In recent posts (such as here), I suggested that the volume, velocity, and variety of modern data projects produce recommendations that need to be acted on so quickly that we must obligate our decision makers to follow them, or more realistically replace human accountable decision making with automation. In those earlier discussions, I focused on the decision maker role as distinct from the population. I pointed out my concern that the loss of accountability could lead to a much less stable society, a society more prone to rebellion. Implicit in that concern is an assumption that the general population remains free to follow the decision maker.
Once we decide to obligate decisions on the result of big data analytics, we essentially enter an era of government by data instead of government by people. Earlier, I coined the phrase dedodemocracy to describe this new type of government by data. Later I made a distinction between dedodemocracy and dedomenocracy where the former allowed for some democratic involvement in the decision making while the latter did not. The democratic involvement occurs when the citizens become data scientists in the sense of having independent ability and capacity to query and analyze data impacting one’s life. For those who lack this skill, they will be subjects of the latter form of government strictly by data.
With that in mind, my earlier concerns about civil unrest may be answered through a regime that obligates participation of the population to follow the recommendations from big data algorithms. Such a regime would not tolerate dissent that could produce instability. Although this is characteristic of ancient regimes, there are modern examples of success authoritarianism being persistent for generations without successful (or even noticeable) civil rebellions. I don’t mean to start writing a futuristic dystopia of rule by data, but it seems to me that it is at least possible to have a peaceable society with rule purely by data and without human accountability. We merely need to mandate the participation of the entire population.
The motivation for today’s post was an intersection of two articles plus a personal complaint.
The personal complaint is the recent legally imposed mandate that I purchase health insurance that complies with the affordable care act. I need to obtain individual health insurance policy and I am very sensitive to the costs involved. I previously enjoyed health insurance that was much cheaper because it didn’t mandate as much as now required. I have no choice but to get the more extensively mandated coverage.
As I read through the differences of the old and new policies, I suspect a good portion of the additional cost involves the services that are required to be offered for no additional costs (such as annual physicals and routine diagnostics) or offered for co-insurance before deductible (such as mental health services). Although the policies describe these as benefits, it appears that the premium is really a prepayment for these services. One intention of the law is to eliminate the barriers for access such wellness care services, but the increased premiums to pay for these services in advance essentially provides an incentive to use these services. We’re paying for annual physicals and routine diagnostics, so we should schedule them into our lives.
Recently, there has been more attention on the electronic health records providing a wealth of information that can be mined to improve health care by customizing health services based on outcomes for other patients with similar traits. This emerging big data project will use the records for analysis to determine future allocations of health services. The quality of the analytics eventually will depend on the extensive participation of the entire population. In order to get the best analysis of outcomes for groups, we need everyone to participate in the health care system. Everyone should be given routine wellness checkups and diagnostics, and everyone should follow through with recommended follow-up examinations and treatments. This will provide us the best data to determine what works best for each category of patients.
I see the health care system heading in the direction I described earlier as government by data: we need to require the entire population participate in the health care system in order to gather the data to improve health care. Initially, we enforce this with premiums that provide incentive to take advantage of free or low co-insurance services made possible by higher premiums. For now, it is merely an incentive with an option to forego these services. However, such non-participation results in a self-selection bias that will impact the accuracy of the predictive analytics. In order to best determine what care to provide unhealthy people, we need data from healthy people. We need to mandate not just the purchase of health care, but the use of the recommended care. Eventually we will require a mandate for utilizing health care in order to provide us the best data to make better allocations of services.
Briefly, the first of the two articles I read yesterday is this description of yet another data-driven ride sharing service. This one is the dynamic deployment of carpooling vans based on data where riders are most likely to start and end their commutes. What makes this different from usual car pooling is the use of external data to identify people who are more likely to use carpooling vans and where they live and work. For today’s post, I observe that the success of this service depends on people actively participating in social media and sharing at least enough information to infer their commuting needs. Without this data, this service can not exist or at least can not compete with the usual spontaneous car pooling arrangements. Initially, the new service is taking advantage of preexisting participation, but as more services like this appear there will be more incentive for more people to participate and to make more information about themselves available.
The second of the two articles I read yesterday is this Time article describing how some doctors decide against treatment for their own terminal illnesses. The examples are doctors who become patients after receiving a diagnosis of a serious illness. Some doctors choose to let the disease take its course without taking advantage of any hospital services that may increase their chance of survival. They understand that the treatment involves a degraded quality of life over what would happen without treatment, and that the chances were still high the disease will kill them anyway. The article presented a sense that this was a good choice with the patients enjoying a longer period of good quality of life despite the fact that they loose the chance to extend their lives.
In light of my earlier observation about the trend to mandating routine health care services, I wonder whether we can allow patients to decline recommended medical procedures. Even in the above cases of doctors declining care, they spent the majority of their lives paying premiums with the expectation that the insurance would be available in precisely these circumstances. Chances are very good that until the terminal illness diagnosis, they paid much more into the insurance programs than they received in services. The premium investment is an implicit incentive to accept the recommended care just based on financial considerations: the patient has the opportunity to get the services that justified paying premiums over the years. In contrast, the article describes that the treatment is no guarantee of success and can be very brutal on the person’s quality of life. The article makes a good case for a rational decision to decline the recommended care even when insurance would fully cover it.
Although such a decision is rational on an individual basis, it may be unacceptable from a societal perspective. In order to have good predictive analysis of data, we need data from everyone who qualifies for the procedure. The increased variety of participation helps the algorithms refine what combinations of physical factors improve or degrade outcomes of procedures. Also, immediate acceptance of the recommended care provides consistency to apply for predicting future outcomes. Much of the objective to manage health care costs is to improve cost effectiveness of actual intervention procedures for all diseases and especially those that are potentially (or even likely) fatal. To achieve this objective, we need data for the new patients. This data is improved when it is free from selection biases.
If the data-driven (evidenced-based) algorithm decides a patient should receive a certain procedure, then in order to continue to refine that algorithm, we need to mandate acceptance of that procedure. We already have similar mandates for accepting care when it is needed. The article describes the case of immediate administration of CPR unless the patient has a “no code” (I assume the same as “do not resuscitate”) necklace or tattoo. If CPR is needed, it will be administered, but the patient is usually already unconscious and unable to object. For a less life-threatening example, there is this Wall Street Journal opinion describing an automatic activation of an expensive trauma team if an emergency room patient has bleeding in the head. If the symptoms indicate a procedure, then that procedure needs to be followed.
The above article about doctors declining treatment for their own terminal illness could actually present a cautionary tale that this option may be taken away from individuals. As we increase the reliance of health care on data analysis, we may enter an era where the patient is obligated to follow the recommendations. Even if the patient realizes that the odds of survival do not compensate for the suffering required for treatment, society needs the data in order to improve the treatment for this condition for populations in a similar circumstance. I suspect the first group to be so obligated would be those in the health care industry.