Human papillomavirus triage of women with persistent borderline or mildly dyskaryotic smears: Comparison of costs and side effects of three alternative strategies

Authors


Abstract

The conventional direct referral to colposcopy of persistent borderline or mildly dyskaryotic (BMD) smears in cervical cancer screening leads to considerable unnecessary referrals and associated anxiety and costs. This may be improved by including testing for oncogenic human papillomavirus (HPV) in the triage. We assessed costs and side effects (referrals, treatments and time in follow-up) for 3 possible HPV triage strategies (immediate HPV testing, a 6-month delay in HPV testing, a 2-stage combination of both) and compared them with the conventional strategy. The assessments are based on recent Dutch data from various national databases and trials. We estimated that the referral rate could be reduced by 49, 58 and 58% with immediate, delayed and 2-stage HPV testing, respectively. As a consequence, the average length of follow-up, as well as average costs, also decrease. Therefore, we advocate including HPV testing before referring to colposcopy. Among the 3 HPV strategies, analysis of additional aspects favors implementation of immediate HPV testing. © 2007 Wiley-Liss, Inc.

Cytology-based cervical cancer screening prevents deaths by treating preinvasive (cervical intraepithelial neoplasia, CIN) and early invasive disease. Low-grade abnormalities described as borderline and mild dyskaryosis (BMD) are the most common type of cytologic abnormalities, ranging in frequency from below 2% in the Netherlands to above 6% in Finland.1 In several countries, it is currently recommended to follow-up these women with a Pap smear in 6 months.2, 3, 4 Women are then referred for a colposcopy if the BMD abnormality does not normalize.3 In the Netherlands, about one third of the women with BMD primary screening smears is eventually referred,1 most often because of BMD persistence and less often because of cytological progression.5

This diagnostic policy of following-up BMD smears induces a considerable amount of side effects in terms of a high number of referrals and a long follow-up period with associated costs and psychologic consequences. About 60–90% of women with BMD persistence have no high-grade lesion that needs to be treated.6, 7, 8, 9, 10, 11, 12, 13 The burden on women and the health care could be reduced if the subgroup at high risk for a significant lesion were identified.

It has been consistently shown in the literature that detection of an infection with one of the high-risk human papillomavirus types (HPV) can be used for a risk stratification of women with low-grade abnormal smears.14, 15 An infection with HPV is a necessary factor in the development of invasive cervical cancer.16 Because no histological progression is seen in women who spontaneously clear the HPV infection,17 women without a detectable HPV infection do not need further follow-up. That could be the case in 40–60% of women with persistent BMD smears.7, 8, 9, 10, 11, 12, 13 The discussion about the optimal strategy and time points of incorporating HPV testing into triage of women with BMD smears is still on-going.18, 19, 20

To inform this discussion, we investigated the costs and the side effects of 3 strategies of managing women with persistent (defined as 2 consecutive) BMD smears using HPV testing as a triage tool. Both the side effects and the costs are compared to those associated with the conventional management (i.e., direct colposcopy of all women without first assessing their HPV status).

Methods

Triage strategies

We analyzed 3 possible HPV triage strategies of women with two consecutive BMD smears (Fig. 1; corresponding to ASC-US+ASC-H+LSIL in the Bethesda 2001 classification21). The difference among these strategies is in the timing of HPV testing, and the consequent referral to colposcopy:

  • A––immediate HPV triage: the co-collected HPV sample is analyzed immediately when BMD persistence is established (t = 0), and all HPV-positive (HPV+) women are referred to colposcopy;

  • B––delayed HPV triage: an HPV sample is collected 6 months after the second BMD smear (t = 6), and all HPV+ women (i.e., those who have not cleared the virus) are referred to colposcopy;

  • C––2-stage HPV triage: the co-collected HPV sample is analyzed immediately when BMD persistence is established (t = 0), HPV+ women are retested for HPV visit at t = 6, and all women who remain HPV+ (i.e., those who have not cleared the virus) are referred to colposcopy.

Figure 1.

Strategies A, B, C and D (conventional strategy) for triage of persistent BMD smears. Shaded rectangles represent women who need to be referred, whereas the rectangles with a bold border represent women who can return to the regular screening programme.

