A systematic review and economic evaluation of subcutaneous and sublingual allergen immunotherapy in adults and children with seasonal allergic rhinitis

Treatment schedule

Owing to the risk of adverse reactions to allergen injection in sensitised patients, conventional SCIT treatment schedules involve a gradual increase in the allergen content of injections, usually involving one or two injections per week over a 3- to 6-month period. 22 Once a prespecified maximum treatment dose has been achieved, or the maximum tolerated dose for any given patient attained, treatment continues with this maintenance dose at regular intervals, usually monthly, for the duration of therapy. 21 Optimal maintenance dosing for a given product is often prespecified by the manufacturer, although substantial evidence suggests that a maintenance dose in the range of 5–20 μg of major allergen per injection is associated with significant clinical improvement. 23 However, the maximum tolerated dose varies between individual patients and may be lower than the target therapeutic dose.

A number of studies have investigated accelerated updosing schedules for SCIT. For example, rush IT involves administering increasing doses of allergen at intervals of between 15 and 60 minutes over a 1- to 3-day period, until the target therapeutic dose is achieved. 22 An alternative form of accelerated schedule is cluster IT, whereby two to three incremental doses are administered on non-consecutive days. Maintenance dose is usually reached at between 4 and 8 weeks. 22

In contrast, treatment schedules for SLIT may or may not include an updosing period, and following initial treatment administration under medical supervision, maintenance dosing is undertaken by the patient in a non-clinical setting. Typically, dosing continues daily for the period of treatment – up to 3 years. However, studies21,24–26 have shown that shorter treatment periods, with SLIT administered for a few months before and during the pollen season, or during the pollen season only, may be as effective as year-round treatment, in terms of symptom and medication reduction and improved QoL.

Optimal treatment schedules for SLIT have yet to be definitely established, and a wide variety of practices are used. 19 Updosing may or may not be necessary, and maintenance schedules ranging from once per day to once per week have been used,19 although daily dosing is the most common.

More recently, rush or cluster regimens for SLIT have been used. A recent meta-analysis27 of individual patient data (IPD) from three open, prospective studies of high-dose SLIT, totalling 1052 adult and paediatric pollen-allergic patients, found no significant difference in rhinoconjunctivitis symptom scores (SSs) or use of rescue medication between perennial or coseasonal schedules, or standard or ultrarush titration. The rate of AEs was also similar between the different treatment schedules. Thus, the major benefit of accelerated IT schedules appears to be in terms of patient convenience. As inconvenience is one of the major reasons for treatment discontinuation,28 accelerated schedules may increase both adherence and therapy uptake. 22

A recent review29 found that accelerated schedules in SCIT may be associated with a higher risk of systemic reactions but also suggested that premedication, for example with antihistamines or corticosteroids, may result in a risk profile similar to that of conventional treatment schedules.

A number of studies25,30,31 have reported that the clinical effects of SIT are additive over time, with increasing benefit following subsequent years of treatment. Based on evidence of sustained clinical benefits after treatment cessation following studies with 3 years of active treatment,17,32 current guidelines recommend this duration of treatment for both SCIT and SLIT. 19 However, there are few double-blind discontinuation studies, and none comparing the long-term effects after different lengths of active treatment for SAR. One prospective controlled study33 evaluated relapse rates following between 12 and 96 months of SCIT treatment in 40 adult and paediatric patients with house dust mite allergy. All patients were symptom free at completion of treatment, but 55% relapsed over the following 3 years. Relapse rate was significantly related to treatment duration, with 62% of those treated for 35 months or less experiencing a recurrence of symptoms, compared with 48% of those treated for > 36 months (p < 0.04).

Specific (allergen) immunotherapy formulations

The immunomodulatory effect of SIT is specific to the allergen used. Although single-allergen IT has proven effective in reducing symptoms on exposure to the specific allergen in polysensitised patients, no additional benefit is obtained in respect of the other allergic triggers. However, there is some indication that SIT may prevent the onset of new sensitisations in monosensitised children,16,34,35 possibly due to cross-reactivity between related allergen species. For example, there is strong reactivity between members of the Festucoideae family of grasses, which includes timothy grass (Phleum pratense L. ), rye grass and orchard grass, and there is extensive cross-reactivity within and between a number of subfamilies of tree pollen. 21

In contrast, mixtures of unrelated allergens have failed to show efficacy in double-blind, placebo-controlled (DBPC) trials of either SCIT or SLIT in multisensitised populations, possibly due to potential interactions between the different enzymatic components and/or dilution of individual allergen dosage. 21,36 For example, extracts from Alternaria species reduce the immunogenicity of timothy grass extract, and studies have shown that extracts of moulds and fungi significantly reduce the potency of grass pollens, some weeds, trees, and a number of perennial allergens when mixed together. 21 Thus, concurrent treatment of multiple sensitivities is not recommended. However, mixtures of related and cross-reacting allergens (e.g. antigens from more than one species of grass pollen) are effective, and a number of commercial products of this type are currently available. 19 It should be noted that multiallergen treatment is commonplace in the USA, where vaccines are formulated for individual use by the treating clinician, but separate vaccines may need to be given for each allergen. 37

A number of modifications in formulation procedures have been made in recent years to improve treatment convenience and/or safety, particularly for SCIT products. The development of depot formulations by adsorption of allergen extract on to depot materials, for example aluminium hydroxide, L-tyrosine or calcium phosphate, is now common practice. 23 This results in prolonged, gradual release of the allergen at the injection site, allowing for a larger maintenance dose to be given at each injection and reducing the number and frequency of maintenance injections required. Another important modification involves chemical modification of the allergen extract with adjuvants such as glutaraldehyde or formaldehyde. 23 The resultant allergoid has reduced specific IgE-binding capacity and therefore lower allergenicity, reducing the risk of treatment-emergent AEs. The reduced allergenicity of allergoid compounds again allows for larger doses to be used, making treatment schedules more convenient. The effect of these modifications on the immunogenicity of the allergoid is unchanged and, hence, clinical efficacy is maintained. Again, this situation differs from that in the USA, where the use of unmodified aqueous allergen extracts is standard practice. 37 More recently, genetically modified allergens or allergen derivatives, or use of allergens conjugated with immunostimulatory molecules has been reported. 12,38

Standardisation

The production of allergen extracts derived from natural allergens can result in highly variable potency of the end product to be used in SIT. Individual manufacturers have therefore developed in-house standardisation procedures for the purpose of quality control and consistency between batches. 23 In addition, European regulations now specify requirements for starting materials, production processes and quality control. 39 Nevertheless, in-house reference standards are based on units of biological activity obtained from immunological assays and/or skin prick tests in a representative population. Thus, sensitivity of the test population, sample size, and the immunological methodologies used may result in differences in potency between products with the same nominal activity. Further, manufacturers use a range of specific units to measure biological response and these are not readily comparable between different commercial products. 23 Given these differences, optimal dosages are product specific and cannot be generalised. Nevertheless, the degree of clinical improvement appears to be dose dependent in both SLIT and SCIT. 40 In injection IT, increased efficacy with higher doses must be balanced with increased risk of systemic reactions. 41 In contrast, a meta-analysis of 25 studies42 in SLIT found that this route of administration did not result in a dose-dependent increase in AEs. These findings were confirmed by a 2011 report from the European Academy of Allergy and Clinical Immunology (EAACI) task force on dose–response relationships in SIT. 40

It has been recommended that manufacturers state the major allergen content (MAC) of their products in mass units (g/ml),43,44 although differences in assay methods may still limit comparability, and the variable contribution of minor allergen content to total biological potency is not accounted for. The use of recombinant allergen products may improve standardisation in the future but these are not yet widely available and few have been tested in large-scale randomised controlled trials (RCTs). 23

Commercial products in the UK

The only aeroallergen SLIT product licensed in the UK for adults and children (5 years) is Grazax® (75,000 SQ-T oral lyophilisate; ALK-Abelló Ltd, Hørsholm, Denmark), a standardised allergen extract of grass pollen from timothy grass. Tablets (one per day) are placed under the tongue and allowed to disperse. Treatment is ideally initiated 4 months before the grass pollen season and continued for a period of 3 years. Where no improvement in symptoms is observed during the first pollen season, there is no indication for continuing the treatment. 45 Grazax costs £66.77 for 30 tablets [source: British National Formulary (BNF) (2012)]. 46

The only SCIT product licensed in the UK is Pollinex® Allergy Therapeutics, Worthing, UK), a standardised L-tyrosine-adsorbed allergoid of grass or tree pollens. Pollinex for grass allergy contains allergen extracts of 12 grass species plus rye, and the tree pollen vaccine contains birch, alder and hazel. Both vaccines may be prescribed to adults and children (≥ 6 years) and are given in six preseasonal injections. 19 An initial treatment set (three vials) and extension course treatment (one vial) of Pollinex costs £450 [source: BNF (2012)46].

A variety of unlicensed products may be prescribed by specialists on an individual ‘named-patient’ basis [see the 2011 British Society for Allergy and Clinical Immunology (BSACI) guidelines for AR19 for an overview].

Non-standard therapies

A number of researchers have investigated highly truncated SIT schedules, including single-injection treatment47,48 and the Rinkel method. 49,50 These are not considered to be standard IT, have proven ineffective in double-blind placebo-controlled studies49,50 and, therefore, have not been included in this review. More recent developments in IT formulations have included the use of peptides fragments of relevant T-cell epitopes of an allergen, as opposed to whole allergens; vaccination with immunostimulatory compounds without a specific allergen attached; and the use of recombinant wild allergens or allergen fragments. 51 Genetically engineered allergens have the potential to reduce allergenicity while maintaining immunogenicity52,53 and are thus a promising avenue for future research. However, these products are generally in the early stages of development and are therefore not currently used in standard practice. These therapies have also not been included in this review.

Patient experience

Lynne Deason developed allergies to different moulds and dog dander in her mid-teens. In her mid-20s she also suffered increasingly with SAR (mainly birch) and allergic reactions to fruit (oral allergy syndrome). In addition, she regularly experienced episodes of anaphylaxis, for which she sought help on several occasions from the accident and emergency (A&E) department; the allergen responsible for these episodes has to date not been identified. No treatment was initiated as a result of visits to the A&E department. As a teenager, LD's parents had paid for her to have private allergy testing and standard medication was recommended (antihistamines and nasal sprays). LD ‘managed’ her anaphylaxis by quickly taking antihistamines whenever signs appeared that an episode was imminent, such as an itching sensation in her ears. More recently, conventional medication provided reasonable relief for SAR, although this still impacted negatively on daily activities, particularly work. Symptoms from both SAR and the oral allergies included puffy eyes, not being able to see very clearly, changes in voice quality, looking like ‘someone had punched me in the face’, feeling ‘groggy’ and an itchy throat; these symptoms made giving presentations at work difficult.

After seeing a nurse at her local GP practice and describing her history of oral allergies, SAR and regular anaphylaxis, LD was referred to hospital and was eventually placed under consultant care for treatment. The nurse expressed disbelief that there had been no earlier referral. Treatment with SCIT was time intensive as it initially involved weekly 2-hour appointments, which then decreased to monthly appointments for approximately 3 years. It also involved a travelling distance of around 25 miles to the hospital. Undergoing treatment was facilitated by having an employer willing to allow time off work. LD did not experience the treatment itself as being particularly unpleasant and felt that professional members of staff who provided good explanations of the procedure were a positive aspect of treatment. LD stated that treatment would be more difficult to incorporate into daily life for parents, as it would involve additional childcare. Additional positive aspects of undergoing SCIT included meeting people with similar experiences at hospital, sharing tips on managing symptoms, feeling less isolated and feeling that people were being empathic. LD also started carrying an EpiPen® (Mylan Speciality L.P., Basking Ridge, NJ, USA) for the first time.

Subcutaneous IT significantly improved LD's SAR symptoms, which became both milder and less frequent. Four years after completing the treatment, significant improvement is still noticeable, with only mild irritation experienced in response to particularly high pollen counts. Medication use also decreased and LD now rarely takes antihistamines. No occurrences of anaphylaxis have arisen since treatment and LD no longer carries an EpiPen. The lessening of symptoms also had a positive knock-on effect on general QoL.

Side effects of the treatment included tiredness (particularly after updosing) and some fairly mild swelling and redness of the arm immediately afterwards. LD noted that other patients had more severe swellings. The treatment has had no effect on the oral allergies. A single allergen was used in the treatment (birch), and LD now knows she is also allergic to almond and hazel. LD felt that there was generally little awareness of severe allergy and treatments with SCIT or SLIT and little empathy for affected individuals. Although, in her own estimation, LD was not among the worst affected, she felt that other people might benefit even more from SCIT or SLIT, for example individuals who are confined indoors during peak allergen times.

Existing evidence for allergen immunotherapy

Searches

Randomised controlled trials of SCIT compared with placebo and SLIT compared with placebo were sought, as were any existing RCTs of head-to-head comparisons (SCIT vs SLIT). A sensitive search strategy, based broadly on those employed in the Cochrane reviews, but with no restriction on routes of IT administration or dates, was used in order to cover both the update and the search for head-to-head trials. There were no language restrictions. Appropriate filters for study design were used where possible. Searches were carried out during April 2011. See Appendix 2 for full details of the search strategies.

The following resources were searched:

• bibliographic databases: MEDLINE (Ovid) 1948–April week 2 2011; EMBASE (Ovid) 1980–April week 15 2011; The Cochrane Library [Cochrane Central Register of Controlled Trials (CENTRAL)] 2011 Issue 1; Cumulative Index to Nursing and Allied Health Literature (CINAHL) (EBSCOhost) 1982–2011; Science Citation Index (Web of Knowledge) 1900–2011. Searches were based on index and text words that encompassed the population and intervention (e.g. ‘seasonal allergic rhinitis’, ‘immunotherapy’)

• ClinicalTrials.gov, UK Clinical Research Network Portfolio Database and metaRegister of Current Controlled Trials (mRCT) (http://controlled-trials.com) were searched for ongoing studies, as well as the lists of ongoing trials identified in the Cochrane reviews

• references lists of relevant reviews and included studies

• consultation with clinical advisors

• selected websites.

Study selection

The following inclusion and exclusion criteria were used (Table 4). These are broadly consistent with those listed in the Cochrane reviews.

The brief specified adults and children (examined separately) with severe hay fever, which does not respond to conventional treatment. We anticipated that not all trials would provide this information, or that they would use different classifications for ‘severe’ or ‘not responding to conventional treatment’. We therefore did not restrict inclusion by severity. We did restrict inclusion to treatment-naive patients; where this was not explicitly stated we noted this but included the study.

Titles and abstracts of retrieved studies underwent an initial screen by one reviewer, and studies that were clearly not relevant were excluded. The remaining studies were independently screened for inclusion by two reviewers. Where it was unclear whether or not studies met the inclusion criteria on the basis of title and abstract, full copies were obtained for assessment. Any discrepancy between reviewers was resolved through discussion or referral to a third reviewer. Reference Manager software, version 11 (Thompson ResearchSoft, San Francisco, CA, USA) was used to track and record study selection decisions and reasons for exclusion. Foreign-language papers were translated, where necessary, by the authors or colleagues.

Assessment of trial validity

Cochrane collaboration guidelines were followed for risk of bias assessment. 135 The following criteria were considered: adequate sequence generation, concealment of allocation, blinding of patients and personnel, completeness of outcome data, selective reporting of outcomes and IT treatment history. This last point is relevant as it is known that SIT can have long-lasting effects. Blinding of outcome assessors was not assessed, as outcomes were largely patient reported. Each item was classified as having a low risk of bias, high risk of bias or unclear risk of bias. Assessment of trial validity was at study level rather than outcome level.

Data extraction

Data extraction, including quality assessment, was conducted by one reviewer and checked by another using a standard, piloted, extraction form. There were no discrepancies that could not be resolved through discussion. Data were extracted on main study characteristics, main patient characteristics, study quality and all included outcomes. Data previously reported in the Cochrane reviews and included in this report were not checked.

