Family and Community Medicine Department, Faculty of Medicine, Jazan University, PO Box 2531, Jazan 45142, Saudi Arabia.
Received date: December 18, 2015; Accepted date: January 08, 2016; Published date: January 14, 2016
Citation: Mahfouz MS, Logistic Regression Models to Predict Factors Associated with Khat Abuse among Students in Jazan Region, Saudi Arabia. Electronic J Biol, 12:1
Khat abuse is associated with many public health problems. Logistic regression analysis is an important tool used in the analysis of the relationship between various explanatory variables and nominal response variables. The objective of this study was to use this statistical method to determine the factors which are considered to be significant contributors to the use or abuse of khat among students at both levels schools and university of Jazan region, Saudi Arabia. The logistic regression models were used to build models for predicting the most important factors associated with khat abuse among study participants. A total number of 7696 students responded to the survey, of which 56.9% were males and 43.1% females. The study results revealed the significant impact of important predictors of Khat chewing in our sample were students’ smoking status (OR=13.597, P value=0.000), gender (OR=5.283, P value=0.000), friends’ using tobacco (OR=3.844, P value=0.000) and friends’ using Khat (OR=1.368, P value=0.000). These results emphasize the importance of peer education interventions as one strategy for changing social norms in the studied population.
Khat chewing; Predictors; Logistic regression.
Research examining prevalence of khat chewing in Jazan region, kingdom of Saudi Arabia, suggests high prevalence of khat abuse among all components of Jazan region population, among them school and university students [1-5]. Khat use has been associated with severe public health problems [6- 10]. In addition to the health problems, khat use is associated with economic and social problems including time wasting, since the amount of time spent in khat chewing is considerable. Also another group of studies established clear association between heavy Khat consumption and psychosis [11-15].
Logistic regression first used during the 1970's as statistical modeling that overcome shortcoming of Ordinary Least square (OLS) . The growing literature on logistic regression as well as its simplicity in assumptions and application assisted in widespread of the use of the technique in most disciplines among which medical research [17-21].
Binary logistic regression is a type of predictive models which is used frequently when the dependent is a dichotomy and the independents or explanatory variables are of any type. The Logistic models uses maximum likelihood estimation after transforming the dependent into a logic variable (the natural log of the odds of the dependent occurring or not). Logistic regression models estimate the probability of a certain event occurring. Moreover logistic regression calculates changes in the log odds of the dependent, not changes in the dependent itself as OLS regression does .
In this article we used logistic regression model to predict the effect of some risk factors on the responses of khat abuse among students at two level of educational system, first intermediate and secondary schools, and second at University level of Jazan region southwest Saudi Arabia.
The data for this analysis was taken from two crosssectional surveys conducted by Jazan Substance Abuse Research Centre, Jazan University, KSA during the academic year 2011/2012. The first focused students in the basic education involving the two stages intermediate and secondary schools of Jazan region, while the second focused on university students in the region. Participants were selected using a three-stage cluster random sampling. A structured self-administered questionnaire was used for data collection.
Equation (1) can be written in the form in equation (2) using algebraic manipulation,
Equation (2), although it is linear in its right side it is nonlinear function of the response variable π. For this reason maximum likelihood is used to obtain these estimates .
Equation (3) provides the good feature of logistic regression, the Odds Ratios, which is also
Given by is known as the odds of the event y=1 occurring.
Data: ni observations at the ith of m distinct levels of the independent variable(s), with yi successes can be written in equation (4).
With Likelihood and log-Likelihood Functions:
The derivative of the log-likelihood wrt β χαν β can be given by:
While the Hessian matrix:
Using the Newton-Raphson-Algorithm, the final coefficients take the shape:
The proposed Logistic regression model in equation (2) was used to assess the significance of the explanatory variables in predicting the dichotomous response variable of khat use among study population. In the initial stage of analysis all study variables were tested, the aim was looking for factors that clearly demonstrated risk or protective properties and also for variables significant in the univariate regression (with a p-value<0.25). Risk factors are those factors believed to have a negative impact on the likelihood of khat abuse while protective factors are those factors that, when in place, are believed to significantly reduce the likelihood of khat abuse. After these factors were identified, the logistic regression procedure was used using stepwise selection method. This enabled us to select those significant variables which impact substance abuse, while at the same time removing those variables which have a lesser or no impact on khat use among study population. The final selected explanatory variables include, gender, tobacco use, students believes about khat effect on academic performance and health, peers use of tobacco and khat, family use of khat which involves whether father, mother, sister and brother khat use or not. All variables studied were dictums either the response yes or no.