We assumed that in strategies A and C the HPV samples are co-collected when the woman presents at the general practitioner (GP) to have a follow-up Pap smear after a BMD primary screening smear (both conventional cytology). These HPV samples are investigated if the follow-up smear is read as BMD. Women with 1 negative HPV test return to a normal screening schedule. The 3 HPV strategies were compared for side effects and costs to:

  • D––conventional strategy: direct colposcopy, i.e. a referral at t = 0 of all women with 2 consecutive BMD smears without a prior assessment of their HPV status.

Quantification of side effects and costs

We assessed the side effects and costs per woman with 2 consecutive BMD smears for the period after the second BMD smear. This period includes all HPV sampling for triage and the complete postreferral management.

In the Dutch cervical cancer screening programme, roughly 1 in 4 women with a primary BMD smear has a follow-up (=second) BMD smear; the remaining 3 in 4 women have either a negative follow-up smear or a highly abnormal (>BMD) smear.5 Because of co-collection at t = 0 in strategies A and C, when the outcomes of cytological testing are not yet known, HPV samples would need to be taken from all women with a primary BMD smear. This represents extra costs compared to the conventional strategy, and needs to be taken into account in the analysis. Therefore, in these 2 strategies we assumed that 3 extra HPV samples need to be collected for each woman with 2 consecutive BMD smears. We further assumed that only those collected samples that are relevant for our analysis, i.e. 1 in 4, are read in the laboratory.

The proportion of women referred, the proportion treated and the time needed to complete the recommended follow-up are quantified from epidemiological data from a recent Dutch trial performed at the Erasmus Medical Center (Erasmus MC) reported earlier.7 This trial aimed to evaluate the potential to prevent unnecessary diagnostic procedures and treatments by doing the HPV triage of women with 2 consecutive BMD smears corresponding to strategy C. At enrolment, the gynecologist took an HPV sample and biopsies from all colposcopic abnormalities but treatment was deferred. Women with an HPV+ high-grade (CIN 2/3) lesion at enrolment were followed-up 6 months later with an HPV test and a colposcopically guided biopsy. If the HPV+ CIN 2/3 lesion persisted, the woman was treated. If instead within 6 months the woman cleared the virus or no CIN 2/3 lesion could be established anymore, she was seen at the exit visit 12 months after enrolment together with those women who tested HPV-negative at enrolment or had (at most) a low-grade lesion (CIN 0/1). At this exit visit, women were tested for HPV and underwent colposcopy. At enrolment, 51% of women had detectable HPV (Table I). Nineteen percent of all enrolled women were found to have a CIN 2/3 lesion at t = 0, i.e. 35% of all HPV+ and 2% of HPV− women. The observed 12-month progression and persistence proportions of (untreated) CIN 0/1 lesions (CIN 0 to >CIN 0, and CIN 1 to ≥CIN 1; both transitions may prompt treatment during post-colposcopic follow-up) were dependent on the HPV status at enrolment (Table II). No cancer was found during follow-up.

Table I. HPV Prevalence and CIN Lesions
 Base caseRange (literature)Outcomes affected
  • t = 0: at time of the second BMD smear; t = 6: 6 months after the second BMD smear.

  • a

    Estimated by the multistate Markov model from the Erasmus MC trial (see Methods).

  • b

    Not varied by HPV status, as observed in the Erasmus MC trial.

