Improved prediction of complex diseases by common genetic markers: state of the art and further perspectives

Abstract

Reliable risk assessment of frequent, but treatable diseases and disorders has considerable clinical and socio-economic relevance. However, as these conditions usually originate from a complex interplay between genetic and environmental factors, precise prediction remains a considerable challenge. The current progress in genotyping technology has resulted in a substantial increase of knowledge regarding the genetic basis of such diseases and disorders. Consequently, common genetic risk variants are increasingly being included in epidemiological models to improve risk prediction. This work reviews recent high-quality publications targeting the prediction of common complex diseases. To be included in this review, articles had to report both, numerical measures of prediction performance based on traditional (non-genetic) risk factors, as well as measures of prediction performance when adding common genetic variants to the model. Systematic PubMed-based search finally identified 55 eligible studies. These studies were compared with respect to the chosen approach and methodology as well as results and clinical impact. Phenotypes analysed included tumours, diabetes mellitus, and cardiovascular diseases. All studies applied one or more statistical measures reporting on calibration, discrimination, or reclassification to quantify the benefit of including SNPs, but differed substantially regarding the methodological details that were reported. Several examples for improved risk assessments by considering disease-related SNPs were identified. Although the add-on benefit of including SNP genotyping data was mostly moderate, the strategy can be of clinical relevance and may, when being paralleled by an even deeper understanding of disease-related genetics, further explain the development of enhanced predictive and diagnostic strategies for complex diseases.Electronic supplementary materialThe online version of this article (doi:10.1007/s00439-016-1636-z) contains supplementary material, which is available to authorized users.

Polygenic risk score for early prediction of sepsis risk in the polytrauma screening cohort

“…Firstly, our study established a risk prediction model through screening and evaluating genetic susceptibility from previous studies that have high prediction accuracy. Furthermore, genetic variants have several advantages as predictors, including remaining unchanged, predictable life-long risk and easy, accurate and cost-effective measurement [14,37]. In addition, the combining of genetic and clinical factors into one model was feasible in clinical practice for trauma patients, which might enhance the discrimination of patients at high risk for sepsis.…”

A Prediction Algorithm for Drug Response in Patients with Mesial Temporal Lobe Epilepsy Based on Clinical and Genetic Information

“…Although we did not replicate these results in a second cohort, the LOOCV statistical approach is able to minimize this limitation. Indeed, one recent review showed that, among 55 prediction studies, nine used cross-validation approach as independent validation, and only six validated in a second cohort [41]. In addition, it would be interesting to know whether specific SNPs may be associated with AED response for different drugs.…”

“…Several studies have observed an improvement of discriminatory accuracy when combining clinical variables with SNP information [41], including breast cancer(from 0.58 to 0.61) [42], nasopharyngeal carcinoma (from 0.68 to 0.74) [43], and venous thrombosis (from 0.77 to 0.82)[44]. In our study, we demonstrate that by including SNP information in addition to only clinical and imaging parameters, there was a significant increase in the discriminatory accuracy from 0.4568 to 0.8177.…”

ATP2C2andDYX1C1are putative modulators of dyslexia‐related MMR

“…Furthermore, prediction due to the addition of the PRS between good and poor readers was analyzed using different measures of reclassification (IDI and continuous NRI) applying the R add‐on package PredictABEL 1.2‐2 (Kundu, Aulchenko, & Janssens, 2014). Reclassification measures indicate if classification of cases and controls improves when adding new information (e.g., genetics) to the model (Müller et al., 2016). …”