SPSS computer program was used to generate indicator variables for the levels of each categorical predictor. Reference groups were chosen for each predictor as well, corresponding to which interpretation would be most useful for understanding khat chewing behavior. All explanatory variables of interest were tested first for possible interactions. Elimination was based on the highest insignificant term at each stage was sequentially removed until all variables fit within the desired 0.05 significance level. By hand elimination was used in order to keep control over which variables were being deleted. This was to make sure the final model would make sense clinically in relation to khat use. For those interactions that were significant, the main effects of the interactions were also kept in the model, regardless of significance.
Parameter estimate and odds ratio probabilities equal to or less than 0.05 were considered statistically significant and kept in the model. Control variables specified by the workgroup were left in the model with the other predictors. Individual parameter estimates were tested by the Wald statistics.
Goodness of fit and model assumptions was also checked. Some of these include multicollinearity of the predictor variables, ROC curves, comparison of AIC values, Likelihood Ratio Test and the Hosmer and Lemeshow goodness-of-fit test .
Table 1 presents some descriptive results about the study participants. According to the Table 7696 students responded to the survey, of whom 56.9% were males and 43.1 were females. The distribution of students according to educational institutions showed that schools students constituted 51.1%, while University students were 48.9%. The prevalence of khat chewing among students showed that 36.7% of males and 4% of females were khat chewers (Table 1).
|Health Related Colleges||600(13.7)||331(10.0)||931(12.1)|
Table 1: Some descriptive results.
Table 2 presents the logistic regression models for the important factors which found are with relationship with the dependent variable khat chewing among the study population. The models were arranged according to education school levels and type of faculties. The table showed that all listed variables were independent risk factors for khat chewing except history of anxiety and depression and the khat chewing status of the mother at the intermediate school level (P value less than 0.05 for all mentioned predictors variables).
|Level||Predictors||B||S.E.||Wald||Sig.||Exp(B)||95% C.I.for EXP(B)|
|Khat affecting Health=No||0.650||0.246||6.946||0.008||1.915||1.181||3.104|
|History of Anxiety=Yes||0.145||0.247||0.344||0.557||1.156||0.712||1.877|
|History of depression=Yes||0.131||0.244||0.288||0.591||1.140||0.706||1.840|
|Friend using Khat=Yes||1.067||0.280||14.481||0.000||2.906||1.677||5.033|
|Friend using tobacco=Yes||0.300||0.271||1.229||0.268||1.351||0.794||2.297|
|Khat affecting Academic=No||1.678||0.258||42.146||0.000||5.352||3.225||8.882|
|Father using Khat=Yes||0.838||0.217||14.931||0.000||2.312||1.511||3.537|
|Brother using Khat=Yes||0.682||0.220||9.606||0.002||1.977||1.285||3.042|
|Khat affecting Health=No||0.609||0.204||8.896||0.003||1.838||1.232||2.743|
|History of Anxiety=Yes||0.590||0.198||8.887||0.003||1.804||1.224||2.659|
|History of depression=Yes||0.258||0.185||1.957||0.162||1.295||0.902||1.860|
|Friend using Khat=Yes||1.207||0.250||23.306||0.000||3.344||2.049||5.460|
|Friend using tobacco=Yes||0.619||0.222||7.805||0.005||1.857||1.203||2.867|
|Khat affecting Academic=No||1.758||0.217||65.881||0.000||5.799||3.793||8.865|
|Father using Khat=Yes||0.290||0.172||2.830||0.093||1.336||0.953||1.873|
|Brother using Khat=Yes||0.441||0.172||6.542||0.011||1.554||1.109||2.179|
|Health related Colleges||Gender=Male||2.301||0.501||21.070||0.000||9.986||3.738||26.674|
|Khat affecting Health=No||1.365||0.290||22.098||0.000||3.914||2.216||6.914|
|History of Anxiety=Yes||0.313||0.369||0.723||0.395||1.368||0.664||2.817|
|History of depression=Yes||0.304||0.361||0.707||0.400||1.355||0.668||2.750|
|Friend using Khat=Yes||1.211||0.397||9.317||0.002||3.357||1.542||7.304|
|Friend using tobacco=Yes||0.458||0.331||1.913||0.167||1.581||0.826||3.027|
