Allergen immunotherapy in MASK‐air users in real‐life: Results of a Bayesian mixed‐effects model

Abstract Background Evidence regarding the effectiveness of allergen immunotherapy (AIT) on allergic rhinitis has been provided mostly by randomised controlled trials, with little data from real‐life studies. Objective To compare the reported control of allergic rhinitis symptoms in three groups of users of the MASK‐air® app: those receiving sublingual AIT (SLIT), those receiving subcutaneous AIT (SCIT), and those receiving no AIT. Methods We assessed the MASK‐air® data of European users with self‐reported grass pollen allergy, comparing the data reported by patients receiving SLIT, SCIT and no AIT. Outcome variables included the daily impact of allergy symptoms globally and on work (measured by visual analogue scales—VASs), and a combined symptom‐medication score (CSMS). We applied Bayesian mixed‐effects models, with clustering by patient, country and pollen season. Results We analysed a total of 42,756 days from 1,093 grass allergy patients, including 18,479 days of users under AIT. Compared to no AIT, SCIT was associated with similar VAS levels and CSMS. Compared to no AIT, SLIT‐tablet was associated with lower values of VAS global allergy symptoms (average difference = 7.5 units out of 100; 95% credible interval [95%CrI] = −12.1;−2.8), lower VAS Work (average difference = 5.0; 95%CrI = −8.5;−1.5), and a lower CSMS (average difference = 3.7; 95%CrI = −9.3;2.2). When compared to SCIT, SLIT‐tablet was associated with lower VAS global allergy symptoms (average difference = 10.2; 95%CrI = −17.2;−2.8), lower VAS Work (average difference = 7.8; 95%CrI = −15.1;0.2), and a lower CSMS (average difference = 9.3; 95%CrI = −18.5;0.2). Conclusion In patients with grass pollen allergy, SLIT‐tablet, when compared to no AIT and to SCIT, is associated with lower reported symptom severity. Future longitudinal studies following internationally‐harmonised standards for performing and reporting real‐world data in AIT are needed to better understand its ‘real‐world’ effectiveness.


Conclusion:
In patients with grass pollen allergy, SLIT-tablet, when compared to no AIT and to SCIT, is associated with lower reported symptom severity. Future longitudinal studies following internationally-harmonised standards for performing and reporting real-world data in AIT are needed to better understand its 'real-world' effectiveness. Clinical Immunology 6 and Allergic Rhinitis and its Impact on Asthma (ARIA; an expert consortium issuing recommendations based on a GRADE evaluation). [7][8][9][10] While RCTs are requested for market authorisation purposes, following the formal regulation by authorities such as the European Medicines Agency, they narrow the study population based on specific criteria as pre-defined in study protocols. 11 It is unclear as to whether the effects of treatments seen in highly-controlled RCTs are similar to those in less-controlled pragmatic study designs, such as large observational studies (often referred to as 'real-world data' [RWD]). 11 Results from clinical trials should therefore be complemented with those from RWD, which can be obtained using data from electronic health records or from monitoring tools such as mobile apps.
Several retrospective studies in administrative databases have suggested the efficacy of AIT in rhinitis and asthma. 12,13 Evidence from mobile apps is more scarce, but RWD obtained from mobile apps is an increasing and demanding field, not only in the allergy domain, but also in several other chronic diseases, including different conditions such as sleep disturbances, 14 rheumatologic diseases 15 or diabetes. 16 This reflects the high potential of mobile apps for scientific purposes, patient self-management and/or adherence, as well as the encouraging results some apps have displayed in improving adherence and/ or clinical trials. 17 A recent proof-of-concept study clearly demonstrated that MASK-air ® (a mobile app with a monitoring questionnaire assessing the impact of allergic symptoms, work and medication use each day 18,19 ) is a valuable tool for assessing the impact of AIT. 20 This first analysis revealed that days under AIT are associated with approximately a 25% improved control of allergic rhinitis symptoms. Interestingly, the same magnitude of effect was observed when comparing days without symptomatic treatment versus those under monotherapy and those under co-medication. 20 However, this study did not compare the different application routes of AIT and the treatment schedules, neither did it take into account the different countries or pollen seasons. Therefore, the aim of the present study was to use MASK-air ® RWD to compare the reported control of allergic rhinitis symptoms in SCIT, SLIT and no AIT users allergic to grass pollen.

