Introduction: In England the 5-year survival for colorectal cancer (CRC) is 50%, significantly lower than in other countries of comparable wealth, largely due to patients presenting late with the disease. Late presentation is strongly correlated with emergency presentation to hospital. The NHS Bowel Cancer Screening Programme (BCSP) has been the most significant nationwide intervention attempting to improve CRC outcomes. This programme began in 2006 and was active throughout the country by 2010. Screening programmes are notoriously difficult to evaluate and so far there is limited robust evidence of its effectiveness. Administrative data from hospitals can be used to study diseases, including cancers. Through careful interrogation of the data, it may be possible to identify factors associated with improved CRC outcomes, including the impact of introducing the BCSP. Aims: To develop a new methodology for identifying a national cohort of incident CRC cases, by applying clinical knowledge to the analysis of administrative data for English hospitals (Hospital Episode Statistics, HES). To use this new methodology to identify which admission signifies a patient’s first presentation to hospital care with CRC and identify which patients present as an emergency (Chapter 2). To establish whether emergency presentation is a valid outcome measure for patients with CRC (Chapter 3). To test whether the benefits of launching a BCSP extend beyond the minority of individuals targeted by the screening program, specifically, if there are early, indirect benefits for the population as a whole resulting from enhanced awareness regarding CRC (Chapter 4). To ensure that the findings in Chapter 4 are robust and not accounted for by confounding (Chapter 5). To determine whether the algorithm that dictates which Faecal Occult Blood test (gFOBt) results are referred for colonoscopy can be improved, to more effectively use the limited colonoscopy resources (Chapter 6). Design and method: All hospital admissions relating to incident cases of CRC in England from 2006-2008 were extracted from HES data to form my main cohort. Traditionally HES-based studies mark a patient’s presentation from the first admission containing a coding of CRC (CRC1). In my thesis, all codes indicative of CRC presentation to secondary care were identified and any admission occurring prior to CRC1, flagged as the presenting admission, and termed “first relevant admission” (REL1). The primary outcome measure was emergency presentation with CRC, while secondary outcomes were major surgical resection and mortality at one-year (Chapter 2). Data was analysed to assess the factors associated with, and outcomes related to, an emergency presentation (Chapter 3). I then matched each patient to the date when their local Primary Care Trust (PCT) began referring to the BCSP to ascertain which patients were living in an area with an active BCSP at the time of their presentation (“exposed” group) and the length of exposure. To exclude confounding based on overall healthcare quality in an area, outcomes for a distinct cancer (Oesophageal and Gastric (OG)) presenting during the same period, were compared, based on their screening exposure (Chapter 4). To further exclude confounding, the effect of living in an area with active screening, yet restricted to those PCTs that began screening during the same time period (the middle year) was analysed (Chapter 5). The Bowel Cancer Screening System database was studied to identify individuals with two consecutive episodes of screening between April 2009 and March 2011. Each test kit was coded depending on the specific FOBt result for each round. The overall percentage of positive results was termed the Spot Positivity percentage (SP%). The results at Episode 1 were analysed for individuals with cancer diagnosed at Episode 2. This could identify a combination of FOBt results that would benefit from earlier colonoscopy (Chapter 6). Results: Chapter 2 identifies 32,299 incident cases of CRC. Older patients and females were more likely to be diagnosed with proximal cancer and deprived patients had lower rates of surgical resection and survival. The overall rates of patient characteristics were similar to matched populations that used a different methodology. However the emergency presentation rate was higher in my cohort (36.1% vs. 32.0%, p < 0.001) and the surgical resection and one-year survival rates were lower. Lower gastrointestinal endoscopy is the gold standard when establishing the CRC diagnosis. My new methodological approach of REL1, rather than the CRC1, increased the proportion of patients identified with an endoscopy at presentation (52.5% vs. 36.8%, p < 0.001) suggesting a more accurate start of a patient’s journey was identified. Chapter 3 shows