|Khat affecting Academic=No||1.401||0.292||22.986||0.000||4.061||2.290||7.202|
|Father using Khat=Yes||0.896||1.482||0.365||0.546||2.449||0.134||44.718|
|Brother using Khat=Yes||0.659||0.276||5.729||0.017||1.934||1.127||3.318|
|Khat affecting Health=No||1.422||0.345||17.022||0.000||4.144||2.109||8.142|
|History of Anxiety=Yes||-0.252||0.408||0.381||0.537||0.777||0.349||1.730|
|History of depression=Yes||-0.223||0.400||0.310||0.578||0.800||0.365||1.754|
|Friend using Khat=Yes||1.067||0.423||6.359||0.012||2.907||1.268||6.665|
|Friend using tobacco=Yes||-0.279||0.359||0.603||0.437||0.756||0.374||1.530|
|Khat affecting Academic=No||1.258||0.302||17.312||0.000||3.520||1.946||6.367|
|Father using Khat=Yes||21.774||19492.824||0.000||0.999||2.860E9||0.000||.|
|Brother using Khat=Yes||1.422||0.315||20.387||0.000||4.145||2.236||7.683|
|Sister= using Khat=Yes||-0.687||0.893||0.592||0.442||0.503||0.087||2.896|
|Khat affecting Health=No||1.681||0.245||46.913||0.000||5.374||3.321||8.694|
|History of Anxiety=Yes||-0.038||0.288||0.018||0.894||0.962||0.547||1.692|
|History of depression=Yes||0.360||0.295||1.486||0.223||1.433||0.804||2.555|
|Friend using Khat=Yes||1.817||0.414||19.298||0.000||6.155||2.736||13.846|
|Friend using tobacco=Yes||0.427||0.291||2.155||0.142||1.533||0.867||2.713|
|Khat affecting Academic=No||1.679||0.230||53.134||0.000||5.360||3.413||8.419|
|Father using Khat=Yes||16.552||40192.970||0.000||1.000||15432668.270||0.000||.|
|Brother using Khat=Yes||1.069||0.233||21.012||0.000||2.914||1.844||4.603|
Table 2: Logistic model for predictors of khat using-according to type of school or college.
For the secondary educational level, all factors remained as significant predictors for khat chewing, except History of depression and having a father chewing khat. For Health related colleges, Arts colleges and scientific colleges of university level, gender, tobacco use, the believe that khat chewing is not affecting health negatively, having a friend use khat or smoke, the believe that khat is not affecting academic performance negatively and having a brother using khat increases the likelihood of abusing khat (Table 2).
Table 3 illustrates logistic models for study participants according to gender. All mentioned variables were significantly associated and increasing the risk of khat chewing for males except having a sister using khat (p value=0.287). The female model in the same table suggested that the important factors were tobacco use which increases the risk khat chewing more than 10 times (OR=10.765; 95% CI: 6.221- 18.627; P value=0.000), the believe that khat has no negative health effect (OR=4.019; 95% CI: 2.28- 7.06; P value=0.000), friend using khat ((OR=3.176; 95% CI: 1.701-5.932; P value=0.000) (Table 3).
|Gender||Predictors||B||S.E.||Wald||Sig.||Exp(B)||95% C.I.for EXP(B)|
|Khat affecting Health=No||1.546||0.117||174.757||0.000||4.692||3.731||5.901|
|History of Anxiety=Yes||1.132||0.118||91.793||0.000||3.102||2.461||3.910|
|History of depression=Yes||0.246||0.128||3.723||0.054||1.279||0.996||1.643|
|Friend using Khat=Yes||0.307||0.129||5.682||0.017||1.359||1.056||1.748|
|Friend using tobacco=Yes||1.361||0.165||68.051||0.000||3.898||2.822||5.386|
|Khat affecting Academic=No||0.533||0.136||15.289||0.000||1.703||1.304||2.224|
|Father using Khat=Yes||0.545||0.110||24.703||0.000||1.725||1.391||2.139|
|Brother using Khat=Yes||0.854||0.110||60.691||0.000||2.350||1.895||2.913|
|Sister= using Khat=Yes||0.372||0.349||1.135||0.287||1.451||0.732||2.878|
|Khat affecting Health=No||1.391||0.288||23.348||0.000||4.019||2.286||7.067|
|History of Anxiety=Yes||0.542||0.353||2.360||0.124||1.719||0.861||3.431|
|History of depression=Yes||0.184||0.316||0.339||0.560||1.202||0.647||2.233|
|Friend using Khat=Yes||0.164||0.338||0.235||0.628||1.178||0.608||2.283|
|Friend using tobacco=Yes||1.156||0.319||13.148||0.000||3.176||1.701||5.932|
|Khat affecting Academic=No||-0.902||0.315||8.173||0.004||0.406||0.219||0.753|
|Father using Khat=Yes||0.216||0.273||0.628||0.428||1.241||0.727||2.119|
|Brother using Khat=Yes||0.646||0.279||5.375||0.020||1.907||1.105||3.292|
Table 3: Logistic model for predictors of khat using-according to gender.
