Search

BACKGROUND: Colorectal cancer (CRC) was the second commonest cancer and the third leading cause of cancer-related deaths worldwide in 2018. In the UK, the overall 5-year survival rate of CRC patients is approximately 60%. Colorectal cancer patients are staged based on the staging system recommended by the American Joint Committee on Cancer (AJCC). The 5-year survival rates vary from approximately 90% for stage I to 10% for stage IV CRC patients. Although the AJCC stage is the main indicator of patients’ prognosis, there is still substantial variation in terms of the survival outcomes of CRC patients within each stage. This merits further examination of other prognostic factors to improve prediction of CRC survival. Previous evidence revealed that germline genetic background plays an important role in determining survival outcomes of CRC patients. However, the human germline genome consists of millions of genetic variants and no specific genetic loci have been robustly mapped in relation to prognosis of CRC patients to date. Firstly, this thesis seeks to systematically review existing literature and explore whether germline genetic variants have been adopted in published multivariable models in attempts to predict CRC survival. Secondly, multiple CRC patient cohorts were leveraged to investigate associations between germline genetic variants and survival outcomes of CRC patients after diagnosis. METHODS: A systematic literature search was conducted in MEDLINE and Embase databases to retrieve published multivariable prediction models that were developed to forecast survival outcomes of CRC. Risk of bias for included models was assessed using published evaluation tools and metrics evaluating model performance were extracted and quantitatively assessed using meta-analysis. Multiple study cohorts were used in this thesis including the Study of Colorectal Cancer in Scotland (SOCCS), incident CRC cases from the UK Biobank cohort and datasets from three previously published clinical trials (QUASAR2, SCOT and VICTOR). Firstly, germline genetic variants associated with CRC survival that were reported by published genome-wide association studies (GWAS) were identified by searching the NHGRI-EBI GWAS catalogue. Associations between these variants and overall and CRC-specific survival were investigated as a replication study using the SOCCS cohort. Then I explored the potential predictive value of these previously reported variants in the UK Biobank study by developing a genetic predictor combining these variants, and evaluated the predictive performance of the predictor along with other variables (age at diagnosis, sex, AJCC stage and tumour grade) using the SOCCS as an external validation cohort. The model performance was assessed in terms of the discriminative ability and model calibration. The next step was to conduct two candidate genetic association studies to test the potential effects of two groups of genetic variants—variants associated with CRC risk and variants associated with prognosis of other cancers—on survival outcomes of CRC patients from the SOCCS study. These two groups of variants were identified from two large GWAS meta-analyses and the GWAS catalogue. Stratified analyses were performed by sex, AJCC stage (stage II/III and IV) and tumour site (colon and rectum). Cox regression models were used to estimate effects—hazard ratios (HRs)--of genetic variants on survival outcomes with age at diagnosis, sex and AJCC stage as covariates. The false discovery rate (FDR) approach was used to correct for multiple testing. Genetic effects were tested under both the additive and recessive genetic models. Finally, I performed a GWAS on both overall and CRC-specific survival by investigating a total of overall eight million autosomal genetic variants throughout the genome using the SOCCS study. The effect estimates for each variant were obtained using a Martingale-residual based approach. Discoveries of the GWAS were then replicated by performing meta-analysis combining effect estimates from the UK Biobank cohort and the three clinical trials. Stratified GWASs were also conducted in SOCCS for stage II/III and stage IV CRC patients separately. Enrichment analyses were employed to detect potential genomic signals enriched in possible genes and gene-sets that are involved in relevant biological pathways. RESULTS: The systematic literature review identified 83 original prediction models and 52 separate external validation studies. Five models (Basingstoke score, Fong score, Nordinger score, Peritoneal Surface Disease Severity Score and Valentini nomogram) were validated in at least two external datasets and showed positive discriminative ability in terms of model performance. No germline genetic variants had been used as prognostic predictors in published prediction models. A total of 5,675 CRC patients from the SOCCS cohort, 2,474 incident CRC cases from the UK Biobank cohort and 4,771 CRC patients from the three clinical trials were included in the main analysis. By searching the GWAS catalogue, I identified 43 independent genetic variants (r2 < 0.2) that were previously linked with CRC survival outcomes. After correcting for FDR, none of these 43 variants, under the additive genetic model, were significantly associated with either overall or CRC-specific survival of CRC patients from the SOCCS cohort. Only three variants (rs17026425, rs17057166 and rs6854845) at nominal significance (unadjusted p < 0.05) showed concordant direction of effects with previously published GWASs, whereas one variant with uncorrected