| Study design
This is a cross-sectional study using MASK-air ® data. We compared SCIT, SLIT-tablet and no AIT for the severity of reported allergic rhinitis symptoms, their impact on work and a combined symptommedication score (CSMS). We took into account the differences across users, countries and seasons, by performing analyses in which the observations were clustered by user, country and season.

| Setting
MASK-air ® was initiated in 2015 and is available in 27 countries (www.mask-air.com). 21,22 For each AIT-specific item, we included data from all MASK-air ® European countries with at least 150 days of reporting.

| Participants
We included users aged 16-90 years, who reported allergic rhinitis and allergy to grass pollen. 21 In the app, they reported whether or not they were under AIT (SCIT/SLIT). All analysed data concerned the period 21 May 2015 to 6 December 2020.

| Ethics
MASK-air ® is CE1 registered and follows the General Data Protection Regulation. 23 An independent review board approval was not required for this specific study as it is an observational study. All data were anonymised prior to the study (including geolocation-related data) using k-anonymity, and users agreed to the analysis of their data in the terms of use (translated into all languages and customised according to the legislation of each country, allowing the use of the results for research purposes).

| Data sources and variables
MASK-air ® comprises a daily monitoring questionnaire which assesses the impact of allergic rhinitis using visual analogue scales (VASs) on a 0-100 scale, which display high intra-rater validity and moderate-high validity, test-retest reliability and responsiveness (Supplementary Table S2). 24 The questionnaire includes a question on how much overall allergic rhinitis symptoms are bothering the user on that day ('VAS Global Allergy Symptoms'), as well as one on how much allergic symptoms are affecting work on that day ('VAS Work'; only presented if the user reports to be working on that day). In addition to the VASs, the MASK-air ® daily monitoring questionnaire asks users whether they took medication or had been under AIT on that day. In the configuration of theirprofile, MASK-air ® users can provide information on their age, sex, country, allergen sensitisation, allergy symptoms, smoking status, and-if under AIT-type of AIT (SCIT or SLIT, including SLITtablet).
When responding to the MASK-air ® daily monitoring questionnaire, it is not possible to skip any of the questions, and data are saved to the dataset only after the final answer. This precludes any missing data.

| Size of the study
For each specific AIT type, we analysed all of the data available from European countries with at least 150 days of use/observations.

| Biases
There are potential information biases related to the self-reported nature of the data collection. Potential selection bias might be introduced due to the fact that app users are not representative of all patients with rhinitis.

| Data analysis
Categorical variables were described using absolute and relative frequencies, while continuous variables were described using medians and interquartile ranges. In MASK-air ® , each reporting day corresponds to an observation. We compared the days of patients under SCIT, SLITtablet and no AIT. The days of these different groups of patients were compared using VAS Global Allergy Symptoms, VAS Work, and a CSMS 37 (mixed hypothesis-and data-driven score calculated by multiplying VAS Global Allergy Symptoms by a medication factor; Supplementary Table S1).
To perform such comparisons, we applied three hierarchical models (also called 'multilevel models' or 'mixed-effects models')-one for each score. For each model, the type of grass AIT was a fixed effect, while random effects included identification of the user (nested within the respective country) and indication as to whether the observation occurred within or outside the grass pollen season (we used Bedard's method to assess the grass pollen season 25 ). In other words, we modelled the association between VAS and AIT type, Hierarchical models were applied using Bayesian methods. We opted for Bayesian approaches as they yield probability distributions of the parameters of interest (posterior probabilities) based on prior probability distributions and on the observed data. 26 That is, in this study, for each comparison, we obtained the posterior probability distribution for the average difference of VAS, retrieving the mean value and the respective 95% credible interval (CrI; range of values within which, with 95% probability, the true VAS difference lies. In Bayesian statistics, uncertainty is expressed through CrI and not through classical confidence intervals or pvalues). This is a methodological advantage as it informs us of the probability of each AIT type being associated with a lower VAS or CSMS, besides allowing for the obtention of a probability distribution that can be graphically plotted. Uninformative prior distri-   Table 1. The mean number of days reported per user is 38 for both no AIT and SCIT, and 47 for SLIT-tablet.

