Development and testing of measures

This chapter describes the development and testing of intervention outcome measures. These were categorised as potential primary outcomes and potential secondary outcomes. We collected data to investigate response rates to the various question types and to assess the utility of the measures for a future trial.

Secondary outcomes are presented descriptively and, in some cases, with measures of reliability of combined question banks. The data generated by these measures are not intended as an assessment of effectiveness of the intervention, rather they provide guidance on the most useful outcomes to take forward to the definitive trial.

Group definitions

Study group was defined as:

• baseline (study participants assessed prior to the start of the intervention)
• follow-up (study participants assessed at the end of the intervention towards the end of the school year)
• control (previous year’s S4, assessed towards the end of the school year).

Exposure to the STASH study intervention was predefined (for follow-up students only) as:

• trained PS
• exposed (i.e. a non-PS student who has been ‘exposed’ in one or more of the following ways: received at least one message via social media, had at least one conversation or visited the STASH study website).

Primary outcome measures

• Have you ever experienced oral sex?
• When was the most recent time you had oral sex? (Within the last 6 months or more than 6 months.)
• In the last 3 months, how many different people have you had oral sex with?
  • And how many of them did you have oral sex with without using a condom or dental dam?
• The last time you had oral sex, was a condom or dental dam used?
• Have you ever had sexual intercourse (penis in vagina)?
• When was the most recent time you had sexual intercourse (penis in vagina)? (Within the last 6 months or more than 6 months.)
• How many times in the last 3 months have you had sexual intercourse (penis in vagina)?
• In the last 3 months, how many different people have you had sexual intercourse with?
  • And with how many did you have sexual intercourse without using a condom?
• How often was a condom used when you had sexual intercourse in the last 3 months? (Always, often, sometimes, never.)
• The last time you had sexual intercourse, was a condom used? (Yes, no.)

For the full range of potential primary outcome measures from the follow-up questionnaire, see Report Supplementary Material 5.

Principal component analysis on potential primary outcomes

Principal component analysis was carried out on the full range of potential primary outcomes on follow-up and control data sets (as listed in Primary outcomes). The main items that loaded onto the first principal component were the number of different partners in the last 3 months and number of partners with whom a condom or dental dam was not used, explaining around 20% of the variability in the sample. Main items loading onto the second principal component were use of a condom at last sex and how often a condom was used in the last 3 months. These explained 18% of the variance in the sample.

The first principal component did not make good sense as a composite outcome, particularly as it was restricted to those who reported having had oral sex or vaginal sex. The second was a more sensible composite, as it contained a greater amount of information on an important outcome, condom use, but it explained < 20% of the variance in the sample.

In general, the PCA of the potential primary outcomes did not provide any strong indications of a data-driven choice of primary outcome. The results were not used further in defining the primary outcome to take forward in the definitive trial.

Descriptive statistics for primary outcomes

Table 13 presents the summary statistics for all of the potential primary outcomes, overall and for the three study groups (baseline, follow-up and control). The rates of ever having had oral or vaginal sex would be expected to increase from baseline to follow-up (due to maturation effects), but would be broadly comparable in the follow-up and control groups, as group members completed the questionnaires at approximately the same time in their school education and were the same age (see Chapter 4 and Table 7).

Table 13 shows that the rates of ever having had oral sex were very similar between the control and follow-up groups. A lower proportion of those in the follow-up group than of those in the control group had oral sex for the first time in the last 6 months. The rates of condom and dental dam use were similarly low in both groups, at 12–13%, with the most common reasons being that students did not think it was necessary or they did not have one to hand. Most students who had oral sex did so with only one person, rendering the measure ‘number of oral sex partners’ uninformative as an outcome.

