Lay, professional and Artificial Intelligence perspectives all say the same – number of lives in risky medical decisions matters in gain frames but not in loss frames

Academic Publication

Using the structure of the ‘Asian Disease’ problem, we asked naïve participants and medical students in India about what drug would they administer for clinical trials for an unknown disease and for COVID-19. This was the only published study we are aware of about framing effects conducted in India during the COVID-19 pandemic and among the few conducted worldwide.

It used a vignette design as in the classical studies on Asian Disease. One drug had sure outcomes on the number of lives that could be potentially saved while the other drug had risky outcomes. We additionally manipulated the number of lives at stake. The same information was presented in a gain-frame (lives saved) to one group of participants or in a loss-frame (lives lost) to another group. Then, we specifically targeted the Corona Virus outbreak and asked naive participants what drug should medical practitioners and medical AI software administer to groups of people who showed positive symptoms of the virus and have volunteered to undergo clinical trials for vaccines. The results helped us test the Prospect Theoretic prediction that one would be risk-averse in the domain of gains (lives saved) and risk-seeking in the domain of losses (lives lost).

The overall question we investigated was whether framing can bias these decisions and if risk preferences shift depending on the number of patients. Hypothetical information about two medicines used in clinical trials having a sure or a risky outcome was presented in either a gain frame (people would be saved) or a loss frame (people would die). The number of patients who signed up for the clinical trials was manipulated in both frames in all the experiments. Using an unnamed disease, lay participants (experiment 1) and would-be medical professionals (experiment 2) were asked to choose which medicine they would have administered. For COVID-19, lay participants were asked which medicine should medical professionals (experiment 3), artificially intelligent software (experiment 4), and they themselves (experiment 5) favour to be administered

The cover story stated there are two alternative medical drugs to be tested on different groups of patients who had voluntarily signed up for clinical trials at various hospitals. Both alternatives can cause serious side effects that can lead to death for some patients. The scientific estimates of the consequences are known and hence the outcomes of the trials can be predicted. The number of lives was manipulated within participants at three levels of magnitude in random order: 18, 402, 912. One group was presented with a gain frame while the other group was presented with the same information in a loss frame .

For example, in the gain frame, when the number of lives was low, it was presented as,

About 18 people suffering from a disease signed up for a medical trial. Two alternative drugs have been developed to treat the disease. Assume that the exact scientific estimate of the consequences of the drugs is as follows:

Drug A: 6 people will be saved and

Drug B: 1/3 probability that 18 people will be saved and 2/3 probability that none will be saved.”

The version with a medium level for the number of lives was similar except saying “About 402 people suffering . . . Drug A: 134 people will be saved and Drug B: 1/3 probability that 402 people will be saved and 2/3 probability that none will be saved.” Similarly, the high level was “About 912 people suffering . . . Drug A: 304 people will be saved and Drug B: 1/3 probability that 912 people will be saved and 2/3 probability that none will be saved.”

In the group presented with loss frames, the options given to choose were framed in terms of lives lost. For example, for 18 people, the options were Drug A: 12 people will die and Drug B: 1/3 probability that none will die, and 2/3 probability that 18 people will die. Likewise for 402 people, Drug A: 268 people will die and Drug B: 1/3 probability that none will die, and 2/3 probability that 402 people will die. For 912 people, Drug A: 608 people will die and Drug B: 1/3 probability that none will die, and 2/3 probability that 912 people will die.

Participants were told, “Assume that the exact scientific estimate of the consequences of the drugs is as follows, which alternative would you have chosen to administer?” They had to choose between two drugs presented in a random order (Drug A and Drug B) wherein one had a risky outcome and the other had a sure outcome. 

A similar experiment was set up to extend the findings for COVID-19 and what we would want Medical AI software to do as the following:

Key Findings: We observed a framing effect both for an unnamed disease and the Coronavirus – tested right when the pandemic was unfolding. This is one of the few studies to have tested the framing effects of a real-world disease on a diverse group of people and from multiple perspectives. Crucially, we found that the number of lives moderated risk-aversion in gain frame (people were risk-neutral for low number of lives and risk-averse for high number of lives). However, in the loss frame, risk-seeking was observed irrespective of how many lives were at stake. So, essentially, the results were broadly consistent with prospect theory: people were more risk-seeking in the loss frames than the gain frames. However, risk-aversion in gain frames was sensitive to the number of lives with risk-neutrality at low magnitudes and risk-aversion at high magnitudes. In the loss frame, participants were mostly risk-seeking. This pattern was consistent across laypersons and medical professionals, further extended to preferences for choices that medical professionals and artificial intelligence programmes should make in the context of COVID-19.

These results underscore how medical decisions can be impacted by the number of lives at stake while revealing inconsistent risk preferences for clinical trials during a real pandemic. As we gear towards dealing with scientific advances and medical trials to counter disease outbreaks, a more nuanced understanding of how lives are affectively valued will pave the way for future medical decisions, technology developments, ethics, and policy.

---