Personalized medicine (or precision or stratified medicine as new terms introduced nowadays) is recently defined as: “Treatments targeted to the needs of individual patients on the basis of genetic, biomarker, phenotypic, or psychosocial characteristics that distinguish a given patient from other patients with similar clinical presentations”. To reach those high goals almost all medical fields need paradigm shift in both research and clinical practice. Health economics and outcome research filed is not exception from this rule.
Traditional approach to developing decision analytic model to inform decision making process under uncertainty conditions is not sufficient for the personalized medicine technologies tailored for the individual patients. The weakest point is the limited possibilities of traditional clinical trials to generate big datasets to distinguish between distinct patient strata with sufficient precision accurately. However big healthcare data are or so-called real world data sources, as electronic medical record databases and claims-based databases, with adequate completeness, appropriate linkages, and robust statistical methods to drown sound conclusions, can inform the decision analytic model with proper inputs. Yet, even big data will not be sufficient when highly individualized risk prediction is required, in conditions with high heterogeneity of patient’s response, non-linear relationships, and the presence of many predictors. Therefore, next generation of the analytic prediction tools utilizing machine learning (ML) methods is rather necessity then advantage. Utilizing class of ML methods called supervised learning, when the outcome is known, and the aim is to learn covariate patterns that predict the outcomes is already confirms the superior classification of the “responders,” based on biomarkers values, outperforming any traditional regression based approach.
After those initial steps to secure “clean signal” from big data to inform the model, instead of development of traditional decision analytic models, new type of probabilistic graphical models called Markov influence diagrams, already commonly used in machine learning, offers entirely new class of solutions able to cope with impelling of personalized medicine technologies. Markov random field over causal graphs offer possibilities to “carry” several variables per cycle gives the opportunity to model spectrum of patient characteristics, if not complete patient history, avoiding state explosion, tunnel states and need for the micro-simulations approaches, envisioned as a necessity in the personalized medicine.
As personalized medicine technologies were envisioned as a tailored approach to meet the needs of the individual patients, described methodological innovations were tailored to meet the methodological needs for the personalized medicine approach.
Synergus is dedicated to continues analytical education and target recruiting to offer to clients cutting age analytics methods to break the barriers of adopting new technologies and secure market access. Therefore, we are proud to announce new artificial intelligence techniques as a part of our standard analytical portfolio for the decision analytic modeling of personalized medicine technologies.