Abstract
Diagnostic errors can pose a serious threat to patient safety, leading to serious harm and even death. Efforts are being made to develop interventions that allow physicians to reassess for errors and improve diagnostic accuracy. Our study presents an exploration of misdiagnosis patterns mined from PubMed abstracts. Article titles containing certain phrases indicating misdiagnosis were selected and frequencies of these misdiagnoses calculated. We present the resulting patterns in the form of a directed graph with frequency-weighted misdiagnosis edges connecting diagnosis vertices. We find that the most commonly misdiagnosed diseases were often misdiagnosed as many different diseases, with each misdiagnosis having a relatively low frequency, rather than as a single disease with greater probability. Additionally, while a misdiagnosis relationship may generally exist, the relationship was often found to be one-sided.
Introduction
Diagnostic errors are a silent crisis in healthcare, affecting an estimated 12 million Americans annually and contributing to nearly 60% of malpractice claims involving serious harm. While many medical AI tools focus on identifying the most likely diagnosis based on symptoms, they rarely account for the "great mimics"—diseases that look like something else. If we can quantify which diseases are most frequently confused, we can build "safety nets" into diagnostic software. These systems could prompt a physician to reconsider a diagnosis by highlighting common pitfalls (e.g., "You've diagnosed X, but in the literature, X is frequently a misdiagnosis for Y").
Methodology
To map these patterns, we developed a pipeline to extract misdiagnosis relationships from the PubMed 2018 annual baseline, a corpus of nearly 28 million citations.
Linguistic Filtering and Entity Resolution
We used Python to parse the corpus, targeting titles containing phrases like "misdiagnosed as" or "masquerading as." Named Entity Resolution was performed using QuickUMLS, mapping disease names to Universal Medical Language System (UMLS) Concept Unique Identifiers (CUIs). We focused specifically on semantic types related to "Disease or Syndrome" and "Neoplastic Process."
Selection and Normalization
To ensure high data quality, we applied strict selection logic. We only selected titles that contained exactly two medical concepts—one preceding the misdiagnosis phrase (the correct diagnosis) and one following it (the incorrect diagnosis). Titles with more complex structures or multiple concepts were excluded to maintain the integrity of the source-destination relationship.
We further normalized the data using the UMLS API to handle synonyms and hierarchical relationships. For instance, specific tumor types were grouped under broader parent CUIs to prevent data fragmentation.
Results and Key Insights
From 5,105 relevant titles, we extracted 2,502 unique misdiagnosis pairs. These were used to build a directed, frequency-weighted graph where nodes represent diseases and edges represent the direction and frequency of misdiagnosis.
The "Great Mimic"
Tuberculosis (TB) emerged as the most frequently misdiagnosed disease. It was mistaken for 46 different conditions, most commonly various forms of carcinoma. This confirms long-standing clinical observations of TB's ability to imitate other pathologies. TB had the highest "out-degree" (46 different incorrect diagnoses) and the highest source frequency (68 mentions).
Asymmetric Relationships and Structural Bias
Our analysis revealed that misdiagnosis is rarely a two-way street. If Disease A is often mistaken for Disease B, it does not necessarily mean Disease B is often mistaken for Disease A. This one-sided directionality suggests that certain diagnostic pathways are inherently more prone to error. Furthermore, "destination nodes" (the incorrect diagnoses) were often broader categories like "Tumor" or "Malignancy," which appeared as the incorrect diagnosis 73 times.
The "Many-to-Many" Problem
Most diseases are not consistently mistaken for just one other thing. Instead, a single disease tends to be misdiagnosed as many different conditions, each with a relatively low individual probability. For example, even for common errors like TB, the normalized frequency was usually less than 13% for any single incorrect diagnosis.
Discussion and Future Work
While this study provides a data-driven foundation for clinical decision support, it has limitations. Analyzing only article titles led to significant data loss. Expanding the scope to abstracts and full-text articles using advanced NLP tools like SemRep is a logical next step. Additionally, the study measures how often a misdiagnosis is reported in literature, not necessarily how often it occurs in the general population. Future models must factor in baseline disease prevalence to provide more accurate risk assessments. The ultimate goal remains the integration of these weighted graphs into Electronic Health Records (EHRs) to provide real-time diagnostic alerts to clinicians.
Summary updated