In a recent study published in PNAS, researchers evaluate the ability of structured information-sharing networks among clinicians to improve clinicians’ diagnostic accuracy and treatment decisions.
Study: Experimental evidence for structured information–sharing networks reducing medical errors. Image Credit: Ground Picture / Shutterstock.com
How can technology improve clinicians’ decisions?
Clinical decision-making is prone to errors in diagnostic assessments, which subsequently leads to incorrect therapy recommendations. The expanding research on clinical collective intelligence demonstrates that a significant percentage of doctors' collective judgment is often more precise than skilled individual clinicians. This ‘clinical crowd wisdom’ has been established across a variety of medical professions.
Recently, researchers have been increasingly interested in the development of medical technology and techniques to utilize crowd wisdom for decision-making processes in clinical settings. Structured data exchange networks with consistent connections can successfully transfer unit-level collective intelligence to actual time improvement in the judgments of individual group members.
About the study
In the present study, researchers explore whether collective medical intelligence among clinicians could improve clinical assessment quality using the network-theoretic approach, which builds on previous research conducted in non-clinical settings.
The study involved 2,941 clinicians from the United States who participated in online clinical vignette-based experiments. The network hypothesis was tested across seven different clinical cases that were conducted in 84 trials.
The collective performance in structured data-sharing networks was compared to the collective performance in independent control groups. The team explored whether the network-theoretic method of using network aggregational dynamics of collective medical intelligence could improve the clinical judgment quality among health practitioners.
There were 12 experiments performed in the study, of which included eight and four network and control trials, respectively. Each study had 40 clinicians, with all clinical vignettes examined by certified physicians from three different specialties including emergency medicine, cardiology, and internal medicine.
Geriatric care, acute cardiac events, pain in the lower back region, diabetes-associated cardiovascular disease, and loss in routine activities’ prevention were among the cases that were included in the study.
For each patient, participating doctors were expected to estimate the patient's risk level and determine a categorical therapy option. Clinicians were randomly allocated in a 2:1 ratio to the network and control conditions.
Networking participants were randomly allocated to one location in a vast, anonymous, and egalitarian web-based peer-to-peer social network with four connections for each therapist.
The non-networking (control) clinicians saw the same scenario and responded to questions answered by the networking clinicians; however, these individuals participated independent of social networks. Participants were given three rounds in the control and network circumstances to submit their risk estimates and therapy guidelines for the provided cases.
The study outcomes included altered preciseness of clinicians' estimated risks and alterations in the frequency of clinical practitioners making the correct therapy recommendations. To assess the impact of data-sharing networks on the precision of clinicians' final therapy recommendations, changes in the percentage of participants providing correct answers in the first round as compared to the third round were evaluated.
A total of 3,360 clinicians were randomized to the network and control conditions, with 70% participating in only one trial. The randomization scheme produced 56 and 28 trial-level observations in the network and control conditions, respectively. The diagnostic accuracy showed a significant improvement among networking and non-networking clinicians.
At study initiation, a 77% and 76% mean accuracy was observed among non-networking and networking clinicians, respectively, thus demonstrating non-significant differences in the conditions. After the last round, the mean accuracy rose to 81% and 79% for networking and non-networking clinicians, respectively.
Structured data-sharing networks significantly reduced clinical errors and improved the clinicians' treatment recommendation quality as compared to control groups of independent clinicians engaged in isolated reflection. The social impact had no negative effect on the most correct physicians, whereas information-sharing networks significantly improved diagnostic accuracy in the least precise group of clinicians.
A statistically significant positive revision coefficient was observed for networking clinicians, thus indicating that less precise clinicians made larger revisions of their responses. Comparatively, more precise clinicians made fewer revisions, thereby granting more accurate clinicians a greater impact on the social network.
Clinical assessment accuracy gains are transferred directly into clinical treatment recommendations, thus increasing the likelihood of doctors shifting from an initially wrong treatment suggestion to the correct advice in their last answer.
The study findings indicate that leveraging networking dynamics of collective learning within egalitarian networks for improving individual clinician decision-making in real-time is a promising approach.
Future research is needed to evaluate the extent to which collective learning among clinical practitioners might engage different cognitive pathways as compared to those utilized by non-networking clinicians, including rapid and delayed thinking. Exploring the influence of interaction technologies on clinical reasoning quality and the impact of providing clinical practitioners with varying data on the network size could also improve clinicians' collective intelligence.