A cluster randomised trial of strategies to increase cervical screening uptake at first invitation (STRATEGIC)

Phase 1 objectives

1. To determine, in an English cohort of 24.75-year-old women in Greater Manchester and a Scottish cohort of 20-year-old women in Grampian, the clinical effectiveness and cost-effectiveness of a pre-first invitation leaflet (pre-leaflet), which had been designed to increase receptivity of young women to cervical screening. The development of this leaflet had been completed prior to the trial, and was based on focus group work with young women, targeting the issues that influence them when thinking about attending for cervical screening.

2. To determine the feasibility, acceptability, clinical effectiveness and cost-effectiveness of offering online booking as a convenient alternative to telephoning the general practice as a means of making an appointment at first invitation. This was available only to the women covered by Manchester Primary Care Trust (PCT) (now redesignated a Clinical Commissioning Group).

3. To determine, among the 20-year-old women in Grampian, whether prior HPV vaccination in the catch-up campaign was associated with an increase or decrease in participation.

Phase 2 objectives

To determine, among those women in the cohort who had not attended after 6 months, the feasibility, acceptability, clinical effectiveness and cost-effectiveness of several novel strategies:

1. the offer of requesting a self-sampling kit (SSK) for HPV status as a determinant of the need for cervical cytology (HPV-negative women would not require cytology)

2. SSKs sent unsolicited to women

3. a specialist NN to help a woman overcome her barriers to screening

4. timed appointments for cervical screening to encourage women to attend

5. a choice of requesting a HPV SSK or the NN.

In addition, a health economic study (see Chapter 3) would be performed to determine the costs and cost-effectiveness of these interventions and a DCE (see Chapter 4) would be performed to determine the importance non-attending women attach to different aspects of the screening experience.

Greater Manchester

Grampian

Phase 1

Initial contact was made with the GPs by the trial team in order to introduce them to the trial. The research nurse attended the clinical commissioning group meetings prior to the STRATEGIC pilot study starting and provided a synopsis of the trial. Representatives from general practices were given the opportunity to make the trial team aware of any potential issues that they could foresee with the trial. Next, a standard letter was sent to the general practice informing them of their trial arm allocation to a pre-leaflet or no pre-leaflet.

Phase 2

The next point of contact made to the general practice by the trial team was prior to the phase 2 interventions when the trial team informed them of the intervention to which their practice had been randomised. The timed appointments required the screening agency to send a standard letter to the general practice asking them to send eligible women an offer letter, detailing a time and date for them to attend for a cytology sample. The general practice then returned the letter to the agency with the date given to the woman. Once the trial team received a list of confirmed timed appointments, the general practice invoiced the trial team for reimbursement of £5 per letter.

Phase 1

Phase 2

Randomisation of general practices

In this trial the interventions were randomised to practices for both the phase 1 and phase 2 interventions. Because all practices in a PCT entered the trial at the same time it was possible to use a minimisation procedure for cluster randomised trials, as described by Raab and Butcher. 15 In this, allocations of interventions to practices were generated using a random number generator and implemented accordingly. For each allocation the imbalance between intervention arms is calculated and the allocations are then sorted according to the magnitude of the imbalance, before selecting an allocation with good balance. We controlled for the cervical screening uptake rate and size of this cohort of women in each practice prior to randomisation based on standard reports of these data provided by the screening agencies involved. We chose the allocation that was the fifth centile of the distribution of imbalance from 10,000 possible allocations. This procedure was carried out separately for each PCT with the names and location of the practices concealed. In phase 2, allocation was also balanced for allocation in phase 1.

Sample size

The primary outcome

The primary outcome for phase 1 was uptake of cervical screening 3 months following the standard invitation, at which time point a reminder would be sent by the screening agency. The primary outcome for phase 2 was uptake of cervical screening at 12 months following the standard invitation. Uptake rates were also calculated at 6 and 18 months to assess any long-term effects. Each primary outcome was calculated using a provided ‘date of test’ variable that related either to cytology test only (phase 1) or to cytology or HPV test (phase 2). In each case a generated binary variable was defined as ‘completed’ (1) and ‘not completed’ (0) by the allotted follow-up time point (e.g. 3 months). A missing date of first test indicated if a test had not been completed by the end of the follow-up period. The data can therefore be considered to be a complete data set for all individuals. Outcome variables were then created by identifying a test present (yes/no) for 3, 6, 12 and 18 months since the standard call.

Secondary outcomes were (1) differential uptake of screening among vaccinated and unvaccinated women in Grampian by 6 months after invitation and (2) uptake 12 months following phase 2 interventions.

Data exclusion and data cleaning criteria

Participants were excluded from the phase 1 data and the subsequent analysis if:

1. the cytology test was prior to the allotted pre-leaflet intervention date (approximately 6 weeks prior to standard invite) assigned to both the intervention and control group participants

2. all intervention dates are missing.

Women were excluded from the phase 2 data and the subsequent analysis if:

1. time in study was > 9 months from when phase 2 interventions started (Greater Manchester)

2. test date was prior to phase 2 interventions start

3. subject had a ‘left trial date’ (i.e. moved home/GP prior to proposed phase 2 intervention sent)

4. subject’s first invitation date was sent outside the recruitment period.

Baseline cervical uptake screening rates

In Greater Manchester, the phase 1 baseline rate for each practice was determined for a baseline data cohort consisting of women for 12 months between October 2010 and September 2011 prior to the trial. The data extraction methodology was the same as that used for the outcome measure. In Grampian, the equivalent cohort was not available and so routinely reported uptake rates for women aged 20–24 years at each practice was used instead. The difference between the methods of determination of the baseline rate is accounted for in the analysis by adding an interaction term between region and baseline rate.

Estimation method

The effect of interventions was estimated using a logistic regression model with covariates for intervention group, baseline rate, PCT (region) and an interaction between baseline rate and PCT/region to account for differences in methods of collection of baseline rate. To account for the clustering effect of practice, a GEE model was used with an exchangeable covariance structure and robust standard errors. This analysis was performed using the Stata procedure xtgee (version 13, StataCorp LP, College Station, TX, USA).

Phase 1 analysis

Phase 1 determined the effect of a pre-leaflet/no pre-leaflet and online/no online booking on uptake of any test, either HPV or cytology. The pre-leaflet was run in Salford, Trafford and Grampian, whereas both the pre-leaflet and online booking was run within a factorial design in the Manchester PCT only. Both interventions were analysed in an ITT analysis, where response was compared between the two groups, as offered, irrespective of whether the subject received the pre-leaflet or the method employed to book the test. A test for evidence of interaction between the pre-leaflet and online booking was performed. Descriptive statistics and response rates were calculated for each intervention and the GEE model estimated the treatment effect. Odds ratios (ORs) would indicate if a significant change in response between the control group and the interventions had occurred (p < 0.05). An OR is expressed on the logarithmic scale meaning a value of 1 indicated no change in response rate, a value > 1 showed an increase in response and a value of 0–1 indicated a decrease in response. Analyses are presented for the uptake at 3 and 6 months post invitation, with the 3-month analysis being the primary end point.

Moderator analyses

Moderator analyses were carried out to test for:

1. an interaction effect between treatment group and the location, with respect to the screening age differential between Scotland and the north-west of England

2. the influence of HPV vaccination status (Grampian PCT only).

Pilot study among non-responders

The pilot study in Manchester was running concurrently with phase 1 in women who had already not taken up their invitation despite reminders. The pilot determined the response rate and effectiveness for the planned interventions for phase 2: HPV self-sampling, choice of two SSKs, NN, choice of SSK or NN, or timed appointments. Each intervention was compared with the control group using the GEE model. Any intervention found to be significant at a one-tailed 25% significance level (i.e. with an OR of an uptake between the intervention and the control larger than 1 with a p-value of < 0.25) would be taken forward into phase 2 of the main study.

Phase 2 analysis

As in phase 1, descriptive statistics in the form of incidence rates described the response to each intervention. Response rates for each intervention were compared with the control response rate, again using a GEE model under an ITT framework. As well as an overall Wald chi-squared test, pairwise tests were carried between each intervention and control. Consistent with the power calculation, a Bonferroni correction of significance levels is used to adjust for multiple testing, with a significant level of 1% in place of 5%. Baseline rates for each general practice used as covariates for the phase 2 analyses were calculated from the uptake rates observed in phase 1. The improved consistency and the direct relationship within general practice of baseline uptake rate in phase 2 compared with phase 1, means an interaction between site and baseline rate should no longer be required.

As the only date available for all women was the standard invitation date, follow-up assessments were timed relative to this. Analyses of phase 2 were carried out at 12 and 18 months after this date, with 12 months post call being the primary end point.

Phase 2 moderator analysis

Moderator analyses were carried out to test for:

1. An interaction effect between the intervention groups and the study location (Greater Manchester vs. Grampian), to investigate the differential screening age and socioeconomic differences between Scotland and the three north-west of England sites.