(1) HPV prevalence at t = 051%740–60%8-13Equals the referral rate in strategy A.
(2) Persistence of HPV infection from t = 0 to t = 682%a60–85%9, 13Multiplied with (1), it equals the referral rate in strategies B and C.
(3) CIN 2+ among women regardless of HPV status at t = 019%9, 11-1310–40%6, 8-13Equals the proportion of women treated immediately upon referral in strategy D.
(4) CIN 2+ among women HPV+ at t = 035%720–50%9, 11-13Multiplied with (1), it equals the proportion of women treated immediately upon referral under strategy A.
(5) CIN 2+ among women with a persistent HPV infection at t = 644%a30–60%Multiplied with (2), it equals the proportion of women treated immediately upon referral under strategies B and C.
(6) CIN 0:CIN 1 ratio, regardless of HPV status44%:56%7 Multiplied with (1–(3)) and the respective progression rates (Table II), it equals the proportion of women treated at follow-up in strategy D. Multiplied with the cost per CIN 0 or CIN 1 case (Table IV), determines the total cost of referral of CIN 0/1 in strategy D.
(7) CIN 0 : CIN 1 ratio, HPV+ women29%:71%7 Multiplied with ((1) × (1 − (4))) for strategy A or ((1) × (2) × (1 − (5))) for strategies B and C, and the respective progression rates, it equals the proportion of women treated at follow-up in strategies B&C. Multiplied with the cost per CIN 0 or CIN 1 case, determines the total cost of referral of CIN 0/1 in strategies A, B and C.
(8) CIN 2:CIN 3 ratio75%:25%b Multiplied with (3), (4) or (5) (depending on the strategy), the recurrence rates after treatment, and the cost per CIN 2 or CIN 3 case, determines the total cost of referral of CIN 2/3 in all strategies.
Table II. Proportion of Women with Persistence or Progression of the Initial Lesion, as Observed During 12 Months in the Erasmus MC Trial
  • a

    Determined by pooling the trial outcomes for HPV+ and HPV− women. Used for evaluation of strategy D.

  • b

    Used for evaluation of strategies A, B and C.

  • c

    Used for evaluation of all four strategies.

Initial CIN 0 lesion to >CIN 0a14%
HPV− at enrolment7%
HPV+ at enrolmentb30%
Initial CIN 1 lesion to ≥CIN 1a38%
HPV− at enrolment26%
HPV+ at enrolmentb48%
Posttreatment recurrent or residual lesions in CIN 2/3c10%26

In the Erasmus MC trial, the HPV status and the distribution of CIN lesions at t = 6 were not directly observed for all women. We applied a multistate Markov model on the longitudinal data of all HPV+ women at enrolment (n = 54), using the HPV and CIN prevalence observed at t = 0 (for all 54 women), at t = 6 (for the subgroup with HPV+ CIN 2/3 at enrolment), and at t = 12 (for women with HPV+ CIN 0/1 at enrolment, and women without persistent HPV+ CIN 2/3 at t = 6). The result of this model was an estimate of HPV and CIN prevalence for all 54 women at t = 6. We used the msm package for the statistical program R version 0.6.3 (Christopher Jackson, Dept. of Epidemiology and Public Health, Imperial College, London). We allowed for the following transitions: HPV clearance, CIN progression and CIN regression, and assumed that women treated at t = 6 would without treatment have remained in the same state until t = 12. This lead to an estimate that 18% of women HPV+ at t = 0 cleared this infection by t = 6 (Table I). With the same model we also estimated that 44% of women who do not clear the HPV by t = 6 have CIN 2/3.

For the HPV prevalence and persistence rates and the CIN 2/3 prevalence, we extracted the plausible ranges from the literature6, 8, 9, 10, 11, 12, 13 and used them as the basis for univariate sensitivity analyses. In all cases where independent observations were available, these formed an interval around the point estimates observed in the Erasmus MC trial (Table I).

We defined the average duration of follow-up as the period following the second BMD smear in which women are further triaged through the HPV test (strategies A, B and C), referred and followed-up. We assumed compliance with the recommended follow-up (surveillance) after colposcopy. In the Netherlands, women with CIN 0/1 at initial referral are not treated but it is recommended to follow them up with 2 smears within 12 months; women with CIN 2/3 at initial referral are offered treatment and are then followed-up with 3 surveillance smears within 24 months.22 If in either case at least 1 surveillance smear is abnormal, the woman is referred for colposcopy again and eventually (re-)treated. We accounted for the extra follow-up time due to surveillance based on the progression/persistence rates of CIN 0/1 observed in the Erasmus MC trial, and on posttreatment residual/recurrence rates for CIN 2/3 extracted from the literature (Table II). In this way, we neglected the extra follow-up time because of negative colposcopies in women with (false-)positive surveillance smears. We assumed that the abnormal surveillance smear is found at the mid-point of the recommended follow-up interval. The assumed time needed between the successive management steps is then as follows: 2 months from the positive triage test to colposcopy, 1 month from colposcopy to treatment, 6 months for postcolposcopy surveillance of CIN 0/1 and 12 months for post-treatment surveillance of CIN 2/3.