Analysis

Where possible, we updated the meta-analyses, including subgroup analyses, from the existing Cochrane reviews. Meta-analysis assumes similarity between trials, and this was explored for both population and study characteristics in the newer studies identified. Studies included in meta-analyses in the Cochrane reviews were assumed to satisfy the similarity criteria.

Meta-analysis was limited to the following outcomes that were consistently reported across a high proportion of trials, and where data suitable for use in meta-analysis were provided or calculable: SSs, MSs, combined SMS and QoL scores. Where not reported, standard deviations (SDs) were calculated from other appropriate measures of variance (e.g. standard error, 95% CIs) according to Cochrane guidelines. 136 Results not suitable for meta-analyses were tabulated and described.

All meta-analyses were undertaken in RevMan software version 5.1 (The Cochrane Collaboration, The Nordic Cochrane Centre, Copenhagen, Denmark) using a random-effects model. SMDs were presented, as there was little overlap between outcome measures across trials. So, for example, although AR symptoms were frequently measured, the number of symptoms measured was different across trials, as was the maximum score that could be achieved. As a rule of thumb, SMDs (effect sizes) can be described as small (< 0.4), moderate (0.4–0.7) or large (> 0.7). 137

In studies with more than one active intervention arm receiving different dosage IT, data for the group receiving the highest dose were used in meta-analysis, consistent with the Cochrane reviews. Where data were reported at different time points, data for the longest treatment duration were used. Outcome values were sometimes reported both for peak pollen season (PPS) and averaged over an EPS. Consistent with the Cochrane review, the mean values for the EPS were used when available.

Indirect comparison meta-analysis (ICMA) and indirect comparison meta-regression (ICMR) were used to compare the efficacy of SCIT and SLIT for each of the four outcomes (SSs, MSs, combined SMSs and QoL). Similarity of trial and population characteristics within (1) SCIT compared with placebo and (2) SLIT compared with placebo trials was assessed qualitatively and statistically (to test the assumption of homogeneity138); similarity between population and trial characteristics between SCIT compared with placebo and SLIT compared with placebo trials was also assessed (to test the assumption of similarity138). Possible sources of heterogeneity were explored and adjusted for using meta-regression, with a number of trial, population and reporting characteristics being used as covariates; specifically, these comprised participant age (adult/child), treatment duration, MAC of IT product and type of allergen (covariates as prespecified in subgroup analyses in the Cochrane review). We also conducted a post hoc exploration using year of publication and number of separate symptoms for which outcome data had been obtained as covariates. Improvement in model fit was expressed using the deviance information criterion (DIC), a compound measure of model fit and complexity, and extent of residual variation was monitored. As scores were measured on different scales, standardised score differences were calculated, except for RQLQ scores. Analyses were conducted in WinBUGS 1.4 (MRC Biostatistics Unit, Cambridge, UK). 139 Full methodological details for the indirect comparison can be found in Appendix 3.

Quantity of evidence (subcutaneous immunotherapy versus placebo studies)

Searches identified 84 publications of DBPC RCTs of SCIT (see PRISMA flow diagram in Appendix 4). Of these, 48 were included in the relevant Cochrane review. 68 Of the remaining 36 publications, 10 pre-dated the Cochrane searches but were not included in that review, and a further four had been excluded by the Cochrane review, despite appearing to meet the inclusion criteria. Details of excluded studies are presented in Appendix 5. The remaining 22 publications,30,140–160 reporting on 18 distinct RCTs, post-dated the final search date listed in the Cochrane review (February 2006). Two151,156 of these publications provided additional data relating to the trial reported in Frew et al. in 2006,161 which was included in the Cochrane review. Thus, 20 publications30,140–150,152–155,157–160 relating to 17 new RCTs were identified. As the purpose of this report was to update, rather than repeat, the Cochrane review, results are presented only for those 17 trials initially published from 2006 onwards. However, all relevant studies have been included in the meta-analyses.

Main study characteristics and risk of bias (subcutaneous immunotherapy versus placebo studies)

The main study and population characteristics, and assessment of risk of bias are detailed for each of the 17 newly identified RCTs30,142–146,148–150,152–155,157–160 in Appendix 5. All studies were DBPC RCTs. Approximately half of the studies (nine trials30,142–144,146,148,152,158,160) involved fewer than 65 participants, but the remaining trials included at least 100 patients, and the largest145 represented over 1000 participants. Skin prick tests were performed in all patients to demonstrate specific sensitivity. Allergy symptoms were described as moderate to severe in 2 out of 17 trials,145,146 whereas level of severity was not stated in 15 out of 17. 30,142–144,148–150,152–155,157–160 Ten studies30,142,143,148–150,152,154,155,160 stated that some patients also had asthma symptoms. Patients with previous IT were excluded in five studies,30,142,152,154,160 four trials143,145,157,159 allowed patients who had not received SIT in the last 3–5 years or ‘recently’, and previous treatment status was unclear in eight trials. 144,146,148–150,153,155,157 Outcomes included SSs, MSs, combined symptom and medication scores, QoL and AEs (note: only outcomes consistent with the inclusion criteria for this report have been listed). Approximately half of the studies (930,145,148,149,152,153,155,157,160 out of 1713,30,142–146,148–150,152–155,157,159,160) reported a combined score, either alone or in addition to individual SSs and MSs. Although SSs were consistently reported across trials, the number and types of symptom assessed varied widely (see Appendix 7 – SSs across studies). Similarly, there were differences in how SMSs were calculated.

Tree and mixed-grass allergens were the most commonly investigated (eight30,142,148,149,154,155,158,160 and three trials,145,150,157 respectively), followed by two trials each of timothy grass (P. pratense)146,153 and ragweed,144,159 one trial in Alternaria142 and one in Russian thistle143 (Salsola kali L. ). The main study characteristics are summarised in Table 5.

Similarity between trials was explored for a range of study and population characteristics. Trial duration and type and amount of allergen used varied between trials; however, this was explored as a source of heterogeneity in subgroup analyses. Where reported, inclusion criteria were very similar across trials, with the majority stipulating a minimum of 2 years' clinical history of moderate to severe SAR, incompletely controlled by standard medication, and no prior experience of SIT. Rates of asthma across trials were largely consistent, comprising between one-quarter and one-third of participants. Prevalence of asthma was higher in one paediatric trial122 (38 out of 50 patients). All studies excluded patients with severe or perennial asthma.

Results of the risk of bias assessment are summarised in Table 6. Full details are given in Appendix 5.

The risk of bias was low in three or more areas of potential bias for 1230,142–146,148,152,154,158,160 out of 17 studies. A lack of detail in the published reports, most notably for those reported in abstract form only, meant that the risk of bias was unclear in many cases. Lack of detail related most often to allocation concealment, sequence generation and whether or not patients were treatment naive. In all studies, apart from those reported as abstracts, there were details on blinding. Study authors were not contacted and an ‘unclear’ rating may be due to a lack of reporting rather than a reflection of poor trial quality. There were only two instances for which a high risk of bias was identified, in the area of data completeness. Only one114 of these studies contributed to any meta-analyses (QoL data); owing to its small sample size (n = 25) it is unlikely to have a large influence on overall results. It should be noted that a certain degree of subjectivity remains in assigning a rating of low/high/unclear risk of bias.

Effectiveness of subcutaneous immunotherapy compared with placebo

Of the 17 newly identified RCTs, only five142–144,152,154 reported data in a form suitable for meta-analyses. Of these, three studies provided new SS and MS data,142,152,154 and two provided QoL data143,144 (Table 7).

Symptom scores

Of the 15 studies91,162–174 included in the Cochrane review meta-analysis, one162 was excluded from this study as patients were not treatment naive. Searches for this report identified 10 new studies142–144,148,152,154,155,157,158,160 reporting this outcome, of which only three142,152,154 provided data suitable for inclusion in meta-analysis. Thus, a total of 17 studies, comprising 659 active and 525 placebo patients, were included (Figure 1).

Figure 1.

Subcutaneous immunotherapy vs placebo: SSs.

The combined SMD was −0.65 (95% CI −0.85 to −0.45; p < 0.00001) favouring SCIT, with evidence of moderate heterogeneity, similar to the findings of the Cochrane review68 (SMD −0.73, 95% CI −0.97 to −0.50; p < 0.00001, based on 15 trials).

Subgroup analyses were conducted by age, study duration, MAC and type of allergen. All favoured the active treatment and were statistically significant. Results of subgroup analyses are shown in Table 8. Forest plots of all subgroup analyses are shown in Appendix 7. Note that the Cochrane review of SCIT did not include subgroup analyses, but that studies identified in that review are included in subgroup analyses here.

TABLE 8 - Subcutaneous immunotherapy vs placebo, subgroup analyses: SSs

IV, inverse variance.

Subgroup	No. of studies	Total n	SMD (IV, random 95% CI)
Age
Adults	16	1140	−0.68 (−0.89 to −0.47)
Duration (months)
< 6	5	274	−1.29 (−2.10 to −0.49)
6–12	2	309	−0.54 (−0.78 to −0.29)
> 12	8	442	−0.51 (−0.70 to −0.32)
MAC
< 5 μg	3	228	−0.43 (−0.69 to −0.16)
5–20 μg	5	231	−0.54 (−0.80 to −0.27)
> 20 μg	3	341	−1.06 (−2.08 to −0.05)
Allergen
Grass	9	552	−0.64 (−0.91 to −0.37)
Parietaria	3	353	−1.15 (−2.09 to −0.21)
Tree	4	235	−0.46 (−0.72 to −0.20)

Sixteen142,154,161,163–174 of the 17 studies were conducted in an adult population and the results did not differ from those of the entire sample. Only one study152 involved a paediatric population. In this study,152 rhinitis, conjunctivitis and asthma SSs were all significantly lower following 3 years of active treatment than with placebo.

Analysis by treatment duration found that studies of ≥ 6 months in duration resulted in similar effect sizes to the sample as a whole. All three142,152,154 of the more recent trials lasted for > 12 months. Shorter studies gave a larger effect size; this was associated with a high degree of between-study heterogeneity.

In line with current guidelines, all three of the newer studies142,152,154 used vaccines with between 5 and 20 μg MAC. Subgroup analyses (see Appendix 7, Figures 31–95) suggest that effectiveness increases with increasing MAC; this finding should be interpreted cautiously, as it is not based on a randomised comparison.

Immunotherapy with grass allergens made up the largest subgroup (47% total sample). Combined effect size was similar to that of the entire sample, with a moderate degree of heterogeneity. Similar effect sizes were found for tree pollen allergy. Only three studies involved Parietaria pollen. 161,165,167 Combined effect size was quite large, but was associated with wide CIs and a high degree of heterogeneity. None of the studies conducted in ragweed that were identified in the Cochrane review were suitable for meta-analysis, and no new studies were found. Only one study152 was conducted in Alternaria.

Details of SS data from the seven studies not included in the meta-analysis143,144,148,155,157,158,160 are presented in Table 9. All favoured the active treatment over placebo.

TABLE 9 - Subcutaneous immunotherapy SSs: studies not in meta-analysis

(A), abstract; AIC, Amb a 1-immunostimulatory oligodeoxyribonucleotide conjugate; IQR, interquartile range; TSS, total symptom score.

Study ID	Results
Ceuppens 2009160	Median nasal SSs (IQR) 0.3 (0.1–0.6) vs 0.7 (0.5–1.1) in the active and placebo groups, respectively (p = 0.041)
Colas 2006143	Over the EPS, median (IQR) TSSs were 4.3 (3.4–4.6) and 6.4 (4.0–8.4) in the IT and placebo groups, respectively (p < 0.001)
Creticos 2006144	During the first ragweed season post treatment, mean rhinitis visual analogue score was 13.8 in the AIC-treated group, vs 35.1 in the placebo group (p = 0.01). Daily nasal SSs were also lower in the active group (p = 0.03)
Hoiby 2010148	Median SS was not significantly different between treatment groups (p = 0.375), despite a reduction of 25% in the SCIT group compared with placebo
Pfaar 2010155	Median SSs were 0.54 (IQR 0.27–0.77) for Depigoid-treated patients completing the study and 0.61 (IQR 0.48–0.75) for placebo [median difference −0.1 (95% CI −0.20 to −0.02); p < 0.01]
Sahin 2011157 (A)	Compared with placebo, the overall SS in the active group was significantly reduced by 36% (p = 0.006)
Ventura 2009158	Not possible to extract data from graphs. Clinical improvements were noted with active treatment compared with placebo

Medication scores

Medication scores suitable for meta-analysis were available in 16 studies (13 from the Cochrane review,91,161,163–166,169–173,17,176 three142,152,154 more recent), which together included 621 active and 483 placebo patients. The combined SMD was −0.55 (95% CI −0.75 to −0.34; p < 0.00001); this was very similar to that reported in the Cochrane review (SMD −0.57, 95% CI −0.82 to −0.33; p < 0.00001). Heterogeneity was reduced slightly, but remained statistically significant (Figure 2).

FIGURE 2.

Subcutaneous immunotherapy vs placebo: MSs.

Subgroup analyses were conducted by age, study duration, MAC and type of allergen. All favoured the active treatment and, with the exception of studies using a < 5 μg MAC, were statistically significant. However, this non-significant result was based on only two trials163,165 totalling 147 patients. Results of the subgroup analyses are shown in Table 10.

TABLE 10 - Subcutaneous immunotherapy vs placebo, subgroup analyses: MSs

IV, inverse variance.

Subgroup	No. of studies	Total n	SMD (IV, random 95% CI)
Age
Adults	15	1059	−0.53 (−0.75 to −0.32)
Duration (months)
< 6	2	143	−0.34 (−0.68 to −0.01)
6–12	3	332	−0.46 (−0.69 to −0.23)
> 12	9	469	−0.67 (−1.06 to −0.29)
MAC
< 5 μg	2	147	−0.31 (−0.63 to 0.02)
5–20 μg	6	254	−0.45 (−0.70 to −0.20)
> 20 μg	2	305	−0.55 (−0.96 to −0.13)
Allergen
Grass	8	483	−0.77 (−1.22 to −0.33)
Parietaria	2	318	−0.43 (−0.67 to −0.20)
Tree	4	235	−0.34 (−0.60 to −0.09)

Fifteen91,126,142,154,161,163–166,169–173,176 of the 16 trials91,126,142,152,154,161,163–166,169–173,176 reporting MSs were conducted in adults, and the results did not differ greatly from the sample as a whole. Only one study152 was conducted in children, and also reported that active treatment resulted in statistically improved MSs compared with placebo.

Nine126,142,152,154,164–16,171,173 of the 14 studies126,142,152,154,161,163–166,169–171,173,176 in the meta-analysis for which treatment duration could be determined lasted for > 12 months. All subgroups favoured active treatment and were statistically significant, although effect size appeared to increase with treatment duration.

Six142,152,154,166,169,176 of the 10 included studies,142,152,154,161,163,165,166,169,173,176 including all three of the recent trials, utilised vaccines with between 5 and 20 μg MAC. Again, combined SMD increased with increasing dosage. Improvements in MS in the lowest dosage group were not significantly better than placebo, but this finding was based on only two studies163,165 (total n = 147).

The most commonly investigated allergen was grass pollen, representing 483 subjects across eight trials,91,126,164,166,170–173 and this was associated with the largest effect size, but also with a high degree of between-study heterogeneity. Tree pollen allergy was studied in four trials,142,154,163,169 and Parietaria (a plant of the nettle family) in two trials. 161,165 Combined SMD favoured active treatment and were statistically significant in all allergen subgroups. Only one study152 was performed in Alternaria (a fungus) and thus meta-analysis was not possible. No studies in ragweed reported MSs suitable for meta-analysis.

Medication score results from five recent studies143,148,155,157,160 were not suitable for inclusion in meta-analysis, and details are shown in Table 11. Two studies reported quite large reductions in MSs in actively treated patients (43%157 and 52%148); however, three studies143,155,160 reported no significant difference between the groups. These contrasting results cannot be explained by differences in sample size, treatment duration, MAC or type of allergen.

TABLE 11 - Subcutaneous immunotherapy MSs: studies not in meta-analysis

(A), abstract; IQR, interquartile range.