The rate of ever having had vaginal sex was 2% lower in the follow-up group than in the control group, with an increase of just under 6% from baseline. Almost 80% of those who reported having had sex had done so in the last 6 months in both groups, the proportion having had sex for the first time in that period being slightly lower in the follow-up group (31%) than in the control group (34%). These observations suggest that the STASH study may be potentially associated with a slightly lower rate of first-time sex among S4 students. Among those who reported having had vaginal sex in the last 6 months, use of a condom at most recent sex was slightly lower in the follow-up group (40.4%) than in the control group (44.1%). The main reasons cited for not using a condom were that students did not think it was necessary, they got carried away and forgot, or they had both been tested for STIs and had another form of contraception. The follow-up group were 3% more likely to have always used a condom when they had sex in the last 3 months, although the percentage who always or often used a condom was similar across the follow-up and control groups, at around 40%. Most students who had had vaginal sex had done so with only one person.

Suitability of primary outcomes

There was much discussion between the TMG and TSC regarding the identification of a suitable primary outcome for any future evaluation. Sexual activity and condom use are clearly important outcomes, which is why they were included as indicative outcomes in this study. However, when considered separately, they provide only partial coverage of the potential impact of the intervention on safe sexual behaviour. Condom use on its own is of limited utility as it is asked only of the subset of the sample who have had sex. The TMG recommended that both dimensions of sexual behaviour (sexual activity and condom use) should contribute to the primary outcome.

A further point of discussion concerned anal sex. This was not originally included in the list of primary outcomes, but it was noted (see Table 13) that the rate of anal sex in the follow-up group was 3.4% lower than in the control group, and the follow-up students were also less likely to have had anal sex in the last 6 months. As this outcome had a low incidence, this was felt to be a considerable drop. The TMG agreed that a reduction in anal sex would be a positive outcome of the intervention, as it is the riskiest practice in terms of STIs120 and as anal sex in young people is associated with coercion.121 The TMG, in discussion with the TSC, also agreed that oral sex should be dropped from the composite measure because it is much less risky in terms of STIs120 and there is significant resistance to condom and dam use in oral sex.122 Our data (see Table 13) confirmed that condom and dam use is much less common for oral sex (14.6% at last occasion of oral sex vs. 44% for vaginal sex). Although messages about condom and dam use for oral sex would be retained in the future intervention, we agreed that both intervention and outcome should focus on the practices with highest risk. This supports simpler messaging, which we found key to PS role effectiveness. In addition, it focuses the intervention on outcomes with the strongest potential for improvement.

The TMG therefore agreed to take forward a composite ‘always practising safe sex’ variable to the definitive trial. This would be defined as, ‘no condomless (vaginal or anal) sex in last 6 months’.

This means that students who were not having sex would also be considered to be safe from STIs (i.e. no sex equals no risk), as would those who were having (vaginal or anal) sex but always used a condom. To ensure consistency and inclusion of both heterosexual and same-sex relationships, it was agreed that the inclusion of anal sex in the ‘safer sex’ outcome should be on the basis that it could be happening, but would always involve condom use. We did not ask students in the current study whether or not they used a condom in anal sex, only whether or not they were having anal sex at all. We also use a 3-month period for reporting condom use in vaginal sex. Therefore, the composite primary outcome variable reported in this study is ‘always practising safer sex’ but is defined as ‘no condomless vaginal sex in the last 3 months and no anal sex in the last 6 months’.

The results from the current study suggest that this composite measure is likely to have greater utility than any of the individual outcomes, as it allows inclusion of everyone in the sample, and as both delaying sex and ensuring that sex is always protected are the safest ways to avoid STIs and unplanned pregnancy. The composite primary outcome is likely to be more sensitive to change because it is measured in more people than contraception use, which is relevant only to those who are having sex. The time marker of 6 months is consistent with the individual outcomes that were measured in this study and allows assessment of the impact of the STASH study intervention, as the study could not have affected behaviour prior to its start.

Therefore, a row is included at the bottom of Table 13 to show the rates of students ‘always practising safer sex’ in the current study and this is the variable that will be taken forward to the formal primary outcome analyses. The rate of this outcome is about 2% higher in the follow-up group than in the control group.

