Application of machine learning methods to guide patient management by predicting the risk of malignancy of Bethesda III-V thyroid nodules
Authors' conclusion
Our ML models performed well in predicting the nature of Bethesda III-V TN. In addition, our freely available online nomogram helped to refine the RM, identifying low-risk TN that may benefit from surveillance in up to a third of ITN, and thus may reduce the number of unnecessary surgeries.
Study context
- Population
- Adults who underwent surgery for ≥1 Bethesda III–V thyroid nodule with full medical records, 2010–2020; training (940 thyroid nodule (TN)) and validation (395 TN) cohorts1,335 TN
- Participants
- 1,288 adults
- Comparison
- Logistic regression model (LR) vs random forest model (RF) for predicting malignancy of Bethesda III-V thyroid nodules
- Outcome
- Diagnostic performance in predicting the nature and risk of malignancy of a TN
Methodological detailsmulticentre, retrospective cohort
- Multicentre
- Retrospective cohort
Performance
As reported in the abstract
Diagnostic performance in the validation cohort
| Test or model | AUC | Sensitivity | Specificity | PPV | NPV |
|---|---|---|---|---|---|
| LR | 84% (95% CI 78.5%-89.5%) | 57.3% | 90% | 73.4% | 81.4% |
| RF | 82.6% (95% CI 77.4%-87.9%) | 54.7% | 87.6% | 68.1% | 80% |
AUC
LR84% (95% CI 78.5%-89.5%)
RF82.6% (95% CI 77.4%-87.9%)
Sensitivity
LR57.3%
RF54.7%
Specificity
LR90%
RF87.6%
PPV
LR73.4%
RF68.1%
NPV
LR81.4%
RF80%
Further findings
Bethesda category distribution and overall risk of malignancy
- Bethesda III
- 253 (19%)
- Bethesda IV
- 693 (52%)
- Bethesda V
- 389 (29%)
- Overall risk of malignancy
- 35%
Original publication
Application of machine learning methods to guide patient management by predicting the risk of malignancy of Bethesda III-V thyroid nodules. Eur J Endocrinol. Published 17 February 2023. doi:10.
Composed by AI from the published abstract and checked against it. How Briefs are made →