How (not) to handle medical certainties produced by algorithms скачать в хорошем качестве

How (not) to handle medical certainties produced by algorithms

1. Predictive algorithms and new medical certainties Recently, more and more algorithms are being used to improve medical treatment. For example, algorithms can help physicians diagnose diseases or find the best treatment for their patients. In addition to the use of algorithms in acute medical settings, algorithms are increasingly used in a predictive way, e.g., to find out which disease a patient will most likely contract at what time and how badly. By determining these individual disease risks and simulating the patients’ future health trajectories, algorithms create “new medical certainties.” By producing new medical certainties, algorithms can enable healthcare systems – which are facing ongoing cost explosion and demographic change – to plan more far-sightedly. E.g., if healthcare systems know in advance how many people will contract certain diseases in the middle to long term future and how many people will need medical treatment, they can hold available the necessary personnel, treatment capacities, or financial resources, or even help prevent those diseases. Thus, by producing new medical certainties, the use of algorithms contributes to improving and sustaining healthcare systems. 2. The challenges of new medical certainties for the social healthcare system On the other hand, however, the medical use of self-learning algorithms may exacerbate existing funding problems in dual public-private healthcare systems like the one in Germany. E.g., if people know with a high degree of certainty what their disease risks are and how their health is most likely to develop, they can use that knowledge to find the most financially advantageous form of health insurance more precisely. For example, there is a high likelihood that people with favorable health trajectories and low disease risks will switch from statutory health insurance – where they have to pay a premium based on their individual income – to private health insurance – where their premium is based on individual risk – while people with overall unfavorable disease risks will stay in or switch to statutory health insurance.. Such dynamics, known as “adverse selection,” challenge the funding of statutory health insurance – which fundamentally relies on a balance of high- and low-risk individuals and an imbalance would lead to an increase in health insurance premiums. Under these circumstances, the mission of the welfare state to ensure healthcare for the needy segments of the population may be at risk. 3. How (not) to counteract the algorithm-produced threats of adverse selection? This scenario raises the question of whether and how to counteract what others have called “the threat of adverse selection.” Is it possible to counter this threat on a technical level, by either: limiting the computing power to prevent the algorithms from producing too many medical certainties; or by setting a maximum level of medical certainty from the outset that the algorithm is allowed to produce? There is a chance that this approach may lead to a moral calculation of disease risk by algorithms. There is also a chance that the problem of adverse selection cannot be solved on the technical level of algorithms, but needs a solution on the societal level. This could range from withholding medical certainties from patients or tying a patients access to medical certainties to agenda-driven conversations, trying to prevent them from switching insurance types to even reconsidering fundamental reforms of the healthcare system – as adverse selection is a fundamental problem of a dual healthcare system that is “merely” exacerbated by self-learning algorithms. We will address these questions and distinctions in our talk, exploring the limits of “moralizing algorithms,” and showing that some algorithm-produced problems require a fundamental solution at the societal level.