|dc.description.abstract||Incidence rates of colorectal cancer (CRC) in New Zealand (NZ) are among the highest in the world. The long-term survival rates in NZ are poor, which has partially been attributed to late diagnosis. Considering that CRC is a curable disease if diagnosed early, conducting research targeted at the improvement of early diagnosis is therefore important. This thesis provides statistical models for the calculation of CRC incidence rates in the entire NZ population and population strata defined by age, gender, ethnicity and diabetes status, and a model for CRC risk in individual patients referred to the secondary care. The models presented here could assist health professionals in the selection of patients for further investigation to facilitate earlier diagnosis. The empirical part consists of three independent observational studies, briefly described below.
Sub-study 1. The objectives were, first, to describe trends in CRC incidence in the NZ population and, second, to investigate whether there are any strata defined by gender and ethnicity with especially increased incidence rates of CRC. To address these objectives, I analysed data from the New Zealand Cancer Registry (years 1994–2018) using an age- period-cohort (APC) model. The overall CRC incidence rates in NZ decreased between 1994 and 2018 by an average of 1.31% per year. However, the decrease was observed only in patients 50 years and older. In those 30-<50 years old, the incidence rates increased between 1994 and 2018 regardless of gender and ethnicity. The increase was similar for proximal, distal and rectal cancers. The APC analyses revealed very strong cohort effects that could explain nearly the entire trends in CRC incidence, pointing out generations born in the 1970s and 80s being affected by the increased incidence rates, rather than individuals 30-<50 years old. The cohort effects were different in Ma ̄ori and non-Ma ̄ori populations. In non-Ma ̄ori born between approximately 1939 and 1955, incidence rates decreased sharply. By contrast, those Ma ̄ori generations have not benefitted from the sharp decrease in rates. However, CRC incidence increased substantially in both M ̄aori and non-M ̄aori groups born in the 1970s and 80s.
Sub-study 2. The objective was to estimate the IRR for CRC in patients with diabetes compared to those without diabetes, with relation to diabetes duration and use of insulin for diabetes control. Registration in the Virtual Diabetes Register (VDR) in the years 2014–2018 was used as a marker of a diabetes diagnosis. Tables with counts of the entire NZ population stratified by age, gender, and ethnicity were obtained from Statistics NZ. In total, data from 310,710 patients with diabetes, corresponding to 1,277,284 person- years and 2512 incident CRC cases were analysed using a Poisson regression model. Diabetes was associated with an overall increased CRC incidence of 13% compared to non-diabetes [IRR=1.13 (95% CI: 1.08, 1.18)]. The IRR was especially high in the first three months after diabetes diagnosis [IRR=2.55 (95% CI 2.02, 3.21)], likely due to detection bias. The association was equally strong in males and females. However, in the analysis by ethnicity, the incidence of CRC was increased only in non-Ma ̄ori patients and restricted only to those younger than 75 years.
Sub-study 3. The objective was to develop a predictive model for CRC risk in individual patients referred to secondary care. To develop such a model, I extracted information from free text included in e-referrals from GPs’ to the Gastroenterology and General Surgery departments in the Waikato Hospital from 2015-2018, including: symptoms; test results; and family history of CRC. The reference test was a full colonoscopy with visualisation of the cecum. Data from 3015 patients, 20-<90 years old were analysed using a logistic regression model. The final model included the following predictors associated with increased CRC risk: anaemia, rectal bleeding, palpable mass in abdomen or rectum, weight loss, age and gender, and a decreased CRC risk: family history of CRC, abdominal pain, and inflammatory bowel disease. The model discriminates patients with low CRC risk well. According to the final cross-validated model, around 20% of patients from our cohort had performed colonoscopy despite a very low CRC risk (less than 1.5%).
In conclusion, the APC analysis revealed an alarming pattern. According to the fitted APC model, the combination of increasing age and cohort effect in generations born in the 1970s and 80s will bring a wave of CRC diagnoses in the near future when the young generations with high CRC incidence rates will replace the old generations with low CRC incidence. The results from this study could therefore help policy-makers to plan the needs for gastroenterology services.
Secondly, CRC incidence rates in diabetes have been found to be slightly increased com- pared to non-diabetes but only in non-Ma ̄ori individuals younger than 75 years.
Third, the results suggest that males underutilise health services. As shown in sub- study 3, males underwent fewer colonoscopies than females, despite having a higher risk of CRC. The higher detection bias in males than in females (sub-study 2) could also suggest underuse of health services by males, but the difference was not statistically significant.
Finally, based on the fitted models for CRC incidence in sub-studies 1 and 2, population- wide CRC screening for Ma ̄ori and patients with diabetes, based on the incidence, instead of age alone, would be proposed to start at age 57.5 years if the screening in the general population starts at age 60 years.||