Primary outcome by study group

The difference in the primary outcome ‘always practising safe sex’ between the follow-up and control groups was estimated using a logistic regression model, adjusting for the clustering by school. The overall OR for the rate of ‘always practising safe sex’ for follow-up compared with control students was 1.18 (95% CI 0.85 to 1.64; p = 0.33). In other words, the prevalence of safe sex was 18% higher among students in the intervention year than among students in the control year. The difference was not significant, as the study was not powered to show a difference in outcomes, but the results are suggestive of a potential positive outcome for the intervention.

As there were more young men in the control group than in the follow-up group, the model was fitted separately for males and for females. Among males, the OR was 1.07 (95% CI 0.67 to 1.73; p = 0.78), whereas among females the OR was 1.37 (95% CI 0.85 to 2.21; p = 0.20). Therefore, there was a larger impact on the safe sex practices of young women than of young men. The interaction between gender and study group was statistically non-significant (p = 0.48), again because the study was not powered to show this. The difference in magnitude of the OR was considered scientifically significant.

In our statistical analysis plan (see Report Supplementary Material 5), we proposed analysis of whether or not there were any differences in primary outcome according to exposure to the intervention (PSs, students reached by PSs and unexposed students). The subgroup sample size for all outcomes was too small to offer a reliable picture. We present the analysis in Appendix 4 for information, but suggest a high level of caution in any interpretation of the data.

Information on the proposed composite primary outcome relevant to a future sample size calculation

The ICC was estimated from models fitted to the composite primary outcome with a random effect for school and no fixed effects. Owing to the small sample size, these ICC estimates were imprecise, with wide CIs. The ICC estimated from the model including all students in the follow-up and control groups was 0.0198, whereas the individual models for the follow-up and control groups estimated the ICC as 0.0066 and 0.0332, respectively. Although set out as a study objective, we opted not to propose a sample size calculation for a definitive trial based on our estimate of the effect on primary outcome from this study. This is consistent with recent guidance,95 as (1) our estimate is highly imprecise and yet any sample size calculation would be highly sensitive to it and (2) other sources of evidence should inform the specification of a difference to detect, including the value of the effect relative to intervention cost.

Secondary outcomes

A variety of potential secondary outcomes are considered here and are summarised into particular domains, each represented by a question bank in the evaluation questionnaires.

To assess reliability of each question bank as a whole, Cronbach’s alpha was calculated (when appropriate) from the control group data and is presented for each set of questions.

Ranges of response rates are given for each question bank and by study group. These represent the proportions of students who answered each question, out of those who completed any of the questionnaire at the relevant time point (and, in some cases, would be expected to have answered the question, following a particular response to a previous question).

As mentioned above, the exploratory and descriptive analysis of these outcomes is intended to provide a guide to their utility for a future, definitive trial of the STASH study, and not as evidence of effectiveness (or lack thereof) of the intervention arising from the current feasibility study.

The data are presented without CIs and the analyses are descriptive because there was no intention or statistical power to test for differences between groups. Planned analysis by exposure was done but is not presented because the sample size was too small for reliable interpretation.

Knowledge

The knowledge questions asked in the questionnaire were as follows (with correct answers in brackets, note ‘do not know’ was considered an incorrect answer):

1. With a condom on, a guy should wait until the penis is soft before pulling out after ejaculation (false).
2. You can make an effective dental dam by cutting up a condom (true).
3. If you get chlamydia you will know because you will have symptoms (false).
4. It is against the law for two 15-year-olds to have sex with each other (true).
5. If someone aged < 16 years goes to the doctor for a STI test or pregnancy test, the doctor has to tell their parents (false).
6. All STIs can be treated with antibiotics (false).
7. A girl cannot get pregnant if the guy pulls out (withdraws) before ejaculation/cumming (false).
8. If you use a lubricant with a condom it should be oil based [e.g. Vaseline^® (Unilever UK Limited, Surrey, UK), massage oil] (false).
9. A girl’s clitoris is inside the vagina (false).

The ‘where to go’ knowledge questions were as follows:

1. where to get free condoms
2. where to get emergency contraception
3. where to get free contraception
4. where to get advice on sex.

