Enhanced Early Detection of Allergic Rhinitis: A Prospective Study on a Symptom‐Based Predictive Model Authors: Ke‐Zhang Zhu, Chao He, Si‐Zhe Zhu, Zheng Liu, Ming Zeng Published: 2026 ABSTRACT ABSTRACTIntroductionAllergic rhinitis (AR) and non‐allergic rhinitis (NAR) share overlapping symptoms but differ in pathophysiology and treatment. Current AR diagnosis relies on skin prick testing (SPT) and serum IgE quantification, both of which are complex. This study aimed to develop a symptom‐based model for early AR detection, explore allergen‐symptom relationships, and evaluate its performance.Material and MethodsA prospective cohort study was conducted at Wuhan Tongji Hospital between June 2024 and October 2024, enrolling 1150 patients with clinically suspected AR. Participants completed a visual analogue scale (VAS) questionnaire evaluating nasal symptoms (itching, congestion, sneezing, rhinorrhea), ocular symptoms, and overall discomfort, and the final diagnosis of AR was confirmed by SPT. Patients were randomly divided into training and test cohorts (8:2). Logistic regression (LR), the classic artificial intelligence‐machine learning algorithm, was used to build a prediction model after analyzing allergen‐symptom associations, with evaluation of discrimination, calibration, and clinical utility.ResultsA total of 758 patients (65.9%) were confirmed AR cases, and showed more severe nasal/ocular symptoms than NAR. Dust mites were the most common allergen, correlated with animal dander (r > 0.45) and negatively with age (r = −0.27), while allergen‐symptom specificity was generally low. Sneezing was the strongest AR predictor (AUC = 0.758) with the highest sensitivity, specificity, and F1 score, and the multivariable model combining all symptoms performed better (AUC = 0.771 training, 0.765 test), outperforming clinician experience. The model showed good calibration and stable performance across subgroups.DiscussionThis study developed a symptom‐based AR predictive model that outperformed clinician experience. Sneezing demonstrated the highest AUC value for prediction, and the multivariable LR model using nasal and ocular symptoms further improved accuracy.ConclusionsThe findings support VAS‐based screening as a practical, cost‐effective tool for early AR detection, therapeutic interventions, and targeted patient education regarding allergen avoidance strategies, helping optimize AR management, minimizing diagnostic delays, and facilitating precision treatment decisions. Introduction Chronic rhinitis, affecting 15%–40% of the global population, is a persistent inflammatory condition of the nasal mucosa lasting ≥ 12 weeks that is classifiable into two etiologically distinct subtypes: allergic rhinitis (AR) and non‐allergic rhinitis (NAR) [1, 2]. While both conditions present with overlapping symptoms, including nasal congestion, rhinorrhea, and sneezing, they differ fundamentally in pathophysiological mechanisms and thus require distinct therapeutic strategies [2, 3]. AR affects 10%–20% of the global population [4, 5, 6], with an estimated 240 million patients in China [7]. Its pathogenesis originates from an immunoglobulin E (IgE)‐mediated immune response triggered in atopic individuals upon exposure to allergens such as dust mites, animal dander, fungi, and pollen, which ultimately leads to non‐infectious inflammation of the nasal mucosa [8, 9]. Typical symptoms include nasal itching, nasal mucosal edema, and clear rhinorrhea, often accompanied by ocular symptoms such as eye itching, tearing, hyperemia, and burning sensation [9]. Severe or persistent cases may develop complications such as sinusitis, otitis media, and asthma [3]. In contrast, NAR is characterized by chronic nasal inflammation through non‐IgE pathways and commonly elicited by environmental irritants (e.g., temperature changes, odors), hormonal fluctuations, or medication effects [6]. Although AR and NAR share core clinical manifestations, their distinct mechanisms pose significant diagnostic challenges, especially given that approximately 50% of patients with rhinitis show no identifiable allergic sensitization [6]. Therapeutically, AR management focuses on allergen avoidance, symptomatic treatments (e.g., intranasal corticosteroids, antihistamines, and antileukotrienes), and causal therapies such as allergen immunotherapy (AIT) [8]. In contrast, NAR lacks specific treatments and primarily focuses on inflammation control and secretion clearance promotion. Some studies have suggested that capsaicin desensitization may have therapeutic potential [10]. Emerging evidence