Improving the Quality of Dentistry (IQuaD): A cluster factorial randomised controlled trial comparing the effectiveness and cost-benefit of oral hygiene advice and/or periodontal instrumentation with routine care for the prevention and management of periodontal disease in dentate adults attending dental primary care

Background

Aim

The aim of this study was to compare the clinical effectiveness and cost-effectiveness of theory-based, personalised OHA or PI at different time intervals (no PI, 12-monthly PI or 6-monthly PI), or their combination, with routine care in improving periodontal health in dentate adults attending general dental practice.

Objectives

The primary objectives were to test the clinical effectiveness and cost-effectiveness of the following dental management strategies:

1. personalised OHA versus routine OHA

2. 6-monthly PI versus 12-monthly PI

3. 6-monthly PI versus no PI.

The secondary objectives were to:

1. test the clinical effectiveness and cost-effectiveness of a combination of personalised OHA and PI at different time intervals

2. measure dentist/hygienist beliefs relating to giving OHA, PI and maintenance of periodontal health.

Recruitment and consent of dental practices

Recruitment of dental practices was achieved by a variety of methods. The IQuaD trial was presented at a number of national conferences and promoted within dental professional publications to raise awareness of the trial within the dental profession. In addition, the Scottish Dental Practice Based Research Network and the Faculty of General Dental Practitioners provided trial information to their members. A series of trial information and recruitment evenings were organised across Scotland and north-east England. Potential dentist participants were sent personalised invitation letters to these events in which the reasons for the trial, design of the trial and practice involvement were described, and dental professionals were given the opportunity to discuss participation with the trial team. Information packs about the trial were posted to dentists who were unable to attend a meeting.

Trial team members telephoned dental practices to follow up the trial information letters and noted interest of involvement. Principal dentists (usually the owner of the practice) who were interested in the trial were asked to provide written consent for the dental practice (cluster) to participate. All participating dentists and hygienists within a dental practice were individually invited to complete a consent form and the clinician belief questionnaire (CBQ) prior to cluster randomisation. Participating dental practices were asked to identify dates in advance for training in trial processes and at least three dates for the screening and recruitment of potentially eligible participants.

Recruitment and consent of participants

The identification of potential participants in each dental practice was supported by the Scottish Primary Care Research Network in Scotland and the UK Clinical Research Network in England. A variety of patient appointment management strategies are utilised within dental practices across the UK. Therefore, the IQuaD trial developed a flexible and pragmatic participant recruitment strategy that aimed to adapt to each dental practice’s usual appointment management system. Some dental practices arrange routine check-up appointments for their patients up to 6 months or 1 year in advance, while other practices send letters or mobile phone text message reminders to their patients when their routine dental examinations are due, asking them to contact the dental practice to make an appointment. Dental practices were asked to send an invitation letter to attend a trial screening session, along with a patient information sheet and baseline questionnaire to potentially eligible participants who were due to attend for a routine dental examination and who were regular attenders and did not have severe periodontitis [Basic Periodontal Examination (BPE) score of 4]. This was sent, at most, 6 weeks in advance of the patient’s routine dental examination appointment. Patients who were not interested in taking part were asked to contact their practice for an alternative appointment.

Trial outcome assessor (OA) teams, consisting of qualified dental hygienists/dental therapists and dental nurses employed by the trial, attended the screening and recruitment sessions in participating dental practices to obtain consent from potentially eligible participants and collect the baseline clinical measurements and questionnaires of consented participants.

At the screening appointment, the dentist was available to discuss the trial with potential participants and answer any questions. The OA was present at this appointment and answered any questions specifically related to the trial. Patients who did not wish to take part were seen by their dentist/hygienist, who provided an examination, OHA and/or PI as normal. Potentially eligible participants provided written informed consent to the trial before the baseline clinical outcomes were measured by the OA to confirm a patient’s eligibility for the trial. Reasons for ineligibility or declining to participate were recorded at this session. Participants excluded from the trial solely because of a BPE score of 4 or * (furcation involvement) were not randomised to a PI allocation and were given the opportunity to consent to follow-up by annual postal questionnaires and clinical follow-up in a separate cohort group (see study documentation, available at www.journalslibrary.nihr.ac.uk/programmes/hta/090145/#/; accessed October 2017). All participants received a baseline PI and OHA from their own dentist or hygienist after the baseline outcome measures and questionnaires were collected by the OA team prior to participants being informed of their randomised PI allocation.

Inclusion/exclusion criteria

Trial outcome assessor training

Before participant recruitment and 3 years’ clinical follow-up, the OA teams were trained in the recording of the trial clinical outcomes. The training was delivered by trial collaborators who have extensive experience of clinical periodontal research (PH, GM). The emphasis of the training was on the consistency of the scoring, both intra- and interassessor, to achieve assessor alignment.

The practical steps of clinical outcome assessment were agreed on in advance, including sequence of outcome assessment; time allocation; sequence around the mouth; isolation as well as the angulation, positioning and pressure of the University of North Carolina (UNC)-15 probe to ensure a standardised approach across the OA teams.

It is widely accepted that methods of assessing gingival inflammation that provoke gingival bleeding do not allow for repeated assessment. 29 The training for the primary outcome of gingival inflammation as bleeding therefore involved OA group discussion of the technique of assessment, scoring definition, as well as demonstrated photograph and clinical examples. This was repeated for the outcome of calculus.

Periodontal probing depths were recorded at six sites on all erupted teeth using a probing force of approximately 25 g on an independent, but similar, cohort of patients to those recruited to the study at the Newcastle upon Tyne and Dundee dental schools. Assessments by an experienced clinical periodontal researcher (LH) were used the reference standard. Intra- and interoutcome assessor scores were recorded and kappa scores of ≥ 0.60 achieved, with over 95% of scores within 1 mm.

Oral hygiene advice

Personalised oral hygiene advice

The personalised OHA intervention used two well-established psychological models for explaining and influencing behaviour (Figure 2). The first was a pre-motivational model, social cognitive theory. 30 This theory proposes that a key variable influencing motivation and behaviour is self-efficacy, assessed as a person’s confidence in their ability to perform behaviour. According to this model, the sources of self-efficacy are doing the behaviour (performance: cleaning their teeth), seeing someone else do it (vicarious example/modelling: the dentist demonstrated to patients how to use the oral health-care tools to clean their teeth on a model of the mouth), being encouraged to perform it (verbal persuasion: the dentist helped the patient make oral hygiene a habit with the right plan) and how we feel afterwards (physical state: the dentist discussed biofeedback, highlighting what clean teeth would feel like). Based on this model, an intervention to influence oral hygiene behaviour should target oral hygiene self-efficacy via these sources. The second model was implementation intention theory,31 a post-motivational model designed to help people put their intentions into practice, bridging the gap between motivation and behaviour. This theory proposes that making an explicit action plan about where and when a behaviour will be performed increases the likelihood of performing it. Action plans work by setting up a cue to remind the individual to perform the behaviour. Therefore, oral hygiene behaviours are likely to be very sensitive to action planning, as they can easily be linked to other behaviours most people do every day, for example tooth brushing after the cue of eating a meal or before going to bed.

FIGURE 2.

The OHA intervention behavioural framework.

The intervention was tested in a study32 consisting of 84 dental practices and 799 patients. The results of this pilot study32 supported the theoretically framed intervention as an effective method of influencing oral hygiene beliefs and behaviour. Participants who received the intervention had significantly better behavioural (timing, duration, method), cognitive (confidence, planning) and clinical (plaque, gingival bleeding) outcomes than the participants in the control group receiving routine care.

Periodontal instrumentation

The definition of PI was as used in standard practice and would include the removal of plaque and calculus from the crown and root surfaces using manual or ultrasonic scalers, with no adjunctive subgingival therapy (e.g. antibiotics)33 and the appropriate management of plaque retention factors.

Practice allocation to oral hygiene advice group

Recruited dental practices were allocated to routine or personalised OHA by minimisation on two factors: (1) practice employed a dental hygienist (yes/no) and (2) practice size (one or two dentists in practice/three or more dentists). This cluster-level randomisation was conducted using the automated, central randomisation service at the Centre for Healthcare Randomised Trials (CHaRT), University of Aberdeen, after the dental practice consent form was received at the Trial Coordinating Office in Dundee (TCOD) and before any potential participant had been approached.