The Cronbach alpha of the knowledge question bank as a whole was 0.70, whereas for the ‘where to go’ question bank it was 0.86.

Study group

The response rates to the knowledge questions were similar across the study groups, ranging from 90% to 98% at baseline, from 88% to 96% at follow-up and from 91% to 97% among the control students.

Figure 13 shows the percentage of the control and follow-up groups with the correct answer for each of the knowledge questions listed above. Lines between control and follow-up group are drawn for ease of comparison and do not indicate change over time. The graphs show that knowledge was generally higher in the follow-up group for both knowledge and knowing where to go for contraception and advice. Of particular note is knowledge question 2, which 8% of control students answered correctly, compared with 25% of follow-up students, and knowledge question 5, which 51% of control students answered correctly, compared with 64% of follow-up students. Both of these relate to key intervention messages. Correct responses to question 4 showed a decrease from the control group (72% correct) to the follow-up group (66%), which may have been due to some confusion around the term ‘sex’. Specifically, PSs were trained to think more broadly about what counts as sex (i.e. it is not just vaginal intercourse) and this may have led to confusion about which types of sex were against the law. This will be addressed in the terminology used for sex in the definitive trial questionnaire.

FIGURE 13 Rates of correct answers to (a) knowledge and (b) ‘where to go’ questions for the control and follow-up groups. Lines are drawn for ease of comparison and do not indicate change over time, each line represents an individual question. (Figure not reproduced here; see the original article.)

Across the nine knowledge questions, the proportion of students getting at least six questions correct was 25% in the control group and 36% in the follow-up group.

Attitudes and norms

A comparison of attitudes and norms across follow-up and control groups is shown in Figure 14. The items were as follows (the positive attitude responses are in brackets):

1. It is OK for someone who is not your boyfriend or girlfriend to ask you to send a nude pic of yourself (disagree).
2. The sex shown in pornography is similar to sex in real life (disagree).
3. Keeping a partner happy is a good reason to have sex (disagree).
4. It is not OK to have sex with someone who is very drunk (agree).
5. As long as you use a condom some of the times you have sex, your risk of getting a STI is very low (disagree).

FIGURE 14 Percentage of (a) control and intervention students reporting positive sexual attitudes and (b) students reporting a positive change in attitudes by intervention exposure. Each line represents an attitude. Lines are drawn for ease of comparison and do not indicate change over time. (Figure not reproduced here; see the original article.)

The response rates to the attitude and norm questions were similar across the study groups, ranging from 96% to 98% at baseline, from 95% to 96% at follow-up and from 96% to 97% among the control students.

The rates of adherence to positive attitudes and norms were slightly higher among follow-up students than among control students, with the exception of question 1, which was lower for follow-up students. The Cronbach alpha of the attitude question bank, as a whole, was 0.52.

We found no difference between follow-up and control students with regard to overestimation of the proportion of sexually active students in their year group (half of both follow-up and control students thought that more than one-third of students in their year were sexually active; the actual figure was 20%).

Confidence

The confidence question bank asked, ‘How confident are you that you could:’

1. ‘get condoms of your own?’
2. ‘put a condom on yourself or a partner?’
3. ‘refuse to have sexual intercourse if they won’t use a condom?’
4. ‘recognise when a partner doesn’t want to do something sexual?’
5. ‘tell a partner that you don’t want to take part in a sexual activity?’
6. ‘tell or show a partner what sexual activities you like or dislike?’

Response options were very confident, quite confident, unsure, not very confident, and not at all confident. These were recoded as a binary variable: confident (very, quite) versus other. Follow-up students were asked questions 1–3 at both baseline and follow-up, whereas the other questions were asked only at follow-up. The response rates to the confidence questions were high across the study groups, with 100% completion of the baseline questions, 86–98% completion at follow-up and 98–99% completion among the control students. The Cronbach alpha for the question bank, as a whole, was 0.77.

Generally, confidence was similar or slightly lower among students at follow-up than among control students. An exception was question 3 – refusing to have sex if a partner will not use a condom – for which follow-up students were more confident than control students (73.4% of follow-up students confident vs. 67.7% of control students confident).