2. An interaction effect between phase 2 intervention groups and phase 1 intervention groups (pre-leaflet vs. control and online booking vs. control), to determine if a cumulative effect was occurring. The interaction analysis including online booking was performed in the Manchester PCT only.

3. The effect of vaccination status and subsequent interaction effects with phase 2 interventions (Grampian only).

4. Test result was modelled as a binary result (positive equals low or high dyskaryosis) to determine the impact any improved screening may have had on identifying cases.

Table 3 shows the intended comparisons of interventions for each cohort, including any suitable interactions (in bold). The table also outlines the follow-up time points for each comparison to be made. These refer to months since participants’ 25th (20th in Grampian) birthday, and hence allows for a further 6–8 weeks for pre-leaflet intervention.

Phase 1

Table 4 shows the number of practices and corresponding number of participants randomised to each phase 1 intervention group. The screening uptake curves over time for the pre-leaflet and online booking, and those of the control practices, are shown in Figures 4 and 5, respectively. The actual percentage uptake at the 3- and 6-month time points following the standard invitation is shown in Table 5. The benefit of pre-leaflet is negligible. Although there was a slight benefit immediately after the standard invite, by 3 months there was no difference. In fact, up until 3 months, both interventions almost overlay the uptake in the controls. At 3 months, a routine reminder was sent out. Between 3 and 6 months the pre-leaflet arm continued to follow the controls, but there was a small increase in the online booking group. Also striking was the steady gradual increase in participation following the invitation and the intervention continuing following the standard reminder at 3 months.

FIGURE 4.

Kaplan–Meier plot showing time to test from standard invitation by pre-leaflet groups (7-month follow-up).

FIGURE 5.

Kaplan–Meier plot showing time to test since standard invitation by online booking groups: Manchester PCT only (7-month follow-up).

Considering the percentage uptake (see Table 8), the first observation to be made is that by 3 months 19.2% of the control group had attended, and by 6 months this had increased to 30.6%. Uptake by women sent a pre-leaflet was almost identical at both time points (3 months, OR 0.967; 6 months, OR 1.014). Among the Manchester cohort, uptake by the control group was marginally lower, at 17.2% and 26.6% at 3 and 6 months, respectively. This is not surprising, as Manchester has poorer coverage than the country as a whole. The online booking group showed a very similar uptake by 3 months. Although uptake at 6 months was 2.2% higher, at 28.8%, in the online booking group, this difference was not statistically significant (p = 0.242). It should be noted that 199 women booked an appointment online, but only 127 (63.8%) actually attended that appointment. The absence of data linkage means that it is not possible to determine the precise proportion of women who attended an online-booked appointment before 3 or 6 months, as some of the online-booked appointments may have occurred after 6 months. It follows, therefore, that < 2.4% (127/5267) of women in this arm attended an appointment booked online by 6 months. The interaction between pre-leaflet and online booking at both 3 and 6 months was not significant (p-value is 0.604 at 3 months and 0.912 at 6 months), indicating no synergistic effect of pre-leaflet and online booking.

Table 6 compares, at 3 and 6 months, the overall uptake in Grampian and Greater Manchester. Both the proportion who attended at each location and the proportion attending at each vaccination level were calculated for subjects within the pre-leaflet group, the control group and in total. The OR relates to the baseline group: vaccination none and location Greater Manchester.

Absolute attendance was similar over time in Greater Manchester and Grampian, although with a slightly greater absolute percentage uptake in Grampian, by 1% and 3.4% at 3 and 6 months, respectively. The ORs of a test for Grampian compared with Greater Manchester were 1.169 [95% confidence interval (CI) 1.030 to 1.326; p = 0.016], and 1.275 (95% CI 1.133 to 1.435; p < 0.001) at 3 and 6 months, respectively. Interaction effects between pre-leaflet and location were found to be not statistically significant, indicating that the effect of location was consistent across intervention groups (a p-value of 0.591 at 3 months and of 0.542 at 6 months).

Table 6 also gives the uptake of screening associated with the level of completeness of vaccination in Grampian. Estimated ORs are given comparing those with ‘full’, ‘incomplete’ or ‘missing’ vaccination status compared with ‘none’. The absolute increase in attendance between ‘none’ and ‘complete’ was 12.62% (23.65 – 11.03%) and 21.96% at 3 and 6 months, respectively. With an approximate doubling of the attendance rate, the ORs for ‘complete’ compared with ‘none’ were 2.074 and 2.571, with both being statistically significant (p < 0.001). Those who were partially vaccinated, that is, with fewer than the three doses, had a smaller but still significant increase in uptake. The three degrees of freedom chi-squared interaction test again indicated that no interaction effect was present between the pre-leaflet group and vaccination status (χ32 p = 0.828 at 3 months and χ32 p = 0.870 at 6 months).

The pilot study of proposed phase 2 interventions

A total of 720 women were targeted, with 120 being offered each of the intervention options, and there was a control group. The piloting required 2–3 months to offer the interventions, 3–4 months of follow-up, and 2 months for analysis.

The piloting interventions were offered as of April 2012 for 3 months; follow-up began in July 2012 for 4 months, ending in November 2012.

The analysis replicated the main study, with the outcome being the performance of a test within 6 months following sent date, compared across interventions. If the cytology date was prior to the intervention, those women were removed from the analysis, as they were duplicates. The following analysis is based on 714 women out of the original 720. The sample sizes per intervention across the north-west of England are summarised in Table 7.

The pilot demonstrated that each of the interventions could be delivered successfully. The uptake of screening following the different interventions is shown in Table 8. Of the 119 control subjects, 14 (11.8%) attended screening. The greatest rates of uptake were achieved by offering timed appointments (22.5%) and a choice of interventions (18.3%), with the NN intervention actually resulting in a decrease in uptake (7.7%) compared with controls. The effect was striking even in this small sample of women, with 21 out of 22 participants offered a choice actually attending for a routine cytology sample, as did 8 out of 20 women who were sent a SSK. This suggests a ‘nudge’ effect being exerted by these interventions, whereby the intervention indirectly influences the woman to attend for screening.

Table 9 reports the OR for each intervention compared with the control group, along with both the 95% and 50% CIs. The 50% CIs are given as the lower limit corresponding to one-tailed 25% significance level test. Timed appointments resulted in a significant increase in uptake, with the odds of any test being administered being almost 2.5 times that of the control group. Choice (1.75) and SSK sent also resulted an increase in uptake, which in both cases was significant at the 25% level. As suggested in Table 8, the NN and offer of a SSK showed a decrease in test uptake, with the NN intervention showing a significant decrease at the 25% level.

Even among the small sample in this pilot study, timed appointments appeared to result in a significant increase in attendance, along with SSK sent and being offered a choice. The combined effect of both SSK offer and NN appeared to result in a strong increase in attendance, even though each individual intervention resulted in a decrease in attendance. Because choice required both NN and an offer of a SSK, the independent Trial Steering Committee advised that all of the piloted interventions should be offered in phase 2.

Phase 2

Each practice was randomly allocated to one of the five phase 2 interventions or to the control group, again using a cluster randomisation that employed a minimisation algorithm to balance each intervention arm. Study participants were excluded from receiving a phase 2 intervention if they had attended screening or moved practices prior to the phase 2 intervention.

Figure 6 repeats the lower half of the STRATEGIC Consolidated Standards of Reporting Trials (CONSORT) diagram. Of the 20,879 participants who entered the study and participated in phase 1, 2330 were not eligible for phase 2 because of a 3-month delay in implementing the phase 2 interventions. A further 6454 attended for cytology screening within 7.5 months of their standard invitation and a further 1969 moved house or GP also within 7.5 months since their standard invitation was sent. This left 10,126 eligible to participate in phase 2.

FIGURE 6.

The CONSORT diagram for phase 2. Note: black numbers in brackets refer to the number of general practices invited during phase 1. a, Delay in starting phase 2 because of a delay in operationalising the interventions.

Phase 2 results

Table 10 shows the number of practices randomised to each phase 2 intervention. Nine practices were not randomised to phase 2 (leaving 269 practices) because all eligible participants in these practices had been screened, and a further two were effectively lost from phase 2 as all remaining eligible women had moved. This left the 267 practices shown in Table 10, which was still enough for 30 practices for each intervention and 100 control practices, as required to achieve adequate statistical power. It should also be noted that the original power calculation had estimated that 50% of the phase 1 participants would be available as ‘non-attenders’ in phase 2, and, despite the loss of almost 10% of the STRATEGIC study cohort because of a delay in initiating phase 2, 10,126 out of 20,879 (48.5%) participants were available for phase 2. Of those entering phase 2, the median [interquartile range (IQR)] follow-up without attendance was 727 days (IQR 630–860 days) [i.e. 24.2 months (IQR 21–28.6 months)]. Note that the minimum follow-up without attendance was 17.8 months, indicating all but 62 (who were within a few days) were followed for a minimum of 18 months. Those 62 were assumed to have not attended during the last few days and were included in the 18-month follow-up outcome. Of those who did attend screening, the median time to attendance was 358 days (IQR 267–500 days) [i.e. 11.9 months (IQR 8.9–16.7 months)] since standard invitation or 4.4 months (IQR 1.9–9.4 months) since phase 2 interventions were sent.