Participant allocation to periodontal instrumentation group

Participants were allocated to the PI trial arms using the automated, central randomisation service at the CHaRT, University of Aberdeen, with access by both telephone and web. Allocation took place once the OA had completed the baseline outcome assessment and was minimised on (1) absence of gingival bleeding on probing (yes/no), (2) highest sextant BPE score (BPE score of < 3/BPE score of 3) and (3) currently smoking (yes/no).

The OAs were informed that allocation had taken place but were blinded to allocation, with the actual allocation transmitted to the TCOD. A letter was sent to all participants to inform them of their PI allocation and all participants received a £25 gift voucher in recognition of their contribution to the study. The TCOD provided each individual recruited dental practice with a list of projected dates for PI and review (no-PI group), according to the individual participant-level allocation, to be delivered throughout the trial for all their trial participants. The list was updated and re-sent by TCOD on an annual basis to each practice.

Outcome measures

Baseline

Dentists and/or hygienists were asked to complete the CBQ at baseline, after each individual dental practice had consented to take part in the IQuaD trial.

Patient-centred outcomes were collected at baseline by self-administered questionnaire.

Arrangement of clinical assessment appointments has been outlined in the recruitment section above.

Clinical outcomes were measured at baseline by trained OAs who were blinded to allocation. Gingival inflammation/bleeding scores, calculus, clinical probing depth and BPE scores were measured by the OA and recorded on the baseline clinical chart by the dental research nurse who was a member of the trial team.

At the baseline appointment, the OAs also collected personal details from participants, including preferred contact method (telephone, e-mail) and contact details.

Annual follow-up

Patient-centred outcomes were collected annually by self-administered postal questionnaires.

Like the baseline questionnaire, the annual follow-up questionnaire contained questions on self-efficacy, PBC, attitude, subjective norm, behaviour, OHIP-14, dental appearance and sensitivity. Questions designed to collect the descriptive measures were also included in this questionnaire, as were questions on dental costs for the health economic outcomes. Annual follow-up questionnaires at year 1 and year 2 post randomisation were posted from the CHaRT trial office to participants’ home addresses, along with a covering letter and reply-paid envelope for return of the completed questionnaire. Those participants who failed to return their questionnaire within 3 weeks were sent a reminder letter, a further copy of the questionnaire and a reply-paid envelope. If questionnaires were returned to the trial office marked ‘return to sender’, every effort was made to obtain updated contact details from the participant’s dental practice.

Three-year follow-up

As at baseline, dentists and hygienists were asked to complete the clinician CBQ. These were collected by the OAs when they visited the dental practices to collect participant clinical outcomes.

Patient-centred outcomes were collected at the 3-year follow-up using the annual follow-up questionnaire, with an additional question on self-reported bleeding. The questionnaire was sent to participants at least 3 weeks before the date of the first 3-year follow-up appointment made at the dental practice where they were recruited. Participants who had not returned a questionnaire by the time of their follow-up appointment were asked to complete a shortened one-page version of the follow-up questionnaire containing questions on the primary patient-centred outcome (self-efficacy) at that appointment. Participants who did not attend the follow-up appointment and who had not returned a questionnaire were sent a reminder letter, a further copy of the questionnaire and a reply-paid envelope to return the completed questionnaire to the CHaRT trial office.

All trial participants were invited to attend a trial follow-up assessment appointment by their dental practice either at the time of their routine check-up or at a separate trial assessment appointment at which the clinical outcomes were measured by trained OAs who were blinded to allocation. As at baseline, gingival inflammation/bleeding scores, clinical probing depths and calculus were measured by the OAs and recorded on the follow-up clinical chart by the dental research nurse who was a member of the trial team.

Participants who could not attend were contacted and given the option of attending on at least one other day or time.

All participants who attended the follow-up assessment received a certificate of appreciation for their participation in the trial, details of when the trial results would be published and a £25 gift voucher in recognition of their contribution.

Oral hygiene advice

Assuming a conservative estimate of the intracluster correlation coefficient (ICC) of 0.05, a cluster RCT of 50 dentists collecting information from 25 patient participants each (25 × 25 = 625 patient participants per arm) was expected to have 90% power to detect a difference of 7.5%. Should the ICC be 0.1, the trial would still have approximately 80% power to detect a difference of 7.5%.

Periodontal instrumentation

Given that the comparison of routine versus personalised OHA required 625 participants in each arm, equal randomisation 1 : 1 : 1 (no PI, 12-monthly PI, 6-monthly PI) of participants implied 208 in each of the six groups. Assuming no interaction effect, the corresponding PI groups could be combined across both routine and personalised OHA groups, requiring 416 patients allocated to each PI group. Based on a sample size of 416 in each group, the trial has in excess of 95% power for each pairwise comparison to detect a difference of 7.5% in the percentage of gingival sites that bleed on probing.

Collection of providers outcomes

Differences in the frequency of PI visits would have an impact on clinicians’ costs and benefits. The effect on incomes, job satisfaction and changes to the level of fees on the provision of PI were assessed using self-reported questionnaires administered to clinicians over the duration of the trial.

Statistical analyses of outcomes

Statistical analyses were based on all participants randomised. Three principal comparisons and their interactions were tested in the factorial design. The principal comparisons were between (1) all those allocated to routine OHA and all those allocated to personalised OHA, (2) all those allocated to no PI and those allocated to 6-monthly PI and (3) all those allocated to 12-monthly PI and those allocated to 6-monthly PI.

Reflecting the clustered nature of the data, the clinical outcomes were compared using mixed models with dental practice as a random effect, and with adjustment for minimisation variables [practice employs dental hygienist (yes/no) and practice size (one or two dentists in practice/three or more dentists) for the cluster-level randomisation and absence of gingival bleeding on probing (yes/no), highest sextant BPE score (BPE score of < 3/BPE score of 3) and currently smoking (yes/no) for the patient-level randomisation] and participant baseline values (when available) as fixed effects. Patient-centred outcomes were compared using a mixed model with a random effect for the participant (to take into account the three-level nested and correlated data structure: a variable number of observations nested within participants and participants grouped in dental practices) and another for the centre and adjusting for the same variables as before. Statistical significance was at the 5% level, with corresponding confidence intervals (CIs) derived.

Pre-planned subgroup analyses on the primary outcome included exploration of participant age (< 45 years, 45–64 years or ≥ 65 years), smoking status (non-smoker or smoker), periodontal disease severity (no clinical signs or presence of gingival bleeding on probing) and intervention provider (dentist or practice hygienist). Post hoc subgroup analyses were undertaken by region (Scotland or England) and clinical pocket depth (four or more sites with clinical pocket depth of ≥ 4 mm, or fewer than four sites). These analyses were conducted by including a subgroup by treatment interaction term in the primary outcome model described above. Conservative levels of statistical significance (p <  0.01) were sought, reflecting the exploratory nature of these pre-planned and post hoc subgroup analyses.

Non-response analysis

Descriptive data are presented, comparing the baseline characteristics of participants who did and who did not attend the 3-year follow-up clinical appointment. The t-test (continuous outcomes) and chi-squared test (dichotomous outcomes) were used to estimate the statistical significance of the differences between responders and non-responders.

Missing items

Missing items were dealt with according to the relevant published recommendations for the specific instruments. 35 For missing items with no published recommendation, self-efficacy, PBC, attitude and subjective norm were calculated as a mean of the items available. Behaviour and intention were calculated using complete cases.

Sensitivity analysis

To assess the possible effect of outliers, treatment effects by individual dental practices were explored to assess the influence of between-centre differences in the primary clinical outcome: bleeding. Any practices where the treatment effect excluded zero from the 99% CI and had at least twice the target difference estimate (either –15% or 15%) were excluded from the analysis comparing no PI with 6-monthly PI, as well as from the subgroup analysis.

Health services costs

Participant costs

Three sources of participant cost were considered for the analysis: (1) charges for NHS dental care, (2) privately purchased dental care products and treatments and (3) time and travel costs (for participants and companions) associated with visits to the dental practice.

