Our meta-analyses confirmed the presence of an effect on symptoms for all drug classes (or single drugs, for MP29-02 and montelukast), including grass pollen SLIT tablets. The Hedges' g values calculated here were compatible with those found in the literature data. For example, we found a value of 0.39 for the antihistamines as a class; this may be compared with Compalati et al.'s value of 0.42 for fexofenadine , Mösges et al.'s values of 0.59 for levocetirizine and 0.21 for loratadine , and Compalati and Canonica's value of 0.37 for rupatadine . In contrast, Matricardi et al. calculated a value of 1.00 for antihistamines as a class; however, the latter meta-analysis included a number of small studies with large effect sizes . The values for montelukast are very consistent: 0.23 in the present study and 0.24 according to both Matricardi et al.  and Rodrigo et al. . Lastly, we calculated a Hedges' g of 0.55 for corticosteroids as a class; Matricardi et al.'s value for mometasone was 0.47 . These similarities indicate that our selected studies form a valid basis for further analysis (that is, calculation of the RCI).
The RCI of SLIT is as large as that of nasal corticosteroids
We studied the degree of symptom relief (relative to placebo) provided by recently approved symptomatic medications and tablet formulations of SLIT products. Despite mechanistic differences in the mechanism of action of these two treatment approaches, the current evidence from the recent, well-powered, stringent, clinical studies analyzed here suggests that grass pollen SLIT tablets provide a greater degree of symptom relief in SAR than certain symptomatic drugs or drug classes (such as the leukotriene receptor antagonist montelukast and second-generation H1-antihistamines) and much the same degree of relief as nasal corticosteroids and an azelastine-fluticasone combination. This finding is especially striking because a number of methodological factors reduce the apparent magnitude of effect in AIT clinical trials. It is problematic to compare the symptomatic medication RCIs calculated in the present study with mean values in the meta-analyses performed and reported by Wilson et al. (-18% for nasal corticosteroids, -7% for oral antihistamines and -5% for montelukast ) and Benninger et al. (-40.7% for nasal corticosteroids and -23.5% for oral antihistamines ) because the latter studies used a different calculation method. However, the order of these drug classes in each analysis is consistent (nasal corticosteroids > oral antihistamines > montelukast). It should be borne in mind that within a given symptomatic drug class, members may differ in their efficacy and tolerance. Using a therapeutic index score, Schäfer et al. suggested that there were differences between intranasal corticosteroids . Likewise, a meta-analysis from open-label prospective observational studies performed by Mösges et al. suggested that levocetirizine is significantly more effective than desloratadine, ebastine and fexofenadine . However, a number of the studies analyzed in the present paper or in the literature made head-to-head comparisons between marketed nasal corticosteroids or between antihistamines. The differences in efficacy were either not significant or were inconsistent from one study to another [33, 34, 44, 50, 51].
Calculation of the RCI
The World Allergy Organization's definition of the RCI (as applied by Matricardi et al. ) is based solely on the relative mean active treatment versus placebo difference in scores calculated over a defined period (usually the pollen season as a whole for the SLIT studies) . The method is easily applicable to SLIT or SCIT trials lacking a low, pre-season baseline score but its use in short-term pharmacotherapy trials (in which a high, peak-season baseline score is available) can be criticized. Indeed, we excluded a number of pharmacotherapy trials in which the RCI was estimated as the percentage difference between reductions in scores (that is, without reporting the baseline and final scores). In the absence of head-to-head SLIT tablet versus pharmacotherapy trials (which would be difficult to design, implement and interpret), we believe that the RCI affords a meaningful comparison.
Methodological differences in the clinical assessment of SLIT products versus symptomatic medications
It was only in 2009 that the European Medicines Agency's guideline on the clinical development of AIT products  came into force after being released as a draft for consultation in 2007. The recent clinical development of tablet SLIT formulations has closely followed these evidence-based guidelines. However, several methodological factors decrease the apparent RCI for SLIT and, conversely, increase the apparent RCI for symptomatic medications (below and Additional file 4: Table S4).
