Employing online social networks in precision-medicine approach using information fusion predictive model to improve substance use surveillance: A lesson from Twitter and marijuana consumption

Highlights

• Twitter data is a source of information to analyse marijuana prevalence.

• A model for marijuana use surveillance to get information fusion to support policymakers.

• Social network and opinion mining to analyse the population thoughts about marijuana.

Abstract

The impact that connected community have on precision health or medicine and vice versa offers opportunities for any type of research and survey information, e. g, to predict trends in health-related issues, specifically people behavior towards drug use. Here, precision medicine influences the way to treat the information and get a better outcome to support the stakeholder decision. Online social networks analysis seems to be good tools to quickly monitor the population behavior where users freely share large amounts of information related to their own lives on day- to- day basis. This novel kind of data can be used to get additional real time insights from people to understand their actual behavior related to drug use (Cortés et al., 2017). The aim of this research is to generated an information fusion model of marijuana use tendency confident enough to be employed by stakeholders. So, we will: (a) collect and process the data from Twitter; (b) design a set of algorithms to estimate the tendency of marijuana use in relation to age, localization and gender, moreover, used a set of processes and activities to verify if our model were performing as expected; and (c) fusion of the information in a model to fully characterize the marijuana use population comparable to the national marijuana consume survey for policy makers utilization to improve drug use prevention. First,we collect the data from Twitter accounts based in Chile using an algorithm for traversing graph data structures, we collected the data from Twitter accounts based in Chile. Then, we estimated marijuana user prevalence during a period from 2006 to 2018 and, within each of the years we predicted the prevalence of user population in relation with age (in range), the localization (regions) and gender. Finally, we built indicators to explore the similarity between data obtained through Twitter (our results) and the actual data collected by the National Service for the Prevention and Rehabilitation of Drug and Alcohol (SENDA) under the same variables analyze in their own survey. When we compare the results of the algorithms and methods developed by us with those provided by the SENDA, we observed that most of the indicators present similar trends, i.e., the variation of the prevalence by years in the age, location and gender, showed similar changes in both analyzes. Also, the algorithms effectiveness and capacity to predict variations of complex cases like marijuana use in Chilean population. This study is a key opportunity to obtain in a faster, low cost and continuous way information about marijuana use, also, is an excellent tool for marijuana surveillance to get information to support policy makers and stakeholder decisions.

Introduction

Precision medicine is a revolutionary field of modern medicine for disease treatment and prevention that include individual variability in genes, environment and lifestyle for each person [2], [3]. The prospect of precision medicine driven by Data Science (machine learning algorithms) and artificial intelligence applications in the modern world in tailoring the diagnostics, treatments, prevention, patient health and outcomes is very promising. Personalized precision medicine includes several benefits such as: (a) ensure that people get the correct treatment every time; (b) the emphasis is in prevention not reaction; (c) improved patient outcomes and (d) offers less cost effective and more efficient health care. In parallel presents ongoing challenges from regulation, reimbursement and clinical adoption to the economic value of the data. A precision medicine approaches, using shared-decision making with patients, policy makers and health workers offers an important opportunity to improve substance-use disorders (SUDs) prevention and treatment outcomes [3], [4]. Health and care experimented an exponential positive growth throughout eight innovative technologies as smart-phone, genome sequencing, machine learning, block-chain and the connected community among them [5]. The impact that connected community would have on precision health offering opportunities on research, distribution and survey information [1], [3], [6]. Everyday users generate large amounts of information to express themselves and behave as they naturally do in their daily lives [7], [8]. This novel type of data can be used to get additional real time insights from people to understand their actual behavior related to drug use. Social media analyses to assess marijuana use included manual labeling and the use of machine learning algorithms [9], [10], [11], [12], [13], [14], [15], [16].

