Is Medicine Precise Enough to Achieve Precision Medicine?

One of the aspects of medicine that struck me as a medical student was its imprecision. I was surprised, sometimes shocked, at decisions that were made based on vague symptoms reported by patients, ambiguous findings detected on physical examination, and even variation in "hard" measurements such as laboratory results. An area of even more imprecision was the data in the patient record, which seemed to matter less when it was scribbled on paper but takes on a whole lot of importance more now that the data is electronic form and advocated for "secondary use."

It is against this backdrop that I view medicine entering the era of "precision medicine" [1]. The vision and potential for precision medicine is compelling and exciting. The notion that we can unravel the myriad of details of a patient's health or disease, and treat the latter more precisely, would be a genuine advance. The charge for precision medicine is being led by the new NIH $215 million Precision Medicine Initiative launched by President Obama.

As a researcher, especially one with an interest in the secondary use of the growing amount of clinical data [2,3], particularly from the electronic health record (EHR), I am naturally excited about contributing to the advance of precision medicine. But as an educator, I usually try to step back and take a more holistic view. We must evaluate not only the specific components of precision medicine, but also the general paradigm. I believe there are challenges for both.

Focusing first on components of precision medicine, let us look, for example, at an area like genomics. Although I find genomics very intellectually fascinating, its impact on patient outcomes has been modest [4]. While our ability to sequence genomes and measure their expression continues to improve while costs fall at a rate exceeding Moore's Law for computers, only a modest amount of what we can measure has been "clinically actionable." Furthermore, although we tend to think of gene sequencing as very precise, it turns out that it too has imprecision. Last year, a study of two commercial whole genome sequencing platforms found that medians of 9-17% of 56 genes recently identified as having potentially high clinical importance were not covered by sufficient numbers of repeated sequencing reads to achieve clinical grade variant detection [5]. While whole genome sequencing is likely to improve, and it is not the only way to assess genomic variation, these data show that even gene sequencing can be imprecise.

Another specific area of challenge is clinical data, whose imprecision has also been long known. In 2013 I authored a paper on its "caveats" [2], and last year I recounted a situation where getting data back to its native form would be like unscrambling eggs. Just recently I heard an overview about our institutional plans for precision medicine, and walking away from a meeting with a clinical colleague, she was lamenting how the switch-over to ICD-10 for coding diagnoses on radiology ordering had just become a whole lot harder at our hospital because of the vastly increased number of codes. Her residents were overwhelmed by the choices, so often sought out the "not otherwise specified" code, which of course was often not the correct one to choose.

Also a concern about the components of precision medicine is how we will figure out what works. Although a proponent of the evidence-based medicine (EBM) approach, I am well aware of the limits of EBM that homogenize patients into large groups in order to determine an effect of a test or treatment. The nature of "best evidence" studies often glosses over individual differences. This provides a benefit in allowing statistical analysis to discern bias and chance from truth, but at the cost of ignoring personal differences. In precision medicine, when each individual is unique, how will we be able to experimentally compare different diagnostic tests and precision-based treatments?

I also believe that we will need to validate the paradigm of precision medicine. Indeed, this may be a way to overcome some of the EBM-related challenges, in that we may be able to apply experimentation to the precision medicine approach rather than any particular (individualized) therapy. Although hopefully there will be some tests and treatments with widespread enough use to conduct clinical trials.

In any case, the era of precision medicine portends an interesting and likely highly beneficial approach to medicine. The role of informatics will be widespread and important. Many of the issues that plague informatics, especially clinical data, currently (e.g., lack of standards and interoperability, ability to aggregate across healthcare systems, need to integrate with genomics and other bimolecular data) will need to be solved for informatics to make its optimal contribution.