The type of screening test performed in phase 2 participants (i.e. HPV test only, cytology only or both, in which case in which order) is shown in Table 11. The result of the HPV test and whether or not the participant went on to have a cytology test, is shown in the lower part of Table 11. Of the 122 women tested for HPV, 33 tested positive, of whom 20 (60%) then went on to have a cytology test. It is interesting that a large proportion of those screened following self-sampling interventions (i.e. sent or offered as part of a choice) did, in fact, have a practitioner-obtained cytology sample. For example, of the 1149 women who were sent a HPV SSK, 292 (21.3%) were screened in phase 2, although 198 of these women (67%) underwent only a cytology test. Of the 279 women who were screened after being offered a SSK, only 21 had a HPV test, the rest having cytology only. Only 7 of the 314 participants screened after being offered a choice, which included SSK, actually had a HPV test.

Figure 7 and Table 12 illustrate the proportion of non-attenders in phase 2 who had any screening test (cytology or HPV) associated with each intervention at 12 and 18 months since the standard invitation. Figure 7, a Kaplan–Meier plot, indicates a differential rate of uptake over time since the standard invite for each of the intervention groups. As this cohort did not attend before the first reminder, all curves are flat for the first 6 months. By 12 months post standard invitation (approximately 6 months of phase 2) between 14.5% and 21% of non-attenders had been screened; 6 months later the proportion had reached between 22.8% and 30.1%. The control group demonstrated a similar trend with 16% and 27% of non-attenders in the control group screened at the same time points.

FIGURE 7.

Kaplan–Meier plot showing time to test since standard intervention for women eligible for phase 2 interventions (non-responders at 6 months).

Table 12 provides the cumulative uptake following each phase 2 intervention at each time point as a proportion along with their corresponding OR (95% CI) and p-value with respect to the control group. A chi-squared test with 5 degrees of freedom suggests that there was a difference between the six study arms at 12 (p < 0.001) and 18 months (p = 0.008) following the standard invitation. Owing to the multi-intervention study design, the study protocol and power calculation specified a Bonferroni correction of significance levels to adjust for multiple testing with a significant level of 1% in place of 5%. The odds of attending increased significantly (p-value = 0.001) for SSK sent (OR 1.512) and timed appointments (OR 1.408) compared with control at 12 months post standard invitation. Although both continued to suggest an increase, only SSK sent (OR 1.286) was statistically significant (p = 0.012) at 18 months. Of the remaining interventions, SSK offered and choice both resulted in a slight, but non-significant, increase in uptake of any test. The NN intervention actually resulted in a non-significant reduction in uptake at 12 months (OR 0.887), which became a statistically significant reduction at 5% significance but not at the Bonferroni-corrected 1% level at 18 months (OR 0.799; p = 0.044). Table 12 also gives the result of a comparison test of the five interventions. The null hypothesis states that there is no difference between the five intervention effects, here p < 0.01 indicates that the null hypothesis is rejected and that there are significant differences between the five interventions at both 12 and 18 months.

Phase 2 planned moderator and predictor analyses

A series of moderator analyses were specified in the statistical analysis plan. By adding a factor by phase 2 treatment interaction, we examined the moderating effect of phase 1 intervention (intervention pre-leaflet vs. control and online booking vs. control), location (Greater Manchester vs. Grampian) and, in Grampian only, prior vaccination status (none, incomplete and complete). The predictive effect of these factors was considered using a main-effects model without interaction terms. There was no evidence that the pre-leaflet intervention moderated the effect of phase 2 interventions at either 12 (χ52 p = 0.740) or 18 months (χ52 p = 0.216) since the standard invitation (see Appendix 7, Table 43). There was no effect of the pre-leaflet during the phase 2 intervention period at either 12 (p = 0.760) or 18 months (p = 0.875) post call, see Appendix 7, Table 43). Similarly, there was no evidence that the online booking intervention moderated the effect of phase 2 interventions at either 12 (χ52 p = 0.594) or 18 months (χ52 p = 0.321) post call (see Appendix 7, Table 44), nor was there any effect of online booking at 12 (p = 0812) or 18 months (p = 0.263). When the effect of location was considered (Greater Manchester vs. Grampian), there was no effect on phase 2 interventions at 12 months post call (χ52 p = 0. 141), but this was significant at a 5% level at 18 months (χ52 p = 0.020) (Table 13). Inspection of the pairwise interaction of each phase 2 intervention against control suggests that the effect of the NN intervention differed between Grampian and Greater Manchester compared with other interventions. The negative effect of this intervention at 18 months (see Table 12) appeared to be moderated in Grampian. Figures 8 and 9 show the uptake rate at 12 and 18 months post call by intervention and location. With the exception of the NN intervention, we see that the uptake rate is higher in Greater Manchester than in Grampian, whereas the NN intervention attendance was higher in Grampian than in Greater Manchester. When a model was fitted without the interaction terms (see Table 13), there was evidence of an overall higher uptake in Greater Manchester than Grampian during phase 2 at both 12 (p < 0.001) and 18 months (p < 0.001) post call. This contrasts with the equivalent analysis during phase 1, which found that uptake in Grampian was higher than in Greater Manchester (see Table 6).

FIGURE 8.

Percentage attendance at 12 months for phase 2 interventions split by location. GM, Greater Manchester; GR, Grampian.

FIGURE 9.

Percentage attendance at 18 months for phase 2 interventions split by location. GM, Greater Manchester; GR, Grampian.

Finally, we considered the relation of vaccination to uptake during phase 2. There was no evidence that intervention effects were moderated at either 12 (χ52 p = 0.147) or 18 months (χ102 p = 0.122) post call (see Appendix 7, Table 45). As with the analysis of phase 2, both ‘incomplete’ and ‘complete’ vaccination status were associated with improved uptake at 12 and 18 months (see Appendix 7, Table 45). As with phase 1, those participants who had experienced some form of vaccination compared with those with none were approximately twice as likely to attend screening in phase 2, with ORs of 2.109 and 2.169 for incomplete and complete, respectively, at 12 months post call, with both effects being statistically significant (p < 0.001).

Adherence with randomisation during phase 2

In both phase 1 and 2, interventions were randomised to practices. The analyses presented above have been by ITT, irrespective of whether or not the intended intervention was actually delivered. The data available do not allow us to determine if a participant could receive or actually received the intended intervention. For example, some women may have changed address, but this information could not be provided. However, the date at which a phase 2 intervention was sent was recorded by the screening agencies for each type of intervention for each woman. These data are summarised in Table 14, by intervention. The column Correct date refers to the number of women recorded as having been sent the randomly allocated intervention for the practice with which they were registered. The column Incorrect date gives the number recorded as having been sent the incorrect intervention for their practice. The percentage of women sent the incorrect intervention ranged from 0.2% for NN to 3.8% for SSK sent. Among the control group, 1.3% received a phase 2 intervention in error. For the timed appointments intervention, 27.6% (449/1629) had no date recorded, largely because one large practice was unable to implement timed appointments. For the other four interventions, between 83% and 89% of women appear to have been sent the correct intervention.

Impact of phase 1 and phase 2 interventions on detection of low- and high-grade cytology

The cervical screening result was available for women who attended for a test. Table 15 gives the rates of women invited by the standard invitation who were tested and found to have low- or high-grade cytology overall, and broken down by phase 1 intervention group and follow-up assessment point. Note that the rates are based on numbers of women randomised and not numbers tested. After 3 months’ follow-up the percentage of women invited for screening and found to have cytology of ‘low grade or above’ or ‘high grade’ was 4.24% and 0.44%, respectively. By 6 months’ follow-up the percentage of women invited for screening and found to have cytology of ‘low grade or above’ or ‘high grade’ cytology was 7.82% and 0.64%, respectively. Rates are given broken down according to intervention. Adjusted ORs were estimated using the same methods and covariates as the analysis of screening uptake. Neither pre-leaflet nor online booking had a significant effect on a low grade or above outcome, or high grade, which is to be expected, as neither intervention had an effect on uptake. The largest, although still non-significant, effect (p = 0.130) occurred for online booking at 6 months (OR 1.57, 95% CI 0.88 to 2.80) for high-grade cytology. It should also be noted that the CIs are now rather wider, which is an indication that the trial is underpowered for plausible effect sizes for these outcomes.

