Incidence, socioeconomic deprivation, volume-outcome and survival in adult patients with acute lymphoblastic leukaemia in England

Study design, area and time span

We used a population (ecological) study design to investigate incidence and a cohort study design to examine survival amongst all adult patients (aged 18 years or more at diagnosis) in England diagnosed from 2001 to 2012. Follow-up was to 31st March 2013. The study was approved by the University of Sheffield’s Research Ethics Committee. Participant consent was not necessary as this study involved the use of de-identified Hospital Episode Statistics (HES) supplied by NHS Digital. NHS Digital is an executive non-departmental public body accountable to Parliament and its statutory role is set out in the Health and Social Care Act 2012.

Data on cases

We used pseudoanonymised HES data with linked mortality data to identify cases with ALL. HES is a data warehouse containing details of all admissions to National Health Service (NHS) hospitals in England [17]. Admissions are recorded as “episodes” where an episode is a period of care under a single consultant during an admission. An admission may comprise more than one episode if a patient is transferred from the care of one consultant to another during the admission, though in practice most admissions comprise a single episode. The admission episode record includes several diagnosis and procedure code fields. We used the first seven diagnosis fields and the first 12 procedure code fields in order to use a set that was consistent throughout the study time span (there were more fields in later years). HES years follow the UK financial calendar and run from 1st April to 31st March the following year. We obtained data from 1st April 2000 to 31st March 2013. We excluded patients with a first admission with ALL in 2000 to exclude prevalent cases.

We used the International Classification of Diseases, 10th revision (ICD-10) code C91.0 to identify cases of ALL. The main advantage of using HES is complete national data capture as every adult in England who developed ALL will almost certainly have been admitted to an NHS hospital on at least one occasion for confirmation of the diagnosis and assessment for treatment and supportive care. However, there is significant scope for error in coding of leukaemia in HES.

We therefore first identified individuals (using the HES pseudoanonymised identifier) with a C91.0 code in any diagnosis field and extracted all their admission episode records. We then carried out a whole series of exploratory analyses which included examining procedure codes for chemotherapy, central line insertion, spinal fluid investigation and bone marrow examination. We also examined numbers of admissions per patient, cause of death and diagnoses likely to be miscoded as ALL. We classified the latter as “ambiguous” diagnoses, which included other leukaemias (C91.1 – C96.9) and lymphomas and myeloma (C81.0 – C90.2).

We arrived at a definition of a case as a patient with all of the following: a diagnosis of C91.0 as the primary diagnosis in at least one episode, an admission count >1; a chemotherapy code recorded in at least one episode; and a ratio of count of episodes with C91.0 code to count of episodes containing an ambiguous diagnosis code >1. Patients admitted for assessment but deemed too frail to withstand chemotherapy, or who died during or following their first admission, would have been excluded by the above definition. We therefore also included, as a separate definition, patients with C91.0 as the cause of death. These patients would also have had a C91.0 code in any diagnosis field in at least one admission episode. Records, in at least one episode, of central line insertion (65.6%), spinal fluid investigation (54.1%) or bone marrow examination (73.7%) were too incomplete to be used in the case definition, given that all patients undergoing diagnosis and treatment would have had these procedures carried out. The average number of cases identified using the above criteria was 243 per year and was comparable to the average estimated count of 240 per year from Office for National Statistics (ONS) statistics based on cancer registries [18].

Other data

We used the Income Domain of the Index of Multiple Deprivation (IMD) 2007 as an indicator of socioeconomic deprivation at the small area level [19]. The IMD is a standard deprivation indicator used by local and national government and is available at the lower layer super-output area (LSOA) level. LSOAs are census areas used in the 2001 and 2011 censuses, with approximately 1500 people per LSOA.

We used the government’s urban-rural classification to classify LSOAs as either urban or rural [20]. Mid-year population counts by five-year age band and sex at the LSOA level were also available from ONS.

Volume and outcome

HES data contain provider unit codes which we used to estimate the volume of adult patients with ALL treated at each provider hospital. Provider units have reconfigured over time on several occasions with consequent changes in codes. We had a consistent set of codes for a limited time span (1st April 2008 – 31st March 2011) available from another project, which we used for this element of the analysis [21]. Estimation of volume was based on new adult patients with ALL diagnosed during this three-year time span. Patients could have been admitted to more than one hospital during this time span and were included in the volume count of each hospital they were admitted to. We used the median to split hospitals into high and low volume hospitals.