| Major results
Overall, patients receiving AIT had a lower median VAS Global allergy symptoms than patients under no AIT (6 vs. 10,     values tend to be higher on the first day of reporting than on subsequent days). 21,22 In addition, it is possible to hypothesise that days with patients feeling worse tend to be more frequently reported than those with patients experiencing no or mild symptoms, although this probably occurs in a non-differential way, irrespective of the AIT type under which the patients may be. However, a differential reporting bias may result from the fact that SLIT-tablet is taken every day, selfadministered by the patient, in contrast to SCIT, which is given every 4-6 weeks. On the other hand, as the MASK-air ® app is to be filled in every day, with each question concerning that specific day, recall bias may not have a substantial impact on our results.

| Strengths and limitations
Third, we did not follow a product-specific approach for each AIT application route (as recommended in current international guidelines on AIT 29,30 ), but analysed differences between routes of administration in a generic way. Moreover, we did not report on SLIT-drops because there was only one country with more than 150 reported days of SLIT-drops and SLIT-tablet use (France), and only one country with more than 150 reported days of both SLIT-drops and SCIT (Lithuania). In this country, the mean number of recorded days per user was substantially higher than that observed in other countries.
Fourth, we used, as previously, a cross-sectional design. [20][21][22] When we launched MASK-air ® , it was expected that patients would use the app regularly and that it would be possible to perform a longitudinal analysis. However, patients use the app for short periods of time (in this study, a mean number of 39 days were reported per user) and intermittently. Analyses for intermittent use (consecutive and non-consecutive data) have been performed. 21 This cross-sectional approach has been shown to be effective in raising new hypotheses, subsequently confirmed by epidemiologic studies. [31][32][33] However, the cross-sectional approach of this study precludes the establishment of a causal relationship between AIT use and reported symptoms, as well as the assessment of AIT adherence and therapy duration. Measurement of the latter variables would be particularly relevant given (i) the possible differences in the time needed for SLIT and SCIT to become effective, and (ii) the fact that SCIT may ensue more physician-patient interaction (as, contrary to SLIT, it cannot be self-administered by the patient at home). If patients tend to report symptoms more often when feeling worse or after establishing care with an allergist, this could imply a bias in the estimation of SLIT and SCIT efficacy (particularly in the first months after AIT initiation). Solutions in addressing these methodological biases of real-world evidence in AIT need to be elaborated in the future.

| Findings
While symptom levels are low, suggesting that AIT patients are not severe, a previous study has found median VAS levels of around 50/ 100 on the first day of reporting (when considering the pollen Abbreviations: CSMS, combined symptom-medication score; SCIT, subcutaneous immunotherapy; SLIT-tablet, sublingual AIT exclusively by tablets; VAS global allergy symptoms, visual analogue scale assessing the overall impact of allergic rhinitis symptoms on the user on that day; VAS work, visual analogue scale assessing the work impact of allergic rhinitis symptoms on the user on that day.
The effect of AIT on work was previously suggested by two studies using the MASK-air ® approach. 19

| Generalisability
The study was carried out in nine European countries and can be extended to the whole of Europe. However, it does not necessarily apply to countries where AIT is used with different allergen extracts and regimens.

| CONCLUSION
When compared to no AIT and to SCIT, SLIT was found to be associated with a better allergic rhinitis symptom control, impact on work and CSMS in patients with grass pollen allergy. By contrast, no such  Abbreviations: CSMS, combined symptom-medication score; SCIT, subcutaneous immunotherapy; SLIT-tablet, sublingual AIT exclusively by tablets; VAS global allergy symptoms, visual analogue scale assessing the overall impact of allergic rhinitis symptoms on the user on that day; VAS work, visual analogue scale assessing the work impact of allergic rhinitis symptoms on the user on that day.
-11 of 13 differences were observed when comparing SCIT to no AIT.
Following these results, future longitudinal studies are needed (taking potentially relevant confounders into account and following internationally harmonised standards for performing and reporting real-world data in AIT) to assess the effectiveness in improving allergic rhinitis control.

ACKNOWLEDGEMENT
MASK-air has been supported by Charité Universitätsmedizin Berlin, EU Grants (EU Structural and Development Funds, POLLAR: EIT Health POLLAR, EIT Health Twinning) and educational grants from Mylan-Viatris, ALK, GSK, Novartis and Uriach.
Open access funding enabled and organized by Projekt DEAL.