Statistical analysis of cost data

Cost data were analysed according to best practice methodology for cluster randomised trials. 52–54 We used multilevel mixed-effects models that account for clustering of the data through a cluster-level random effect. 52 Models for cost data were estimated using the meglm command in Stata^® 14 (StataCorp LP, College Station, TX, USA) specifying a gamma family and log-link function for cost data to account for skewness of the distribution. The decision to use this model for costs was based on a combination of the results of a modified Park’s test, which identified both an inverse Gaussian and gamma distribution to be acceptable, and software restriction (only gamma family allowed within Stata’s meglm command). The log-link function was decided on as it allowed easier model convergence and had a preferred Akaike information criterion score than an identity link. The models were estimated by maximum likelihood. Incremental costs were estimated as the mean cost difference between randomised policies (relative to the policy most representative of standard care: routine OHA with 6-monthly PI), together with 95% CIs. All models were adjusted for baseline characteristics (age, age-squared and sex) as well as minimisation variables (cluster level: hygienist employed at practice and practice size; individual level: smoker, gingival bleeding at baseline and BPE score band).

Owing to the high level of completeness, descriptive analyses of NHS dental costs and participant co-charges were based on complete cases. Participant costs were presented and analysed using multiple imputation of missing cost data from participant annual questionnaires. For the analysis of total costs within the CBA, the small proportion of missing routine data were also imputed to complete the data set. The imputation process followed best practice guidelines. 55,56 All imputation was completed using Stata’s multiple imputation procedure. Missing cost data were imputed at the level of cost component (e.g. missing data on private PI imputed for each annual questionnaire). Costs were imputed using predictive mean matching accounting for the repeated measures nature of the costs from the annual participant questionnaires. The average of the five closest values was used for each imputed data point. Data were imputed separately for each randomised group to preserve the allocation of participants. Imputation models were adjusted for cluster, age and sex as well as minimisation covariates. Sixty imputed data sets were used to ensure stable results and were combined using Rubin’s rules to generate estimates of costs. 57

Selection of attributes and levels

The relevant attributes were, to a large extent, determined by the trial. Preferences needed to be elicited for the specific interventions (PI and personalised OHA) and for the outcomes that may vary across the arms of the trial (i.e. self-reported bleeding gums and perception of appearance and feel of cleanliness). We followed recommended practice to assess whether or not these attributes were important to individuals and whether or not there were any other important attributes that should be included in the DCE. We conducted a literature review and focus groups (FGs) with the general population to develop the final list of attributes and levels to present in the DCE (see Appendix 2, Section 2: research conducted for questionnaire development). We also engaged with, and sought advice from, a range of stakeholders, including clinical dental experts, practising dentists and hygienists, and patient representatives on our trial advisory group about the DCE content. Table 4 shows the final list of attributes and levels included in the DCE.

Additional levels were added for provider of service (dentist/hygienist) as this was found to be an important consideration in the FGs. The FGs showed that respondents would have difficulty understanding and attaching value to the primary clinical trial outcome (gingival inflammation/bleeding). Therefore, we included the self-reported frequency of bleeding measure from the participant questionnaire instead. Respondents were informed that having bleeding gums increases the risk of tooth loss in the future. The look and feel attribute was also based on the measures used within the same questionnaire. The cost attribute was framed as an annual commitment over a 3-year time period (trial follow-up). The levels of the cost attribute were determined from a number of sources, including the FGs, baseline trial data on WTP for PI and a supplementary questionnaire asked to FG respondents. Figure 3 provides an example choice set included in the DCE.

FIGURE 3.

Example choice task. Note: the example choice set shown here is for a respondent categorised as having poor dental health. See Appendix 2, Section 1: additional detailed methods for costing, the discrete choice experiment and mapping discrete choice experiment valuations to the trial outcomes, for more information on segmentation of choice sets based on dental health category. This hypothetical respondent has chosen Dental Care Package B and is willing to pay £100 per year for the package.

Experimental design

Choices presented to respondents in the DCE were drawn from the set of all possible attribute and level combinations. With a single three-level attribute and four five-level attributes, the total number of combinations in the full factorial is given as 3¹ × 5⁴ = 1875, resulting in [(1875 × 1874) ÷ 2] = 1.76 million unique choice sets. Following standard practice in DCE literature, a D-efficient design was used. 58 Ngene version 1.1.2 software (Ngene, ChoiceMetrics Pty Ltd, Sydney, NSW, Australia) was used to create a main effects design consisting of 30 choice set questions. To ensure a manageable number of choice tasks, the tasks were split into three blocks, each with 10 questions. Two further choice questions were added to each block to check the validity of responses as follows: (1) choice 11 repeated choice 5 to test for consistency of preferences and (2) choice 12 was a non-satiation (dominance) test in which one option should be preferred over its alternatives if individuals value PI polish and personalised OHA. Therefore, each respondent completed a total of 12 choice tasks. The responses to the internal validity choice tasks were not included in the analysis; therefore, a total of 10 choice sets were analysed for each respondent.

As respondents were likely to vary considerably in terms of how many PIs they usually receive and the extent to which they experience bleeding gums, a segmented pivoted design was used to increase realism. For example, it is unrealistic that a respondent who has no bleeding gums and never has a PI would suddenly experience bleeding very often if they opt out. Respondents were assigned to one of three segments [(1) good, (2) moderate or (3) poor dental health] based on self-reported frequency of bleeding gums and the look and feel of their teeth (both attributes in the DCE). The pivoted design tailored the presented levels based on the segment to which the respondents were assigned. For example, respondents categorised as having ‘good’ dental health would never be presented with the worst level of bleeding gums and unclean teeth, as it would be highly unlikely for dental health to deteriorate so quickly because of forgoing PI or detailed OHA. The list of attribute levels included in the experimental design for the good, moderate and poor dental health segments can be found in Appendix 2.

Questionnaire development

Think-aloud interviews were used to test the wording and understanding of the questionnaire. Extensive piloting was used to assess the best way of reducing hypothetical bias and to ensure that attributes and levels were sensible, realistic and tradable to the majority of respondents. Full details of all methods used to select attributes and levels (literature searches and FGs) and test their feasibility in the survey (think aloud and piloting) are outlined in Appendix 2, Section 2: research conducted for questionnaire development.

The survey was divided into three sections. Section 1 asked about respondents’ experiences of dental care, how often they attend their dental practice and who they are treated by. Section 2 included questions about bleeding gums and aesthetic appearance to determine the appropriate version of the DCE to present. Section 3 concluded the survey with demographic questions and a number of debriefing questions about respondents’ views and attitudes to dental topics. The questionnaire was designed using Qualtrics (Provo, UT, USA).

Data collection

Data were collected using Qualtrics online panels. The survey was nationally representative, with population census quotas for age (among adult population), sex and region sought. We oversampled in Scotland (n = 125) to enable subgroup analysis across the main participating regions in the trial (England/Scotland). We further sought to achieve a mix of respondents with recent experience of dental services and without. During the pilot phase, we attempted to achieve 30% of responses from non-regular attenders, defined as not having seen the dentist in the previous 2 years. 11 However, we were unable to achieve this sample in the pilot and relaxed the target to a more achievable 10% for the main phase of data collection. Responses were anonymised and respondents were free to leave the survey at any point without having to give a reason for doing so. Respondents who completed the survey were reimbursed in a manner determined by the survey panel from which they were partaking.

Data analysis

The DCE data were analysed using best practice methods59 and followed random utility maximisation theory. 60 The utility of the option (V_j) is a linear function of the attributes and levels presented to the respondents, where:

Alternative-specific constant (ASC) is accounting for latent or unobserved utility associated with choosing any package, regardless of the attribute levels, β represents the marginal utilities associated with the attributes and levels. All categorical variables were included in the model using effects coding. The advantage of effects coding is that the reference level of an attribute is uncorrelated with the ASC. The reference categories for the effects coded attributes are no OHA, no PI, never bleeding and having teeth that look and feel very clean. For example, the value for ‘no advice’, β_{no advice}, is retrieved as the negative sum of β₁ Advice D + β₂ Advice H.

The marginal WTP of an attribute level is equal to the marginal rate of substitution between that level and the cost attribute. For ease of interpretation, we also estimate the difference in WTP between the reference level and the attribute level. For example, the annual marginal WTP for OHA from the hygienist compared with the reference level (no OHA) is equal to:

Mixed logistic regression models in preference space were used to analyse the data to determine the impact of the attributes and levels on preferences for dental care services and outcomes. The ASC was assumed random and normally distributed for all models, reporting mean and SD. All other attribute levels were fixed effects, reporting mean coefficients only. The final model was estimated using the maximum simulated likelihood method, with 2000 Halton draws. All models were estimated using Stata 14 software.

