[Skip to Content]
[Skip to Content Landing]
Views 282
Citations 0
Original Investigation
April 2017

Association of Malignancy Prevalence With Test Properties and Performance of the Gene Expression Classifier in Indeterminate Thyroid Nodules

Author Affiliations
  • 1Department of Otolaryngology–Head & Neck Surgery, University of Iowa Hospitals and Clinics, Iowa City
  • 2Department of Pediatrics, Tulane University School of Medicine, New Orleans, Louisiana
  • 3Division of Endocrinology and Metabolism, Department of Medicine, Tulane University School of Medicine, New Orleans, Louisiana
  • 4Department of Biostatistics and Bioinformatics, Tulane University School of Public Health and Tropical Medicine, New Orleans, Louisiana
  • 5Division of Endocrine and Oncological Surgery, Department of Surgery, Tulane University School of Medicine, New Orleans, Louisiana
  • 6Department of Pathology, Tulane University School of Medicine, New Orleans, Louisiana
JAMA Otolaryngol Head Neck Surg. 2017;143(4):403-408. doi:10.1001/jamaoto.2016.3526
Key Points

Questions  What is the validity of the gene expression classifier, and is there variability in test performance among different institutes, and, if so, what is the cause of this variability?

Findings  In this study, the negative predictive value of the gene expression classifier was lower than expected. The variability in test performance was not associated with cancer prevalence alone and may have been associated with inconsistencies in the intrinsic properties of the test.

Meaning  The variability in the test performance limits its utility in practice; large clinical trials are warranted to better define the test’s intrinsic properties.


Importance  It is crucial for clinicians to know the malignancy prevalence within each indeterminate cytologic category to estimate the performance of the gene expression classifier (GEC).

Objective  To examine the variability in the performance of the GEC.

Design, Setting, and Participants  This retrospective cohort study of patients with Bethesda category III and IV thyroid nodules used single-institution data from January 1, 2013, through February 29, 2016. Expected negative predictive value (NPV) was calculated by adopting published sensitivity and specificity. Observed NPV was calculated based on the true-negative rate. Outcomes were compared with pooled data from 11 studies published January 1, 2010, to January 31, 2016.

Results  A total of 145 patients with 154 thyroid nodules were included in the study (mean [SD] age, 56.0 [16.2] years; 106 females [73.1%]). Malignancy prevalence was 45%. On the basis of this prevalence, the expected NPV is 85% and the observed NPV is 69%. If the prevalence is assumed to be 25%, the expected NPV would be 94%, whereas the observed NPV would be 85%. Pooled data analysis of 11 studies comprising 1303 participants revealed a malignancy prevalence of 31% (95% CI, 29%-34%) and a pooled NPV of 92% (95% CI, 87%-96%).

Conclusions and Relevance  In this study, variability in the performance of the GEC was not solely a function of malignancy prevalence and may have been attributable to intrinsic variability of the test sensitivity and specificity. The utility of the GEC in practice is elusive because of this variability. A better definition of the GEC’s intrinsic properties is needed.