Table 16 gives the corresponding result by phase 2 interventions. Of this cohort, 5.1% and 7.6% were detected as having low grade or above at 12 and 18 months post call, with 0.3% and 0.5% of the cohort being high grade. At 12 months post call the intervention groups differed in the proportion of low-grade cytology detected (p = 0.043). Of the phase 2 interventions, timed appointments showed a significant increase (p = 0.001) in low grade or above (OR 1.57, 95% CI 1.21 to 2.03) at 12 months that persisted with a borderline significant increase at 18 months (OR 1.24, 95% CI 0.98 to 1.57; p = 0.074). SSKs sent showed a borderline increase at 12 months (p = 0.065). Inference from these data should be viewed with caution because of the possibility of multiple testing artefacts. The significant increase because of timed appointment, and to a lesser extent for SSK sent, nevertheless reflects the increased uptake rates of these interventions (see Table 12).

Measurement and valuation of costs

Measures of effects and follow-up

The primary within-trial outcome measure for all interventions was attendance for cervical screening. This outcome was measured at 3 and 6 months post invitation for phase 1 interventions (i.e. pre-leaflet and online booking) and at 12 and 18 months post invitation for the interventions provided in phase 2. The primary end point for phase 1 interventions was uptake of screening at 3 months and for phase 2 interventions at 12 months post invitation. These end points were used in the main economic evaluation and the secondary end points (6 months for phase 1 interventions, 18 months for phase 2) in a secondary economic analysis.

Lifetime costs and quality-adjusted life-years of screening attendance

A systematic literature review was performed to obtain information about the lifetime costs and quality-adjusted life-years (QALYs) of women who attended cervical cancer screening and those who did not. We developed the literature search strategy based on the PICOS (population, intervention, comparator, outcomes, study design) framework, in order to retrieve economic evaluations of screening strategies. These included no screening as a comparator and reported lifetime costs and outcomes in the form of life-years and/or QALYs for women eligible for cervical cancer screening. Following the recommendations of the Centre for Reviews and Dissemination26 and Cochrane Central Register of Controlled Trials, the following computerised bibliographic databases were searched on 15 May 2016: MEDLINE, MEDLINE In-Process, EMBASE, EconLit and the NHS Economic Evaluation Database. Letters, editorials, historical articles, animal studies, studies published before 1995 and studies published in languages other than English were excluded from the search.

A two-stage selection process was followed. The title and abstract of the retrieved studies were scanned for relevance and the full text was then accessed if a paper was judged to be an economic evaluation related to cervical cancer screening and had employed a disease progression model with a lifetime time horizon. Where exclusion based on the titles/abstracts was not possible, the full publication was retrieved and evaluated. When all relevant full texts were obtained, they were assessed for eligibility against the PICOS criteria (Table 17).

The reviewing process was documented in Microsoft Excel, and reasons for inclusion and exclusion detailed to facilitate updates of the review. The websites of agencies including NICE, the Medical Research Council and National Cancer Screening Programmes were also scanned for relevant reports. A template was developed and used to extract the information from the most relevant studies to the STRATEGIC trial. The review was performed in May 2015.

The quality of the selected studies from the literature review was assessed using a checklist of good practice in decision-analytic modelling studies developed by Philips et al. 27 and the CHEERS statement for reporting economic evaluation. 22 In this way, the selected studies were assessed both on specific modelling and on broader evaluation grounds. In case of disagreement between the assessors, the mean assessment score was assigned. In studies where more than one cancer screening strategy was compared with no-screening strategy, we used a selection process to retrieve lifetime costs and outcomes of screening compared with no screening. In this process, the next more-costly strategies to no screening as well as strategies that had similar population (i.e. age groups), interval of screening and discount rates (i.e. 3.5%) to the UK were preferred. Where results from cancer screening in different countries were reported, we selected the UK-based estimates.

A meta-analysis was performed to pool the lifetime costs and outcomes reported in the selected studies. A random-effects model was specified to pool the estimates assuming real differences in treatment effects because of heterogeneity in screening strategies, population and other factors. 28 The study quality score was used in this model to weight each study’s contributed information to the pooled estimate. The NICE appraisal studies were assigned the highest possible weight on the grounds that they were of high relevance to our study and of high methodological quality.

Cost-effectiveness and cost–utility analyses

A decision model was constructed in Microsoft Excel to calculate the lifetime costs and outcomes of each intervention. Each of the six interventions was compared with the control group. The probabilities of attendance in each intervention and control group were informed by the trial results. Costs in this model included the intervention costs, screening costs during the study, and lifetime screening and disease-specific costs (which were informed by the meta-analysis described in Lifetime costs and quality-adjusted life-years of screening attendance).The incremental cost-effectiveness ratios (ICERs) were expressed as incremental costs per women attending a screening test and incremental costs per QALY gained taking the NHS perspective.

Uncertainty and scenario analysis

Probabilistic sensitivity analysis was performed to address the uncertainty in the ICERs by performing 5000 draws of all cost and effect parameters using pre-specified distributions, and recording incremental costs and incremental QALYs from each draw. These results were then plotted on cost-effectiveness planes. Mean costs and outcomes, as well as their standard errors as reported by the trial’s statistical analysis, were used to define distributions for the attendance rate. The reported mean estimated and standard errors of lifetime costs and QALYs from the meta-analysis were used to estimate the distribution of these parameters. The intervention costs per attendee, unit costs and probabilities of having a HPV test, cytology test and colposcopy were also included in the probabilistic sensitivity analysis. Cost-effectiveness acceptability curves were derived from these results, to express the uncertainty in these cost-effectiveness estimates in graphical form. This was done by displaying the probability that each intervention is cost-effective as the ceiling ratio for the maximum acceptable ICER varies from £0 to £75,000 per QALY gained.

Two univariate sensitivity analyses were also performed. First, the impact of using secondary end points (6 months instead of 3 months for phase 1 and 18 months instead of 12 months in phase 2) on the results of the economic evaluation was tested. In another univariate analysis, the study quality scores based on Phillips et al. 27 that were used to weight the pooled estimates in the meta-analysis were replaced with the study quality scores based on the CHEERS statement.

A scenario analysis was also performed to explore the adoption of these interventions for a population of 365,087 women in phase 1 interventions, which is approximately the total number of women annually invited for cervical screening for the first time in England, Scotland and Wales. A similar scenario analysis was performed for phase 2 interventions applied to a population of 255,561 women, that is, assuming that 70% of women do not attend phase 1 interventions. In these analyses, intervention costs were categorised into fixed one-off costs, semifixed scalable costs, which are incurred in steps as the scale increases, and variable costs. It was assumed that large-scale purchasing of materials and equipment would allow a cost discount to be obtained; this is set at 20% in the sensitivity analysis. In these scenario analyses, therefore, only the components of total cost that are not variable will affect the cost per woman and the ICERs.

Using discrete choice experiments to explore individual preferences

Human beings consciously or subconsciously make choices in their everyday lives. Researchers interested in exploring how individuals make choices between alternatives face methodological, theoretical and practical challenges. It is very complex to find a theoretical model that explains exactly an individual’s decision-making process, as this is highly variable among individuals. 45 Researchers cannot get into people’s minds and observe exactly what happens when a decision is made. Therefore, analysts need to find a strategy for identifying, capturing and using as much as possible the information that an individual considers when processing a scenario that leads to a choice. Because the analyst will never be able to identify all data and information processed by an individual, there is always a margin for error. This error of measurement is an important component of any choice model. There are two main alternatives that have been used to elicit people’s preferences in the economic literature: revealed preferences (we observe the choices made by people) and stated preferences (we directly ask people about their preferences). 46 Revealed preferences can be used only when an individual has made a real decision, and thus far in the context of novel and new therapies its use in health care has been very limited. Stated preferences have so far been found to be a more appropriate strategy to obtain preferences for alternatives that individuals may or may not have experienced in real life. To obtain preferences for different strategies to increase cervical screening uptake by non-responders, we used a stated preference method called a DCE. 47

A DCE is an elicitation technique to obtain individual preferences using a series of hypothetical choices. DCEs are implemented using surveys that require a series of choices to be made between alternatives where each alternative, although described using a set of common attributes (i.e. characteristics or traits), has different values or levels for these attributes. Alternatives are usually presented in pairs known as choice sets. For each choice set, individuals are asked to select which profile they prefer based on the information presented. The response data from a DCE are analysed within a random-utility model using statistical modelling for choice data, such as logit or probit models. A technical summary of the random-utility model and the theoretical framework employed in this DCE study is presented in Appendix 12.

Development of discrete choice experiment attributes

A number of different methods can be used to obtain and develop attributes for DCEs including literature reviews,48 expert reviews49 and focus groups or interviews. 50 Experts in the field do, however, recommend that qualitative work is conducted in order to develop such attributes47 and interviews are considered by Coast et al. 51 as providing the greatest ‘richness’ of attributes. A small qualitative study complemented with the views from the qualitative and quantitative research teams was undertaken to identify the attributes for this DCE exploring young women’s preferences for cervical screening programmes.