suggests 71% of NAR patients, classified as having vasomotor rhinitis mediated by neurogenic mechanisms, may exhibit suboptimal response to conventional AR therapies, underscoring the imperative for accurate differentiation [11]. This diagnostic dichotomy highlights the clinical need for improved early‐stage screening tools to guide mechanism‐specific therapeutic interventions and optimize long‐term outcomes. Emerging evidence indicates that prompt discrimination of AR versus NAR prevents treatment delays, optimizes pharmacotherapy selection, and mitigates secondary complications including asthma, while enhancing patient‐reported outcomes [12]. Currently, the preliminary diagnosis of AR relies on clinical assessment and physical examination by physicians [13, 14]. Skin prick tests (SPT) and serum IgE measurements remain the gold‐standard diagnostic procedures, typically requiring access to an allergy specialty clinic [9, 15], which are costly, time consuming, and labor‐intensive. The onset and severity of allergic diseases are significantly influenced by the specific type of allergen exposure. Common environmental allergens, including dust mites, pollen, and fungal spores, demonstrate distinct geographic and seasonal distribution patterns, which directly correlate with variations in allergic symptom presentation across different regions and time periods [16, 17, 18]. Although previous studies have made significant progress in developing AR predictive models, their clinical applicability remains constrained by several limitations. These include insufficient validation across diverse populations, variability in allergen exposure patterns, and challenges in integrating predictive algorithms into routine clinical practice. Existing models mainly rely on self‐reported symptom scores, which are often obtained through retrospective questionnaire surveys, making them susceptible to recall bias and subjective judgment errors [19, 20]. More importantly, many of these studies lack a clinical gold standard (SPT/serum‐specific IgE testing) for diagnosis [21, 22], raising concerns about the specificity and sensitivity of the models. Second, model development in these studies has largely overlooked the intrinsic heterogeneity of AR, and the symptom differences caused by different allergens (such as dust mites vs. pollen) have yet to be quantified [23, 24], making it difficult to support individualized treatment decisions. Furthermore, the relationship between AR symptom phenotypes, disease subtypes, and demographic characteristics (age and gender) remains unexplored. In this study, we analyzed the distribution patterns of various allergens and the prevalence of poly‐sensitization among patients, followed by an investigation of symptom profiles and their co‐occurrence patterns. We then explored the correlations between specific allergen exposures and clinical symptom phenotypes. Based on these findings, we developed and tested a symptom‐based diagnostic model for distinguishing AR from NAR with machine learning algorithms, while model performance rigorously evaluated against preliminary clinician experience using standardized metrics. Furthermore, we conducted stratified analyses to assess how demographic factors (age, gender) and symptom characteristics influenced patients' quality of life. Through this large‐scale prospective cohort study, we attempted to establish a relatively accurate, potentially clinically applicable diagnostic model that may help improve AR/NAR differentiation but also provide new insights into allergen‐symptom relationships and their impact on disease phenotypes. Our integrated approach successfully combined multidimensional clinical variables to create a scientifically robust yet practical tool for real‐world clinical decision‐making. Materials and Methods Study Design and Participants This prospective study of anonymous data collected from Wuhan Tongji Hospital between June 2024 and October 2024 (Figure 1), was registered with ClinicalTrials.gov NCT06474923. Flow chart of patient enrollment in the prospective study. The patients were randomly divided into the training cohort and test cohort in an 8:2 ratio. AR, allergic rhinitis; NAR, nonallergic rhinitis. All participants initially presented with AR symptoms as specified in the Allergic Rhinitis and its Impact on Asthma (ARIA) guidelines [9], including nasal congestion, rhinorrhea, and sneezing. After reviewing the medical history, otolaryngologists made a preliminary diagnosis of AR based on clinical experience [20, 25], and patients with acute diseases and other