Cervical cancer is the commonest female cancer, a major cause of death and an important public health problem in many developing countries.1 Screening by routine cytology, using Papanicolaou stain (Pap), has had a major impact on cervical cancer rates in many developed countries, but not in any developing country despite its widespread use.2, 3, 4 A number of other cervical screening tests have been proposed. Here, we wanted to identify the most appropriate test for use in developing countries with high rates of cervical cancer and limited resources.
In Peru, as in most Latin American countries, cervical cancer is usually detected in an advanced incurable stage, despite widespread use of cytology screening.5, 6 Others have documented numerous problems of screening in Peru. Samples may be poorly collected, slides may be incorrectly labelled and some never reach the laboratory. Cytology results are sometimes wrongly transcribed or are not reported.6 Cytology laboratories suffer from deficient infrastructure, inadequate training and supervision, staff shortages, lack of follow-up procedures and lack of internal and external quality control leading to suboptimal reading of cytology for which considerable expertise is required. This study was set up to investigate a variety of screening tests that may be more effective in such a setting.
Liquid-based cytology is replacing routine cytology in many countries and, if costs can be minimised, could offer several advantages for a developing country. Persistent human papilloma virus (HPV) infection is the primary cause of cervical cancer and an international group of experts has recently declared that “there is sufficient evidence that testing for HPV infection as the primary screening modality can reduce cervical cancer incidence and mortality rates”.7 Visual inspection after application of acetic acid (VIA) offers a low-cost alternative and has the advantage of instant results. The results of early studies of visual inspection as a screening test have been encouraging.8, 9, 10, 11, 12 This study aimed to compare screening using liquid cytology (LBC), HPV testing, VIA, VIA combined with magnified triage of positive VIA (VIA+M) and routine conventional cytology (CC) in 5,435 women. Here we report the results of the different screening tests.
This study was set up within the TATI project, a screening intervention in the Peruvian Amazonia in the Department of San Martin. The Department of San Martin, with a population of about 800,000,13 has its own Directorate of Health (DIRES), which is divided into 11 health networks. Within each health network, a lead health centre and a team of 1 doctor and 1–4 midwives was trained in screening and treatment of cervical lesions. Teams screened women as part of their routine activities. Referral centres for colposcopy and specialised treatments were established in 3 main towns of the region, and women with confirmed cancer were referred to the National Cancer Hospital in Lima.
Women aged 25–49 years enrolling into the TATI project between February and December 2001 were invited to participate in this study whenever collection kits for LBC and HPV testing were available. Women who said they were pregnant or virgins and those who had had a hysterectomy or conization of the cervix were ineligible. This study was approved by both the Pan American Health Organization (PAHO) and Peruvian Ministry of Health Ethics Committees.
All women who consented to the study were screened with 4 tests. Midwives performed first a gynaecological exam after introducing a speculum.
A cervical sample for routine cytology was collected by means of an Ayre spatula, according to standard (Peruvian Ministry of Health) procedures and was read in a local laboratory. A second sample for liquid cytology was collected using a Rovers-Cervex® brush rotated 5 times clockwise. The brush head was transferred directly into a vial containing CytoRich® Preservative Liquid and kept in a refrigerator (4°C). A third sample for HPV DNA testing was collected using a Digene cervical brush rotated thrice anticlockwise. Collected samples were immediately stored in tubes containing Digene Sample Transportation Medium (STM) and kept in a refrigerator (4°C).
Next, midwives applied a solution of 5% acetic acid, and after waiting for about a minute, examined the cervix for acetowhite areas close to the squamocolumnar junction. VIA was considered positive if any acetowhite lesion was observed in or close to the transformation zone; otherwise it was considered negative. Midwives referred women to the doctor if VIA was positive or directly to the colposcopist if they found evidence of suspected invasive cervical cancer.
Doctors performed an additional visual inspection using an AviScope™ device (4x magnification, white and green light) after reapplying acetic acid (VIAM). VIAM was considered positive if acetowhite lesions were observed, and negative if otherwise. Women testing positive were offered immediate treatment with cryotherapy or referral to a gynaecologist for colposcopy as appropriate. The combination of VIA followed by VIAM if the VIA was positive is referred to as VIA+M.