Study ID	Results
Ceuppens 2009160	No significant differences between groups (p = 0.155)
Colas 2006143	Over the EPS, median (IQR) MSs were 0.8 (0.7–0.8) and 0.9 (0.5–1.1) in the IT and placebo groups, respectively (p = 0.115)
Hoiby 2010148	Median MS were significantly different between treatment groups after 18 months: median (IQR) 2.1 (0.6–3.7) in the SCIT group and 4.4 (1.9–14.0) in the placebo group (p = 0.016), a reduction of 51.9% in the SCIT group compared with placebo
Pfaar 2010155	Median MSs were 1.36 (IQR 0.4–2.9) for Depigoid-treated patients completing the study and 2.95 (IQR 1.7–3.9) for placebo [median difference −1.3 (95% CI −1.87 to −0.34); p = 0.09]
Sahin 2011157 (A)	Compared with placebo, the overall MS in the verum group was significantly reduced by 43% (p = 0.002)

Symptom and medication scores

Only the eight studies91,163–168 (total n = 617) previously reported in the Cochrane review reported this outcome in a manner suitable for meta-analysis. Thus, the combined effect size calculated in that review remains valid (SMD −0.48; 95% CI −0.67 to −0.29; p < 0.00001) (Figure 3). 68 However, it was possible to conduct a number of subgroup analyses on this sample.

FIGURE 3.

Subcutaneous immunotherapy vs placebo: SMSs.

All of the eight included studies91,163–169 were conducted in adults. Thus, subgroup analyses were conducted for study duration, MAC and type of allergen. All favoured the active treatment and were statistically significant. Results of subgroup analyses are shown in Table 12.

TABLE 12 - Subcutaneous immunotherapy vs placebo, subgroup analyses: SMSs

IV, inverse variance.

Subgroup	No. of studies	Total n	SMD (IV, random, 95% CI)
Duration (months)
< 6	3	221	−0.47 (−0.91 to −0.02)
> 12	3	253	−0.51 (−0.76 to −0.25)
MAC
< 5 μg	3	228	−0.39 (−0.70 to −0.07)
Allergen
Grass	5	435	−0.43 (−0.62 to −0.24)
Parietaria	2	77	−0.96 (−1.44 to −0.49)

Treatment duration could not be determined for two studies. 91,177 Three of the included studies164–166 were of < 6 months' duration. Combined effect size was similar to that in the overall sample, although significantly more heterogeneity was indicated in this subgroup (I² = 59% vs 22%). A further three studies163,167,168 lasted over 12 months, and effect sizes were again similar. However, these studies163,167,168 were more homogeneous (I² = 0%). None of the studies reporting this outcome lasted between 6 and 12 months in length.

Major allergen content could not be determined for three studies. 91,164,177 Three studies163,165,168 utilised a dose of < 5 μg major allergen. Effect size was a little smaller than for the sample as a whole but remained significant. Only one study each used a vaccine with 5–20 μg166 and > 20 μg167 major allergen, and thus meta-analysis was not possible in these subgroups.

Again, the largest subgroup of studies was for grass pollen: five studies,91,164,166,168,177 (total n = 435) resulted in a moderate effect size in favour of active treatment. Two studies165,167 were conducted with Parietaria allergen, with the combined effect size strongly favouring the active treatment. Only one study164 was conducted with tree allergen and meta-analysis was therefore not possible.

None of the nine newer studies30,145,148,149,152,153,155,157,160 reporting combined SMSs was suitable for meta-analysis. Results from all nine studies30,145,148,149,152,153,155,157,160 favoured active SCIT; details of the data from these studies are shown in Table 13. Studies that involved >1 year of treatment reported that the effect size increased with each year of active treatment.

TABLE 13 - Subcutaneous immunotherapy SMSs: studies not in meta-analysis

(A), abstract; ANOVA, analysis of variance; IQR, interquartile range; ITT, intention to treat; MATA MPL, modified allergen tyrosine adsorbate monophosphoryl lipid A.

Study ID	Results
Chakraborty 200630	The SIT group had a 33.5% (p < 0.01) and 57% (p < 0.001) decrease in the SMS during the first and second treatment seasons of 2000 and 2001 when compared with the baseline peak month. There were no significant changes in the control group
Ceuppens 2009160	Clinical index score (a combined score of symptoms and medication use) was reduced in the verum group by 24% compared with placebo [median (IQR) 0.5721 (0.2759–1.2830) vs 1.0322 (0.5882–1.6080), respectively; p = 0.055]
DuBuske 2011145	Median (SD) combined SMS during the four peak weeks of the grass pollen season was 6.00 (5.57) in the Grass MATA MPL-treated subjects and 7.06 (5.57) in the placebo-treated subjects. In the ITT analysis (n = 1028), the least squares mean combined SMS was reduced by 13.4% with Grass MATA MPL compared with placebo (p = 0.0038). Similar differences were seen during the EPS
Hoiby 2010148	After 18 months, the median (IQR) combined SMS of the SCIT and placebo groups were 8.0 (5.8–10.3) and 12.6 (8.6–16.2), respectively (p = 0.004) – a 36.5% reduction in the SCIT group compared with the placebo
Kettner 2007149 (A)	After 1.5 years of therapy, a highly significant and clinically relevant reduction in the median AUC of the SMS from 389.6 to 207.8 in the active group compared with the placebo group (from 382.5 to 306.5; p = 0.0137) was observed in the full analysis set
Kuna 2011152	Reductions in combined SMS after therapy, compared with placebo, were 10.8%, 38.7%, and 63.5% after the first, second, and third years of SIT, respectively (p = 0.73, 0.102, < 0.001 and < 0.001 for baseline, 1, 2 and 3 years of SIT, respectively, active therapy vs placebo, one-way ANOVA test)
Ljorring 2009153 (A)	The estimated treatment effect on combined SMS over peak season was −4.45 (95% CI −6.84 to −2.06) in favour of active treatment. (Abstract; patient numbers in each group not reported)
Pfaar 2010155	At 18 months, the median AUC for the combined SMS was 2.3 (IQR 0.9−3.2) for the actively treated patients and 2.6 (IQR 2.2–4.4) for placebo-treated patients [median difference −0.4 (95% CI −1.22 to −0.03); p < 0.04] for the ITT population
Sahin 2011157 (A)	There was a significant reduction in total combined scores in the active group compared with the placebo group (p = 0.005)

Quality of life

Quality-of-life data suitable for meta-analysis were available for eight RCTs (five from the Cochrane review,161,164–166,173 three more recent;143,144,146 Figure 4). The addition of the three newer studies resulted in a nearly 70% increase in sample size for this outcome (total n = 955). Nevertheless, the results (SMD −0.53, 95% CI −0.66 to −0.39; p <0.00001) were almost identical to those reported in the Cochrane review (SMD −0.52, 95% CI −0.69 to −0.34; p <0.00001). No heterogeneity between studies was found.

FIGURE 4.

Subcutaneous immunotherapy vs placebo: QoL.

Where subgroup analyses were possible, treatment duration, MAC and type of allergen did not appear to affect the outcome (Table 14). All of the studies included in the meta-analysis were conducted in adults and thus analysis by participant age was not possible. However, based on results from one paediatric study,152 SCIT appears to be effective for the improvement of QoL in children and adolescents when used long term.

TABLE 14 - Subcutaneous immunotherapy vs placebo, subgroup analyses: QoL

IV, inverse variance.

Subgroup	No. of studies	Total n	SMD (IV, random, 95% CI)
Duration (months)
> 12	6	662	−0.47 (−0.63 to −0.31)
MAC (μg)
5–20	3	381	−0.52 (−0.74 to −0.31)
> 20	3	379	−0.65 (−0.87 to −0.43)
Allergen
Grass	4	561	−0.44 (−0.61 to −0.26)
Parietaria	2	316	−0.63 (−0.87 to −0.39)

As all eight of the included studies143,144,156,161,164–166,173 assessed QoL using the Juniper RQLQ, an additional meta-analysis was conducted to calculate weighted mean difference (MD) (Figure 5). Active treatment had a significant positive effect on QoL, equivalent to a 0.74 unit reduction in RQLQ score compared with placebo.

FIGURE 5.

Subcutaneous immunotherapy vs placebo: QoL (RQLQ scores).

Three new studies142,148,152 reported QoL data in a manner not suitable for inclusion in meta-analysis. Details of data from these studies142,148,152 are shown in Table 15. The smallest142 of the three studies (total n = 28) found no difference in overall QoL between treatment groups using SF–36. The other two studies148,152 used disease-specific instruments (age-appropriate versions of the RQLQ) and both reported clinically significant improvements in QoL in active- but not placebo-treated subjects. However, in the 3-year paediatric study,152 improvements became statistically significant only in later years of treatment.

TABLE 15 - Subcutaneous immunotherapy QoL scores: studies not in meta-analysis

Study ID	Results
Charpin 2007142	Assessment measure was SF–36. There was no significant difference between the two treatment groups except for variations in social function after the first season
Hoiby 2010148	At baseline, there was a statistically significant difference between the two treatment groups in the mean RQLQ (SCIT: 2.58; placebo: 2.7; p < 0.0001). During the study, RQLQ improved in both groups. The change in mean RQLQ between baseline and assessment during pollen season 2005 after 6 months of treatment was significantly different between the two groups (SCIT −3.8; placebo −0.1; p < 0.0001)
Kuna 2011152	Active treatment was associated with an improvement in QoL for children (up to 12 years of age) with rhinoconjunctivitis. The mean baseline RQLQ score was 1.7 and decreased significantly in consecutive years of therapy to 1.4, 1.0, and 0.7 (p = 0.009, 0.008 and 0.003 after the first, second and third years of SIT, respectively). In the comparable group of children receiving placebo, the baseline QoL score was 2.0 and increased in consecutive years of therapy to 2.3, 2.3 and 2.7 (p = 0.08, 0.09 and 0.019, respectively). The group of adolescents who received active treatment also showed significant improvements in QoL: a baseline QoL score of 2.7 decreased significantly in consecutive years of therapy to 2.0, 1.4 and 0.9 (p = 0.0018, 0.0006 and 0.0006, respectively). The comparable placebo group showed no change during the entire study, with a baseline mean QoL score of 2.0, and scores of 1.9, 1.9 and 2.2 after the first, second, and third years, respectively (p = 0.034, 0.5 and 0.68, respectively). Comparisons of the actively treated and placebo groups showed statistically significant differences for children after the second and third years of SIT (p = 0.015 and 0.001, respectively) and for adolescents after the third year of SIT (p = 0.03)

Quantity of evidence (sublingual immunotherapy vs placebo studies)

Searches identified 85 publications of DBPC RCTs of SLIT (see PRISMA flow diagram in Appendix 4). Of these, 52 publications relating to 44 RCTs were already included in the relevant Cochrane review83 (note: the other RCTs in the Cochrane review related to dust mites or animal dander). Of the remaining 33 publications, 15 were within the Cochrane search period but appear not to have been identified and two had been excluded from the Cochrane review despite apparently meeting the inclusion criteria. Details of the excluded studies are presented in Appendix 5. Sixteen publications25,32,158,188–199 post-dated the Cochrane searches; thirteen25,158,188–199 of these related to 11 new RCTs, three were updates of trials previously included in the Cochrane review: two32,200 related to the Dahl et al. (GT–08) trial201 and one202 provided additional data from the trial described in Didier et al. 24 As the purpose of this report was to update, rather than repeat, the Cochrane review, results are presented for only the 11 studies25,158,189–199 published from 2009 onwards, although all relevant studies were included in the meta-analyses.

Main study characteristics and risk of bias (sublingual immunotherapy compared with placebo studies)

The main study and population characteristics, and assessment of risk of bias, are detailed for each of the 11 new studies25,158,189–199 in Appendix 6. All studies were double-blind placebo-controlled RCTs. The largest four trials25,189,192,195 had between 276 and 633 participants, with the number of participants in the smaller trials ranging from 20 to 115. Skin prick tests were performed in all patients to demonstrate specific sensitivity. Allergy symptoms were described as moderate to severe in 6 out of 11 trials,189–191,195–197 with no indication of severity given in 5 out of 11 trials. 25,158,192–194 Seven studies25,189,190,192,193,195,197 stated that some patients also had asthma symptoms. Two trials specified that patients were treatment naive,190,191 four reported no details,25,158,189,197 and in five trials192–196 it was stated that patients had not received SIT within the last 3–5 years, although it was unclear if any patients had ever been treated with SIT. This may be important, as it is known that SIT can have long-term effects. Outcomes included SSs, MSs, combined SMSs, QoL and AEs (note: only outcomes consistent with the inclusion criteria of this report have been listed). Six25,189–192,196 of the eleven studies reported SMSs compared with none of the studies in the Cochrane review. As for SCIT compared with placebo trials, the number and types of symptom assessed varied widely (see Appendix 7 – SSs across studies).

The most commonly tested allergen was timothy grass (P. pratense), investigated in four trials,174,177–179 followed by tree pollen (three trials),158,191,197 a mix of several grasses (two trials),25,194 one trial196 with ragweed and one190 with Alternaria. Length of treatment varied between 8 and 10 weeks and over three pollen seasons, and there was also variation in treatment schedules (e.g. daily or weekly dosing). One study189 was in children and adolescents, one190 in both children and adults, with the remaining studies all conducted in adults. The main study characteristics are summarised in Table 19.

Similarity between trials was explored for a range of study and population characteristics. Trial duration and type and amount of allergen used varied between trials; however, this was explored as a source of heterogeneity in subgroup analyses. Where reported, inclusion criteria were very similar across trials, with the majority stipulating a minimum of 2 years' clinical history of moderate to severe SAR, incompletely controlled by standard medication; actual SAR history of included patients ranged from approximately 5 to 18 years. Four studies158,191,194,196 did not report whether or not patients with previous SIT were included. Rates of asthma across trials were largely consistent, with none greater than one-quarter (range 7–26%). All studies excluded patients with severe or perennial asthma.

Results of the risk of bias assessment are summarised in Table 20 (full details are given in Appendix 5).

Overall, the risk of bias was low for three or more areas of potential bias in most of the studies, although a lack of detail in the published reports meant that risk of bias was unclear in many cases. Study authors were not contacted, and an ‘unclear’ rating may be due to lack of reporting rather than a reflection of poor trial quality. All but one study191 indicated that endeavours were made to maintain blinding. There were only two instances191,197 in which a high risk of bias was identified; neither of these studies contributed to any meta-analyses. It should be noted that a certain degree of subjectivity remains in assigning a rating of low/unclear/high.

Effectiveness of sublingual immunotherapy compared with placebo

Of the 11 newly identified RCTs,25,158,189–197 only five reported data in a form suitable for meta-analysis. 25,189,190,192,196 All five studies provided SS, MS and combined SMS data,25,189,190,192,196 and three additionally provided QoL data25,189,192 (Table 21).

Symptom scores

A total of 39 SAR studies24,26,38,84–91,94–118,197,200 were included in the meta-analysis in the Cochrane review. Of these, two were excluded by this review,35,92 one92 reported combined results for patients with SAR and SAA, and one35 was restricted to patients with SAA. Further, 3-year results of the GT–08 trial200 superseded the 1-year results reported by Dahl,93 which was consequently removed from the meta-analysis. Searches for this review identified eight25,158,189–191,196,197,200 new studies reporting this outcome, of which six25,189,190,192,196,200 were included in the meta-analysis. In total, 2440 active (SLIT) and 2379 patients receiving placebo were included in 42 studies.

The combined SMD following SLIT was −0.33 (95% CI −0.42 to −0.25; p < 0.00001) favouring active treatment (Figure 6). This result is very similar to that found in the Cochrane review (SMD −0.34, 95% CI −0.44 to −0.25; p < 0.00001). Heterogeneity was also similar to the earlier review.

FIGURE 6.

Sublingual immunotherapy vs placebo: SSs.