Relationship well-being

A bank of items asking about relationship well-being was designed specifically for the STASH study, after we failed to identify any suitable measures of relationship well-being in young people. PCA was carried out on the seven relationship well-being indicators to assess whether or not a summary score could be constructed. The relationship indicators were (with the answer representing a healthy relationship in brackets):

1. They get very angry with me (not often/never).
2. We do sexual activities that I do not feel comfortable with (not often/never).
3. They get jealous and want to know what I’m doing and who I’m with (not often/never).
4. I tell them how I really feel (often/all the time).
5. I feel happy when we are together (often/all the time).
6. I feel comfortable talking about sexual things with them (e.g. whether to do something sexual, using condoms, sexual activities I like) (often/all the time).
7. They respect my opinions and ideas (often/all the time).

Only students who were in a relationship were asked to answer the questions. The response rates for the relationship indicators were similar across the study groups, with a 97–100% completion rate at baseline, a 98–99% completion rate at follow-up and a 98–99% completion rate among control students.

The PCA showed that there were two clusters of questions: questions 1–3 and questions 4–7. Together, these two clusters explained > 63% of the variability in the data and this was consistent whether the analysis was carried out on the control data alone, the follow-up data alone or both combined. The Cronbach alpha for the first measure (questions 1–3) was 0.66, whereas for the second measure (questions 4–7) it was 0.79. The question bank overall had a Cronbach alpha of 0.66. Future validation work on the measure could consider dropping the first subscale (items 1–3) to give a four-item measure of positive relationships. This four-item measure correlated as expected with distress about sex life [those reporting higher well-being were less likely to report distress about their sex life: OR 0.882 (95% CI 0.788 to 0.987); p = 0.029]. Interestingly, those with higher relationship scores were less likely to report safe sex (OR 0.869, 95% CI 0.786 to 0.961; p = 0.006), perhaps because they were in a longer-term relationship and were using another form of contraception.

As well as summarising the individual indicators, we included these two subscales of relationship well-being. These were calculated as a sum of the constituent question scores, with the question options ‘all of the time/often/sometimes/not often/never’ being given scores of 0–4 for the first (negative) measure and 4–0 for the second (positive) measure. The range for each could therefore be from 0 (unhealthy) to 12 (healthy) for the negative measure and 0 (unhealthy) to 16 (healthy) for the positive measure.

Generally, the rates of the indicators of a healthy relationship among follow-up students were similar to, or slightly higher than, rates among control students. The subscales of relationship well-being were very similar between the groups, with the control mean for the negative scale being 9.1 (SD 2.5), the follow-up mean being 9.1 (SD 2.7), and the control and follow-up means for the positive measure being the same, at 13.5 (SD 2.8).

Other secondary outcomes

Scores on the SWEMWBS measure of mental well-being101 had response rates of 99% at baseline, 90% at follow-up and 98% among the control students. Well-being scores were similar between the study groups [control mean score 21.0 (SD 4.6), follow-up mean score 21.3 (SD 4.2)]. As one of our hypothesised outcomes for the trial was increased well-being among PSs, we looked at differences in well-being by exposure (PSs, exposed students and non-exposed students) but found no differences in scores [mean total score for PSs 20.8 (SD 4.7), mean total score for exposed students 20.8 (SD 4.6) and mean total score for unexposed 21.2 (SD 4.7)].

The response rate for distress about sex life was 92% at baseline, 91% at follow-up and 95% among controls. The proportion of students distressed about their sex life was low across the study population, with 8.2% reporting distress in the follow-up group compared with 10.3% in the control group.

The response rates for the social media use questions were 90–92% at baseline, 90–91% at follow-up and 94% among controls. Cronbach’s alpha for the question bank was 0.78. Rates of using social media to view sexual images or to send, receive or be asked for nude or sexual pictures or videos were slightly lower in the follow-up group than in the control group. Rates of sharing a nude or sexual picture of someone else were very low and almost identical between the control (5.2%) and follow-up (5.0%) groups. The proportion of students reporting that they had used the internet or social media to find information about sexual health was slightly higher in the follow-up group (18.4%) than in the control group (15.4%).