References

1. Collins, FS and Varmus, H (2015). A new initiative on precision medicine. New England Journal of Medicine. 372: 793-795.
2. Hersh, WR, Weiner, MG, et al. (2013). Caveats for the use of operational electronic health record data in comparative effectiveness research. Medical Care. 51(Suppl 3): S30-S37.
3. Hersh, WR, Cimino, JJ, et al. (2013). Recommendations for the use of operational electronic health record data in comparative effectiveness research. eGEMs (Generating Evidence & Methods to improve patient outcomes). 1: 14. http://repository.academyhealth.org/egems/vol1/iss1/14/.
4. Green, RC, Berg, JS, et al. (2013). ACMG recommendations for reporting of incidental findings in clinical exome and genome sequencing. Genetics in Medicine. 15: 565-574.
5. Dewey, FE, Grove, ME, et al. (2014). Clinical interpretation and implications of whole-genome sequencing. Journal of the American Medical Association. 311: 1035-1044.

Semantic Drift and the Persistence of Informatics

Being concerned with representation of information and knowledge, researchers in informatics sometimes express concern with the concept of "semantic drift," where the meaning of words and concepts changes over time. Semantic drift happens for a variety of reasons, most commonly due to advancing and changing knowledge of health and biomedicine. Another type of semantic drift occurs in many industries, including the information technology (IT) industry, where new terms come along reflecting evolution in technology, although sometimes the new terms are just a different name for a similar or sometimes the same thing. Not infrequently, the new terminology reflects marketing and hype as much as substantive change.

Some terms withstand the test of time, and I am pleased to note that "informatics" fits into that category. The word traces its origins back to the 1960s, and the importance of the discipline has withstood the test of time. As with all fields, the leading edge has changed substantially, but the core function and definition of the field - the use of data, information, and knowledge to improve human health - has not.

Like many fields, informatics has seen the emergence of areas of work that overlap with its work, in essence that provide semantic drift not only from the core definition of informatics but also the description of work that rightfully belongs to it. I am referring to some of the emerging "hot topics" in recent years, such as data science, data analytics, and precision medicine. I suspect that some may argue these are different from informatics, but I would rebut that they really fit under the broad umbrella of informatics.

I also believe these new sub-disciplines need to prove their work, just as informatics has (or in some cases has not). Like most established disciplines, informatics has a long trail of science. Not all of it is strong methodologically, particularly the portion that evaluates systems in the real world. But we can point to techniques and implementations that have been studied enough to demonstrate where they do and do not work [1-4]. Informatics also provides a good deal of experience and perspective in having tried to address some of what these new sub-disciplines are trying to accomplish.

The current hot topic is precision medicine [5-6]. While I share the excitement and recognize its potential, I also know that it is still an unproven science. In other words, there are still few "products" of precision medicine that demonstrated any large-scale success. This does not mean precision medicine will not have such benefit, or that further research should not be pursued. But we also need to look for its results, especially those that lead to improved health and of outcomes from treatment of disease. The same holds true for the previous hot topic before precision medicine, namely data analytics and other aspects of Big Data.

In the meantime, I would encourage those who are pursuing these emerging areas to find a home in the larger science of informatics. Indeed, those from the informatics community are working in them (myself included), and we should show there is a solid trail of science leading into them and eschew that they are somehow completely brand new.

References

1. Chaudhry, B, Wang, J, et al. (2006). Systematic review: impact of health information technology on quality, efficiency, and costs of medical care. Annals of Internal Medicine. 144: 742-752.
2. Goldzweig, CL, Towfigh, A, et al. (2009). Costs and benefits of health information technology: new trends from the literature. Health Affairs. 28: w282-w293.

3. Buntin, MB, Burke, MF, et al. (2011). The benefits of health information technology: a review of the recent literature shows predominantly positive results. Health Affairs. 30: 464-471.

4. Jones, SS, Rudin, RS, et al. (2014). Health information technology: an updated systematic review with a focus on meaningful use. Annals of Internal Medicine. 160: 48-54.

5. Collins, FS and Varmus, H (2015). A new initiative on precision medicine. New England Journal of Medicine. 372: 793-795.

6. Ashley, EA (2015). The Precision Medicine Initiative - a new national effort. Journal of the American Medical Association, Epub ahead of print.