Experimental design

A full factorial design using the attributes and levels in Table 32 resulted in 2 × 2 × 2 × 4 = 32 possible profiles and 1024 (322) possible pairwise choice sets to select from. Given that it was impossible to ask a respondent all the choice sets from the full factorial design, a selection of choice sets was generated using a D-optimal method in Ngene 1.1.2 software (ChoiceMetrics, Sydney, NSW, Australia). 54 This method is designed to provide stated choice experiments with optimal statistical properties and to elicit the maximum information from respondents to inform the estimation of model parameters. 55–57 We based the choice of the D-optimal approach on recent guidance about constructing experimental designs for DCEs. 58 Only main effects were considered, with non-linearities and interactions assumed negligible. Table 34 shows the D-optimality levels for DCE designs with different numbers of choices. The figures indicated that the marginal benefit of increasing the number of choices above 16 choice sets was minimal. We found little guidance about the maximum number of choice sets that a person could endure in a single DCE survey, but a recent study suggested that the number of choice sets depends on the complexity of the DCE study and the context of the research questions. 59 In our case, 16 choices might have been a large number to ask our population as they are not taking up the offer of cervical screening. In addition, the qualitative team already had struggled to recruit participants for their interviews and it was felt that non-responders (the target population for this DCE) were a ‘hard-to-recruit’ sample. Therefore, a trade-off between number of choices and D-optimality levels was made and 12 choice sets were selected for the DCE survey. This provided a D-optimality level of 90.86%, which was still considered high. This would ensure that the survey would be quicker to complete and likely to encourage participation and reduce missing responses.

The final 12 choice sets used in the DCE are presented in Table 35, with an example actual choice set (in this case set number 1 in Table 33) presented to participants shown in Table 36.

Questionnaire design and data collection

Participants were given the option of completing a paper-based or online version of the DCE survey. A copy of the paper-based version and the patient information leaflet prepared to accompany the survey are presented in Appendices 14 and 15, respectively. Prior to the piloting phase, and to assess acceptability and ease of completion, the questionnaire was given to female members of staff at the Health Economics Research Centre within the University of Oxford to complete. Minimal changes to the questionnaire wording were made as a result of this exercise. The DCE questionnaire comprised a section containing a number of choice scenarios that incorporate the attributes and levels developed from the qualitative analysis presented above, a section on women’s general views about the value of the current Cervical Screening Programme and a section where women ranked each of the four attributes in order, from most important to least important. The questionnaire also included questions on basic demographic information such as age, education and employment status. Women were also asked about any difficulties encountered when completing the DCE questionnaire and were given the opportunity to feed back any additional comments. The English and Scottish screening agencies, in conjunction with the STRATEGIC research study team, identified non-responding women who were eligible to participate in the DCE study and sent a participant information leaflet, the paper-based version of the survey and a Freepost return envelope. As women had the choice of whether to complete the paper-based version or the DCE online, the web link for the online version was provided in the patient information leaflet and the paper-based version.

A custom-made online survey was developed using the open source application LimeSurvey (www.limesurvey.org). The online version was an exact replica of the paper-based version. The online version was tested thoroughly before, and during, the pilot study phase to minimise any potential problems during the main data collection. The web version complied with the required information governance standards of the University of Manchester and University of Oxford to ensure anonymity of responders and compliance with data protection. To ensure anonymity, a token number was created in LimeSurvey for each potential participant. Participants had to use this token number to access and complete the online survey. When the system recognised that a token had been introduced, it recorded only that the number had been used and not any other information about the participant. The token number was also included in the paper-based version. Once a token was used and a survey was completed, participants were not able to access the survey again.

To encourage participation in the survey, women were offered a £10 high-street voucher as an incentive. In order to claim the voucher, an online form was created and the web address to access the form was included in the patient information leaflet and paper-based survey.

First phase (pilot discrete choice experiment data collection) and second phase (main discrete choice experiment data collection)

In a first phase, a pilot study was conducted from July 2014 to September 2014 to assess the working interface between the English and Scottish screening agencies, and the STRATEGIC study research team. During this period, we also assessed the robustness of the online survey and the likely response rate for the second phase (main data collection). A total of 1000 questionnaires were sent to women in Greater Manchester (n = 650) and Grampian (n = 350) during this period.

The feedback from the pilot suggested that communications between the parties was clear and that the same arrangement could be used with no modifications in the main study. The online survey also worked well and no major issues were identified. The response rate was low (as described in detail in Results) but, overall, women felt that they completed the questionnaire without difficulties. Some women reported that it was not clear when the nurse would be available to help, so the words ‘prior to appointment’ were added to the description of the attribute for the second phase. We consulted a colleague with extensive experience in population surveys about possible additional ideas to improve the response rate during the main data collection. It was clear that there were difficulties in reaching the population of interest and it was recommended that the look of the questionnaire could be improved, ideally with the assistance of a graphic designer. This was undertaken and the final questionnaire is presented in Appendix 14. The research team also decided to maintain the £10 voucher during the main DCE data collection. No other changes to the questionnaire were made between study phases. The data from the pilot were entered once and subsequently checked by the quantitative STRATEGIC study research team.

The main DCE data collection (second phase) was carried out between mid-July 2015 and mid-September 2015, and 3000 questionnaires were circulated to women in Greater Manchester (n = 2000) and Grampian (n = 1000) during this period. Double data entry of paper responses received was used during this phase.

Sample size

Sample size calculations in statistics are traditionally used to minimise biases and ensure generalisability to the population of interest. There are approximately 300,000 women in England and 30,000 women in Scotland who are eligible to be screened for the first time every year. The STRATEGIC study cohort suggested that around 70% of women do not respond to the first invitation, resulting in 210,000 and 21,000 non-responder women in England and Scotland, respectively. With these figures, obtaining a representative sample size of the preferences of women to the Cervical Screening Programme would be very expensive using traditional sampling theories. Therefore, sample size calculations in stated choice experiments take a different angle: to identify the minimum number of choice observations needed to obtain reliable parameter estimates for discrete choice models using stated choice data. The latter approach is described in a recent paper60 and was used in this study. Briefly, the method suggests that, if prior information on the coefficients of a discrete choice model is available, it is possible to identify the theoretical minimum sample size (called S-error) required for a design. This is achieved combining the asymptotic variance-covariance matrix from the experimental design with the available prior information and specifying a significance level that normally corresponds to 5% and a t-ratio of 1.96. We used the information from the pilot phase to estimate the prior coefficients associated to each attribute using a conditional logit without a constant, as suggested by Rose and Bliemer. 60 The S-error was estimated to be 1154, but we also evaluated the minimum sample size requirements for each parameter separately. Figure 20 shows the asymptotic t-ratios for different sample sizes associated with each of the attribute coefficients, with the dashed line indicating a t-ratio of 1.96 (e.g. a coefficient significantly different from 0 at 5% level).

FIGURE 20.

Asymptotic t-ratios for different sample sizes for each parameter in the discrete choice model.

The figure suggested that, with sample sizes > 150, we were able to estimate significant parameters in the action, nurse and cost attributes but not on location. To obtain a significant parameter in the location attribute using our prior information we needed to obtain the sample identified as S-error = 1151. This information was very useful for the team to understand; given the response rate in the pilot study, it still appeared possible to obtain significant coefficients using the data from the second phase. As the prior information was based on a small sample of women, the results from this analysis were used for guidance, but did point out that a minimum sample size of 150 was needed to reliably estimate three attribute parameters. It was uncertain at that time whether or not the parameter for the attribute location was going to be significant with that sample size.

Statistical analysis

Descriptive statistical analyses were used to present participant demographic information, women’s views of cervical screening in general and any feedback about the stated choice questions. Choice data were analysed first using a conditional logit, and the reader is referred to Appendix 12 for its theoretical and formal derivation to analyse DCE data. One of the main criticisms of this model is the assumption of independence of irrelevant alternatives, which states that the ratio of the probabilities of choosing one alternative over another does not change with the introduction of a new alternative or the removal of an existing one. 47 We therefore also estimated a multinomial conditional probit that relaxed the independence of irrelevant alternatives assumption. Conditional logit and probit models are analysed using maximum likelihood estimation methods, but the former assumes a standard logistic distribution of errors, whereas the latter assumes a normal distribution. In addition, the multinomial conditional probit allows for a general covariance structure in the errors and hence does not impose the independence of irrelevant alternatives property present in conditional logistic models. The coefficients from logit and probit models are difficult to interpret directly, but the signs associated to each coefficient have a more straightforward interpretation. A positive coefficient means that an increase in the attribute leads to an increase in the predicted probability of selecting a particular scenario. A negative coefficient means that an increase in the attribute leads to a decrease in the predicted probability. The aim of the choice model was to identify the effect of each attribute level on women’s preferences for potential interventions to increase the uptake to the Cervical Screening Programme. The levels for action, location and nurse were defined using dummies, whereas the cost variable was included as a continuous variable. The ‘no’ level was used as the reference level for the action and nurse attributes and ‘home’ was used as the reference level in the location attribute. Costs were expressed in 2013/2014 UK pounds. A main-effects specification, using attribute levels only and main effects plus women characteristics, was evaluated in the models. The latter model was used to determine any propensity to select a particular scenario in the choice set given the characteristics of the participants. In our unlabelled DCE, it was expected that all these covariates would be non-significant and that the final model would be either the main-effects conditional logit or probit. Characteristics were included as binary dummies (0/1) in the model (i.e. employed vs. other current main activities or higher education vs. other highest level of education). Robust standard errors associated with attribute coefficients were estimated in recognition that each participant completed 12 choice scenarios. The final selected model was based on the model specification that yielded best goodness-of-fit measures (lowest value) based on the Akaike information criterion (AIC) and the Bayesian information criterion (BIC). The coefficients of the final selected model were employed to predict the probability of selecting each of the STRATEGIC study interventions, identified in Table 33, as potential strategies to increase screening uptake. All analyses were carried out in StataMP version 13 (StataCorp, College Station, TX, USA).