The model was re-estimated separately using interaction terms for several pre-determined subgroups including (1) sex; (2) region, to determine if system of reimbursement has an impact on preferences; (3) smoking status, to determine if current or previous smokers’ preferences differ from the overall group; and (4) household income, to determine if greater ability to pay (annual income of ≥ £20,800) had an impact on preferences. Further subgroup analyses explored if the preferences of those with experience of PI or of treatment from a hygienist differed from the overall sample. We planned to conduct subgroup analysis by dental attendance (regular/non-regular attenders) but were unable to achieve sufficient numbers of respondents attending the dentist less often than every 2 years to power such an analysis. The impact that subgroups had on the main attribute level effects was assessed by the significance of interaction terms between subgroup identifier and attribute level coefficient. Likelihood ratio tests were used to test the joint significance of the interaction terms.

Estimating benefits (willingness to pay) for each trial participant

Benefits measured as WTP were estimated for each trial participant by attaching WTP values from the DCE to the relevant treatments provided (PI and personalised OHA) and to the relevant trial outcomes (self-reported bleeding when brushing and aesthetic outcome). The DCE provided estimates of WTP for zero, one and two PIs provided per year or a total of zero, three or six PIs over the trial period. A step-wise linear utility function was used to estimate WTP for the number of PIs other than those presented in the DCE. It was assumed that the provider of the baseline PI (dentist or hygienist) also delivered the intervention over follow-up. The impact that the results of this assumption have is also explored in a sensitivity analysis.

Willingness to pay for personalised OHA was attached based on randomised group, as all participating practices complied with their allocation to provide either personalised or routine OHA at the baseline visit. For the remaining attributes, including aesthetic outcome and frequency of bleeding gums when brushing, WTP values were directly attached at an individual level in the trial. The full mapping process for each attribute to the trial interventions and outcomes can be found in Appendix 2.

Discounting was applied at a rate of 3.5% per annum. 42 Discounted marginal WTPs were summed to generate a total WTP for each trial participant. Total WTP data were analysed using multilevel mixed-effects models with a random effect capturing cluster effects to account for the hierarchical nature of the trial. A normal distribution was assumed for benefits data; hence, the analyses were implemented using Stata’s mixed command. All models were adjusted for individual- and cluster-level minimisation variables, age and sex. The base-case analysis undertakes statistical imputation of missing data on benefits (i.e. WTP mapped to trial outcomes). Data were imputed for annual questionnaires at the level of WTP for each outcome (e.g. bleeding gums), following a similar procedure to that outlined for costs.

Statistical analysis of (incremental) net benefit

The imputed data sets were used to estimate net benefits and incremental net benefits. Net benefit was calculated for each participant in the trial according to Equation 3:

Multilevel mixed-effects models were used to analyse the data, implemented using Stata’s ‘mi estimate: mixed’ command following the approach of Gomes et al. 52,53 These models can explicitly recognise clustering in parameter estimates though the incorporation of cluster-level random effects (ujc, uje), while also adjusting for cluster- and individual-level covariates (minimisation variables: age and sex). The model addresses both individual- and cluster-level correlation between costs and benefits, using a bivariate normal distribution of the error terms. The model assumes linear additive effects for treatments and covariates. Restricted maximum likelihood was used to estimate the model for the base-case analysis. Interaction terms were included to retrieve the true impact of randomisation on costs/benefits. 54 Stata’s mimargins command was used to retrieve the values of the reference category of cost/benefit from the imputed data sets and the nlcom command was used to calculate incremental costs, incremental benefits and incremental net benefits across groups, together with 95% CIs.

Presentation of cost–benefit analysis results

The base-case perspective is that of the UK NHS. A wider perspective, including both NHS and participant costs, was considered as a secondary analysis. All results are presented as incremental costs, incremental benefits and incremental net benefits, relative to standard care. Note that exclusion of dominated policy options is not required in the absence of a threshold value of a ratio between costs and benefits (as would be the case in a cost–utility analysis).

Confidence ellipses, based on parametric calculations using the model variance–covariance matrix, were used to illustrate the uncertainty in each comparison of costs and benefits graphically on the cost–benefit plane. 61 The confidence ellipses illustrate the probability that each intervention is less/more costly and less/more beneficial than standard care. All policies with positive incremental net benefits are associated with higher levels of welfare than standard care. Note that this can be generated by a higher level of benefit and/or a lower level of cost of the policy. The strategy with the greatest incremental net benefit would be considered the preferred policy from a welfare perspective. The analysis is solely based on incremental changes in net benefits between policies and, as is common for economic evaluations alongside trials, makes no statement about the magnitude of the underlying net benefits of specific policies.

Analysis of uncertainty

The confidence ellipses of the cost–benefit plane illustrate the joint uncertainty surrounding estimates of incremental costs and incremental benefits on results. Further deterministic sensitivity analyses, outlined in Table 5, investigated the impact of assumptions around cost data, benefit estimation and model structure on the results.

Cost–consequences analysis

In addition to the results of the CBA conducted for the primary economic analysis, we further considered all of the available information across all components of the study, using a narrative cost–consequences analysis (CCA) framework. The CCA is a useful approach to summarise all the study information succinctly for decision-makers. We use a balance sheet approach summarising the advantages and disadvantages of each policy option under consideration. The CCA includes the following information:

• the results of clinical effectiveness, using both primary and secondary outcomes defined in the protocol

• the results of all economic analyses, including costs to the NHS and participants, general population preferences for PI/OHA (results from the DCE) and overall incremental net benefits

• a summary of information sourced from the CBQ investigating the practitioner’s perception of the advantages and disadvantages of different policy strategies.

No attempt was made to quantitatively synthesise the information across different categories. The results of the CCA are presented qualitatively as a means to aid decision-makers’ understanding of the overall outcomes of the trial as well as the trade-offs between clinical and economic outcomes.

Dental practices

Sixty-eight dental practices were initially recruited in total, with 49 in Scotland and 19 in England: 34 practices were allocated to give routine OHA and 34 to provide personalised OHA. From the 68 practices, four were excluded post randomisation in the routine OHA arm and one in the personalised OHA arm. The exclusions were due to a trial procedural error in the early recruitment phase of the study whereby practices were randomised when a dentist in a practice had verbally given consent to take part but before the practice principal dentist (usually the owner of the practice) had signed a consent form for the practice to take part in the trial. The principal dentist did not give permission to use their practice. The practice study intervention allocation was not known to any potential dentist, practice principal dentist (or patient) participants at the time of non-consent. Selection bias due to the post-randomisation exclusions was not, therefore, possible. Hence, there were 63 recruiting practices in total.

Participants

A total of 2341 patients were screened for trial entry and 183 (8%) of the screened patients were found to be ineligible (Table 6). The primary reason for ineligibility was due to a BPE score of 4 or * (160 patients). From those ineligible as a result of a BPE score of 4 or *, 144 (90%) patients agreed to join a separate cohort group. Further details about the cohort participants are given in Chapter 6. There were 281 patients eligible for the study who were not recruited. The main reason for non-recruitment was that the patient did not attend the baseline appointment (see Table 6).

Participants were recruited in 63 dental practices (see Appendix 3). A total of 1877 participants were recruited to the study, with 867 in the routine OHA arm (n = 289 allocated to receive no PI, n = 290 allocated to receive 6-monthly PI and n = 288 allocated to receive 12-monthly PI) and 1010 in the personalised OHA arm (n = 334 allocated to receive no PI, n = 337 allocated to receive 6-monthly PI and n = 339 allocated to receive 12-monthly PI). Of the total participants, 1348 (72%) were recruited across Scotland, including 121 (6%) recruited from the Scottish islands. No dental practice recruited > 2.5% of the participants. The average cluster size was 30.6 and 28.9 participants for personalised OHA and routine OHA practices, respectively. Three participants were subsequently identified to be non-NHS patients at consent (and therefore ineligible) and were thus excluded post randomisation.

Practice characteristics

Table 7 provides the dental practice characteristics and beliefs. The majority of practices employed hygienists and had three or more dentists. Participating dentists and hygienists completed a questionnaire at baseline (the CBQ) on their beliefs about providing OHA. Scores for belief variables varied from 1 to 7, where 7 represented the most positive outcome (e.g. the highest level of confidence or the most positive attitude). Intention and subjective norm were the exceptions: intention varied from 0 to 100, where 100 represented the highest level of intention, and subjective norm varied from 1 to 49, where 49 represented the highest level of subjective norm. The results suggested that dental professionals had positive attitudes towards, and high levels of intention of, providing OHA. There was little perceived influence of peers in the ability of the respondents to provide OHA (subjective norm) (see Table 7). Overall, there were no important differences between groups on any of the practice characteristics or beliefs at baseline.

