Oral splints for patients with temporomandibular disorders or bruxism: a systematic review and economic evaluation

Description of the intervention

Oral splints are removable appliances that can cover all or some of the teeth in either the maxillary or the mandibular arches. The term ‘oral splint’ is used colloquially in (UK) dentistry and is really a misnomer, as oral splints do not actually splint (i.e. immobilise) anything. Splints can also be known variously throughout the literature and the world as oral appliances, devices, orthotics or biteplates.

Oral splints can resemble a device similar to a mouthguard used in contact sports, overlaying the biting surface of the teeth with some type of material. Numerous types of oral splints are available, varying in design, material, coverage and application. Splints cover either the upper teeth (upper splints) or the lower teeth (lower splints) and can be classified by the type of material they are made from: hard (hard acrylics), soft (soft polymers or plastics), or composite amalgams of the two aforementioned materials. 8 They can then be subdivided into whether they cover all the surfaces of the teeth in one jaw (full coverage) or only some of the teeth surfaces (partial coverage, e.g. covering only the front six to eight teeth, or two to four of the anterior incisor teeth), and whether or not they provide an adjusted biting surface to equalise the way the teeth meet the splint (‘occlusally adjusted’ surface). 9,10 Finally, they may be made from impressions of a patient’s teeth (custom made) or adapted directly onto the teeth from a non-specific blank (prefabricated or non-custom made).

It should be noted that there are multiple names for different types of splints, and many variations on a design theme. For example, an upper hard stabilisation splint is also known as a Michigan splint, and a Lucia jig is similar in design to the proprietary Nociceptive Trigeminal Inhibition Tension Suppression System (NTI-tss)™ (National Dentex LLC, Palm Beach Gardens, FL, USA) splint.

Traditionally, oral splints recommended by dentists have been custom made, often in dental laboratories, sometimes requiring a number of appointments. More recently, a vast array of prefabricated splints have become available, either for provision by the dentist or health-care worker at a single appointment, or as over-the-counter purchases for patients who wish to self-manage their symptoms. 11 Prefabricated splints include soft, rubber splints (which function by separating the teeth); hydrostatic splints, which are cushioned with fluid to redistribute occlusal force; and the NTI-tss device (semi-customisable).

The aims, duration of treatment, need for adjustments, perceived mode of action and the costs of the splints vary across splint types.

How the intervention might work

There is continuing debate about the exact mechanism of action of oral splints. However, mechanisms include:

• muscle relaxation/habit-breaking for patients with increased parafunctional or muscle-tightening habits

• protection of teeth and jaws, particularly when teeth clenching and grinding may lead to damage of teeth, resulting in the need for restorative treatment

• normalising periodontal ligament proprioception, by utilising a splint to spread the forces placed on individual teeth

• repositioning of the jaws and condyles into centric relation

• central effects that are yet to be fully understood. 12

The mode of action varies according to the type of splint used, with some splints (permissive) allowing the teeth/jaw to move or glide over the biting surfaces unimpeded (permissive splints) and others having indentations that hold the jaw in a fixed position (directive or non-permissive).

Why it was important to do this review

This systematic review arose from a National Institute for Health Research Health Technology Assessment programme call addressing the research question: ‘what is the clinical effectiveness and cost-effectiveness of prefabricated oral splints and custom-made splints for the treatment of orofacial symptoms?’. Our application was successful and we received funding to conduct this systematic review and economic evaluation, so the objectives of this review have been driven by this.

It should be noted that the original call focused on treatment for orofacial symptoms. The causes of orofacial pain are varied, but splint therapy for orofacial pain is primarily limited to pain resulting from TMD. Splint therapy is also used for non-painful TMD and bruxism. In order to reflect the use of oral splints in dental practice in the UK, the review will focus on TMD (pain related and non-pain related) and bruxism.

Although we used Cochrane methods, this was not undertaken as a Cochrane review; however, we will share all data from the screening of studies, data extraction forms and correspondence with authors of any future Cochrane reviews, or review updates that overlap with the scope of this review.

Dentists in the NHS in both primary and secondary care are currently providing oral splints for patients who have orofacial signs (such as tooth wear in patients with bruxism) or symptoms (primarily pain). In Scotland alone, the number of splints provided in NHS primary care is increasing from 1985 custom-made hard splints in 2005/06 to 3521 custom-made hard splints in 2015/16. Dentists in Scotland have also recently been allowed to provide custom-made soft splints on the NHS; 16,888 were provided in 2015/16. Oral splints are also provided privately and directly to patients, with a growing industry reported. 11

Despite the frequent use of splints for the management of orofacial sign and symptoms, their clinical effectiveness and cost-effectiveness remain uncertain. This research proposal will inform the NHS, dentists and patients as to whether or not oral splints provided by dentists or other health-care workers are effective in reducing orofacial symptoms (primarily pain) and when they are indicated to prevent tooth wear. If oral splints are found to be effective, then the effectiveness of prefabricated splints compared with custom-made splints (laboratory made, requiring more than one visit to the health-care worker to fit) will be evaluated to help inform care pathways for the target population.

If prefabricated splints are found to be at least as effective as custom-made splints, then there is the potential for a cost saving to both the NHS and directly to patients. Currently, in primary care, the provision of custom-made oral splints for these patients is a band 3 charge to the patient under the 2016 NHS dental fee scale (£256.50). Prefabricated splits are a much cheaper alternative to custom-made splints as they require only one visit for fitting rather than two, do not require laboratory costs and are a band 2 charge in the NHS (£59.10 in 2016 values). Over-the-counter splints can be purchased for < £10.

Types of studies

We included randomised controlled trials (RCTs) but did not include crossover studies as we do not feel that this is an appropriate design owing to the transient nature of the TMD symptoms, or bruxism in patients (which may be due to external factors such as stress).

Types of participants

Inclusion criteria: children (aged > 11 years) and adults who have either TMD or bruxism, and the dentist or other health-care worker is considering treating the patient with an oral splint, in either primary or secondary care.

Exclusion criteria: studies in which the majority of participants were undergoing fixed or removable orthodontic treatment.

Types of interventions

Two comparisons are made:

1. Splints versus no splints, which included any type of splint provided for patients, as described in Types of participants. The no-splint group also included a control splint, which is used in some trials, watchful waiting or minimal treatment. Minimal treatment included advice/counselling, education or self-performed exercises (but could not involve multiple visits/appointments).

2. Prefabricated splints versus custom-made splints. No other head-to-head comparisons were included between different splint types.

For clarity, we refer to a splint according to the jaw in which it is used (upper/lower), its material (hard/soft/composite), its degree of coverage of teeth (full/partial) and then its most generic name, unless the proprietary name is particularly pertinent.

Types of outcome measures

Electronic searches

The following databases were searched:

• Cochrane Central Register of Controlled Trials (CENTRAL) in The Cochrane Library (to issue 9, 2018, searched on 1 October 2018)

• MEDLINE via OvidSP (from 1946 to 1 October 2018)

• EMBASE via OvidSP (from 1980 to 1 October 2018)

• Cumulative Index to Nursing and Allied Health Literature (CINAHL) via EBSCOhost (1937 to 1 October 2018).

When appropriate, the searches of these databases were linked to study design search filters developed by Cochrane for identifying reports of randomised and controlled clinical trials. They were undertaken without restrictions on language or date of publication.

Searching other resources

Unpublished data on clinical trials was sought via searches of the US National Institutes of Health trials register (ClinicalTrials.gov) and the World Health Organization International Clinical Trials Registry Platform, which includes trials data from the European Union, the UK, Australia, China, the Netherlands, Brazil, India and Republic of Korea (South Korea). Conference proceedings were searched via EMBASE in the main literature search, and the Web of Science. Abstracts of dissertations and theses were searched via the ProQuest database. Searches of these databases were also undertaken on 1 October 2018, without any restrictions on date of publication or language.

Additional grey literature was sourced through the American Academy of Dental Sleep Medicine website. 13 The International Association of Dental Research (IADR) annual conference abstracts were searched via the IADR website14 on 1 October 2018. The protocol stated that we planned to search the conference proceedings of the American Academy of Orofacial Pain and the European Academy of Craniomandibular Disorders; however, these were not available to us.

Data collection and analysis

Data extraction and management

Two review authors independently extracted the following data from the included trials:

• location/setting, type of provider, number of centres, recruitment period, trials registry identifier

• inclusion/exclusion criteria, age and sex of participants, number randomised/analysed, any other important prognostic factors (i.e. comorbidities, concomitant prescription medicines/co-interventions)

• population characteristics – age, sex, presenting condition [bruxism, TMD (plus subtype) or mixed] and severity, duration since presenting condition began, comorbidities

• intervention – primary purpose of splint (e.g. pain reduction, bruxist motor activity reduction, to aid functional rehabilitation, to decrease tooth damage, jaw repositioning); type of splint in terms of jaw worn in (upper/lower), material (hard/soft/composite), teeth coverage (full/partial), design (prefabricated/custom made); duration of splint use

• detailed description of comparator

• details of the outcomes reported, including method of assessment and time(s) assessed

• details of sample size calculations, funding sources, declarations/conflicts of interest.