disorders that might compromise the accuracy of symptom self‐assessment were excluded from the study, as detailed in Supporting Information. Subsequently, they completed a self‐assessment symptom questionnaire [26, 27], evaluating nasal itching, congestion, sneezing, clear rhinorrhea, ocular symptoms (such as eye itching, tearing, and hyperemia), and overall discomfort using a visual analogue scale (VAS) from 0 to 10, where 0 indicated no impact on daily life and 10 represented a severe impact. Following questionnaire completion, all participants underwent SPT for 10 regionally prevalent inhalant allergens, with full methodological details available in the supplementary materials [28]. The final diagnosis of AR was confirmed by requiring both positive allergen test results and concordance with characteristic allergic symptoms [9], while other patients were ultimately diagnosed as NAR. This study complied with the Declaration of Helsinki and was approved by the Ethics Committee of Wuhan Tongji Hospital (TJ‐IRB202405023), with written informed consent obtained for medical data collection and use. Correlation Analysis To examine the relationship between symptom severity (measured by VAS scores ranging continuously from 0 to 10) and allergic sensitivity (assessed by SPT grades ordinally categorized from 0 [negative] to 4 [strongly positive]), we first analyzed the distribution characteristics of VAS scores across the study population. Statistical analysis was performed as follows: for normally distributed data, Pearson correlation coefficients were computed to evaluate linear relationships, while Spearman's rank correlation coefficients were employed for non‐normally distributed data to assess monotonic associations. To further enhance the robustness of the results [29, 30], the Bootstrap method (with 1000 resampling iterations) was applied to calculate the 95% confidence interval (95% CI) for the correlation coefficient, thereby quantifying the precision of the correlation estimate. The strength of the effect size was interpreted based on Cohen's criteria: (i) |r | < 0.10: no correlation; (ii) 0.10 ≤ |r | < 0.30: weak correlation; (iii) 0.30 ≤ |r | < 0.50: moderate correlation; (iv) |r | ≥ 0.50: strong correlation. Model Development and Evaluation All patients were randomly split into the training and test cohorts at an 8:2 ratio. After conventional machine learning algorithms‐univariable logistic regression (LR) analysis [31], a predictive model for distinguishing AR from NAR was constructed using multivariable LR algorithm with class weight adjustment in the training cohort and then tested in the test cohort [31]. Finally, the model performance was evaluated: (i) discrimination: receiver operating characteristic (ROC) curve with the corresponding area under curve (AUC) value; (ii) calibration: calibration curve and Hosmer‐Lemeshow test; (iii) clinical utility: decision curve analysis (DCA). The accuracy, sensitivity, specificity, and F1 score were calculated, and the nomogram and confusion matrices were also generated. Stratified Analysis Age and gender were further stratified to test multivariable LR models, assessing the association between stratified variables and AR diagnosis [32]. AR patients were divided into two groups based on symptom presentation: those with nasal symptoms alone and those with both nasal and ocular symptoms, to examine potential associations with allergen types and disease severity. Further stratified analyses were then conducted for more detailed comparisons between these subgroups. Finally, patients with AR were categorized into seasonal AR and perennial AR in accordance with the ARIA guidelines to validate whether the symptom‐based LR model could achieve favorable predictive performance across these AR subtypes [9]. Statistical Analysis All analyses were performed using R software (R Foundation for Statistical Computing, version 4.4.2) (Supporting Information: Table S1). Statistical tests were two‐tailed, and p < 0.05 was considered statistically significant. More information is provided in supporting material. Results Patient Characteristics This study analyzed 1150 patients in Wuhan Tongji Hospital from June 2024 to October 2024 as shown in Figure 1. Of these patients, 758 with high clinical suspicion for AR were subsequently confirmed through diagnostic testing, yielding a correct predictive ratio of 65.90% for the clinician experience (Table 1 and Figure 2A). No significant baseline differences were found in age and gender between AR and NAR, while AR patients exhibited significantly more severe nasal