Participant characteristics

Clinical characteristics

Table 8 displays the results of the clinical assessment at baseline. The mean number of teeth per participant was 24. The highest sextant BPE score was ≤ 2 in two-thirds of the participants. The mean per cent of sites bleeding was 33%, and 35% of teeth had calculus present. The mean clinical probing depth was 1.8 mm and between 10% and 12% of participants in each group had four or more sites with a clinical probing depth of ≥ 4 mm.

There were no important imbalances across the randomised groups.

Patient-reported outcomes

The baseline patient-reported outcomes summary is shown in Table 9. More details about individual questions that produced the scales are given in Appendix 1.

Self-efficacy, the patient-reported primary outcome, was, on average, 5.2 points (on a scale of 1 to 7 points with 7 points meaning very confident).

Cognition variables

Perceived behaviour control was, on average, 4.5 points; attitude towards oral health behaviour was very positive and was, on average, 5.8 points and subjective norm was 5.3 points (scales of 1 to 7, with 7 meaning very confident).

Behaviour

The mean self-reported oral hygiene behaviour score was 4.7 points. The scale varies from zero (poorest hygiene behaviour) to 9 (perfect hygiene behaviour). The behaviour scale had space for improvement across the three behaviours measured: (1) frequency of brushing, (2) duration of brushing and (3) frequency of interdental cleaning (either flossing or the use of interdental brushes). In both brushing frequency and duration, around 10% of participants scored the maximum of 3 points (brushed more than twice a day, for longer than 2 minutes). Regarding interdental cleaning, around 20% of participants scored the maximum of 3 points (floss or use interdental brushes every day).

Intention

The mean intention to perform good OHA practice was 5.5 points (on a scale from 0 to 9, with 9 being the best outcome).

Quality of life

The mean score on the oral health QoL scale, OHIP-14, was 6.3 points on a scale form 0 to 56 (with 0 indicating perfect oral health QoL). The OHIP-14 75th percentile was 9 and the 99th percentile was 39, illustrating a generally good oral health-related QoL in this population.

The baseline participant self-reported data suggested that participants were confident that they could perform good toothbrushing practice, thought oral hygiene practice was a positive thing to do, were influenced by their peers to do it and intended to do it. However, the participants did not currently actually perform good oral hygiene practice and reported that there were external factors, PBC, that limited their ability to do so. Overall, across the randomised groups, there were no apparent imbalances in the patient-reported outcomes.

Attendance at 3-year clinical examination

The primary clinical outcome was collected at the 3-year clinical follow-up. Overall, 71% of the participants attended the appointment (see Appendix 1). Twelve participants were known to have died by the end of the 3-year follow-up. The main reasons for non-attendance were that the practice was unable to contact the participant (41%), the participant was unable to attend (30%) and the participant did not want to attend (13%). There were no important differences between groups for the reasons of non-attendance.

A comparison of baseline characteristics between those who did and those who did not attend the final appointment was undertaken. The results in Table 10 show that only a few factors differed between attenders and non-attenders. Attenders were, on average, older (50 vs. 43 years). Although there was a statistically significant difference in the OHIP-14 score, with the mean score reported by non-attenders higher than that reported by attenders, the difference of 1.3 points (0.15 of the SD of the score) was not thought to be clinically important. There was no evidence that non-attenders had different levels of disease, with BPE scores, bleeding, calculus and probing depths all very similar to those seen in attenders.

Annual questionnaire returns at years 1, 2 and 3

Approximately 77% of participants completed a follow-up questionnaire at 3 years. There were no substantive differences in response rates between the randomised groups. The overall rates of return of follow-up questionnaires at 1, 2 and 3 years are shown in Appendix 1.

Participant dental characteristics at 3 years

The participant dental characteristics at 3 years are shown in Table 11. The profile of characteristics is similar across the groups and similar to baseline responses (e.g. smoking status, type of toothbrush used, preferences for two PIs per year; see Tables 1 and 8). Characteristics for years 1 and 2 are given in Appendix 1, Section 2: participant dental characteristics (years 1 and 2).

Primary outcomes

Gingival inflammation/bleeding

The pre-chosen clinical primary outcome was mean gingival bleeding at the 3-year follow-up. The means and 95% CI for baseline and follow-up for each randomised group are displayed in Figure 5. All randomised groups followed a similar pattern: bleeding increased from baseline (mean 32% of sites) to 3-year follow-up (mean 39% of sites). The differences between groups and corresponding 95% CIs are shown in Table 12.

FIGURE 5.

Gingival inflammation/bleeding (mean and 95% CI), by randomised allocation. (a) OHA delivery and (b) PI interval.

TABLE 12 - Treatment effects for primary outcomes

Primary outcomes	Estimate, p-value (95% CI)
	No PI vs. 6-monthly PI	12-monthly PI vs. 6-monthly PI	Personalised OHA vs. routine OHA
Gingival inflammation/bleeding
Unadjusted	0.71 (–2.0 to 3.7), 0.579	–0.53 (–3.4 to 2.3), 0.715	0.02 (–6.3 to 6.4), 0.985
Adjusted for baseline bleeding	0.8 (–1.7 to 3.2), 0.534	0.1 (–2.4 to 2.5), 0.947	–1.9 (–8.1 to 4.3), 0.549
Adjusted for baseline bleeding and minimisation variables	0.87 (–1.6 to 3.3), 0.481	0.11 (–2.3 to 2.5), 0.929	–2.5 (–8.3 to 3.3), 0.393
Self-efficacy
Unadjusted	–0.017 (–0.133 to 0.098), 0.767	–0.082 (–0.198 to 0.033), 0.162	0.002 (–0.144 to 0.147), 0.984
Adjusted for baseline self-efficacy	–0.098 (–0.189 to –0.007), 0.035	–0.027 (–0.118 to 0.064), 0.564	0.003 (–0.110 to 0.115), 0.962
Adjusted for baseline self-efficacy and minimisation variables	–0.028 (–0.119 to 0.063), 0.543	–0.097 (–0.188 to –0.006), 0.037	0.017 (–0.089 to 0.123), 0.750

Three types of analyses are reported: (1) an unadjusted analysis, (2) an adjusted-for-baseline bleeding analysis and (3) a fully adjusted model accounting for baseline bleeding and minimisation variables (this is the prespecified primary model). Under ITT analysis, there was no evidence of a statistical or clinical difference between those randomised to receive 6-monthly PI and those randomised to receive no PI (difference 0.87%, 95% CI –1.6% to 3.3%; p = 0.481). Similarly, there was no evidence of a difference between 6-monthly PI and 12-monthly PI (difference 0.11%, 95% CI –2.3% to 2.5%; p = 0.929). The 95% CIs were small enough to exclude the prespecified clinically important difference of 7.5% in bleeding. There was also little evidence of a difference between participants randomised to personalised OHA and those randomised to routine OHA (difference –2.5%, 95% CI –8.3% to 3.3%; p = 0.393), although the 95% CI did not entirely rule out a 7.5% reduction. The results were robust to other adjusted/unadjusted models. The interaction between personalised OHA and 6-monthly PI was 1.7% (95% CI –3.8% to 7.3%). The interaction demonstrated that a participant receiving both personalised OHA and 6-monthly PI would have a further 1.7% bleeding reduction compared with just adding the individual effects of 6-monthly PI with personalised OHA. The interaction term ruled out an additional 7.5% reduction. The ICC at follow-up for bleeding for all participants was 0.23 (95% CI 0.16 to 0.31). A full description of the means (SD) for gingival inflammation at each time point is given in Appendix 1, Section 3: clinical and patient-reported outcomes by year of follow-up.