Assessment of risk of bias in included studies

The assessment of risk of bias was done independently and in duplicate, using the Cochrane Risk of Bias tool. 15 The following domains were assessed: sequence generation (selection bias), allocation concealment (selection bias), blinding of participants and personnel (performance bias), blinding of outcome assessors (detection bias), incomplete outcome data (attrition bias), selective outcome reporting (reporting bias) and other bias. We realised that it would be difficult or impossible to blind participants and personnel to whether or not a participant had been randomised to receiving a splint. This could potentially introduce performance bias, and, in the case of subjective outcomes, detection bias.

The overall risk of bias of individual studies was categorised as being low, high or unclear according to the following:

• low risk of bias (plausible bias unlikely to seriously alter the results) if all domains had a low risk of bias

• unclear risk of bias (plausible bias that raises some doubt about the results) if one or more domains had an unclear risk of bias

• high risk if one or more domains had a high risk of bias.

Measures of treatment effect

For continuous outcomes [e.g. pain on a visual analogue scale (VAS)], we used the means and standard deviations (SDs) reported in the trials to express the estimate of effect as mean difference (MD) with a 95% confidence interval (CI). In the event that different scales were used, we expressed the treatment effect as a standardised mean difference (SMD).

For dichotomous outcomes (e.g. jaw clicking/no jaw clicking), we expressed the estimate of effect as a risk ratio (RR) with a 95% CI.

Unit-of-analysis issues

The patient was the unit of analysis for all included studies.

Dealing with missing data

We attempted to contact the author(s) of all included studies, if feasible, in the event of missing data. Missing SDs were estimated according to the methods for estimating missing SDs described in section 7.7.3 of the Cochrane Handbook for Systematic Reviews of Interventions. 15

Assessment of heterogeneity

If a sufficient number of studies were included in any meta-analyses, we planned to assess any clinical heterogeneity by examining the following characteristics of the studies: the similarity between the types of participants [TMD, bruxism; age (< 18 and ≥ 18 years)], the type of health-care worker providing the splints, the type of splint, the control intervention and the outcomes.

We assessed heterogeneity statistically by using a chi-squared test, in which a p-value of < 0.1 indicates statistically significant heterogeneity. We quantified heterogeneity by using the I² statistic. A guide to the interpretation of the I² statistic, as given in the Cochrane Handbook,15 is as follows:

• 0–40% – might not be important

• 30–60% – may represent moderate heterogeneity

• 50–90% – may represent substantial heterogeneity

• 75–100% – considerable heterogeneity.

Assessment of reporting biases

If a sufficient number of studies had been included in any meta-analyses, publication bias would have been assessed in accordance with the recommendations on testing for funnel plot asymmetry,16 as described in section 10.4.3.1 of the Cochrane Handbook for Systematic Reviews of Interventions. 15 If asymmetry had been identified, other possible causes of asymmetry would have been assessed, as outlined in table 10.4.a of the Cochrane Handbook. 15 We were unable to undertake funnel plot analysis on the main primary outcome because the effect estimate was reported as SMD.

Data synthesis

We carried out meta-analyses only if there were studies of similar comparisons reporting the same outcomes. We performed meta-analyses using Cochrane’s Review Manager software (RevMan version 5.3, The Cochrane Collaboration, The Nordic Cochrane Centre, Copenhagen, Denmark) and exported the forest plots into this document to graphically display the results. We combined MDs (or SMDs when different scales were used) for continuous data, and RRs for dichotomous data. Our general approach was to use a random-effects model. With this approach, the CIs for the average intervention effect are wider than those that would have been obtained using a fixed-effect approach, leading to a more conservative interpretation.

We used additional tables (see Appendix 2, Tables 26–29) to report the results from studies not suitable for meta-analysis.

For the meta-analysis of splints versus no splints, we planned to include prefabricated and custom-made splints as subgroups; however, there was an insufficient number of studies including prefabricated splints. Pooling across subgroups depended on the degree of heterogeneity/subgroup differences. As an additional analysis, if we had determined that there was evidence that the prefabricated splits, when placed by any health-care professional, are effective for the primary outcomes, then we planned to look at any head-to-head RCTs comparing the delivery of prefabricated splints by different types of health-care workers. There was insufficient evidence to undertake this.

We planned to consider undertaking a network meta-analysis for different splint types; however, there were insufficient data to undertake this.

Subgroup analysis and investigation of heterogeneity

For the meta-analysis of splints versus no splints, we planned to include the following subgroups:

• prefabricated

• hard custom-made splints that alter occlusion (jaw relationship)

• hard custom-made splints that do not alter occlusion (jaw relationship)

• soft custom-made splints that do not alter occlusion (jaw relationship).

There were insufficient data to undertake this.

Sensitivity analysis

For TMD patients, we undertook a sensitivity analysis restricted to trials in which the inclusion criteria were based on, or could be clearly mapped to, one of the following sets of diagnostic criteria:

• Research Diagnostic Criteria for Temporomandibular Disorders (RDC/TMD) guidelines17

• Diagnostic Criteria for Temporomandibular Disorders (DC/TMD) guidelines18

• American Association of Orofacial Pain (AAOP) guidelines. 19

Similarly, for bruxism patients, we planned to undertake a sensitivity analysis restricted to trials for which there was a clear diagnosis of bruxism. 4 The study should have used polysomnography to diagnose the bruxism. There were insufficient trials to do this.

We planned to test the robustness of our results by performing sensitivity analyses based on excluding studies deemed to be at high and unclear risks of bias from the analyses. However, we knew this was unlikely to be possible for the splint versus no splint comparison if we judged that there was a high risk of performance bias or detection bias or both.

If any meta-analyses had included several small studies and a single very large study, we planned to undertake sensitivity analyses comparing the effect estimates from both random-effects and fixed-effects models. If these were different, we intended to report on both analyses as part of the results section, and consider possible interpretation.

Study design

All included studies were of parallel design. In one of these studies, non-responders from the control group were allowed to cross over after 6 weeks, but we report data up to 6 weeks, thus treating the study as parallel (Wassell et al. 22).

Number of arms

Many studies had more than two arms as they assessed more than one splint type or more than one control type or both. Twenty-four studies had two arms, 19 had three arms, eight had four arms and one had five arms.

Setting

Fifty-one studies were conducted in universities or public hospitals/clinics. The remaining study was carried out at the Mexican Institute for Clinical Research (Tavera et al. 23).

Eleven studies were carried out in Brazil,24–34 10 in Sweden,35–44 seven in the USA,45–51 three in Turkey,52–54 two in India,55,56 two in Egypt,57,58 two in China,59,60 two in Germany,61,62 two in the UK,22,63 two in Italy,64,65 two in Japan,66,67 one in Canada,68 one in the Netherlands,69 one in Mexico,23 one in Poland70 and one in Finland. 71 The remaining two studies72,73 were carried out in both Sweden and Finland.

Forty-seven studies were conducted at a single centre. One study was conducted at 11 general dental practices in the UK (Wassell et al. 22), one study was conducted at two locations in Sweden (Lundh et al. 40), one study was conducted at two locations in the USA (DeVocht et al. 45), one study was conducted at two locations (Sweden and Finland; Nilner et al. 73) and one study was conducted at three locations (two in Sweden and one in Finland; Christidis et al. 72).

Sample size calculation

Ten studies reported sample size calculations that were met, although one of studies was not powered on a relevant outcome (Gomes et al. 32), one sample size was met at 10 weeks but not at 6 and 12 months (Nilsson et al. 43) and one stated only that a sample size calculation had been done and that it had been met (Zhang et al. 60). A further study reported a sample size calculation but it was unclear if it was met (Costa et al. 29). Four studies50,66,71,72 reported sample size calculations that were not met. Three studies reported only post hoc sample size calculations (Giannakopoulos et al. ,62 Michelotti et al. 64 and Sharma49). One study did not perform an a priori sample size calculation as it was a feasibility study, so it was not powered to detect differences between groups (DeVocht et al. 45). In the remaining 33 studies, sample size calculations were not mentioned so it was unclear whether or not they were done.

Funding and conflicts of interest

Twenty-three studies22,24–27,29,31,35,36,38,43,44,46,47,50,53,66,68–73 declared what appeared to be public funding. Five studies39–41,45,63 reported both public and industry funding. One study declared only industry funding (Ficnar et al. 61). Five studies32,52,54,55,67 declared that they received no funding. One study reported the funding source but it was unclear whether this represented public or industry funding (Yu and Qian59). The remaining 17 studies did not mention funding.

Sixteen studies27,29,32,52,54,55,57,61,62,64,65,67,70–73 declared that the authors had no conflicts of interest. However, in one of those studies, one of the authors had designed and patented the splint used in the study (Rampello et al. 65). In a further study (DeVocht et al. 45), one author declared instructing for the manufacturers of one of the interventions. However, that intervention was excluded from the review because it was ineligible. The remaining 35 studies did not mention conflicts of interest.

Number randomised/analysed

The studies randomised 3229 participants to the arms we included in this review (i.e. some trial arms were not eligible or were not used; therefore, those participants are not included in this number). The number of participants included in analyses varied by the time at which the outcomes were assessed, and sometimes it was unclear how many were analysed.

Age and sex

The reported age range of the participants was 10–76 years. In the majority of studies (31 studies), the participants’ mean or median age range was 30–39 years. The vast majority of participants were female.