and ocular symptoms, as well as overall discomfort and higher VAS total scores compared to NAR patients (p < 0.001, Table 1, Supporting Information: Table S1, and Figure 2B,C). Table 1Demographic characteristics and symptom scores of AR and NAR patients.AR (n = 758)NAR (n = 392) p‐valuea Age, median (IQR), years23.0 (15.0–36.0)28.5 (14.0–40.0)0.152Gender, male, n (%)437 (57.5)213 (54.4)0.315Symptom, median (IQR)Nasal itching5.0 (3.0–7.0)3.0 (1.0–5.0) < 0.001 Nasal congestion6.0 (4.0–8.0)5.0 (2.0–8.0) < 0.001 Sneezing7.0 (4.0–8.0)3.0 (1.0–5.0) < 0.001 Clear rhinorrhea6.5 (4.0–8.0)4.0 (2.0–6.0) < 0.001 Ocular symptoms4.0 (1.0–7.0)2.0 (0.0–4.2) < 0.001 Overall discomfort7.0 (5.0–8.0)5.0 (3.0–8.0) < 0.001 Total VAS score34.0 (25.0–42.0)24.0 (15.0–32.0) < 0.001 Note: Data are presented as median for continuous variables and numbers (percentage) for categorical variables, analyzed by Mann–Whitney U test between two groups.Abbreviations: AR, allergic rhinitis; IQR, interquartile range; NAR, non‐allergic rhinitis; VAS, visual analogue scale. a p < 0.05 indicate significant differences.John Wiley & Sons, Ltd. Demographic characteristics and symptom scores of AR and NAR patients. Note: Data are presented as median for continuous variables and numbers (percentage) for categorical variables, analyzed by Mann–Whitney U test between two groups. Abbreviations: AR, allergic rhinitis; IQR, interquartile range; NAR, non‐allergic rhinitis; VAS, visual analogue scale. p < 0.05 indicate significant differences. Patients' clinical data, symptom scores, and allergens. (A) Proportion of AR and NAR patients. (B) Proportion of male and female patients. (C) Symptom scores of all patients. (D) Correlation heatmap of allergens, age, and symptom scores. The color scheme of Spearman's rank correlation symbols indicates the degree of correlation. AR, allergic rhinitis; Der p, Dermatophagoides pteronyssinus; Der f, Dermatophagoides farinae; NAR, non‐allergic rhinitis; VAS, visual analogue scale. Among the 758 patients who were ultimately diagnosed with AR, 280 (36.94%) were exclusively allergic to dust mites. A further 152 patients (20.05%) did not have an allergy to dust mites but demonstrated sensitivity to other allergens. Meanwhile, 326 patients (43.01%) were allergic to both dust mites and other allergens. These findings suggest that relying solely on the SPT for dust mite allergy detection would likely result in a missed diagnosis for over 80% of AR patients. The study population exhibited distinct symptom patterns: 149 patients (19.66%) manifested isolated nasal symptoms, while the majority (609 patients, 80.34%) presented with concurrent nasal‐ocular symptoms. Notably, nearly half of these patients (356 cases, 46.97% of total cohort) reported severe ocular involvement (VAS ≥ 5), demonstrating that ocular symptom burden constitutes a clinically significant component of AR presentation in approximately 50% of affected individuals. Correlation analysis of SPT grading, age, and symptom severity (Figure 2D) showed that sensitization levels were highly correlated between the two dust mite species (r = 0.97). Cat and dog dander sensitization also demonstrated strong mutual correlation (r = 0.72) and moderate correlation with dust mites (r > 0.45). Among pollen allergens, the three weed species exhibited moderate correlations in sensitization levels (r ≥ 0.40). Analysis of allergen sensitization patterns revealed distinct age‐related associations: dust mite sensitization demonstrated a weak inverse correlation with age (r = −0.27), indicating higher prevalence among pediatric populations. In contrast, no clinically meaningful correlations were observed between other allergen sensitizations and either age or symptom severity measures (all |r | < 0.20). The analysis revealed minimal associations between common allergen types and specific symptoms (all |r | < 0.12). In contrast, symptom scores showed moderate to strong intercorrelations, with particularly high collinearity observed between the total VAS score and individual symptom components (all r ≥ 0.60). Univariable Logistic Regression Analysis of Symptom VAS Score To evaluate the diagnostic utility of symptom profiles in differentiating AR from NAR, we performed univariable LR analyses in the training cohort. All assessed symptoms, including nasal pruritus, congestion, paroxysmal sneezing, rhinorrhea, ocular manifestations, global discomfort, and composite VAS score, demonstrated statistically significant positive associations with AR diagnosis (all p < 0.001, Figure 3A). Notably, sneezing episodes showed the