Brain Tumours: Rise in Glioblastoma Multiforme Incidence in England 1995–2015 Suggests an Adverse Environmental or Lifestyle Factor

1Children with Cancer UK, 51 Great Ormond Street, London, WC1N 3JQ, UK
2Powerwatch, Cambridgeshire, UK
3Professor Emeritus, University of Bristol, UK
4Professor Emeritus, Vice–Chancellor’s Office, University of Sunderland, UK

Abstract

Objective. To investigate detailed trends in malignant brain tumour incidence over a recent time period.Methods. UK Office of National Statistics (ONS) data covering 81,135 ICD10 C71 brain tumours diagnosed in England (1995–2015) were used to calculate incidence rates (ASR) per 100k person–years, age–standardised to the European Standard Population (ESP–2013). Results. We report a sustained and highly statistically significant ASR rise in glioblastoma multiforme (GBM) across all ages. The ASR for GBM more than doubled from 2.4 to 5.0, with annual case numbers rising from 983 to 2531. Overall, this rise is mostly hidden in the overall data by a reduced incidence of lower-grade tumours. Conclusions. The rise is of importance for clinical resources and brain tumour aetiology. The rise cannot be fully accounted for by promotion of lower–grade tumours, random chance or improvement in diagnostic techniques as it affects specific areas of the brain and only one type of brain tumour. Despite the large variation in case numbers by age, the percentage rise is similar across the age groups, which suggests widespread environmental or lifestyle factors may be responsible. This article reports incidence data trends and does not provide additional evidence for the role of any particular risk factor.

1. Introduction

The causes of brain tumours in adults remain largely unknown [1]. In 2011, the World Health Organisation (WHO) prioritised the monitoring of detailed brain tumour incidence trends through population–based cancer registries [2]. This article reports recent changes in malignant brain tumour incidence in England that include age, sex, morphology and tumour location.

2. Materials and Methods

2.1. Data

The International Classification of Diseases for Oncology (ICD–O) is a dual classification, with coding systems for both topography and morphology [3]. The relevant topology codes are listed in Table 1, along with the number of tumours diagnosed in 1995 and 2015

Table 1: ONS WHO ICD10 brain tumour data for England.

There are 102 different ICD–O–3.1 morphology codes used in the data set, though many have few cases. The morphology code describes the cell type and its biological activity / tumour behaviour.

WHO last updated their classifications in 2016, but their changes have minimal impact on our analysis of the data [45]. Malignant brain neoplasms without histology are recorded as ICD–10 D43 (D43.0 & D43.2 supratentorial).

We used anonymised individual–level national cancer registration case data from the UK Office of National Statistics (ONS) for all 81,135 ICD10–C71 category primary malignant brain tumours diagnosed in England for the years from 1995 to 2015, plus 8,008 ICD10–D43 supratentorial malignant tumours without histology/morphology data from 1998–2015. The initial data is supplied by the National Cancer Registration Service (NCRS). The ONS then apply further validation checks and the UK Department of Health use the ONS data to inform policy making. The ONS state their cancer data are generally within 2% of the correct values [6]. Until about 2005, some cases in the oldest age–groups will not have been recorded in the cancer registries. Since 2005 this error is likely to be small.

Glioblastoma Multiforme (GBM), the most common and most malignant primary tumour of the brain, is associated with one of the worst five–year survival rates among all human cancers, with an average survival from diagnosis of only about 1 year. This ensures that few cases will be unrecorded in the ONS database and we show that their number of GBM tumours is similar to NHS hospital inpatient numbers. The data include the year of diagnosis, age at diagnosis, sex of patient, primary site and morphology code. National population estimates of age and gender by calendar year were also obtained from ONS data [7] and age–specific incidence rates per 100,000 person–years and for a wide variety of tumour types were calculated in 5-year age group bins for males and females separately.

Some published incidence analyses have used different criteria as to which glioma and astrocytoma should be considered malignant. WHO considers Grades I to IV as biologically malignant even if they have not been graded histologically malignant. We have taken the WHO/IARC morphology behaviour codes /3, /6 and /9 as being histologically malignant which means that Grade I and II tumours are classed as low–grade malignancies.

We are not aware of any specific bias in the ONS data. There is a slight data–lag in cancer registry data, which are regularly checked and updated if necessary, but are generally stable after 3 to 5 years. Our ONS data extract is dated  July 2017.

