Presentation
HE6 - Counteracting Cognitive Biases in Healthcare Decision-Making: A Scoping Review of Technological Debiasing Strategies
DescriptionBackground
In high-stakes settings such as healthcare, sustained high-level cognitive performance relies on designing working environments to optimise humans’ cognitive capacities. Cognitive biases in medical decision-making are widely documented (Blumenthal-Barby et al., 2014; Saposnik et al., 2016; Whelan et al., 2020) and can degrade decision quality, potentially leading to life-limiting outcomes in domains like surgery (Armstrong et al., 2023; Graber et al., 2018). A large portion of cognitive bias mitigation or debiasing research has employed strategies that aim to elicit analytical reasoning in decision-makers to minimise cognitive errors due to bias. However, the efficacy of strategies such as cognitive forcing (e.g. O'Sullivan & Schofield, 2019) and instruction- or training-based interventions (e.g., Sherbino et al., 2011; 2014) is contingent on the time and cognitive resources available to individuals, which are scarce in high-stakes medical settings with high clinical workload.
To tackle this theory-practice divide, we draw from the distributed cognition approach (Hutchins, 2000) which attributes decision-making to the interactions between humans and their material environments over time, not merely to individual cognition. In doing so, the present study extends cognitive bias mitigation to the decision environment focusing on strategies called technological debiasing strategies (Larrick, 2004), which use or modify systems, workflows, artefacts and agents (human and non-human), that are external to individual decision-maker(s) to minimise biased reasoning. Technology, here, is defined as “the specific methods, materials and devices used to solve practical problems” (Hinsz, 2015); it includes but is not limited to digital technology.
Study summary
Using a systematic search strategy, this scoping review aimed to synthesise the extant empirical literature on technological debiasing strategies. In this presentation, we focus on the implications of a subset of findings specific to healthcare contexts.
Searches were conducted across six databases (IEEE Xplore, ACM Digital Library, PsycINFO, Emerald, Web of Science, PubMed). Two reviewers (HD, EH) screened for articles meeting the following inclusion criteria: peer-reviewed, experimental studies that explicitly aimed to minimise cognitive bias in healthy adults using technological strategies. Studies examining debiasing for behaviour change, and cognitive or motivational strategies were excluded. Debiasing strategies were classified into either: (i) Group composition and structure, (ii) Information design, or (iii) Procedural debiasing, based on each of the three distributed cognitive processes outlined by Hollins et al. (2010) Cognitive biases were further classified into eight categories identified by Fleischmann et al. (2014) following a scientometric analysis of biases in information systems research.
Within healthcare, twenty-six articles were identified, addressing thirty-one debiasing interventions for twenty-nine individual cognitive biases. Information design (n=26) accounted for 84% of the debiasing investigations, followed by procedural debiasing (n=5). None examined the impact of modification of group composition and structure.
Overall, 80.6% of the strategies were effective (n=25), 12.9% were ineffective (n=4) and 6.4% were partially effective (n=2). Effective strategies were those that resulted in significant reductions or complete mitigation of biases, and ineffective strategies were those that failed to do so. Partially effective debiasing strategies were defined as those that mitigated one bias but not another (within the same study), or mitigated biases under specific experimental conditions but not others.
Pattern-recognition biases (n=9) were studied the most, followed by perception (n=8) and decision biases (n=7). Four studies recruited representative samples, i.e., medical professionals and medical students, while the rest involved unspecified and lay samples, focusing on judgement and decision-making from patient or lay perspectives.
Discussion and key takeaways
This study synthesised the technological debiasing strategies applied to healthcare contexts using a distributed cognition framework. Technological debiasing strategies were found to be largely effective in minimising bias (e.g., graphical information presentation, decision support systems). Integration of debiasing considerations into design of healthcare environments, tools and teams, in light of advances in clinical human factors, can help to minimise cognitive errors caused by bias that may be ingrained in the healthcare system. Based on the evidence, we argue that technological strategies can also be beneficial in aiding the prevention of cognitive overload, allowing healthcare workers to maximise their cognitive resources for the task at hand, further enhancing patient safety.
Findings highlight several research gaps and avenues for future work, as follows:.
1. Group composition and structure: Further research into the strategic composition of medical teams, utilising cognitive diversity to counteract commonly prevalent biases, in order to facilitate open communication and widen available knowledge, information pool and range of cognitive styles. Previous research outwith healthcare has shown groups outperform individuals when composed and supported appropriately (e.g. Meissner et al., 2018; Mojzisch et al. 2008; O’Leary, 2011).
2. Information design: Using human factors principles to guide design of electronic health records, visualisation dashboards and displays that allow for unbiased perception and accurate information interpretation under fast-paced conditions.
3. Procedural debiasing: Careful reconsideration of workflow design, including task sequences to ensure initial events or information do not inappropriately skew processing of subsequent events (as in anchoring or availability bias) and that biases are not left unchecked at any stage of the decision process.
4. Research in naturalistic healthcare settings systematically identifying when and where cognitive biases are most likely to occur, to target specific decision points prone to cognitive errors.