most pronounced association, yielding the highest odds ratio among all evaluated symptoms. To assess the predictive capacity of individual symptoms and the total symptom score for AR, we conducted univariable LR analyses. As shown in Figure 3B,C, sneezing emerged as the most predictive symptom, achieving the highest area under the curve (AUC = 0.758) with an optimal cutoff score of 3.483. Meanwhile, its corresponding sensitivity, specificity, and F1 score were also higher than other single symptoms as shown in Table 2, which confirmed its optimal predictive performance. This finding is consistent with clinical observations, as sneezing represents both a highly specific AR symptom and one that patients can readily self‐evaluate objectively. The analysis yielded an odds ratio of 1.415 (95% CI: 1.339–1.499) for sneezing. Subsequent predictive symptoms included clear rhinorrhea (cutoff score = 5.486, AUC = 0.662) and ocular symptoms (cutoff score = 4.485, AUC = 0.654), both of which also demonstrated relatively high specificity for AR diagnosis. However, clear rhinorrhea may overlap with acute rhinitis symptoms, while ocular symptoms are often mild in half of AR patients and may overlap with other eye conditions. Nasal itching (cutoff score = 3.489) showed relatively lower predictive value (AUC = 0.635), possibly due to subjective reporting differences. Notably, nasal congestion (cutoff score = 3.493) had the lowest predictive accuracy (AUC = 0.588), as this symptom commonly occurs in various nasal disorders including chronic rhinosinusitis, nasal septal deviation, and nasal tumors. Univariable logistic regression analysis of symptom VAS score. (A) Age and symptom VAS score associated with allergic rhinitis diagnosis. Error bars represent limits of the 95% confidence interval for the odds ratio. (B, C) Receiver operating characteristic curves for univariable symptom VAS score to predict AR in training and test cohorts. AR, allergic rhinitis; AUC, area under the curve; CI, confidence interval; NAR, non‐allergic rhinitis; OR, odds ratio; ROC, receiver operating characteristic; VAS, visual analogue scale. Table 2Predictive ability of the multivariable and univariable LR models.CohortModelAUC (95% CI)TPFNFPTNACCSensitivitySpecificityF1 score p‐valuea Training cohortMultivariable model0.771 (0.738–0.804)4671391072060.7320.8140.5970.792Nasal itching model0.635 (0.597–0.673)3882181301830.6210.7490.4560.690 < 0.001 Nasal congestion model0.588 (0.549–0.628)4651411931200.6370.7070.4600.736 < 0.001 Sneezing model0.758 (0.724–0.793)4981081311820.7400.7920.6280.806 0.012 Clear rhinorrhea model0.662 (0.625–0.700)4261801422120.6500.7500.4870.726 < 0.001 Ocular symptoms model0.654 (0.618–0.690)281325742390.5660.7920.4240.585 < 0.001 Overall discomfort model0.616 (0.578–0.655)3922141471660.6070.7270.4370.685 < 0.001 Total VAS score model0.713 (0.678–0.748)398208962170.6690.8060.5110.724 < 0.001 Test cohortMultivariable model0.765 (0.699–0.831)1104225540.7100.8150.5620.767Nasal itching model0.614 (0.535–0.692)985439400.5970.7150.4260.678 < 0.001 Nasal congestion model0.588 (0.510–0.665)1124052270.6020.6830.4030.709 < 0.001 Sneezing model0.731 (0.661–0.801)1173536430.6930.7650.5510.767 0.006 Clear rhinorrhea model0.660 (0.585–0.736)906229500.6060.7560.4460.664 0.005 Ocular symptoms model0.653 (0.578–0.727)767624550.5670.7600.4200.603 0.002 Overall discomfort model0.590 (0.511–0.669)1044841380.6150.7170.4420.700 < 0.001 Total VAS score model0.709 (0.638–0.780)995328510.6490.7800.4900.710 0.003 Abbreviations: ACC, accuracy; AUC, area under curve; CI, confidence interval; FN, false negative; FP, false positive; LR, logistic regression; NPV, negative predictive value; PPV, positive predictive value; TN, true negative; TP, true positive; VAS, visual analogue scale. a DeLong' test between the multivariable model and each univariable model in both the training and test cohorts, while p < 0.05 indicate significant differences.John Wiley & Sons, Ltd. Predictive ability of the multivariable and univariable LR models. Abbreviations: ACC, accuracy; AUC, area under curve; CI, confidence interval; FN, false negative; FP, false positive; LR, logistic regression; NPV, negative predictive value; PPV, positive predictive value; TN, true negative; TP, true positive; VAS, visual analogue scale. DeLong' test between the multivariable model and each univariable model in both the training and test cohorts, while p < 0.05 indicate significant differences. Multivariable Logistic Regression Analysis of Symptom VAS Score To address the significant collinearity between the Total