Opinion Mining techniques have been used to label tweets related to marijuana [17], [18], [19], [20], [21]. Regardless of certain limitations, the system proposed in this project, works as complement to the already existent monitoring apparatus by identifying new trends in people behavior towards drug [1], [13], [14], [15], [16], [17]. United Nation Members are engaged in intensive discussions on the way forward to address the world drug problem. Risk factors and circumstances that can render people more vulnerable to illicit drugs (long-term effect on the brain), as well as facilitate the establishment and expansion of illegal markets, are often related to issues of development, rule of law and governance. Policies can never be pursued in isolation, and drug control is no exception [11], [17], [18]. Significant changes are taking place in the policy landscape surrounding cannabis legalization, production, use, prescription cannabis-derived medicinal products to patients with certain medical conditions and whether recreational cannabis use is associated with physical health problems later in life [20], [21], [22]. Most of the time the policymakers and stakeholders require ways to check or fast track data referring to drug use and control [23], [24], [25].

In Chile, marijuana has been given a particular attention due to extensive debate and relevance where Chile has the highest marijuana prevalence rate in Latin America¹ [26]. The National Service for the Prevention and Rehabilitation of Drug and Alcohol (SENDA) and its Drug Chilean Observatory conduct studies to collect information on the extent of use and risk perception, among other variables, every two years. SENDA, (2016) reports showed a systematic increase of drug consumption from 2010 and the prevalence is higher in the male population rather than the female. The group of people from 19 to 25 years showed the higher increased in drug consumption passing from 24,0% in 2014 to 33,8% in 2016 in comparison with the rest of population. The social environment can increase the risk of substance use and school presented a high risk of consumption [27]. In late adolescence and early adulthood, people who relates to others who use marijuana increase their risk of utilization [28]. Social Network Analyses (SNA) shows that greater proximity to peer substance users may be related to substance use and people who connect groups are especially affected [18]. Studies on the relationship between online network features and substance use by young adults showed that drug uses was associated with more connections and a high proportion of the network that discusses and accepts the drug use [10], [13], [15], [21], [28], [29], [30]. Marijuana use by peers on social media networks has also been associated with individual marijuana use among young adults [21], [28].

The system proposed in this project works as a novel source of fast track information fusion and control. SENDA conducts studies every two years, in a very time-consuming cycle and with continuous obstruction of the monitoring system. Therefore, our approaches works as complementary sources of data to improve and enrich the quality of information to the existent SENDA monitoring apparatus. Furthermore, we propose a new way to increasing the frequency of data collection from Twitter. Such data must be transformed into information to support future decision making. It is important to highlight the economic value of the data in low income countries depends on cleaver ways to lower the cost and digital technology opens this possibility. In this context, to get information in a fast way about people substance use can help us to indirectly prevent population addiction or dependence and also is an excellent tool for marijuana surveillance to support policymaker decisions.

This paper is facing the research problem of characterizing a complex phenomenon such us the marijuana consume through the analysis of related tweets, considering as research goal the construction of a confidence model in comparison with the national marijuana consume survey. In that sense, we have two main research questions to answer: Q1 Is it possible to extract from the texts related with marijuana written by Twitter user’s valuable information about the marijuana use? Q2: Is it possible to develop a model whose outcome is comparable with the data provided by the National marijuana consume survey? We are working only with Twitter user’s account holders living in Chile. Also, we considered only the people whose language expressed Chilean slang consequently we are certain to include only Chilean people. This means if we want to applied a similar analysis in other country in South America we must considered the special slang language used only by them. We assumed the relation between the Twitter user’s account holder tweet about marijuana and the probability of marijuana consumption.

The remainder of this work is structured as follows. In Section 2 we show an overview of social network analysis in social analysis, in particular its application on data collected from Twitter for studying a complex phenomenon as the marijuana use. Section 3 shows the methods and algorithms applied for analyzing the marijuana use of account holders in Twitter. The main results of this work are summarized in Section 4. Finally, Section 5 exhibits the conclusions from this work.