Diagnosis

Fifty-two studies were included in this evidence synthesis. The majority of studies [47/52 (90%)] focused on people with TMD, with only four studies recruiting people with bruxism (8%). One study evaluated the use of splints in people with bruxism with comorbid TMD.

For the studies evaluating the effectiveness of splints for people with TMD, the diagnostic criteria for TMD varied. However, the predominantly used criteria were the RDC/TMD, used in 26 studies. Two studies used the DC/TMD criteria (Sharma49 and Tatli et al. 54) and an additional five studies used criteria that approximated to the RDC/TMD (either by citing the instrument and/or their description matched a similar process) (Conti et al. ,25 de Felício et al. ,30 Ekberg et al. ,35 Wassell et al. 22 and Wright et al. 51). One study used the AAOP criteria (Alencar and Becker24).

The remaining studies used criteria that we had not prespecified in our protocol (RDC/TMD, DC/TMD or AAOP)17–19 or were undefined/unclear:

• Three had used the Helkimo index74 (Daif,57 Johansson et al. 37 and List et al. 38).

• Two used arthrography (Lundh et al. 41 and Lundh et al. 40).

• One used MRI (Haketa et al. 66).

• One had defined myofascial pain dysfunction syndrome (Rubinoff et al. 48).

• Six used diagnostic systems that it was not possible to classify (Elsharkawy and Ali,58 Leeson,63 Lundh et al. ,39 Magnusson and Syrén,42 Rampello et al. 65 and Zuim et al. 34).

If studies had not clearly used the prespecified criteria (RDC/TMD, DC/TMD or AAOP),17–19 an expert reviewer examined the information available in the paper, alongside any correspondence from authors, to identify the probable subgroup of TMD included in the study. When possible, a ‘probable’ RDC/TMD (sub)group diagnosis was assigned. If a (sub)group diagnosis was not possible, then the sample was regarded as ‘painful TMD’ (Conti et al. ,25 de Felício et al. ,30 Elsharkawy and Ali,58 Johansson et al. ,37 Katyayan et al. ,56 Leeson,63 Lundh et al. 39 and Rampello et al. 65).

Table 1 provides an overview of the number of studies, including participants for each probable RDC/TMD subgroup diagnoses.

All studies that did not use the prespecified diagnostic criteria were excluded from the sensitivity analyses.

The four studies (Gomes et al. ,33 Karakis et al. ,53 Pierce and Gale46 and van der Zaag et al. 69) examining the effects of splints on bruxism all used the Lobbezoo et al. 4 criteria for likelihood of a bruxism diagnosis: ‘possible’ self-report of bruxism, ‘probable’ clinical evidence of bruxism with or without self-report, and ‘definite’ defined by polsomnography. On this basis, one study examined ‘definite’ sleep bruxism and all the other studies examined ‘probable’ sleep bruxism.

The study that examined bruxism with comorbid TMD used the Fonseca index75 for TMD and examined ‘probable’ bruxism (Gomes et al. 32). This study was classified as examining ‘painful TMD’ and excluded from the sensitivity analyses.

Splint versus no splint for temporomandibular disorder

Custom-made splint versus prefabricated splint for temporomandibular disorders

Six studies compared custom-made splints with prefabricated splints for TMD patients:

1. Amin et al. 55 – (1) prefabricated readily available liquid occlusal splint (Aqualizer^®; Bainbridge Island, WA, USA), (2) hard occlusal splint and (3) soft occlusal splint.

2. Christidis et al. 72 – (1) prefabricated occlusal splint (Relax; Unident AB, Falkenberg, Sweden) and (2) stabilisation splint.

3. Ficnar et al. 61 – (1) prefabricated, semi-finished occlusal splint (SOLUBrux) and (2) stabilisation splint.

4. Giannakopoulos et al. 62 – (1) prefabricated oral splint with water-filled elastic pads (Aqualizer) and (2) vacuum-formed splint.

5. Nilner et al. 73 – (1) prefabricated occlusal splint (Relax) and (2) stabilisation splint.

6. Truelove et al. 50 – (1) prefabricated soft thermoplastic athletic mouthguard splint and (2) flat-plane hard splint.

Splint versus control splint for temporomandibular disorders

Ten studies compared control splints that did not alter the occlusion with active splints for TMD patients. In six of the studies, the active splint was described as a stabilisation splint (Dao et al. ,68 Ekberg et al. ,35 Ekberg et al. ,36 Rubinoff et al. ,48 Wassell et al. 22 and Zhang et al. 60). In one study it was described as a flat-plane splint (Raphael and Marbach47) and in another only as an occlusal splint (Nilsson et al. 43). The remaining studies compared two active splints against the control splint:

• Alencar and Becker24 – (1) hard occlusal splint and (2) soft occlusal splint.

• Conti et al. 26 – (1) modified stabilisation splint and (2) conventional stabilisation splint.

Splint versus no splint for bruxism

Three studies compared splints with no splints for bruxism patients:

1. Gomes et al. 32 – Michigan splint + massage versus massage.

2. Gomes et al. 33 – (1) Michigan splint versus no treatment and (2) Michigan splint + massage versus massage (i.e. two pairwise comparisons).

3. Pierce and Gale46 – flat-plane splint versus no treatment.

Custom-made splint versus prefabricated splint for bruxism

One study compared custom-made stabilisation splints with prefabricated splints (Bruxogard™; Myofunctional Research Europe B.V., Waalwijk, the Netherlands) for bruxism patients (Karakis et al. 53).

Splint versus control splint for bruxism

One study compared stabilisation splints with control splints for bruxism patients (van der Zaag et al. 69).

Characteristics of the outcomes

Nine of the 52 studies did not contribute any outcome data to this review, either in the meta-analyses or the data analysis presented in the additional tables (Conti et al. ,25 Conti et al. ,26 Dao et al. ,68 Ficnar et al. ,61 Gomes et al. ,32 Karakis et al. ,53 Pierce and Gale,46 Rampello et al. 65 and Zuim et al. 34).

Primary outcomes

Secondary outcomes

Risk of bias in included studies

A summary of the risk-of-bias assessments for the seven domains is given in Figure 2.

FIGURE 2.

Summary of risk-of-bias assessments for each study. Adapted from Riley et al. 1 This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/. This includes minor additions and formatting changes to the original text.

Adapted from Riley et al. 1 This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/. This includes minor additions and formatting changes to the original text.

Allocation (selection bias)

Blinding of participants and personnel (performance bias)

Forty-eight studies22–42,44–46,49–59,61–73 were rated as having a high risk of performance bias because of the inability to blind patients and personnel to splint/no splint or splint type. Four studies43,47,48,60 were rated as having an unclear risk of bias. These studies all compared splints against control splints, and attempts were made to blind the personnel and/or patients; however, it was not clear if both were blinded.

Blinding of outcome assessment (detection bias)

Forty-one studies22–31,33,34,37–42,44,45,49–52,54–59,61–67,70–73 were rated as having a high risk of detection bias based on the primary outcome of pain. This was because the patients were aware of their assigned group in the studies and would then subjectively rate their own pain.

Six studies were rated as being at a low risk of detection bias. In two of these studies, comparing splints with control splints, the patients were blinded (Dao et al. 68 and Rubinoff et al. 48). Two studies used objective assessment of bruxism while the participants slept (Pierce and Gale46 and van der Zaag et al. 69). Two studies did not assess any outcomes of this review; therefore, detection bias was irrelevant (Gomes et al. 32 and Karakis et al. 53).

The remaining five studies were rated as having an unclear risk of bias. These all compared splints with control splints and it was not clear whether or not the patients were blinded (Ekberg et al. ,35 Ekberg et al. ,36 Nilsson et al. ,43 Raphael and Marbach47 and Zhang et al. 60).

Incomplete outcome data (attrition bias)

Thirty-four studies22,24,26,30,32,34–42,45,47,48,51,52,54,56,57,59–69,71 had limited or no attrition and were rated as being at a low risk of attrition bias. Twelve studies23,27–29,33,43,44,50,58,70,72,73 were rated as being at a high risk of attrition bias because of high attrition rates, substantial differences between groups in attrition rate, or both. The remaining six studies25,31,46,49,53,55 were rated as having an unclear risk of attrition bias owing to poor reporting of numbers randomised or analysed.

Selective reporting (reporting bias)

Twenty-eight studies24,31,34,37,42,43,45,47–52,54–56,59,61–67,69–72 reported outcome data adequately and were assessed as being at a low risk of reporting bias. The remaining 24 studies22,23,25–30,32,33,35,36,38–41,44,46,53,57,58,60,68,73 had problems with the way in which the data were reported and were rated as being at a high risk of reporting bias.

Other bias

For 45 studies,22–24,29–36,39–64,66–73 we did not identify any other potential source of bias and rated them as being at a low risk of bias. Three studies were rated as having a high risk of bias because outcomes were followed up at different times for the two groups (Johansson et al. ,37 List et al. 38 and Rampello et al. 65). For one of those studies, there was also a substantial sex imbalance between groups, potentially indicating that the randomisation process was inadequate or did not work (List et al. 38). The remaining four studies were rated as having an unclear risk of bias because the reporting was poor and we were unable to properly assess them (Conti et al. ,25 Conti et al. ,26 Conti et al. 27 and Conti et al. 28).