Brodbelt et al. (2015) [8] reported an analysis of treatment and survival for 10,743 GBM cases in England over the period 2007–2011, which had an overall median survival of only 6.1 months, rising to 14.9 months with maximal treatment. Brodbelt et.al.’s GBM case total from English hospital data is only 0.5% higher that our ONS GBM total of 10,687 cases for the same time period; this suggests that a very complete UK cancer diagnosis and registration system is now in place. In contrast, Ostrom et al. (2015) [9] reporting on USA SEER brain tumour data provide a scatter–plot that shows a median complete registration and histological confirmation level of only about 65%, with the best examples returning less than 75% full completion in 2012.

2.2. Confounding

We had a large number of categories and sub–categories in the data. It was necessary to combine some of these to increase the resolving power. We ran analyses separately for each site (C71.0 to C71.9), for each main type of tumour, and for tumour grade (I to IV). It was immediately obvious that the most significant change was in the incidence of GBM in frontal and temporal lobes. The obvious potential confounders would be the C71.8 (overlapping) and C71.9 (unspecified) categories due to better imaging techniques and we discuss this later.

2.3. Standardisation

Incidence rates rise dramatically with age and standardisation is necessary as population age profiles are changing with time. We calculated age–standardised incidence rates (ASR) per 100k person–years to the current recommended European Standard Population (ESP–2013), as it best represents the reality of the case burden on society [10]. Adjusting European cancer incidence to the World Standard Population is not helpful as the age-spectra are so different.

Table 2 lists the morphology codes with the highest case numbers, totalling 80354 tumours. Included in our analyses are an additional 781 cases in 78 other categories, each with fewer than 100 cases over the 21 years. A full listing of all the cases in the data set is provided in the Supplementary File.

Table 2: ICD-O-3 morphology codes with more than 100 cases between 1995-2015 inclusive. (A full listing of all the morphology codes and cases is present in the Supplementary file).

We needed to group data to improve resolution and reduce random data noise. We examined infant and child neoplasms separately, but did not find any statistically significant time–trends. Three age-groups seemed reasonable. We chose a child, teenage and young-adult group (0-29), a main middle-age group (30-54) and an older age group (over 55 years of age). These reasonably split the population into three roughly equal (20, 18 and 16 million) groups of people. The case totals in the three groups were about 9.5k, 19.5k and 52k respectively. We tested moving the cut-point boundaries by 5 years in both directions and it made little difference to the overall results.

2.4. Analysis

The cases were analysed by morphology, topology, sex, age, age–specific and age–standardised incidence. The Annual Average Percentage Change (AAPC) and corresponding 95% CI and p–values were calculated using Stata SE12.1 (StataCorp). A linear model on the log of the age–standardised rates, which tests for a constant rate of change (), best fitted the data. See Supplementary File sections S2 and S3.

2.5. Background

In a major 2013 review article, Hiroko Ohgaki and Paul Kleihues [11] wrote “Glioblastoma is the most frequent and malignant brain tumor. The vast majority of glioblastomas (~90%) develop rapidly de novo in elderly patients, without clinical or histologic evidence of a less malignant precursor lesion (primary glioblastomas). Secondary glioblastomas progress from low-grade diffuse astrocytoma or anaplastic astrocytoma. They manifest in younger patients, have a lesser degree of necrosis, are preferentially located in the frontal lobe, and carry a significantly better prognosis.”

Overall primary malignant brain tumour ASRs are only rising slowly and are often considered fairly static. Figure 1 shows the age–standardised trends from 1971 to 2015. From the 1970s to about 2000 there was a fairly steady rise in recorded overall incidence, however since then the rise has slowed, though clinicians have been reporting a rise in high-grade, aggressive tumours.

Figure 1: Age–standardised overall trends from 1971 to 2015 using data in ONS MB1 series, including a smaller number of supratentorial neoplasms without histology or morphology data coded D43.0 & D43.2. The data table for this figure is in the SI file as [S4].

Overall adult survival for all malignant brain tumours after diagnosis during 2006–2010 was about 35% for one year and 15% for five years, falling to about 3% for aggressive grades–III and IV tumours. ONS data show age-standardised death rates from malignant brain tumours (C71) have increased by 7% between 2001 and 2015, showing that improvements in treatment alone are inadequate and that there is a need to find ways of preventing brain cancer.

3. Results

Comparing new case numbers in 2015 with 1995 shows an extra 1548 aggressive GBM tumour cases annually. Figure 2 and Table 3 show that up to about 2004 the overall rise in GBM incidence (Annual Average Percentage Change (AAPC) 5.2%, 95% CI 3.7–6.6, p < 0·00003) could be mostly compensated for by the fall in incidence of all lower grade astrocytoma and “glioma, malignant, NOS, ICD10–93803”. This leaves a fairly steady rise in the GBM ASR from 2004 to 2015 (AAPC 2.2%, 95% CI 1.4–3.0, p < 0.0001).