Both routine and liquid cytology were read using a standard Peruvian classification. Women with moderate dysplasia or worse on either type of cytology were referred for colposcopy. HPV testing was done in London using the Hybrid Capture II kit according to Digene's standard protocol. Women with any positive result on liquid cytology or HPV, who had not already been treated, were contacted by health personnel and “repeat testing” was arranged for ∼12 months after the original sample was taken. Repeat testing included VIA+M, CC, LBC and HPV as on entry.
Biopsies were collected before cryotherapy or at colposcopy. Histology was carried out in a laboratory in Lima and was reported using a standard Peruvian classification (which includes moderate and severe dysplasia, carcinoma in situ and cancer). Roughly speaking, severe dysplasia and carcinoma in situ should correspond to CIN3 and moderate dysplasia to CIN2. Some histology slides were reviewed by an expert gynae-pathologist in England. The original results are used in this paper.
Liquid-based cytology was prepared and read in Lima at the National Cancer Hospital (INEN) using the AutoCyte-Prep® manual system (TriPath/SurePath).
HPV DNA testing was performed, in London using Hybrid Capture II (HC-II, Digene Corporation). Specimens containing the target DNA were denatured and hybridised with an HPV RNA probe cocktail (13 full-length RNA probes recognising oncogenic HPV types 16, 18, 31, 33, 35, 39, 45, 51, 52, 56, 58, 59 and 68). Results are expressed as the ratio of the relative light units (RLU) given by the test specimen to that given by a 1 pg/ml HPV DNA control. For management, the standard threshold of 1 (RLU) identified HPV positive samples. Some analyses used more stringent thresholds of 2, 4 and 10 RLU.
A questionnaire was answered by all women in TATI. Age, ethnicity, education (years at school), province of residence, age at first intercourse, number of lifetime sexual partners, parity, abortion, use of contraceptives (method, time), lifetime number of Pap smears, date and result of the last Pap smear and smoking status (never, former, current) were collected.
On the repeat testing, any woman with positive VIAM, HPV test or with moderate dysplasia or worse on cytology was referred for treatment or colposcopy. All others (including some with mild dysplasia) were considered to be free of high-grade disease and were discharged.
A woman was considered fully evaluated for high-grade disease if she: (i) tested negative on initial VIA+M (including those who were VIA negative), LBC and HPV testing, and had no high-grade disease on routine cytology; (ii) was negative on repeat VIA+M, had low-grade disease or less on LBC and tested negative on HPV; (iii) had a satisfactory colposcopy in which no biopsy was collected because it was considered negative (less than mild dysplasia); (iv) had an adequate biopsy before cryotherapy or during colposcopy or surgical treatment; or (v) was referred to the cancer hospital in Lima for treatment of invasive cancer.
A fully evaluated woman was considered to be free of high-grade disease if no such disease was observed on histology (or no biopsy was taken).
Analysis was carried out using STATA 9.1 (StataCorp, USA).
Sensitivity, specificity and positive predictive values were obtained after adjustment for imputation of missing data and undetermined disease status. A number of women were not examined with VIAM despite having been positive on VIA, and some women did not attend for repeat testing. Therefore, missing results of VIAM in the initial screening and of repeat testing (VIA, VIAM, LBC and HPV) were imputed prior to obtaining sensitivity and specificity of screening tests.
Imputing missing data
Multinomial logistic regressions were used to estimate the probability of different VIAM results (negative, cryotherapy or colposcopy) and of different management (negative, treat or refer) resulting from repeat testing conditionally on the results of the initial tests. Observations on women with missing results were replaced by pseudo-observations with each possible value for the missing test. These pseudo-observations were weighted by the estimated probability of women having the particular result. For women with undetermined disease status, the probability of high-grade disease given the (actual and imputed) screening results was estimated taking into account the probability of being fully evaluated which was dependant on the screening results. This approach is valid under the assumption that the lack of VIAM, repeat testing and biopsy was conditionally independent of the underlying disease status given the available test results (“missing at random”). Confidence intervals for screening parameters were estimated by bootstrapping (10,000 replications) the data and repeating the imputation of missing results.