Total versus partial symptom scores
The etiological nature of AIT means that clinical trials in this field generally use total symptom scores, in which nasal symptoms, ocular symptoms and (sometimes) other local or individual parameters (coughing, wheezing, ear itch, and so on) are taken into account. In symptomatic medication trials and depending on the drug's pharmacological action, certain symptoms are sometimes excluded from the efficacy scores. The failure to score individual symptoms that are at least partly treatment-refractive (for example, nasal symptoms for antihistamines) may thus prompt overestimation of the RCI for some symptomatic medications.
Rescue medication use
For ethical reasons, rescue medications cannot be prohibited in month- or year-long SLIT or SCIT trials. The experimental data (that is, medication scores) show that rescue medication use is greater in placebo groups than in active treatment groups. This factor reduces the difference in mean symptom scores between the SLIT and placebo groups and thus leads to underestimation of the RCI for SLIT products. Overall, there were few marked differences between the RCIs calculated from symptom scores and those calculated from combined scores in a given trial (respectively, -27.4% and -30.8% for Didier et al. , -27.9% and -26.0% for Wahn et al. , and -18.3% and -20.05% for Nelson et al. , for example).
Trial design and duration, patient recruitment, randomization and baseline scores
AIT products and pharmacotherapy products differ markedly in terms of the typical study period in SAR. Symptomatic medication trials typically evaluate symptom relief over a two-week period during the pollen season. In contrast, the efficacy of SLIT (and indeed SCIT) is studied over a whole pollen season (up to two months).
The mean disease severity in SLIT (and SCIT) trials is usually lower than in symptomatic medication trials, for two main reasons: trial duration and patient recruitment. Firstly, allergen exposure (and thus disease severity) in SLIT trials will fluctuate over the month- or year-long study period, giving peaks and troughs of disease activity. In contrast, symptomatic medications are tested over short periods at or around the pollen peak, when disease severity is high and highly symptomatic patients can be easily recruited. In a SLIT trial, treatment is initiated before the start of the expected pollen season (that is, when patients are asymptomatic). Hence, investigators can never be sure that randomized patients will actually be symptomatic during the coming study. This limitation ‘dilutes’ the level of disease severity. This SLIT versus symptomatic drug difference can be exemplified by estimating the relative SAR severity in the placebo group. In the trials selected in the present meta-analysis, we expressed the mean symptom score as a percentage of the maximum possible symptom score. In the SLIT trials, the mean (range) relative SAR severity score in the placebo group was 24.7% (18% to 34%). In the symptomatic medication trials, the mean (range) relative SAR severity score was 48.7% (30% to 67%) in antihistamine trials, 52.7% (20% to 65%) in nasal corticosteroid trials, 62.2% (58% to 64%) in montelukast trials and 66.5% (64% to 68%) in trials of the azelastine-fluticasone combination.
