Investigation of house dust mite induced allergy using logistic regression in West Bengal, India.

Background: The diagnosis of house dust mite (HDM) allergy based on Skin prick test (SPT) is not accurate, especially in lower risk cases. Our aim is to develop and validate a predictive model to diagnose the HDM allergic symptoms (urticaria, allergic rhinitis, asthma).

Methods: A forward-step logistic regression model was developed using a data set of 537 patients of West Bengal, India consisting of clinical variables (SPT based on 6 allergens of house dust and house dust mites, total IgE) and demographic characteristics (age, sex, house conditions). The output probability was estimated from the allergic symptoms shown by the patients. We finally prospectively validated a data set of 600 patients.

Results: The gradual inclusion of the variables increased the correlation between observed and predicted probabilities (correlation coefficient (r) = 0.97). The model development using group-1 showed an accuracy rate of 99%, sensitivity and specificity of 99.7% and 88.6% respectively and the area under the receiver operating characteristics (ROC) curve (AUC) of 99%. The corresponding numbers for the validation of our model with group-2 were 87%, 95.6% and 66% and 86% respectively. The model predicted the probability of symptoms better than SPTs in combination (accuracy rate 0.76-0.80), especially in lower risk cases (probability< 0.8) that are highly difficult to diagnose.

Conclusion: This is perhaps the first attempt to model the outcome of HDM allergy in terms of symptoms, which could open up an alternative but highly efficient way for accurate diagnosis of HDM allergy enhancing the efficiency of immunotherapy.