VAS score and individual symptom scores (Figure 2D), we developed a multivariable LR model incorporating specific symptom measures. The final predictive model for the logit of AR occurrence probability was expressed by the equation: logit (P) = −0.375 − 0.070 × nasal itching + 0.049 × congestion + 0.386 × sneezing + 0.021 × clear rhinorrhea + 0.102 × ocular symptoms − 0.105 × overall discomfort (Figure 4A). This model demonstrated robust diagnostic performance with an optimal cutoff value of 0.748, yielding 81.4% sensitivity and 59.7% specificity. The AUC reached 0.771 in the training cohort and 0.769 in the test cohort (Figure 4A). Statistical comparison confirmed the model's superior predictive accuracy over all univariable LR analyses (DeLong test, all p < 0.015; Table 2). Multivariable logistic regression analysis of symptom VAS score. (A) Receiver operating characteristic curves for multivariable symptom VAS score to predict AR in training and test cohorts. (B, C) Confusion matrix demonstrating the evaluation of the multivariable LR model for discriminating AR and NAR in training and test cohorts. The horizontal axis represents the actual diagnosis of patients, while the vertical axis represents the predictions of the multivariable LR model, and the color scheme on the right indicates the numbers of patients. (D) The nomogram developed on the basis of the multivariable LR model. (E) Calibration curves of the multivariable LR model in training and test cohorts. The diagonal gray dashed line indicates a perfect prediction, and the solid lines indicate the performance of the models. When the solid line is close to the dotted line, the model works well. (F) The decision curve analysis of multivariable LR model in training and test cohorts. The gray line represents the assumption that “all patients are AR”, and the dashed black line represents the assumption that “none are AR”. The multivariable LR model shows high net benefit in a larger range of threshold probability, indicating the excellent clinical utility ability. AR, allergic rhinitis; AUC, area under the curve; CI, confidence interval; DCA, decision curve analysis; NAR, non‐allergic rhinitis; OR, odds ratio; ROC, receiver operating characteristic; VAS, visual analogue scale. The final analysis demonstrated that 838 patients (as detailed in the confusion matrices: true positives [467] + true negatives [206] in training cohort and true positives [54] + true negatives [110] in test cohort, Figure 4B,C) were correctly diagnosed, representing 72.78% of the total patient cohort. This diagnostic accuracy rate significantly surpassed (p < 0.001) the 65.90% of clinicians' experience. These results indicate that the VAS‐based symptom scoring system provides superior diagnostic performance compared to clinicians' experience, while maintaining advantages as a rapid, cost‐efficient tool that requires no additional diagnostic procedures. To enhance clinical utility for early screening, we developed an easy‐to‐use nomogram that visually represents the scoring system (Figure 4D). The model demonstrated excellent calibration in both training and test cohorts, as evidenced by non‐significant Hosmer‐Lemeshow test results (p > 0.05) and corresponding calibration curves (Figure 4E). Decision curve analysis (DCA) confirmed the model's strong clinical value, showing significant net benefit across a wide range of threshold probabilities (Figure 4F). Finally, we implemented this scoring system as an online webtool (https://tongjiardata.shinyapps.io/DynNomapp/), enabling self‐assessment for patients and facilitating outpatient screening. Stratified Analysis Results Stratified analyses revealed distinct patterns across demographic groups. Age‐based stratification (using 20th percentiles as the stratification cutoff, Supporting Information: Figure S1) showed that 80% of enrolled patients were younger than 40 years, suggesting that adolescents and young adults experience higher AR incidence rates, greater disease burden, and more proactive healthcare‐seeking behavior, with all age strata maintaining AUC values > 0.700. Similar predictive performance patterns were observed in gender‐based analyses (Supporting Information: Figure S2) and when comparing patients with nasal symptoms alone versus those with both nasal and ocular symptoms (Supporting Information: Figure S3). According to ARIA guidelines, AR patients were classified into seasonal and perennial subtypes, with stratified comparative analyses conducted for seasonal AR versus all NAR and perennial AR versus all NAR. Both the univariable symptom‐based