以時間序列模式早期預測躁鬱症患者使用鋰鹽的治療反應

Abstract

躁鬱症(雙極性情感疾患)是一種情緒反覆在狂躁和憂鬱兩極之間擺盪的精神疾病，影響全世界數百萬的人口的健康和生活品質。藥物對於預後是一個重要的因素，其中鋰鹽被視為第一線預防復發的治療方式。由於其副作用影響七成到九成的病患，且僅有三成的人對治療有良好反應，科學家試著找出和療效相關的特徵和因子，但受限於測量方法和研究納入變項的不同，結果往往無法相互比較。再者，若能在病患早期使用鋰鹽便預測療效，可以幫助患者減少不必要的藥物嘗試。本篇研究目的以早期病程軌跡和相關基線預測變項結合，以其作為臨床上及時反饋之應用。 本研究招募來自台北三家醫院共232名服用鋰鹽的躁鬱症患者，以Alda準則測量鋰鹽的治療反應，分數大於7分者為良好反應，3到6分為部分反應，小於2分為沒有反應。病人的資訊藉由問卷和病歷所取得，變項包含臨床變項(如家族精神疾病史、發病年齡)，社會變項(如人格特質)以及人口學變項(如職業)等，根據不同時間獲得的資訊，可分成基準變項、短期變項及長期變項，若有遺失值則以K nearest neighbor演算法補值。我們使用逐步羅吉斯迴歸從病患初診可獲得的基準因子挑選與鋰鹽反應相關變項，並建立疾病發作後24個月的情緒狀態納入序列分析。 病患的臨床特徵會影響醫師的處方行為，被診斷為第一型躁鬱症、躁症發作、精神病家族史、有精神病症狀、較早發病且較不神經質的人較容易開立鋰鹽做為預防發作用藥。而在接受鋰鹽治療的病患當中，將基線變項放入羅吉斯模型，相比沒有反應或是部分反應者，對鋰鹽有良好反應者有較高的比例為首次鬱症發作、有躁鬱症家族史、有工作、較低的神經質；而有良好或是部分反應者則有較高比例為第一型躁鬱症、有工作、無抽菸習慣、較低的逃避傷害人格和外向特質。從發病後24個月序列分析的分類當中將病患分為三組：情緒穩定但未進行治療、接受精神科藥物治療和接受鋰鹽處方治療。首次發作後即接受鋰鹽治療組有較高的比例對鋰鹽有治療反應，未進行治療組則有較高的比例沒有反應，且他們多為首次鬱症發作且有抽菸習慣；對照服用鋰鹽前後的序列分析結果，對鋰鹽的反應並無顯著差異，這段時間情緒變化的軌跡能提供給臨床醫師的幫助十分有限。 在本研究當中，早期躁鬱症病程當中發現幾個可能在台灣族群中影響鋰鹽治療效果的相關因子和病程型態。使用羅吉斯回歸和序列分析幫助臨床醫師進行鋰鹽治療效果之預測，未來的研究更需要這樣的早期預測模式幫助躁鬱症患者實現個人化用藥處方。

Bipolar disorder is a serious and recurrent psychiatric illness, leading to cognitive and functional impairment. Lithium is one of good standard mood stabilizer treatments, but the clinical factors relevant to responder are inconsistent. As the number and variety of medication increase, reliable estimate of comparative efficacy is needed. The goal of the study is to combine the disease course trajectories and relevant baseline predictors in early illness phase so as to feedback to psychiatrists in a timely fashion. Characteristics of subjects with or without lithium prophylaxis lifetime may influence the decision whether psychiatrist to prescribe lithium or not. Lithium-treated subjects in our sample have higher proportion were BPD I and mania onset patients with psychotic symptom, family history of psychiatric disorder, younger age at onset and lower neuroticism. 232 lithium-treated subjects are included and assessed the degree of lithium response through Alda scale. They are classified into 3 treatment group: good responder (GR, total score=7-10), partial responder (PR, total score=3-6), and non-responder (NR, total score=0-2). The information of demographic are collected through personal interview and medical chart review. The K nearest neighbor algorithm are used for data imputation. We introduce two kinds of methodological approach: (1) a stepwise logistic model with baseline factors, and (2) sequence analysis by optimal matching to build typologies of the period after disease onset and before/ after lithium treatment. The logistic model reveals compared to PR and NR, GR have higher proportion of depression onset, BPD family history, employment and less neuroticism. Compared to NR, PR and tends to be diagnosed with BPD I with employed status, lower tobacco use, less harm avoidance and extraversion personality. In terms of the result of sequence analysis, there are three-cluster typology of disease onset, including stable without treatment, stable with treatment and lithium use since onset subgroup. Among the subgroups, the patients receiving lithium treatment soon after first episode are more likely to be GR while stable mood without treatment subgroup have highest proportion of being NR. When compare the typologies before/ after lithium treatment, the response rate has no different, which may imply that the transition has limit information for psychiatrist to make the clinical decision. The original combination of sequence analysis and logistic models is a powerful process that could be applied in the clinic as a timely prediction for lithium response to assist guiding patients’ treatment strategies. Further research is needed to bring us closer to achieve personalized medicine in bipolar disorder patients.