Issue: July 25, 2018

Yu K, et al. J Natl Cancer Inst. 2018;doi:10.1093/jnci/djy044.

June 22, 2018
3 min read

Automated approach may improve cervical cancer screening

Issue: July 25, 2018

Yu K, et al. J Natl Cancer Inst. 2018;doi:10.1093/jnci/djy044.

You've successfully added to your alerts. You will receive an email when new content is published.

Click Here to Manage Email Alerts

We were unable to process your request. Please try again later. If you continue to have this issue please contact

An automated risk score algorithm of cervical screening and triage demonstrated similar sensitivity and specificity to cytology for targeting precancerous conditions and identifying women with HPV most in need of treatment, according to study findings.

“The results of this study showed that a computer algorithm matches or exceeds cytology triage performance, confirming our previous proof-of-principle study conducted in a referral population,” Kai Yu, PhD, from the biostatistics branch, division of cancer epidemiology and genetics at NCI, and colleagues wrote. “Thus, the findings strongly support the feasibility of totally automated cervical screening without cytology.”

Cervical screening programs include screening of the general population and triage of screen-positive women to target treatment of precancers. In general screening, HPV testing is becoming increasingly used over cytology — or Pap testing — due to its sensitivity detection capabilities. However, a challenge remains in how to identify the minority of women who test positive with HPV screening who are most likely to have precancer requiring treatment, because many women clear the infection without treatment.

In higher resourced regions, triage consists of partial HPV typing followed by cytology as a second test among HPV-positive women. However, conventional cytology is time consuming. Yu and colleagues designed and evaluated a complete automatable cervical screening strategy to conclude whether an automated algorithm could triage women with HPV as effectively as cytology.

Researchers used data from the Kaiser Permanente Northern California cervical screening program to design a cytologic risk score algorithm based on liquid slide features.

They compared the ability of the algorithm to predict precancer with abnormal cytology

in a cohort of 1,839 women positive for HPV in 2010, as determined by Hybrid Capture 2 (Qiagen).

For prospective validation, researchers compared the algorithm with cytology results among a separate cohort of 243,807 women screened from 2016 to 2017 to determine association between higher risk scores and HPV positivity.

Investigators used cervical intraepithelial neoplasia grade 2 (CIN2), CIN3 and adenocarcinoma in situ as measurements of precancer for algorithm training.

For direct comparison between the algorithm and cytology strategies, two cut-points in the automated risk score determined three risk groups — high, intermediate and low risk — of same sizes as an original cytology strategy:

normal cytology — negative for intraepithelial lesion or malignancy;

minor abnormalities related to HPV — atypical squamous cells of undetermined significance/low-grade squamous intraepithelial lesion; or

higher risk — higher than low-grade squamous intraepithelial lesion.


Cytology results of the 2010 cohort indicated 49% were normal, 45% had minor abnormalities related to HPV and 5.9% were higher risk.

The risk groups defined by each strategy appeared strongly associated with case-control status (P < .001).

Receiver operating characteristic curve analysis based on the risk score for diagnosis of CIN2/CIN3 or adenocarcinoma in situ diagnosis had an area under the curve of 0.71 (95% CI, 0.68-0.74).

The risk score that corresponded to low vs. intermediate/high-risk groups had 0.75 sensitivity (95% CI, 0.7-0.8) and 0.54 specificity (95% CI, 0.51-0.57), whereas cytology had 0.71 sensitivity (95% CI, 0.66-0.76) and 0.53 specificity (95% CI, 0.5-0.56). The difference between the automated and cytology sensitivity and specificity was not statistically significant.

When used for threshold to define highest-risk patients, the algorithm had modestly lower sensitivity and specificity than cytology, but these were not significant.

The algorithm matched the triage performance of abnormal cytology among women with HPV. Combined with HPV16, HPV18 or HPV45 typing, the algorithm referred 91.7% of HPV-positive CIN3 and adenocarcinoma in situ cases to immediate colposcopy and deferred 38.4% of all women with HPV to 1-year retesting, whereas typing combined with triage referred 89.1% of HPV-positive CIN3 and adenocarcinoma in situ cases and deferred 37.4% to retesting.

In the validation cohort, the predicted risk scores strongly correlated with cytology (P < .001), based on Spearman’s rank correlation test. The scores were low among all HPV-negative slides except rare HPV-negative atypical squamous cells rule out high grade, atypical glandular cells and high-grade squamous intraepithelial lesion.

Further researcher is needed to determine whether the algorithm may be improved and used to target missed cases, and to determine whether other systems will exceed the automated approach.

“The ultimate goal remains integration of affordable and high-quality screening/triage with vaccination to promote comprehensive cervical cancer control worldwide,” the researchers wrote. – by Melinda Stevens

Disclosure s : NCI Intramural Research Program funded the study. The authors report no relevant financial disclosures.