they comprise only about 0.2% of children, 7 those with SNI accrue approximately 25% of PICU admissions, 8 are more likely to be readmitted than children with other chronic conditions, 9 and are at increased risk of respiratory failure from acute respiratory distress syndrome. 10For caregivers of children with SNI, end-of-life decision-making is common; in a survey of 103 bereaved parents, 42% reported decisions to withhold or withdraw life-sustaining therapies. 11kelihood of survival is an important factor in these decisions, 12 and patterns of PICU admissions are included in the estimation of mortality risk.A common clinical belief is that increasing frequency of PICU admissions represents a health trajectory portending death.However, to our knowledge, the association between recurrent PICU admissions and subsequent mortality for children with SNI has not been quantitatively evaluated.
We assessed survival after critical illness episodes (CIEs) among children with SNI born from 2002 to 2019 in Ontario, Canada.Specifically, we tested the hypothesis that more frequent CIEs in the preceding year would be associated with higher 1-year mortality.

Study Design and Data Sources
This population-based retrospective cohort study used Ontario health administrative data, which included all children and hospitals in the province.The most populated Canadian province, Ontario has 14.6 million residents who receive all acute care through a single-payer public health care system.
We used linked health administrative databases available at ICES to identify and follow up data over time on Ontario children with SNI.ICES is an independent, nonprofit research institute with special status under Ontario's Personal Health Information Protection Act (PHIPA), which allows researchers to access health and demographic data to evaluate and improve the health care system without individual consent.This project's use of data was authorized under section 45 of PHIPA, approved by ICES' Privacy and Legal Office, and did not require additional review by a research ethics board.We accessed health administrative records (Canadian Institute for Health Information Discharge Abstract Database and the Ontario Health Insurance Plan database) and demographic data (Registered Persons Database).Datasets were linked using unique encoded identifiers.We followed the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guideline for cohort studies. 13

Descriptive Analyses
We identified children's SNI diagnostic categories 15 based on the SNI codes in the hospitalization record for the initial CIE.Among children without SNI codes in the hospitalization record for their initial CIE, we derived the SNI diagnostic category from their first SNI hospitalization.Area-level indicators of rurality and socioeconomic status were derived from census data using the postal code listed in the hospitalization record for the initial CIE.We defined rural or urban residence using the Rurality Index of Ontario, 16 which assesses population density and health care accessibility.Income quintiles were assigned based on population averages from the census dissemination area (400-700 people).Sex was obtained from the administrative dataset Registered Persons Database.Details about race and ethnicity are not available in the data sources and thus were not assessed.We defined comorbidities according to ICD-10-CA codes identified as CCCs, 17 omitting ICD-10-CA codes that were cross-listed on the SNI diagnosis list.Similarly, we used a list of ICD-10-CA codes describing medical technology assistance that were developed by the Canadian Institute for Health Information specifically for use in categorizing children with medical complexity. 7Because medical technology assistance and CCCs are inconsistently recorded in hospitalization records, we included a lookback to the initial hospitalization with an SNI diagnosis to improve sensitivity in their identification. 7We also described features of the CIEs, including the number of episodes per child, the frequency of high-risk markers, and the time to event from first to second and from second to third CIE.

Survival
Using death data from the Registered Persons Database, which captures all deaths reported to the province irrespective of location (hospital, home, hospice, or other facility), we calculated survival from the date of PICU discharge until date of death.Surviving children's data were censored at study end (March 31, 2020) or if they moved out of Ontario (eAppendix 2 in Supplement 1).

Statistical Analysis Baseline Long-Term Survival
Statistical analysis was conducted from November 2021 to June 2023, using SAS Enterprise, version 7.1 (SAS Institute Inc).All P values were from 2-sided tests, and results were deemed statistically significant at P < .05.Our primary analysis was designed to test the general assumption that the recent frequency of CIEs is associated with mortality, irrespective of individual-level characteristics.
Conditional survival estimates, however, require a base case for comparative interpretation, so we calculated long-term survival estimates for the cohort from the date of first CIE discharge in Ontario after or concurrent with first SNI diagnosis.Because the baseline risk of childhood mortality is highest during the first year of life, 18 we chose a priori to stratify by age (<1 year and Ն1 year).To understand the association of individual-level factors with long-term survival, we created age-stratified Cox proportional hazards regression models from discharge date of initial PICU-CIE.Our Cox proportional hazards regression models included baseline demographic characteristics defined at the time of initial PICU-CIE: year, sex, rurality, income quintile, presence of non-SNI-associated CCCs, and use of medical technology assistance.Because many children had multiple SNI and non-SNI complex CCCs and diagnoses were inconsistently recorded across episodes, we did not include diagnostic categories in models.

