Trends in the contemporary incidence of colorectal cancer and patient characteristics in the United Kingdom: a population-based cohort study using The Health Improvement Network

In Europe, colorectal cancer (CRC) is the third most commonly diagnosed cancer in males and the second most commonly diagnosed cancer in females [1]. In the United Kingdom (UK), approximately 100 new cases of CRC are diagnosed each day [2]. Cancer registry data show that CRC incidence rates in the UK have remained relatively stable for over a decade [3] yet improvement in survival has been small [4]. Furthermore, survival rates in the UK have been poor compared with most other countries in Western and Central Europe [5, 6], Scandinavia [5, 6] and the US [7] with a relatively higher number of excess deaths in the first few months following diagnosis [8, 9] particularly among older age groups [9, 10].

While cancer registries are suitable for determining population-level incidence and survival – including adjustment for age, stage at diagnosis and socio-economic status – data such as the prevalence of comorbidities, medication use before cancer onset and other potential confounders are not systematically recorded. This hinders adjustment for these factors and limits the robustness of comparisons that can be made when making comparisons of CRC survival between populations. In the UK, prospectively collected data on patient comorbidities, prescribed medications and lifestyle factors can be ascertained from primary care databases of electronic medical records (EMRs). Using a validated UK primary care database, we conducted a retrospective cohort study to evaluate trends in the contemporary incidence of CRC in the UK and characteristics preceding CRC diagnosis, including specific comorbidities and medication use. The study protocol was approved by an independent scientific review committee (reference number 14-088A1).

The annual incidence of CRC per 10,000 person-years remained relatively constant across the study period; incidence rates increased from 9.27 (95% CI: 8.59–1.01) in 2000 to 10.65 (95% CI: 10.15–11.18) in 2007, followed by a decreased trend during the later years in the study period, falling to 8.37 (95% CI: 7.93–8.83) in 2014 (Fig. 1a). In men, the incidence of CRC per 10,000 person-years was 11.44 (95% CI: 10.35–12.66) in 2000 and 9.39 (95% CI: 8.74–10.10) in 2014; corresponding rates for women were 7.40 (95% CI: 6.59–8.32) and 7.38 (95% CI: 6.81–8.00). Incidence rates by age at diagnosis across the study period are shown in Fig. 1b for men and Fig. 1c for women. The declining incidence of CRC in the later years of the study period was mainly driven by decreasing rates in older men (aged ≥60 years) from 2012, while overall rates in women remained relatively stable during these study years. As shown in Table 1, incidence rates of CRC were of similar magnitude and followed a similar trend to those reported by the Office for National Statistics (ONS) for the UK over the same study period. Rates were slightly higher in THIN but this could be owing to the fact that our study population was among adults aged 40–89 years while the ONS rates are among all individuals.

In this large population-based study set in a representative primary care setting, we have described the contemporary epidemiology of CRC in the UK and characterized CRC patients at the time of diagnosis in two calendar years more than a decade apart. Few studies have described changes in the comorbidity profile of CRC patients over time [15] and we are unaware of any other study to describe a wide range of patient comorbidities and medication use.

While we have shown that the incidence of CRC has remained relatively stable in the UK over the last 15 years, incidence rates appear to have declined in more recent years, particularly in men aged ≥60 years. Our study also found increased uptake of the National Bowel Screening programme over the study period, which could explain the finding of a slightly earlier mean age at diagnosis in 2014 compared with 2002. The national faecal occult blood test screening programme was rolled out in England in June 2006 and in Scotland in June 2007 with the aim of reducing the number of incident cases through detecting pre-cancerous CRC adenomas. Primary care practitioners involved in the programme were informed of participants who failed to complete the programme, and this may have contributed to an increase in the detection or reporting of new cases that may have previously gone unnoticed – this could explain the peak in CRC incidence rates in this current study in 2007. The incidence rates of CRC in this present study are in line, albeit slightly higher, with those from UK Cancer Registry data [3] and similar age- and gender-specific trends were seen over the study period.