We estimated that from the moment the referral advice is given it takes on average 10, 14 and 17 months to complete follow-up for CIN 0, CIN 1 and CIN 2/3, respectively. These estimates, used for the evaluation of the conventional strategy D, are based on the combination of the observed persistence/progression proportions (Table II), and the recommended length of postreferral follow-up, both per CIN stage and regardless of the HPV status. In strategies A, B and C all referred women are HPV+. Untreated HPV+ CIN 0/1 lesions are more likely to be referred once again because of higher persistence/progression proportions of the lesion than the HPV− CIN 0/1 lesions (Table II). In these strategies, therefore, the estimated average follow-up time per CIN stage increases to 13 and 15 months for CIN 0 and CIN 1 lesions, respectively.

Direct medical costs, and the time and travel cost incurred by women were included in the analysis. The costs per procedure, and the average number of diagnostic and treatment procedures during postreferral management (diagnostics, treatment and follow-up) per CIN grade regardless of the HPV status, represent the recent Dutch situation23, 24 (Tables III and IV). The average number of procedures per CIN stage23 are based on the assumptions that (i) the given CIN grade is the maximum CIN grade, and that therefore women with CIN 0 are never treated, (ii) 44% of women whose CIN 1 lesion is expected to persist or progress25 are subsequently treated, (iii) all women with CIN 2/3 are treated immediately, and (iv) 10% of the women treated will need retreatment.26 These assumptions and the estimated treatment modalities published earlier24, 27 were validated against the most recent available individualized national data on diagnoses, diagnostic procedures and treatments.5, 28 For the purpose of the present cost analysis, we also accounted for progression of CIN 0, and at least persistence of CIN 1 lesions by HPV status (see Table II). We assumed that all women showing progression (or persistence of CIN 1) are treated.

Table III. Average Number of Diagnostic and Treatment Procedures Per CIN Grade Regardless of HPV Status, Based on National Data for the Netherlands23
 Histology
CIN 0CIN 1CIN 2CIN 3
  • a

    Loop excision of the transformation zone.

Diagnostic procedures:    
 Colposcopy2.114.454.775.05
 Smear by a gynaecologist1.612.593.103.13
 Biopsy0.370.630.650.62
 Endocervical curretage0.130.370.350.38
 Smear by a GP0.400.510.440.32
Treatment procedures:    
 LETZa0.130.420.930.75
 Conization0.010.040.150.30
 Hysterectomy0.000.020.020.04
Table IV. Unit Costs (€ 2005) of Medical Procedures, Visits and Hospital Stays, Based on the Recent Dutch Data23, 24
  • a

    Includes the visit at the GP and the collection of sample material (€ 21), laboratory cost (€ 17) and costs of the woman (€ 6).

  • b

    The laboratory cost. Collection of sample material and costs of the woman are included in the cost of colposcopy.

  • c

    Includes visit at the GP and collection of sample material (€ 21), laboratory costs (€ 33), and costs of the woman (€ 6).

  • d

    Treatment costs include the charge per type of treatment (LETZ € 294, conization € 477, hysterectomy € 1,062), cost of outpatient visit (if the procedure is performed in an outpatient setting; € 64), cost of hospital days (day care € 229, hospital day € 359), preoperative diagnostics (for conization or hysterectomy € 98) and costs for the woman (€ 9 for an outpatient visit, and € 42 per treatment day).

Procedure 
 Pap smear, taken by a GPa45
 Pap smear, taken by a gynaecologistb25
 Co-collected HPV test1
 HPV test at an extra GP appointmentc59
 First colposcopy86
 Second or later colposcopy63
 Biopsy50
 Endocervical curettage81
 LETZd490
 Conizationd1,195
 Hysterectomyd4,176
Average total cost (referral + postreferral management):
 CIN 0336 (420 if HPV+)
 CIN 1828 (884 if HPV+)
 CIN 21,239
 CIN 31,432

