Reducing misdiagnoses and cognitive errors using virtual patients and automated feedback in a clinical reasoning curriculum

TL;DR

A medical curriculum using virtual patients and automated feedback improves diagnostic accuracy and reduces cognitive errors in students.

Contribution

A new clinical reasoning curriculum using virtual patients and automated feedback significantly improves diagnostic skills in early medical students.

Findings

Misdiagnoses dropped threefold and cognitive errors halved with repeated practice.

Diagnostic justification and test ordering were strongest predictors of accurate diagnoses.

First-year students who completed 20 cases outperformed second-year students who did 10.

Abstract

Diagnostic errors remain prevalent across all specialties, driven largely by deficits in clinical reasoning (CR). Although CR is a core competency, most medical schools lack structured pre-clerkship CR training. Virtual patients (VPs) with automated feedback offer scalable, simulation-based training to improve diagnostic skills and reduce faculty workload. The aim of this study was to assess whether a CR curriculum using VPs with automated scoring and deliberate practice improves diagnostic accuracy and CR.

We conducted a multi-site observational study across five North American medical schools. First- and second-year students completed up to 20 diagnostic VP cases on TeachingMedicine.com, each with automated scoring to inform individualized feedback. We analyzed 1.55 million datapoints from 12,400 cases completed by 1,066 students to assess differences in CR performance between correctly and incorrectly diagnosed cases, associations between CR components and diagnostic accuracy, and learning gains over time.

Misdiagnoses occurred in 20.1% of cases. Correct diagnoses were associated with higher diagnostic justification (DxJ) scores (+ 50%), better test ordering (+ 51%), and fewer cognitive errors (–89%). Multivariate analysis identified DxJ and cognitive errors as the strongest predictors of diagnostic accuracy. With repeated practice, students improved DxJ by 72%, test ordering by 40%, and reduced misdiagnoses threefold and cognitive errors by half, with no plateau observed after 20 cases. By end of pre-clerkship, first-year students who completed 20 cases outperformed second-year students who completed 10 in all CR metrics. All results were statistically significant with p < 0.0001.

This curriculum shows that CR skills are highly trainable through deliberate practice. Improved DxJ and reduced cognitive errors are strongly associated with lower misdiagnosis rates. In contrast to a common misperception, training CR diagnostic skills is successful when started in the beginning of 1st year medical school prior to students’ acquisition of significant medical knowledge.

The online version contains supplementary material available at 10.1186/s12909-026-08647-4.

Figures