Several factors should ideally be taken into account when analyzing trends in survival of patients with CRC across time and when making comparisons with other geographical populations, including the comorbidity profile and medication use of patients at the time of diagnosis. Several studies have shown that high comorbidity levels are associated with poorer survival [10, 16‐22]. In the National Cancer Data Repository, Downing et al. [10] showed that 24.2% of colon cancer patients who died within the first month following diagnosis had a Charlson comorbidity score of ≥3, compared with 17.6%, 14.2% and 7.2% with a score of two, one or zero, respectively. In a population-based study in Denmark, Erichsen et al. [23] showed that comorbidities interacted with CRC to increase mortality beyond that explained by CRC and comorbidities acting independently, particularly in the first year after CRC diagnosis. In the Netherlands, overall comorbidity among over 27,000 CRC patients in the Eindhoven cancer registry increased from 47% to 62% between 1995 and 2010, with hypertension increasing from 16% to 29% and cardiovascular disease increasing from 12% to 24% [15]. Diabetes, hypertension and atrial fibrillation and obesity, were the main characteristics observed to be more prevalent among CRC cases diagnosed in 2014 than among those diagnosed in 2002. Although these conditions were also more prevalent among non-cases in these two study years, the level of increase was notably higher among CRC cases, especially for diabetes. It is possible that more frequent PCP visits among patients presenting with CRC symptoms, or a higher level of monitoring among high-risk patients, could have led to better recording of other comorbidities and lifestyle factors. However, a genuine greater increase in these conditions among CRC patients could translate into more CRC patients having fewer treatment options and ultimately worse survival. In a large study of older age CRC patients investigating population-attributable risks, a substantial proportion of deaths were attributable to congestive heart failure, diabetes and chronic obstructive pulmonary disease [22]. In our study, a 8.8% increase in the prevalence of diabetes at the time of CRC diagnosis was observed between study years, although only minor changes in the prevalence of heart failure and chronic obstructive pulmonary disease were seen. In relation to medication use at the time of diagnosis, higher increases in the use of PPIs, anti-hypertensives and warfarin among CRC cases compared with non-cases between 2002 and 2014 suggests the growing pill burden among CRC patients, a factor that may also influence treatment options in addition to adherence and outcomes. With an aging general population, non-cancer related health status could play an increasingly important role in the survival of cancer patients, highlighting the importance of data such as those obtained in this study.

Strengths of our study include the large sample size representative of the UK population as a whole; Blak et al. [13] have shown both the demographic of individuals and the prevalence of major medical conditions in THIN to be generalisable to the UK. In addition, previous validation work has shown a high level of completeness and validity of the recorded CRC diagnoses in THIN [14]. Although we did not link to cancer registry data in our prior validation study [14] another study comparing incidence rates in THIN with those in a UK national cancer registry found the age- and sex-standardized incidence rates of CRC to be very similar during the latter part of their study period (2005–2007) [24]. In addition, a study using data from the UK Clinical Practice Research Datalink, which contains very similar primary care data to THIN, reported a 98% PPV for the CRC diagnosis in the primary care data when linked to cancer registrations [25]. A limitation of our study is the possible improvement in recording in THIN over time, resulting in an under-estimation of CRC incidence rates and patient characteristics in the early years of the study period. Indeed, there was a notable improvement in the proportion of both CRC cases and non-cases with a recorded BMI, smoking status or alcohol intake between 2002 and 2014. Prior to 2003, PCPs contributing data to THIN did not have specific recording instructions, although any improvements in recording will likely have affected CRC cases and non-cases equally. The year 2003 also saw the introduction of a cancer quality improvement measures in the UK by the NHS, which may not only have improved recording of CRC diagnoses but also of other morbidities and patient characteristics. This would have led to a more complete profile in the patient’s records.

Advanced stage at CRC diagnosis is clearly associated with worse survival [26, 27], and accounts for the highest proportion of early colon cancer deaths; Downing et al. [10] reported 15.4% of colon cancer deaths in the month following diagnosis to be Duke’s Stage D, compared with approximately 5% or less for less advanced stages of the disease. There was insufficient information in our dataset to enable us to ascertain Duke’s stage of CRC cases at diagnosis and thereby describe any potential changes over time in the proportion of CRC cases diagnosed at advanced stage. Failing to reduce the proportion of CRC cases diagnosed at advanced stage compared with other countries could be a possible explanation for the lower relative survival rates and small improvements in survival rates over the last decade. Diagnostic codes in THIN are not specific for Duke’s stage CRC; rather information on CRC histology and stage is entered into patients’ medical notes as free-text comments, which were not obtained in this study. We have, however, previously evaluated CRC stage at diagnosis in another study in THIN [28], which included 3033 incident cases of CRC diagnosed between 2000 and 2011 with Duke’s stage ascertained through manual review of patients EMRs including the free-text comments. We found that Duke’s stage D accounted for 12.7% of CRC cases diagnosed in 2002 and 16.3% of CRC cases in 2011, whereas reductions were seen in the proportion of CRC cases diagnosed at Duke’s stages A to C between these two calendar years (6.7% to 4.1% for Duke’s stage A; 14.4% to 9.2% for Duke’s stage B; 11.0% to 8.9% for Duke’s stage C; unpublished data). It should be noted that 55.1% of the CRC cases in 2002 and 61.4% of CRC cases in 2011 did not have details on stage recorded in the free text.

In conclusion, our findings suggest that the increased prevalence of obesity, hypertension, atrial fibrillation and the pill burden among CRC cases should be considered when evaluating and making comparisons in patterns of CRC survival.