LR model (sneezing as the key symptom) and the multivariable LR model exhibited good discriminatory performance for the two AR subtypes (Supporting Information: Figure S4 and Table S3). Discussion AR and NAR present with overlapping clinical manifestations yet require distinct therapeutic approaches. Current diagnostic protocols require physicians to initiate SPT or serum‐specific IgE testing based on preliminary clinical judgment. However, empirical test selection often lacks precision, leading to a high rate of negative results, which waste resources and place unnecessary burdens on patients. This underscores the critical need for accurate early screening strategies to optimize diagnostic efficiency and minimize redundant testing. In this large‐scale prospective cohort study, we developed a predictive diagnostic model for AR through comprehensive analysis of patient‐reported VAS scores and SPT results, while exploring the interrelationships between clinical symptoms, disease manifestations, and various allergens. Our study successfully developed an AI‐based VAS symptom preliminary predictive model for AR, showing better diagnostic accuracy than conventional clinicians' experience. The model maintained robust discrimination in the test cohort and demographic subgroups, with calibration and decision curve analyses evaluating its clinical reliability. This tool effectively bridges the gap between symptom reporting and objective diagnostic confirmation, offering a standardized approach to prioritize confirmatory testing. In our cohort, the overall prevalence of both AR and NAR was slightly higher in males than in females, consistent with previous epidemiological studies [33]. However, we found that the majority of patients in our study were under the age of 40, suggesting that younger individuals may be more symptomatic, more affected in terms of quality of life, or more likely to seek medical care. Similar patterns have been reported in large‐scale surveys, which show that around 80% of AR symptoms begin before age 20, peak between 20 and 40, gradually decline thereafter [34], potentially due to higher allergen exposure and immune responsiveness during this life stage. Age was also associated with differences in sensitization profiles. Consistent with prior studies [35], children in our cohort were more frequently sensitized to dust mites, while adults tended to show polysensitization involving both indoor (e.g., animal dander) and outdoor allergens (e.g., pollen). These age‐specific patterns may reflect both cumulative allergen exposure and evolving immune responses over time. Approximately 43.01% of patients exhibited polysensitization, a finding comparable to previous reports ranging from 30% to 60% in AR populations [36, 37]. Polysensitization complicates both allergen avoidance and immunotherapy due to the broader range of triggers involved. Notably, our data revealed strong correlations between Dermatophagoides farinae and Dermatophagoides pteronyssinus, which differs somewhat from patterns observed in Europe, where pollen sensitization rates are typically higher [38]. Interestingly, we also identified moderate correlations between dust mites and animal dander, likely reflecting environmental overlap, such as dust mites residing in pet fur or shared indoor habitats. These findings highlight the importance of comprehensive environmental control measures and targeted patient education. Previous studies have found that allergen‐specific IgE levels do not always correlate with clinical symptoms, especially in polysensitized individuals [39]. Further analysis of the SPT results and symptom characteristics shows that different allergens have low symptom specificity and limited association with symptom severity. This lack of one‐to‐one correlation supports the continued use of broad‐spectrum symptomatic treatments, such as intranasal corticosteroids, oral antihistamines, and leukotriene receptor antagonists, as the mainstay of AR management, independent of allergen type. Overall, our findings reinforce prior observations regarding demographic patterns, sensitization complexity, and the clinical management of AR, while offering updated insights based on a real‐world cohort. Future work may explore how molecular or immunological endotyping could further improve the precision of AR diagnosis and treatment. Some AR scoring systems are also applied in clinical practice, such as the International Study of Asthma and Allergies in Childhood (ISAAC) questionnaire for children and adolescents [40, 41]. Although they have performed well