that patients presenting as an emergency were significantly less likely to have surgical resection, (35.4% vs. 59.1%, p < 0.001). Furthermore, the Odds Ratio for mortality at one year following emergency presentation was 4.11 (p < 0.001). This was the single strongest predictor of adverse outcomes. Within the first 30 days of presentation, the mortality rate was almost four times higher in emergency patients (6.5% vs.1.7%, P < 0.001). In Chapter 4 patients “exposed” to local screening had significantly lower crude emergency presentation rates than the “non-exposed” group (34.9% [2,492/7,142] vs. 37.0% [7,599/20,520], p=0.002). Using exposure as a continuous variable, there was a 2% reduction in emergency presentations for every month “exposed” to local screening (OR =0.98 [CI: 0.97-0.99], p < 0.001). By studying the non-screening age population, indirect effect of screening could be tested. Those with greater than 6 months screening exposure had an Odds Ratio (OR) of 0.85 (CI: 0.77-0.94, p < 0.001) for emergency presentation compared to those non-exposed. During the same period, 9,319 patients presented with OG cancer. These patients showed no significant difference in emergency presentation rates between patients “non-exposed” (28.0%), and those “exposed” for less than six months (29.7%) or longer (28.0%). Chapter 5 analyses 48 PCTs that began screening in the middle year and demonstrated the same association with screening exposure as chapter 4. There were more daycase colonoscopies in the six months after screening was introduced (30,347 vs. 31,805). The 4.8% increase in activity, implies either extra capacity and/or extra referrals followed the introduction of screening. Analysis of PCTs that began screening in the first 6 months of the year, showed that prolonged exposure to screening (6-12 months) was associated with a significantly lower emergency presentation rate (34.1 vs. 38.9%, p < 0.001). Chapter 6 examines the data of 284,261 subjects that completed gFOBTs, of which 3,891 (1.4%) had a colonoscopy at Episode 2. As the SP% increased from 11 to 100%, so the CRC detection rate increased from 4 to 25%. At the lower SP%s, from 11% to 25%, the CRC risk was relatively static at 4%. Above an SP% of 25%, every 10-percentage points increase in the SP%, was associated with an increase in cancer detection of 2.5%. Conclusion Clinical knowledge may be applied to routine administrative data (HES) to more accurately identify incident cases of CRC. Using this clinical knowledge creates a better understanding than the traditional methods of when the presenting admission occurred. In particular, patients often present before CRC is recorded in the administrative data, evidenced by the higher rate of endoscopy coded at presentation, using the new methodology. Chapter 2 also demonstrated that more patients than previously thought had presented as an emergency. There is strong evidence to suggest emergency presentation is a valid independent outcomes measure; one strongly associated with reduced access to surgical resection and an increased one-year mortality. Specifically, it is associated with an excess of early deaths after presentation, suggesting an association with the late presentation of cancer. Areas with active screening were significantly associated with a reduced risk of emergency presentation. This reduction extended outside the screening age group, implying an indirect benefit to the population. The reduction became apparent within 6-12 months of the start-up of screening. Longer exposure was incrementally associated with improved CRC outcomes. The outcomes for OG cancer were unaffected by screening exposure, implying the effect was not due to broader differences between local services. There was no obvious confounding factor to the findings in chapter 4 and increasing colonoscopy activity in the immediate six months after screening began, was likely to relate to increased CRC awareness. The BCSP data demonstrated a strong correlation between SP% and cancer detection. Some subjects with an SP% of 11% proceed to colonoscopy, whereas others with an SP% of 22% do not. This suggests that the programme could be adjusted to increase the detection of CRCs without substantially increasing the colonoscopy workload. Overall conclusion My thesis describes how routine administrative data, augmented with clinical knowledge may be used to study CRC. Specifically, it describes how the introduction of the BCSP was associated with an early and population-wide reduction in the risk of emergency presentation. This demonstration of indirect benefits for symptomatic cases of CRC is likely to reflect enhanced public and/or professional awareness of CRC, leading to more timely clinical presentation and investigation in the symptomatic population.