菸酒慣用者之危險性決策行為與非法藥物使用：以認知模型分析愛荷華賭博測驗

Abstract

物質使用者傾向做短視近利的決策，而非長期能帶來益處的決策。藥物濫用者的這類帶有風險的決策傾向，已能用愛荷華賭博測驗（Iowa Gambling Task, IGT）量測出。然而，在用藥進程相對較早期的物質使用者上，對於這種危險決策行為的了解相對較少。這篇研究的目的為：檢驗在物質使用的程度與決策行為特性上是否存在相關性，在三群不同物質使用程度參與者，未曾吸菸和未用非法藥物的控制組（共45人）、經常吸菸和飲酒者（共185人）、和使用非法或非處方藥物者（共55人）上使用增強式學習模型（reinforcement learning models）分析愛荷華賭博測驗的決策行為結果，並將男女性分層分析。主要研究對象來自2008和2009年在台灣的大台北地區針對常用菸酒者以回應者引介抽樣法收案的樣本，在2009和2010年做追蹤。我們使用展望評價學習模式（Prospect-Valence Learning model, PVL model）來分析並以階層性貝氏（Bayesian）估計法來估計決策行為中主要三個潛在的因素：避險行為參數（其值低則傾向有危險決策）、學習參數、和選擇中穩定性參數。主要的結果為：一、使用非法或非處方藥物者對最近期的事件學習力，相較於未曾吸菸和未用非法藥物者、與經常吸菸和飲酒者少；二、經常吸菸和飲酒者、與使用非法或非處方藥物者，都較未曾吸菸和未用非法藥物者傾向做風險性的決策；三、相較於其他兩群人，經常吸菸和飲酒者做決策時較不固定於某樣決定，男性特別有此傾向。這些主要的結果顯示用這些認知行為模式參數，能辨別用藥進程相對較早期的藥物使用者和合法物質使用者之間的差異。這些藥物使用者對於最近期的事件學習力較低，且他們會較固執地做帶有風險性的決策。經常吸菸和飲酒者也會做帶有風險性的決策，但他們在選擇上較不固定，且被最近期的事件影響較大。未曾吸菸和未用非法藥物者較固定選擇不具風險的決策。這些結果對於藥物使用的預防與介入提供有用的啟示。

Substance users tend to pay more attention to short-term gains yet neglect long-term losses. Performance deficits on the Iowa Gambling Task (IGT), which is widely used to measure such risky decision-making, have been found among drug abusers. However, it remains rarely known about whether such risky decision-making is present among subjects who are still in the early stage of their illegal drug use career. This study aimed to examine the associations between the magnitude of substance usage and decision-making characteristics as measured on the IGT using reinforcement learning models by comparing tobacco- and drug-naive controls (N = 45), regular tobacco and alcohol users (N = 185), and illegal/non-prescription drug users (N = 55), with stratification by gender. Study subjects were the participants recruited via respondent-driven sampling (RDS) among regular tobacco and alcohol users in Taipei metropolitan area in Taiwan in 2008, 2009, and 2010. We used the Prospect-Valence Learning model to estimate three main latent components in decision-making, i.e., individuals’ loss aversion (less risky tendency), learning ability, and choice consistency in trials, with Bayesian hierarchical estimation procedure implemented. The results showed that 1) illegal/non-prescription drug users learned less from the most recent event than both tobacco- and drug-naive controls and regular tobacco and alcohol users; 2) drug users and regular tobacco and alcohol users were more risk-seeking than tobacco- and drug-naive controls; 3) regular tobacco and alcohol users made choices more randomly than the other two groups, particularly in men. These results reveal that drug users in relatively early stage of their drug use career could be differentiated from legal substance users by cognitive modeling parameters. These drug users may have lower learning ability in the most recent event and be more obstinate of their relatively risky choices. The regular tobacco and alcohol users may make risky decisions, but change their decisions as a result of being more influenced by the most recent event. The tobacco- and drug-naive subjects are more fixed on their non-risky choices. These results have useful implications for the prevention and intervention of drug use.