Results

While all women undergo a colposcopy under the conventional strategy (strategy D), only 51, 42 and 42% would have been triaged to it by HPV under strategies A, B and C, respectively (Table V). The expected total detection rate of women with high-grade CIN lesions (CIN 2/3) would be 18, 18, 18 and 19% of the eligible women under strategies A, B, C and D, respectively. The avoided referrals would therefore predominantly concern women with at most low-grade CIN lesions (CIN 0/1). While 81% of all women with 2 consecutive BMD smears turned out to have CIN 0/1 at referral in the trial (strategy D), this would only be 33, 23 and 23% under strategies A, B and C, respectively. As a consequence, the total treatment proportion would amount to 32, 28, 28 and 41% under strategies A, B, C and D, respectively. This proportion includes immediate treatment of women with CIN 2/3 lesions at referral, and later treatment of initially untreated women (CIN 0/1 at referral) who show abnormalities in follow-up.

Table V. Results: Number of Procedures, Side Effects, and Costs (€ 2005) Due to HPV Triage, Diagnostic Assessment and Treatment Per Strategy, and Per Woman with Two Consecutive BMD Smears. Base Case Assumptions
 Triage strategy
A (immediate triage)B (delayed triage)C (two-stage triage)D (direct colposcopy)
  • t = 0, immediately after the second BMD smear; t = 6, 6 months after the second BMD smear. n.a., not applicable.

  • a

    E.g. for strategy C: It is observed in the Netherlands that 25% of all BMD primary smears have a repeat BMD smear.5 This means that for every woman with the second BMD smear, 4 cocollected HPV samples need to be taken in total.

  • b

    E.g. for strategy C: 51% of women with two BMD smears are HPV+ at t = 0, and are retested for HPV at t = 6.

  • c

    E.g. for strategy C: 51% × 82% HPV persistence rate (Table I).

  • d

    Directly observed in the Erasmus MC trial for strategies A and D (see Table I), and estimated for strategies B and C.

  • e

    E.g. for strategy C: women with HPV+ CIN 2/3 (18%, see Table V) + persistence or progression in women with initial CIN 0/1 lesions (7% × 30% + 17% × 48%, see Tables II and V).

  • f

    E.g. for strategy C: 51% HPV+ at t = 0 × 6 months in triage + post-triage follow-up (7% × 13 months for CIN 0 + 17% × 15 months for CIN 1 + (14% + 5%) × 17 months for CIN 2/3, see Table V & Methods).

Number of procedures, side effects
 HPV tests–cocollected samplesa4.000.004.00n.a.
 HPV tests–samples collected at extra GP visitsb0.001.000.51n.a.
 Proportion of women referredc51%42%42%100%
 Detection of CIN 0d10%7%7%36%
 Detection of CIN 1d24%17%17%45%
 Detection of CIN 2d13%14%14%14%
 Detection of CIN 3d4%5%5%5%
 Proportion of women treatede32%28%28%41%
 Average time in follow-up (months)f7.712.49.512.9
Costs
 HPV tests375967n.a.
 Referrals to colposcopy478412412740
 Total515472479740
HPV tests (extra GP visits)
 Per avoided referral0.01.70.9n.a.
 Per avoided treatment0.07.84.0n.a.

We estimated that the total follow-up after the second BMD smear takes 13 months on average under the conventional strategy D (Table V). This period predominantly reflects the time needed to complete the recommended management after referral. Therefore, the average length of follow-up per strategy is strongly affected by the lower referral rates in the HPV strategies. These outweigh the prolongation of the prereferral period due to the extra HPV triage, so that in the end the completion of follow-up after the second BMD smear would on average take 8, 12 and 10 months under strategies A, B and C, respectively. The expected average difference between strategies A and C is small though it should be noted that the subgroup that is referred based on delayed HPV testing (strategy C) is at a disadvantage. This is because they spend extra time in triage while the fact that they are referred and their postreferral management do not change.

We estimated that the total cost to manage a woman with 2 consecutive BMD smears under the conventional strategy (D) is € 740 (Table V). HPV testing itself would add to the triage costs, but these extra costs would be lower than the savings because of fewer referrals. The resulting difference is most favourable for strategy B (a decrease of 36% compared to strategy D), followed by strategies C (35%) and A (30%).

