Comparing routine inpatient data and death records as a means of identifying children and young people with life-limiting conditions

Background: Recent estimates of the number of children and young people with life-limiting conditions derived from routine inpatient data are higher than earlier estimates using death record data. Aim: To compare routine inpatient data and death records as means of identifying life-limiting conditions in children and young people. Design: Two national cohorts of children and young people with a life-limiting condition (primary cohort from England with a comparator cohort from Scotland) were identified using linked routinely collected healthcare and administrative data. Participants: A total of 37,563 children and young people with a life-limiting condition in England who died between 1 April 2001 and 30 March 2015 and 2249 children and young people with a life-limiting condition in Scotland who died between 1 April 2003 and 30 March 2014. Results: In England, 16,642 (57%) non-neonatal cohort members had a life-limiting condition recorded as the underlying cause of death; 3364 (12%) had a life-limiting condition-related condition recorded as the underlying cause and 3435 (12%) had life-limiting conditions recorded only among contributing causes. In all, 5651 (19%) non-neonates and 3443 (41%) neonates had no indication of a life-limiting condition recorded in their death records. Similar results were seen in Scotland (overall, 16% had no indication of life-limiting conditions). In both cohorts, the recording of life-limiting condition was highest among those with haematology or oncology diagnoses and lowest for genitourinary and gastrointestinal diagnoses. Conclusion: Using death record data alone to identify children and young people with life-limiting condition – and therefore those who would require palliative care services – would underestimate the numbers. This underestimation varies by age, deprivation, ethnicity and diagnostic group.


Cause of death
The Scottish death records contained underlying and other causes of death. Whether underlying cause of death was a life limiting condition was checked using the coding framework. If not a life limiting condition, underlying cause was assessed to see whether it was related to a life limiting condition identified in the inpatient data. For example, nonspecific cerebral palsy as cause of death was considered related to quadriplegic cerebral palsy; unspecified congenital malformations of heart to tetralogy of Fallot. Finally, for those with underlying cause neither a life limiting condition nor life limiting condition-related, contributing causes of death were checked against the life limiting condition coding framework. Where life limiting condition was recorded as a contributing cause, trauma-related underlying causes were determined (all codes starting S; T0; T1; T2; T30; T31; T32; T5; T6; T7; T9; V; W; X; Y1; Y2; Y3). The analyses were split by age group (at death), by ethnic group, by deprivation category, by diagnostic category and by financial year of death.

Statistical modelling
Predictors of life limiting condition being present in death records were explored. A binary outcome variable was defined indicating presence of life limiting condition in a death record, set to 1 if the underlying cause was a life limiting condition or was related to a life limiting condition or a contributing cause of death was a life limiting condition and to 0 if there was no indication of life limiting conditions. Candidate predictor variables were: age group at death, primary diagnostic group, deprivation category and ethnic group. Multivariable logistic regression models were fitted, with candidate predictors added in turn and retained if their odds-ratios were significantly (p < 0.05) different to 1 or inclusion reduced Schwarz's Bayesian Information Criterion for the model by more than 2. 5,6 Interactions between deprivation and ethnic group were also considered (using the same inclusion criteria). Individuals with data missing for any included predictors were excluded from the corresponding models.

Cohort size
In Scotland, 20436 children and young people with a life limiting condition were identified between 1 April 2009 and 31 March 2014 in the Scottish inpatient and birth records. Of these, 2320 had death records in the period. 71 death records were excluded as there were one or more inpatient admissions after the recorded date of death, leaving 2249 individuals in the final cohort. Numbers of deaths in each year and cohort demographics are shown in Table S1.

Missing data
There were large numbers of missing ethnicity data in the Scottish cohort, over half for those dying in the first few years of the study (Table S1). There were no missing data for age group or diagnostic group and missing data for deprivation were at or below 1%.

Cause of death
In Scotland, 1291 cohort members (57%) had a life limiting condition recorded as underlying cause of death (Table S2); 319 (14%) had a life limiting condition-related underlying cause and 268 (12%) had life limiting conditions only among contributing causes, of which 10 had a trauma-related underlying cause (Table S2). 371 (16%) had no indication of life limiting conditions in their death records.

Cause of death by financial year of death
There was only minor variation in recording of life limiting conditions across financial year of death (Table S1), with no clear trend over time.

Cause of death by age at death
In Scotland, recording of a life limiting condition as underlying cause of death was lowest (239, 41%) amongst under 1 year olds, but they had the highest percentage of life limiting condition-related deaths (150, 26%) (Table S2)

Cause of death by diagnostic group
In Scotland, 94% of individuals with a haematology or oncology diagnosis had a life limiting condition as underlying cause of death; only 2% had no indication of life limiting conditions in their death records (Table S5). Only 36% of patients with a genitourinary diagnosis had a life limiting condition as the underlying cause of death and 33% no life limiting condition among any cause of death.

Multivariable model
The multivariable model for the Scottish data (Table S6) showed no significant variations by age.
Neither were there significant variations by ethnic group or deprivation category (although for deprivation category there was an apparent trend towards increased likelihood of life limiting condition recording for individuals from less deprived areas, albeit not significant). Primary diagnostic group showed similar patterns to those seen in the univariable analyses: haematology and oncology diagnoses were most likely to be associated with life limiting condition recording in death records and genitourinary, gastrointestinal and perinatal diagnoses least likely.