Projects

. Introduction

Cancer cervix is amenable to screening and early detection. Although Pap smear is the most widely used tool for cervical cancer screening, it has its own drawbacks in terms of low sensitivity (52%)1 and the need for efficient networking between smear collection and cytology laboratories.2 Besides, varying levels of sensitivity and specificity has been reported based on the grade of intraepithelial neoplasia [(sensitivity and specificity of 93%and 73% in low grade intraepithelial neoplasia (LSIL) versus 64% and 84% in high grade intraepithelial neoplasia (HSIL)]3

Visual Inspection with Acetic acid (VIA) is a simple, cost effective test with a sensitivity that varies from 62.5 to 80%, and a specificity of 80–98.8% and is most suited for resource poor settings.4, 5, 6 Depending on the skills of the staff who perform the test, accuracy of VIA varies widely; the screening performance of physician's assessment was significantly better than the nurse's.7,8

As none of the screening tools have been considered as optimal, researchers have even suggested a combination of VIA/Pap as it increased the sensitivity and specificity for detection of cervical cancer.9,10

A few researchers have tried the feasibility of using commercially available smartphones for acquisition and evaluation of VIA images. Their results have been similar to direct VIA, which shows a high inter observer variability.11, 12, 13, 14

With this background in mind, the current study was designed to evaluate the inter observer variability among gynecologists in assessing cervical images acquired using a specified device.

. Discussion

The study was carried out to evaluate the inter observer variability in interpretation of cervix images among seven gynecologists of varying clinical expertise and experience. The results indicate that there was a wide variation in sensitivity (57.1%–92.9%) and specificity (54.3%–94.5%) among individual specialists in image interpretation. Kappa value varied between 0.11 and 0.493 indicating slight to moderate agreement among them.

Shankaranarayanan R et al. reported pooled sensitivity, specificity, positive and negative predictive values of 80%, 92%, 10% and 99%, respectively with VIA, for detecting cervical intraepithelial neoplasia grade 2 or worse lesions.17 Ghosh et al. found visual inspection after Lugol's Iodine (VILI) (expansion) to be a useful test in screening for cervical neoplasia either as a single tool or in combination with Pap test. They found a high sensitivity of 100% and a specificity of 93.3% with VILI.18

In the present study evaluation was just restricted to VIA and VILI was not considered. Pap smear with its low sensitivity could not be considered as the gold standard and it would have been ideal to have colposcopy guided biopsy. However, the purpose of the analysis was to document the degree of variability in reporting cervix images and not testing the accuracy of reporting; hence the consideration of an expert as the gold standard. The gold standard had used strict criteria while reporting the images, while the others reported based on their learning and experience. Prior to reporting, none of the specialists in any of the three groups had any focused training regarding the interpretation of cervical findings. Our observations indicate the need for periodic training and skill updating of the health care providers in simple techniques such as the interpretation of VIA images.

However, from the reported literature it appears that training the raters also would not minimise the inter observer variability. In their study Vedantham et al. found the highest inter observer variability among the two Gynecologists who were the trainers for other four raters.2 In their study there was a marked variation in the rates of VIA positivity among the six gynecologists who conducted the examinations. VIA positivity ranged from a low of 4% for two gynecologists to a high of 18% for one gynecologist and 31% for another gynecologist. The greatest variation was between providers 1 and 5, the two providers who had received the most training. Similarly in the present study, the group 1 specialists showed highest intragroup variation.

When the gold standard based on similar lines of IFCPC categorisation was compared with the false positive reporting of the specialists, the discrepancies noted were observed in categories, ‘normal’ and ‘squamous metaplasia’. Gland openings, Nabothian follicles in the metaplastic zone were the common reasons for false positivity. In a small number even mucus had compromised the image quality and made the field appear ‘white’.

Vedantham et al. found that among control group VIA positivity was 15.5% in women with inflammation and 6.1% in women without inflammation and the association was found to be statistically significant. (p value < 0.001).2

In their commentary on the issue of inter observer variability, Parashari A and Singh V attributed it to variation in man power training, light source used for visualization, and preparation of diluted (4–5%) acetic acid and its storage.19 In the present study the issue of light variation was overcome by the use of a single device with the same light source.

The challenges in introducing high-quality cytology screening in Low and Middle Income countries (LMIC) have led to the WHO recommendation of screen-and-treat programs using HPV tests and/or VIA followed by cryotherapy for comprehensive cervical cancer prevention and control programs in LMICs.20,21 With such high degree of interobserver variability, using only VIA as screening tool may result in performing an unacceptable number of otherwise unindicated cervical ablative procedures in young women. Hence it is worth exploring the possibility of making the reporting more accurate and precise.

To minimise the interobserver variability, it is important to train the person doing VIA/image acquisition regarding the procedure, clearing of mucus completely without causing abrasions over the surface, specifications during image capturing and about the various categories when there is high possibility of false positive/negative reporting. To bring down the subjectivity, the need for a device with a decision support system that can report the cervix images, is emphasized which probably would provide a more objective and precise report.

Conclusion

There was significant inter observer variability among gynecologists with varying levels of expertise while reporting manual cervix image analysis for detection of cervical epithelial abnormalities. An automated system for cervix image analysis might be able to overcome inter observer variability, and provide for a more reliable and valid interpretation.