Response rate and final sample size

Figure 21 describes the flow of women through the study. The left-hand side presents figures for the pilot DCE data collection and the right-hand side for the main DCE data collection. In total, 4000 questionnaires were sent out. Figure 21 shows that response rates were similar for both phases at around 5.5%, as were the proportion of respondents from Greater Manchester and Grampian. In both phases, the majority of respondents also chose to complete and return the paper-based version of the questionnaire. There was a general concern that using only data from the second phase would risk obtaining a reasonable sample size to estimate the models as indicated in Sample size. Changes made to the DCE from the first to second phase were minor and the same experimental design was used in both phases with the only amendment of adding the words ‘prior to appointment’ to the nurse attribute. Therefore, data from the first and second phases were combined to obtain a final sample size of 222 responses.

FIGURE 21.

Flow diagram of women’s participation through the study.

Characteristics of the sample

Table 37 describes the characteristics of the 222 participants in the stated choice questionnaire study. Almost three-fifths were from England and two-fifths from Scotland, and the mean overall age was 24.6 years. Women from Scotland were, on average, 4 years younger than women from England, reflecting the lower age at which women are first invited for cervical screening in Scotland (20 years vs. 24.5 years). The majority of women (86.4%) were of white ethnicity, with women of Pakistani, Indian and African descent making up a further 9%. Almost half of the respondents had been educated up to a university level, and a further 40% had either studied to the age of 18 years or undertaken vocational qualifications. Just 6% of women had left school aged 16 years and 5% chose not to disclose information on their education levels. In response to the question on main activity, 57% of women reported being in employment, just over one-quarter were undertaking full- or part-time studies, and 8% were homemakers looking after their families. The percentage of women who were unemployed, undertaking voluntary work, or who were long-term sick or disabled was 3.2%, 1.8% and 1.8%, respectively.

Women’s views of cervical screening

To gauge women’s views of cervical screening in general, respondents were presented with the statement ‘screening for cervical cancer is important’, and were asked to select one of the following responses: strongly agree, agree, neither agree nor disagree, disagree or strongly disagree. Figure 22 shows that, despite not attending for their screen, almost two-thirds of responding women strongly agreed with the statement that screening for cervical cancer is important, and a further 30% agreed. A total of 4.5% of women appeared indifferent, and just 1.8% disagreed that screening for cervical cancer is important.

FIGURE 22.

Responses to the statement ‘screening for cervical cancer is important’.

Ranking of attributes

When asked to rank each of the four attributes included in the DCE in order from most important (rank 1) to least important (rank 4). Figure 23 shows that most women identified the location of the test as being most important (93/204, 46%), whereas others were most concerned about whether or not they personally needed to take any action to arrange a test (63/204, 31%) or whether or not a nurse was available as part of the screening process (34/204, 17%). Although a small number of women (14/204, 7%) considered the cost to the NHS to be most important, for most women (91/204, 45%) cost was the least important attribute.

FIGURE 23.

Ranking of four attributes (action required to arrange a test, location of test, nurse available for help and advice, and cost of test) from most (1) to least (4) important. Denominator is 204, as 18 participants did not complete this question in the survey.

Stated choice data analysis

Table 38 provides a summary of the proportion of participants selecting choice 1 across the 12 choice sets as presented in Table 35. There was a clear trend to select choice 1 in scenarios 4–6 and 10, and a trend to select choice 2 in scenarios 1 and 7. These results provided some insight about the interpretation of the constant in the modelling results in the next table.

The results of the conditional logit and multinomial probit models using main effects and main effects plus selected women characteristics are presented in Table 39. The table reports the coefficients, associated standard error and p-values for each attribute or women characteristic. It also reports information about the goodness of fit to assist in the identification of the final selected model. For both types of models there were marginal improvements in terms of the AIC and BIC using the probit model, although model coefficients were similar between both model types. As expected, including women’s characteristics in the main-effects models did not add any significant information as all demographic coefficients were not significant in both models. Given the lower AIC and BIC, our final selected model for our stated choice data was the multinomial probit model with main effects. The coefficients in the selected model indicate the contribution of each attribute to the probability of selecting a particular scenario (described by a combination of levels for the four attributes). Negative coefficients indicate that women preferred these attributes to take lower values for continuous variables (costs) or preferred the reference attribute for qualitative attributes (action, location and nurse). For example, the signs for the main-effects attributes were significant (p < 0.001), suggesting that women preferred scenarios where minimal personal action was required to arrange a test, the test was done at home, there was a nurse available for discussion or help, and the test was cheaper to the NHS. The constant was statistically significantly different from 0 indicating that potential sources of utility were not captured by the information in the DCE or a potential bias towards selecting always choice 1 in the choice set.

The coefficients from the main-effects probit model were employed to estimate the probability of selecting each of the interventions assessed in the STRATEGIC study based on our choice data and are presented in Table 40. The intervention with a higher probability was the unsolicited SSK, with an estimated figure of 0.263. Although no intervention clearly dominated, strategies involving SSKs accumulated > 75% of the total joined probability across all interventions. This seemed to indicate that women had a clear preference for interventions that include the SSK.

Feedback discrete choice experiment survey

Discrete choice experiments have previously been criticised by respondents for being difficult to understand and to complete. To explore if this was the case within the context of the STRATEGIC study DCE, women were presented with each of the three statements shown in the first row of Table 41 and were asked to indicate whether or not they were in agreement, using the same five categories given in the preceding question on the importance of screening.

Table 41 shows that 86% of respondents agreed or strongly agreed with the statement that the questions were easy to understand, and just 5% disagreed or strongly disagreed. Over three-quarters of women agreed or strongly agreed that the difference between the choices on each question was clear, although around one-fifth of women appeared unsure, reporting that they neither agreed nor disagreed. In response to the statement about it being easy to make a choice on each question, two-thirds of women agreed or strongly agreed, and around one-fifth again appeared unsure. Almost 15% of women disagreed or strongly disagreed with the statement that it was easy to make a choice on each question.

Women were also given the possibility of writing any comment about the stated choice questions or any other aspect of the questionnaire using a free-text box. The literal qualitative comments received during phases 1 and 2 are presented using themes in Appendix 16.

Chief investigator

Henry C Kitchener.

Principal investigator

Henry C Kitchener (Greater Manchester) and Margaret Cruickshank (Grampian).

Trial co-ordinators

Rebecca Albrow, Jamie Oughton and Carley Moseley (Institute of Cancer Sciences, University of Manchester) worked sequentially as STRATEGIC trial co-ordinators.

Nurse navigators

Samantha Fletcher and Julie Burnett (Greater Manchester).

Judith Wilson and Diane Ledingham (Grampian).

Statistics

Chris Roberts and Matthew Gittins (Institute of Population Health, University of Manchester).

Virology

Alexandra Sargent (Department of Virology, Central Manchester University Hospitals NHS Foundation Trust).

Economic modelling

Apostolos Tsiachristas and Alastair Gray (University of Oxford).

Discrete choice experiment

Oliver Rivero-Arias, Helen Campbell and Alastair Gray (University of Oxford).

David Torgerson, Cath Jackson and Jude Watson (York Trials Unit, University of York).

The study was conducted under the guidance of a Trial Steering Committee and a Data Monitoring Committee. The independent members are as follows.

DCE comments

As a (junior) doctor I think it is important to reiterate the value or cervical screening in SEXUALLY ACTIVE women[.]

As I’ve already written I feel the first set of questions were too much[.]

I am very grateful this questionnaire was sent to me I genuinely forgot/I was too busy to book a cervical smear test and this has encouraged/reminded me to do so Thank you, good luck in your studies[.]

I found the question is a bit confused, but, anyway I tried to answer[e]d it. I hope you understand me[.]

I have decided to take the questio[n]naire, as my case is a bit different, so I don’t know what bracket I fall under[.]