p < 0.05 showed opposite direction of effect (rs11138220). The polygenic risk score (PRS) combining the 43 variants was not associated with CRC survival outcomes. No significant associations after adjusting for FDR were found in the stratified analysis. Although four variants (rs17280262, rs16867335, rs6854845 and rs17057166) showed potential effects when the recessive model of inheritance was used in SOCCS, I failed to replicate these effects using data from the UK Biobank cohort. With respect to the predictive performance of the 43 variants in the UK Biobank cohort, the genetic predictor combining the 43 variants did not show statistically significant C statistics after internal validation, with the 95% confidence intervals (CIs) including the null (overall survival: C=0.510, 95%CI=0.498-0.521; CRC-specific survival: C=0.518, 95%CI=0.498-0.530). Similarly, non-significant C statistics were observed for the 43- variant predictor in the external validation analysis using the SOCCS cohort. Moreover, the prediction model composed of the 43 variants was poorly calibrated in both the UK Biobank and the SOCCS cohorts. The model performance remained nearly unchanged when combining the genetic predictor with other variables including age at diagnosis, sex, AJCC stage and tumour grade in the SOCCS cohort, suggesting no incremental predictive value had been introduced by the addition of genetic variants. Regarding the other two groups of candidate genetic variants, a total of 128 independent variants (r2 < 0.2) associated with CRC risk and 82 independent variants (r2 < 0.2) associated with survival outcomes of other cancers were included. Overall, none of the variants were observed in statistically significant associations (after FDR correction) with CRC survival under the additive model using the SOCCS cohort. The CRC-risk PRS was not significantly associated with either overall or CRC-specific survival. Stratified analysis did not identify any significant associations after correcting for FDR. Three CRC-risk variants (rs10161980, rs9537521 and rs7495132) showed significant genetic effects (recessive model after FDR correction) on survival outcomes of CRC patients from the SOCCS, and a significant association between the TT genotype of the variant rs7495132 and CRC-specific survival was also observed in the UK Biobank cohort (HR=1.69, 95%CI=1.03-2.79, p=0.038).

This thesis studies the institutional investor background in order to understand the working of hedge fund activism: how institutional investors affect hedge fund activists target selection and how activists share information and build alliances through social connections to achieve their goals. Chapter 2 utilizes a rich literature on institutional investors' governance roles and develops simple measures of institutional discontent expressed through holding, trading and voice channels, to predict hedge fund activism target selection. Discontent expressed through all three channels leads to subsequent targeting. Medium sized dissatisfied owners and sellers seem to be the main driving force, and institutions' discretionary disagreements on management compensation and governance related proposals have the highest explanatory power among other voice channels. Activists are more likely to gain higher announcement returns and threaten to take hostile actions against management with more discontented institutional investors in the target companies. Discontented institutions are more likely to vote pro-activist in the subsequent annual meetings after campaigns. Chapter 3 uses a social network framework to study information dissemination during activist campaigns. Actively managed funds whose managers are socially connected to the lead activist are more likely to increase their ownership in the target firms around the activist disclosure. In the cross sectional analysis, we find that the effect is stronger if the activists have better track records and if the ties are established via club membership, charity works, and other small circles. Connected institutions also earn significantly higher announcement returns relative to non-connected funds. The presence of connected institutions contributes to the activist's campaign success. Additional tests are performed to rule out alternative explanations such as fund manager ability or similarity in portfolio choices. Chapter 4 goes one step further to study alliance building among activist investors and institutional investors during the campaign period. A socially connected institution is 1.1 percentage points more likely to increase its ownership in the target firm during the campaign period, compared to funds that are not socially connected to the activist. We use a subsample that includes all institutions subject to M&As before activism events to identify plausibly exogenous shocks to social connections and find similar results. Furthermore, connected institutions also perform significantly better on their investments than non-connected institutions and they are more likely to vote pro-activist in routine proposals, especially director election proposals. The effect is stronger if connected institutions also purchase target stocks during a campaign. The thesis contributes to the literature by developing measures of revealed institutional governance preference based on theoretical and survey evidence in the literature. It also uncovers a channel through which hedge fund activists share information and build alliances and push for corporate changes facilitated by mutual benefits amongst their fellow institutional allies.