Statistical analysis

We used Poisson regression to examine incidence, and survival plots and Cox regression to examine survival. Age (20–29 years for incidence, 18–29 years for survival, then 10-year bands to 80+ years), gender, rurality, time-period (3 four-year periods) and socioeconomic deprivation (categorised by tertile) were entered as categorical variables. The logs of population counts were entered as the offset in the Poisson regression, which was restricted to 31,672 (97.5%) of the 32,482 LSOAs in 2001 which remained unchanged in 2011. Log [−log] plots were consistent with the proportionality assumption for Cox regression. There was no overdispersion in the Poisson model.

Date of diagnosis and age at diagnosis were based on the admission date of the first admission with a C91.0 code in any diagnosis field. Survival time was calculated in years from the date of diagnosis to date of death or censored at the end of the follow-up period. Analysis of the effect of volume on survival was examined on the subset of new patients described above. Results are presented as rate ratios or hazard ratios, adjusted for all other variables, with 95% confidence intervals.

Characteristics of patients

There were a total of 2921 adults aged 18 or more years diagnosed with ALL in the 12-year time span examined, giving a crude annual incidence of 0.61/100,000 population. There were 1870 deaths in 7078 person-years of follow-up time.

Incidence rate ratios

Survival

Figure 1 shows the survival plot for adults diagnosed with ALL in England from 2001 to 12. The overall five-year survival for this cohort of 2921 patients was 32%. When restricted to the 2559 patients who had a record of chemotherapy, the five-year survival was 36%. The adjusted hazard ratios for mortality are shown in Table 3. These were based on analysis of all 2921 patients. Hazard ratios increased with increasing age but there was no evidence of a gender difference in survival. Compared with patients living in the least deprived areas, survival was poorer in patients living in intermediate and most deprived areas, with mortality hazard ratios 21% (8–35%) and 16% (3–30%) higher respectively. In terms of five-year survival, predicted rates for the least, intermediate and most deprived areas were 36% (33–38%), 30% (28–33%) and 31% (29–34%) respectively.

Characteristic	Adjusted hazard ratio (95% CI) for all 2921 patients
Age (y)
- 18-29	1
- 30-39	1.38 (1.15–1.66)
- 40-49	1.69 (1.42–2.01)
- 50-59	2.23 (1.89–2.62)
- 60-69	3.72 (3.18–4.34)
- 70-79	5.50 (4.68–6.46)
- 80+	9.02 (7.43–10.95)
Gender
- Women	1
- Men	1.04 (0.95–1.14)
Socioeconomic deprivation (by tertile)
- Least deprived	1
- Intermediate	1.21 (1.08–1.35)
- Most deprived	1.16 (1.03–1.30)
Urban or rural areas
- Urban	1
- Rural	1.00 (0.89–1.13)
Time period
- 2001-04	1
- 2005-08	0.93 (0.84–1.03)
- 2009-12	0.70 (0.62–0.79)
	Adjusted hazard ratio (95% CI) for subset of 465 patientsa
Record of chemotherapy
- Yes	1
- No	5.53 (3.98–7.69)
Hospital provider volume of ALL patients
- >15	1
- = <15	1.33 (1.03–1.73)

^aPatients diagnosed between 1st April 2008 and 31st March 2011 with all their admissions at a single hospital provider unit during this period. This analysis included chemotherapy and volume in addition to the other variables

There was no association with rurality. Survival improved over the time period examined, and the mortality hazard ratio was 30% (21–38%) lower in patients presenting in 2009–12 compared with patients presenting in 2001–04. In terms of five-year survival, predicted rates for patients presenting in 2001–04, 2005–08 and 2009–12 were 28% (26–31%), 30% (28–33%) and 39% (36–42%) respectively.

Volume-outcome association

There were 727 new patients admitted between 1st April 2008 and 31st March 2011, the period for which we had a consistent set of provider unit codes available. Of these, 465 had all their admissions to the same provider unit during this time span. The adjusted hazard ratio for mortality in this subset of 465 patients was 33% (3–73%) higher in low volume hospitals (Table 3). This analysis additionally adjusted for chemotherapy treatment as patients with no record of treatment for chemotherapy had much poorer survival.