I have now attended my smear thanks to this form, luckily my tests came back negative:)!

I’d encourage more insightful questions regarding the Cervical Screening Process when it comes to overseas students choosing to do the test within the UK[.]

If you are going to hold surveys like this, could you ph[ra]se ‘women who have screenings but have not attended’ better or elaborate further? Not all women have skipped these screenings because they’re afraid, costs or laziness[.]

It would be good to have a section for comments and reasons for peo[p]le to provide for refusing cervical screening test[.]

Obviously the options were not what I would pick overall. The answers were based on what I thought were important if doing the test at home or by GP[.]

Personally I’m not sure your proposals will make me more inclined to have a smear.

Questionnaire does not address my main reason for non-participation in Cervical Screening Programme[.]

Some of the options I didn’t think were viable, and I was asked to choose between two un-realistic options. I think the questions would have been better in a different format, to get peoples opinions on specific options/thoughts of the NHS, instead of making them choose between two options that they would not necessarily choose as an option.//Individual questions would have been better. I do not feel my t[ru]e opinion has been expressed in this questionnaire due to the format[.]

Some of the questions repeat in a way that is confusing, I’m not sure that my answers will be interpreted in a way that reflects my real opinions. The questionnaire in general I found somewhat confusing[.]

Sometimes difficult to make a distinction between questions - variety of questions might be easier to understand.//Otherwise thank you!

The question[n]aire was really easy to complete and understand[.]

The questionnaire has showed me how important cervical cancer screening is I will book my appointment with my GP to do my test. Thanks[.]

The questionnaire indicates I am yet to attend a screening test but does not provide questions asking why women may shy away from undergoing a test. Overall, good questionnaire items[.]

The questionnaire was very easy to follow[.]

This is a great idea [I] am pleased to post my opinions personally[.]

This questionnaire was very easy to answer[.]

[T]his was a fairly complicated questionnaire and can see how some people may become easily confused[.]

DCE attributes

[A]n optional question on whether the participant is sexually active or not can affect your study in terms of the participant possessing adequate knowledge/importance of screening.//

Tweaking the cost etc - would have made more sense to ask people what there options would be i.e. action require by you? ? No// ?Yes.

as much as it gives service users input into methods best used for cervical screening, it doesn’t ask why there was a delay in uptake, which [I] think is also important[.]

It’s great that you are doing a study into this, which may encourage better uptake of cervical screening. However feel it isn’t comprehensive. Clearly you have set parameters which you are working in. But from my point of view the reasons for not undergoing screening are more complex than the 4 issues you are studying. Psychological and cultural factors as well as childhood abuse make it incredibly unlikely that I could allow anyone near me, voluntarily. I see a lack of recognition within the NHS that for some people screening just isn’t for them, and that there are complex, human reasons for this.

The options presented to choose from are based around 4 possible reasons why women not want to undertake cervical screening, but would it not be more pertin[e]nt to ask wo[m]en their own reasons why? My own reasons for not atten[d]ing screening were not covered by these choices. Also, the 4 simple bullet points do not convey the full pros and cons of each choice, and readers may not be aware of them. For example, sending a kit home for patients to take the test themselves may avoid embar[r]assment and time off work to make an appointment, but there are risks that the woman might do the test wrong, or hurt herself, or it may become contaminated. A professional nurse is trained to take the test. These pros and cons might not be immediately obvious to people taking the survey, and may not be taken into account when they are providing their answers.

The questions in section 2 could offer a further option because personally the main reason I chose an option was the location it was carried out although I don’t agree on the higher costs either therefore I wouldn’t want there to be a high cost yet also I wouldn’t want the rest to be carried out at the local surgery therefore I picked out the best option out of the two[.]

DCE process

SITE DID NOT WORK, I DID-NOT GET MY £10 VOUCHER, UNHAPPY WITH.

General

There is no use in screening women who have never been sexually active!

AWARENESS OF THIS SCREENING, NEED IMPROVENING./

Cervical screening (as well as breast examination) is so very important. More and more young adults are finding they have abnormal smears. My Mum and Grandma were diagnosed with carrying the BRAC2 gene.//(I have been tested and luckily I am not a carrier.)//

Cervical Screening age should be lowered to 20. This will save young lives.

Cervical screening is a very intrusive procedure which is not necessary for all women.

I recently went to a ‘lady doctor’s’ appointment, but it was in regards to contraception . . . She had noticed I hadn’t had my smear test so asked if she could do it at the same time - but it messed up the contraception by doing so (the coil-copper coil... the wires were lost).//

I Have Not yet had a smear test, but thinking about it. I feel that starting having smears early is important. I am not sure what age your first test should happen.

I THINK THE TEST AGE SHOULD BE LOWERED TO 16-18[.]

I think you are doing a bril[l]iant job + keep up the hard work, we need more information to be able to tackle this issue head on. Also I think the age for screening should be lower to help save more lives.

I WHOLLY SUPPORT AND ENCOURAGE CERVICAL SCREENING FROM AN EARLY AGE[.]

Your cervical screening records do need to be updated, as I told my doctor I was screened in a different country.

Location

I WOULD GLADLY BE SCREENED IF I HAD THE OPT FOR MY LOCAL SURGERY./

I like the idea of a home test as I feel more women would be more [i]nclined to get tested.

I believe more women, especially younger, would take the test if given the option to take the test at home. Knowing from past experience, some women can feel a bit shy & not vey keen at being poked & prodded.

I feel if testing was to be done at home by individuals it would be a huge succes[s], avoiding embarras[s]ment and time issues. It’s a good idea and something I would use.

I feel many women are at risk due to the screening being carried out by CP [GP] which is embar[a]ssing.

I had a cervical screening, 1 week ago. If there was an option of doing the test at home I would have done it MANY years ago.

I never new that there was even an option being considered where we would be able to do this from home[.]

I think offering a home testing kit as an alternative would encourage women who avoid testing (through embar[a]ssment/forgetting) to test themselves, and could save lives[.]

I WOULD ALWAYS CHOOSE HOME, PROVIDED THAT IT WOULD NOT COST THE NHS TOO MUCH, AND I FEEL GUIDANCE SHOULD ALWAYS BE AVAILABLE - EVEN IF DONE VIA AN ONLINE VIDEO/FAQ [frequently asked questions] SECTION.

However I feel I would also be willing to try a home kit first, before this survey I did not know that was a possible option in the future I hope if becomes available.

Personally I would prefer to take a test at home as I get embarras[s]ed and have a lot of anxiety’s that’s why I have never atten[d]ed for a test to be done.

In summary: I would prefer a home testing kit, that I would have to request (rather than just being sent automatically).

AND I’D PREFER A METHOD THAT IS MORE PRIVATE I.E. THE TEST KIT SENT TO MY HOUSE FOR SELF TESTING.

THE OPTIONS WHEN CHOOSING WHICH CHOICE I WOULD ALWAYS CHOOSE THE OPTION TO DO THE TEST AT HOME.

Therefore, testing myself at home would be 100% better all around.

I believe that sending tests in the post so women can do the test themselves would get a lot more women to do the test.

A LOT of people get put off the idea because of having To go To gp For it but doseNt bother me Either WAY the Leaflets sh[ou]ld be Eas[i]er To understand and [i]f Anyone cAnt mAke it To gp should be offer[e]d To have a home Test sent out if they can do it them self’s.

[I] would be more will to do a test at home if it was automatically posted when it was due.

I would prefer if an appointment was booked for me and I would rather have it at a clinic than do it myself.

Would always say to have screening at surgery.

Action

Young girls are having sexual intercourse at an early age and so I feel they need scree[n]ing and if a test kit could be sent through the post and is not as invasi[ve] as a normal smear, then I think this is the way forward.

Not an excuse but appointment times at a GPs doesn’t help, this problem.

I would highly recommend a time/date being sent to women as this would have definitely given it a higher sense of importance for me personally.

Booking appointments has its pro’s + con’s but if people aren’t going to make an appointment themselves then it’s likely they won’t call to reschedule or simply won’t turn up.//I would prefer an online/SMS [short messaging service] booking process. Allowing me to book an appointment at any time not when the phone line is open.

HOWEVER I DO THINK IF I RECEIVED AN APPOINTMENT ALREADY ARRANGED FOR ME I WOULD HAVE BEEN MORE LIKELY TO ATTEND.// THE REASON TO BE ABSENT FROM WORK WOULD BE COVERED BY A CONFIRED DOCTORS APPOINTMENT.//

A LOT of people get put off the idea because of having To go To gp For it but doseNt bother me Either WAY the Leaflets sh[ou]ld be Eas[i]er To understand and [i]f Anyone cAnt mAke it To gp should be offer[e]d To have a home Test sent out if they can do it them self’s.

I would prefer if an appointment was booked for me and I would rather have it at a clinic than do it myself.