Overall risk of bias

Fifty studies22–46,49–73 were rated as having a high risk of bias overall because they received at least one high risk-of-bias rating for the above domains. The remaining two studies47,48 were rated as having an unclear risk of bias because they did not receive any high risk-of-bias ratings for the above domains, but received at least one unclear risk-of-bias rating. Therefore, no study included in this review was considered to be at a low risk of bias.

Studies excluded from the review

Six studies were excluded from the review for the following reasons: not random allocation (Al Quran and Kamal,77 Alpaslan et al. 78 and Gavish et al. 79), inappropriate study design [4-month difference in timing of outcomes between the groups (Madani and Mirmortazavi80)], the counselling group had multiple reinforcement sessions and therefore not considered minimal treatment (Manfredini et al. 81) and we were unable to obtain a full-text copy (Castroflorio et al. 76).

Patients with temporomandibular disorder

One of the main questions posed in this investigation is whether or not there is evidence that splints are effective for reducing pain when compared with no splints. We undertook two analyses. One was for the splint group compared with no/minimal intervention, such as watchful waiting or minimal treatment or self-management. A second analysis was conducted for comparisons with a placebo/control splint, which was used in some trials. There was consensus among clinicians and methodologists that 0–3 months was an appropriate time point to use for the primary analysis of the data. The primary pain outcome was any continuous scale that was sensible to combine (e.g. VAS, NRS, CPI). VAS was the most frequently reported outcome, and 0–3 months was the most frequently reported time point. Other time points, 3–6 months and 6–12 months, were also analysed and reported.

Pain (splint versus no splint/minimal intervention)

Thirteen trials of 16 pairwise comparisons (three of the studies assessed more than one type of splint), all rated as having a high risk of bias, with 1076 patients contributed to the results for the no/minimal interventions at 3 months (Figure 3). There was considerable heterogeneity and the overall SMD was –0.18 (95% CI –0.42 to 0.06). Using a rule of thumb for SMD effect estimates, 0.18 would be considered a small effect15 and, as this was not statistically significant, there is insufficient evidence, which is of very low quality,20 to show that oral splints reduce pain (Table 3). Owing to differences in splint type, the control group no/minimal interventions and different types of TMD diagnoses between the individual studies, we were unable to investigate the heterogeneity any further. There were fewer studies and patients for the other time periods (3–6 months: two trials, 160 patients; and 6–12 months: two trials, three pairwise comparisons, 246 patients), and the effect sizes were SMD –0.31 (95% CI –1.31 to 0.68) and 0.11 (95% CI –0.16 to 0.38) for the 3- to 6-month and 6- to 12-month time periods, respectively, which also fail to demonstrate that oral splints reduced pain (Table 4) (see also Appendix 4, Figures 15 and 16).

FIGURE 3.

Forest plot of comparison: TMD, splint vs. no/minimal treatment; outcome – pain: any combinable scale (higher = more pain), 0–3 months. Risk of bias: A, random sequence generation (selection bias); B, allocation concealment (selection bias); C, blinding of participants and personnel (performance bias); D, blinding of outcome assessment (detection bias); E, incomplete outcome data (attrition bias); F, selective reporting (reporting bias); and G, other bias. a, Current pain intensity 0 to 100 mm VAS (custom anterior repositioning); b, muscle pain 0 to 10 for when (1) waking, (2) chewing, (3) speaking, (4) at rest (score summed = 0 to 40 scale); c, current pain intensity 0 to 10 NRS converted to a 0 to 100 scale (prefabricated splint); d, current pain intensity 0 to 10 NRS converted to a 0 to 100 scale (custom splint); e, current maximum daily pain intensity 0 to 100 mm VAS; f, current pain intensity 0 to 100 mm VAS; g, current pain intensity 0 to 10 cm VAS converted to 0 to 100 mm; h, 0 to 100 mm VAS, recorded three times daily with average calculated on weekly basis (appears to be reported in cm – we converted this to mm); i, current orofacial pain 0 to 10 NRS converted to a 0 to 100 scale; j, current facial pain intensity 0 to 10 cm VAS (we converted this to mm); k, current pain intensity – 0 to 10 cm VAS (we converted this to mm); l, current pain intensity – 0 to 10 cm VAS (we converted to mm); m, CPI 0 to 10 converted to 0 to 100 scale – SD is median value from range of SDs reported in the paper; n, CPI 0 to 10 converted to 0 to 100 scale – SD is median value from range of SDs reported in the paper (custom-made splint vs. control); o, current pain intensity 0 to 10 VAS – we converted to 0 to 100 (splint vs. control); p, current pain intensity 0 to 10 VAS – we converted to 0 to 100 (splint + manipulative and physical therapies vs. manipulative and physical therapies). Adapted from Riley et al. 1 This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/. This includes minor additions and formatting changes to the original text.

Adapted from Riley et al. 1 This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/. This includes minor additions and formatting changes to the original text.

The results for the other pain outcomes are shown as forest plots (see Appendix 4, Figure 17) and summarised in Table 4. There was no convincing evidence that the oral splints reduced pain (apart from a single study rated as having a high risk of bias that showed a statistically significant difference in incidence of 50% reduction in VAS pain, in favour of the control group, between 6 and 12 months), although the quality of the evidence was assessed as being very low.

Pain was also measured and reported in other ways that were not possible to meta-analyse, with mixed and inconclusive results (see Appendix 2, Table 26).

Pain (splints versus control splints)

Three trials (159 patients) were included in the comparison between splints and control splints for the 0- to 3-month time period (Figure 4). The SMD effect size was –0.67 (95% CI –1.16 to –0.17), which indicated a possible benefit for the oral splint compared with a control splint in reducing pain (very low-quality evidence). This result was not confirmed at the two longer-term time points, although the same single study was included in both (Nilsson et al. 43) (see Appendix 4, Figures 21 and 22) (Table 5).

FIGURE 4.

Forest plot of comparison: TMD, splint vs. control splint; outcome – pain: any combinable scale (higher = more pain), 0–3 months. a, Worst pain experienced 0 to 100 mm VAS; b, mean daily pain in the 2 weeks prior to follow-up – 0 to 10 scale (we converted to 0 to 100); c, current pain intensity 0 to 100 mm.

Pain was also measured and reported in other ways that were either not possible to meta-analyse or were not VAS/NRS/CPI, with mixed and inconclusive results (see Appendix 2, Table 28).

Other outcomes (splint versus no intervention/minimal intervention/control splint)

Several other outcomes were measured for these comparisons; these are summarised in Tables 6 and 7. When comparing splints with no/minimal interventions or with control splints, there was no evidence that they reduced TMD clicking or increased mouth-opening at any of the time points measured. There was no evidence that splints improved quality of life at any time point when compared with no/minimal interventions. There was also no evidence of a difference in compliance between the splints and the control splints at any time point. The quality of the evidence for all these other outcomes was assessed as being very low.

Analysis of the robustness of the results (sensitivity analyses)

For TMD patients, we planned to undertake a sensitivity analysis restricted to trials for which the inclusion criteria were based on, or could be clearly mapped to, one of the following sets of diagnostic criteria: RDC/TMD guidelines,17 TMD (DC/TMD) guidelines18 or AAOP guidelines. 19 For the primary analysis of splints versus no/minimal intervention in the 0- to 3-month time period (see Figure 3), there was no difference in the result when removing those trials that did not use the above diagnostic criteria: SMD –0.24 (95% CI –0.52 to 0.04; p = 0.09, I² = 71%; 851 participants) (Figure 5).

FIGURE 5.

Forest plot of comparison: TMD, splint vs. no/minimal treatment; outcome – pain: any combinable scale (higher = more pain); sensitivity analysis of studies using the recommended diagnostic criteria, 0–3 months. Risk of bias: A, random sequence generation (selection bias); B, allocation concealment (selection bias); C, blinding of participants and personnel (performance bias); D, blinding of outcome assessment (detection bias); E, incomplete outcome data (attrition bias); F, selective reporting (reporting bias); and G, other bias. a, Current pain intensity 0 to 100 mm VAS (custom anterior repositioning); b, muscle pain 0 to 10 for when (1) waking, (2) chewing, (3) speaking, (4) at rest, score summed = 0 to 40 scale; c, current pain intensity 0 to 10 NRS converted to a 0 to 100 scale (custom splint); d, current pain intensity 0 to 10 NRS converted to a 0 to 100 scale (prefabricated splint); e, current pain intensity 0 to 100 mm VAS; f, current orofacial pain 0 to 10 NRS converted to a 0 to 100 scale; g, current facial pain intensity 0 to 10 cm VAS (we converted this to mm); h, current pain intensity – 0 to 10 cm VAS (we converted this to mm); i, current pain intensity 0 to 10 cm VAS (we converted this to mm); j, CPI 0 to 10 converted to 0 to 100 scale – SD is median value from range of SDs reported in the paper; k, CPI 0 to 10 converted to 0 to 100 scale – SD is median value from range of SDs reported in the paper (custom-made splint vs. control); l, current pain intensity 0 to 10 VAS – we converted to 0 to 100 (splint + manipulative and physical therapies vs. manipulative and physical therapies); m, current pain intensity 0 to 10 VAS – we converted to 0 to 100 (splint vs. control).

We also carried out a sensitivity analysis restricting the meta-analysis in Figure 3 to studies using stabilisation splints. Again, this did not change the result: SMD 0.04 (95% CI –0.13 to 0.22; p = 0.62, I² = 27%; 750 participants) (Figure 6). This removed much of the heterogeneity seen in the other analyses.

