Publications

Company blog post

https://www.geneia.com/blog/2020/january/interpretability-and-the-promise-of-healthcare-ai

Academic publications

Medical Concept Representation Learning from Claims Data and Application to Health Plan Payment Risk Adjustment. Zhong QY, Fairless AH, McCammon JM, Rahmanian F. arXiv:1907.0660.

Risk adjustment has become an increasingly important tool in healthcare. It has been extensively applied to payment adjustment for health plans to reflect the expected cost of providing coverage for members. Risk adjustment models are typically estimated using linear regression, which does not fully exploit the information in claims data. Moreover, the development of such linear regression models requires substantial domain expert knowledge and computational effort for data preprocessing. In this paper, we propose a novel approach for risk adjustment that uses semantic embeddings to represent patient medical histories. Embeddings efficiently represent medical concepts learned from diagnostic, procedure, and prescription codes in patients’ medical histories. This approach substantially reduces the need for feature engineering. Our results show that models using embeddings had better performance than a commercial risk adjustment model on the task of prospective risk score prediction.

---

Publications at Pubmed

Development of home cage social behaviors in BALB/cJ vs. C57BL/6J mice. Fairless AH, Katz JM, Vijayvargiya N, Dow HC, Kreibich AS, Berrettini WH, Abel T, Brodkin ES. Behav Brain Res. 2013 Jan 15;237:338-47. doi: 10.1016/j.bbr.2012.08.051. Epub 2012 Sep 5. PMID: 22982070

• Hierarchical linear mixed effects models showed that one mouse strain (C57BL/6J) huddled together (were passively sociable) in their home cages more than the other strain (BALB/cJ) before adulthood, but their time spent huddling decreased by adulthood
• BALB/cJ mice reared more than C57BL/6J mice, while C57BL/6J mice groomed themselves more than BALB/cJ mice did
• Home cage social behaviors between littermates did not correlate with sociability towards a stranger mouse in a novel environment

Sociability and brain development in BALB/cJ and C57BL/6J mice. Fairless AH, Dow HC, Kreibich AS, Torre M, Kuruvilla M, Gordon E, Morton EA, Tan J, Berrettini WH, Li H, Abel T, Brodkin ES. Behav Brain Res. 2012 Mar 17;228(2):299-310. doi: 10.1016/j.bbr.2011.12.001. Epub 2011 Dec 9. PMID: 22178318

• Robust (from Wilcox 2005) and classical regression showed that the sociability of one mouse strain (BALB/cJ) increased across development;
• Robust and classical ANOVAs (and classical ANCOVAs) showed that one mouse strain (C57BL/6J) was more sociable than the other (BALB/cJ) before adulthood
• Linear mixed effects models compared using Akaike’s Information Criterion showed that mouse sociability clustered by litter, so that littermates resembled each other in sociability more than non-littermates
• This litter clustering was partly accounted for by perinatal litter size and sex ratio (classical ANCOVAs and robust regression)
• Outlier detection and robust regression identified mice with an unusually small brain structure (corpus callosum), but this abnormality did not correlate with sociability
• The brain weights of the mouse strains differed before adulthood (robust and classical ANOVAs)

Deconstructing sociability, an autism-relevant phenotype, in mouse models. Fairless AH, Shah RY, Guthrie AJ, Li H, Brodkin ES. Anat Rec (Hoboken). 2011 Oct;294(10):1713-25. doi: 10.1002/ar.21318. Epub 2011 Sep 8. PMID: 21905241

• Correlations (Pearson’s r) quantified the test-retest reliability and the ecological validity of a behavioral test measuring mouse sociability (after controlling for mouse strain, sex, and age)
• Comparisons of the correlations (using methods from Steiger 1980) showed that one sociability measurement (cylinder scores) showed higher reliability and validity than another measurement (chamber scores)
• Separately, intracluster correlations showed that automated measurement of cylinder scores was as accurate at measurement by human raters

Low sociability is associated with reduced size of the corpus callosum in the BALB/cJ inbred mouse strain. Fairless AH, Dow HC, Toledo MM, Malkus KA, Edelmann M, Li H, Talbot K, Arnold SE, Abel T, Brodkin ES. Brain Res. 2008 Sep 16;1230:211-7. doi: 10.1016/j.brainres.2008.07.025. Epub 2008 Jul 15. PMID: 18662677

Association between sociability and diffusion tensor imaging in BALB/cJ mice. Kim S, Pickup S, Fairless AH, Ittyerah R, Dow HC, Abel T, Brodkin ES, Poptani H. NMR Biomed. 2012 Jan;25(1):104-12. doi: 10.1002/nbm.1722. Epub 2011 May 25. PMID: 21618305

Longitudinal in-vivo diffusion tensor imaging for assessing brain developmental changes in BALB/cJ mice, a model of reduced sociability relevant to autism. Kumar M, Kim S, Pickup S, Chen R, Fairless AH, Ittyerah R, Abel T, Brodkin ES, Poptani H. Brain Res. 2012 May 21;1455:56-67. doi: 10.1016/j.brainres.2012.03.041. Epub 2012 Mar 25. PMID: 22513103

Loss of CDKL5 disrupts kinome profile and event-related potentials leading to autistic-like phenotypes in mice. Wang IT, Allen M, Goffin D, Zhu X, Fairless AH, Brodkin ES, Siegel SJ, Marsh ED, Blendy JA, Zhou Z. Proc Natl Acad Sci U S A. 2012 Dec 26;109(52):21516-21. doi: 10.1073/pnas.1216988110. Epub 2012 Dec 10. PMID: 23236174

The mitigating effects of suspicion on post-identification feedback and on retrospective eyewitness memory. Neuschatz JS, Lawson DS, Fairless AH, Powers RA, Neuschatz JS, Goodsell CA, Toglia MP. Law Hum Behav. 2007 Jun;31(3):231-47. Epub 2007 Jan 26. PMID: 17253155