Table 4 provides two logistic models for school and university students. According to the table all factors were statistically significant predictors for khat chewing among school students, whereas for university students, some of the variables were turned to be insignificant predictors for khat abuse, they were history of anxiety and depression and the khat chewing of the status of the sister and the believe that khat is not affecting academic performance negatively (Table 4).
|Control||Predictors||B||S.E.||Wald||Sig.||Exp(B)||95% C.I.for EXP(B)|
|Khat affecting Health=No||1.730||0.166||108.488||0.000||5.640||4.073||7.811|
|History of Anxiety=Yes||0.584||0.158||13.623||0.000||1.794||1.315||2.447|
|History of depression=Yes||0.484||0.158||9.351||0.002||1.622||1.190||2.211|
|Friend using Khat=Yes||0.320||0.153||4.355||0.037||1.377||1.020||1.859|
|Friend using tobacco=Yes||1.465||0.199||54.383||0.000||4.327||2.932||6.387|
|Khat affecting Academic=No||0.484||0.178||7.371||0.007||1.623||1.144||2.303|
|Father using Khat=Yes||0.486||0.140||12.120||0.000||1.626||1.237||2.139|
|Brother using Khat=Yes||0.677||0.140||23.325||0.000||1.968||1.495||2.591|
|Khat affecting Health=No||1.489||0.151||97.083||0.000||4.431||3.295||5.958|
|History of Anxiety=Yes||1.521||0.161||89.089||0.000||4.578||3.338||6.279|
|History of depression=Yes||0.012||0.194||0.004||0.949||1.013||0.692||1.482|
|Friend using Khat=Yes||0.156||0.194||0.650||0.420||1.169||0.800||1.708|
|Friend using tobacco=Yes||1.370||0.230||35.443||0.000||3.935||2.507||6.178|
|Khat affecting Academic=No||0.218||0.184||1.408||0.235||1.243||0.868||1.782|
|Father using Khat=Yes||0.527||0.150||12.286||0.000||1.694||1.262||2.275|
|Brother using Khat=Yes||0.985||0.151||42.688||0.000||2.677||1.992||3.597|
Table 4: Logistic model for predictors of khat using-according to gender.
Table 5 summarizes predictors of khat chewing among all study participants at both levels schools and university. The table suggested that the most important independent predictors of Khat chewing in our sample were students’ smoking status (OR=13.597, P value=0.000), gender (OR=5.283, P value=0.000), friends’ use tobacco (OR=3.844, P value=0.000) and friends’ use of Khat (OR=1.368, P value=0.000) (Table 5).
|Predictors||B||S.E.||Wald||Sig.||Exp(B)||95% C.I.for EXP(B)|
|Khat affecting Health=No||1.509||0.107||198.007||0.000||4.521||3.664||5.579|
|History of Anxiety=Yes||1.078||0.110||95.188||0.000||2.938||2.366||3.649|
|History of depression=Yes||0.213||0.117||3.308||0.069||1.238||0.984||1.558|
|Friend using Khat=Yes||0.281||0.119||5.581||0.018||1.324||1.049||1.671|
|Friend using tobacco=Yes||1.346||0.147||84.223||0.000||3.844||2.883||5.124|
|Khat affecting Academic=No||0.328||0.126||6.812||0.009||1.388||1.085||1.776|
|Father using Khat=Yes||0.512||0.101||25.586||0.000||1.668||1.368||2.034|
|Brother using Khat=Yes||0.806||0.101||63.211||0.000||2.240||1.836||2.733|
|Sister= using Khat=Yes||541||0.275||3.876||0.049||1.718||1.002||2.944|
Table 5: Logistic model for predictors of khat using-ll study participants.
The aim of this study was the use of logistic regression models, to determine the significant contributory factors that result in the use or abuse of khat among Jazan students at both levels of schools and universities. These factors were examined in two stages, stage; univariate analysis was used to determine the most important significant factors and then the use of these significant factors order to determine what were important among them in predicting khat use among the study participants. The study results highlight the significant impact of independent predictors of Khat chewing in our sample as follows: student's smoking status (OR=13.597, P value=0.000), gender (OR=5.283, P value=0.000), friends’ use tobacco (OR=3.844, P value=0.000) and friends’ use of Khat (OR=1.368, P value= 0.000). This result is further supported by other findings investigating factors associated with khat abuse in Saudi Arabia [2-4] in Yemen  and in Ethiopia [24,25].
Khat is a socially acceptable habit in Jazan and our findings suggest that khat control program efforts need to focus on peers to reduce the prevalence of the habit and its unfavorable consequences. Peer impact is higher among male users than among females. With both genders, these results emphasize the importance of health educational interventions as an important strategy for changing social norms in Jazan population.
The authors declare that they have no competing interests.
The author would like to thank Substance Abuse Research Center (SARC). The survey was funded by the Substance Abuse Research Center (SARC), Jazan University, Saudi Arabia under grant project no. 100.