In the sensitivity analysis, we varied the epidemiologic assumptions with ranges from the literature reported in Table I. In Table VI, we present the effects on the referral rate, the average length of follow-up and total costs. The ranking of strategies does not change. All 3 HPV strategies would remain more favorable than the conventional strategy. Under all investigated possibilities except when a lower HPV persistence rate is assumed (60% instead of 82%), strategy A would remain the most favorable in terms of the time needed to complete the total recommended follow-up. When the lower HPV persistence rate is assumed, strategy C could decrease the average time in follow-up to the same level as strategy A. The proportion of women referred and treated, as well as the total cost, would remain the lowest under strategies B and C: the referral rate in the range of 31–49%, and the cost decrease compared to strategy D in the range of 33–51%. Our results are most affected by the changes in the assumptions on HPV prevalence and its persistence within 6 months (direct observations for the latter are not available). This is not surprising since these determine how many women will be ultimately referred for colposcopy.

Table VI. Results: Triage Outcomes (Referral Rate, Length of Follow-Up) After Changes in Assumptions on HPV Prevalence and Persistence, and CIN 2/3 Prevalence. Sensitivity Analysis
 Strategy AStrategy BStrategy CStrategy D
% ReferredFollow-up (months)Total cost% ReferredFollow-up (months)Total cost% ReferredFollow-up (months)Total cost% ReferredFollow-up (months)Total cost
  • a

    Baseline assumption, 51%; range, 40–60%.

  • b

    baseline assumption, 82%; range, 60–85%;

  • c

    baseline assumption, 19% (all at t = 0), 35% (HPV+ at t = 0) and 44% (HPV+ at t = 6); range, 10, 20, 30% (low), 40, 50, 60% (high).

Baseline assumptions517.75154212.4472429.547910012.9740
Lower HPV prevalencea406.14123311.0383337.438410012.9740
Higher HPV prevalencea609.15994913.55444911.155710012.9740
Lower HPV persistenceb517.75153110.7361317.836910012.9740
Higher HPV persistenceb517.75154312.6487439.749410012.9740
Lower CIN 2/3 prevalencec517.64744212.3440429.444810012.6679
Higher CIN 2/3 prevalencec517.95564212.6508429.651510013.9882

Discussion

Our analyses showed that compared to direct referral of women with 2 consecutive BMD smears to colposcopy based on HPV testing can prevent at least 1 out of 2 colposcopies and treatment in 1 of 3 women with at most low-grade lesions. It can decrease the average follow-up time by half a year, and reduce the associated average total costs by a third. In the population of 3.4 million women at risk (i.e., with a cervix) aged 30–60 in the Netherlands, around 8,700 annually have a BMD primary screening programme smear in the screening programme, of whom around 2,200 are referred because of BMD persistence.5 Even if strategy A with the lowest expected cost decrease would be adopted instead of the currently recommended direct colposcopy (strategy D), total annual savings of close to € 0.5 million could be attainable in this group of women in the Netherlands. BMD primary smears outside of the screening programme account for roughly half of the BMD primary smears in the Netherlands,5 so savings could double if the recommendation for HPV triage would extend from smears within the screening programme to all smears. Savings per screened woman could be higher in areas where the proportion of BMD primary smears is higher (e.g. >5% in Finland and England compared to <2% in the Netherlands1).

None of the 3 analysed HPV strategies is optimal for both side effects and costs. Delaying the HPV testing by 6 months (strategy B) may have the lowest total costs (Δ = −36% compared to direct colposcopy in strategy D) but for the combination strategy C the costs are only slightly higher. The risk selection is equally good for strategies B and C (both strategies establish HPV persistence before a referral), but the period of time in which women are kept in triage is on average 3 months shorter for strategy C. Strategy A eliminates the need for an extra triage period and extra GP visits, but it is less powerful in selecting women at higher risk for progression to cancer than strategies B and C.