FIGURE 6.

Forest plot of comparison: TMD, splint vs. no/minimal treatment; outcome – pain: any combinable scale (higher = more pain); sensitivity analysis of studies using only stabilisation splints, 0–3 months. Risk of bias: A, random sequence generation (selection bias); B, allocation concealment (selection bias); C, blinding of participants and personnel (performance bias); D, blinding of outcome assessment (detection bias); E, incomplete outcome data (attrition bias); F, selective reporting (reporting bias); and G, other bias. a, Muscle pain 0 to 10 for when (1) waking, (2) chewing, (3) speaking, (4) at rest, score summed = 0 to 40 scale; b, current maximum daily pain intensity 0 to 100 mm VAS; c, current pain intensity 0 to 10 cm VAS converted to 0 to 100 mm; d, 0 to 100 mm VAS, recorded three times daily with average calculated on weekly basis (appears to be in cm – we converted this to mm); e, current orofacial pain 0 to 10 NRS converted to a 0 to 100 scale; f, current facial pain intensity 0 to 10 cm VAS (we converted this to mm); g, current pain intensity 0 to 10 cm VAS (we converted to mm); h, current pain intensity 0 to 10 VAS – we converted to 0 to 100 (splint + manipulative and physical therapies vs. manipulative and physical therapies); i, current pain intensity 0 to 10 VAS – we converted to 0 to 100 (splint vs. control).

We had also planned to test the robustness of the results by performing sensitivity analyses based on excluding studies deemed to be at high and unclear risks of bias from the analyses. This was not possible as all the studies in this comparison were assessed as being at a high risk of bias.

Current pain intensity on visual analogue scale/numerical rating scale

For the purposes of the economic modelling, the main pain results as SMDs needed to be presented as MDs. To do this, we undertook further sensitivity analyses including only studies that measured pain at the time of assessment (current pain), measured on a 0–100 VAS or NRS. Two studies (DeVocht et al. 45 and Michelotti et al. 64) that reported the results as change scores, and therefore were not possible to include in the main SMD analysis, were added to this analysis for the 0- to 3-month time period because they reported current pain intensity on a VAS or NRS. The results were consistent with the main SMD results, as the point estimate represented a very small, clinically unimportant, reduction in pain for splints, with imprecision in the CI that included a benefit for both using splints and not using splints: MD –4.48 (95% CI –11.59 to 2.64; p = 0.22, I² = 94%; 874 participants) (Figure 7). However, the extremely high heterogeneity means that the results should be interpreted with caution.

FIGURE 7.

Forest plot of comparison: TMD, splint vs. no/minimal treatment; outcome – pain: sensitivity analysis of studies reporting current pain intensity on a 0–100 VAS/NRS (higher = more pain), 0–3 months. Risk of bias: A, random sequence generation (selection bias); B, allocation concealment (selection bias); C, blinding of participants and personnel (performance bias); D, blinding of outcome assessment (detection bias); E, incomplete outcome data (attrition bias); F, selective reporting (reporting bias); and G, other bias. a, Change score reported.

For the 3- to 6-month time period, DeVocht et al. 45 was again added to the analysis and the result was again consistent with the SMD analysis: MD –3.43 (95% CI –11.77 to 4.90; p = 0.42, I² = 82%; 202 participants).

For the 6- to 12-month time period, when considering only studies that measured pain at the time of assessment (current pain) measured on a 0–100 VAS or NRS, this reduced the analysis to a single study. There was, again, insufficient evidence of a difference: MD 8.70 (95% CI –4.30 to 21.70; p = 0.19; 78 participants).

Patients with bruxism

An overview of the findings for bruxism is given in Table 8.

Two trials that focused on patients with bruxism provided usable outcome data for the 0- to 3-month time period; however, neither study looked at the primary outcome of tooth wear. The results for the other outcomes are presented in Table 9 (compared with minimal intervention) and Table 10 (compared with control splints). There is some very low-quality evidence that splints, when compared with minimal intervention, reduced pain intensity (see Table 9); however, there was insufficient evidence to determine whether or not the splints led to shorter bruxism times, or fewer episodes than control splints (see Table 10).

Patients with temporomandibular disorder

Pain

Table 11 indicates that three trials (178 patients) were included in the meta-analysis comparing custom-made with prefabricated splints for pain on a combinable scale (0–100) for the 0- to 3-month time period. There was no evidence of any heterogeneity and the pooled SMD was –0.14 (95% CI –0.44 to 0.15) (Figure 8). The evidence was assessed as being of very low quality (Table 12) and there was insufficient evidence to determine whether or not there were any differences between custom-made and prefabricated splints with respect to pain measured on a combinable scale.

TABLE 11 - Oral splints provided for TMD: custom-made vs. prefabricated splints

GCPS, Graded Chronic Pain Scale.

Downgraded as all three studies were rated as having a high risk of bias and a lack of precision in the pooled estimate.

Event rate for custom-made splint group.

Downgraded as a single study was rated as having a high risk of bias, with lack of precision.

Downgraded as two small studies were rated as having high risks of bias, substantial heterogeneity, with lack of precision.

Notes

Oral splints for patients with orofacial signs or symptoms to reduce orofacial pain.

Patient or population: patients provided with oral splints for TMD.

Setting: primary or secondary care.

Intervention: prefabricated oral splint.

Comparison: custom-made oral splint.

Outcomes	Illustrative comparative risks (95% CI)		Relative effect (95% CI)	Number of participants (n studies)	Quality of the evidence (GRADE)	Comments
	Assumed risk	Corresponding risk
	Custom-made splints	Prefabricated splints
Pain SD units: Pain measured on combinable scale 0–3 months	The pain score in the custom-made oral splint group was, on average, 0.14 SDs lower (0.15 higher to 0.44 lower) than that of the prefabricated splint group			178 (3 studies)	⊕⊝⊝⊝ very lowa	Insufficient evidence to determine if either splint type leads to less pain, at this and other time points, and also for pain measured on the GCPS, and pain on palpation As rule of thumb, 0.2 SD represents a small difference, 0.5 a moderate difference and 0.8 a large difference
Clicking of joint at 0–3 months (yes/no)	500b per 1000	500 per 1000 (350 to 720)	RR 1.00 (0.70 to 1.44)	110 (1 study)	⊕⊝⊝⊝ very lowc	Insufficient evidence to determine if either splint type leads to a reduction in joint clicking, at 0–3 months and at 6–12 months
Maximum mouth-opening at 0–3 months (mm)	The mean maximum mouth-opening in the custom-made splint group was 41 mm	The mean maximum mouth-opening in the prefabricated splint group was 4.47 mm higher than for the custom-made splint group (6.13 lower to 15.07 higher)		68 (2 studies)	⊕⊝⊝⊝ very lowd	Insufficient evidence to determine if either splint type leads to an increase in maximum mouth-opening at any time point
Quality of life: SCL-90-R – depression 0–4 (higher = worse) at 0–3 months	The mean quality-of-life score in the custom-made splint group was 0.743	The mean quality-of-life score in the prefabricated splint group was 0.03 higher (0.46 lower to 0.53 higher)		44 (1 study)	⊕⊝⊝⊝ very lowc	Insufficient evidence to determine if either splint type leads to an increase in quality of life at any time point
Quality of life: SCL-90-R – non-specific physical symptoms 0–4 (higher = worse) at 0–3 months	The mean quality-of-life score in the custom-made splint group was 0.685	The mean quality-of-life score in the prefabricated splint group was 0.02 higher (0.46 lower to 0.5 higher)		44 (1 study)	⊕⊝⊝⊝ very lowc	Insufficient evidence to determine if either splint type leads to an increase in quality of life at any time point
Adverse events	Two studies reported that there had been no adverse events. A further study reported on an increased overbite in one patient in the prefabricated splint group, which was treated and not present at 12 months

FIGURE 8.

Custom splint vs. prefabricated splint: pain – any combinable scale (higher = more pain), 0–3 months. Risk of bias: A, random sequence generation (selection bias); B, allocation concealment (selection bias); C, blinding of participants and personnel (performance bias); D, blinding of outcome assessment (detection bias); E, incomplete outcome data (attrition bias); F, selective reporting (reporting bias); and G, other bias. a, Assessed daily in a 1-week pain diary for the week prior to each assessment point using 0 to 100 (pain at rest); b, current pain intensity 0 to 10 NRS converted to a 0 to 100 scale; c, CPI 0 to 10 converted to 0 to 100 scale – SD is median value from range of SDs reported in the paper.

TABLE 12 - Effect estimates for TMD pain: prefabricated splint vs. custom-made splint

GCPS, Graded Chronic Pain Scale; N/A, not applicable.

