DESIGN: A prospective, observational study.

SETTING: A single-center study.

PARTICIPANTS: 200 patients undergoing vascular surgery under general anesthesia.

INTERVENTIONS: Preoperative (resting- and nighttime), intraoperative, and postoperative blood pressure measurements were performed.

MEASUREMENTS AND MAIN RESULTS: Intraoperative and postoperative hypotension were defined as time with mean arterial pressure (MAP) < 65 mmHg and the area from the MAP curve up to MAP 65 mmHg (mmHg × min). The burden of postoperative complications was quantified using the Comprehensive Complication Index 30 days after surgery. A hurdle statistical model was used due to the zero-inflated outcome data. In multivariable analyses every minute with MAP < 65 mmHg during anesthesia and in the postanesthesia care unit was associated with a 0.15- and 0.09-point increase in Comprehensive Complication Index at 30 days, respectively. Preoperative resting- and nighttime blood pressures did not significantly modify the associations between intra- or postoperative hypotension and postoperative complications.

CONCLUSIONS: Intraoperative and postoperative hypotension were significantly associated with postoperative complications in vascular surgery patients. There were no statistically significant interactions between these associations and preoperative nighttime blood pressure, or ambulatory resting blood pressure.

Papers 1 and 2 were motivated by the hypothesis that defects of the immune system may increase risk of psychiatric disorders. We consider two components from the lectin pathway of activation: mannan-binding lectin (MBL) and MBL-associated serine protease-2 (MASP-2) via continuous traits (protein level), dichotomous trait (disease status) as well as genetic markers including GxG interactions. We use Tobit regression to handle data below the detection limit of MBL.

The involvement of the immune system may also be less direct as seen by the findings how infections impact disorders, e.g. via interaction between genes and maternal infection by virus. Paper 3 presents the initial steps (mainly data construction) of an ongoing simulation study aiming at guiding decisions by comparing methods for GxE interaction analysis including both traditional two-step logistic regression, exhaustive searches using efficient algorithms, and data mining or machine learning methods like model-based multifactor dimensionality reduction (MB-MDR) and logic regression with feature selection (logicFS).

The analysis of sex chromosomes may require different approaches than those commonly used for autosomes. In paper 4 we include a marker from the X chromosome and discuss how to analyse with and without the assumption of inactivation of one of the female X chromosomes early in development. In addition this paper includes analysis of the interaction between genetic markers and age and sex.

Haplotype analysis and other multilocus approaches may increase the power to detect disease association but introduce also the problem of determining the gametic phase. In papers 1 and 2 we analyse multilocus genotypes and haplotypes but assuming known phase as linkage disequilibrium (LD) implies only few haplotypes to be commonly observed using these markers. However, the validity of the identified haplotypes is also checked by inferring phased haplotypes from genotypes. Haplotype analysis is also used in paper 5 which is otherwise an example of a focused approach to narrow down a previously found signal to search for more precise positions of disease causing mutations and functional implications.

In stark contrast to such a focused approach stand genome-wide studies (GWAS). Here it is truly important to address the enormous increase in type I error introduced when performing hundreds of thousands or even millions of statistical tests. The commonly accepted genome-wide threshold for single-marker association tests has become 5e-8 but searching for interactions genome-wide results in drastically many more tests and thus the need of an even lower p-value threshold. Lowering the threshold comes at the unfortunate but inevitable expense of increasing the probability of type II errors and thus lowering the power to detect association. Statistical procedures where the test statistics initially are grouped according to some criteria, e.g. by candidate regions or functional pathways, may be one way to decrease the number of tests instead of lowering the threshold for significance. Yet, in paper 6 we propose the Landscape method to summarise a series of sequentially ordered test values without the need of more or less arbitrary prior grouping.