If personalized medical care is the goal, how does that fit with the concept of “population health,” the darling of the health care industry’s drive toward better results and lower costs? Are these two concepts really at odds, or do they work in tandem? This is not a rhetorical question; in the current environment of keeping costs under control, lives are at stake.
How Personalized Medicine Should Work
We know that best outcomes occur when individuals are appropriately assessed and allowed to make choices based on their personal characteristics. Personalized medicine is not a concept of averages; it is a concept of variations around the average. Individuals must know how they differ from the group and if their differences matter to their care. At present, even medical research is focused on averages, and the technologies brought to bear on medical care, like the randomized controlled trial (RCT), are technologies of the average.
In a randomized controlled trial, the process of randomization does two things: limits treatments and balances variations in confounding characteristics. The first component is a good one. Observational research does not limit treatment options; hence, it fails as a scientific method. People are free to choose, and that freedom diminishes inference to others who would not chose the same. The only way observational research may help is if those free to choose knew why they were making their choices. They don’t, and present observational research methods do not account for this. The RCT is an advance over observational research, limiting options by assigning them.
The second component of the RCT, balancing confounding factors, is where the technology can and will fail individuals (I will ignore the fact that this balancing may fail, also). Rarely will a RCT have enough people parsed along the lines of measured confounders to allow inference in subdivided, confounded groups of patients. As I described in an earlier post, a patient of mine had some confounders and not others, and the RCT for her disease could not answer her medical questions. Again, the RCT purposely averages confounders, thereby blunting any chance for individuals and their variations.
Is Population Health for the Average Patient?
So, what is wrong with applying the average outcomes of a trial to an individual? This is obvious, but let’s nail it down. First, not all subdivided patients will benefit, or gain the same amount of benefit. The odds of harm and benefit vary across groups, but are averaged. Look at any report of a RCT, at the table showing how the outcome varies within measured subgroups of patients, and you will see different confidence intervals for benefit. Hence, applying the studied treatment to all and expecting the average benefit would only provide potential harm to those who would not benefit; nor would this approach allow individuals to balance their personal differences in outcomes should some amount of benefit be possible. It will always be better to limit treatments to those who would benefit. My patient had characteristics that imposed restrictions on the averaged-out treatment effects. The outcome effect varied wildly and widely for her characteristics and even suggested more harm than benefit was possible for her subgroup.
To be realistic, the goal of most Population Health programs is not, per se, aimed at delivering medical treatment. Rather, Population Health is usually part of a performance measurement process that attempts to categorize patients by condition and then determine whether those patients’ outcomes are acceptable. It may also map them over time. Applying analytics to patient groups is not negative and can be instructive. Nor are the vast improvements in public health by vaccinations and efforts to eradicate infectious disease at issue here.
The problem is, however, that the Population Health definition as used in research, and then applied to actual treatment such as cancer care, is interventional. The intent is to apply the average benefit across the population to individuals. That extrapolation of value based on such groupings has many flaws. While analytics can provide a baseline as to how patients are doing, Population Health cannot help differentiate effective solutions to complex clinical problems. It also cannot, if we are honest, determine whether one provider did or did not provide good care.
The realm of most Population Health is process improvement, not outcomes improvement. Yet the concept of Population Health sometimes leaks into the actual treatment arena, supported by quality measures of what “should” happen to patients who are assumed to be all the same. We ask providers to do the same for everyone, and then punish them for poor outcomes. The fact is that our knowledge base of what works to improve care is very dependent on individual factors and cannot be dictated across a population.
Crafting Medical Care for Individuals Instead of the Average
The real solution to development of effective medicine is to support our “population health” analytics with a change in the way we do research, and thereby vastly expand the data associated with patients as we evaluate their outcomes.
We need research methods other than randomized controlled trials to deal with the vagaries of the human experience. But these personalized research methods will have requirements. First and foremost, all patients must be in the study, not just those who agree to be in a RCT. There are plenty of observations that show that people who enter trials may differ from those that don’t. When all patients are involved, the vagaries will amass in numbers that may help us detect differences in outcomes for different groups. So, how do we get all patients involved?
We get them involved via clinical data repositories. A clinical data repository is one that captures all outcome data on a full population of patients, for a physician group, a group of physicians’ groups, an insurer, a Medical Home, an Accountable Care Organization, and so on. Outcomes will include if the patients are alive, where they reside and all of the medical care they receive; every bill, every diagnosis code and every medication they take—in other words, every event that occurs in their medical care must be captured and dated. This sort of data most likely must be aggregated from several sources, including insurance claims and health provider source systems. Any organized group can develop an outcomes-based clinical repository.
In addition to outcomes data, there are data required for diagnoses. For example, HA1C is collected on all diabetics. Women with breast cancer will have standard measures collected. However, it is not necessarily a requirement of any single clinical data repository to collect all this data for every diagnosis. A practice group or single health care system may never do research accounting for the vagaries of people in their practice, as no single group or system may have enough patients. However, they will be capturing data on these diseases somewhere in their system. Since the data exists somewhere, it can be collected and collated.
In the early 1990’s I saw an Electronic Health Record (EHR) that made a lot of sense. The EHR did not exist except in “virtual reality.” This EHR was constructed on the spot when a patient visited care. The data for the record resided in multiple, disparate databases. The entirety of the EHR was not a single database. The EHR drew data from existing databases via an “event database.” Each of the more than 90 sites of care in the medical system would send a “flag” to an event database for that patient. The flags served to find every source of data for that patient and presented it in a web interface.
Clinical Data Repositories Needed for Personalized Medicine
This sort of planning will be needed for research in personalized care. There will need to be collated, flagged, if you will, clinical data repositories for every disease we know of. These can be developed to capture all information about a certain disease, and then research can be planned in the collated subgroups. This is the only way we will begin to find differences in groups of patients; we need lots of patients within subgroups.
All of clinical medicine should be research with clinical data repositories. In fact, we believe that research should form the basis of doing clinical care. Clinical care is the care of individuals, and research methods must match that type of care. Waiting for information from RCTs is not the answer to clinical, individualized care. Nor is waiting for genetic markers, even as complex genetic markers set the next step to personalized care. Better data repositories and systematic experimentation are needed—and needed now. Clinical research will be useless if it is not contemporary and tailored to the vast range of individuals under our care.
Founded in 2002, ICLOPS has pioneered data registry solutions for improving patient health. Our industry experts provide comprehensive Solutions that help you both report and improve your performance. ICLOPS is a CMS Qualified Clinical Data Registry.
Image Credit: Saïvann