Article Text

Download PDFPDF
A Prospective Validation of the IOTA Logistic Regression Models (LR1 and LR2) in Comparison to Subjective Pattern Recognition for the Diagnosis of Ovarian Cancer
  1. Natalie Nunes, MBBS, MRCOG*,
  2. Gareth Ambler, PhD,
  3. Wee-Liak Hoo, MBBS*,
  4. Joel Naftalin, MBBS*,
  5. Xulin Foo, MBBS*,
  6. Martin Widschwendter, PhD and
  7. Davor Jurkovic, MD*
  1. *Gynaecological Diagnostic Outpatient Treatment Unit, University College Hospital;
  2. Department of Statistical Science, University College London; and
  3. Department of Women’s Cancer, University College London, Elizabeth Garrett Anderson, Institute for Women’s Health, London, United Kingdom.
  1. Address correspondence and reprint requests to Natalie Nunes, MBBS, MRCOG, Gynaecological Diagnostic Outpatient Treatment Unit, -1 Floor, Elizabeth Garrett Anderson, University College Hospital, London NW1 2BU, United Kingdom. E-mail: npnunes{at}


Objectives This study aimed to assess the accuracy of the International Ovarian Tumour Analysis (IOTA) logistic regression models (LR1 and LR2) and that of subjective pattern recognition (PR) for the diagnosis of ovarian cancer.

Methods and Materials This was a prospective single-center study in a general gynecology unit of a tertiary hospital during 33 months. There were 292 consecutive women who underwent surgery after an ultrasound diagnosis of an adnexal tumor. All examinations were by a single level 2 ultrasound operator, according to the IOTA guidelines. The malignancy likelihood was calculated using the IOTA LR1 and LR2. The women were then examined separately by an expert operator using subjective PR. These were compared to operative findings and histology. The sensitivity, specificity, area under the curve (AUC), and accuracy of the 3 methods were calculated and compared.

Results The AUCs for LR1 and LR2 were 0.94 [95% confidence interval (CI), 0.92–0.97] and 0.93 (95% CI, 0.90–0.96), respectively. Subjective PR gave a positive likelihood ratio (LR+ve) of 13.9 (95% CI, 7.84–24.6) and a LR−ve of 0.049 (95% CI, 0.022–0.107). The corresponding LR+ve and LR−ve for LR1 were 3.33 (95% CI, 2.85–3.55) and 0.03 (95% CI, 0.01–0.10), and for LR2 were 3.58 (95% CI, 2.77–4.63) and 0.052 (95% CI, 0.022–0.123). The accuracy of PR was 0.942 (95% CI, 0.908–0.966), which was significantly higher when compared with 0.829 (95% CI, 0.781–0.870) for LR1 and 0.836 (95% CI, 0.788–0.872) for LR2 (P < 0.001).

Conclusions The AUC of the IOTA LR1 and LR2 were similar in nonexpert’s hands when compared to the original and validation IOTA studies. The PR method was the more accurate test to diagnose ovarian cancer than either of the IOTA models.

  • IOTA
  • Ultrasound
  • Ovarian cancer
  • Adnexal tumor
  • Pattern recognition
  • Risk of malignancy index
  • Logistic regression
  • LR1
  • LR2

Statistics from

Request Permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.


  • Dr Gareth Ambler received a proportion of funding from the Department of Health’s NIHR Biomedical Research Centres funding scheme.

  • The authors declare no conflicts of interest.