Samples were collected from 5,595 women recruited by TATI between February and December 2001. Women were recruited from all 11 health networks across the region. Inclusion of TATI participants in this study depended only on availability of resources to offer additional testing at the time of the visit, and so health centres started recruitment at different months in 2001. Overall, 160 women were excluded from the analysis due to ineligibility or lack of screening, leaving 5,435 women with a complete set of screening tests.
Table I shows the characteristics of the participants of the study: 54% of the women in this study were younger than 35 years, and only 9% were aged 45–49. In all respects, those participating in this study were similar to TATI participants who were not recruited to this study.
Table I. Selected Characteristics of Participants
No. of women in TATI1 project but not in this study
No. of women in this study
Data were missing on a = 18, b = 18, c = 5, d = 0, e= 26, and f = 16 women.
Women participating in TATI from February to December 2001.
Age Group (years)
Years of education
Number of pregnancies
Number never pregnant
Number of lifetime sexual partners
Last Pap smear test
Before the study
Ever using hormonal contraception
Table II shows the proportion testing positive for each test (the 4 individual screening tests and VIAM in VIA positive women) both overall and in relation to age and the other test results. The proportion testing positive was highest for VIA (24%) and LBC (16.9% plus 5.5% inadequate) and somewhat lower for HPV (12.6%) and VIA+M (10.1%). The positivity rate for conventional cytology was very low (1.8%), but the inadequate rate was high (11.4%).
Table II. Positive Results of Liquid Cytology (LBC), Conventional Cytology (CC), HPV, Visual Inspection (VIA) and Magnified Triage (VIAM)
Percentage of those with an adequate test having the particular result: Number divided by (5,435-number of missing values) × 100.
Percentage of those in row who were positive on each test (or age group). Those with missing or inadequate test samples have not been deducted from the denominator.
Positivity rates adjusted for VIAM missing (VIA positive and VIAM not done).
Liquid-based Cytology (LBC)
Conventional Cytology (CC)
Not done but VIA+
CIS or worse
There was very poor agreement between the different tests although they are clearly correlated. Only 59% (n = 3213) of women were negative on all 3 new tests (VIA, LBC and HPV) and 2.2% (n = 117) were positive on all 3 tests. Although all but 4 of the 48 women with high-grade conventional cytology had abnormal LBC, there were 15 times as many low-grade abnormalities on LBC as on conventional cytology. Treating each test as either positive or negative, the kappa statistic comparing the 2 types of cytology was only 0.13. The kappa statistics comparing VIA to the other tests was even smaller: 0.03, 0.05 and 0.07 for conventional cytology, LBC and HPV respectively. The best agreement was between HPV testing and LBC (κ= 0.32).
Comparison of VIA positive women who did and did not receive VIAM showed no significance differences in terms of the results of the other screening tests or the demographic variables of Table I (details not shown), suggesting that the assumption of missing at random is reasonable. A quarter (24%) of women were positive on VIA and 518 of 1246 (42%) of those examined on VIAM were also positive, yielding an overall positive rate of 10.1% for VIA+M.
Also presented are the test positivity rates in women with moderate and severe dysplasia and CIS or worse on histology. (Note that sensitivities using imputed histology are given in Table IV).
The relationship between positivity rate and age was quite different for different tests. The proportion of women with abnormal LBC increased slightly with age, and this was mostly attributable to increasing rates of moderate dysplasia or worse (Fig. 1). Abnormal conventional cytology also increased with age. By contrast, HPV and VIA+M positivity decreased with age, as expected.
Imputed disease status
Table III shows how disease status was imputed for the 355 women who were not fully evaluated. Of the 5,435 women in the study, 768 (14%) had a biopsy, 106 (2%) had a negative colposcopy, 3,733 (69%) were negative after initial screening and 473 (9%) were deemed negative on repeat testing (see methods for definition of negative at repeat). There were 148 cases of biopsy-confirmed high-grade disease (79 CIS or worse) and an estimated 26.8 additional cases (6.5 CIS or worse) among women without biopsy.
