In the 1930s, the evolutionary biologist, Sewall Wright, developed the concept of the fitness (adaptive) landscape as a visualization of evolution. The fitness landscape, conceived as a topographic map that resembles a mountain range with peaks and valleys, described different phenotypes of an organism that can vary over a continuous range of genotypes. The vertical dimension in the landscape was conceptualized as the fitness value of a particular trait. Fitter traits, those more optimally adapted to differentially survive and reproduce, are represented by higher peaks. Natural selection is the mechanism that moves a population (of traits, organisms) toward a local peak and removes less fit outliers that are too far downslope. Randomized mechanisms such as genetic drift and sexual recombination enable individuals to jump from a local peak to a more global maximum. These mechanisms also enable species to adapt to a dynamic environment with its associated fitness functions that are mostly changing gradually but also abruptly at times. This biological concept of the fitness landscape has not only extended in scope and dimensions within evolutionary biology but also serves as a grounding framework to define system fitness in machine learning and social science models. In prior essays, I discussed adjacent possibles and the Nash equilibria to describe how the healthcare system is moving towards an inferior optimum. In this essay, I add the related model of fitness landscapes and fitness functions as a framework to explain the trajectory of the healthcare system.
The fitness landscape is a flexible concept that can be extended within and across many dimensions to evaluate phenomena temporally and spatially. It can be applied to parts of a system or to whole systems. For example, it can be used to trend an individual’s health temporally or it can be used to compare populations at a specific time and place. Similarly, it can be used to evaluate stakeholder populations – physicians, health insurers, pharmaceutical companies – within the healthcare ecosystem at a specific time. However, in order to create a fitness landscape, there must be a criterion that defines fitness. In evolutionary biology, the orthodox fitness function for an organism is survival and reproduction. In machine learning applications, a fitness function is a predetermined objective that enables a model to converge on an appropriate solution. However, in complex social domains fitness functions are difficult to define. For example, what is the fitness function of the healthcare system? How does the fitness function for the whole cohere with the fitness function for its parts? What if the fitness functions of the parts conflict (not in kind but in degree) with the fitness function of the whole?
The Institute of Healthcare Improvement introduced the triple aim framework of cost, quality, and access of care to optimize health system performance. The objective being to improve care outcomes (quality) and access to healthcare services, while simultaneously decreasing cost. Thereafter barriers to insurance were reduced in order to improve access and the framework of value-based reimbursements was introduced to decrease cost by tying reimbursements to outcomes rather than to the volume of services provided. Even then, how do you define “outcomes?” Does an improved vital sign (ie blood pressure) or an improved lab value (ie hemoglobin a1c) equate to improved outcomes? Empirically, we know that overall life expectancy has decreased in sequential years and the morbidity of chronic diseases continues to rise. If life expectancy is the fitness function tied to improved outcomes, we can state that the system has not moved to a higher peak. Maybe more interestingly, the introduction of fitness functions can move “populations” to unexpected and undesired peaks on the landscape. A case in point is the example of the population of physicians. Physicians have become increasingly disenchanted by the practice of medicine and are suffering from what is termed as an “epidemic” of professional exhaustion syndrome.
Therein lies the inherent challenge of complex systems. Although goals require a precise definition and must be measured against invariable fitness functions, the social systems we inhabit are complex and multidimensional and oftentimes resistant to simplifications. As top down fitness functions are introduced, each stakeholder in the ecosystem responds and adapts with their local interests and becomes perched on new (higher or lower) peaks. The top down introduction of the triple aim and value-based care moved some stakeholders to higher peaks but other stakeholders to a lower peak. The end result is a system that is persistently entrenched in an inferior optimum. How do you align the various goals of the populations to harmonize with the overall goal of the system? How do you define criteria where outcomes are more connected to intentions? [Next essay]