In high-stakes settings such as healthcare, sustained high-level cognitive performance relies on designing working environments to optimise humans’ cognitive capacities. Cognitive biases in medical decision-making are widely documented (Blumenthal-Barby et al., 2014; Saposnik et al., 2016; Whelan et al., 2020) and can degrade decision quality, potentially leading to life-limiting outcomes in domains like surgery (Armstrong et al., 2023; Graber et al., 2018). A large portion of cognitive bias mitigation or debiasing research has employed strategies that aim to elicit analytical reasoning in decision-makers to minimise cognitive errors due to bias. However, the efficacy of strategies such as cognitive forcing (e.g. O'Sullivan & Schofield, 2019) and instruction- or training-based interventions (e.g., Sherbino et al., 2011; 2014) is contingent on the time and cognitive resources available to individuals, which are scarce in high-stakes medical settings with high clinical workload.
To tackle this theory-practice divide, we draw from the distributed cognition approach (Hutchins, 2000) which attributes decision-making to the interactions between humans and their material environments over time, not merely to individual cognition. In doing so, the present study extends cognitive bias mitigation to the decision environment focusing on strategies called technological debiasing strategies (Larrick, 2004), which use or modify systems, workflows, artefacts and agents (human and non-human), that are external to individual decision-maker(s) to minimise biased reasoning. Technology, here, is defined as “the specific methods, materials and devices used to solve practical problems” (Hinsz, 2015); it includes but is not limited to digital technology.
Study summary
Using a systematic search strategy, this scoping review aimed to synthesise the extant empirical literature on technological debiasing strategies. In this presentation, we focus on the implications of a subset of findings specific to healthcare contexts.
Searches were conducted across six databases (IEEE Xplore, ACM Digital Library, PsycINFO, Emerald, Web of Science, PubMed). Two reviewers (HD, EH) screened for articles meeting the following inclusion criteria: peer-reviewed, experimental studies that explicitly aimed to minimise cognitive bias in healthy adults using technological strategies. Studies examining debiasing for behaviour change, and cognitive or motivational strategies were excluded. Debiasing strategies were classified into either: (i) Group composition and structure, (ii) Information design, or (iii) Procedural debiasing, based on each of the three distributed cognitive processes outlined by Hollins et al. (2010) Cognitive biases were further classified into eight categories identified by Fleischmann et al. (2014) following a scientometric analysis of biases in information systems research.
Within healthcare, twenty-six articles were identified, addressing thirty-one debiasing interventions for twenty-nine individual cognitive biases. Information design (n=26) accounted for 84% of the debiasing investigations, followed by procedural debiasing (n=5). None examined the impact of modification of group composition and structure.
Overall, 80.6% of the strategies were effective (n=25), 12.9% were ineffective (n=4) and 6.4% were partially effective (n=2). Effective strategies were those that resulted in significant reductions or complete mitigation of biases, and ineffective strategies were those that failed to do so. Partially effective debiasing strategies were defined as those that mitigated one bias but not another (within the same study), or mitigated biases under specific experimental conditions but not others.
Pattern-recognition biases (n=9) were studied the most, followed by perception (n=8) and decision biases (n=7). Four studies recruited representative samples, i.e., medical professionals and medical students, while the rest involved unspecified and lay samples, focusing on judgement and decision-making from patient or lay perspectives.
Discussion and key takeaways
This study synthesised the technological debiasing strategies applied to healthcare contexts using a distributed cognition framework. Technological debiasing strategies were found to be largely effective in minimising bias (e.g., graphical information presentation, decision support systems). Integration of debiasing considerations into design of healthcare environments, tools and teams, in light of advances in clinical human factors, can help to minimise cognitive errors caused by bias that may be ingrained in the healthcare system. Based on the evidence, we argue that technological strategies can also be beneficial in aiding the prevention of cognitive overload, allowing healthcare workers to maximise their cognitive resources for the task at hand, further enhancing patient safety.
Findings highlight several research gaps and avenues for future work, as follows:.
1. Group composition and structure: Further research into the strategic composition of medical teams, utilising cognitive diversity to counteract commonly prevalent biases, in order to facilitate open communication and widen available knowledge, information pool and range of cognitive styles. Previous research outwith healthcare has shown groups outperform individuals when composed and supported appropriately (e.g. Meissner et al., 2018; Mojzisch et al. 2008; O’Leary, 2011).
2. Information design: Using human factors principles to guide design of electronic health records, visualisation dashboards and displays that allow for unbiased perception and accurate information interpretation under fast-paced conditions.
3. Procedural debiasing: Careful reconsideration of workflow design, including task sequences to ensure initial events or information do not inappropriately skew processing of subsequent events (as in anchoring or availability bias) and that biases are not left unchecked at any stage of the decision process.
4. Research in naturalistic healthcare settings systematically identifying when and where cognitive biases are most likely to occur, to target specific decision points prone to cognitive errors.
Event Type
Poster Presentation
TimeTuesday, March 264:45pm - 6:15pm CDT
LocationSalon C
Digital Health
Simulation and Education
Hospital Environments
Medical and Drug Delivery Devices
Patient Safety Research and Initiatives