Table III. Stage of Screening Program at Which Disease Status Became Known or Had to be Estimated
The sensitivity, specificity and PPV of the various screening tests were computed from the imputed data. The results are given in Table IV for moderate dysplasia or worse, severe dysplasia or worse and for CIS+. These are illustrated as a receiver operating characteristics (ROC) curve in Figure 2. Here we concentrate on sensitivity for detecting CIS+, and specificity for absence of moderate dysplasia (or more) on histology. HPV testing was the most sensitive (95.8%). It has been suggested that a threshold of 2 RLU could be used in primary screening, but in this study that would have resulted in a marked drop in sensitivity (to 89.8%) with little gain in specificity. LBC (ASCUS or worse) was the next most sensitive test (80.3%), but at the standard threshold had very poor specificity (83.7%). Changing the threshold to LSIL or worse improves the specificity, but the resulting reduction in sensitivity is of similar magnitude. VIA and CC both performed poorly in terms of both sensitivity and specificity for CIS+, identifying slightly fewer than half the cases. The sensitivities of the different screening tests for moderate dysplasia or worse were similarly ordered, but with a marked drop of sensitivity for HPV testing and CC and with significant improvements for VIA and VIA+M. Interestingly (when excluding CIS+), VIA was the most sensitive for detecting moderate and severe dysplasia (67.5% compared to 59.6% for both LBC and HPV), whilst CC failed to detect the vast majority of these lesions (sensitivity 10.5% for moderate and severe dysplasia).
Table IV. Sensitivity, Specificity and Positive Predictive Values of Screening Tests in 5,435 Women
Apart from laboratory testing, this study was conducted under routine screening conditions in an area of Peru with poor roads and little infrastructure. Population lists were not available so women were screened either opportunistically when they attended a health post or as part of a campaign targeting all women in a small well-defined area. Additionally, community promotion teams specially trained to educate women about cervical cancer prevention encouraged women to participate. Although over 80% of participants reported a previous Pap smear, these smears were only rarely reported as dysplastic and very few women had been treated for cervical intraepithelial neoplasia.5 Screening was done by midwives as part of their routine workload. It is unlikely that a national screening programme in Peru could be based on specialised screening personnel, and this is perhaps the only cervical screening study not to have used dedicated personnel. Despite the difficult conditions including lack of refrigeration during transport so that samples reached temperatures above 30°C, there were no major technical difficulties with any of the screening tests. The biggest challenges for LBC and HPV were related to communication and management of women with positive results who had not already been treated (due to positive magnified triage). Tracing women to return results and arrange follow-up care could be a major challenge to programs using either of these methods for routine screening, as has been the case with CC.
In the study, CC and VIA, both performed less well than in many other studies. The results, however, are very similar to those of the LAMS study carried out in Brazil and Argentina.14
When follow-up and referral for treatment are difficult, there is a substantial advantage of having an immediate result from screening. This, however, has to be balanced against the over-treatment that would result from offering cryotherapy (or LEEP without prior biopsy) to all women who screen positive. Whilst triage tests may be thought of as a luxury for settings such as these, it needs to be determined whether treating a quarter of screened women would be acceptable to the women concerned (24.2% were positive on VIA and 12.6% on HPV). In our study the advantages of “see and treat” were often not realised because doctors were unavailable at the time of screening.
Visual inspection methods
VIA is a simple and fast exam giving immediate results. It does not require the examiner to be highly qualified. The positivity rate of 24% for VIA is similar to that of other studies,15, 16, 17, 18 the sensitivity was relatively poor but not exceptional: 3 of the 11 centres in the IARC VIA multi-centre study had sensitivities of <60%.12
Three cervical samples were collected from the cervix before the VIA was done. The process of collecting these samples combined with the high prevalence of vaginal infections (26% of LBC slides showed inflammatory changes) sometimes caused bleeding which made visual inspection difficult.
Doctors performed magnified triage of VIA to confirm the midwife's positive result and to offer immediate treatment with cryotherapy if appropriate. Addition of this triage substantially improved the specificity of visual inspection, but further eroded the sensitivity.
This was the first time LBC was used in Peru and all slides were read by a single cyto-pathologist which would not be possible in routine screening. Converting the cytology laboratory in Lima to prepare AutoCyte slides was straightforward. Since the smear-taker was not responsible for slide preparation or fixation, slides were unlikely to be assessed as inadequate, equivocal or borderline because of poor fixation.19 The inadequate rate in this study (5%) was high compared to most other reports. The majority (65%) were due to bleeding, with the rest being either hypocellular or acellular.