Subgroup analyses in this report were limited to studies in SAR, whereas data from the earlier Cochrane review included studies of SLIT for the treatment of PAR. Small to moderate effect sizes favouring active SLIT were found in all subgroup analyses, and these did not differ significantly with age, study duration, MAC or type of allergen. All were statistically significant, with the exception of the four studies in Parietaria allergy;99,110,115,125 however, this was based on a total of only 124 participants. Results of subgroup analyses are shown below in Table 22. Forest plots of all subgroup analyses are shown in Appendix 6.

TABLE 22 - Sublingual immunotherapy vs placebo, subgroup analyses: SSs

CR, Cochrane review; ES, effect size; IV, inverse variance.

Subgroup	No. of studies	Total n	SMD (IV, random, 95% CI)	Comparison with CR
Age group
Adults	33	3476	−0.38 (−0.49 to −0.27)	Similar
Children	9	1343	−0.24 (−0.35 to −0.13)	ES smaller, but remained statistically significant
Duration (months)
< 6	15	1882	−0.34 (−0.47 to −0.20)	ES smaller, but remained statistically significant
6–12	15	1539	−0.31 (−0.47 to −0.16)	Similar
> 12	12	1398	−0.35 (−0.52 to −0.18)	ES smaller, but remained statistically significant
MAC
< 5 μg	6	229	−0.53 (−1.03 to −0.03)	ES larger; becomes statistically significant
5–20 μg	13	2287	−0.29 (−0.37 to −0.20)	Similar
> 20 μg	12	1088	−0.33 (−0.48 to −0.18)	Similar
Allergen
Grass	25	4042	−0.31 (−0.39 to −0.24)	34% increase n; ES similar
Ragweed	3	244	−0.44 (−0.69 to −0.18)	Similar
Parietaria	4	124	−0.27 (−0.62 to 0.09)	ES smaller and became non-significant
Tree	9	380	−0.42 (−0.77 to 0.06)	Data as per Cochrane review

Compared with the Cochrane review, subgroup analysis by age resulted in very similar effect sizes in adult participants but a much smaller effect in children (SMD −0.24 compared with −0.52 in the Cochrane review), although this remained statistically significant. Although only nine paediatric studies have been included here,26,84–90,189 compared with 15 in the Cochrane review,26,35,84–90,92,203–207 total participant numbers were very similar (1343 vs 1392 children, respectively) and heterogeneity was significantly reduced (I² = 0%, compared with 92% in the Cochrane review).

Analysis by treatment duration found reduced effect sizes in trials lasting < 6 months and over 12 months (the latter associated with a 69% increase in sample size) compared with the Cochrane review, but all remained statistically significant.

Few differences were found compared with the earlier review in subgroup analysis by allergen content, with the exception of studies using < 5 μg of major allergen. 88,99,111,116,190 Despite a small reduction in both study number and total sample size, effect size was larger in the present review (SMD −0.53, compared with SMD −0.32 in the Cochrane review), and became statistically significant. There was no apparent dose–response relationship for SLIT.

Subgroup analysis by allergen type gave effect sizes very similar to those in the earlier review, despite a 34% rise in sample size in the grass allergen subgroup. This latter was associated with a reduction in heterogeneity, which became non-significant. Effect size decreased slightly in Parietaria allergen, becoming non-significant, but total participant numbers were small (total n = 124 in present review).

Two studies158,197 reported results in a manner not suitable for meta-analysis and details are shown in Table 23. Data could be extracted from only one study,197 which showed a significant improvement in SSs compared with placebo.

TABLE 23 - Symptom score results: studies not in meta-analysis

Study ID	Results
Ventura 2009158	Not possible to extract data from graphs. Clinical improvements were noted with active treatment compared with placebo
Voltolini 2010197	Median baseline rhinorrhoea score in both groups was 2 (5–95 percentiles: 1–3); after 1 year IT median scores were 1 (5–95 percentiles: 0–2) with active SLIT and 1.5 (5–95 percentiles: 0–2) with placebo (p < 0.05)

Medication scores

A total of 32 SAR studies26,35,84–86,88–109,111,114,116,117,197 were included in the meta-analysis in the Cochrane review. Of these, two35,92 were excluded by this review, as described above. Further 3-year results of the GT−08 trial200 superseded the 1-year results reported by Dahl et al. ,93 which was consequently removed from the meta-analysis. Six new studies25,189,190,192,196,200 reported this outcome and were included in the meta-analysis.

In total, 1934 active and 1845 placebo patients were included in 35 studies. 25,26,84–86,88–91,94–109,111,114,116,117,189,190,192,196,197,200 The combined SMD was −0.27 (95% CI −0.37 to −0.17; p < 0.00001) favouring active treatment (Figure 7). This result is very similar to that found in the Cochrane review (−0.30, 95% CI −0.41 to −0.19; p < 0.00001). Heterogeneity was also similar to the earlier review.

FIGURE 7.

Sublingual immunotherapy vs placebo: MSs.

Small to moderate effect sizes favouring active SLIT were found in all subgroup analyses. MSs in children were not significantly better than with placebo treatment. This finding was consistent with that of the earlier Cochrane review, and effect size was decreased further with the addition of the more recent studies. Of the eight included studies,26,84–86,88–90,93 only one favouring placebo treatment was statistically significant. All others favoured active SLIT, but did not reach significance either alone or when combined. All other subgroup analyses were statistically significant, and most did not differ greatly from effect sizes reported in the earlier review, despite sometimes large increases in participant numbers (e.g. 31% increase in grass allergen; 69% increase in studies with duration of > 12 months). Analyses in two subgroups – > 20 μg major allergen and ragweed pollen SLIT – became statistically significant in the present review. Results of subgroup analyses are shown in Table 24.

TABLE 24 - Sublingual immunotherapy vs placebo, subgroup analyses: MSs

CR, Cochrane review; ES, effect size; IV, inverse variance.

Subgroup	No. of	Total n	SMD (IV, random, 95% CI)	Comparison with CR
Age group
Adults	27	2604	−0.35 (−0.47 to −0.23)	Similar
Children	8	1175	−0.08 (−0.25 to 0.08)	ES smaller; remained non-significant
Duration (months)
< 6	14	1517	−0.33 (−0.47 to −0.19)	Similar
6–12	13	1223	−0.31 (−0.53 to −0.08)	Similar
> 12	8	1039	−0.16 (−0.31 to −0.01)	69% increase n; ES smaller, but remained significant
MAC
< 5 μg	5	194	−0.63 (−1.08 to −0.18)	Similar
5–20 μg	12	2285	−0.18 (−0.30 to −0.05)	Similar
> 20 μg	10	708	−0.26 (−0.47 to −0.06)	ES similar; however, became statistically significant
Allergen
Grass	19	3028	−0.20 (−0.32 to −0.08)	31% increase in n; ES similar
Ragweed	3	244	−0.34 (−0.60 to −0.09)	ES similar; however, became statistically significant
Parietaria	4	124	−0.49 (−0.85 to −0.13)	No new studies; one study from CR excluded;34 ES smaller, but remained statistically significant
Tree	9	380	−0.38 (−0.62 to −0.13)	Data as per CR

Symptom and medication scores

The Cochrane review did not report on this outcome. We identified seven new studies25,189–192,196,200 that reported combined SMS, of which six25,189,190,192,196,200 were suitable for meta-analysis. Details of the remaining study191 are shown in Table 25. Although inadequate reporting means the results are difficult to interpret, active treatment appeared to result in improved outcomes in the second year of treatment.

TABLE 25 - Symptom and medication score results: studies not in meta-analysis

Study ID	Results
Fujimura 2011191	Data presented in graphical form only; summary statistic used unclear. No apparent difference in scores between treatment groups during first pollen season, but in second season better scores were reported in the active group

The six studies25,189,190,192,196,200 included in the meta-analysis represented a total of 690 patients who received active treatment and 704 receiving placebo (Figure 8). Combined SMD was −0.40 (95% CI −0.55 to −0.25; p < 0.00001) in favour of SLIT. Some heterogeneity between studies was indicated, but this was not statistically significant.

FIGURE 8.

Sublingual immunotherapy vs placebo: SMSs.

Moderate effect sizes favouring active SLIT were found in all subgroup analyses conducted (Table 26), and these were similar between studies. Combined SMD in the two studies190,191 lasting between 6 and 12 months was larger but a high degree of heterogeneity was indicated and this result was not statistically significant. Only one study189 conducted in children (n = 307) reported SMS and, therefore, meta-analysis was not possible. However, SMD for this study favoured active treatment and was statistically significant. Only one study190 was conducted using < 5 μg of major allergen and in Alternaria allergy, and only one study196 used > 20 μg of major allergen and was conducted in ragweed. Meta-analyses were, therefore, not possible in these subgroups. However, both of these studies190,196 favoured active treatment. No studies of SLIT for tree allergy reported usable data and no new studies were conducted in Parietaria allergy.

TABLE 26 - Sublingual immunotherapy vs placebo, subgroup analyses: SMSs

IV, inverse variance.

Subgroup	No. of studies	Total n	SMD (IV, random, 95% CI)
Age group
Adults	5	1087	−0.44 (−0.62 to −0.27)
Duration (months)
< 6	2	376	−0.31 (−0.51 to −0.11)
6–12	2	417	−0.66 (−1.63 to 0.30)
> 12	2	601	−0.48 (−0.64 to −0.31)
MAC
5–20 μg	3	985	−0.32 (−0.45 to −0.19)
Allergen
Grass	4	1299	−0.36 (−0.48 to −0.24)

Quality of life

Eight studies25,102,112,189,191,192,200,202 reported QoL data (three from Cochrane review,102,112,200 five new25,189,191,192,202). All assessed QoL using versions of the disease-specific RQLQ.

Seven studies25,102,189,192,200,202 suitable for meta-analysis included 927 active and 951 placebo patients in total, a 473% increase from the Cochrane review (Figure 9). Nevertheless, the effect size (SMD −0.37; 95% CI −0.52 to −0.22; p < 0.00001) was very similar to that in the earlier review (SMD −0.42; 95% CI −0.73 to −0.12; p = 0.0063). Heterogeneity between studies was reduced, but remained significant (χ² = 14.62; p = 0.02; I² = 59%).

FIGURE 9.

Sublingual immunotherapy vs placebo: QoL.

Summary of subgroup analyses are shown in Table 27.

TABLE 27 - Sublingual immunotherapy vs placebo, subgroup analyses: QoL

IV, inverse variance.

Subgroup	No. of studies	Total n	SMD (IV, random, 95% CI)
Age group
Adults	6	1658	−0.37 (−0.52 to −0.22)
Duration (months)
< 6	4	908	−0.45 (−0.74 to −0.17)
> 12	2	601	−0.37 (−0.54 to −0.21)
MAC
5–20 μg	2	507	−0.32 (−0.49 to −0.14)
> 20 μg	2	323	−0.70 (−0.93 to 0.48)

One189 of the new studies reported the first QoL results for children and adolescents, and these were statistically significant in favour of SLIT. Subgroup analysis of the remaining six studies25,102,112,192,200,202 did not lead to significantly different results from those reported above.

Four102,112,189,202 of the seven studies25,102,112,189,192,200,202 reporting QoL data lasted for < 6 months. Subgroup analysis found a moderate effect size in favour of SLIT, although heterogeneity between these studies was high (χ² = 11.54, p = 0.009, I² = 74%). One study192 was of between 6 and 12 months duration and, therefore, meta-analysis of this subgroup was not performed; however, the SMD was small and failed to reach statistical significance. In contrast, two studies25,200 presented data from trials lasting > 12 months, comprising a total of 601 participants. These longer studies25,200 reported a moderate effect size for QoL, which was statistically significant and no between-study heterogeneity was detected (I² = 0%).

None of the studies included in the meta-analysis used vaccines with < 5 μg MAC. Two studies each used medium and high doses, and results are suggestive of a positive correlation between dose and effect size, although, given the small number of studies involved, particularly at the higher dose, these results should be interpreted with caution.

Four of the included studies25,102,200,202 used the full version of the disease-specific RQLQ to measure QoL, and random-effect meta-analysis was conducted for these studies (Figure 10). Two studies182,192 used alternate versions of the RQLQ (age-specific or standardised activities), which do not use the same scale or domains as the original, and these could therefore not be included in the meta-analysis. It was not possible to identify the instrument used in one study. 112

FIGURE 10.

Sublingual immunotherapy vs placebo: QoL (RQLQ scores).

Weighted MD indicated an overall reduction in RQLQ scores of 0.34 units in the active SLIT group, indicating a positive effect of treatment.

One study191 did not present data in a manner suitable for meta-analysis (Table 28), but active treatment resulted in significantly improved QoL scores compared with placebo during the second year of treatment.

TABLE 28 - Quality-of-life results: studies not in meta-analysis

Study ID	Results
Fujimura 2011191	Data presented in graphical form only; summary statistic used unclear. Total scores on the Japanese Allergic Rhinitis QoL Standard Questionnaire No. 1 were significantly better in the active group (p <0.01) during the second pollen season. Data are not presented for the first pollen season

Quantity of evidence sublingual immunotherapy compared with subcutaneous immunotherapy

Searches identified only one randomised, double-blind, double-dummy comparison study210 of SLIT with SCIT (n = 71). Three further studies158,211,212 were initially identified as potentially relevant, but were subsequently excluded. One study158 (n = 40) was not a double-blind, double-dummy comparison; it included a comparison of SCIT with placebo and SLIT with placebo but no adequately blinded direct comparison of SCIT with SLIT, and results for SLIT compared with SCIT were not reported (this study is included in the relevant sections on placebo-controlled studies in this report). A second study211 (n = 47) compared a SLIT with a SCIT treatment arm, but there was no blinding (no treatment with placebo). Finally, a small, double-blind, double-dummy study (n = 20) by Quirino et al. 212 used 10 ‘matched pairs’ and does not appear to have used random allocation. The three studies that reported results for SCIT compared with SLIT found no significant differences between the two. Results for the one study210 meeting inclusion criteria are reported in more detail below.

Main study characteristics sublingual immunotherapy compared with subcutaneous immunotherapy

The main study and population characteristics of the included study, and the assessment of risk of bias are detailed in Appendix 5.

The main characteristics are summarised in Table 31.

Results of the risk of bias assessment are summarised in Table 32 (full details are given in Appendix 5).

As in many of the placebo-controlled studies, details of randomisation and blinding are reported but a lack of detail on other aspects of quality make it difficult to judge the overall quality.

Effectiveness of subcutaneous immunotherapy compared with sublingual immunotherapy

Results were reported for SSs, MSs, QoL and AEs and are summarised below (Box 3). Results were not presented in a way that is consistent with most other RCTs in this area, in that absolute mean SSs or MSs post treatment are not presented; instead median changes relative to the preceding pollen season are given. Nevertheless, the findings are consistent with other studies showing a benefit in terms of symptoms and MSs with both SCIT and SLIT compared with placebo. No significant differences between SLIT and SCIT groups were identified in this small study. 210 This does not mean that such differences do not potentially exist.

Indirect comparison of subcutaneous immunotherapy with sublingual immunotherapy

Given the paucity of direct comparisons (head-to-head trials) between SCIT and SLIT, it was decided to undertake an indirect comparison of SCIT with SLIT using data from the separate meta-analyses of SCIT compared with placebo and SLIT compared with placebo. Indirect comparison analyses rely on the assumptions (explored below) that there is sufficient similarity both within the SCIT compared with placebo and the SLIT compared with placebo trials (homogeneity assumption138), and that the true treatment effect comparing any two interventions would be similar across all trials, irrespective of whether they included one or both of those interventions (similarity assumption138).

The direct (head-to-head) evidence from the one small SCIT compared with SLIT trial210 included in this report could not be incorporated into this analysis, as the outcomes were not reported in a way that would allow this.

This section presents the most relevant results of the statistical analysis. The full list of parameter estimates and modelling approach is reported in Appendix 3.