in specific populations, they generally face challenges in broader application. Our model is built on a simplified symptom scale and thus may offer some advantages in routine clinical practice over more complex clinical scoring systems, serving as a supplementary approach to existing diagnostic protocols and scoring methods. Compared to models relying on immunological testing, biomarker detection, or machine learning algorithms [17], which often require specialized equipment, high‐cost assays, or complex data input, symptom‐based tools are simpler and more feasible for large‐scale preliminary screening. Prior models, such as those using retrospective total nasal symptom score (rTNSS) or ocular symptom score (rTOSS) [42], involve numerous assessment items and face challenges including recall bias and questionnaire fatigue, limiting their generalizability and real‐time applicability. In contrast, by focusing on core nasal symptoms and using a straightforward scoring method, symptom‐based LR models can still achieve competitive predictive performance (AUC = 0.758 in training cohort, and 0.731 in the test cohort). Although models based on objective indicators and machine learning may offer higher precision in select cohorts, symptom‐based models like ours are particularly valuable in primary care or public health settings where rapid, low‐cost screening tools are needed. This highlights their potential to serve not only as diagnostic aids but also as tools for monitoring treatment response and guiding early clinical intervention. This study has important implications and broad applications for the screening, initial diagnosis, and health management of AR patients. In addition, the webtool has further enhanced the practicality of this predictive model. In terms of screening, patient self‐assessment is convenient and straightforward, allowing for rapid identification of high‐risk individuals. For initial diagnosis, a questionnaire‐based prediction model (proportion 72.78%) offers higher accuracy than preliminary physician assessments (proportion 65.90%), making it suitable as a potential valuable adjunct tool for primary assessments. In guiding diagnosis and treatment, the model helps clarify the impact of different symptom phenotypes on patients' quality of life, and clinicians can precisely select treatment methods and medications based on individual patient symptoms. In health management, the symptom scoring system has the potential to evaluate the reduction in the frequency and severity of allergic episodes in patients after interventions and protective measures. Despite yielding several important findings in AR research, this study has limitations that suggest directions for future improvement. First, patients were recruited from a single center, which may not fully represent the AR population nationwide or globally. Second, clinical data were limited to gender, age, and self‐reported symptoms. Since symptom self‐assessment is subjective, it may not comprehensively reflect the objective severity of symptoms. Additionally, allergen type distribution is influenced by climate and seasonal factors, and our limited recruitment period may affect the observed primary allergen types and the robustness of the analysis in this study. Future studies should aim to broaden the recruitment scope and duration, and include patients from diverse regions and ethnic backgrounds to improve the generalizability of the findings. Conclusion This study provides important evidence supporting symptom‐based screening of AR, suggesting that early diagnostic prediction using VAS scores may effectively control symptoms and facilitate targeted treatment, thereby helping to prevent disease progression. A deeper understanding of the pathogenesis and clinical features of AR will further strengthen efforts to improve prevention and treatment strategies. Author Contributions Ke‐Zhang Zhu: data acquisition, data analysis, and manuscript preparation. Chao He: manuscript preparation. Si‐Zhe Zhu: data acquisition. Zheng Liu and Ming Zeng: study design and manuscript preparation. All authors approved the manuscript. Ethics Statement This study was approved by the Ethics Committee of Wuhan Tongji Hospital (TJ‐IRB202405023). Conflicts of Interest The authors declare no conflicts of interest. Professor Zheng Liu is a member of World Journal of Otorhinolaryngology – Head & Neck Surgery (WJOHNS) editorial board and is not involved in the peer review process of this article. Supporting information for typesetting‐Supplementary Figures. for typesetting‐Supplementary Information.