Despite limited efforts to improve routine cytology (re-training of cyto-technicians in Lima, centralisation of the Pap laboratories, improvement in the information system and intensification of supervision), positivity rates were extremely low by any standards and in particular in comparison with concurrent LBC.
Many studies in developed countries have shown the greater sensitivity of HPV testing (compared to cytology) for detecting high-grade CIN,20, 21, 22 but it is unclear whether and how HPV testing should be used in a low-resource setting. The performance of 4 screening tests (HPV testing, LBC, CC and cervigrams) and their combinations were evaluated in a study in Costa Rica. No significant differences were found between the sensitivity and specificity of HPV testing (85.3%, 85.7%) and LBC (85.7%, 87.8%), while the sensitivity of CC (read by an expert cyto-pathologist) was 63%, substantially higher than in our study.23
The overall prevalence of high-risk HPV infection in this study was 13%, similar but slightly higher than that of other studies in South America.24, 25, 26 Studies of self-sampling suggest that HPV infection tends not to be limited to the immediate lesion, so high-quality sampling is less critical than it is for cytology. Additionally, the Hybrid Capture II test can be automated; the results are not subject to human judgement; they have been successfully implemented in several Latin American laboratories and have shown excellent inter-laboratory agreement.14, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36
There was only one insufficient sample for testing, a remarkable achievement compared to inadequate rates for cytology (5% for LBC and 11% for CC).
HPV testing was substantially less sensitive for moderate dysplasia (77.3%) than it was for severe dysplasia or worse (89.4%). The relevance of moderate dysplasia is called in to question by the results of the pathology review, which suggests there was some overgrading in the Peruvian laboratory. Of 23 biopsies originally classified as moderate dysplasia, only 2 (9%) were classed moderate dysplasia or worse on review. By contrast, of 37 biopsies originally classified as CIS or worse, 32 (91%) were moderate dysplasia or worse on review.
The cost per test of using an HPV assay or liquid cytology is much greater than that of routine cytology or visual inspection with limited quality assurance. Nevertheless, an occasional accurate but expensive test may be more cost-effective than frequent use of routine cytology or VIA. A new HPV DNA test for developing countries is under development (in collaboration between Digene Corporation and PATH). The test will be fast (results will be available within 3 h allowing a “see and treat” approach) and inexpensive. It will only require basic laboratory infrastructure, a source of electricity (battery) and could be performed by any trained technician.37
1Routine cytology as done in San Martin is ineffective in preventing cervical cancer. If routine cytology is to be used, substantial investment is required to improve its quality.
2VIA performed by midwives during their routine work has many advantages, but its sensitivity in this study was around 50%. The potential of VIA carried out by experienced or specialised screening personnel with good quality assurance is unknown in this setting.
3Liquid based cytology offers some advantages over routine cytology, but at a cost. Slides are still subject to human interpretation and high-quality cyto-screening may be difficult to achieve nationally.
4HPV screening is feasible in low-resource settings even when samples are subjected to high-temperatures and transport delays. Hybrid Capture II is not geared for this setting and results using the faster and cheaper test that is under development will be of great interest.
In summary, cytology and visual inspection methods both suffer from their dependence on human skill and judgement. In that respect HPV testing offers the best prospect for cervical screening, although it is currently costly and would require establishing laboratory and transport networks.
We thank Dr. Juvenal Sanchez for reading our histology; the cytology laboratory at INEN for testing our LBC samples, and Dr. Linda Ho and Dr. Philip Londesborough for testing our HPV samples in the CRUK Centre for Epidemiology, Mathematics and Statistics Laboratory, London.
We are very grateful to all the women who participated in this study, the midwives and doctors who performed visual screening techniques, colposcopists, the TATI office staff (in particular Dr. M. Gonzales and Mr. Jose Delgado for data management and follow-up monitoring), the Promotoras and PATH team for ensuring recruitment, the Direccion Regional de Salud (DIRES) San Martin for infrastructure, the Maes-Heller Cancer Research Centre and the PAHO Office in Lima (OPS) for their collaboration with preparing and sending HPV samples to the UK.
The TATI project was conducted with the generous support of the Bill & Melinda Gates Foundation to the Pan American Health Organization and PATH through the Alliance for Cervical Cancer Prevention.