Short-Term Survival Conditioned on Number of Recent Episodes
To assess the strength of the association between number of recent PICU-CIEs and short-term mortality, the analysis was anchored at the level of the episode (ie, a child with multiple episodes contributed data for each episode).However, not all PICU-CIEs are likely associated with the same risk of mortality.For example, some children with SNI require intensive care to support their baseline medical technology.To identify PICU admissions that more likely indicate worsening of a child's baseline health, we stratified the short-term survival analyses by risk (standard risk vs high risk).We defined high-risk admissions as those with increased risk of mortality based on either prolonged stay 19 or illness severity.In prior studies of PICU admissions for children with CCCs, prolonged length of stay was defined as more than 15 days. 20We chose intubation as the marker of illness severity because invasive mechanical ventilation is a risk factor for mortality 21 and a marker of illness severity on the Pediatric Index of Mortality. 22For children with chronic ventilatory needs, we considered only prolonged admissions to be high risk; their other episodes were classified as standard risk.
To understand the association between CIEs and subsequent mortality, our primary analysis estimated 1-year survival conditioned on the number of high-risk episodes in the preceding year.For each CIE, we looked back 1 year from the date of PICU admission to assess the number and risk of preceding episodes in the preceding year.We did not differentiate between initial admissions and unplanned readmissions; a child could accrue multiple PICU-CIEs within a single hospitalization.However, if more than 1 year elapsed between 2 PICU-CIEs, both would contribute to the analysis for the first discharge.We assumed that increased mortality risk would persist 1 or more years after a high-risk episode.Thus, an episode was classified as high risk if it or any of the preceding episodes in the prior year were high risk (length of stay >15 days or use of invasive mechanical ventilation); all other episodes were standard risk.Within the high-risk and standard-risk categories, we assessed change in survival conditioned on number of previous episodes.Approximately 40% of the cohort (1893 [39.7%]) had a cardiac CCC.Among children with 2 or more CIEs (n = 1728), the median time from the first to second episode was 2.5 months (IQR, 0.3-11.3months).For children with 3 or more episodes (n = 864), the median time from the second to third  3).

Short-Term Survival Based on PICU Episode Characteristics
Across all conditions, 1-year survival was greater than 80% (Table 2; eFigure 2A-D in Supplement 1).

Discussion
In a large population-based sample, 17.2% of children with SNI had CIEs requiring PICU admission between 2002 and 2019.Among children who survived their first CIE, 79% were alive 15 years later; increased complexity was associated with higher mortality.One-year survival conditioned on the number and severity of episodes in the preceding year ranged from 81% to 94%.Among high-risk episodes, there was a modest dose-dependent association, with 1-year survival decreasing from 90% after the first episode to 81% after the fourth episode.
Our findings conform to other comparable studies.Reported 5-year survival rates of children with SNI after gastrostomy tube placement ranged from 76% 23 to 86%. 24In a prior study of mortality after PICU admission, children with neurologic CCCs had 1-year survival ranging from 95% for children with a single discharge to 89% for those with multiple PICU discharges, as estimated This forest plot does not show missing categories for rurality or income quintile (Յ2%).AHR indicates adjusted hazard ratio; CCC, complex chronic condition; PICU, pediatric intensive care unit; and SNI, severe neurologic impairment.Abbreviation: PICU-CIEs, pediatric intensive care unit critical illness episodes.
a An episode was classified as high risk if it or any of the earlier episodes in the preceding year included invasive mechanical ventilation or had a length of stay more than 15 days.The focus of this analysis is on the episode, not the child; a child with multiple critical illness episodes would contribute data for each individual episode.
from Kaplan-Meier plots. 2 Most studies describing long-term survival focus on survival from birth, rather than from an event that can occur at any age, and thus are not comparable.
Our findings highlight the challenges in using population-level estimates to inform individuallevel care.Severe neurologic impairment includes diagnoses with variable prognoses, so interpretation of findings for individuals requires nuance; our results must be contextualized based on a child's specific clinical situation.For example, an infant with trisomy 18 has different projected survival than a teenager with cerebral palsy and recurrent PICU-CIEs.Furthermore, health administrative data provide no details about why clinical decisions were made.Decisions to withhold or withdraw life-sustaining technology are common before pediatric deaths-occurring prior to 40% of inpatient deaths 25 and 70% of PICU deaths 26 -so mortality data do not indicate natural history.
Finally, while prognosis is often part of advanced care planning conversations, survival estimates provide one piece of information; our study gives no insight into quality of life or other important outcomes. 27