The rationale for screening programmes is early detection and treatment of disease. In our analyses, we assumed equal effectiveness of each triage strategy, i.e. that there are not more cervical cancer deaths in the HPV strategies than in the direct colposcopy. This assumption can be challenged for 2 reasons. First, in the HPV strategies only women with detectable HPV infections, i.e. women at risk for cancer, are referred for colposcopy. A recent meta-analysis estimated that 95.5% of all CIN 2+ lesions can be identified if women with primary ASCUS and LSIL smears (which approximately correspond to BMD smears) are tested for HPV.29 Because at the time of histological testing some women may have already cleared the HPV without yet having their lesion regress, or might still clear the HPV later, only (an unknown) part of the remaining 4.5% may represent potential loss of sensitivity in detecting CIN 2+ compared to direct colposcopy (strategy D). On the other hand, recent data from the ALTS trial suggests that HPV testing may perform no worse than, or may even outperform colposcopy in identifying high-grade CIN lesions.30 Follow-up data from the currently on-going randomised trials, e.g. that from the POBASCAM study expected shortly,31 will shed more light on the loss of sensitivity of HPV triage because of less frequent referral. Second, in the Netherlands the conventional guidelines of following up the BMD primary smear by a follow-up smear in 6 months implicitly accept the risk of postponing treatment to occult underlying cancers in 0.4 per 1,000 women with BMD primary screening smears.5 When HPV testing is delayed for another 6 months (i.e., to 12 months after the BMD primary smear), the risk of diagnosing an invasive cancer is a further 0.9 case per 1,000 women with a BMD primary screening smear.5 Should strategies B or C be adopted, postponing referral would miss another 6 months of lead-time for treatment in these 0.9/1,000 women. Given that these are screen-detected cancers, one may assume that despite this delay it is likely that they are still found at an early enough stage of invasion to retain the good 5-year prognosis of 90% survival.32 To sum up, we can reasonably assume that the effectiveness of the 3 analysed HPV triage strategies and of the conventional strategy are comparable.

An alternative to the analysed triage strategies could be to drop the requirement of first establishing the persistence of the BMD smear and to instead triage solely through HPV that is, directly after 1 primary BMD smear. Per 1,000 women with primary BMD smears, such an approach would expectedly increase the referral rate by 100, the CIN 2/3 detection rate by 10–15, and prevent 1–3.5 cancers (see Appendix). Given the very long (>10 years) average duration of preinvasive lesions,33 some of these 1–3.5 per 1,000 women will still have the chance to be managed early at the next screening round in less than 5 years. This balance is not straightforward, and remains uncertain. Again, the expected long-term follow-up data from randomised trials such as the POBASCAM study should in the near future help improve these estimates. Also, adding further markers to HPV-testing (e.g. typing for HPV 16 and 18, testing for mRNA) is currently under investigation as feasible ways to improve specificity, which would more likely favour HPV-testing above cytological triage.

This study has some limitations. First, some women who in the Erasmus MC trial had a persistent HPV+ CIN 2/3 lesion from t = 0 to t = 6 (this prompted treatment and censoring from further follow-up) could have cleared the HPV by t = 12. Therefore, the assumption we used to fit our interpolation model with that-women treated at t = 6 would have remained in the same state until t = 12 had they not been treated––may give an overestimate of HPV and CIN 2/3 persistence, and an overestimate of the referral rate and costs for strategies B and C. On the other hand, the quantification of strategies B and C is based on the data from a trial in which biopsies were taken at enrolment from all colposcopic abnormalities. Biopsies may have interfered with the development of the disease, e.g. by removing HPV-infected lesions which would not have cleared and/or regressed if left unbiopsied in the observed period. Consequently, the estimated 6-month HPV persistence rate may be too low, the CIN 2/3 prevalence rate at t = 6 too low, and the expected decrease in the referral and total costs too high. Though especially strategy C seems interesting for implementation, we are less certain exactly how well it would in reality perform relative to strategy A. Given our assumptions, strategy C could avoid a referral to colposcopy in at most 9% more women than strategy A. The cost for this advantage is an extra GP visit and a 6-month longer triage period for 51% of women in strategy C. In the base-case calculation the extra GP visits in strategy C save 0.18 referrals and 0.07 treatments compared to strategy A. Strategy C would be preferred over strategy A only when it could be shown that the perceived burden of an extra GP visit and 6-month waiting time during triage will be less than 18% of that associated with a referral for colposcopy, and also less than 7% of that of treatment. The relative burden is thus far unknown but it could be studied in an implementation trial in which the 6-month natural history of HPV could be monitored for this group of women without early biopsy interference.