In SLIT trials, there are several lines of evidence to suggest that greater mean disease severity in patients is associated with a greater RCI (and, conversely, that low mean disease activity in patients reduces the apparent RCI). It is possible to identify high-severity patients within a SLIT trial, so that this subpopulation can then be more fairly compared with high-severity patients in symptomatic medication trials. Firstly, Bufe et al.'s study  in a pediatric population found RCIs of -24% for the grass pollen season as a whole, -25% for the 15-day peak grass pollen season and -28% for the ‘high-level’ grass pollen season (the period over 30 grains/m3). Hence, for SLIT products, low disease activity at the start and end of the pollen season reduces the mean RCI calculated over the season as a whole. Secondly, a novel way to focus on patients in SLIT trials with high disease burdens (thus mirroring experimental conditions in symptomatic medication trials) involves a prespecified, post hoc tertile analysis. Howarth et al. has applied this approach  to three large SLIT clinical trials [19, 59, 60]. Study centers were grouped into low, middle and high tertiles according to the average RTSS or AAdSS observed in each center's placebo patients. The high-severity tertile (in which the relative severity of SAR was 34%, rather than 27% for the placebo group as a whole) corresponds most closely to the population typically recruited in symptomatic medication trials. After calculating the RCIs on the basis of the average RTSS and the AAdSS for all three studies, Howarth et al. found that the greatest RCI was always observed in the high tertile, that is, the centers in which patients were most strongly affected by pollen . When calculated from the AAdSS for the high tertile in four studies of five-grass pollen SLIT tablets, we found that the weighted mean (range) RCI was –37.1% (-26% to -45%). Thirdly, Durham  published an analysis of ‘days with severe symptoms’ in clinical trial patients taking a timothy grass tablet. Even though this analysis was based on individual severity scores (rather than groups of centers), Durham et al. came to the same conclusion: the more severe the symptoms, the greater the clinical impact of SLIT. Recently, Durham et al.  have further shown that the size of the treatment effect over five pollen seasons in a long-term trial of timothy SLIT tablets was highly correlated with the cumulative pollen exposure at the start of the season. In particular, the SLIT versus placebo difference in the weighted rhinoconjunctivitis combined symptom and medication score increased as the pollen count increased (reaching about 33% for the highest pollen count).
An interesting question relates to whether the RCI for SLIT products and symptomatic medications changes over time during long-term use (that is, from one year or treatment season to another). The multiseason studies of grass pollen SLIT tablets provide a few indications [18, 19]. In the trial by Durham et al., the RCIs for the treatment years one, two and three (based on the RTSS) were -0.31, -0.36, and -0.29, respectively . On this basis, Durham et al. considered that the ‘reductions in rhinoconjunctivitis symptom and medication scores and the increase in quality of life and percentage symptom- and medication-free days one year after treatment were all similar to the treatment effect at the end of the three-year treatment period’ . Based on the three-season data for the AAdSS in the study by Didier et al., the respective RCI for seasons one, two and three were -0.20, -0.34, and -0.37 . However, it must be borne in mind that the mean pollen count (and, thus, the severity of disease) varied from one treatment year or season to the next. This factor is likely to be the major factor involved in the variation of the measured RCI (see below). Due to the absence of a persistent, long-term effect of symptomatic drugs, there is no reason to believe that their efficacy in SAR will change year-on-year.
In summary, an ‘unbiased’ comparison between SLIT and symptomatic medications would have to be performed with the most similar possible levels of pollen exposure and symptom severity. As things stand, one can hypothesize that trials of SLIT (generally performed in patients with mild-to-moderate symptoms) tend to underestimate the RCI for these formulations. Estimation of SLIT's effect size on the basis of the RCI observed for the high disease tertile is far from perfect. However, in the absence of robust, large-scale, head-to-head clinical trials, this tertile is an approximation of the conditions encountered in a symptomatic drug trial.
Limitations of the RCI approach
Although we restricted our selection to investigations of pollen-induced SAR, the studies of symptomatic medications (notably the H1-antihistamines) were variously performed in spring, summer and fall in patients with SAR induced by tree, grass and/or weed pollens. This is an additional source of heterogeneity. In contrast, the SLIT studies all concerned grass-pollen-induced SAR occurring in late spring/early summer. As mentioned above, SLIT trials and symptomatic medication trials differ in terms of the characteristics of the study population and the scoring systems used. The scores in SLIT trials tend to be averaged over a treatment season, whereas those in symptomatic medications trials tend to be point measurements at the end of a short treatment period. Furthermore, the RCI takes account of differences in scale because the comparison is always made with the placebo group in the same trial. In the term ‘relative clinical impact’, the word ‘relative’ means ‘the clinical impact in the active group relative to the placebo group.’ Hence, the RCI can provide a valid (albeit indirect) comparison between SLIT and symptomatic medications.