Strengths and Limitations
Our results should also be interpreted in light of the study's strengths and limitations.Strengths include the validity and generalizability of our findings, which are enhanced by population-level data that capture cross-provincial hospitalizations and deaths occurring in any location (eg, home, hospital, or hospice).We identified the cohort using a published list of diagnosis codes associated with high-intensity neurologic impairment 15 ; this list and analogous lists 28 have been used in many prior studies about children with SNI 23,[29][30][31] ; in general, diagnostic code lists 14 work well to describe population-level trends in health care use. 32wever, several limitations are important to note.First, because diagnosis codes do not differentiate children based on functional level, this cohort likely includes children with SNI as well as those with milder impairments 14 (eg, gross motor function classification system level 1 and level 2 cerebral palsy).However, inclusion in the cohort required 1 or more PICU-CIEs, which likely reduced the proportion of children with milder impairments.Furthermore, children with milder functional impairments would be less likely to have recurrent admissions, so their inclusion would likely exacerbate a dose-dependent mortality gradient between first and subsequent admissions.Second, there is likely a spectrum of mortality risks associated with a PICU episode, which we have treated as binary (high risk vs standard risk) and assessed with imperfect surrogates (length of stay and invasive mechanical ventilation).Some episodes were potentially misclassified as high risk if the reason for prolonged length of stay or mechanical ventilation was not severe illness (eg, delayed discharge for logistical reasons or planned ventilation between surgical stages); others were misclassified as standard risk because of earlier discharge to intermediate care units.Further evaluation of mortality using more specific markers for severe critical illness-such as admission type (medical or surgical; planned or unplanned), vasoactive medications, increased respiratory settings for children with tracheostomy, or acute critical illness diagnoses (eg, status epilepticus)-would be valuable.Similarly, mortality data for common subpopulations (eg, children with SNI and congenital heart disease) would also be helpful.Third, we made multiple methodological decisions that likely influenced the magnitude of the survival estimates.For example, we considered episodes less than 15 days among children with chronic ventilatory needs as standard risk, we assessed episodes as high risk if any preceding episodes were high risk, and we did not differentiate between static and progressive SNI diagnoses.Although specific estimates might vary based on analytic choices, because survival was more than 80% for all conditions, our interpretation-that most children with SNI discharged from the PICU, including those with multiple recent admissions, will survive at least 1 year-is likely robust.Fourth, our data could not differentiate anticipated from unanticipated deaths, so we could not assess the association of decisions to limit life-prolonging therapies with mortality.
Fifth, this study did not assess the likelihood of future PICU admissions based on the number of recent PICU admissions.Because PICU admissions have significant quality-of-life implications, 33 understanding this association is important for clinical decision-making and warrants further study.

Conclusions
Critical illness episodes requiring PICU admission are common among children with SNI.
Approximately 80% of children with SNI are alive 15 years after their first CIE.Short-term survival decreases with increased number of recent high-risk PICU-CIEs, from 90% after the first discharge to 81% after the fourth.Although these findings support the common hypothesis that recurrent PICU episodes are associated with increased mortality, the absolute difference in estimated survival (and the converse of estimated mortality) is modest.In the context of counseling parents regarding potential major decisions to transition from life-prolonging to comfort-focused therapy, how this finding is framed will be critical; while the risk of mortality nearly doubles (from 10% to 19%), the likelihood of survival (expressed as natural frequencies) decreases only slightly, from 9 of 10 patients to 8 of 10 patients.The modest association identified in this study challenges the common narrative that increasing frequency of recent PICU admissions indicates a substantially higher risk of subsequent mortality.

Figure 3 .
Figure 3. Factors Associated With Mortality After Children's First Critical Illness Episodes

Table 1 )
. They represented 17.2% of the 27 731 children in Ontario with SNI-associated diagnosis codes in a hospital discharge record from 2002 to 2019 (Figure 1).Most of the cohort (3050 [63.9%]) had additional CCCs not associated with SNI; many children (1240 [26.0%]) had CCCs affecting 2 or more organ systems (Table 1).

Table 1 .
Clinical Characteristics of Children With Severe Neurologic Impairment Discharged Alive After Their First Critical Illness Episode was 3.1 months (IQR, 0.6-11.9months).Most individuals accrued 1 or 2 CIEs during childhood (3910 [81.9%]); 306 children (6.4%) had 5 or more episodes.Among all CIEs (n = 8995), 4158 episodes (46.2%) included invasive mechanical ventilation, 1267 episodes (14.1%) lasted more than 15 days, and 4552 episodes (50.6%) included 1 or more severity markers.hazards regression model, the shapes of the hazard function varied, but the coefficients were the same in both strata.Adjusting for demographic characteristics (year of admission, sex, rurality, and income quintile), there was an increased risk of death for children who had long-term medical technology use (adjusted hazard ratio [AHR], 2.32 [95% CI, 1.92-2.81])and additional CCCs not associated with SNI (AHR, 1.70 [95% CI, 1.43-2.02])(Figure Abbreviation: SNI, severe neurologic impairment.aChildren may be included in multiple categories.episode4774 Children with SNI discharged alive after first critical illness episode 2636 Aged <1 y 2138 Aged ≥1 y For this study, we excluded brief postprocedural critical care unit admissions occurring within 24 hours of a surgical procedure and lasting less than 48 hours.SNI indicates severe neurologic impairment.Figure 2. Long-Term Survival of Children With Severe Neurologic Impairment After Their First Critical Illness Episode proportional

Table 2 .
Short-Term Survival Conditioned on Number and Intensity of PICU-CIEs in the Preceding Year Two-Year Survival of Children Younger Than and Older Than 1 Year With Severe Neurologic Impairment Following Their Initial Discharge From the Pediatric Intensive Care Unit eFigure 2. One-Year Survival After PICU Discharge Based on Number of PICU Admissions in the Preceding Year (1st to 4th Episode) Stratified by High Compared to Standard Risk eAppendix 1. Diagnosis and Intervention Codes Defining Children With Severe Neurologic Impairment eAppendix 2. Notes on Methodology eReference.