Second, because we wanted to study the optimal strategy to offer to women, we assumed compliance with follow-up in all strategies during the triage period. Lack of follow-up will decrease screening effectiveness. It has been shown in the literature that up to one third of the women do not comply with follow-up,34 and that the compliance decreases with longer time lags in recommended follow-up.35 Since women in our study have already had to wait for 6 months for the follow-up smear, these findings are especially challenging for those strategies that involve extra prolonged periods of triage in which HPV is allowed to clear (strategies B and C). In the Netherlands, cervical cancer screening is performed at the primary health care level by GP. A large audit of GP practices in the Netherlands has shown that complex decision trees are important barriers to compliance of the GP with diagnostic guidelines.36 All 3 HPV strategies increase the diagnostic complexity for the GP as they add to the currently routine practice a test of a different type, strategies B and C also at extra time points.

Third, women's preferences should play a role in optimising the triage strategies. In principle, triage through HPV could decrease anxiety in women with abnormal smears by decreasing the number of false-positive (i.e., HPV-negative) referrals. Still, follow-up of a large UK screening cohort showed that a negative HPV result does not significantly reassure women with a BMD smear.37 Moreover, women who received a positive HPV report showed even higher anxiety levels after an abnormal smear.38 Several surveys have shown that a significant proportion of women with cytologic abnormalities may prefer an early referral to wait-and-see approaches.39, 40 and that a higher level of psychological distress in a woman is an important factor contributing to such a choice.41 It seems then that it is especially the postponement of action in women who know that they tested HPV-positive in strategy C that could negatively affect their well-being, making strategy A more appealing to implement.

In conclusion, our analysis provides further evidence that HPV can improve the specificity of referral for colposcopy of women with persistent BMD smears, and decrease the burden on both the women and the health-care system. Given the established high sensitivity of HPV testing for progressive cervical neoplasia, we therefore advocate including HPV testing before referring women. For these women, who are already in follow-up for 6 months, analysis of additional aspects favours implementation of immediate HPV testing without waiting another 6 months for clearance.

Acknowledgements

The authors wish to thank Caspar Looman for statistical advice, and Marie-Louise Essink-Bot for giving comments on an earlier draft.

Appendix–expected referral and detection rates for HPV triage, with and without cytology in triage

Assume strategies A1, B1 and C1 such that relative to the primary BMD smear (found at t = −6) the timing of HPV testing remains equal as in strategies A, B and C but the cytologic triage (at t = 0) is not done. As a consequence, HPV testing is done on all women with a BMD primary smear in strategies A1, B1 and C1, whereas it is only done on women with 2 consecutive BMD smears in strategies A, B and C.

The expected referral rates, expressed as the proportion of all women with a BMD primary smear, are:

  • Strategy A: 18% (25% with BMD in follow-up5 × 51% HPV-positive7 + 5% with >BMD in follow-up5; Fig. A1);

  • Strategies B and C: 15% (25%5 × 51%7 × 77% (Table I) + 5%5);

  • Strategy A1: 28% (35% of the women with a BMD primary smear are HPV positive31 × 81% HPV persistence rate in 6 months42; Fig. A1); and

  • Strategies B1 and C1: 24–28% (35%31 × [68% (18 months)42 to 81% (6 months) HPV persistence rate42]).

The expected increase in the detection rate of CIN 2/3 lesions in strategies A1, B1 and C1 over those of strategies A, B and C is 13.5 per 1,000 women with a BMD primary smear ((95–80%) difference in sensitivity for CIN 2+ of HPV vs. cytology triage29 × 9% prevalence of CIN 2/3 lesions in BMD primary smears13 × 1,000).

It has been estimated that 10–24% CIN 2/3 lesions eventually progress to cancer.25, 43 Therefore, substituting the combination of a follow-up smear and a HPV test for a stand-alone HPV test in women with BMD primary smears could prevent at most 1.4 to 3.2 ([10 to 24%] × 13.5) invasive cancers.

Figure A1.

Calculation of expected referral rate in strategies A and A1, after a single BMD primary smear. Shaded rectangles represent women who need to be referred, whereas the rectangles with a bold border represent women for whom we assumed that they return to the regular screening programme.

Ancillary