Table 3: ICD10-C71 and (D43.0 + D43.2) cases and age-standardised (ESP-2013) incidence rates.
Figure 2: Age–standardised incidence rates for all C71 glioma cases diagnosed between 1995 and 2015 analysed by type and year (Data in Table 3). Grouping details: (1) = 94403–94433 (2) = 93843, 94003–94303 (3) = 93803 (4) = 93813, 93823, 93903–93943, 94503–94733.

Ohgaki and Kleihues [11] reported that most secondary GBMs are found in younger middle-age people and most primary GBMs are in over 60s. We tested our (30–54) and (>54) age group data, splitting the total GBM into de novo and promoted tumours. We estimated the maximum possible number of promoted tumours using the change in the grades II and III diffuse and anaplastic astrocytomas. The results are shown in Figures 3(a)and 3(b). These are discussed later.

Figure 3: Age–standardised rates for two age groups. The possible split between de novoand secondary promoted GBMs is based on incidence change of Grades II and III diffuse and anaplastic astrocytoma.

We found a large decrease of ASR over time for Grade–II diffuse astrocytoma, a slight rise in ASR for WHO Grade–III anaplastic astrocytoma (94013; 2832 cases). There was little change in rates of anaplastic oligodendroglioma (94513; 1339 cases), anaplastic ependymoma (93923; 313 cases) Grade–II oligodendroglioma (94503; 2671cases), embryonal, or ependymal tumours.

Figure 4 shows the relative increase in age-specific GBM incidence between the averaged periods (1995–1999) and (2011–2015) for 5–year age–groups. This 1.5-fold change is remarkably similar across the age–groups, suggesting a universal factor.

Figure 4: Relative change in GBM age–specific incidence rates (ASpR) averaged over two five-year periods 1995-1999 and 2011-2015 in 5-year age bands and gender.

Figure 5 shows ASR GBM rates for frontal lobe, temporal lobe, unspecified & overlapping (C71.8 & C71.9) and ‘all other brain regions’. Most of the rise is in the frontal and temporal lobes, and most of the cases are in people over 55 years of age, with a highly statistically significant overall AAPC of 7.6% (see Table 4). There was an extra rise in frontal and temporal GBM incidence between 2006 and 2008, which coincided with a slight reduction in the GBM ASR in overlapping and unspecified regions and may be due to improved imaging.

Table 4: Age standardised incidence rates to ESP-2013 (/100k people).
Figure 5: Frontal and temporal lobe GBM age–standardised incidence rates by tumour site and year (data table in the SI as).

4. Discussion

Using sufficiently high–quality data, we present a clearer picture of the changing pattern in incidence of brain tumour types than any previously published. We report a sustained and highly statistically significant ASR rise in GBM across all ages and throughout the 21 years (1995–2015), which is of importance both for clinical resources and brain tumour aetiology.

Dobes et al. (2011) [13] reported a significant increase in malignant tumour incidence from 2000 to 2008 in the ≥65–year age group. In a second article they noted an increasing incidence of GBM (APC, 3.0; 95% CI, 0.5–5.6) in patients in the same age group, especially in temporal and frontal lobes [14]. De Vocht et al. (2011) [15] reported a rise in temporal lobe tumour incidence in ONS data, but dismissed its significance. In a 2016 paper he claimed no increase in GBM incidence, but later published a major correction to the paper that shows an increase.

Zada et al. (2012) [17] using USA SEER data for 1992–2006 reported a rising trend in frontal and temporal lobe tumours, the majority of which were GBM, with a decreased incidence of tumours across all other anatomical sub–sites. Ho et al. (2014) [18] reported a 2.2–fold increase in glioblastoma incidence in the Netherlands over the period 1989–2010 (APC 3.1, p<0.001).

There were no material classification changes over the analysis period that might explain our findings, though multidisciplinary team working was strengthened (2005 onwards) and better imaging has resulted in improved diagnosis along with a more complete registration of brain tumours in the elderly. We analysed our data in 5-year age group categories to look for evidence of improved diagnosis; the data do suggest diagnosis and registration have improved in people aged over 70. However, at earlier ages the incidence rate of ‘all’ glioma (and all C71) registrations have remained almost constant, whereas the rates for lower–grade tumours fell until about 2006 and have since remained fairly static as the rate for GBM has risen steadily.