Self-efficacy

The pre-chosen patient-centred primary outcome was self-efficacy at the 3-year follow-up. The means and 95% CIs for baseline and follow-up for each randomised group are displayed in Figure 6, by randomised allocation. All groups followed a similar pattern. The score was similar from baseline (mean 5.2 points) up to the year 3 questionnaire (mean 5.3 points) in both the PI and the OHA groups. The differences between groups and corresponding 95% CIs are shown in Table 12. Under ITT analysis, there was no evidence of a difference between those randomised to receive 6-monthly PI and those randomised to receive no PI (difference –0.028, 95% CI –0.119 to 0.063; p = 0.543). Between those randomised to receive 6-monthly and those randomised to receive 12-monthly PI, there was a statistically significant difference, favouring the 6-monthly PI (difference –0.097, 95% CI –0.188 to –0.006; p = 0.037). However, this result is not clinically significant. There was also no evidence of a difference between participants randomised to personalised OHA and those randomised to routine OHA (difference 0.017, 95% CI –0.089 to 0.123; p = 0.750). The interaction coefficient between personalised OHA and 6-monthly PI was –0.007 (95% –0.22 to 0.20), suggesting no evidence of important interaction between interventions. The self-efficacy ICC for all participants was 0.06 (95% CI 0.04 to 0.09). A full description of the means (SD) for self-efficacy at each time point is given in Appendix 1, Section 3: clinical and patient-reported outcomes by year of follow-up.

FIGURE 6.

Self-efficacy score (mean and 95% CI), by randomised allocation. (a) OHA and (b) frequency of PI.

Secondary outcomes

A full description of the means (SD) for the secondary outcomes at each time point is given in Appendix 1, Section 3: clinical and patient-reported outcomes by year of follow-up.

The clinical secondary outcomes were calculus and mean clinical probing depth collected at the 3-year follow-up. The patient-reported secondary outcomes were PBC; attitude; subjective norm; behaviour and intention collected at 1, 2 and 3 years; and private PI received throughout the trial collected at year 3. Self-reported bleeding was a post hoc secondary outcome collected at 3 years. The differences between groups, and corresponding 95% CIs, are shown in Table 13.

Calculus

The calculus means and 95% CIs for baseline and follow-up for each randomised group are displayed in Figure 7. The mean level of calculus decreased across the no PI, 12-monthly and 6-monthly PI groups. The calculus level was statistically significantly higher in the no-PI group than in the 6-monthly PI group (mean 8.0%, 95% CI 5.4% to 10.7%, p < 0.001). There was no evidence of a difference in mean calculus between 12-monthly PI and 6-monthly PI or between personalised and routine OHA (Table 13).

FIGURE 7.

Calculus percentage (mean and 95% CI), by randomised allocation. (a) OHA and (b) frequency of PI.

Clinical probing depth

The mean clinical probing depth and 95% CI for baseline and follow-up for each randomised group are displayed in Figure 8. There was no evidence of changes from the baseline clinical pocket depths and no evidence of differences between any of the randomised comparisons (see Table 13).

FIGURE 8.

Mean clinical probing depth (mean and 95% CI), by randomised allocation. (a) OHA and (b) frequency of PI.

Patient-reported outcomes

The patient-reported outcomes (PBC, attitude and subjective norm) are shown in Figures 9–11. There was no evidence of any clinically important differences between the randomised groups in any of the cognitive variables at any follow-up time (see Table 13). Attitude was statistically significantly different between participants randomised to 6-monthly PI and those randomised to no PI, favouring a better attitude to oral hygiene behaviour in the no-PI group (difference –0.137, 95% CI –0.273 to –0.001; p = 0.048). However, the 95% CI excludes the possibility of a clinically important difference.

FIGURE 9.

Perceived behaviour control score (mean and 95% CI), by randomised allocation. (a) OHA and (b) frequency of PI.

FIGURE 10.

Attitude score (mean and 95% CI), by randomised allocation. (a) OHA and (b) frequency of PI.

FIGURE 11.

Subjective norm score (mean and 95% CI), by randomised allocation. (a) OHA and (b) frequency of PI.

Behaviour score

The mean behaviour score and 95% CI for baseline and follow-up for each randomised group are displayed in Figure 12. The behaviour score was significantly different between participants randomised to the 6-monthly PI and those randomised to 12-monthly PI, favouring those in the 6-monthly group (difference –0.164, 95% CI –0.296 to –0.032; p = 0.015) (see Table 13). However, the 95% CI excludes the possibility of a clinically important difference. Therefore, there was no evidence of any clinically important differences between the randomised groups.

FIGURE 12.

Behaviour score (mean and 95% CI), by randomised allocation. (a) OHA and (b) frequency of PI.

Intention score

The mean intention score and 95% CI for baseline and follow-up for each randomised group are displayed in Figure 13. There was no evidence of any clinically or statistically significant differences between the randomised groups for the intention score at any of the follow-up times (see Table 13).

FIGURE 13.

Intention score (mean and 95% CI), by randomised allocation. (a) OHA and (b) frequency of PI.

Oral Health Impact Profile-14

The OHIP-14 score was overall very low (representing a good oral QoL) across the randomised groups at all follow-up time points (Figure 14), varying from mean 6.0 at baseline to mean 5.0 at 3 years. There was no evidence of any clinically important differences between the randomised groups for OHIP-14 scores at any of the follow-up times (see Table 13).

FIGURE 14.

The OHIP-14 score (mean and 95% CI), by randomised allocation. (a) OHA and (b) frequency of PI.

Tertiary outcomes

Appendix 1, Section 4: tertiary outcomes (at baseline and 3 years) provides the self-reported data on dental sensitivity and appearance for baseline and 3 years. Possible answers to all the questions varied from 1 point (not clean at all/not at all pleasant) to 7 points (could not get any cleaner/extremely pleasant). At baseline, patients reported feeling that their teeth were clean after brushing (on average, 5.8 points), but cleaner after PI (on average, 6.6 points). The same was reported for how clean their teeth look (after brushing 5.3 points vs. after a PI 6.2 points), how pleasant they feel (5.8 points after brushing vs. 6.3 points after PI) and how pleasant they look (5.1 points after brushing vs. 5.7 points after PI). Around half of the patients reported feeling sensitivity in their teeth. Dental sensitivity and appearance characteristics were balanced across randomised groups.

At 3 years, participants reported that their teeth felt and looked clean and pleasant after brushing (on average, around 5.5 points out of 7 points) and even more clean and pleasant after PI (on average, around 6.4 points out of 7 points). The average scores were similar across randomised groups, indicating that having fewer PIs throughout the trial did not influence the participants’ perceptions.

Fidelity of interventions

Subgroup analyses

Prespecified subgroup analyses for gingival bleeding included currently a smoker (yes/no), BPE score (of < 3 or of 3), age (< 45/45–64/> 64 years) and whether or not the practice employed a hygienist (yes/no). Post hoc subgroup analyses were undertaken for participants with four or more sites with a clinical probing depth of ≥ 4 mm (yes/no) and region (Scotland/England). Figures 15 and 16 show the means and 99% CIs for the differences in bleeding at 3 years in the subgroups for PI frequency and OHA, respectively. There was no evidence that any of the subgroups were statistically significantly different at the 1% level (Table 16).

FIGURE 15.

Subgroup results for PI allocation: difference between arms, by subgroup.

FIGURE 16.

Subgroup results for OHA allocation: difference between arms, by subgroup.

Sensitivity analyses

NHS dental care resource use

Data were linked for 1337 out of 1346 (99%) trial participants randomised in Scotland (after post-randomisation exclusion) and 477 out of 525 (91%) trial participants randomised in England. It was not possible to link data for 11 participants in Scotland, but data were available for two participants randomised in England having some dental treatment in Scotland. Linkage was not possible for 48 participants in England. A further participant had claims in both Scotland and England. For this individual, claims were summed across both data sets, with the participant remaining allocated to their region of randomisation for the purposes of reporting results.

Table 17 describes the total number of treatment claims provided to all randomised participants with linked data over the 2.25-year follow-up period (excluding baseline and final study visits) by randomised group. The majority of claims in England fell into band 1 treatments. There were relatively few emergency claims and free treatments across all groups. The highest number of claims in Scotland was for diagnostic assessments, including clinical examinations and radiographs. There were 2516 periodontal treatments provided, accounting for 23% of claims. Among the remaining categories, the greatest number of claims were for conservative treatments [2506 claims (23%)] such as fillings, with relatively few surgical extractions or more intensive treatments. There were clear differences across groups in terms of numbers of periodontal treatments reflecting the different number of PIs delivered in each randomised trial arm. Overall, owing to the restrictions of the reimbursement system, there were only 11 claims for ‘preventative care’ (including intensive hygiene instruction), all in the personalised OHA cluster. Detailed information on items claimed in Scotland can be found in Appendix 2, Section 3: detailed claims for Scottish routine data.