Symptom scores

When covariates were not included in the model (unadjusted model), the standardised score difference was 0.351 (95% CrI 0.127 to 0.586), a statistically significant result in favour of SCIT. Probabilistic analysis suggests that SCIT has a greater probability of being the best treatment compared with SLIT overall (unadjusted model) and also when participant age, study duration, MAC and type of allergen are accounted for (Table 33). However, not all the standardised score differences were statistically significant. For Alternaria allergy, the best estimate probability suggests that SLIT is the preferred treatment; however, this is based on only a single study for each intervention and a non-significant standardised score difference. Residual heterogeneity in the model could not be accounted for by participant age or type of allergen, while including MAC as a covariate reduced heterogeneity slightly (as shown by a decrease in σ²).

TABLE 33 - Indirect comparison of SSs: probabilistic analysis and random-effects meta-regression of SCIT vs SLIT under different modelling conditions

A, number of patients receiving active treatment; P, number of patients receiving placebo.

Not statistically significant.

One study91 had separate arms for tree and grass allergens.

Covariate	SCIT, no.of trials (A/P)	SLIT, no. of trials (A/P)	Probability that treatment is best (%)			ICMR SCIT compared with SLIT
			Placebo	SCIT	SLIT	Standardised score difference (95% CrI)	Heterogeneity, σ2 (95% CrI)
None	17 (659/529)	42 (2440/2379)	00.0	99.9	00.1	0.351 (0.127 to 0.586)	0.089 (0.027 to 0.586)
Age group
Adult	16 (634/506)	33 (1768/1708)	00.0	99.6	00.4	0.328 (0.088 to 0.579)	0.091 (0.028 to 0.192)
Child	1 (25/23)	9 (672/671)	00.9	54.9	44.2	0.059 (−0.837 to 0.966)a
Trial duration (months)
< 6 months	5 (136/138)	15 (942/940)	00.0	> 99	00.0	0.822 (0.379 to 1.299)	0.1112 (0.040 to 0.221)
6–12	2 (203/106)	15 (759/780)	00.0	80.1	19.9	0.252 (−0.357 to 0.862)a
> 12	8 (227/215)	12 (739/659)	00.0	83.4	16.6	0.187 (−0.199 to 0.577)a
MAC (µg)
< 5	3 (112/116)	6 (118/111)	00.0	47.4	52.6	−0.016 (−0.516 to 0.483)a	0.053 (0.005 to 0.13)
5–20	5 (117/114)	13 (1163/1124)	00.0	92.0	08.0	0.262 (−0.108 to 0.634)a
> 20	3 (222/119)	12 (564/524)	00.0	99.7	00.3	0.582 (0.167 to 1.060)
Type of allergenb
Grass	9 (295/257)	25 (2050/1995)	00.0	99.4	00.6	0.396 (0.090 to 0.721)	0.096 (0.030 to 0.20)
Parietaria	3 (227/126)	4 (60/64)	00.0	99.1	00.9	0.775 (0.140 to 1.446)
Ragweed	0 (0/0)	3 (118/126)	01.2	50.0	48.8	0.011 (−197.0 to 196.70)a
Tree	4 (112/123)	9 (197/183)	00.0	64.3	35.7	0.092 (−0.408 to 0.597)a
Alternaria	1 (25/19)	1 (15/11)	00.1	04.6	95.3	−1.133 (−2.460 to 0.199)a

The ICMR model included comparisons where there were no studies (no data) in either the SCIT or the SLIT arm (e.g. no SCIT trials for ragweed compared with three SLIT trials95,97,196 for ragweed); here the extremely large CrIs around the standardised score difference reflect the uncertainty resulting from this absence of data.

In order to explore unexplained residual heterogeneity further, two post hoc meta-regressions were conducted. In the first analysis, year of publication was used as a covariate in the model, and a strong effect for year was identified. An improvement in the fit of the model to the data was also observed (DIC 502, compared with 508 for the null model) and the heterogeneity parameter estimate was also lower [σ² 0.067 (95% CrI 0.017 to 0.147) compared with 0.089 (95% CrI 0.027 to 0.187), respectively].

Figure 11a shows pooled SMD point estimates for placebo-controlled trials and suggests that results from older studies show a greater benefit for SCIT, with more recent studies finding a decreasing benefit (more negative SMD values indicate a greater improvement in SSs compared with placebo). In contrast, SMD estimates for SLIT compared with placebo appear to remain more stable over time. The ICMR with year as a covariate (Figure 11b) finds that from approximately 2007 there is an increased probability that SLIT is more beneficial than SCIT. Note that standardised score differences at different time points are not all statistically significant (see Appendix 3).

FIGURE 11.

Symptoms scores: year of publication as covariate in the model. (a) theoretical and observed SMD and credible boundaries by year of publication for SCIT vs placebo and SLIT vs placebo; (b) probability of intervention being superior.

Given the huge variability in symptoms recorded between trials (see Appendix 7 for types and numbers of symptoms scored in different trials), a second post hoc analysis was conducted to explore any effect of this variable on trial outcome. It was found that increasing numbers of symptoms being measured in a trial appeared to favour SCIT over SLIT (Figure 12). Again, standardised score differences are not all statistically significant (see Appendix 3) and residual heterogeneity increases in this analysis.

FIGURE 12.

Number of symptoms measured as covariate in the model. (a) theoretical and observed SMD and credible boundaries by year of publication for SCIT vs placebo and SLIT vs placebo; (b) probability of intervention being superior.

Medication scores

The unadjusted model (no covariates included), found a standardised score difference of 0.273 (95% CrI 0.027 to 0.529) in favour of SCIT (Table 34). This was associated with a > 99% chance of SCIT being the best treatment. Where meta-regressions resulted in standardised score differences in favour of SLIT, these were not statistically significant. It appears that some of the heterogeneity could be reduced by introducing MAC and particularly age (adult/child) as a covariate.

TABLE 34 - Indirect comparison of MSs: probabilistic analysis and random-effects meta-regression of SCIT vs SLIT under different modelling conditions

A, number of patients receiving active treatment; P, number of patients receiving placebo.

Not statistically significant.

One study91 had separate arms for tree and grass allergens.

Covariate	SCIT, no.of trials (A/P)	SLIT, no. of trials (A/P)	Probability (%) that treatment is best			ICMR SCIT vs SLIT
			Placebo	SCIT	SLIT	Standardised score difference (95% CrI)	Heterogeneity, σ2 (95% CrI)
None	16 (621/483)	35 (1934/1845)	00.0	> 99	00.0	0.273 (0.027 to 0.529)	0.099 (0.030 to 0.211)
Age group
Adult	15 (596/464)	27 (1353/1251)	01.8	43.1	55.1	−0.009 (−0.124 to 0.106)a	0.002 (0.001 to 0.005)
Child	1 (25/19)	8 (581/594)	00.0	68.1	31.9	0.009 (−0.031 to 0.050)a
Trial duration (months)
< 6	2 (69/74)	14 (766/751)	00.0	50.9	49.1	0.006 (−0.634 to 0.634)a	0.120 (0.039 to 0.255)
6–12	3 (215/118)	13 (609/614)	00.0	75.6	24.3	0.192 (−0.376 to 0.750)a
> 12	3 (244/225)	8 (559/480)	00.0	99.1	00.9	0.511 (0.094 to 0.950)
MAC (µg)
< 5	2 (71/76)	5 (98/96)	00.0	15.1	84.9	−0.286 (−0.849 to 0.266)a	0.042 (0.006 to 0.108)
5–20	6 (129/126)	12 (1167/1118)	00.0	95.3	04.7	0.283 (−0.049 to 0.622)a
> 20	2 (203/102)	10 (383/325)	00.0	90.4	09.6	0.298 (−0.164 to 0.765)a
Type of allergenb
Grass	8 (263/220)	19 (1544/1461)	00.0	99.8	00.2	0.504 (0.169 to 0.873)	0.094 (0.024 to 0.209)
Parietaria	2 (209/109)	4 (60/64)	00.1	42.1	57.9	−0.068 (−0.761 to 0.619)a
Ragweed	1 (11/12)	3 (118/126)	00.8	67.8	31.4	0.251 (−0.809 to 1.325)a
Tree	4 (113/123)	8 (197/183)	00.0	43.0	57.0	−0.45 (−0.550 to 0.453)a
Alternaria	1 (25/35)	1 (15/11)	00.0	18.0	82.0	−0.608 (−1.933 to 0.710)a

As for SSs above, adjusting for year of publication found changes in benefit from SCIT and SLIT over time, although the effect was not as pronounced and more difficult to interpret (Figure 13).

FIGURE 13.

Medication scores: year of publication as covariate in the model. (a) theoretical and observed SMD and credible boundaries by year of publication for SCIT vs placebo and SLIT vs placebo; (b) probability of intervention being superior.

Ongoing trials

Searches identified 22 (12 SLIT, 10 SCIT) Phase III double-blind, randomised placebo-controlled trials that are still ongoing or had recently finished but for which no published results were yet available (see Appendix 9). The majority of trials are being conducted in adults (some including adolescents), comprising over 4500 adult subjects in total, with only two SLIT trials (n = 1450) and one SCIT trial (number of participants not stated) recruiting children and adolescents specifically. We identified only one Phase II/III double-blind, double-dummy study of SLIT compared with SCIT, initiated in March 2011 and due for completion September 2014. 215 Four studies216–219 are investigating the efficacy and safety of SCIT with recombinant allergens, of which one includes off-treatment follow-up:216 a SCIT trial will collect data for one season after 2 years of treatment; a further trial will look at long-term allergy and asthma outcomes over a 5-year period in 1000 children aged 5–12 years (the Grazax Asthma Prevention study59). However, in general, the majority of trials do not appear to differ extensively from previously published studies in terms of patients and treatment regimens. Although beyond the scope of this report, it would be of interest to investigate the type of outcome measures being used given recent recommendations in this area.

Methods

Quantity of evidence

Searches identified 406 potentially relevant publications, of which 330 were excluded at the title/abstract stage. Of the 76 publications that were potentially relevant, full texts could not be obtained for eight. Full-text copies of 68 papers were examined, of which 52 were excluded (reasons for exclusion are detailed in Appendix 5). Sixteen publications were included. Of these, 14 were primary EEs and two were reviews of EE studies. In addition, three studies were identified that reported utility-based outcomes for SAR.

Main characteristics of economic evaluations

The main study characteristics and findings are shown below. Further details can be found below (see Table 40) and Appendix 9.

Quality of economic evaluations

Quality assessment of the EEs identified a number of methodological issues, which are outlined below (see Appendix 10 for completed checklists and see Table 39 for key issues for each study).

Results of economic evaluations

Seven studies134,223,226–228,231,232 compared SLIT with standard care, and all found that SLIT was more cost-effective. Studies reporting results of EEs based on data from randomised studies134,222,225,228,231,232 seemed to have been the most robust. The quality of reporting for resource use and unit cost data in these trial-based studies also appeared to be adequate. Four studies228,230–232 reported cost per QALY (Table 39); all were based on the same multinational trial (GT–0893) and included populations from different European countries in the respective analyses. All found that Grazax was cost-effective (below a threshold of £20,000), providing that annual costs of Grazax remain below £2200. The current annual cost in the UK is £814 (at £2.23 per tablet). 46 The evaluations appear to be well conducted; however, there was a general lack of detailed reporting on the outcome side, for example lack of disaggregated baseline and follow-up EQ-5D values and QALY gains. All four studies were funded by the manufacturer of Grazax.

Six studies222,227,229,233–235 compared SCIT with standard treatment. All found that SCIT was associated with better outcomes and/or lower long-term costs. Two studies229,233 calculated a cost per QALY for SCIT (see Table 39). Brüggenjürgen et al. 229 found a low ICER, but the robustness of this result is uncertain, as there was insufficient detail on the instrument used to derive utilities, and on cost and resource use data. The evaluation by Keiding and Jørgensen233 found that SCIT either dominated ST or was associated with very low ICERs, although robust sensitivity analyses were not conducted.

Where SCIT and SLIT were directly compared against each other, SCIT was found to be both more effective and more cost-effective over the long term. The sample size of the only trial225 that has directly compared the cost-effectiveness of SCIT and SLIT was, however, small (n = 64), and therefore more studies based on larger samples are needed. As this was a CCA, there is no combined cost-effectiveness measure. Assumptions made in the other study224 that directly compared the two interventions were not robust enough and variations of such assumptions should ideally be tested within a sensitivity analysis.

Other useful studies

Three studies242–244 were identified that reported utility-based outcomes for AR. Tamayama et al. 242 used the rating scale and time trade-off methods to estimate utility weights for four AR severity levels (mild, moderate, severe and severest) based on a Japanese sample. From mild to severest, time-trade-off estimates were 0.96, 0.94, 0.89 and 0.83, respectively. The rating scale counterpart weights were 0.82, 0.71, 0.56 and 0.43. Chen et al. 243 collected EQ-5D data for three groups in the USA and reported the following values: 0.76 (for individuals with asthma and rhinitis), 0.76 (for those with asthma only) and 0.92 (for individuals with rhinitis only). In Wasserfallen et al. ,244 the responsiveness of the EQ-5D when used on asthma patients with AR was compared with that of the McMaster Asthma Quality of Life Questionnaire (MAQOL). The EQ-5D was found to be less responsive than the MAQOL.

Two abstracts245,246 reporting EEs of Grazax based on populations of children with AR were identified at a late stage of writing this report and have thus not been formally included. A detailed appraisal was not possible, as the information was in abstract form only, but the results of these studies were consistent with those found for adults in that Grazax was found to be cost-effective compared with ST below a £20,000 threshold.

Reviews of economic evaluations

Both reviews of EEs237,247 considered studies that applied simple cost analyses as well as full EEs from societal, health service and patient perspectives (Table 40).

Nine EEs222,223,225–228,230,233,235 included in Berto et al. 247 were also identified in the present review. All of the EEs reported in Hankin et al. 237 were included in our review with the exception of Buchner and Siepe,248 which seems to be a cost analysis rather than a CBA. Neither review presents a detailed search strategy. An additional seven EEs224,225,229–232,234 were identified for this report.

Consistent with our findings, both reviews found that IT (SCIT or SLIT) was (just) more effective or, in some cases, both more effective and cost-effective when compared with standard care. Hankin et al. 237 also highlighted limitations associated with the data sources used in the EEs, a finding similar to ours. In particular, there were issues around generalisability and short time horizons associated with trial-based studies, and concerns about the quality of data obtained from observational studies were also raised.

Conclusions

Economic evaluations varied widely in the type of analysis used, outcome measures, sources for cost and effectiveness data, and adequacy of reporting for different parameters (Table 41). They did find consistently that where either SCIT or SLIT was compared with standard therapy, IT was (just) more effective or, in some cases, both more effective and cost-effective. The most robust studies228,230–232 found that SLIT is likely to be cost-effective at thresholds of £20,000; these studies did not, however, report all of the utility data in a disaggregated form and all were funded by a manufacturer of SIT products. SCIT was also found to be cost-effective at a threshold of £20,000 (based on two studies229,233); however, these results were associated with slightly greater uncertainty.

Only two studies224,225 looked at the cost-effectiveness of SCIT compared with SLIT. Although suggestive of greater cost-effectiveness for SCIT, both studies had some methodological concerns associated with them and neither presented combined cost-effectiveness measures.

Adult Markov model

The long-term progress of a hypothetical cohort of AR patients moving along the three alternative pathways of care (SCIT, SLIT and ST) was to be compared. The treatment schedules for SCIT, SLIT and ST are presented in more detail in Chapter 1 (see Allergen immunotherapy). Briefly, patients would receive weekly subcutaneous injections (SCIT) for 16 weeks in a clinical setting, followed by monthly injections for up to 3 years. Patients receiving SLIT would take daily tablets or drops at home, also for up to 3 years. Symptomatic treatments may be oral, topical or intranasal. Occasionally, systemic corticosteroids may be prescribed.

Patients would then follow clinical pathways designed to mirror the natural progression of the condition in the population, with health resources use for all three groups also modelled. The aim was to enable model-based predictions of costs and outcomes to be compared for the SCIT, SLIT and ST groups in a CUA from the UK NHS and patient perspectives.

Adult Markov model structure and model inputs

The structure of the adult Markov model is shown in Figures 14–16. Only health states for the SCIT and ST arms are presented. The health states for the SLIT arm are identical to those in the SCIT arm.