Outcome	Number of studies (n participants)	Effect estimate (95% CI) (random effects)	p-value for effect estimate	Heterogeneity
				χ2 p-value	I2 (%)
Pain: any combinable scale (higher = more pain)
0–3 months (see Figure 8)	3 (178)	SMD –0.14 (–0.44 to 0.15)	0.35; favours custom splint	0.47	0
3–6 months (see Appendix 4, Figure 36)	1 (37)	SMD 0.71 (–9.12 to 10.55)	0.89; favours prefabricated	N/A	N/A
6–12 months (see Appendix 4, Figure 37)	2 (153)	SMD –0.18 (–0.50 to 0.14)	0.26; favours custom splint	0.43	0
Pain: GCPS (incidence of grade III or IV)
0–3 months (see Appendix 4, Figure 38)	1 (44)	RR 1.64 (0.30 to 8.89)	0.56; favours prefabricated	N/A	N/A
3–6 months (see Appendix 4, Figure 39)	2 (85)	RR 1.48 (0.29 to 7.41)	0.64; favours prefabricated	0.63	0
6–12 months (see Appendix 4, Figure 40)	2 (82)	RR 1.00 (0.03 to 33.30)	1.00	0.09	65

Very low-quality data for pain on a combinable scale at the other time points also failed to determine whether or not there were any differences between custom-made and prefabricated splints (see Appendix 4, Figures 36 and 37).

A summary of all the pain outcome data comparing custom and prefabricated splints is shown in Table 13, with the forest plots shown in Figure 8 and in Appendix 4, Figures 36–40.

TABLE 13 - Effect estimates for TMD outcomes other than pain for custom-made splints vs. prefabricated splints

N/A, not applicable.

Outcome	Number of studies (n participants)	Effect estimate (95% CI) (random effects)	p-value for effect estimate	Heterogeneity
				χ2 p-value	I2 (%)
Change in restricted mouth-opening: maximum mouth-opening (mm)
0–3 months (see Appendix 4, Figure 41)	2 (68)	MD (mm) –4.47 (–15.07 to 6.13)	0.41	0.07	70
3–6 months (see Appendix 4, Figure 42)	1 (37)	MD (mm) –1.00 (–6.74 to 4.74)	0.73	N/A	N/A
6–12 months (see Appendix 4, Figure 43)	1 (33)	MD (mm) –1.00 (–7.82 to 5.82)	0.77	N/A	N/A
Quality of life: SCL-90-R – depression, 0–4 (higher = worse)
0–3 months (see Appendix 4, Figure 44)	1 (44)	MD –0.03 (–0.53 to 0.46)	0.89	N/A	N/A
3–6 months (see Appendix 4, Figure 45)	2 (89)	MD 0.04 (–0.31 to 0.39)	0.83	0.22	35
6–12 months (see Appendix 4, Figure 46)	2 (82)	MD 0.11 (–0.54 to 0.75)	0.75	0.95	0
Quality of life: SCL-90-R – non-specific physical symptoms, 0–4 (higher = worse)
0–3 months (see Appendix 4, Figure 47)	1 (44)	MD –0.02 (–0.50 to 0.46)	0.93	N/A	N/A
3–6 months (see Appendix 4, Figure 48)	2 (89)	MD –0.07 (–0.47 to 0.33)	0.73	0.17	48
6–12 months (see Appendix 4, Figure 49)	2 (82)	MD 0.17 (–0.14 to 0.49)	0.29	0.34	0
Adherence to treatment: use of appliance for several nights per week or more
0–3 months (see Appendix 4, Figure 50)	2 (109)	RR 0.99 (0.91 to 1.08)	0.84	0.38	0
3–6 months (see Appendix 4, Figure 51)	1 (37)	RR 1.02 (0.71 to 1.47)	0.92	N/A	N/A
6–12 months (see Appendix 4,Figure 52)	1 (33)	RR 1.09 (0.65 to 1.82)	0.74	N/A	N/A

Pain was also measured and reported in other ways that were not possible to meta-analyse, with mixed and inconclusive results (see Appendix 2, Table 27).

Other outcomes

Several other outcomes were measured for this comparison; these are summarised in Table 13. When comparing custom-made splints with prefabricated splints, there was no evidence that either improved maximum mouth-opening, quality of life or adherence to treatment at any of the time points measured. Some outcomes were measured for which data were reported that were not possible to meta-analyse in the forest plots; these are reported in Appendix 2, Table 27. There was no evidence of a benefit for either type of splint for any of these additional analyses, and the quality of the evidence was assessed as being very low.

Patients with bruxism

One study including patients with bruxism compared prefabricated splints with custom-made splints, but provided no data for this review (Table 14).

Harms

Harms/adverse events are reported for all comparisons in Appendix 2, Table 29. These were generally poorly reported and minor in nature.

Review methods

This section provides detailed methods used for the search strategy, inclusion criteria and exclusion criteria.

Review results

The number of studies identified from the database searches is provided in Figure 9.

FIGURE 9.

Flow diagram for the identification of studies (cost-effectiveness).

A total of 38 studies were identified from the literature searches. After screening titles and abstracts, 29 studies (76%) were excluded as they did not include an economic evaluation. Full-text versions were obtained for 9 studies (24%),11,84–91 but none were included in the review because (1) they did not conduct a formal economic evaluation (n = 7 studies) or (2) they did not evaluate splints for the treatment of orofacial signs or symptoms (n = 2).

Introduction

As there was no evidence available from the systematic review to inform the cost-effectiveness of splints, or different types of splints for patients with orofacial signs and symptoms with TMD or bruxism, it was decided to develop a de novo decision-analysis model to answer the research question.

The overall project aim was to assess the cost-effectiveness of splints for orofacial signs and symptoms. From the outset, TMD and bruxism were considered as distinct entities that required different economic models to evaluate. However, as highlighted in Chapter 3, there is insufficient evidence regarding the effects of splints to populate a meaningfully structured model for bruxism, and the limited data that do exist cannot readily be translated into meaningful, patient-relevant health states. However, in consultation with clinical expert opinion, we propose an outline structure of a Markov cohort state-transition decision-analysis model that might be used in the future if more data become available. The suggested structure, provided in Appendix 5, might be used to guide the data collection in future research studies that could be used to help inform the cost-effectiveness of splints for treating bruxism.

Given the lack of data, this chapter focuses solely on the model developed to determine the cost-effectiveness of splints for treating TMD. The economic analysis first seeks to determine the cost-effectiveness of all splints compared with none (as defined in Chapter 3) for treating TMD and to determine if sufficient data exist to determine the most cost-effective form of splints by comparing custom-made with prefabricated splints. It is important to note from the outset that the clinical effectiveness evidence base is limited and these uncertainties inevitably translate into the economic model. The cost-effectiveness results should, therefore, be considered as exploratory in nature. The model is, however, informative in determining the key parameters that drive the cost-effectiveness results, and importantly, value-of-information (VOI) analysis is conducted to steer the future research agenda to minimise decision uncertainty.

Model structure

A Markov cohort state-transition model was developed in TreeAge Pro (TreeAge Software Inc., Williamstown, MA, USA) to evaluate the cost-effectiveness of splints in patients with TMD. The comparator was no splints. The model population was the adult population with TMD, with a starting age of 25 years, which is a common age for symptoms of TMD to start, as the 18–35 years age group are significantly more likely to experience first-onset persistent orofacial pain. 92,93 The proportion of the cohort that are male is taken from the Developing Effective and Efficient care pathways in chronic Pain (DEEP) study (19.1% male). 94 Figure 10 outlines the model structure.

FIGURE 10.

State-transition diagram and pain NRS.

The model simulated a cohort through the health states depicted in Figure 10. The health states include pain tertiles, ‘low pain’, ‘moderate pain’ and ‘high pain’, and ‘death’. The health states defined using a NRS of pain intensity (from 0 to 10) in which low-intensity pain was definied as a NRS score of 0–3, moderate pain was defined as a NRS score of 4–6 and high-intensity pain was defined as a NRS score of 7–10.

The preferred instrument to measure the impact of TMD is the Graded Chronic Pain Scale (GCPS), of which the CPI scale is a subcomponent. GCPS is preferable because it incorporates both the pain intensity and disability associated with TMD. 95 However, there was insufficient evidence comparing the use of splints with no splints in relation to the GCPS; therefore, it was not possible to parameterise the model using the preferred measure of treatment effect. In the absence of sufficient clinical effectiveness data to structure a model around GCPS, it was felt that pain tertiles provided the most feasible and practical balance between a meaningful structure to capture the potential impact of treatment on outcomes (pain) that could be populated using clinical effectiveness data, as well as cost and utility data from the DEEP cohort study. The state classification was chosen for practicality, but future research should aim to collect sufficient data to populate a TMD model structured around GCPS.

The economic model has therefore been designed to allow population of the health states (transition probabilities, effect sizes, costs and utilities) using two alternative definitions of pain (CPI and current pain intensity). CPI is a combination of three factors measuring current pain, average pain (in the previous 6 months) and worst pain intensity (in the previous 6 months) using a NRS, with the final score based on an average of the three domains. Current pain intensity asks patients to report their present pain state, on, for example, a NRS or VAS.

The proportion of the cohort entering the model in each of the pain states is determined by the corresponding proportions from the DEEP study cohort. Using a pain definition of ‘current pain’, 34%, 34% and 32% enter the cohort in low, moderate and high states, respectively. For pain defined as CPI, the corresponding proportions are 41%, 27% and 31%, respectively.

