As a healthcare technologist I spend a lot of time thinking about where healthcare and medicine is headed in order to understand what role technology can play in realizing the envisioned future. For some time now I have been contemplating Dr. Leroy Hood’s P4 vision of medicine, which stands for predictive, preventive, personalized and participatory. In brief, Hood’s P4 vision entails the integration of genetic data with knowledge of how the environment modulates the expression of genes to create the phenotype of health or disease. It is a bold optimistic vision of the future of healthcare that implies data – lots of it.
Intrigued with Hood’s vision – and its ramifications for the future of healthcare IT – I decided to test it on myself. Could P4 help us formulate a go-forward strategy in government healthcare? So I took a nine-month period off work (from July 2011 through March 2012) to educate myself on the constructs of P4 medicine. It was not long before I was introduced to a veritable zoo of new knowledge disciplines most of which ended in “-omics”. These included genomics, proteomics, transcriptomics, metabolomics/metbanomics, epigenomics, microbiomics, pharmacogenomics, toxicogenomics, nutrigenomics and functional metagenomics in addition to systems biology and functional medicine. Most of these disciplines seem to have sprung forth since I graduated from medical school in the early 1990s.
As I plowed through journal articles and textbooks on these subjects I also decided to obtain more data on my own genetics. I signed up for 23andMe’s personal genetic service to obtain my personal set of single-nucleotide polymorphism (SNP) data, which covered more than a million data points, and began combing through it to see what new and wondrous things I might discover about myself. When it became available, I also signed up for getting my exome sequenced, which was another 50 million base pairs of data that gave me the part of my genome that codes for proteins. Data like these drive home the concept of biochemical individuality and I gained a greater appreciation of uniqueness and the need for personalization of therapeutic approaches.
While it was interesting I still felt the need to dig more deeply into how to apply what I was learning to a clinical problem to see what difference, if any, all of this could have on an outcome; how it could impact my own healthcare. So I decided to see if what I was learning about my genetics could have a direct impact on the management of my diabetes. For a decade I have been managing my Type 2 diabetes with diet and exercise and have managed to avoid taking oral anti-diabetic drugs or insulin. Shortly before I embarked on this journey into P4 medicine, however, I had experienced an upswing in my hemoglobin A1C (a measure of blood sugar control) that warranted the addition of medications to achieve better control. My doctor and I also had begun discussing whether and when we would need to be adding insulin injections to my treatment regimen. It seemed I had found my high impact target.
I dug into my SNP data and found that I had data on 11 SNPs that were known to be associated with diabetes. Of these 11 variants 6 were known to be associated with increasing my likelihood of having diabetes. I spent a great deal of time running down studies to understand what metabolic pathways these genes coded for and how the changes were related to diabetes.
Out of this a clearer picture of what the genetic basis of my diabetes was emerged: the variants I had were associated with impaired baseline insulin secretion, altered first phase insulin response, decreased response to dietary glucose, increased levels of fasting plasma glucose levels (especially overnight) and finally altered fat cell metabolism and insulin resistance. My experience over the years with managing my diabetes had taught me that the foods we eat have a significant impact on our state of health. Nutrigenomics had further clarified this by providing me with a more detailed understanding that the foods we eat are actively modulating our genes, sometimes switching them on and sometimes switching them off. Armed with a better understanding of my unique physiology and the means by which the food we eat modulates gene expression – per P4 – I set out to determine a treatment protocol that would return me to optimal function or as close to it as I could get.
Given all of this newfound knowledge I realized that I needed to develop a data driven approach to making decisions about what to eat and what not to eat. To do this I had to identify a biomarker that I could monitor and a simple rule for decision making, which I would use to decide what foods to eat and what to avoid. I also investigated what effect pharmaceuticals, nutraceuticals and herbals had on glucose control. So I collected data, a lot of data. I used a glucometer to measure glucose levels and set up a lab to determine hemoglobin A1C levels and lipid levels monthly at home. In addition, I measured weight, blood pressure and pulse oximetry and collected sleep data using a Zeo sleep monitor. During this experimentation phase, I would measure a baseline glucose level prior to eating, then I would eat whatever item I was testing. Immediately upon putting the fork down I would measure another glucose level and measure glucose levels every 5 to 10 minutes afterward until my glucose level returned to its pre-meal levels. I would then plot the glucose response curve for that food item. I did this for the foods I ate individually and also in combinations. Once I had this data I also collected data on the effect pharmaceuticals, nutraceuticals and herbals had on glucose levels with the various foods individually and in combination as meals. I recorded all of this data into my lab notebook for analysis.
Out of this research I was able to develop an algorithm for making food choices. For any food item, if it caused a greater than 40 mg/dL increase in glucose level or took longer than 2 hours to drop below 126 mg/dL or took longer than four hours to return to baseline, it either needed to be eliminated from the diet or the portion size reduced until it fell within these parameters. I also realized that optimal blood glucose levels fall in the range of 60 to 85 mg/dL for the fasting state. This is the equivalent of an Hgb A1C of less than 4.9 percent. By applying the algorithm to my own diet and determining the impact of specific supplement combinations I have established a diet and treatment regimen to manage my diabetes that is data-driven and personalized to account for my biochemical individuality.
The outcomes have been remarkable: Hgb A1C levels have dropped from over 7 percent back down to the 4.9 percent to 5.5 percent range; I have lost over 50 lbs of weight, which has resulted in a change to my body mass index (BMI) from morbidly obese to obese to overweight now; significantly improved lipid levels, blood pressure and resting pulse rate; improved sleep quality and exercise tolerance; and finally, just feeling better than I have in years. As to the treatment regimen, I now have a data-driven approach for assessing whether or not a specific intervention works for me. Using this approach I have designed a treatment regimen that uses a combination of prescribed drugs, nutraceuticals and herbal compounds proven effective for me.
So what have I learned during this foray into P4 medicine and what does this have to do with Health IT?
It is about data, granular data, lots of data, big data. During the time I have been engaged in this exploration I have amassed huge datasets, gigabytes of genetic data and thousands of data points collected by hand in my lab notebooks. This is data I use to manage my health. It is data I own; I paid for the genetic data out of my own pocket and have collected the rest in my quest to restore myself to a better state of health. My world is now the world of big data and personal analytics.
Admittedly, we cannot expect individuals to have the ability to test themselves in this way. But my experience points to the role healthcare IT can play in this new future of personalized medicine: We have a need for sensors that are less invasive and require less attention for collecting the data we need to help us manage our health. We have a need for applications that have a rich user experience that enables us to manage big data on a personal level and to provide us with a dashboard of my health status. And we also need the ability to share this data with our providers when it is appropriate without overwhelming them with data that is not relevant to their decision making process.
[See also: Public health's 5 big data hurdles.]
As we move into the world of P4 medicine it is apparent that we will not do less testing of patient biomarkers but rather more, a lot more. This testing will not be done in the primary care provider’s office setting, however, it will instead be conducted in the patient’s home setting. This means real time or near real time event streams will be generated requiring aggregating and processing into meaningful alerts for the patient and their care team. I suspect that as we move forward, testing that is now the exclusive province of research institutes will become ubiquitously available in the primary care provider setting, such testing as protein expression and metabalomic testing, which will enable providers to more quickly determine the impact of a specific therapeutic intervention or change to a therapeutic regimen.
Healthcare that is personalized, preventive, predictive and participatory entails big data, personal analytics, ubiquitous digitization through bio-sensors, event processing, filtering and context aware presentation of information that is relevant at the point decisions are made by both the recipients and providers of healthcare. It looks to be an interesting and exciting future!
David Riley is chief of informatics for Harris Healthcare Solutions.