CS-N (why do not attend cervical screening)

The HPV virus is spread by sexual transmission hence if you have never been active sexually it feels like you are being violated. Also the risk is extremely low to warrant a check in my personal opinion.//I did try to be screened by I was unable to let the nurse proceed as it was extremely uncomfortable and violating. This is why I refuse to be screened.

Finding time to attend an appointment is my downfall.

I actually went to my GP surgery to arrange a test, but since I was not at the top of the list to be sent a reminder I could not organise one. I also did not received any invitation to attend, and to date have still not. Thus it has been a bit of a nightmare to actually get a test. If I wasn’t presented with administrative barriers I would have gone ages ago.

I am very grateful this questionnaire was sent to me I genuinely forgot/I was too busy to book a cervical smear test and this has encouraged/reminded me to do so Thank you, good luck in your studies.

I believe more women, especially younger, would take the test if given the option to take the test at home. Knowing from past experience, some women can feel a bit shy & not vey keen at being poked & prodded.

I don’t want anyone to look at my private parts, as they are private . . .

I ignored previous invites to a smear test, as it sounds painful! However in real life, is nothing like the stories! This lack of information . . . is all it was. It’s a painless procedure! Should be a leaflet!// I wrote on the first page. You need a leaflet stating is not scary and it’s painless!

I have not attended screening because I suffer from PTSD [post-traumatic stress disorder] related to sexual traumas so the test would be psychologically traumatic to me. I have never had sex so the chances of cervical cancer are low. [H]aving this test would likely to therefore be painful and unnecessary. I also have been diagnosed with high-functioning autism. I do not cope well with touch. I find many NHS professionals do not make allowances for this.

I can guarantee that most people, certainly my age, don’t go as A. They’re embarrassed and B. We’re always told by others how painful it is which I know is not true.

I think offering a home testing kit as an alternative would encourage women who avoid testing (through embar[a]ssment/forgetting) to test themselves, and could save lives[.]

I think the study needs to look further into having tests done by trained gynaecologist rather than GPs. The reason why I personally haven’t been back to arrange a screening test is because I have previously had a test done (probably before I was 25 and that’s why I didn’t show up in your records) and it was horrible. Having lived abroad up until I was 21, I went for regular smear tests (around every 6 months) as this was the norm for any woman who was on the pill. The tests were done by a gynaecologist, who I knew and trusted, in his clinic. Of course the test isn’t comfortable but I never felt in pain. However, when I had the test done at the GP practice I’m registered at now, it was quite painful and uncomfortable. I didn’t have the impression the doctor was doing this on a regular basis and I didn’t feel she had the right equipment, such as the correct facilities for me to sit on, to make the procedure less uncomfortable. I would much rather prefer to pay privately to have the procedure done by a gynaecologist in their practice, than having this done at my GP, as I don’t think they are equipped for this. Having a test like that done is very personal and I can imagine that a lot of young women are self conscious about themselves when going for such a test, so having this done in the right environment, with the correct equipment and by a doctor who is an expert in that field would, in my opinion, be much more beneficial and effective. (expertise, pain privacy?)

I went to my first smear test just the other week at aged 25. For the sake of a couple of minutes and life. I’m ashamed I’d put it off so long. There need to be much more awareness for how important this is. (time)

HOWEVER, THE REASON WHY I HAVE NOT YET ATTENDED A SCREENING IS LESS TO DO WITH CONVENIENCE OR LOCATION BUT MY PERSONAL CIRCUMSTANCES. I AM A VIRGIN WITH VERY LITTLE EXPERIENCE AND I AM VEY ADVERSE TO BEING TOUCHED BY STRANGERS, TO THE POINT OF PANIC ATTACKS. AN INTIMATE EXAM WITH A COMPLETE STRANGER WOULD, BY ITS NATURE, MAKE ME INCREDIBLY UNCOMFORTABLE AND WOULD LIKELY MAKE SUCH A TEST PAINFUL. THERE ISN’T A LOT OF SUPPORT FOR THIS, AND IF THERE IS, NONE OF IT HAS BEEN OFFERED TO ME. (risk, location perception-pain)

My first experience of a smear was painful, each attempt after (3 attempts) have been painful due to being unable to relax. I found the nurses at the G.P. surgery unsympathetic. Oral sedative (valium) did not help. I feel I would benefit from I.V. [intravenous] [sed]ation (midazolam) as this is also a muscle relaxant, I would be more likely to return for screening if this was an option.

Reason why I have not gone for cervical screening is because I am not sexually active and therefore at low risk of cervical cancer.

The main reason for ignoring these screenings is the pain, and that’s the main reason, so just by having a nurse to help you guide through the process may help the most. A nurse who is encoura[g]ing and not to put you off. Because the pain puts you off!

THE MAIN REASON FOR NOT BOOKING AN APPOINTMENT FOR SCREENING IS THAT IT’S A BIT EMBARASSING[.]

I would like to e[m]phasise that I fully understand the importance of screening, but I find the whole idea of the process far too uncomfortable.

It is mainly embarrassment that has made me put off having the test so far.

The reason I did not accept the invitation to have a cervical screening last year was because I have not yet ever been sexually active. When I am, I will ask for a cervical screening.

The screening Test is A good thing I didn’t go To Mine because I WAS pregnant[.]

I turned 25 in June and have put off going but I will go its just finding the time. I then had the implant put in and have been bleeding for 5 weeks so that[‘]s another reason why I have not been.

While I know that cervical screening is important I always have put off booking an appointment, because I always have other activities I’d put first. I go to the doctors about 4/5 times a year. However, only once in about 20 appointments has the doctor mentioned that I have not had my screen. This, in an inadvertent way reaffirms to me that the fact I have delayed my screen isn’t that important/urgent.

Privacy

I feel if testing was to be done at home by individuals it would be a huge succes[s], avoiding embarras[s]ment and time issues. It’s a good idea and something I would use.

I feel many women are at risk due to the screening being carried out by CP which is embar[ra]ssing.

AND I’D PREFER A METHOD THAT IS MORE PRIVATE I.E. THE TEST KIT SENT TO MY HOUSE FOR SELF TESTING.

Time

I feel if testing was to be done at home by individuals it would be a huge success, avoiding embarras[s]ment and time issues. It’s a good idea and something I would use.

CS-Y (why do attend cervical screening)

I would highly recommend a time/date being sent to women as this would have definitely given it a higher sense of importance for me personally.

I had a cervical screening, 1 week ago. If there was an option of doing the test at home I would have done it MANY years ago. I only got the test one week ago as it was a requirement of the fertility clinic before I could get treatment.

Risk

I find it irritating to be persistently invited for a screening when I am not in an at risk group.

I’ve attended my appointments and was told that was low risk of C.C. [cervical cancer] because I’ve not had sexual intercourse, therefore no screening had to be done. I understand that I have to make my own appointments now but the fact that I am getting letters about me ‘missing’ or ‘refusing’ to go to my screening is getting annoying. I’m sure you don’t mean disrespect you not all women are simply refusing to go, some our us have gone to our appointments but DO NOT need screening!!!

I am a qualified doctor (surgery) & have taken steps to minimise my risk of cervical cancer (HPV vaccination prior to being sexually active, protected sex, small number of sexual partners). I therefore feel my risk of cervi[c]al cancer is low; sufficiently low that the disadvantages of screening (false positives & importantly time off work to attend appoint for screening test) outweighs the benefits in my personal case. I support the cervical screening program[m]e on the whole but based on my personal risk/benefit analysis I have elected not to participate at this stage.//Good luck with your study.

Reason why I have not gone for cervical screening is because I am not sexually active and therefore at low risk of cervical cancer.

I DO BGELIEVE THERE ARE ALOT OF MYTHS AND STORIES WHICH DISCOURAGE ALOT OF PEOLE TO ATTEND A SMEAR TEST.

[I] have heard horrific things about smear tests which has put me off but [. . .]

Cost

But on the other hand if it is gonna cost the NHS a lot more money then I think it should be carried on at your GP by the nurse or the nurse to be there and you do it yourself behind the curtain.

I really don’t want the NHS to invest heavily in this. Although a smear is important there are other areas of the NHS that need investment.

the cost to the NHS may make me change my view if it was vastly more expensive (to the NHS) to send out a home test rather than to have an appoin[t]ment at a clinic.

Perception/pain

The thought of a DIY [do it yourself] kit is quite daunting.

The main reason for ignoring these screenings is the pain, and that’s the main reason, so just by having a nurse to help you guide through the process may help the most. A nurse who is encoura[g]ing and not to put you off. Because the pain puts you off!

Nurse

I would really think if someone was available to discuss the test this would make one comfor[t]able to have it done.

It wouldn’t concern me if I wa[sn’t] able to talk to a nurse PRIOR to the test[.]

The main reason for ignoring these screenings is the pain, and that’s the main reason, so just by having a nurse to help you guide through the process may help the most. A nurse who is encoura[g]ing and not to put you off. Because the pain puts you off!