At the end of each 3-monthly model cycle, a proportion of the cohort move between pain states. A proportion of the cohort are also assumed to die in the model following age- and sex-adjusted general population all-cause mortality rates. 96 The cycle length defines the fixed period of time, at which point the cohort is introduced to a new set of transition probabilities, costs and utilities. The model estimates the accumulated costs and QALYs using a UK NHS perspective, over an 85-year time horizon in the base case, running for 340 (3-monthly) cycles in the base case up until a maximum of age 110 years to reflect all the costs and outcomes associated with pain states over a lifetime horizon. Half-cycle corrections were applied to the costs and outcomes, to reflect that, on average, events occur in the middle of a cycle rather than at the start or end of a cycle. Costs and QALYs occurring into the future are discounted at a rate of 3.5% per annum, as recommended by the National Institute for Health and Care Excellence (NICE) in England and Wales. 97

Model parameters

The model was populated using best-available data on transition probabilities, costs, utilities and clinical treatment effects. Clinical treatment effects were estimated as MDs using random-effects meta-analysis of studies included in the clinical effectiveness review (see Chapter 3) to obtain MDs between splints and no splints at 3, 6 and 12 months. Longer-term effect size estimates were unavailable, or insufficient for populating the model. The transition probabilities, costs and utilities were sourced from a reanalysis of the DEEP study data. The DEEP study was an observational study in the UK, including 35 dental and medical practices, with a total of 198 patients. 94,98,99 The cohort was followed up over a 2-year period. The reanalysis, performed and provided by the DEEP study chief investigator (Professor Durham) was tailored to provide model parameter estimates (transition probabilities, costs and utilities) applicable for a population of the desired model health states. DEEP study data are the most appropriate source of model parameter estimates, being the largest available cohort, with the longest follow-up and from a UK perspective.

Assessment of cost-effectiveness

A CUA was conducted using QALYs as the measure of benefits. The results presented are over the lifetime of the simulated cohort (with a starting age of 25 years in the base case). The model is fully probabilistic, with model outputs calculated using Monte Carlo simulation with 1000 repetitions, sampling from distributions for transition probabilities, MDs, utilities and costs as described in Tables 16, 17, 19 and 21, respectively. The probabilistic analysis is advantageous because it varies all parameters simultaneously. Model results are reported as expected values of costs and QALYs over the modelled time horizon. An incremental comparison of costs and QALYs between splints and no splints is reported and ICERs are calculated as the MD in costs divided by the MD in QALYs. The ICER is then compared with a commonly used cost per QALY threshold recommended by NICE. 97 If the ICER is within the desired range (£20,000–30,000, or below), an intervention would generally be considered cost-effective. However, in all cases, determination of cost-effectiveness must also consider the variability around the point estimates of incremental costs, incremental QALYs and, hence, the ICER. Consideration of such uncertainty is important to determine if current evidence is sufficient for decision-making, and if decision-makers can be confident that the ICER is likely to fall below commonly accepted threshold values.

As all models are run probabilistically, it is possible to determine the probability of cost-effectiveness at threshold values of willingness to pay (WTP) for a QALY gained (e.g. £20,000 per QALY) for each scenario analysis. Uncertainty in the base-case results is also illustrated using scatterplots of the cost-effectiveness plan and cost-effectiveness acceptability curves (CEACs). Scatterplots and CEACs are particularly informative as they demonstrate the uncertainty arising due to the combined statistical variability in all the models’ parameter inputs. The CEAC shows the probability that splints or no splints are the most efficient use of resources at difference threshold values of society’s WTP for a QALY gain.

A range of scenario analyses were also conducted to account for structural and methodological uncertainty as well as heterogeneity. In economic models, some structural assumptions are made that come with some uncertainty. An example of this would be to vary the time horizon or vary the frequency at which individuals are assumed to replace their splints. There is also uncertainty surrounding the methods that could be used in the model. For example, there are uncertainties regarding the discount rates that should be applied to costs and benefits. The results from the analysis might be subject to heterogeneity. To account for heterogeneity, conducting the analysis in, for example, different age groups would be beneficial. All sensitivity and scenario analyses are listed in Table 22. It should be noted that all analyses, including the base-case and scenario analyses, are reported for different definitions of pain (current and CPI).

Value-of-information analysis

Value-of-information analysis is a useful tool for identifying what contributes to the decision uncertainty in the model. The expected value of perfect information (EVPI) is the difference between the expected value with perfect information and the expected value with current information, and can be used to determine whether or not future research to resolve current decision uncertainty is a worthwhile investment. The EVPI is calculated using the net monetary benefit (NMB). The NMB is calculated from the results of the probabilistic analysis. The intervention with the highest NMB is the most cost-effective intervention. The EVPI is the maximum NMB of each iteration from the Probabilistic Sensitivity Analysis, averaged:109

where λ = threshold, and

The population EVPI is the value of doing further research in the population of interest, namely those who would benefit from the intervention, for example the value of conducting further research on the relative effectiveness of splints compared with no splints in the TMD population in the UK. The expected value of perfect parameter information (EVPPI) estimates which parameters contribute to the decision uncertainty. The EVPPI is conducted by taking an iteration from the Monte Carlo simulation of a parameter in the outer loop and running the model for a defined number of simulations (e.g. 1000 runs), and repeating this exercise for all the parameter iterations from the probabilistic analysis.

To calculate the population EVPI, a number of assumptions were made both regarding the population likely to benefit from the intervention and the lifetime of the health technology evaluated. The population likely to benefit each year was assumed to be 2,311,407 (based on an annual prevalence of examiner-verified TMD of 3.5% obtained from Slade,110 and a UK-wide population of approximately 66 million111). The expected lifetime of the technology was assumed to be 10 years, in line with typical practice for the conduct of a VOI analysis, and the threshold value of WTP for a QALY gained was assumed to be £20,000, in line with typical UK decision-making processes. 97

The VOI analysis was conducted on the Sheffield Accelerated Value of Information (SAVI) application. 112 If the EVPI is positive, further research might be worthwhile. If this is the case, EVPPIs would be calculated to identify the source(s) of uncertainty in the results and the value of future research to resolve decision uncertainty surrounding the most important drivers of cost-effectiveness results.

Base-case results

The base-case analyses used either current pain as the measurement of pain intensity or CPI, because the majority of included RCTs focused on current pain or CPI (see Chapter 3 for further details). Table 23 reports the base-case analysis for both current pain and CPI specifications of the model.

The point estimate of the ICER indicates that splints are not cost-effective using a model structured around current pain. Splints are less likely to be cost-effective using the CPI specification of the model, in which, on average, splints appear to be more costly and less beneficial in terms of QALYs gained. The unfavourable results for splints are driven by the point estimate of the MD for CPI at 6 months, which non-significantly favours the no-splint group. However, these results should be interpreted with caution and must be considered in the light of the very poor quality of the trials used to generate the effect size estimates used in the model. The optimal strategy is highly uncertain and estimates of the ICER should be considered as exploratory in nature.

For all analyses, considering the point estimates of the ICER in isolation may be misleading regarding the true cost-effectiveness of splints using either definition of pain. It may be more appropriate to consider the illustrations of cost-effectiveness provided by scatterplots of the cost-effectiveness plane and CEACs, which more adequately characterise the substantial uncertainty surrounding the results. Figures 11 and 12 illustrate the scatterplot of the cost-effectiveness plane and the CEACs, illustrating the probability that each strategy (splints/no splints) is the most efficient use of resources at alternative threshold values of society’s WTP for a QALY gained.

FIGURE 11.

Cost-effectiveness plane for splints vs. no splints (current pain specification).

FIGURE 12.

Cost-effectiveness acceptability curve for current pain specification.

The scatterplot indicates a high level of uncertainty surrounding the incremental QALYs gained, driven by uncertainty in the effect size for pain MDs identified in the review of clinical effectiveness (see Chapter 3).

The CEAC indicates substantial uncertainty regarding the optimal strategy, with probabilities of the cost-effectiveness of splints never increasing above 60% at WTP threshold values of between £10,000 and £40,000 per QALY.

Figures 13 and 14 report similar data for the CPI specification of the model.

FIGURE 13.

Cost-effectiveness plane for splints vs. no splints (CPI specification).

FIGURE 14.

Cost-effectiveness acceptability curve for CPI specification.

The scatterplot for CPI also indicates a high level of uncertainty surrounding the incremental QALYs gained, again driven by uncertainty in the effect size from poor-quality studies regarding the MDs in CPI between splints and no splints used to populate the model (see Chapter 3). Although the results indicate a lower likelihood of splints being cost-effective using the CPI specification, it should be noted that this is driven by the MD favouring no splints at 6 months. However, as discussed in Chapter 3, Results of the systematic review, this finding should be interpreted with caution and in the light of the poor methodological quality of the included studies in the clinical effectiveness review.

Scenario and sensitivity analyses

The base-case analysis should be interpreted with caution. A range of scenario and sensitivity analyses have also been undertaken, the results of which are presented in Table 24 and serve to illustrate the wider variation in the ICER depending on the assumptions applied in the model. All scenario and sensitivity analyses were conducted probabilistically and report results for both current pain and CPI. These analyses are informative in determining the key assumptions that drive the cost-effectiveness results.

