Research

For those who know about my academic education (see CV), it should come as no surprise that my research interest and projects aren’t limited to one topic. This page lists the topics that I am currently working on.

The topic of polygenic risk prediction is the main topic of my research, but was not the first. My PhD project was a psychological study on the perception of prognostic risk in patients who were recently diagnosed with multiple sclerosis. This project investigated patients’ expectations about their future disease and the relation between expectations and psychological well-being, but my thesis had a strong emphasis on how to measure perception of risk, including qualitative validation studies to verify survey responses.

The first project I worked on as a postdoctoral researcher was on the sequential diagnostic tests and logistic regression. I developed a new logistic regression approach for the evaluation of diagnostic test results that could take into account individual patient characteristics (see here). The rationale for the method was that the performance of diagnostic test results changes when they are used as a second test following the results of a screening test.

Polygenic Prediction of Common Diseases

My first paper was a simulation study on the predictive performance of polygenic risk models and tests. The use of simulated data was a necessity at the time because the polymorphisms for common diseases were still unknown–the genome-wide association studies were in the making. I developed a simple simulation algorithm that mimicked the calculation of weighted risk scores that later became the standard, and we modeled various scenarios that differed in the number of polymorphisms, their frequency in the population and their impact on disease risk (in terms of their odds ratio). We showed that a high number of low-impact susceptibility genes were needed to reach appreciable predictive performance, and that a few variants with strong effect would make a positive difference. We later applied this to breast cancer, heart disease, type 2 diabetes, and multiple sclerosis. Not a single empirical study on polygenic risk has refuted this early simulation work since.

The publication of that first simulation study has had major influence on my work. Having seen in simulated data how many polymorphisms would be needed to reach appreciable levels of predictive performance made me less interested in empirical studies. Simulation studies give much more opportunities to investigate the drivers of predictive performance and their impact. That is why my work mostly focused on methodological questions in prediction research.

  1. ACJW Janssens, MC Pardo, EW Steyerberg, CM van Duijn. Revisiting the clinical validity of multiplex genetic testing in complex diseases. Am J Hum Genet 2004;74:585-8. PMCID: PMC1182273.
  2. ACJW Janssens, YS Aulchenko, S Elefante, G Borsboom, EW Steyerberg, CM van Duijn. Predictive testing for complex diseases using multiple genes: fact or fiction? Genet Med 2006;8:395-400. PMID: 16845271.
  3. ACJW Janssens, CM van Duijn. Genome-based prediction of common diseases: advances and prospects. Hum Mol Genet 2008:17;R166-73. PMID: 18852206.
  4. R Mihaescu, MJ Pencina, A Alonso, KL Lunetta, SR Heckbert, EJ Benjamin, ACJW Janssens. Incremental value of rare genetic variants for the prediction of multifactorial diseases. Genome Med 2013; 5:76. PMID: 23961719.

Assessment of Predictive Performance

The first empirical studies on polygenic risk were published right after the discoveries from the genome-wide association studies, in 2007. At that same time, researchers had proposed a new metric for the assessment of predictive performance, the reclassification measures. These measures were adopted instantly because they offered the statistically significant results that the traditional measure, the area under the ROC curve (AUC), couldn’t. We published several studies on the simultaneous assessment of the multiple metrics to investigate the meaning of contradictory findings. The AUC is a vastly misunderstood and misinterpreted metric, and I have written and am working on articles to clarify its interpretation. We proposed the prediction impact curve as a complementary way to add a clinical perspective to the interpretation of the AUC. Part of my current work consists of further analyses on the inference of predictive performance when metrics give discordant results as well as a proposal for a more intuitive interpretation of the AUC.

  1. R Mihaescu, M van Zitteren, M van Hoek, EJ Sijbrands, AG Uitterlinden, JC Witteman, A Hofman, MG Hunink, CM van Duijn, ACJW Janssens. Improvement of risk prediction by genomic profiling: reclassification measures versus the area under the receiver operating characteristic curve. Am J Epidemiol 2010;172: 353-61. PMID: 20562194.
  2. W Campbell, A Ganna, E. Ingelsson, ACJW Janssens. Prediction impact curve was a new graphical approach integrating intervention effects in the evaluation of prediction model utility. J Clin Epidemiol 2015. PMID: 26119889.
  3. ACJW Janssens, JPA Ioannidis, CM van Duijn, J Little, MJ Khoury. Strengthening the reporting of genetic risk prediction studies: the GRIPS statement. Plos Med 2011;8:e1000420. PMID: 21403077.
  4. ACJW Janssens, JPA Ioannidis, S Bedrosian, P Boffetta, SM Dolan, N Dowling, I Fortier, AN Freedman, JM Grimshaw, J Gulcher, M Gwinn, MA Hlatky, H Janes, P Kraft, S Melillo, CJ O’Donnell, MJ Pencina, D Ransohoff, SD Schully, D Seminara, DM Winn, CF Wright, CM van Duijn, J Little, MJ Khoury. Strengthening the reporting of genetic risk prediction studies (GRIPS): explanation and elaboration. J Clin Epid 2011;64:e1-22. PMID: 21431409.

Co-Citation Ranking: a Novel Search Method for Scientific Literature

I recently developed a new method for searching scientific literature. The method, which uses citations rather than keywords to find literature, proved to be highly efficient and effective in finding articles related to one or more known articles. The method has several promising applications, from finding studies for systematic reviews and meta-analysis to finding experts for reviewing manuscripts and grants. We are currently investigating whether the proposed configuration of the method is the most optimal one, as well as investigating its application for more complex search questions.

  1. ACJW Janssens, M Gwinn. Novel citation-based search method for scientific literature: application to meta-analyses. BMC Med Res Methods 2015;15:84. PMID: 26462491.

Critical assessment of science methods

The overarching topic that connects all projects and publications is my interest in research methodology and analysis. Science is not an objective process, but an art and an expertise. Researchers design studies and the choices they make in the methodology and analysis might impact the results and conclusions. Their expertise determines, and limits, the inference of data. Scientific research requires advanced understanding of (statistical) methods to be able to interpret findings in the context of the messy reality of heterogeneity, complexity, misclassification, and other ‘imperfections’ in research data. I regularly reflect on these topics in editorials and commentaries.

  1. ACJW Janssens. Raw personal data: access to inaccuracy. Science 2014;343(6174):968. PMID: 24578562
  2. ACJW Janssens. The hidden harm behind the return of results from personal genome services: A need for rigorous and responsible evaluation. Genet Med 2015;17:621-2. PMID:25412399.
  3. ACJW Janssens, P Deverka. Useless until proven effective – the clinical utility of preemptive pharmacogenetic testing. Clinical Pharmacology and Therapeutics 2014;96:652-4. PMID: 25399713.
  4. ACJW Janssens, P Kraft. Research Conducted Using Data Obtained through Online Communities: Ethical Implications of Methodological Limitations. PLoS Med. 2012;9:e1001328. PMID: 23109913.