FIGURE 14.

Adult Markov model structure: SLIT arm. Note that the health states for the SLIT arm are identical to those in the SCIT arm.

FIGURE 15.

Adult Markov model structure: SCIT arm. [+], same structure as the first expanded structure directly above but with appropriate changes in probabilities.

FIGURE 16.

Adult Markov model structure: ST arm. [+], same structure as the first expanded structure directly above but with appropriate changes in probabilities.

Subcutaneous immunotherapy arm

The Markov process for the SCIT arm begins with the initial health state called ‘Start IT’ representing individuals with SAR eligible for treatment with SCIT. Following the ‘Start IT’ health state, the model then distinguishes between patients who, in addition to SCIT, would receive treatment to control all, or some, of their symptoms (symptoms controlled and symptoms part controlled), as well as patients who would not receive any ST at all as they are assumed to be symptom free (no ST).

In the second or third year, patients who complete the previous year of SCIT (with or without ST) may then continue with, or quit, SCIT. In total, patients can move to one of 10 possible health states. The first three represent variations of patients continuing with SCIT with or without ST (‘on IT no ST’, ‘on IT symptoms controlled’ or ‘on IT symptoms partly controlled’). The next three health states are for patients who have completed SCIT and then continue with or without ST (i.e. ‘complete IT no ST’, ‘complete IT symptoms controlled’ or ‘complete IT symptoms partly controlled’). Patients who remained on SCIT for 3 years move to one of these states at the end of the third year. Another three health states depict options for patients who continue with or without ST having quit SCIT in the first or second year (i.e. ‘Early quit IT no ST’, ‘Early quit IT symptoms controlled’ or ‘Early quit IT symptoms partly controlled’). The last health state is ‘Dead’. Moving on from nine of these 10 health states (excluding ‘Dead’), the model also allowed for a change in ST. For instance, given that an adult with AR completed treatment with SCIT in the first or second year and was not on any other treatment, what was the probability that they would add treatments to control all, or some, of their symptoms at some point within the following year?

The symptomatic treatment-only arm

The starting point for the Markov process in the ST-only arm is the health state called ‘Start Symptomatic Rx only’ representing individuals with SAR eligible for treatment with SCIT or SLIT, but only receiving treatment to relieve their AR symptoms. Following the ‘Start Symptomatic Rx only’ health state, the model then distinguishes between patients who continue with treatment to control all or some of their symptoms (symptoms controlled and symptoms part controlled) as well as patients who stop receiving any ST completely (no ST). In the second or third year, patients who survive the previous year with or without ST can then move to one of four possible health states: have no ST (No ST), have treatment to control all of their symptoms (Symptoms controlled), have treatment to control some of their symptoms (Symptoms partly controlled) or die (Dead).

Child Markov model

Methods of delivering SIT were similar to those in the adult model, but with appropriate adjustments in dosages. The resultant clinical pathways were also designed to mirror the natural progression of AR in the children with health resources use for all three groups also modelled. The goal for this model was also to enable model-based predictions of costs and outcomes to be compared for the SCIT, SLIT and ST groups in a CUA from the UK NHS and patient perspectives.

Child model structure and model inputs

The structure of the child Markov model is similar to that for adults except it incorporates asthma in all outcomes in order to account for the probability of developing the condition among children with AR (Figures 17–20). Only health states for the SCIT and ‘Symptomatic treatment’ arms are presented. The health states for the ‘SLIT’ arm are identical to those in the ‘SCIT’ arm.

FIGURE 17.

Child Markov model structure. Note that the health states for the SLIT arm are identical to those in the SCIT arm.

FIGURE 18.

Child Markov model structure: SCIT arm (subtrees 1–6). [+], same structure as the first expanded structure directly above but with appropriate changes in probabilities.

FIGURE 19.

Child Markov model structure: SCIT arm (subtrees 7–18). [+], same structure as the first expanded structure directly above but with appropriate changes in probabilities.

FIGURE 20.

Child Markov model structure: ST arm. [+], same structure as the first expanded structure directly above but with appropriate changes in probabilities.

Subcutaneous immunotherapy arm

The Markov process for the SCIT arm begins with the initial health state called ‘Start IT’, representing children with SAR who are eligible for treatment with SCIT. Following the ‘Start IT’ health state, the model then distinguishes between children who, in addition to SCIT, would receive treatment to control all, or some of their symptoms (symptoms controlled and symptoms part controlled), as well as children who would not receive any ST at all (no ST). In the second or third year, asthmatic and non-asthmatic children who complete the previous year following SCIT (with or without ST) may then continue with, or quit, SCIT. Therefore, patients can move to 1 of 19 possible health states. The first six represent variations of asthmatic and non-asthmatic children continuing with SCIT with or without ST (‘on IT no ST with asthma’, ‘on IT no ST without asthma’, ‘on IT symptoms controlled with asthma’, ‘on IT symptoms controlled without asthma’, ‘on IT symptoms partly controlled with asthma’ or ‘on IT symptoms partly controlled without asthma’). The next six present states for asthmatic and non-asthmatic children who complete SCIT and then continue with or without ST (i.e. ‘complete IT no ST with asthma’, ‘complete IT no ST without asthma’, ‘complete IT symptoms controlled with asthma’, ‘complete IT symptoms controlled without asthma’, ‘complete IT symptoms partly controlled with asthma’ or ‘complete IT symptoms partly controlled without asthma’). Patients who remained on SCIT for 3 years move to one of these states at the end of the third year. Another six health states depict options for asthmatic and non-asthmatic children who continue with or without ST having quit SCIT in the first or second year (i.e. ‘Early quit IT no ST with asthma’, ‘Early quit IT no ST without asthma’, ‘Early quit IT symptoms controlled with asthma’, ‘Early quit IT symptoms controlled without asthma’, ‘Early quit IT symptoms partly controlled with asthma’ or ‘Early quit IT symptoms partly controlled without asthma’). The last health state is ‘Dead’. Moving from 18 of these 19 health states (excluding ‘Dead’), the model also allowed for a change in ST. For instance, given that an asthmatic child with AR completed treatment with SCIT in the first or second year and was not on any other treatment, what was the probability that they would add treatments to control all or some of their symptoms at some point within the following year?

The symptomatic treatment-only arm

The starting point for the Markov process in the ST-only arm is the health state called ‘Start Symptomatic Rx only’, representing children with SAR who are eligible for treatment with SIT but who are receiving treatment to relieve only their AR symptoms. Following the ‘Start Symptomatic Rx only’ health state, the model then distinguishes between children who continue with treatment to control all or some of their symptoms (symptoms controlled and symptoms part controlled), as well as children who would not receive any ST at all (no ST). In the second or third year, both asthmatic and non-asthmatic patients who complete the previous year with or without ST can then move to one of seven possible health states. These include states for asthmatic children who have no ST (‘No ST with asthma’), non-asthmatic children who have no ST (‘No ST without asthma’), asthmatic children who have treatment to control all of their symptoms (‘Symptoms controlled with asthma’) and non-asthmatic children who have treatment to control all of their symptoms (‘Symptoms controlled without asthma’). The rest of health states are for asthmatic children who have treatment to control some of their symptoms (‘Symptoms partly controlled with asthma’), non-asthmatic children who have treatment to control some of their symptoms (‘Symptoms partly controlled without asthma’) and ‘Dead’.

Resource use and costs

All costs were to be reported in UK pounds at 2011–12 unit prices and, where appropriate, discounted at 3.5% as recommended by the UK NICE. 249 Resource use data were to be obtained from studies reviewed, as well as from expert opinion. Unit cost data were to be derived from nationally representative sources, such as the BNF (2012),250 the National Schedule for Reference Costs and the Unit Costs of Health and Social Care (PSSRU). 251

Utility values

As the primary outcome was to be expressed in terms of costs per QALYs gained, studies were of interest if they reported utility scores to reflect the health-related QoL associated with each health state in the model.

Transition probabilities

To complete the process of populating the model, it is necessary to obtain transition probabilities governing movement between the different health states in both adult and child models from studies in this review. A full list of transition probabilities required for the two models is provided in Appendix 11.

Cost data

Two studies228,231 were conducted using samples that included subjects from the UK, both of which compared SLIT with standard care. They presented detailed UK costs associated with AR, based on resource use data such as visits to the GP, allergy specialist, and A&E department. In addition, costs of symptomatic and asthma medication, as well as indirect costs associated with production loss, were also presented. The cost estimates reported in these studies lend themselves to replication in our model and we would therefore consider these estimates to be credible inputs to our model. Despite these cost estimates all being based on the GT–08 GRAZAX trial,93 it is likely that these results would be transferable to the UK generally. Other potentially useful cost estimates were identified in four other European studies. 225,226,230,234

Outcome measures

Over 15 different types of outcome measures have been reported in the studies evaluated.

As described above (see Quality of economic evaluations), the vast number of these outcomes were not expressed as utilities but rather as ‘natural’ outcomes, and, as such, cannot be used in the Markov model. These include the number of asthma and rhinitis exacerbations, number of hospital visits and absence from nursery or school, SSs and MSs or the RQLQ.

Some studies did report utility-based outcome measures that could be converted into QALYs. As outlined above (see Main characteristics of economic evaluations), these utility-based outcomes were based on EQ-5D for four of the studies228,230–232 and on an unspecified instrument for one study. 229 Studies whose QALYs were based on EQ-5D did not present EQ-5D data in sufficient detail to allow for their replication in our model. In particular, no baseline or follow-up values were reported in three of these studies,228,230,231 with only the final results being reported in terms of cost per QALY gained. In the Beriot-Mathiot et al. study,232 EQ-5D could be read off a diagram, but these were not disaggregated according to SLIT or ST. Brüggenjürgen et al. 229 reported utility scores for the following health states: mild AR (0.7579); moderate to severe AR (0.7378); severe AR and mild allergic asthma (0.7317); severe AR and moderate to severe allergic asthma (0.6985); no symptoms (0.7841); and death (0.0). Much higher utility scores were reported by Tamayama et al. 242 for similar AR health states: 0.96 (mild), 0.94 (moderate), 0.89 (severe) and 0.83 (severest). Provided that information on transition probabilities could have been obtained, these values could potentially have been used. In order to explore the possibility of using utility data based on a UK population, study authors from the GT–08 trial93 were contacted, but it was not possible to obtain the necessary data.

Model structure and transition probabilities

Only five studies224,226,227,229,235 were model-based studies (EEs or effectiveness studies): three226,227,235 were based on decision tree models, whereas the other two224,229 used Markov models.

Berto et al. 226 compared SLIT (with standard care) to standard care only, in a decision tree framework based on a sample of 2230 patients. A model structure was presented but the health states analysed are not the same or adaptable to those in our model. More importantly, information of transition probabilities depicting movements between health states is not presented in the study,226 making it impossible for the model to be replicated using UK cost data.

The second decision tree model-based evaluation235 was based on a cohort of 2000 patients and compared SCIT with ST. This study235 presented some potentially useful data on probabilities in terms of asthma symptoms: the proportion of patients with AR and asthma compared with those with AR only; patients with asthma symptoms compared with those without asthma symptoms following either symptomatic or specific IT treatment; proportion of patients who develop asthma compared with proportion who do not among patients previously without asthma; and the proportion of patients who stop having asthma symptoms compared with proportion who continue with the symptoms among patients previously with asthma. The target outcome was the number of additional patients free from asthma symptoms after 10 years. However, it was difficult to ascertain how robust these probabilities were as the sources were not described in detail. Changes in symptoms or severity of AR were not incorporated into this analysis.

The third evaluation based on a decision tree model226 compared SCIT, SLIT and ST for adults and children with AR and asthma. Some potentially important transition probabilities were also presented in this study. 226 However, most of these were based on expert opinion from a Delphi panel, whereas some information was obtained from an unpublished report. All epidemiological data (‘hypotheses’) for children were determined by the Delphi panel. For both children and adults, for example, the study227 gives proportions that help determine the following probabilities: having rhinitis only as opposed to having rhinitis and asthma; having moderate compared with having severe rhinitis; and having mild compared with moderate asthma, as well as the percentage decrease in ST associated with IT by type of allergy and by SCIT or SLIT. However, attempts to replicate the model using UK cost data are not possible, as not all of the probabilities required to populate the model are presented, for example the proportion of patients who were either asymptomatic, who had an improvement in symptoms or those for whom the response to IT was inadequate.

Brüggenjürgen et al. 229 presented an analysis based on a Markov model. This study229 was based on a hypothetical cohort of 2000 patients and compared SCIT in addition to ST to ST alone. An overview of the Markov model was given in the paper but the actual structure or the parameter estimates are not presented. This makes it difficult to assess the suitability of the model inputs for use in our model or for replicating the model using UK cost data.

The second Markov model-based study224 was a cost analysis comparing SCIT, SLIT and variations of SCIT with standard care. The structure of the model was presented in the paper showing various pathways taken by patients following different treatment options. The health states presented in this model are not the same as those in our preferred model, and, more importantly, there was a lack of detail on the transition probabilities governing movement between health states, which precluded attempts to replicate the model using UK cost data.

Suitable data on transition probabilities, i.e. proportion of patients moving between different health states or levels of disease severity depending on treatment, were also not identified in any of the clinical effectiveness studies reviewed for this report.

Summary

Detailed information was available on cost data. A number of studies (both UK and non-UK) reported direct and indirect costs in sufficient detail to allow for their replication in our model. In particular, data from Bachert et al. 228 and Nasser et al. 231 were useful, supplemented with those obtained from expert opinion. Furthermore, data on utility scores associated with some severity levels of AR were also available based on German229 and Japanese230 populations. It may therefore have been possible to adapt these data for use in our models.

The biggest challenge faced, however, was the lack of information on transition probabilities. Although some model structures were presented in sufficient detail in some studies,224,227 and utility data were provided in others,229 not enough information on transition probabilities was given in all of the studies to allow for model replication using UK cost data.

Methods

Results

Cost-effectiveness results

Results based on both direct comparisons and indirect comparisons of the difference in RQLQ between the three groups are presented below.

Direct comparisons

As shown in Table 50, the mean change in RQLQ for the comparison between SLIT and ST was 0.340 (95% CI 0.18 to 0.49) in favour of SLIT. Based on a cost difference between the two groups of £2668, an ICER of £7848 per unit improvement in RQLQ was obtained. Table 50 also shows that between 3 and 10 years after the start of IT, the cost difference between SLIT and ST reduces from £2668 to £1820. Over the same period, QALY gains favouring SLIT increase from 0.0440 to 0.1305. The ICERs based on QALYs gained over this period therefore reduce from £60,704 to £13,951. An ICER of £27,269 was obtained 6 years after the start of IT. This reduction in costs per QALY gained is shown graphically in Figure 21.

TABLE 50 - Economic evaluation results: SLIT vs ST (based on direct comparisons)

Cost difference (£) at
3 years	2668
4 years	2534
5 years	2405
6 years	2280
7 years	2159
8 years	2042
9 years	1929
10 years	1820
RQLQ difference	0.340 (95% CI 0.180 to 0.490)
ICER (£): costs/unit improvement in RQLQ	7848 (95% CI 5445 to 14,823)
Total QALY gain in 3 years	0.0440 (95% CI 0.0233 to 0.0633)
ICER (£): costs per QALY	60,704 (95% CI 42,121 to 114,663)
Total QALY gain in 4 years	0.0576 (95% CI 0.0305 to 0.0830)
ICER (£): costs per QALY	43,977 (95% CI 30,514 to 83,067)
Total QALY gain in 5 years	0.0708 (95% CI 0.0305 to 0.0830)
ICER (£): costs per QALY	33,948 (95% CI 23,556 to 64,124)
Total QALY gain in 6 years	0.0836 (95% CI 0.0443 to 0.1205)
ICER (£): costs per QALY	27,269 (95% CI 18,921 to 51,508)
Total QALY gain in 7 years	0.0959 (95% CI 0.0508 to 0.1383)
ICER (£): costs per QALY	22,504 (95% CI 15,615 to 42,508)
Total QALY gain in 8 years	0.1078 (95% CI 0.0571 to 0.1554)
ICER (£): costs per QALY	18,935 (95% CI 13,139 to 35,767)
Total QALY gain in 9 years	0.1194 (95% CI 0.0632 to 0.1720)
ICER (£): costs per QALY	16,164 (95% CI 11,216 to 30,532)
Total QALY gain in 10 years	0.1305 (95% CI 0.0691 to 0.1880)
ICER (£): costs per QALY	13,951 (95% CI 9680 to 26,351)

FIGURE 21.