Table 18 shows that the numbers of PIs delivered to all randomised participants with linked data were evenly balanced across clusters. The mean number of reported PIs was lower in England than in Scotland. It should be noted that the number of PIs in England may be under-reported as only one qualifying procedure for a band 1 treatment needs to be reported when filling in the clinical data set (e.g. a clinical examination).

NHS dental care costs

Table 19 presents complete-case NHS dental costs for the UK and at the regional level based on the appropriate region-level unit costs applied to resource use data. Incremental differences are presented for each randomised policy, compared with standard care.

The mean cost per trial participant over the 2.25-year follow-up period (excluding baseline and final study visits) ranged from £62.42 to £85.43 across the randomised groups. The SDs are generally large, suggesting relatively large variation in costs across participants. Data were somewhat skewed to the left, with some outliers incurring very high costs; however, these were spread evenly across groups.

The average cost is lowest for the no-PI groups. However, none of the differences in mean cost between standard care (routine OHA and 6-monthly PI) and the other policies considered was statistically significant at a 5% level. The largest difference in costs is between standard care and a policy of no PI with personalised OHA (mean difference –£14.91; 95% CI –£34.18 to £4.36).

Table 19 also reports the NHS costs by region. This is important given the differences in reimbursement systems and the level of detail on resource use. Average per-participant costs are higher in England than in Scotland. However, the overall results are the same across the two regions: there are no statistically significant differences in costs between standard care and the other five groups. The CIs for incremental costs are generally wider for England, which is not surprising given the smaller sample sizes.

The Scottish data provide much more detail on resource use than the English data and can, therefore, provide additional insights into differences in different cost categories. The costs of periodontal treatments were expected to be lower for no PI and 12-monthly PI than for standard care, and this was found to be the case. Mean cost differences of periodontal treatments in Scotland for each randomised group, compared with routine OHA and 6-monthly PI were as follows: routine OHA, no PI, –£7.47 (95% CI –£11.22 to –£3.72); routine OHA, 12-monthly PI, –£5.11 (95% CI –£8.96 to –£1.26); personalised OHA, no PI, –£9.31 (95% CI –£13.30 to –£5.32); personalised OHA, 6-monthly PI –£2.00 (95% CI –£6.76 to £2.75); and personalised OHA, 12-monthly PI, –£3.78 (95% CI –£8.29 to £0.73). There were no significant differences across groups in any of the other categories of cost. Given the relatively low average NHS costs of PI, the differences in PI costs did not lead to differences in the total costs.

Participant exemptions from NHS co-charges can have an impact on the split of dental charges between NHS and participant. Therefore, it is important to explore whether or not there are differences in the proportion of participants exempt across the randomised groups. Table 20 shows that around 24% of participants were exempt from charges. There were no significant differences across randomised groups in terms of the proportion of participants exempt from payment (χ² 6.82; p-value of 0.234). As a robustness check, we also analysed the data assuming that all participants are exempt, that is, the NHS pays the full cost of dental care provided. Relative to standard care, for this exploratory analysis, incremental costs were as follows: routine OHA, no PI, –£4.76 (95% CI –£30.53 to £21.01); routine OHA, 12-monthly PI, –£1.89 (95% CI –£28.04 to £24.27); personalised OHA, no PI, –£25.65 (95% CI –£53.15 to £1.85); personalised OHA, 12-monthly PI, £10.95 (95 CI –£20.17 to £42.06); and personalised OHA, 6-monthly PI, £11.21 (95% CI –£19.97 to £42.40). Furthermore, using similar regression models as for the main results, there was no evidence that exemption status had an impact on the number of PIs delivered (p = 0.518). The main conclusion was the same: there was no statistically significant difference in costs across the randomised groups.

Other NHS-incurred costs

The analysis protocol set out to measure costs to the NHS of primary and secondary care, based on annual questionnaire data, for problems related to participants’ teeth. There were 193 out of 1630 (12%) participants who reported at least one contact with secondary care services over the 3 years of follow-up. This relatively high number of participants reporting secondary care use was unexpected given the relatively good dental health of the trial population. A validation exercise was conducted by comparing secondary care contacts from the trial with the general population using routine data and by checking self-report data against practice records. In terms of the routine data, the average annual rate of outpatient consultations in the trial (1 : 19) was six times greater than the corresponding number of consultations in the general population in Scotland (1 : 121) and three times greater than the corresponding rate in England (1 : 70). A similar result was found when inpatient admissions were compared, with an average annual rate among trial participants (1 : 70) 10 times higher than the general population in Scotland (1 : 722) and over three times higher than the general population rate in England (1 : 249). It should be noted that the general population rates for England and Scotland should not be directly compared as the data are based on different assumptions. For 80 out of the 193 trial participants reporting resource use (23 in England and 57 in Scotland), data were checked against dental practice records for evidence of referral for secondary care treatment (across all hospitals, dental or otherwise). Fewer than one in four reports were validated against practice records. Table 21 summarises the results of the validation exercise.

Given the poor level of validation of other NHS resource use against routine data, the lack of evidence of dental care referrals in practice records and the fact that the general population may be expected to have, on average, poorer rather than better dental health, the data on use of NHS resources have been excluded from the base-case analysis. Appendix 2, Section 4: participant-reported contact with non-dental health services presents descriptive data for primary and secondary care resource use falling outside the dental budget for information only. The data are excluded from all subsequently reported results.

Participant-incurred costs

Discrete choice experiment

Results of analysis models

The results of the mixed logit model are presented in Table 26.

The positive value for the ASC indicates a preference among the general population to have any package of dental care as opposed to none. The significance of the SD around the constant term indicates the presence of significant preference heterogeneity for opting into a dental care package, with some respondents more likely to opt out (choosing no dental care package). Across all choice sets, 288 (43%) respondents always opted in to the choice task and 45 (7%) always opted out, choosing no dental care package.

Respondents valued having both personalised OHA and PI regardless of provider. They preferred to receive personalised OHA and 12-monthly PI from the dentist. However, who provided the care was less important for the 6-monthly PI. As expected, respondents prefer to have gums that do not bleed and teeth that look and feel clean. The negative coefficient on the annual cost attribute indicates that respondents prefer dental packages that cost less.

Willingness-to-pay estimates, together with their CIs, should be interpreted as the general population’s valuation of each attribute and level. The differences between WTP values indicate how much the general population values moving from one health-care package to another. For example, the general population would be willing to pay an additional £90.53 per year over 3 years to move from having bleeding gums very often (very often: –£54.13) to never having bleeding gums (never: £36.40). They would be willing to pay £145.50 per year to move from having teeth that look and feel very unclean to teeth that look and feel very clean. The difference between these valuations indicates that the general population attaches a greater value to shifts in average aesthetic outcome than to changes in frequency of bleeding 3 years after signing up to the dental package.

The WTP estimates can thus be used to calculate the general population’s WTP for packages of dental care. Assuming that a participant in the trial received no OHA/6-monthly PI from the dentist, hardly ever had bleeding gums at the final annual questionnaire and had teeth that, on average over the period, looked and felt moderately clean, the general population would be willing to pay £93.93 per year for this package (i.e. £43.95 – £13.50 + £29.83 + £20.55 + £13.10). Following a similar approach, the results of the DCE are mapped to each individual to enable a CBA to be completed alongside the trial.

Validity of responses

The estimates of annual WTP for PI (relative to none) fall within a feasible range of prices charged in the private market for PI (i.e. between £49 and £86 per PI provided by the dentist). In terms of the internal validity checks, 535 (80%) respondents passed the consistency of preferences check (i.e. answering the same way for exactly the same choice twice). A total of 498 (75%) respondents passed the non-satiation (dominance) test. Very few respondents failed both tests (≈5%). The high percentage passing such tests is in line with other DCE studies. 65

Analysis of subgroups

Subgroup analyses were conducted for participant characteristics by sex, region and income, as well as experience of PI and dental hygienists. Table 27 describes the impact of subgroup membership on preferences, through interaction terms between effects coded categorical variables and attribute-level main effects. Subgroup effects were considered significant at a p-value of < 0.01.

