Research Projects

The menace of substance addiction among the youth of India is a severe threat and predicament for the overall development of the nation. Across the country several de-addiction clinics and centres are emerging to curb the menace through extensive counselling and medications. But the vexing fact is that often the recovery from substance use disorder is riddled with relapse. In a cohort of patients struggling with addiction, those who recover without experiencing a relapse within a certain time frame are considered to be cured. The remaining ones develop the relapse, termed as susceptible. A longitudinal study is essential to gain insight about the long-term outcomes of the concerned treatments and therapies. In this study, the objective is two-fold- to efficiently predict whether a patient will achieve complete recovery or if experience a relapse, then to estimate the time to relapse at any stage of treatment or thereafter based on a longitudinal study. A prediction model will be established which can embody the long-term cured introducing a fraction of immunes in the population and a latency distribution representing the relapse time of the susceptible. Moreover, it is evident that in addition to the impact of the administered treatments, various socio-economic factors and environmental stresses (termed as covariate) significantly influence the process of relapse recovery journey. Consequently, such effects must be incorporated into the prediction model for an accurate outcome. In this project, our main objective is to implement some flexible machine learning classifier to predict the decision boundary of cured and susceptible groups and assuming a semi-parametric form of the latency distribution for the relapse time, we will infer the relapse recovery experience of the patients with substance use disorder. Further, with the arising of time varying covariates, estimation of relapse time can be accomplished leveraging a dynamic proportional hazard model. Moreover, to accommodate individual-specific heterogeneity, incorporation of random effects can be implemented in the prediction model. For further advancement, this project proposes the integration of time-aware neural network architectures capable of capturing complex temporal patterns in longitudinal data. Another concern in the longitudinal study is that often the data is right censored, resulting in the patients' addiction status being unknown after the censoring time. Instead of discarding those data points and loosing information, advanced techniques like expectation maximization method will be implemented to impute those missing data. To the best of our knowledge, there is no such dynamic prediction model proposed and implemented within the substance addiction treatment framework. This work will substantially pave a new way on that direction. The model development and implementation will be accomplished based on real data collected from the Institute of Psychiatry-A Centre of Excellence, Kolkata and various other de-addiction clinics. In presence of huge data set and with application of such advanced tools, prediction of the recovery journey of addicted patients can be performed quite efficiently. With greater depth of analysis, the outcome of the study will effectively benefit the further decision making. Currently, there is a huge gap in knowledge regarding what constitutes evidence-based programs for managing substance abuse. This study must lead to gathering of evidence to guide the policymakers in formulating plans for the prevention of substance use. The findings of the proposed study enable the treatment team to plan the follow-up frequency of recovered patients and to identify the cases where possibility of relapse may be frequent. These identifications will enable the policy makers to make effective treatment decision so that relapse may be prevented and therefore advancing the process of bringing back the addicted youth into mainstream life.