Cost-effectiveness of SLIT vs ST (direct comparisons).

Table 51 shows the results comparing SCIT with ST. The mean change in RQLQ in favour of SCIT was 0.740 (95% CI 0.56 to 0.92) and this resulted in an ICER of £7483 per unit improvement in RQLQ (based on a cost difference of £5537). Between 3 and 10 years after starting IT, the cost difference between SCIT and ST reduced from £5537 to £5198, whereas the QALY gains in favour of SCIT increased from 0.0957 to 0.2840. Consequently, the cost per QALY gained between the same time period reduced from £57,883 to £18,304 as shown in Figure 22. The value of this ICER was £29,579 6 years after the start of IT.

TABLE 51 - Economic evaluation results: SCIT vs ST (based on direct comparisons)

Cost difference (£) at
3 years	5537
4 years	5484
5 years	5432
6 years	5382
7 years	5334
8 years	5287
9 years	5242
10 years	5198
RQLQ difference	0.740 (95% CI 0.560 to 0.920)
ICER (£): costs/unit improvement in RQLQ	7483 (95% CI 6019 to 9889)
Total QALY gain in 3 years	0.0957 (95% CI 0.0724 to 0.1189)
ICER (£): costs per QALY	57,883 (95% CI 46,558 to 76,488)
Total QALY gain in 4 years	0.1254 (95% CI 0.0949 to 0.1559)
ICER (£): costs per QALY	43,722 (95% CI 35,168 to 57,776)
Total QALY gain in 5 years	0.1542 (95% CI 0.1167 to 0.1917)
ICER (£): costs per QALY	35,233 (95% CI 28,340 to 46,558)
Total QALY gain in 6 years	0.1820 (95% CI 0.1377 to 0.2262)
ICER (£): costs per QALY	29,579 (95% CI 23,792 to 39,087)
Total QALY gain in 7 years	0.2088 (95% CI 0.1580 to 0.2596)
ICER (£): costs per QALY	25,545 (95% CI 20,547 to 33,756)
Total QALY gain in 8 years	0.2347 (95% CI 0.1776 to 0.2918)
ICER (£): costs per QALY	22,524 (95% CI 18,117 to 29,764)
Total QALY gain in 9 years	0.2598 (95% CI 0.1966 to 0.3230)
ICER (£): costs per QALY	20,178 (95% CI 16,230 to 26,664)
Total QALY gain in 10 years	0.2840 (95% CI 0.2149 to 0.3531)
ICER (£): costs per QALY	18,304 (95% CI 14,723 to 24,188)

FIGURE 22.

Cost-effectiveness of SCIT vs ST (direct comparisons).

Indirect comparisons

Table 52 presents the results of the comparison between SLIT and ST. The mean change in RQLQ favouring SLIT was 0.247 (95% CrI −0.156 to 0.729) and this resulted in an ICER of £10,802 per unit improvement in RQLQ (based on a cost difference of £2668). The cost differences between the two groups reduced from £2668 to £1820 over a period of 3–10 years post commencement of IT. As in the direct comparison, the QALY gains favouring SLIT increased from 0.0319 to 0.0948 over the same period. This resulted in a reduction in the costs per QALY gained from £83,560 to £19,203 over the same period of time as depicted in Figure 23. The value of the ICER 6 years after the start of IT was £37,537.

TABLE 52 - Economic evaluation results: SLIT vs ST (based on indirect comparisons)

‘ST dominates’ means that ST is more effective and also cheaper.

Cost difference (£) at
3 years	2668
4 years	2534
5 years	2405
6 years	2280
7 years	2159
8 years	2042
9 years	1929
10 years	1820
RQLQ difference	0.247 (95% CI −0.1560 to 0.729)
ICER (£): costs/unit improvement in RQLQ	10,802 (3660 to ST dominates)
Total QALY gain in 3 years	0.0319 (95% CI −0.0202 to 0.0942)
ICER (£): costs per QALY	83,560 (28,312 to ST dominates)
Total QALY gain in 4 years	0.0419 (95% CI −0.0264 to 0.1236)
ICER (£): costs per QALY	60,535 (20,510 to ST dominates)
Total QALY gain in 5 years	0.0515 (95% CI −0.0325 to 0.1519)
ICER (£): costs per QALY	46,730 (15,833 to ST dominates)
Total QALY gain in 6 years	0.0607 (95% CI −0.0384 to 0.1792)
ICER (£): costs per QALY	37,537 (12,718 to ST dominates)
Total QALY gain in 7 years	0.0697 (95% CI −0.0440 to 0.2057)
ICER (£): costs per QALY	30,977 (10,496 to ST dominates)
Total QALY gain in 8 years	0.0783 (95% CI −0.0495 to 0.2312)
ICER (£): costs per QALY	26,065 (8831 to ST dominates)
Total QALY gain in 9 years	0.0867 (95% CI −0.0548 to 0.2559)
ICER (£): costs per QALY	22,250 (7539 to ST dominates)
Total QALY gain in 10 years	0.0948 (95% CI −0.0599 to 0.2798)
ICER (£): costs per QALY	19,203 (6506 to ST dominates)

FIGURE 23.

Cost-effectiveness of SLIT vs ST (indirect comparisons).

In the comparison between SCIT and ST (Table 53), the mean change in RQLQ favouring SCIT was 0.764 (95% CrI 0.425 to 1.116). Based on a cost difference of £5537, an ICER of £7248 per unit improvement in RQLQ was found. As in the direct comparisons, the cost difference between the two groups reduced over a period of 8 years (3–10 years after the start of IT) from £5537 to £5198. The QALY gains in favour of SCIT increased from 0.0988 to 0.2932 over the same period. As depicted in Figure 24, the resultant ICERs therefore decreased from £56,064 to £17,729 over this period. The ICER, 6 years after the start of IT, was £28,650 per QALY gained.

TABLE 53 - Economic evaluation results: SCIT vs ST (based on indirect comparisons)

Cost difference (£) at
3 years	5537
4 years	5484
5 years	5432
6 years	5382
7 years	5334
8 years	5287
9 years	5242
10 years	5198
RQLQ difference	0.764 (95% CI 0.425 to 1.116)
ICER (£): costs/unit improvement in RQLQ	7248 (95% CI 4962 to 13,029)
Total QALY gain in 3 years	0.0988 (95% CI 0.0549 to 0.1443)
ICER (£): costs per QALY	56,064 (95% CI 38,381 to 100,784)
Total QALY gain in 4 years	0.1295 (95% CI 0.0720 to 0.1892)
ICER (£): costs per QALY	42,349 (95% CI 28,992 to 76,128)
Total QALY gain in 5 years	0.1592 (95% CI 0.0885 to 0.2325)
ICER (£): costs per QALY	34,126 (95% CI 23,362 to 61,347)
Total QALY gain in 6 years	0.1879 (95% CI 0.1045 to 0.2744)
ICER (£): costs per QALY	28,650 (95% CI 19,613 to 51,502)
Total QALY gain in 7 years	0.2156 (95% CI 0.1199 to 0.3149)
ICER (£): costs per QALY	24,743 (95% CI 16,939 to 44,479)
Total QALY gain in 8 years	0.2423 (95% CI 0.1348 to 0.3540)
ICER (£): costs per QALY	21,816 (95% CI 14,935 to 39,218)
Total QALY gain in 9 years	0.2682 (95% CI 0.1492 to 0.3918)
ICER (£): costs per QALY	19,544 (95% CI 13,380 to 35,133)
Total QALY gain in 10 years	0.2932 (95% CI 0.1631 to 0.4283)
ICER (£): costs per QALY	17,729 (95% CI 12,137 to 31,871)

FIGURE 24.

Cost-effectiveness of SCIT vs ST (indirect comparisons).

The last comparison was between SCIT and SLIT (Table 54). The mean change in RQLQ (favouring SCIT) was 0.517 (95% CrI −0.0710 to 1.045) resulting in an ICER of £5550 per unit improvement in RQLQ, based on a cost difference of £2869. The cost differences between the two groups reduced from £2869 to £3378 over a period of 3–10 years following commencement of IT. QALY gains favouring SCIT increased from 0.0668 to 0.1984 over the same period. This resulted in a reduction in the costs per QALY gained from £42,928 to £17,025 over the same period of time as depicted in Figure 25. The value of the ICER, 6 years after the start of IT, was £24,404.

TABLE 54 - Economic evaluation results: SCIT vs SLIT (based on indirect comparisons)

‘ST dominates’ means that ST is more effective and also cheaper.

Cost difference (£) at
3 years	2869
4 years	2950
5 years	3027
6 years	3102
7 years	3175
8 years	3245
9 years	3312
10 years	3378
RQLQ difference	0.517 (95% CI −0.0710 to 1.045)
ICER (£): costs/unit improvement in RQLQ	5550 (2746 to SLIT dominates)
Total QALY gain in 3 years	0.0668 (95% CI −0.0092 to 0.1351)
ICER (£): costs per QALY	42,928 (21,238 to SLIT dominates)
Total QALY gain in 4 years	0.0876 (95% CI −0.0120 to 0.1771)
ICER (£): costs per QALY	33,661 (16,653 to SLIT dominates)
Total QALY gain in 5 years	0.1077 (95% CI −0.0148 to 0.2177)
ICER (£): costs per QALY	28,105 (13,904 to SLIT dominates)
Total QALY gain in 6 years	0.1271 (95% CI −0.0175 to 0.2569)
ICER (£): costs per QALY	24,404 (12,074 to SLIT dominates)
Total QALY gain in 7 years	0.1459 (95% CI −0.0200 to 0.2948)
ICER (£): costs per QALY	21,764 (10,768 to SLIT dominates)
Total QALY gain in 8 years	0.1640 (95% CI −0.0225 to 0.3315)
ICER (£): costs per QALY	19,787 (9789 to SLIT dominates)
Total QALY gain in 9 years	0.1815 (95% CI −0.0249 to 0.3668)
ICER (£): costs per QALY	18,251 (9030 to SLIT dominates)
Total QALY gain in 10 years	0.1984 (95% CI −0.0272 to 0.4010)
ICER (£): costs per QALY	17,025 (8423 to SLIT dominates)

FIGURE 25.

Cost-effectiveness of SCIT vs SLIT.

Six-weekly treatment schedule

Using a 6-weekly maintenance schedule instead of a 4-weekly one resulted in lower costs (fewer clinic visits and fewer injections) (Table 55). The effectiveness was assumed to be the same, therefore the cost-effectiveness was increased. The differences in ICERs (at 6 years) between the two maintenance schedules are shown below (full details are shown in Appendix 12).

TABLE 55 - Incremental cost-effectiveness ratios for 4- and 6-week maintenance schedules (at 6 years)

Comparison	Alutard maintenance
	4-weekly	6-weekly
SCIT vs ST (direct comparisons)	£29,579	£21,599
SCIT vs ST (indirect comparisons)	£28,650	£20,920
SCIT vs SLIT	£24,404	£12,982

NHS perspective only

Costs were considered from the NHS and patient perspectives because of the burden that AR places on both the NHS and patients. For purposes of reimbursement, however, only NHS costs are important. We therefore also conducted a CEA (results not shown) based on NHS costs only, and the results were not altered significantly. IT (when compared with ST) was found to be cost-effective around 7 years after the start of treatment (1 year later than when productivity costs are included). The only exception was the comparison between SLIT and ST based on indirect comparison results where this threshold increased to more than 10 years after the start of IT. In the comparison between SCIT and SLIT, SCIT was shown to be more cost-effective as early as 4 years after the start of treatment (1 year earlier than when productivity costs are included).

Interpretation of the cost-effectiveness results

Staff costs for SCIT were higher compared with those for SLIT owing to the greater number of clinic visits made by patients in this group over a 3-year period, i.e. 46 compared with 13. The costs of SCIT medication per unit were also higher than those for SLIT thereby driving the costs for SCIT further up. As would be expected, SCIT was associated with higher productivity losses owing to the nature of treatment that warranted absence from work. Individuals in the ST group had highest overall productivity losses. This finding is consistent with those found in other studies. 228,230,231 Costs associated with asthma and AE medication were not included in the analysis because the cost differences in medications between SCIT, SLIT and ST were assumed to be negligible. This was based on expert clinical opinion and a review of the literature.

In the results based on direct and indirect comparisons, SCIT and SLIT were both found to be more effective than ST as shown by the difference in RQLQ. The results from the indirect comparison also suggest that SCIT may be more effective than SLIT; however, this is associated with uncertainty (a non-significant result) and must be interpreted cautiously (see Chapter 3, Indirect comparison of subcutaneous immunotherapy versus sublingual immunotherapy). As the QALYs used in this analysis were based on an algorithm that mapped the RQLQ to EQ-5D, the same direction of effect observed in the RQLQ was also seen in the QALY gains.

In terms of cost per unit improvement in RQLQ, ICERs of £7848 and £10,802 were obtained in the comparison between SLIT and ST, whereas ICERS of £7483 and £7248 were estimated for the comparison between SCIT and ST. As IT in the two sets of comparisons was both more costly and effective, it would be considered to be cost-effective only if decision-makers were willing to pay at least £11,000 for each unit improvement in RQLQ. For the comparison between SCIT and SLIT, SCIT would be considered a more cost-effective alternative if decision-makers were willing to pay at least £5600 for a similar improvement in RQLQ. A unit change of 0.5 may be considered clinically significant.

Cost-effectiveness results based on costs per QALY gained can be assessed against a threshold of £20,000–30,000, the conventional threshold adopted by decision-makers in the UK NHS, such as NICE. 249 The results of the analysis show that IT, when compared with ST, becomes cost-effective around 6 years after the start of treatment (7 years for NHS perspective only). The only exception is the comparison between SLIT and ST based on indirect comparisons, where this threshold increased to 7 years (10 years for NHS perspective only). In the comparison between SCIT and SLIT, SCIT was shown to be more cost-effective as early as 5 years after the start of treatment (4 years for NHS perspective only). These results are consistent with those for studies that reported outcomes in terms of ICERs (shown in Table 39), although the ICERs in our study were much higher.

In view of the many simplifications required in performing the CEA, these results must be regarded as indicative. The results using direct comparisons suggest that either SLIT or SCIT may be cost-effective compared with symptomatic treatment (ST), applying usual UK standards of cost-effectiveness. However, this tentative conclusion depends on there being a good reason to believe that clinical effectiveness will be sustained for somewhat longer than the 3 years' period following cessation of treatment.

When the results from the indirect comparisons were used, the cost-effectiveness results for SCIT compared with ST were largely unchanged. For SLIT compared with ST, however, the results for a difference in RQLQ were no longer statistically significant at the 95% confidence level. It is clear that SLIT is more costly than ST. Accordingly, there is no finite upper confidence limit for the ICER: the CI stretches into the region where ST dominates (is less costly and more effective than) SLIT. Figure 26 illustrates this point. It is shown in terms of mean improvement in RQLQ, but exactly the same principles apply to the results when effectiveness is estimated in QALYs.

FIGURE 26.

Incremental cost-effectiveness plane for SLIT vs ST using the results of the indirect comparison. The incremental cost is fixed at approximately £2700 per patient. The points L, M and U show this incremental cost combined with the lower, mean and upper limits of the CI for improvement in RQLQ, respectively. The gradient of the line from the origin to M is the point estimate of the ICER. The gradient of the line from the origin to U is the lower confidence limit of the ICER. The dashed line from the origin to L has a negative gradient indicating that, if L were the true representation of the cost and effect difference, ST would dominate (be less costly and more effective than) SLIT. In such a case, the magnitude of the slope of the dashed line is of no importance: ST would be preferred in any case. It should also be noted that the negative ICER represented by point L should be considered as higher than all positive ICERs, not lower. For the reasons given in the preceding sentences, it is not appropriate to quote a numerical upper confidence limit for the ICER, and the statement ‘ST dominates’ has been placed instead of such a limit in the results tables.