The response rates to the questions about finding it easy to talk to parents, friends and partners about sex were 90–97% at baseline, 89–96% at follow-up and 93–96% among controls. The Cronbach alpha for this question bank was 0.5, suggesting poor correlation between comfort communicating with parents and communicating with friends and partners. The rates of students finding it easy to talk about sex was very similar in the control and follow-up groups (parents: control 17.6% vs. follow-up 19.3%; friends: control 58.3% vs. follow-up 59.5%; partners: control 30.6% vs. follow-up 28.9%), with students being most likely to talk to friends.

Questions assessing sexual competency at first sex were asked of the control group, but not of the follow-up group. This was an adaptation of a measure used in the National Survey of Sexual Attitudes and Lifestyles.99 These measures were summarised as being the right time, both partners being equally willing, contraception being used and a positive reason for having had sex. The response rates were around 94% for these items. PCA showed that there were two clusters among these measures, explaining 70% of the variance in the data. The first cluster was the right time and equal willingness, and the second cluster was use of contraception and a positive reason for having sex. The Cronbach alpha was 0.53 for the first cluster and 0.63 for the second cluster. These may therefore be useful for taking forward as overall measures of sexual competence in the definitive trial. At follow-up, we asked the item on willingness and regret, plus a new one on alcohol and drug use. These items were asked about last sex rather than first sex (to increase the potential number of participants for whom this question would be relevant). The Cronbach alpha for these three items was 0.66, suggesting good utility in a definitive trial.

Effect modifiers

Aside from gender, as previously discussed, various potential effect modifiers of the difference between the follow-up and control groups in the composite primary outcome variable were investigated.

Although none of the interactions with study group reached statistical significance, those which showed potential differences in magnitude were:

• FSM [yes, OR 1.21 (95% CI 0.85 to 1.73); no, OR 0.92 (95% CI 0.37 to 2.30)]
• ethnicity [white Scottish/British, OR 1.14 (95% CI 0.80 to 1.61); other, OR 1.54 (95% CI 0.54 to 4.36)]
• educational attainment [six or more National 5s, OR 1.53 (95% CI 0.92 to 2.55); other, OR 0.91 (95% CI 0.59 to 1.41)]
• religiosity [very/quite important, OR 1.57 (95% CI 0.63 to 3.90); not important, OR 1.11 (95% CI 0.78 to 1.59)]
• number of friends having sex [all/most/some, OR 1.14 (95% CI 0.74 to 1.74); a few/none, OR 1.70 (95% CI 0.86 to 3.36)]
• number of friends who drink alcohol regularly [all/most/some, OR 1.25 (95% CI 0.84 to 1.87); a few/none, OR 1.92 (95% CI 0.87 to 4.23)].

This indicates that the intervention potentially was more effective on behaviour among students on FSM, non-white Scottish/British students, those with higher educational attainment, those to whom religiosity was important and those with peer groups who were less likely to report risky behaviours.

These are therefore variables that should be collected in the definitive trial, to ensure that these potential effect modifiers can be assessed rigorously.

Acceptability of data linkage

We explored the acceptability and feasibility of measuring a biological outcome in the longer term via linkage of student questionnaires to routine health service data on STI diagnosis. To assess acceptability, students were asked a question about data linkage at each survey administration. At follow-up, the question was worded as follows:

In the future it would be useful to link the answers you’ve given us today with information about you that health professionals (doctors, nurses) in the NHS (National Health Service) collect.

This might help us to see if taking part in the STASH study might improve future health.

To link this information we need to get your name, address and date of birth from your school so that it can be linked to your NHS data.

Your details would be kept strictly confidential the linked data would be anonymised (will not identify you) and would only be used for research.

We would like to ask for your consent to link your NHS record with your questionnaire answers.

