The genome is particularly useful for assessing inherited disease risk and the modulation of drug response. To improve the precision of diagnosis and treatment for personalized medicine, multiple sources of information, including genomic information, will likely need to be combined (Altman, 2013).
However, to be successful in personalized medicine, the integration of genomic information into the clinic will also need to involve integration of a broader set of population-based data sources (Altman, 2013). This will allow for the use of Bayesian reasoning to estimate the a posteriori probability of a clinical event as a function of the a priori probability of that event as well as information contained in any newly measured data.
For example, to estimate the probability a drug works given genetic data for a particular patient, such as a particular single nucleotide polymorphism or SNP, based on Bayesian reasoning, clinicians may simply apply Bayes rule:
P(drug works | has SNP) = (P(has SNP | drug works)*P(drug works))/P( has SNP) where
P(has SNP | drug works) and
P(drug works) are estimated from previous genomic and clinical data and
P(has SNP) is estimated from population statistics.
Therefore, genomic data with population data may be used to help estimate the a posteriori probability of disease, drug response, surgical outcomes, etc. and both are critical for the implementation of personalized medicine in a clinical setting.
The electronic health records (EHRs) can be adapted to enable said integration of both genomic information and a broader set of population-based data sources into the clinic and ultimately bring such application of Bayesian reasoning into clinical practice at the point of care with clinical decision support (Ullman-Cullere & Mathew). However, prior to such application of Bayesian reasoning can be made possible, several technical challenges must be addressed to enable the development of such a genomics aware EHR.
In particular, we need to develop a coherent, consistent, and uniform naming convention for genetic data, such as the reference to a particular SNP of interest (Chute & Kohane, 2013). Else, the accuracy with which we may estimate the probability a clinical event given genetic data will be highly contingent on any biases due to incomplete representation of genetic data. The development of such a naming convention will allow users to query the EHR more easily and effectively. Similarly, the same coherent, consistent, and uniform naming convention for clinical events must be developed for accurate clinical interpretation (Ullman-Cullere & Mathew).
Given the vast quantity of particular genetic data that may influence clinical events, authenticated, well-annotated, curated, and freely accessible knowledge base of genomic associations, risks, and warnings must be developed to reduce the search space (Chute & Kohane, 2013). Thus, clinicians will be able to issue tests for and focus on particular genetic data that may most likely influence clinical events for a particular patient.
Even beyond these technical hurdles, the EHR carries numerous inherent challenges. Data are largely missing and frequently inaccurate (Hripcsak & Albers, 2013). Particularly for Bayesian reasoning on rare SNPs where P(has SNP) and P(has SNP | drug works) will be very small, Bayesian reasoning will not be sufficiently robust to cancel out noise from inaccuracies or biases due to missing or inaccurate data. Similarly, data may be systemically error prone due to billing standards, resulting in multiple clinical events or genetic data being grouped together when they should not be, further complicating Bayesian reasoning (Hripcsak & Albers, 2013).
Additionally, modular, standards-based decision-support rules must be developed for easy integration into current EHR environments (Chute & Kohane, 2013). Clinical organizations must also decide whether the offered decision-support rules from Bayesian reasoning should supplement or override existing clinical decision support rules (Chute & Kohane, 2013). Various legal and liability issues will be associated with either option and must be addressed accordingly.
It is also important to note that clinicians will not only be using the decision-support rules offered from Bayesian reasoning but will also contribute to the continuing collection, documentation, and interpretation of risk assessments (Scheuner et al., 2008). Collection and insertion of genetic and genomic information into the clinical workflow will need to be streamlined (Ullman-Cullere & Mathew). This continuing maintenance will remain an inherent challenge to any EHR and likely progressively intensify as healthcare becomes more complex (Hripcsak & Albers, 2013).
Furthermore, social challenges stand as potentially the biggest barrier towards the adoption of a genomics aware EHR. Recent reviews suggest that clinicians are not prepared to integrate genetic information into routine clinical practice (Scheuner et al., 2008). Similarly, recent reviews suggest that clinicians are uncomfortable with recommending risk-specific interventions and even offering genetic tests (Scheuner et al., 2008). Thus, a clinician may not even use such a genomics aware EHR because they do not have any particular genetic data from patients to query or the desire to use the resulting information. Despite the potential utility of a genomics aware EHR, there is still a limited demand from within the medical community in the United States (Ullman-Cullere & Mathew). That said, while overcoming technical barriers will be necessary to develop such a genomics aware EHR, overcoming social barriers will be the key to ultimately bringing the genomics aware EHR into clinical practice.
Still, in spite of all these challenges, the genomics aware EHR remains a goal worth striving towards. Data from a genomics aware EHR provide numerous potential for primary as well as secondary uses and may lead to discoveries that improve the understanding of biology, aid in the diagnosis and treatment of disease, and improve patient outcome and reduce healthcare costs through targeted therapy, risk reduction, and personalized care (Hripcsak & Albers, 2013; Ullman-Cullere & Mathew). Through continuing and continuous efforts to address barriers, both technical and inherent, social and political, forseen and unexpected, this long-term vision can be achieved.
Ullman-Cullere MH, Mathew JP. 2011. Emerging landscape of genomics in the Electronic Health Record for personalized medicine. Human mutation 32(5):512-6.