Using the point estimate of difference in effectiveness based on RQLQ, SCIT appears to be cost-effective compared with SLIT. However, as for SLIT compared with ST, the CI for difference in effectiveness crosses zero, so again there is no finite upper limit for the CI in the ICER.

It is acknowledged that there is considerable relative uncertainty in the inputs concerning costs relating to symptomatic medication and productivity loss. However, these form a sufficiently small part of the overall cost difference between treatments that making plausible changes to those values would not make any appreciable difference to the results quoted. Formal sensitivity analysis on these inputs has not been carried out, as such analysis would add nothing of value to the illustrative results already quoted, and risks being quoted out of context as giving some spurious indication of the robustness of the results. In particular, it is not possible, on the evidence available to the research team, to produce a meaningful estimate of the probability that each treatment is cost-effective at any given threshold ICER, or the value of perfect information at any such threshold.

Repeating the CEA using a 6-weekly (rather than 4-weekly) maintenance schedule for Alutard resulted in slightly lower ICERs. This was based on an assumption that clinical effectiveness remained the same and was undertaken only to illustrate the potential impact of a reduction in costs. The use of shorter treatment courses, such as preseasonal treatment using Pollinex, is likely to reduce costs even further, but there is uncertainty around the long-term effectiveness.

No analysis based on RQLQ was possible for children. One study that reported a paediatric version of the RQLQ was identified;152 however, differences compared with the adult RQLQ meant that equivalent mapping to the EQ-5D could not be undertaken.

One of the limitations of the CEA is that it does not take into account any health benefits or potential cost savings from future cases of asthma prevented. This would be particularly relevant when considering the treatment of children with SAR, as they are more likely to develop asthma than children without SAR. A simple calculation based on costs for SCIT and SLIT as outlined above, and number of asthma cases avoided based on the PAT study54–56 (for SCIT) and Novembre et al. 57 (for SLIT) is presented below [see also Chapter 1, The role of specific (allergen) immunotherapy in asthma prevention]. Both studies are in children with SAR, with no asthma at the start of treatment. The numbers needed to treat (NNT) to prevent one case of asthma were derived from the number of patients with and without asthma in the respective treatment arms, and CIs were calculated around these (method given in Armitage et al. 256). The NNT was multiplied by the cost difference (at 3 years) between SCIT (or SLIT) and symptomatic treatment only (Table 56).

TABLE 56 - Cost per asthma case avoided

Study	NNT	Cost difference IT and ST (£)	Cost (£) per asthma case avoided (95% CI)
SCIT: PAT study,55 3-year data (n = 205)	5	5537	27,500 (15,598 to 104,660)
SCIT: PAT study,56 5-year data (n = 183)	4	5537	22,000 (14,374 to 68,807)
SCIT: PAT study,54 10-year data (n = 147)	5	5537	27,500 (14,745 to 183,571)
SLIT: Novembre et al. 2004,57 3-year data (n = 113)	4	2668	10,627 (6348 to 63,004)

The results suggests that if each case of asthma costs over £10,627 (SLIT) or over £22,000–27,500 (SCIT) in treatment costs over a lifetime (appropriately discounted) then SLIT or SCIT could be potentially cost saving compared with symptomatic treatment. These results should be seen as indicative only, as they are based on effectiveness data from relatively small open-label studies. Health benefits and costs associated with a reduction in SAR symptoms are not considered in these calculations.

What is required as a top priority is to establish the extent to which the data already collected in past primary research can be used to populate a useful cost-effectiveness model, which captures all relevant benefits and costs. This means making the data available to independent researchers who will be able to analyse it using a common, and economically useful, framework. Only when this has been done will it be possible to carry out a more meaningful analysis, possibly using a value of information framework, to determine whether or not further primary research is worthwhile and to set the priorities for such research.

Main findings

A total of 28 new DBPC RCTs of SCIT (n = 17) or SLIT (n = 11) compared with placebo for SAR were identified, bringing the total number of relevant RCTs in this area to around 128. Assessment of the risk of bias was hampered by a lack of reporting of all relevant criteria, but, while risk of bias was often unclear, there were very few instances of high risk of bias. Overall, results were unlikely to be affected by the studies reporting instances of high risk of bias.

The updated findings are consistent with the Cochrane reviews and find statistically significant benefits for both SCIT compared with placebo and SLIT compared with placebo across all outcome measures and for the majority of subgroup analyses.

In trials of SCIT for SAR, the total number of studies and participants included in the present review was only slightly higher than those for the earlier Cochrane review, and results of meta-analyses remained very similar, with moderate effect sizes in favour of SCIT for all outcomes. Greater improvements in both SSs and MSs, compared with placebo,were found with increasing vaccine allergen content, consistent with the recognised dose–response relationship in SCIT. Note that this was based on non-randomised comparisons from subgroup analyses and should therefore be interpreted cautiously. It was beyond the scope of this report to look at randomised comparisons of allergen content, although such comparisons exist. There was only one small trial152 of SCIT in children and this found significantly lower SSs and MSs, and improved QoL, in the actively treated group (after 3 years of treatment).

Consistent with previous literature, randomised, double-blind trials of SLIT show efficacy in all major outcomes compared with placebo. Few differences were found compared with the earlier Cochrane review, despite restriction to SAR and increased sample sizes in many cases. However, a number of previously non-significant results reached statistical significance in the present review – namely, SSs in studies with < 5 μg of MAC, MSs in studies with > 20 μg of MAC, and MSs in ragweed allergy. Only one previously significant result became non-significant: of the five studies in Parietaria allergy included in the Cochrane review, removal of the study by Pajno et al. 35 (which was restricted to SAA) resulted in loss of significance; however, all of these studies had very small sample sizes (total number in remaining studies was 124).

Perhaps more importantly, despite a small increase in total sample size, and limitation to SAR, MSs in children still failed to show a significant improvement with active treatment compared with placebo. The clinical significance of this finding is unclear. The MS in isolation may not be able to reflect the effectiveness of SLIT, which is why the combined SMS is a preferred measure. It is also possible that medication use differs in children compared with adults, as it may be influenced by parental preferences. Another possibility is that SLIT is less effective in children than in adults.

Consistent with this hypothesis, the pooled SMD in SSs in children decreased by over 50% compared with the Cochrane review of SLIT. However, the result did remain statistically significant in favour of SLIT. The Cochrane review included studies of both SAR and PAR, and subgroup analysis of SAR studies only (while maintaining very similar participant numbers of > 1300 children) appears to have been associated with a reduction in treatment effect. It may be possible that SLIT treatment is more appropriate for perennial allergens in paediatric populations than for seasonal allergens. Consistent with this, exploratory subgroup analysis of SSs from paediatric studies in PAR only indicates a much larger effect size than for SAR (combined SMD −0.89 vs −0.24, respectively). Although the findings in the PAR studies were more variable than those in SAR, and the pooled result failed to reach statistical significance, this was based on a much smaller sample size overall (total n = 328, compared with n = 1343 for SAR).

None of the trials in children included in this review take into consideration future benefits from SLIT, such as avoidance of asthma or new sensitisations, although there are trials currently under way to evaluate this (see Chapter 3, Ongoing trials). Ten-year data from the PAT study54 (open-label RCT) suggests that the incidence of asthma in the treatment arm was lower than that of the control arm. Further follow-up of this study is ongoing.

In contrast with SCIT, no relationship between increased MAC and effect size was apparent from subgroup analyses, but this finding may be related to different sample sizes or due to other sources of heterogeneity in the allergen content subgroups. The estimate for the low-allergen group had wider CIs and is associated with more uncertainty. Again, these findings are not based on randomised groups and should thus be interpreted cautiously.

Overall, both SLIT and SCIT resulted in statistically significant improvements in QoL scores. Although SCIT improved RQLQ scores by 0.74 points compared with placebo, SLIT resulted in only a 0.31-point improvement compared with placebo, despite both of these meta-analyses including a similar number of participants (955 compared with 924, respectively).

All of these studies were conducted in adults (alternative versions of the RQLQ have been validated for use in paediatric and adolescent populations) and the findings are consistent with those for SSs and MSs, suggesting a greater clinical benefit from SCIT, at least in adults.

Quality-of-life data in children are scarce; eight studies of SCIT25,102,112,189,191,192,200,202 and 11 of SLIT142–144,148,152,156,161,164–166,173 reported QoL outcomes (not all included in the meta-analyses), but only one study in each intervention was restricted to paediatric or adolescent152,189 participants. SLIT is more commonly prescribed than SCIT in this population, largely due to a perceived reduced risk of potentially severe AEs. Given the possibility that efficacy of SLIT in children is poorer in terms of SSs and MSs than in adults, the presence or absence of benefit in terms of QoL in children should be further explored. In addition, more studies using the paediatric or adolescent version of the RQLQ are needed in order that clinical as well as statistical significance may be assessed. Nevertheless, both studies (one for SCIT and SLIT, respectively) found statistically significant benefits in terms of QoL in children with active treatment.

The overall incidence of AEs following treatment with SCIT was slightly higher than for SLIT (79% vs 65%, respectively, experienced at least one AE); however, as SCIT is administered under clinical supervision, reporting of AEs could be expected to be more stringent. With both routes of administration, the majority of AEs were local reactions at the site of injection (SCIT) or in the oral cavity or gastrointestinal system (SLIT), and resolved spontaneously without treatment. Where severity of systemic AEs was reported, most were graded as mild or moderate, and, again, many did not require special treatment; however, 19% of systemic AEs occurring following SCIT treatment were graded as severe, compared with only 2% following SLIT treatment.

Based on six trials155,166,179,181,185,257 that reported rates of events per injection, 129 systemic reactions occurred following 2909 administrations of active SCIT (4.4%). This number is much higher than the 0.2% suggested by a recent review of the literature. 258 However, that review included studies of both seasonal, perennial and venom IT, IT for other atopic conditions besides AR, and reports from trials, retrospective studies, surveys, and clinical observations. It seems likely that the strict reporting requirements of RCTs may provide a more realistic representation of the true rate of systemic reactions. Similar incidence-per-dose data were not available from the studies of SLIT included in the present review. No fatalities as a result of treatment were reported for either SCIT or SLIT.

Funnel plot evaluations of SSs for both SCIT and SLIT trials showed slight evidence of plot asymmetry (see Appendix 13), with larger effect sizes tending to be associated with smaller trial size. Possible sources of asymmetry include publication bias or poorer methodological quality in smaller studies, sampling variation or chance,259 and these findings should be interpreted with caution. Further, not all studies reported large effect size estimates in favour of the active treatment, and comparison of combined SMD using both fixed-effect and random-effect meta-analyses identified little difference between the two methods. Thus, any small-study effects are unlikely to impact significantly on the overall effect estimates for the interventions. 259

In contrast to the large number of placebo-controlled trials, only one small double-blinded head-to-head RCT of SCIT compared with SLIT was identified; this study did not find a significant difference between the two types of treatment, although both were better than placebo. Given the paucity of this type of data, an indirect comparison was conducted.

As there was some evidence of heterogeneity both within and between sets of placebo-controlled trials, ICMR with various covariates was performed in order to explore and potentially reduce heterogeneity. However, adjusting for type of allergen, allergen content and duration of treatment did not substantially reduce heterogeneity. Statistically significant findings favouring SCIT over SLIT (for SSs and MSs, not adjusted for covariates) were associated with substantial heterogeneity. In the analysis using combined SMS, arguably a preferred outcome measure, none of the differences in adjusted and non-adjusted analyses were statistically significant, but were associated with reduced heterogeneity. The difference in RQLQ score (0.517; 95% CrI −0.071 to 1.045) was also found to be not statistically significant. Many of the best estimate probabilities found that SCIT was most likely to be the best treatment, but this needs to be interpreted in the context of the standardised score difference results.

Year of publication appeared to account for a degree of heterogeneity (for SSs). Analyses were suggestive of earlier studies showing greater benefit for SCIT compared with studies published at a later date finding less benefit. It seems unlikely that SCIT has become less effective over time, and alternative explanations for the apparent reduction in effectiveness of SCIT include improved trial protocols and reduced dosages of SCIT being administered for safety reasons. Another possibility is that with increasingly stringent requirements for registration of clinical trials, the potential for publication bias may have reduced over time. However, these results are based on use of a post hoc defined variable and must be considered exploratory in nature.

The use of standardised units makes interpretation difficult, and where significant differences were found favouring SCIT, it is difficult to gauge how much of a clinical difference this would make.

A strength of this review is that a robust review methodology was used, including a comprehensive search strategy. It is unlikely that many relevant studies were missed in this update.

One limitation of both the Cochrane reviews and this review is that many studies do not contribute to the direct and indirect comparison meta-analyses. In this review only 5 out of 17 SCIT and 5 out of 11 SLIT RCTs contributed to at least one meta-analysis. This was partly because different outcome measures were used or that data were not presented in a way that was suitable for meta-analysis. Findings from any studies not represented in a meta-analysis were presented and were found to be consistent with the meta-analyses in their findings (benefit from IT over placebo). By far the largest of the new trials identified for SCIT compared with placebo (DuBuske et al. ,145 n = 1028) did not contribute to meta-analyses, as outcomes were expressed as number of ‘well-days’ and ‘bad-days’. This study145 evaluated the effect of an ultrashort course of SCIT and found significant benefits for SCIT over a 4-week period; longer-term outcomes were not reported.

A major limitation is the inconsistent use of outcome measures across studies [see Chapter 1, Outcome measures in randomised controlled trials of specific (allergen) immunotherapy, for further details]. Although almost all studies use the same four-point scale to assess symptom severity, the number and type of symptoms measured in different studies is so diverse as to preclude any useful comparison of results (see Appendix 7). The inconsistency in choice of outcome measure is also reflected in MSs and combined SMSs. A consequence of the highly variable outcome reporting across studies is that the pooled summary measures from meta-analyses can only be reported as SMDs, which are difficult to interpret clinically. Although effect sizes can be classified as small, moderate or large, these do not necessarily correspond to clinically meaningful changes. So, although it can be stated with some certainty that allergen IT shows consistent benefit, and that pooled results are statistically significant, there is no good estimate of the proportion of patients who, for example, have more ‘well-days’ or fewer ‘worst-days’.

The COMET (Core Outcome Measures in Effectiveness Trials) Initiative260 is attempting to develop standardised sets of outcomes (‘core outcomes’) that represent the minimum that should be measured and reported in all clinical trials of a specific condition, although this does not mean that outcomes have to be restricted to only core ones. Ultimately, this would make it easier for the results of trials to be compared and combined. Similarly, guidelines on AE reporting23,134,261 should be followed more consistently.

A further limitation was the inclusion of studies of variable methodological study design or risk of bias. This is in addition to the clinical heterogeneity observed. We did not use a quality threshold for including studies, as quality criteria were often poorly reported and inclusion would therefore have been on the basis of reporting rather than actual quality. Contacting all study authors would have been beyond the scope of this report.

The impact of IT in patients with severe, uncontrolled SAR was of particular interest when conducting this review; however, this information was not always included in the published reports of trials, and it is conceivable that there was a degree of variation in severity across trials. The concept of ‘severe chronic upper airway disease’ has recently been proposed;262 this defines patients with uncontrolled AR despite adequate pharmacological treatment based on guidelines, and could potentially be used as an inclusion criterion for future trials.

It was beyond the scope of this review to address the issue of optimum dosing and treatment regimen (see Chapter 1, Treatment schedule, for further details). Similarly, the review did not address issues around the efficacy of different SIT products [e.g. differences in depot formulations, modification of the allergen extract with adjuvants, use of allergen fragments; see Chapter 1, Specific (allergen) immunotherapy formulations and Non-standard therapies]. Routes of administration other than sublingual or subcutaneous were also not explored (e.g. epicutaneous, intralymphatic).