The model results for the current pain configuration were most sensitive to varying structural assumptions about the long-term MD and long-term transition probabilities. Assuming a long-term MD of zero, namely that splints have no additional effectiveness beyond 1 year of usage, generates additional costs to the NHS with mean QALY losses, and thus a very low probability of cost-effectiveness. However, assuming that the MD observed at 6 months continues over a patient’s lifetime, splints would be a highly cost-effective use of resources, with an ICER of just over £2000 per QALY gained. These analyses serve to illustrate the sensitivity of the model to long-term assumptions about differences in pain, and highlight the need for further research to adequately determine the appropriate long-term trajectory of the effectiveness of splints. The CPI-configured model is also sensitive to this assumption, but less so given that the distribution of the MD in CPI at 12 months is centred closer to zero. It should also be noted that extrapolation of MD data to the longer term are based on poor-quality shorter-term MDs (at 3, 6 or 12 months), adding a further layer of uncertainty to the results. It is imperative that high-quality data are obtained for MD in pain related to TMD to generate more robust estimates of cost-effectiveness from the modelling analysis.

The model is also somewhat sensitive to assumptions around the costs of splints and the frequency of splint replacement. There is substantial uncertainty over the actual banded change that might be applied to splints, particularly for prefabricated splints, which could feasibly incur a band 1 or 2 charge, with custom-made splints more likely to incur a band 2 or 3 charge. As anticipated, lower banding improves the cost-effectiveness case for splints. Similarly, there was much uncertainty among the project’s clinical advisors regarding the most probable frequency of splint replacement. This is an important driver of incremental costs in the model, with more frequent replacement generating lower likelihoods of cost-effectiveness. For the current pain configuration, the probability of cost-effectiveness (at £20,000 per QALY) drops from 60% when never replaced to 54% when replaced every 2 years. The CPI results are more sensitive to this assumption, decreasing from 40% to only 17% for the same replacement frequencies.

The scenario indicates that the model results were not particularly sensitive to the discount rate or time horizon chosen.

Value-of-information results

This section reports the results of the VOI analysis. Given the high level of uncertainty described in the previous section, it is important to determine the value of additional research to determine the cost-effectiveness of splints. The EVPI results in Table 25 indicate that future research is worthwhile. The EVPPI results build on this to identify the parameters in the model for which further research is most valuable to reduce uncertainty regarding the optimal treatment decision (splints or no splints).

The VOI analysis further emphasises the decision uncertainty, and identified the key drivers of the cost-effectiveness results. The large positive values of EVPI indicate that further research would be a good investment. EVPPI helps to indicate the parameters in the model that contribute most to decision uncertainty. Large EVPPI values (for current pain in particular) indicate that further research should be prioritised to resolve decision uncertainty about the impact of splints on pain (i.e. the MDs used in the model). This should also encompass an assessment of the long-term effectiveness to more accurately guide extrapolation in the model. In addition, further research regarding the long-run trajectory of disease would be worthwhile, to determine transition probabilities over the full life course. Research on the costs of splints and, in particular, the replacement cost of splints over time is also worthy of further research, although EVPPIs are lower than for the clinical effectiveness data. Although generating positive values of EVPPI, further research with regard to health-state utilities is less valuable, and should be considered only after decision uncertainty in clinical effectiveness and costs of splints has been resolved.

It is noted that parameter EVPPI is positive, but remains consistently lower in models parameterised around CPI than in those parameterised around current pain. This finding must be considered in the light of the other findings in the report, and, in particular, the poor methodological quality of the underlying effect size studies. VOI is a useful tool in prioritising future research, but is driven solely by sampling uncertainty around the input parameters. It does not consider the methodological quality of the clinical effectiveness studies. Details of the economic evaluation results, including VOI results, comparing custom-made and prefabricated splints with no splints can be found online. 113

Quality of the evidence

The quality of the evidence for comparing splints with no splints/minimal interventions/control splints in patients with all subtypes of TMD was downgraded to ‘very low’ owing to the studies being at a high risk of bias, heterogeneity and a lack of precision in the estimates. Most studies were assessed as being at a high risk of bias because of the inability of researchers to blind patients to wearing a splint or not, or wearing different types of splint. As the primary outcome for the TMD patients was pain assessed by the patients, this meant that this outcome measurement was also assessed as being at a high risk of bias. It is difficult to design trials to overcome this problem. We were unable to investigate the heterogeneity of the effect estimates because of the different splint types used, different patient diagnoses and the different minimal interventions being used as control groups.

There were no studies looking at tooth wear, and very few studies and a lack of useable other data for patients with bruxism, so we were unable to determine whether or not splints are effective in these patients. The quality of the evidence was therefore deemed to be very low for the reported outcomes.

The risk of bias of 5022–46,49–73 of the 52 studies was high. Although patient blinding is not possible when comparing oral splints with no splints or a minimal intervention, there were also problems with selective reporting bias and incomplete outcome data. The majority of the studies were assessed as having an unclear risk for selection bias, with researchers not reporting the trial methodology and data according to the Consolidated Standards of Reporting Trials (CONSORT). 121

Patient and public involvement

The project benefited from the establishment of a patient advisory group during the development of the application. At least one member of the patient advisory group attended each of the face-to-face meetings of the research team held in Manchester, and took part in most of the monthly teleconferences. On reflection, involving patients in discussions about the questions and outcomes for the protocol, and in the readability of the Plain English summary was helpful. It is more difficult to involve patients in the more detailed work of the effectiveness review, as this is a specialist methodological exercise with clinical involvement.

Recommendations for future research

Further research is urgently needed to determine whether or not the use of splints is clinically effective, generates meaningful patient benefit and whether or not splints offer an efficient use of scarce NHS resources for both bruxism and TMD. There is a need for well-conducted RCTs involving both TMD and bruxism patients. These trials should compare oral splints with an agreed minimal intervention such as advice/counselling, education or self-performed exercises (applied to both the intervention and control groups). Multiple trials will be required to answer questions about patients with different subtypes of TMD. The selection of patients for inclusion in these studies should, ideally, conform to the DC/TMD diagnostic guidelines to ensure that patients have well-defined conditions and are a homogeneous group. Trials should be conducted in those settings that reflect the current provision of splints provided in the NHS. Triallists should carefully report the data for the patients included in each TMD subgroup separately, to ensure that the data can be pooled for each subgroup in future meta-analyses.

The results of the EVPI analysis indicate that there is substantial decision uncertainty and that future research is worthwhile. The EVPPI analysis identified two important areas of future research to reduce decision uncertainty with regards to cost-effectiveness, as well as two areas of research in which further research is unlikely to reduce decision uncertainty. The priority areas for further research are:

1. The treatment effectiveness of splints and, consequently, that of custom-made versus prefabricated splints. Future economic evaluations should also include provision for extended follow-up to resolve longer-term decision uncertainty in pain trajectory and clinical effectiveness.

2. Further data are required to determine the long-term costs to the NHS of different types of splint provision. This includes generating evidence regarding the appropriate banding for different splint types on the NHS in England, as well as determining the frequency at which splints will require replacement if rolled out to TMD patients.

The VOI analysis indicated that further research to determine the costs or utilities associated with pain states would not be worthwhile, as current evidence obtained from the DEEP study94 is sufficient to aid decision-making.

Cochrane Central Register of Controlled Trials

Date range searched: inception to issue 9 2018.

Date searched: 1 October 2018.

MEDLINE (via OvidSP)

Date range searched: 1946 to 1 October 2018.

Date searched: 1 October 2018.

Cochrane search filter for MEDLINE Ovid

Cochrane Highly Sensitive Search Strategy (CHSSS) for identifying randomized trials in MEDLINE: sensitivity maximising version (2008 revision) as referenced in Chapter 6.4.11.1 and detailed in box 6.4.c of The Cochrane Handbook for Systematic Reviews of Interventions Version 5.1.0 [updated March 2011].

1. randomized controlled trial.pt.

2. controlled clinical trial.pt.

3. randomized.ab.

4. placebo.ab.

5. drug therapy.fs.

6. randomly.ab.

7. trial.ab.

8. groups.ab.

9. or/1-8

10. exp animals/ not humans.sh.

11. 9 not 10

EMBASE

Date range searched: 1980 to 1 October 2018.

Date searched: 1 October 2018.

Cumulative Index to Nursing and Allied Health Literature (via EBSCOhost)

Date range searched: 1937 to 1 October 2018.

Date searched: 1 October 2018.

ProQuest Dissertations & Theses

Date range searched: no restriction.

Date searched: 1 October 2018.

Conference Proceedings Citation Index (via Web of Science)

Date range searched: no restriction.

Date searched: 1 October 2018.

US National Institutes of Health trials registry (ClinicalTrials.gov)

Date range searched: no restriction.

Date searched: 1 October 2018.

World Health Organization International Clinical Trials Registry Platform

Date range searched: no restriction.

Date searched: 1 October 2018.

American Academy of Dental Sleep Medicine website

Date range searched: no restriction.

Date searched: 1 October 2018.

International Association of Dental Research conference abstracts

Date range searched: no restriction.

Date searched: 1 October 2018.

MEDLINE (via OvidSP)

Date range searched: 1946 to 1 October 2018.

Date searched: 1 October 2018.

EMBASE (via OvidSP)

Date range searched: 1980 to 1 October 2018.

Date searched: 1 October 2018.

NHS EED Database (via OvidSP)

Date range searched: inception to issue 1 2016.

Date searched: 1 October 2018.

Cumulative Index to Nursing and Allied Health Literature (via EBSCOhost)

Date range searched: 1937 to 1 October 2018.

Date searched: 1 October 2018.