The likelihood ratio tests show no overall effect on preferences (jointly across all model parameters) for UK region or smoking status. For subgroups for which the tests indicated an impact of subgroup on preferences, the following findings were observed:

• Respondents with experience of the hygienist valued the dental care packages more highly (disutility associated with cost attribute was lower). They also gained less utility from teeth that look and feel only moderately clean.

• The significant interaction effect on the ASC with experience of PI indicates that those with experience of PI are more likely to commit to a dental care package. They also valued the dental care packages more highly.

• Neither experience of seeing the dental hygienist or of having PI affected the preferences of respondents for having these services, or for whether it is the dentist or hygienist who provides the service.

• Females in the sample gained greater utility from dental care packages in which teeth look and feel clean or very clean. Similarly, they experienced greater disutility from teeth that look and feel moderately clean or unclean.

• As expected, the cost attribute had less of an impact on the preferences of higher-income respondents.

The subgroup analyses indicate that, although there were some differences across subgroups, in general, the direction of attribute-level effect remained similar with respondents preferring PI and personalised OHA, less bleeding and teeth that look and feel clean and healthy, as well as preferring lower-cost dental care packages.

Benefits: willingness to pay

Table 28 presents the mean WTP values across randomised groups, including the component parts of total WTP based on the mapping of DCE results to treatment received and trial outcomes. The average total benefits in terms of WTP ranged from £79.09 to £207.65 across the groups. The policy of 6-monthly PI and personalised OHA is associated with the highest average benefit. The difference in benefits between this policy and standard care (routine OHA and 6-monthly PI) is statistically significant. Compared with standard care, having no PI or a 12-monthly PI and routine OHA is associated with significantly lower benefits. There was no significant difference between the benefits for personalised OHA with either 12-monthly PI or no PI and standard care, implying that a reduction in PI can be compensated for by the provision of personalised OHA. The findings were consistent across regions. The average benefit was generally lower in England, as PIs were more likely to be provided by hygienists in England than in Scotland (following the assumption that the provider of baseline OHA/PI would consistently provide the trial intervention over follow-up).

The differences in benefits observed between groups were driven primarily by the differences in WTP for PI and personalised OHA. For example, the 6-monthly PI with routine OHA had a higher WTP (–£27.43) associated with PIs than no PI with routine OHA (–£82.92), a difference of £55.49. Note that the negative mean WTPs for PI across all groups are due to the relatively few PIs provided across all groups, and the use of effects coding in which WTP values are estimated relative to the mean of the attribute. In terms of interpretation, the focus should therefore be on differences in WTP for PI across groups. Similarly, for example, 6-monthly PI with personalised OHA had a higher WTP for the OHA component (£19.88) than did 6-monthly PI with routine OHA (–£30.38), an additional WTP of £50.26 for the OHA provided in the personalised group. There were some differences in benefits across regions, driven by the provider of care (more treatments assumed to be provided by hygienists in England than in Scotland). The differences across groups appear to be driven mainly by the services provided rather than the bleeding or aesthetic outcomes. This observation matches the findings reported in the clinical effectiveness chapters.

Cost–benefit analysis

Cost–consequences analysis

The balance sheet of costs and outcomes outlined in Table 32 narratively summarises the advantages and disadvantages of the different policy considerations within a cost–consequences framework. The analysis makes no attempt to conduct any further quantitative synthesis of economic findings.

The data presented in the balance sheet indicate the trade-offs decision-makers need to consider when determining the most efficient policy of PI and OHA to implement. It shows that, although personalised OHA with 6-monthly PI is the most cost-beneficial policy, owing to the high value placed on the service by the general population, there is no clear evidence of clinical benefit, or changes to provider beliefs. The preferred policy depends on the viewpoint taken. If the aim is to maximise welfare from a fixed NHS budget, then the CBA analysis shows that personalised OHA with 6-monthly PI is the preferred policy. If the aim is to maximise health from a fixed NHS budget, then it could be argued that no PI is the preferred policy given the lack of evidence of clinical benefit and given that it was associated with the lowest average costs (although it should be noted that differences were not significant).

Baseline characteristics

Of the 2341 potential participants who attended for a baseline clinical outcome assessment, 160 were found to have a BPE score of 4 and/or *. These potential participants were invited to participate in the IQuaD cohort study, with 16 (10%) declining to participate (reasons outlined in Table 6), resulting in 144 (90%) participants being recruited to the longitudinal study.

Clinical outcomes

Clinical outcome assessment, measurement and outcome mean calculations followed the same protocol as the IQuaD trial (outlined in Chapter 2, Collection of clinical outcome measures). At baseline, the 144 cohort participants had a mean number of 22 (SD 4.7) teeth present, with gingival inflammation/bleeding recorded in 35.1% (mean) (SD 23.4%) of sites. Supragingival calculus was recorded on 40.6% (mean) (SD 30.2%) of teeth and the mean clinical probing depth of the cohort patients was 2.2 mm (SD 0.4 mm).

Similar to the IQuaD trial, a number of recruited participants were not available for follow-up clinical assessment; 85 (59%) IQuaD cohort participants attended for follow-up clinical assessment at approximately 3 years following recruitment. When considering only the 85 participants who attended for both clinical assessment appointments, the baseline clinical characteristics were as follows: a mean number of 21.6 (SD 5) teeth present, 37.0% (mean) (SD 24%) of sites with recorded gingival inflammation, mean supragingival calculus of 36.0% (SD 27%) and mean clinical probing depth of 2.2 mm (SD 0 mm). A total of 97% (n = 82) of these participants had a BPE score of 4 or * at baseline; the remaining three participants, who were ineligible due to medical history reasons, had a BPE score of 3 (1%, n = 1) or 2 (2%, n = 2).

Patient-centred outcomes

The collection, assessment and mean score calculations of the participant-centred outcomes are the same as in the IQuaD trial outlined in Chapter 2.

At baseline, the cohort participants’ self-efficacy, assessed as confidence in one’s ability to perform a behaviour and measured using a 7-point scale scored from 1 (not at all confident) to 7 (extremely confident), was, on average, 5.0 points (SD 1.2 points; completed the questionnaire, n = 131).

Cognitions

Perceived behaviour control score, assessed in terms of perceived ease or difficulty of performing the behaviour and measured using a 7-point scale varying from 1 (strongly agree) to 7 (strongly disagree), was, on average, 4.1 points (SD 1.4 points; completed the questionnaire, n = 131).

Attitude towards oral health behaviour, measured using a scale varying from 1 (strongly agree) to 7 (strongly disagree), was very positive, averaging 5.6 points (SD 1.2 points; completed the questionnaire, n = 131).

Subjective norm, assessed as attitude towards the behaviour and perceptions of social pressure to perform the behaviour and measured using a scale of 1 (strongly agree) to 7 (strongly disagree) scored, on average, 4.9 points (SD 1.3 points; completed the questionnaire, n = 131).

The score of mean intention to perform good oral hygiene practice was 5.5 points (SD 1.8 points; completed the questionnaire, n = 114), which was measured using questions to which responses were scored from 0 to 3 (the best possible intention), with 9 being the best possible score.

Behaviour

On the self-reported oral hygiene behaviours scale, which ranges from 0 to 9, with 9 being the best possible behaviour, the cohort group had an average score of 4.9 points (SD 1.7 points; completed the questionnaire, n = 130). Like the IQuaD trial participants, the cohort group showed room for improvement on this scale.

Quality of life

The self-reported QoL was measured using the OHIP-14, a 14-item oral health-specific patient-centred outcome referring to symptoms in the past 12 months. Each item is scored from to 0 to 4 (very often) and the scores are summed to produce a summary score ranging from 0 to 56, with 56 being the worst outcome. The cohort participants’ mean score at baseline was 8.5 points (SD 9.5 points; completed the questionnaire, n = 126).

Clinical

Follow-up examinations were conducted from May 2015 to July 2016. At the 3-year clinical follow-up, the 85 participants had an average of 21.2 (SD 5.4) teeth with mean gingival inflammation/bleeding recorded at 41% (SD 25.5%) of sites. More teeth (mean 43.5%, SD 33.1%) exhibited supragingival calculus and the mean clinical probing depths had increased minimally by 0.1 mm to 2.3 mm (SD 0.5 mm).

Patient-centred outcomes

At the 3-year follow-up, the cohort participant patient-centred outcomes were similar to the baseline scores. The average self-efficacy score was 5.1 points (SD 1.1 points; completed the questionnaire, n = 101) on a scale of 1–7.