No, I don’t give my permission  1

Yes, I give my permission  2

In the control year, 280 (43%) students said that they would be willing to give permission for access to their records. In the intervention year, 332 (53%) students said they would be willing at baseline and 352 (66.5%) students said they would be willing at follow-up. Greater willingness at follow-up may be related to higher levels of awareness (and possibly trust) in the study, but is also likely to be because of improved wording with each questionnaire iteration. The question was increasingly simplified and less hypothetical, which may have an impact on acceptability. It is also worth noting that there were slightly more missing data at follow-up, with 82 students not completing the question, compared with 59 students at baseline and 46 control students.

With further improvement in wording, it may be possible to achieve an acceptability of about 70%, which could make data linkage feasible. However, linkage to biological outcomes would be difficult for three reasons. First, prevalence of STIs is low, requiring a large sample to detect a difference. At age 18–19 years, the population prevalence of chlamydia is 4.7% in women and 0.5% in men.2 Second, linkage to health-care records would pick up only individuals attending a clinic for STI testing and would miss individuals with asymptomatic STIs who do not present. If the STASH study is effective in raising awareness of STIs then participants might be more likely to seek help, making differences in diagnosis difficult to interpret. Third, there are limitations to data sources. At present, in Scotland, data on STI diagnosis are captured in the National Sexual Health System. A unique Community Health Index (CHI) patient identifier number is currently recorded for only a proportion of patients, leading to incomplete data. An alternative option would be to link to abortion and pregnancy before age 18 years as a proxy for unsafe sex, but this is not ideal for an intervention primarily addressing STIs, particularly as unplanned pregnancy is declining,123 whereas STIs are not.124

Summary of key findings

Primary outcome (behaviour)

• The suggested primary composite outcome for the definitive trial is ‘always practising safe sex,’ defined as ‘no condomless (vaginal or anal) sex in last 6 months’.
• There was an increase in odds of the (modified) primary composite outcome of 18% following the intervention (compared with the previous year’s control group), with a larger effect among girls.

Knowledge

• High response rates and high Cronbach’s alpha for the question banks.
• Knowledge generally increased regarding both sexual health and where to go for contraception and advice, over baseline and compared with the control group.
• There seemed to be some confusion around the law regarding 15-year-olds having sex with each other, which needs to be addressed in the delivery of the intervention.
• Likely to be a useful measure in the definitive trial.

Attitudes

• High response rates and moderate Cronbach’s alpha for the question bank.
• Attitudes were generally observed to be more positive following the STASH study intervention than among control students.
• May be a useful measure in the definitive trial.

Confidence

• Generally high response rates but slightly lower among students not exposed to the intervention. High Cronbach’s alpha for the question bank.

Relationship well-being

• Very high response rates and good Cronbach’s alpha, particularly for the ‘positive’ relationship well-being subscale.
• Students had similar or slightly higher rates of healthy relationship indicators in the follow-up group than in the control group.
• The composite relationship well-being scores were almost identical, both between the study groups and across exposure groups among follow-up students.
• The relationship well-being measures may not be useful in the definitive trial, although the positive composite measure, in particular, may be useful more generally as an assessment of relationship well-being in research.

Other measures

• Response rates were generally high for these measures.
• Well-being scores were similar between study groups and by exposure.
• Rates of distress about sex life were low and slightly lower among follow-up students than control students. There was little difference by exposure.
• Rates of using social media or internet for positive purposes were slightly higher following the STASH study intervention than among controls, whereas use for negative purposes was slightly lower following the STASH study intervention than among controls.
• Students found it easier to talk about sex with friends than parents or partners, and there was little difference between the study groups.
• Sexual competency may be usefully assessed by two measures: one combining the right time to have sex and equal willingness, and the other combining use of contraception and a positive reason for having sex.
• Well-being scores, distress about sex life, social media use and ease of talking about sex did not show strong utility for the definitive trial. Sexual competency at last sex may have some utility.

Data linkage

Around two-thirds of students at follow-up were willing to provide permission to access their health records, but at this point linkage to STI diagnosis does not seem feasible.